RESEARCH IN ECONOMIC HISTORY
i
RESEARCH IN ECONOMIC HISTORY Series Editor: Alexander J. Field
ii
RESEARCH IN ECONOMIC HISTORY VOLUME 23
RESEARCH IN ECONOMIC HISTORY EDITED BY
ALEXANDER J. FIELD Department of Economics, Santa Clara University, USA CO-EDITED BY
GREGORY CLARK Department of Economics, University of California, Davis, USA
WILLIAM A. SUNDSTROM Department of Economics, Santa Clara University, USA
Amsterdam – Boston – Heidelberg – London – New York – Oxford Paris – San Diego – San Francisco – Singapore – Sydney – Tokyo iii
ELSEVIER B.V. Radarweg 29 P.O. Box 211 1000 AE Amsterdam, The Netherlands
ELSEVIER Inc. 525 B Street, Suite 1900 San Diego CA 92101-4495 USA
ELSEVIER Ltd The Boulevard, Langford Lane, Kidlington Oxford OX5 1GB UK
ELSEVIER Ltd 84 Theobalds Road London WC1X 8RR UK
r 2005 Elsevier Ltd. All rights reserved. This work is protected under copyright by Elsevier Ltd, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier’s Rights Department in Oxford, UK: phone (+44) 1865 843830, fax (+44) 1865 853333, e-mail:
[email protected]. Requests may also be completed on-line via the Elsevier homepage (http://www.elsevier.com/locate/permissions). In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 20 7631 5555; fax: (+44) 20 7631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of the Publisher is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier’s Rights Department, at the fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made. First edition 2005 British Library Cataloguing in Publication Data A catalogue record is available from the British Library. ISBN-10: 0-7623-1262-9 ISBN-13: 978-0-7623-1262-7 ISSN: 0363-3268 (Series)
∞ The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.
Working together to grow libraries in developing countries www.elsevier.com | www.bookaid.org | www.sabre.org
iv
CONTENTS LIST OF CONTRIBUTORS
vii
INTRODUCTION Alexander J. Field
ix
A SOVIET QUASI-MARKET FOR INVENTIONS: JET PROPULSION, 1932 –1946 Mark Harrison
1
NETWORK QUALITY IN THE EARLY TELEGRAPH INDUSTRY Tomas Nonnenmacher
61
THE SPANISH INFRASTRUCTURE STOCK, 1844–1935 Alfonso Herranz-Lonca´n
83
HAVE AMERICAN WORKERS ALWAYS BEEN LOW SAVERS? PATTERNS OF ACCUMULATION AMONG WORKING HOUSEHOLDS, 1885–1910 John A. James, Michael G. Palumbo, and Mark Thomas
127
WORKER ABSENTEEISM UNDER VOLUNTARY AND COMPULSORY SICKNESS INSURANCE: CONTINENTAL EUROPE, 1885–1908 John E. Murray
177
v
vi
CONTENTS
URBAN REAL WAGES AROUND THE EASTERN MEDITERRANEAN IN COMPARATIVE PERSPECTIVE, 1100–2000 S; evket Pamuk
209
JAPANESE UNSKILLED WAGES IN INTERNATIONAL PERSPECTIVE, 1741–1913 Jean-Pascal Bassino and Debin Ma
229
RELATIVE BRITISH AND AMERICAN INCOME LEVELS DURING THE FIRST INDUSTRIAL REVOLUTION Marianne Ward and John Devereux
249
LIST OF CONTRIBUTORS Jean-Pascal Bassino
Faculty of Mathematics and Social Science, Paul Vale´ry University, Montpellier, France
John Devereux
Department of Economics, Queens College, NY, USA
Alexander J. Field
Department of Economics, Santa Clara University, CA, USA
Mark Harrison
Department of Economics, University of Warwick, Coventry, UK
Alfonso Herranz-Lonca´n
Department of Economic History, University of Barcelona, Barcelona, Spain
John A. James
Department of Economics, University of Virginia, VA, USA
Debin Ma
GRIPS/FASID Joint Graduate Program, National Graduate Institute for Policy Studies, Tokyo, Japan
John E. Murray
Department of Economics, University of Toledo, OH, USA
Tomas Nonnenmacher
Department of Economics, Allegheny College, PA, USA
Michael G. Palumbo
Board of Governors, Federal Reserve System, DC, USA
S; evket Pamuk
Ataturk Institute for Modern Turkish History and Department of Economics, Bogazici University, Istanbul, Turkey
Mark Thomas
Department of History, University of Virginia, VA, USA
Marianne Ward
Department of Economics, Loyola College, MD, USA vii
This page intentionally left blank
viii
INTRODUCTION This volume of Research in Economic History (REH) includes eight papers, five of which were submitted and evaluated through our regular channels. An additional three were solicited from among those presented at the conference ‘‘Toward a Global History of Prices and Wages,’’ held in Utrecht in August of 2004. Because of the emphasis of these papers on data and the relevance of their findings for our understanding of long-run economic growth and development in different parts of the world, we encouraged a number of authors from this conference to submit their work to REH. Associate editor Gregory Clark took responsibility for soliciting, refereeing, selecting, and editing the submissions. We anticipate publishing up to three more of these in the next volume, enriching both REH and our understanding of economic history. Our lead article is Mark Harrison’s fascinating study of the ‘‘quasi-market’’ for inventions that emerged as the Soviet Union struggled to develop jet aircraft between 1932 and 1946. This was a period in which it was quite unclear exactly what technological route would lead to a successful vehicle. Within this uncertain environment, Soviet rulers were especially dependent on the ingenuity and brilliance of their engineers. How could they be best motivated? Drawing principally from archival sources, Harrison shows how Soviet rulers exploited engineers’ desires for inventive priority to extract creative effort. When the ‘‘correct’’ path forward became clearer after the war, and innovation became more routinized, the Soviet system had to resort to very large cash awards to continue to keep research engineers engaged. We follow with another paper on industrial organization, Tomas Nonnenmacher’s study of the early years of the telegraph industry in the United States. Between 1845 and 1866 the industry’s organization moved from monopoly to competition to cartels and then back again to monopoly. Nonnenmacher explains why, in a network industry with local monopolies, the phenomenon of double marginalization led to higher prices and poorer quality. For long-distance transmissions, firms had to rely on the actions of other firms down the line to ensure that the message got through and thus provide satisfaction for their customers. It was very hard to write effective ix
x
INTRODUCTION
contracts that could be monitored at reasonable costs, and thus there were strong pressures and large benefits, even for customers, of reverting to a national monopoly in the form of Western Union. Our third paper is Alfonso Herranz-Lonca´n’s pathbreaking estimates of the growth of the Spanish infrastructure between 1844 and 1935. As in many other countries, capital formation in the years before 1895 was dominated by railroad investment. Subsequently, infrastructural saving flows shifted to other types of assets. His numbers allow us to chart the process of catch up and eventual convergence, seeing clearly how and when some of the foundations for late twentieth century economic expansion in Spain were established. We now shift to two papers based on microeconomic data. The first is the investigation by John James, Michael Palumbo, and Mark Thomas of late nineteenth century saving among working class families in the United States. They find that behavior of these households was not dissimilar from what we observe today among similarly situated families, casting doubt on the view that New Deal social insurance programs have been responsible for the continuing decline in the aggregate saving rate in the US during the twentieth century. Instead, that decline appears to have taken place largely among middle and upper income and professional households for whom the disincentive effects for saving of social security and medicare wealth must have been smaller. John Murray studies the operation of pioneering sickness insurance schemes in several European countries between 1895 and 1908. He describes a process of experimentation with voluntary and compulsory systems, documenting the degree to which the problem of adverse selection plagued the voluntary systems, leading to a shift in most instances to programs requiring mandatory enrollment. His study is based on data from systems in five countries: Austria, Belgium, France, Denmark, and Germany. We now move to the three papers from the Prices and Wages conference. Again, I owe a special debt to Gregory Clark for refereeing and editing these manuscripts. In the first of these papers, Sevket Pamuk studies trends in urban construction workers’ wages in the Eastern Mediterranean over almost a millennium. He finds that the Black Death doubled real wages, an increase that persisted until well into the sixteenth century, when population growth reversed the uptick. The second major influence on their wages has been the industrial revolution and the era of modern economic growth, although the largest effects were not experienced until the second half of the twentieth century. Pamuk also uses his time series to provide a comparative
Introduction
xi
perspective on urban wage trends elsewhere in Europe. Between the sixteenth and eighteenth century, real wages in Istanbul converged with those in Europe, with the exception of cities in northwestern Europe, which retained an advantage. In the nineteenth century, a more systematic gap opened between the Eastern Mediterranean and the rest of Europe, with partial closing in some but not all of the regions under study in the twentieth. A similarly motivated paper by Jean-Pascal Bassino and Debin Ma examines wages of Japanese unskilled workers between 1741 and 1913. Rather than continuity, they find a substantial increase in real wages comparing Meiji (post-1868) Japan with the preceding Tokugawa period. In their international comparisons, they conclude that real wages in Tokyo and Kyoto ran about a third of wages in London, although these series display levels and trends comparable to those observed in wage series for southern and central Europe. In the final paper, Marianne Ward and John Devereux present estimates of the relative income of the United Kingdom and Great Britain in comparison with that of the United States for 1831, 1839, 1849, 1859, and 1869. They find that for per capita income, the US was ahead of or equal to the UK for all five benchmarks, and ahead of the UK for all years save 1869. Examining output per worker, they find the US ahead in all instances. The paper includes a detailed discussion of their methodology and an appendix describing their sources. We continue to welcome innovative, provocative, well written, and carefully considered contributions to economic history in any area. As has been true in the past, we have more flexibility than other outlets in publishing work which is data rich, or runs to greater length. Potential authors may submit their work in hard copy or as attachments in an email addressed to the editor (
[email protected]). We are flexible on matters of style and formatting for the first round of submission. Alexander J. Field Series Editor
This page intentionally left blank
xii
A SOVIET QUASI-MARKET FOR INVENTIONS: JET PROPULSION, 1932–1946 Mark Harrison ABSTRACT This paper is about how a command system allocated resources under profound uncertainty. The command system was the Soviet economy, the period was Stalin’s dictatorship, and the resources were designated for military research & development. The context was formed by the limits of the existing aviation propulsion technology, the need to replace it with another, and uncertainty as to how to do so. We observe the formation of a quasi-market in which rival agents proposed projects and competed for funding to carry them out. We find rivalry and rent seeking, imperfectly regulated by principals. As rent seeking spread and uncertainty was reduced, the quasi-market was closed down and replaced by strict hierarchical allocation and monitoring. In theory, a dictator cannot commit to refrain from taxing the returns from today’s effort tomorrow; therefore, we expect agents in a command system to seek only short-term returns from quasi-market activity. Agents’ willingness to invest in the Soviet quasi-market for inventions is ascribed to a reputation mechanism that enforced long-run returns.
Research in Economic History, Volume 23, 1–59 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23001-X
1
2
MARK HARRISON
1. INTRODUCTION This paper is about an artificial market within a centralized command economy. It is generally thought that markets allocate resources more efficiently than hierarchies when information is dispersed and there is no easy way of making bureaucrats pay for mistakes, in other words, in most circumstances (Hayek, 1945). But when political and economic power is already concentrated in the hands of an authoritarian ruler there is also no easy way for society to promise him compensation in return for surrendering power (Acemoglu & Robinson, 2000). In this case excessive hierarchy is likely to persist and society will continue to lose from it. Beyond a point, however, the losses may detract from the income not just of society but also of the ruler. In that case, why should the ruler not gain by selective delegation to others of those decisions that he is least equipped to make efficiently? From the early 1930s, Soviet officials began to discuss whether it was possible to nest the informational and incentive advantages of markets within the hierarchical structures of the command economy behind closed doors (Davies, 1996, pp. 201–228). From the 1960s they pursued this quest openly under the banner of socialist economic ‘‘reforms.’’ These reforms aimed to devolve use rights over selected assets to selected agents who would be converted from hierarchical subordinates into financially independent stakeholders, and from rent-sharers to profit-makers. Given a clearer interest in the results of their own efforts, they would be motivated to achieve the leaders’ objectives at lower cost as a result. The reforms were ultimately a failure, however. The literature understands this as the result of a dynamic commitment weakness. From one point of view it is about taxation. Litwack (1991) describes the problem as the inability of Soviet planners to commit to a long-term scheme to tax the results of initiative; rather, the tax on tomorrow’s productivity is determined only after observing that of today. As a result, today’s effort goes into negotiating tomorrow’s taxes, not improving productivity. From another point of view, selective delegation is not sustainable in the long run. Williamson (1996, p. 17) considers the problem of ‘‘credible selective intervention’’ where intervention is the opposite of delegation: a ruler who is in a position to intervene selectively, i.e., at some particular point, cannot commit not to intervene at any point. If a hierarchical principal were able to commit to a policy of hands-off when decentralization yields the first-best outcome, and to intervene only when the market outcome was second best, then socialist planning could have the best of everything. But it cannot.
A Soviet Quasi-Market for Inventions
3
Up to this point the problem is framed in terms of control over streams of expected future taxes on and rewards to effort. Control over future income streams is also an aspect of property rights. For subordinate agents to become autonomous stakeholders, and to change from rent-sharing to profit-making, they must seize their stake and acquire irrevocable disposal rights over it. As long as their control over assets remained at the discretion of the ruler, they would rationally fear future expropriation. By the same token, the ruler, as long as he stood above the law, could not bind himself not to take back his assets and the profits earned on them in the future (Williamson, 1996, p. 113). He had specific production objectives for the sake of which he remained willing to override his own previous decision to delegate. The use rights that he delegated could only be temporary in principle, even if continually extended in practice. Consequently agents who were granted these temporary rights continued to seek to share the rents generated in the command system rather than make profits from production. Moreover, they could exploit these rights to extract a rent from the dictator up to the level of the cost saving that he aimed to achieve by decentralizing. At first sight this seems to imply that selective delegation can never work. Yet, sometimes it does. One area where it seems to have worked was in the invention phase of Soviet military research and development. The invention phase was marked by technological uncertainty and information bias: while nobody yet knew the answers, the specialists knew the problems better than the officials who funded them. Under these conditions inventive activity could not be regimented from above. Instead, rival inventors competed for funding and rewards (Holloway, 1982a, pp. 317–319). Decentralization seems to have worked in the sense that Soviet designers did keep pace with the global technological frontier as it expanded and did occasionally expand it themselves. The results included high-quality tanks and aircraft, atomic weapons and intercontinental missiles, satellites in space, and a remotely guided vehicle on the surface of the moon. In short, the Soviet command system could apparently solve some problems through selective delegation even though permanent disposal rights over assets were not devolved and future taxes remained undetermined. An empirical question then arises: what mechanism made selective decentralization work in this specific context? The present paper is a study of decentralization in the Soviet command system. I consider the process that led to a new aerospace technology based on jet propulsion. The period is the Stalin era, before decentralizing ‘‘reforms’’ became fashionable. I begin at the dawn of the process in 1932 and I conclude in 1946 when the invention phase had come to an end. Throughout this period there existed an artificial ‘‘quasi-market’’ for inventions. The
4
MARK HARRISON
problem of high-speed, high-altitude aviation was placed in this quasi-market at the beginning of the 1930s when uncertainties and information biases were at their greatest. The problem was removed from the quasi-market and returned to direct hierarchical regulation between 1944 and 1946 when the concept had been implemented and the uncertainties removed. I explore this topic on the basis of the recently declassified records of the defence industry held by the Russian State Economics Archive (RGAE), supplemented by those of the Red Army held by the Russian State Military Archive (RGVA), and a few records from the State Archive of the Russian Federation (GARF), all in Moscow. These are supplemented by available memoirs and a wide secondary literature that narrates the stories of lives and deeds. The official documents are marked, however, by qualities that the available narratives tend to lack: they are written without hindsight, and they are not selected to tell a story with an uplifting moral or to show someone in a good or bad light. Of course, they are selective in other ways. For present purposes the archives make it possible to do something that is new: to describe the Soviet market for inventions as an economic institution, and analyse the conditions for it to succeed. A note on Soviet terminology seems inescapable. Soviet ministries were called ‘‘people’s commissariats’’ until they were renamed ministries in 1946. I call them ministries throughout. I generally refer to the USSR Council of People’s Commissars (Ministers) as ‘‘the government,’’ and its chairman as the prime minister. Where necessary I use Russian official acronyms as follows: KB, OKB: NII: NKVD (MVD):
Design Bureau, Experimental Design Bureau Scientific Research Institute People’s Commissariat (Ministry) of Internal Affairs
It may also help to know of the major reorganizations of industrial ministries in the late 1930s: in December 1936, the ministry of heavy industry was split among several specialized branches of which one was the defence industry; 2 years later, in January 1939, the ministry of the defence industry was itself split into new ministries of the aircraft, ammunition, armament, and shipbuilding industries. The paper is organized as follows: Much of it is description. Section 2 sets out the technological problem that confronted the Soviet economy. Section 3 evaluates the scale and scope of the effort invested in solving it. Section 4 discusses the nature of quasi-markets and describes the structure of the quasi-market for inventions within the framework of a command system in
A Soviet Quasi-Market for Inventions
5
terms of principals and agents. Section 5 illustrates the role of the agents in forming the market, entering it, and securing initial funding. Section 6 looks at the interplay between principals and agents when projects came up for refinancing. How did principals intervene in the market? To what extent did agents’ competition limit or embody rent seeking? Section 7 shows that the quasi-market for inventions gave rise to an informal secondary market in long-lived research assets. As Section 8 briefly recounts, once the technological uncertainties had been removed, the authorities closed the quasimarket for inventions down. Section 9 returns to analysis of the problem posed in the introduction, that of credible selective delegation: how was successful invention motivated in this temporary market, and how could agents enforce their payoffs for success? A final section concludes.
2. THE PROBLEM OF JET PROPULSION In the interwar period, the airscrew propeller driven by a reciprocating piston engine reached its limits in terms of speed and altitude of aircraft performance (Grigor0 ev, 1994, p. 189). This prompted intensive efforts in several countries to develop new types of aeroengine based on a continuous thermal cycle. Some efforts succeeded and others failed. The success story was that of jet propulsion. The parallel story of the aviation steam turbine described by Harrison (2003a) is also of interest but ended in failure. Jet propulsion or reaction is the principle underlying both rocket motors and jet engines: action and reaction are equal and opposite. A rocket is a jet that does not need to breathe air. Small solid-fuelled rockets had been used for hundreds of years in many countries for display and signals. A survey prepared in the USSR ministry for ammunition in 1939 reminds us that rockets were used for the first time in European warfare in the Napoleonic wars as an incendiary siege weapon and later as ammunition against troops; by the middle of the nineteenth century most armies carried substantial stocks of rocket artillery.1 As a result of improvements in conventional artillery rockets were little used in World War I, but interest in them revived in the interwar period in connection with new fuels and artillery uses. Accompanying developments in aviation played a role because rockets could be used for both air-to-surface and surface-to-air artillery. In the 1930s, rockets also began to be used in a new way as aviation boosters. Since rocket fuel contained its own combustion ingredients, rockets were capable of performing at limitless altitudes but for limited duration. To create a primary aviation power plant therefore required a rocket motor of
6
MARK HARRISON
unprecedented size and complexity, using more powerful liquid fuels with pressurized or pumped fuel delivery; in turn these required substantial advances in material and fuel sciences and control systems. In the interwar years it was unclear whether a rocket aircraft would ever be practicable. The air-breathing jet engine was a more recent concept. The simplest version was a hollow tube or ramjet; when the tube was moving at high speed air would enter it at one end and be compressed, mixed with fuel, and burnt; the exhaust gases left the other end in a jet stream that drove the tube forward. The ramjet was no good as a primary power plant because something else had to accelerate it to high speed before ignition. Thus a primary power plant had to be able to suck in and compress air when stationary. In Frank Whittle’s patent of 1930 this need was met by a supercharger or compressor attached to a turbine driven by the exhaust gases; hence turbojet. In the 1930s, theorizing ran far ahead of practice. In theory jets were capable of higher speeds at higher altitudes than airscrew engines, though not of space flight, and for much longer duration than rockets. The theory of the gas turbine was also not a problem; it arose naturally from existing applications of steam turbines, principally in electricity generation and marine engineering (Voronkov, 1984, p. 115). The other major concepts that would power military and commercial aircraft for the next half century were also worked out at this time. For example, the thermodynamic efficiency of the turbojet was already understood to be poor at low speeds and altitudes, so designers were already thinking about a turboprop in which the gas turbine would drive both the compressor and reduction gears linked to an airscrew for slower long-range aircraft. Between the turbojet and turboprop lay the turbofan or turbojet engine with a bypass chamber, universally applied in modern jet airliners; a bypass engine was patented by the Soviet turbojet designer Arkhip Liul0 ka in April 1941 (Liul0 ka & Kuvshinnikov, 1981, p. 91). To put a gas turbine into a jet engine in practice, however, raised requirements on material and fuel sciences and control systems that were far above the level of the time. In the interwar years no one knew for sure that a turbojet could be made to work. If anything, it looked further from realization than the idea of a rocket aircraft. Expectations, positive and negative, were very important. In each country faith in the future of the turbojet was reinforced by the belief that rivals in other countries were making equal or greater efforts. These beliefs helped to make the turbojet a reality and in this sense (MacKenzie, 1996, p. 57) it was a self-fulfiling prophecy. But there were many failures en route; these were double-edged. In the long run the failures were positive: they promoted
A Soviet Quasi-Market for Inventions
7
learning and were a necessary cost of ultimate success. In the short run, however, the setbacks were just that; they set the process back and fuelled scepticism and conservatism as even committed believers understood. As an investigation into faulty parts for a Soviet prototype gas turbine reported in 1938: ‘‘to start the engine in the form produced by the factory – if it were possible – could only end in an accident and destroy the idea of building a gas turbine at its very inception.’’2 In Britain, Frank Whittle (1953, p. 78) feared that each mishap would destroy his sponsors’ faith in himself and his engine. Scepticism was an obstacle everywhere. According to Liul0 ka disbelief persisted in the Soviet Union until the end of 1943 (Liul0 ka & Kuvshinnikov, 1981, p. 89). In the United States in June 1940, the National Academy of Sciences announced that the turbojet was technically infeasible; a passage from its report reproduced by Golley (1987, facing page 114) bears Whittle’s handwritten comment: ‘‘Good thing I was too stupid to know this.’’ Actually, a Heinkel turbojet aircraft had already flown in secret in Germany the year before the American report. Even in Germany, as the inventor Hans von Ohain and the BMW engineer Peter Kappus recalled independently (Ermenc, 1990, pp. 8, 89), the scepticism evaporated only after the principle had been demonstrated in flight. But it may be that inventors also puffed up their reputations by stressing the resistance they had had to overcome. A natural response to the great gap between theory and practice was to compromise. To get around the fundamental difficulty of the turbocompressor, clever and inventive people all over Europe were exploring a variety of intermediate steps and hybrid solutions; the Soviet Union was perhaps unusual in the number of alternatives that were pursued simultaneously (Egorov, 1994, pp. 424–436; Serov, 1997; Gordon & Dexter, 1999). The simplest stopgap was to strap auxiliary ramjet boosters onto an otherwise conventional aircraft, but there was little practical gain because the boosters did not add much speed and caused large aerodynamic losses. A more advanced compromise was to use an auxiliary piston engine to supercharge a jet engine. This hybrid engine resulted in a primary power plant that was less demanding than the turbojet because the elements were all established technologies; Whittle (1953, p. 39) filed a patent for such an engine in the early 1930s. There was also interest in refining the existing steam turbine technology to drive a conventional airscrew more cheaply and reliably (Harrison, 2003a): if oil-fired steam turbines had replaced the reciprocating engine at sea, why not in the air as well? Finally, there was still rocketry: if the turbojet turned out to be impractical, then rocket motors must be developed far beyond their existing limits.
8
MARK HARRISON
Thinking along this line remained influential even after the turbojet had become a proven concept. For example, concluding a long report to Stalin’s deputy Malenkov in July 1944, minister of the aircraft industry Shakhurin compared the respective potentials of jet and rocket aviation as follows: Aircraft with motors of the liquid-fuelled [rocket] type have a bigger future in terms of achieving greater efficiency [a larger KPD, coefficient of useful action] and a sharp increase in the tactical flight characteristics of jet-propelled devices. The power of such a motor does not depend on altitude and it can operate (even better) in airless space. However, extremely high fuel expenditures at the present stage of their development (cf. the Me[sserschmitt]-163) sharply limit the great potential of this type of aircraft. This is why existing jet-propelled aircraft have big engines of the air[-jet] type.3
From a mission standpoint, what were the appropriate responses to the extent of technological uncertainty? The best chances of progress would result from an open-ended commitment to advance on many fronts at the same time. Many problems demanded simultaneous technological solutions. Many applications would not be detected without free-ranging exploration of new technologies. The state needed to fund many projects, accepting a high probability of failure in any one of them, in order to ensure that at least some successful projects would be included. As Joel Mokyr (1990, pp. 176–177) has taught us, many failures could be expected as part of the cost of success. As far as the scale of effort is concerned, no government would accept a completely open financial commitment, but there were different degrees of open-handedness across countries. For each country the feasibility of a given commitment varied with economic size, development level, and mobilization capacity. The interwar British, German, and Soviet economies were of about the same size in gross domestic product (GDP). Britain and Germany were more developed in science and industry than the USSR; the Soviet economy could claim an advantage, however, in its superior mobilization capacity (Harrison, 1998).
3. SCALE AND SCOPE This section gives a brief account of the scale and scope of jet propulsion research for aviation in the Soviet Union from 1932 to 1946. Table 1 lists 18 major projects over the period; the research in rocketry, limited to aviationrelated activities, excludes numerous artillery projects some of which were highly successful. It is compiled mainly from items reported in the plans, reports, and memoranda of the ministries of defence, internal affairs, and
A Soviet Quasi-Market for Inventions
9
the heavy, defence ammunition, and aircraft industries; these are listed more completely in the appendix. The ‘‘major project’’ as a unit of measure is convenient but simplified or fuzzy in some dimensions. First, research strategies varied; for example Glushko, Merkulov, and Uvarov were often to be found engaged in multiple designs for power plants of varying size and capacity based on similar principles at the same time, whereas Liul0 ka committed everything to a single design that evolved through time. To make sense of the data, I have counted the work of each as a single major project, and this has imposed some aggregation on the ‘‘minor’’ ones. (I have made an exception for Glushko whose collaboration with Sergei Korolev on a rocket aircraft in 1936 and 1937 was clearly different from his other projects in scale and significance.) Second, the official documents do not capture the informal or peripheral involvement of designers before they moved to the centre of the field. The documents must also fail to reflect the disruption or continuity in designers’ work as they moved from one organization to another. It is not clear, for example, what Uvarov did in 1940 or whether, when he left VTI in 1939 and reappeared at TsIAM in 1941, he brought with him his assistants and equipment or only his personal intellectual capital. The great advantage of Table 1 is that the data underlying it were created without foresight. In contrast, the memoirs and biographies of the designers such as Liul0 ka and Uvarov, who were eventually successful, naturally pay scant attention to the projects of their lesser rivals. This makes the table substantially more complete than previous narrative accounts. Table 1 shows six major projects in aviation rocketry and 12 in jet aviation over the period, but this does not give a clear impression of relative importance since some projects were long-lived, while others appeared and disappeared within a year. Fig. 1 measures the overall investment in cumulative ‘‘major project-years.’’ It shows that until 1937 research in rocketry and jet engines advanced more or less in step; after that, the balance shifted away from rocketry and by 1946 jet engine development had accomplished almost two-thirds of the 76 project years accumulated in total. Also of interest is the turnover of projects, especially the termination of those regarded as unsuccessful. Fig. 2 illustrates this for the aviation field as a whole. It shows clearly the buildup of research and design activity in the second half of the 1930s. It shows that project closures were concentrated in 1937–1939, and in the first two and a half years of the war. Closer inspection of Table 1 reveals that there was a complete break in rocketry development at the end of 1938, when all projects then current were terminated. The
10
Table 1.
Major Soviet R&D Jet Propulsion Projects for Aviation, 1932–1946.
1932
1933
1934
GDL
-
RNII
1935 1936 1937
1938
1939
1940
1941(1)
1941(2)
1942
1943
1944
1945
1946
KB Z-16
-
-
-
-
GIRT, OKB-293
-
NII-1
NII-1
-
Rocket motors Smaller Glushko rocket motors and aviation boosters Large Glushko rocket motor leading to RP-218 aircraft
- NII-3
-
RNII NII-3
KB-7 NII-3, OSK Z-1
-
NII-3
-
NII-3
VTI MAI
GIRT
MARK HARRISON
KB-7 rocket motor Dushkin, Isaev rocket motors leading to RP-318 and BI fighter NII-3 rocket booster Dushkin rocket motors Jet Engines VTI gas turbine MAI gas turbine
-
Pobedonostsev, Merkulov ramjet
NII-3 hybrid jet Bas-Dubov, Zaslavskii ramjet Abramovich hybrid jet Fadeev, Kholshchevnikov hybrid jet Tolstov hybrid jet
-
-
-
TsIAM
-
-
-
OKB-293
-
-
GIRT
Z-28
-
GIRT
TsIAM
-
-
-
OKB-293, TsIAM Z-84
NII-1
-
-
TsIAM
-
-
TsIAM
-
-
KB-7
GIRD
-
RNII
-
- NII-3
-
TsKTI, SKB-1 Z-18 NII-3, NII-3, OSK TsKTI, Z-1 OSK Z-1, Z-18 NII-3 -
TsAGI
A Soviet Quasi-Market for Inventions
VTI
Uvarov gas turbines leading to turboprop engine KB-7 ramjet Liul0 ka turbojet
Source: The appendix, supplemented by Egorov (1994, pp. 424–436). The documentation supporting the appendix comprises plans, reports, and memoranda of the ministries of defence, internal affairs, heavy industry, the defence industry, ammunition, and the aircraft industry.
11
12
MARK HARRISON 50 45
Jet Engines Rocket Motors
Major Project Years
40 35 30 25 20 15 10 5 0 1931
Fig. 1.
1936
1941
1946
Soviet R&D in Jet Propulsion, 1932–1946: Cumulative Investment in Major Project Years. Source: Table 1.
figure concludes with the year 1946 when, with activity now at a high level, no projects were closed down and no new ones were started. Table 1 showed that some 15 distinct research establishments were involved in developing jet propulsion for aviation; these are detailed in Table 2. The true total of establishments listed in the table is less than those that appear in the table because there were many reorganizations and changes of name, usually driven by a desire to break with the errors of the past. Behind these are a strife-torn story of the fall and rise of the idea of jet-propelled aviation. The central organization in this story was the Jet Propulsion Research Institute (RNII) formed in 1934. Its turbulent history has been described recently by Harrison (2000, pp. 127–130) and Siddiqi (2000, pp. 1–14). The low point fell between 1937 and the second half of 1941. In 1937, discredited by association with ‘‘enemies of the people’’ charged with wasting state funds on useless dreams of rocket-powered flight, RNII was converted into NII-3, an adjunct of the ammunition industry, and ordered to concentrate on designing rocket shells and mortars. Interest in rocket aviation was revived in August 1941 and in jet engines in July 1942. NII-3 was converted at first to a State Institute for Jet Propulsion Technology (GIRT) reporting directly to the
13
A Soviet Quasi-Market for Inventions 8
Starting
Jet Propulsion: Major Projects
7
In progress at start
6
Finishing
5 4 3 2 1 0 -1 -2
Fig. 2.
1946
1945
1944
1943
1942
1941(2)
1941(1)
1940
1939
1938
1937
1936
1935
1934
1933
1932
-3
Soviet R&D in Jet Propulsion, 1932–1946: Major Projects. Source: Table 1.
central government. It was fully rehabilitated in the spring of 1944 as the Research Institute for Jet-Propelled Aviation (NIIRA) and then Research Institute no. 1 (NII-1) of the aviation industry. Its many transformations are shown for reference in Fig. 3. The organizations involved were of modest size. Table 3 shows that RNII-NII-3 had 400 or so staff in the mid-1930s of whom around one quarter could be classed as specialist ‘‘engineering and technical employees.’’ By the end of the decade this number had risen to more than 800. In the same period, the RNII budget trebled, although not all of this was real growth; price and wage inflation was especially rapid between 1937 and 1940 (Bergson, 1961, pp. 367–368, 422). In comparison with other establishments of a similar profile, RNII-NII-3 was medium-sized; the other main rocketry establishment, KB-7 was much smaller. This is shown in Table 4, which compares them with other research outfits of the ammunition industry in 1938. In considering these figures, it is important to bear in mind that most of the work done in RNII-NII-3 and KB-7 was concerned with rocket artillery, not aviation. Aviation projects accounted for two-fifths of the value of research and experimentation planned by RNII in 1937.4 By 1940 their
Acronym GDL
a
a
GIRD
GIRTa
KB Z-16 KB-7a
MAI a
NII-1
NII-3a NIIRAa
R&D Organization
Fundholder
Gas Dynamics Laboratory Jet Propulsion Study Group
Red Army Administration for Military Inventions Society for Cooperation in Air and Chemical Defence; transferred to Red Army Administration for Military Inventions in 1933 USSR Council of People’s Commissars
State Institute for Jet Propulsion Technology Factory no. 16 Design Bureau Design Bureau no. 7
Moscow Aviation Institute Research Institute no. 1 Research Institute no. 3 Research Institute for Jet-Propelled Aviation Experimental Design Bureau no. 293
Period Involved
Location
Notes
A
To 1934
Leningrad
A
To 1934
Moscow
Merged with GIRD into RNII in 1934 Merged with GDL into RNII in 1934
D
1942–1944
Moscow
NKVD Fourth Special Department
D
From 1942
Kazan0
Red Army Administration for Military Inventions, then Artillery Administration; transferred to Defence (later Ammunition) Industry in 1938 Heavy (later Defence, later Aircraft) Industryb Aircraft Industry
A
1938 and 1939
Moscow
I
1934
Moscow
I
From 1944
Moscow
Formerly NIIRA
Formerly NII-3
Spin-off from RNII in 1935
Defence (later Ammunition) Industryb Aircraft Industry
I
1937–1942
Moscow
Formerly RNII
I
1944
Moscow
Formerly NII-3
Aircraft Industry
I
1942–1944
Khimki, Moscow district; evacuated temporarily to Bilimbai, Sverdlovsk district, October 1941– January 1942
Merged into NIIRA-NII-1 in 1944
MARK HARRISON
OKB-293a
Soviet Jet Propulsion R&D Organizations and their Fundholders, 1932–1946.
14
Table 2.
Factory no. 1 Department of Special-Purpose Designs Jet-Propulsion Research Institute Special-Purpose Design Bureau Central AeroHydrodynamic Institute
Aircraft Industry
I
1939–1942
Moscow
Heavy (later Defence, later Aircraft) Industryb Defence (later Aircraft) Industryb
I
1934–1937
Moscow
I
1940–1941
Leningrad
Aircraft Industry
I
1943
Central Institute for Aeroengine Building Central Boiler and Turbine Institute
Aircraft Industry
I
1941 and from 1943
Stakhanovo (later Zhukovskii), Moscow district, evacuated temporarily to Kazan0 and Novosibirsk, 1941–1942 Moscow
Electricity Generation Industry
I
1939
VTI
Dzerzhinskii AllUnion ThermalTechnical Institute
Electricity Generation Industry
I
1932 and 1936– 1939
Z-18
Factory no. 18
Defence (later Aircraft) Industryb
I
1939
Z-28
Factory no. 28
Defence (later Aircraft) Industryb
I
1941
RNIIa SKB-1 TsAGI
TsIAM
TsKTI
15
Leningrad, with a subsidiary in Podol0 sk, Moscow district Moscow, evacuated temporarily to Keremovo, 1941–1943 Voronezh, evacuated to Kuibyshev in October 1941 Moscow, evacuated to Sverdlovsk in October 1941
Formed by GDL and GIRD in 1934
A Soviet Quasi-Market for Inventions
OSK Z-1
Acronym Z-84
R&D Organization Factory no. 84
Fundholder Defence (later Aircraft) Industry
b
I
16
Table 2. (Continued ) Period Involved
Location
1943
Khimki, near Moscow, evacuated to Tashkent in October 1941
Notes
Note: A (Army); D (Dictator); I (Industry). Source: Table 1 and Fig. 1. a See Fig. 1. b On December 8, 1936 the ministry of heavy industry was divided into a number of specialized branches of which one was the defence industry; on January 11, 1939 the ministry of the defence industry was divided into new ministries of the aircraft, ammunition, armament, and shipbuilding industries.
MARK HARRISON
17
A Soviet Quasi-Market for Inventions
Table 3.
RNII-NII-3, 1935–1941: Personnel by Employment Status and Gross Value of Output. 1935 1936 1937 plan 1938 1939
1940 1941 plan
Persons employed, annual average Engineering and technical employees Manual employees Nonmanual employees Accounting and clerical Production and planning Junior service personnel Total Gross value of output (thousand rubles)
— — 76 — — 37
102 196 84 — — 64
118 295 88 — — 63
— — — — — —
— — — — — —
— — — — — —
215 385 — 125 79 32
403
446
476
514
799
—
836
— 3,377
4,482
6,111 11,434 11,233
—
Source: Employment: 1935 from RGAE, 8162/1/16, 16, and 1936–1937 from ibid., 4 (no date but about February 1937); 1938–1939 from RGAE, 8162/1/240, 32 (January 13, 1940); 1941 from RGAE, 8162/1/449, 144 (November 18, 1941). Gross value of output: for 1936 and 1937 plan see RGAE, 8162/1/16, 2–3 (February 28, 1937), for 1938–1939 RGAE, 8162/1/240, 32 (January 13, 1940), and for 1940 RGAE, 8162/1/449, 3 (January 14, 1941). The figure for 1940 that had been planned and approved by KO was slightly higher at 11,725 thousand rubles. However, towards the end of that year an investigation disclosed that the responsible fourth chief administration of the ministry of the aircraft industry had illegally planned a much higher figure of 13,162 thousand rubles; see RGAE, 7516/1/692, 3 (November 21, 1940).
nominal value had fallen absolutely, while their share in the much larger NII-3 budget was now only 6%.5 These research expenditures were a very small fraction of overall defence procurement. In the late 1930s, they ran at less than a million rubles a year. Equipment orders for the army and navy in 1937 were 5.7 billion rubles (Davies & Harrison, 1997), and 14.5 billion rubles for the army alone in 1940 (Harrison, 1996, p. 281). The Soviet outlays appear trivial in comparison with the resources that Germany devoted to the development of V-weapons. A postwar American estimate puts the total development costs of the jet-powered V-1 cruise missile to Germany at approximately $200 million in wartime US prices, and those of the V-2 rocket at 10 times that amount, about the same as the $2 billion cost of the Manhattan Project (Ordway & Sharpe, 1979, pp. 242, 253). Given that one prewar ruble was worth at most 35 US wartime cents (Harrison, 1996, 275), the resources invested in Soviet aviation jet propulsion in the 1930s and 1940s can hardly have totalled more than $10 million.
18
MARK HARRISON
GIRD pri TsS Osoaviakhima: Group for the Study of Jet-Propelled Motion of the Central Council of the Society for Cooperation in Air and Chemical Defence
GDL UVI RKKA: Gas Dynamics Laboratory of the Red Army Administration for Military Inventions (Leningrad)
October 1933
GIRD UVI RKKA: Group for the Study of Jet-Propelled Motion of the Red Army Administration for Military Inventions
January 1934
RNII NKTP: Jet Propulsion Research Institute of the People’s commissariat for Heavy Industry
August 1935 January 1937
NII-3 NKOP: Research Institute no. 3 of the People’s Commissariat for the Defence Industry
December 1938
NII-3 NKB: Research Institute no. 3 of the People’s Commissariat for Ammunition
January 1942
KB-7 AU RKKA: Design Bureau no. 7 of the Red Army Artillery Administration
January 1938
KB-7 NKOP: Design Bureau no. 7 of the People’s Commissariat for the Defence Industry
December 1938
KB-7 NKB: Design Bureau no. 7 of the People’s Commissariat for Ammunition
GIRT pri SNK SSSR: State Institute for Jet Propulsion Technology of the USSR Council of People’s Commissars
February 1944
NIIRA NKAP: Research Institute for Jet– Propelled Aviation of the People’s Commissariat for the Aircraft Industry
May 1944
NII-1 NKAP: Research Institute No. 1 of the People’s Commissariat for the Aircraft Industry
OKB-293 NKAP: Experimental Design Bureau of Factory no. 293 of the People’s Commissariat for the Aircraft Industry, Khimki, Moscow oblast’
A Soviet Quasi-Market for Inventions
19
4. DICTATOR, ARMY, AND INDUSTRY IN A QUASI-MARKET In Section 4, I discuss the differences between ‘‘real’’ and quasi-markets and provide a simplified description of the structure of the market for inventions. We think of real markets as formed by buyers and sellers who enter the market independently, motivated by their own self-interest. In real markets, prices are set by interpersonal negotiation or impersonal bidding, or are preset by one side in the presence of market power. At least one equilibrium is possible and the interaction of supply and demand generally leads to it. Contract disputes are resolved by custom or law. In the outcome, the market steers resources in the general direction of their most profitable use. Quasi-markets, in contrast, are created by the state to allow its own agents to engage in decentralized transactions with each other. The agents enter the quasi-markets because they are told to. They are not supposed to behave in an independently self-interested way but to follow contingent rules. If they find themselves in dispute, the principal determines whether or not to intervene and which side to uphold. Prices and incentives in quasi-markets are formed by the principal’s decision; the process is not equilibrating and adjustment typically involves ‘‘false’’ or cross-trading, described by Morishima (1984, p. 15). The
Fig. 3. The Evolutionary Path of RNII, 1932–1944. Sources: For details see Siddiqi (2000, pp. 1–18). In addition:GIRD, originally sponsored by Osoaviakhim, was taken over by the Red Army administration for military inventions in October 1933 before being merged with GDL, renamed RNII, and transferred to the ministry of heavy industry (RGVA 4/14/1171, 33: memorandum dated 23 January 1934). Siddiqi states that NII-3 was handed over to the ministry of the ammunition industry in November 1937, but this ministry was only created on the dissolution of the ministry of defence industry on 11 January 1939. Various documents indicate that the ammunition industry also acquired KB-7 from the Red Army’s artillery administration at the beginning of 1938. A memorandum from deputy defence minister Fed0 ko to prime minister Molotov dated 15 February 1938 refers to ‘‘the former KB no. 7 of the AU RKKA [Red Army artillery administration], transferred to NKOP [people’s commissariat of the defence industry]’’ (RGVA, 4/14/1925, 22), and KB-7 is listed among the establishments of the thirteenth chief administration of the ministry of the ammunition industry in its report of work for the year 1938 (RGAE, 8162/1/89, 101). KB-7 was apparently dissolved in 1939. GIRT is described as ‘‘pri SNK SSSR [attached to the USSR Council of People’s Commissars]’’ in its deed of transfer to the ministry of the aircraft industry, not dated but in 1944 (RGAE, 8044/1/1182, 11-16).
20
MARK HARRISON
Table 4. Research Institutes and Design Bureaux of the Thirteenth Chief Administration of the Ministry of Defence Industry, 1938. Budget (thousand rubles)
Planned Research Topics
Scientific Workers Employed
NII-24 Leningrad filial KB-47 NII-3 KB-7 KB-31
12,764 11,052 8,006 5,667 1,200 700
178 81 94 39 9 6
60 55 55 44 13 19
Total
39,389
407
246
Source: RGAE, 8162/1/299, 9 (no date but 1938). The thirteenth chief administration of the ministry of defence industry was the future ministry of the ammunition industry.
principal usually tries to calibrate incentives in advance so when his agents allocate resources in detail the results will conform to broad limits already set out in centralized plans. The idea of quasi-markets is widely applied to describe decentralized allocation within large private and public-sector organizations, for example, the British welfare state since Margaret Thatcher (Le Grand, 1991; Le Grand & Bartlett, 1993; Bartlett, Roberts, & Le Grand, 1998). Its conceptual origins, however, lie with Ludwig von (Mises, 1949/1998, pp. 701–706), who employed the term to describe the market-like rules for decentralized allocation that contemporary socialists such as Oskar Lange wanted to embed within a planned economy. Mises regarded the idea as inherently unworkable; he argued that, since the agents would have no property of their own to lose, quasimarkets would be dominated by the ‘‘audacity, carelessness, and unreasonable optimism’’ of ‘‘the least scrupulous visionaries or scoundrels.’’ We shall find that, while he had an element of prophetic truth on his side, the reality of the quasi-market for inventions was much more complex and interesting than this would suggest. In particular, the quasi-market began to blur into a real market at two points: the sellers played a significant role in influencing the principal to create it; and enforcement relied significantly on a reputation mechanism that was independent of the principal. The structure of the market was roughly as follows: There were four main categories of players and Fig. 4 illustrates the quadrilateral relationships among them. Stalin, the dictator, was personally represented by central government. On Stalin’s behalf, high-level government agencies established
21
A Soviet Quasi-Market for Inventions
Dictator : Stalin, his government and security agents (NKVD-MVD)
Funding Agency, usually Army (Defence Ministry)
Fundholder, usually a branch of Industry (e.g. Defence Industry, Aircraft Industry)
R&D O rganization , e.g. RNII, NII-3
Key Funder contracts with fundholder. Funder substitutes for fundholder. Dictator substitutes for fundholder. Fig. 4.
Players in the Quasi-Market for Inventions.
the framework within which the other players operated by issuing strategic directives from time to time. Formally, these were usually the government subcommittee’s responsibility for defence matters or, in wartime, the war cabinet. In reality, regardless of formal authority, Stalin decided many of these things personally in consultation with a varying circle of members of the party politburo, usually after receiving representations from other stakeholders (Khlevniuk, Kvashonkin, Kosheleva, & Rogovaia, 1995; Khlevniuk, 1996). Examples of the major directives affecting R&D for jet propulsion during the period under review are shown in Table 5. The actors at the next level down were the defence agencies to whom Stalin delegated the power to fund research, and the specialized fundholders or suppliers of R&D services. The defence ministry was the prospective purchaser of jet propulsion technology and acted as the funding agency. The funder commissioned R&D services from the fundholder who was paid to supply them. The funder’s contract could implement a government decree
22
MARK HARRISON
Table 5.
Major Directives of the Soviet Government Concerning Jet Propulsion, 1932–1944.
Date
Issuing Authority
Decision
July 1932
Expand research on jet propulsion Establish RNII Develop the Liul0 ka turbojet
November 1942
Government Defence Commission Council of Labour and Defence Government Defence Committee State Defence Committee (the war cabinet) State Defence Committee
February 1944
State Defence Committee
May 1944
State Defence Committee
September 1933 July 1940 August 1941
Develop the BI (BerezniakIsaev) rocket fighter Produce the 302 (Kostikov) rocket fighter Reorganize NII-1 as a research institute of the aviation industry Develop a number of rocket and jet aircraft and engines
Source: Danilov (1981, p. 71).
using centralized funds earmarked from the USSR state budget. Alternatively, as a budgetary institution the defence ministry could enter into decentralized contracts with industrial institutes and design bureaus for R&D services on its own initiative. The fundholder was the legal owner of the R&D organization. The main fundholders supplying jet propulsion research are listed in Table 2. If the list appears complex at first sight, the detail can be simplified by grasping the three main types of fundholder, indexed in the table as A for Army, D for Dictator, or I for Industry. The ‘‘normal’’ arrangement of the quasi-market for inventions was that Industry, i.e., the ministry of heavy industry and its successors responsible for the various branches of the defence industry, in particular the aircraft and ammunition industries, was the fundholder. But other industrial interests could also take on sideline responsibilities, for example, the electricity generation industry. In varying circumstances the Army and the Dictator intervened. As the principal funder, the Red Army made various attempts to bypass the quasi-market for inventions and substitute itself for industry as the fundholder by establishing its own in-house jet propulsion research and development facilities, for example, KB-7. In Fig. 3 this is shown by the fine dashed arrow. Second, Stalin’s NKVD sometimes acted directly for him by substituting itself for the fundholder, seizing R&D personnel and assets and managing them on a prison basis in a sharaga or sharashka (Albrecht, 1993,
23
A Soviet Quasi-Market for Inventions
pp. 133–135; Starkov, 2000, pp. 255–260); this is indicated by the heavy dashed arrow. The fundholder of factory no. 16 in Kazan0 , for example, was the ministry for the aircraft industry, but its aeroengine design bureau was a sharaga staffed by prisoners and run by the NKVD fourth special department. Normally, however, funding departments and fundholders formulated independent operational plans that were then coordinated through a contracting process. The most important planning horizon was annual. The Red Army had an annual plan for the development of military inventions most of which it contracted out to other organizations through the quasimarket for inventions. Industrial ministries also had their own R&D plans, for example, the annual plan for aeroengine research and experimentation to be carried out by the institutes and bureaus of the aviation industry, part of which was made up by contracts accepted from the Red Army. But some R&D was also financed by the industry. The point is illustrated by the sources of funding of a design organization such as NII-3. Table 6 shows that NII-3 planned to undertake 11,725 thousand rubles of expenditure on research in 1940. This sum was to come from three sources, one part from the state budget, another from decentralized contracts with outside funding departments, mainly the Red Army, and the rest from the NII-3 fundholder, the responsible chief administration in the ministry of the ammunition industry. These arrangements imposed the following structure on competition among designers. There were many designers and many design organizations. Since R&D projects had to fall under one or other design organization, the design organization was the main vehicle for this competition. Designers competed for funding from a limited number of sources. In principle the defence industry, later the aviation industry, was the monopoly fundholder but in practice its monopoly was limited and threatened by other parties: industries with sideline interests in potential diversification, military men interested in the scope for
Table 6. Source of Funding
Planned Funding of NII-3, 1940. Number of Projects
Thousand Rubles
State budget Contracts Chief administration
18 15 12
5,790 3,440 2,495
Total
45
11,725
Source: RGAE, 8162/1/449, 3 (January 14, 1941).
24
MARK HARRISON
vertical integration, and the dictator who could revoke the delegation of his powers to industry at any time and impose direct control under the NKVD.
5. ACTION AND REACTION The quasi-market for inventions was not atomistic or impersonal. Rather, it was driven by the designers. David Holloway (1982a, p. 288) has suggested that the balance between supply side or ‘‘discovery-push’’ factors and ‘‘demand-pull’’ factors determined how major innovations were diffused through the Soviet defence sector. The stereotype of a command economy might lead one to expect that the active side of the market was the demand side in the sense that the funding principal issued compulsory contracts on the basis of a high-level plan, and the fundholders then complied with the contracts. The difficulty with this stereotype is that, given the technological uncertainty surrounding the future of aviation in the 1930s, the principal did not know what contracts to offer. The specialist agents on the supply side knew the answers better than the principal, and the command system had to adapt to this reality. Thus the active side of the market was the supply side and decision-making on the demand side was mainly reactive. In Section 5, I describe briefly the supply side activity, which took the form of proposals for funding from three groups of actors: established designers, backyard inventors, and foreign specialists. I note separately the uses that these proposals made of information from the foreign press and foreign commercial information. The reactive character of decision-making on the demand side is shown by the character of funding decisions: reports and resolutions that consolidated or cancelled existing rival projects greatly outweighed the number of decisions that authorized new ones. Designers’ control of the initiative resulted in a tendency for projects to proliferate that the funding principals found difficult to control. We will look more closely at this in Section 5.5. But the scope of the designers’ initiative went beyond this; it was not just that they held the initiative within the market, they also helped to create it. In this sense, the distinction between real and quasi-markets was not completely watertight. 5.1. Established Designers How did projects win a place in the plan? There were a variety of routes, but in each case the initiative lay with the designer. This was not a process whereby all-seeing and all-knowing planners identified needs from above,
A Soviet Quasi-Market for Inventions
25
sought out designers, and put them together with resources to meet the needs identified. Rather, proposals came first from below. As minister of the aircraft industry Shakhurin explained to deputy prime minister Voznesenskii in February 1941, ‘‘Work on the creation of jet propulsion engines at home in the USSR [y] began on the initiative of a few engineers taking the form of inventors’ proposals.’’6 This was the case in other defence projects too. For example, the first movers were the designers in both the failed attempt to build an aviation power plant around a steam turbine (Harrison, 2003a) and the successful project to build a Soviet atomic bomb (Holloway, 1994, pp. 72–95). Liul0 ka, father of the Soviet turbojet, began his work in 1936 on the back of a steam turbine project at the Khar0 kov Aviation Institute (Egorov, 1994; Sultanov, 1998); this experience soon convinced him that steam power had no future in aviation. The following account is based on Berne and Perov (1998, pp. 78–81). Liul0 ka wanted to develop work on a turbojet but his path was strewn with obstacles. His superiors in the Khar0 kov institute refused to back his work directly and referred him to Moscow. There he met Uvarov, who helped him get some funding from the aviation industry. But money opened few doors without adequate technical support; back in Khar0 kov his project languished. In 1938, Liul0 ka contrived ‘‘with great difficulty’’ to meet minister for the defence industry Mikhail Kaganovich who was impressed enough to convene a night-time meeting with his deputies; the outcome was to offer Liul0 ka the facilities of SKB-1 at the Leningrad Kirov factory. Back in Khar0 kov, Liul0 ka began to build a new engine with an axial compressor in the expectation of Air Force sponsorship. When this failed to materialize he wrote to prime minister Viacheslav Molotov in March 1939. At the end of this year he finally secured proper funding from the aviation industry and a proper base at the Central Boiler and Turbine Institute (TsKTI) of the electrical industry (see Table 1). During 1940 and 1941, Liul0 ka designed an engine and experimented with its turbine, compressor, and combustion chamber. However, within a few days of the outbreak of the war this work was mothballed in favour of an alternative plan to build a liquid-fuelled rocket fighter, the BI, pursued to a catastrophic conclusion in early 1943. Designers worked to secure ministerial approval and the funding that followed. If refused at one level, they appealed to the next. If necessary they began work without waiting for authorization; they illegally diverted resources of their own design organizations that had been allocated to other uses and then used the preliminary results to support subsequent attempts to gain official backing. The principle is illustrated by an anecdote (Perakh, 1998, y9.1): in the 1930s a group of young scientists who wanted to embark
26
MARK HARRISON
on research in atomic physics approached A.F. Ioffe, director of the famous Fiztekh, the Institute of Physics and Technology in Leningrad. Ioffe saw the potential of their proposal but was under pressure from above to give more resources to applied research instead. He realized it might be difficult to justify the proposal to the party authorities, and resolved to go ahead by means of a ruse. He gave laboratory space to the atomic physics project on an unofficial basis, and posted a sign on the door: ‘‘Stockroom.’’ As he expected, at the next inspection the party officials walked straight past without curiosity. The project was safe until the time came for Stalin to recognize the importance of atomic physics. Such behaviour caused projects to proliferate in an uncontrolled way. The consequences were outlined by NII-3 director A.G. Kostikov at an internal meeting in May 1942 when the strain on resources was at its most intense:7 As an example of how we are forced to diffuse the attention of our cadres I will take the first research department. There are 26 [research] topics for 10 engineers. Some of these topics are incidental to our institute and do not match its profile or specialization. These topics arose because there were people to put them forward and instead of passing them on to those organizations for whom such topics were more appropriate we engaged in them ourselves. [y] It’s characteristic of such topics that working on them involves unnecessary investigations since [we have] no corresponding experience. Often what is done is done many times, and all because we took on what was not our business, because we have neither experience nor cadres to work on items that don’t match the profile of our institute.
Successful proposals required investments in lobbying. Such investments could bring the designer not only success with an individual proposal but also a privileged long-term relationship with the government officials responsible for funding. To win support for their projects and adoption of their designs, designers had to be ‘‘heterogeneous engineers’’ capable of reshaping organizational as well as technological constraints (MacKenzie, 1996, p. 13). To create a demand for new designs they had to build coalitions with soldiers or industrialists to overcome interests vested in markets for products that already existed (Holloway, 1982a, p. 292). In particular, they had to overcome the preference of industry for the undisturbed mass production of weapons in long serial runs, which was often at odds with radical product innovation and the risks and requirements of continual upheaval in production (Berliner, 1976, pp. 534–538; Albrecht, 1993, pp. 195–197, 207–208). This could make it difficult to establish where the initiative lay and blurred the whole distinction between discovery-push and demand-pull. It also provides an exception to the rule that Soviet producers did not need to hire marketing agents, only supply facilitators or tolkachi.
A Soviet Quasi-Market for Inventions
27
The aircraft designer Sukhoi, for example, is said to have won success in having his designs adopted only after he took on a partner, E.A. Ivanov, who had the political and bureaucratic skills to push his product through the military and party-state apparatus (Ozerov, 1973, p. 53). Almquist (1990, pp. 70–73) has described the political ‘‘connectedness’’ of successful postwar Soviet designers. The German turbojet pioneer Hans von Ohain prospered on account of the alliance he forged with the aircraft manufacturer Otto Heinkel; according to Kappus (Ermenc, 1990, p. 91), the support of Heinkel was critical to Ohain’s success with the turbojet in Germany and the lack of similar support explains why Whittle took twice as long in the UK. Finally, Siddiqi (2000, p. 7) has described the rocket pioneers’ alliance with Marshal Mikhail Tukhachevskii. Tukhachevskii, Red Army chief of armament from 1931 to 1936, was the most important patron of jet propulsion in the Soviet Union between the wars.8 At the outset he took the Leningrad Gas Dynamics Laboratory (GDL), founded in 1929, under the aegis of the Red Army’s administration for military inventions to develop solid-fuelled rocket ammunition. In Moscow in 1931, a voluntary society of rocket scientists, the Jet Propulsion Study Group (GIRD), began to promote the cause of space exploration based on liquid-fuelled rocketry.9 The group was led by Sergei Korolev, the future chief designer of ballistic missiles and space launch vehicles, and was sponsored by the civil defence organization Osoaviakhim. In September 1933, Tukhachevskii sponsored a merger of GDL and GIRD in a new establishment subordinated directly to him, the RNII.10 He seems to have hoped to monopolize the development of jet propulsion as both funder and fundholder. He was frustrated by a decree of the Council for Labor and Defence which almost immediately transferred the new establishment to Sergo Ordzhonkidze’s ministry of heavy industry.11 This followed the precedent of arrangements for TsAGI and TsIAM, recently established to focus research in aircraft and aeroengine design, respectively. Although no longer the fundholder, Tukhachevskii personally, and the Red Army as funder, retained close involvement with RNII, contracting for much of its R&D output. RNII was an unhappy marriage, and divisions soon emerged between the weapons specialists of GDL and the space enthusiasts of GIRD. The new director I.T. Kleimenov, formerly head of GDL, curtailed work on liquid-fuelled rockets on the grounds of its low expected military utility, sidelining Korolev and the other GIRDers. The result was a huge row that embroiled RNII with the local party organization and pitched Korolev and Tukhachevskii against Kleimenov and Ordzhonikidze.12 In 1935, Tukhachevskii exploited these divisions to recruit some former GIRDers led
28
MARK HARRISON
by Korneev, a dissident engineer whom Kleimenov had sacked, to set up KB-7 as a Red Army design bureau for liquid-fuelled rocketry (Siddiqi, 2000, p. 8). Thus Tukhachevskii succeeded in becoming a fundholder in jet propulsion by other means, although without a monopoly. For several reasons unrelated to this theme, Tukhachevskii was arrested in May 1937 and was executed as a traitor with many other officers (Stoecker, 1998; Samuelson, 2000). If there is a connection, it is that Stalin distrusted Tukhachevskii in part for his monopolizing ambitions which were strongly suggestive of a rent-seeking military-industrial lobby (Harrison, 2003b). After this, the cause of aviation jet propulsion lacked a high-level sponsor until Malenkov began to take an interest in 1943 and was briefed on the issues by aircraft industry minister Shakhurin.13
5.2. Citizen Initiatives Many ordinary citizens with and without technical qualifications wrote to the Red Army with unsolicited ideas and suggestions for work on highspeed, high-altitude aviation, a few of which were taken up. The files of the Red Army administration for military inventions show, for example, that in April 1932 and again in March 1936, E.A. Blau submitted proposals for different aeroengines based on jet propulsion. These proposals were reviewed and rejected. One referee judged that ‘‘despite the fact that the author is an engineer [his designs] are distinguished by their naivety and demonstrate a complete absence of elementary information concerning jet propulsion.’’14 Another file collects 51 proposals for jet, rocket, or turbine engines, and airframes that were submitted in 1937, some professionally executed, some handwritten and childishly illustrated. One author proposed to lighten his super-heavy airframe by filling the wings with hydrogen, neglecting the fact that a cubic metre of hydrogen generates only one kilogram of buoyancy; another proposed a winged cruise missile but omitted to allow for a guidance system or automatic stabilization. Depending on its merits the response to each proposal was either a request for further information or a curt rejection.15 A special case is presented by GIRD, a voluntary society until it was merged with GDL to form RNII in 1933. The GIRDers were civilian engineers drawn together by a common interest; an official report describes them as ‘‘enthusiasts for the cause of rocketry who had no material base and no staff.’’16 At first they won backing from Osoaviakhim, a state-sponsored voluntary association for civil defence. Their great achievement was the
A Soviet Quasi-Market for Inventions
29
successful launch of the Soviet Union’s first liquid-fuelled rocket in August 1933.17 Then Tukhachevskii tried to take them over for the Red Army. This was the planned-economy equivalent of a friendly corporate takeover of a private high-technology start-up. 5.3. Foreign Specialists In the late 1920s and early 1930s, sympathetic foreigners with a technical interest wrote to the Red Army drawing its attention to the military significance of work on rocketry going on abroad and offering to promote such work in the Soviet Union. In 1932, for example, Rolf Engel, a German specialist in rocketry and communist party member or sympathizer, was referred to Tukhachevskii. According to biographical notes he had worked in German astronomy and as a member of the Verein fu¨r Raumschiffart (Association for Space Travel) at its test firing range outside Berlin. Engel volunteered a report on developments in rocketry in Germany and abroad, emphasizing the breadth and depth of German developments. He also proposed to bring a group of specialists to the Soviet Union to collaborate with Soviet rocketeers.18 5.4. The Foreign Press Established designers monitored the foreign press and worked up the information they found in order to demonstrate foreign progress, promote the cause of aerospace experimentation, and support bids for funding. If foreign press information was lacking, however, they still argued for increased funding on the grounds that foreign powers were evidently forging ahead of the Soviet Union in secret. The Red Army chief of armament’s files for 1931 testify to the pressure from below to take note of progress abroad and emulate it at home. This pressure was clearly related to funding decisions. For example, in May 1931 GDL director Petropavlovskii reported to the Red Army on work on rocketry abroad, mainly in Germany and the United States.19 A number of German research groups and firms, including Junkers and Opel, were described as competing for patents and funding under the umbrella of a voluntary society for space travel including armed services representatives. The American scene was said to be characterized by a similar mix of commercial and military motivations. Petropavlovskii noted that rocketry could be applied to aviation as well as artillery, with the possibility of an aircraft with a
30
MARK HARRISON
primary rocket power plant capable of speeds of 1,000 or more kilometers per hour. In a similar survey submitted at the same time the GDL rocket engineer Glushko emphasized that there was intense activity in western countries and that the basic difficulties in building rocket aircraft were close to solution; he concluded that ‘‘in the West both industrial and, particularly, military circles are keenly interested in the question of creating rocket shells and apparatuses.’’20 In the mid-1930s, the absence of foreign press information was used to promote bids to fund foreign commercial trips. This was the reason that RNII director Kleimenov, for example, used three times in 1936 when asking permission to send his engineers abroad generally, on a tour of Germany, France, Britain, and America, and to the Paris air show.21 The foreign information available, although limited, was analysed exhaustively to support funding claims. In 1939 an article on developments in rocketry that had appeared 3 years earlier in the Italian journal Revista Maritima finally reached NII-3 director Slonimer.22 He cited the article to demonstrate intensive German work on rocket munitions and asked for more money, supplies, and engineering personnel. He requested that the ‘‘appropriate organizations,’’ presumably diplomats and spies, should seek out more information abroad, and applied to send a delegation from the institute to an armaments exhibition in New York.23 The next year he made a similar request to send two specialists to Germany to find out more about work on rocketry there, and again he cited the Revista Maritima article in support.24 A survey of the historical applications of rocketry written in the ministry of the ammunition industry in 1939 noted that a veil of secrecy had descended over most military aspects of rocketry abroad; the little that was being published pointed to intense international rivalry in rocket technology. The examples cited were from a French work translated into Russian in the defence ministry.25 A translation of an American article, also from 1939, listed the potential uses of rockets as ranging from field artillery to intercontinental bombardment and space exploration, and emphasized their ease of construction and use.26 The outbreak of war did not cut off press information. Following the maiden flight of the Whittle jet-powered Gloster E.28/39 in April 1941, Flight magazine published a series of articles about jet propulsion in London. These articles were collected in a booklet and republished by the magazine editor (Smith, no date). It seems to be in this version that they reached the Soviet Union. Their impact was significant. The booklet was circulated among designers; according to Gordon and Dexter (1999, p. 150), the description of the Italian hybrid jet of 1940, the Caproni-Campini N.1, encouraged staff at
A Soviet Quasi-Market for Inventions
31
TsAGI in a similar design (see Table 1). In July 1944 minister of the aircraft industry Shakhurin copied many original Flight drawings to illustrate both existing and futuristic jet and turbine projects in a long briefing report for Malenkov.27
5.5. Foreign Commercial Information Trade links gave some information to designers, but its value in lobbying for funding of jet propulsion projects was limited. In 1935, the aircraft designer Tupolev visited the United States for a second time, his first visit having taken place in 1929/30. He toured a number of aircraft factories. He saw nothing of American progress in military rocketry and his report was silent on the whole issue of jet propulsion.28 This does not reflect secrecy; there was nothing to report because the Americans had nothing to show or hide. Surveys of the German aircraft industry were also carried out in the framework of the August 1939 nonaggression pact and these were reported to the ministry of the aircraft industry in September 1940. Here secrecy did play a limiting role; the Soviet delegations did not catch the least glimpse of the immense German activity in relation to new jet and rocket aircraft and artillery.29 There is no indication in the files that Soviet spies gained any information about progress in jet engines or rocketry in other countries. If they did, it did not reach the aeroengine designers.
6. REFINANCING When projects are long term, projects in progress require periodic refinancing. Alternatively, they must be discontinued. In this section, we look at refinancing decisions affecting projects in progress to learn more about the incentives facing designers and funding principals and the calculations they made. When refinancing decisions were disputed, there was also a mechanism for conflict resolution: designers could and did appeal adverse decisions to higher authority, ultimately right up to Stalin.
6.1. Project Evaluation and Soft Budget Constraints There were good reasons for governments to ration the funding of military aviation R&D. Most important was the fact that funding opportunities
32
MARK HARRISON
attracted both bad and good projects, which those allocating research funding could not tell apart beforehand. This made it efficient for principals to ration the available funding across projects and through time. The various research establishments reported regularly to higher authority on each project in progress.30 Through time, funders could compare the progress of alternative projects in the hope of identifying the bad projects that should not be refinanced. In this way a form of rivalry similar to yardstick competition (Shleifer, 1985) could be exploited to increase principals’ information. By monitoring the progress of long-lived projects frequently, the funder could always obtain more information about the quality of projects than was available initially. Looking forward, by providing funding in instalments and tying refinancing decisions to intermediate progress reports, the funder aimed to use the additional information to restrict financing to good projects. However, the funder could not always act on the information obtained. This was because of a weakness in the funder’s commitment to act on this information after the event. Faced with poor intermediate results it could still be efficient to go on paying for a project that, in hindsight, the funder would prefer not to have initiated in the first place (Dewatripont & Maskin, 1995). This was because of sunk costs: since part of the project had already been paid for, the likely return was now increased relative to the fraction of costs not yet sunk. One result was that despite the funder’s intentions budget constraints became soft ex post. Another was adverse selection: it gave R&D agents an incentive to understate needs and overstate expected returns so as to obtain the first instalment of funding. Once the first instalment was paid and had become a sunk cost, the payment of the next instalment became more likely. This meant that projects could continue to be refinanced even when they were known to be bad. Soviet funding arrangements thus offered a degree of protection for selfserving interests. There was clearly rent seeking; was it intentionally tolerated at any level? Were bad projects deliberately fostered, for example, to share rents and promote loyalty? Some allegations of this nature concern the rocket designer and NII-3 director Kostikov. The background is important: Kostikov remains a controversial figure. His accusers resent the fact that he took the public credit for developing the famous Katiusha rocket mortar from its true inventor Langemak who was executed (Medvedev, 1978, pp. 36–37). They argue that Kostikov was not an accidental beneficiary of the purge at RNII but a willing instrument of Ezhov and Stalin, a renegade GIRDer who turned against his former comrades. They hold him at least partly responsible for the repression of Korolev and others (Siddiqi, 2003). According to Golovanov (1994, p. 512), Korolev carried a lifelong grudge
A Soviet Quasi-Market for Inventions
33
against him for this reason. But Raushenbakh (1998, p. 66) considers the charge of complicity in the purge to be unfounded. Serov (1997, p. 4) has suggested that Kostikov was unduly favoured by Stalin in the wartime allocation of project funding. In November 1942, Stalin authorized the development of Kostikov’s unproven design for the 302 rocket fighter, at a time when the development of new weapons in other fields was being ruthlessly suppressed in favour of mass production of existing ones. Serov notes the ‘‘practically unlimited financial possibilities’’ at Kostikov’s disposal: 25 million rubles for NII-3-GIRT in 1943, compared with a similar sum for the Iakovlev and Mikoian aircraft OKBs put together. It is true that subsequently Kostikov was punished for the 302’s failure: in the spring of 1944 he was sacked, then arrested. On the other hand his punishment was mild: he was released after a year in prison, and retained his military rank and medals. Golovanov (1994, p. 511) claims that ‘‘Stalin needed Kostikov, since [the latter] was one of the bearers of the Stalinist world order.’’ Whatever that means, however, what we know of Stalin suggests that by this time he did not regard anyone as indispensable. 6.1.1. Adverse Decisions In the quasi-market for inventions, projects developed out of initiatives at lower levels. The role of funding principals and planning decisions was reactive and tended to validate these initiatives. Consequently the refinancing of projects in progress was normal and we do not usually see explicit decisions to that effect except in those rare cases when verbatim minutes of discussion meetings were preserved (Harrison, 2003a). A decision not to refinance a project in progress is illustrated from a file of the Red Army department of inventions.31 In October 1937, engineer R.G. Sergeev of the design department of aircraft factory no. 22 at Fili submitted a proposal to design a 500–1,000 kg thrust rocket motor for an auxiliary flight booster, aircraft launcher, or rocket fighter. He based his proposal on a suggestion by the German specialist Eugen Sa¨nger that had been published in an unnamed Swiss journal in 1936. He signed an agreement with the department of inventions on August 15, 1938 for the sum of 5,000 rubles. He failed to complete the work promised, so his expenses were paid off in the sum of 1,000 rubles only on September 26, 1940. We learn from this that small sums were easily written off. Bigger decisions were more complicated. This is illustrated by the turbulent history of RNII. Frustration with the results of military R&D boiled over in the purges of 1937–1938 (Harrison, 2000, pp. 128–130; Siddiqi, 2000, pp. 10–11; for previous accounts see Medvedev, 1978, pp. 34–37, 42–43;
34
MARK HARRISON
Holloway, 1982b, pp. 387–388). In May 1937, Tukhachevskii was arrested. The purge of RNII began in October with the arrest of director Kleimenov, Glushko, and others including the rocket mortar designer G.E. Langemak. In June 1938, work on the Korolev–Glushko rocket glider was suspended, the reasons given being the need to concentrate resources for rearmament on projects of more immediate military utility. A few days later Korelev was arrested, accused of being a Trotskyist saboteur, and sentenced to 10 years’ forced labour. Impatience with the lack of results of Korolev’s work on rocket aviation was clearly a factor. The testing of liquid-fuelled rocket aircraft was suspended while the rocket artillery programme was stepped up. Why did these conflicts flare with such intensity? The conflict between artillerymen and space enthusiasts at RNII had simmered through the mid1930s before the purge of 1937–1938 swept the GIRDers away, taking several of their opponents with them. Siddiqi (2003) suggests that the technological uncertainties were simply too large to be settled scientifically on the basis of the limited funding provided by principals. This heightened the risks of R&D activity, and high stakes plus limited resources fed back into bitter infighting. The end of KB-7 was decided by a combination of factors. The arrest of its sponsor Tukhachevskii created an immediate threat. At first, KB-7 director Korneev staved off repression by joining in the destruction of the leading figures of RNII; he sent slanderous allegations to Stalin about Kleimenov.32 In January 1938, with Tukhachevskii gone, KB-7 was taken away from the Red Army and handed over to the ministry of the defence industry where, like RNII (now NII-3), it was attached to the thirteenth chief administration for ammunition. But Tukhachevskii had devised KB-7 for the far-off development of liquid-fuelled rocketry, not the quick results now sought for immediate armament. For 3 years KB-7 had produced nothing to show for its outlays. The annual report of work of the thirteenth chief administration listed the projects completed under each institute or bureau and the weapons officially adopted by the Red Army for armament as a result. Under KB-7 the report for 1938 says only: ‘‘for armament in 1938 nothing supplied, in view of the long-term [perspektivnyi] character of work.’’33 In the atmosphere of the time KB-7 became an easy target. In early 1939, the Red Army resolved to close down a wide range of projects in ammunition R&D, not just those concerned with aviation jet propulsion, but including the work it was funding in KB-7. According to reports made to ammunition minister Sergeev, and forwarded by him to deputy defence minister Kulik, navy minister Frinovskii, and prime minister Molotov, the
A Soviet Quasi-Market for Inventions
35
aggregate plan for ammunition research and experimentation for 1939 had been agreed among these ministries with the general staff the previous year.34 In the course of disaggregating the plan and agreeing to individual contracts with R&D establishments, however, the Red Army had unilaterally reneged on commitments worth 40 million rubles (out of 52.5 millions) and the Navy on 7.5 millions (out of 25 millions). Even after immediate cutbacks, 25 million rubles worth of research and experimentation remained without a sponsor, including two institutes that were entirely without funding. One was KB-7. Among the projects without funding at NII-3 were the Korolev–Glushko rocket glider and a ramjet project. The effectiveness of these cancellations is not clear-cut. We do not know that either Voroshilov or Molotov gave Sergeev’s protests a hearing. By the end of the year, KB-7 had been closed down; according to Siddiqi (2003) the staff, starved of funding, turned on each other and eventually on Korneev too, who was arrested and imprisoned. On the other hand the rocket glider and ramjet projects at NII-3 were evidently reinstated, and the Korolev– Glushko RP-318 made its maiden test flight in 1940 although the designers were absent and others got the credit. To judge from the annual returns shown in Table 3, NII-3 continued to expand during 1939 at a high rate, perhaps by absorbing the staff of KB-7. But the funding of NII-3 was squeezed in real terms in 1940 and its expansion was brought to a sudden halt. In short, in 1939 the funder lost patience with the designers’ lack of results and tried suddenly to enforce a harsh constraint on the budget for research and experimentation. In the short term, this attempt was only partly successful because the constraint was softened again by the fundholder’s lobbying to reverse cutbacks, or by drawing on the fundholder’s budget, or by some combination of the two. The efforts to squeeze R&D outlays may have continued in 1940. Aircraft design yields a few cases of design organizations that were closed because of lack of results (Albrecht, 1993, pp. 214–215). More frequently, the chief designers were imprisoned along with their teams, for example, Bartini, Grigorovich, Miasyshchev, Petliakov, Polikarpov, Sukhoi, and Tupolev. Kalinin was executed (Albrecht, 1993, pp. 133–136). These and episodes such as the purge of RNII gave credibility to subsequent threats of extreme penalties for failure. Those charged with designing the Soviet Union’s first atomic bomb, for example, all expected to be arrested if it failed to detonate (Holloway, 1994, p. 215; Simonov, 2000, p. 154). The credible penalization of individuals for R&D failures helped to compensate for the weakness of the principal’s commitment to penalize organizations financially for lack of results.
36
MARK HARRISON
6.1.2. Appeals against Adverse Decisions Some designers appealed to higher authority against threats to project financing. In a related context Markevich and Harrison (2004) note that ‘‘written appeals to higher authority were a general feature of life in a society with underdeveloped legal enforcement, and citizens in all walks of life used them to seek truth and justice.’’ Disputes within RNII were a plentiful source of petitions. In May 1934, for example, both Korneev and Korolev complained to party and military authorities over RNII director Kleimenov’s suppression of liquid-fuelled rocket projects.35 After the RNII purge, Korolev appealed from prison to the prosecutor (Raushenbakh, 1998, pp. 61–64), he also wrote to both Beria and Stalin personally (Golovanov, 1994, pp. 286–289) to protest his innocence and ask to be allowed to return to work. For a petition to carry weight at higher levels the appellant had to invest something in the outcome; thus it was normal to support an appeal by listing the writer’s progress in the cause at hand. In effect the appellant offered her specialist reputation as a hostage (Williamson, 1996, pp. 120–144) to support the transaction sought. But most inventors do not usually have much reputation in their lifetimes. Most people are lucky to have even one great idea, let alone carry it out; as a result, the past is a poor guide to an inventor’s future performance and this is true both before and after she has actually invented something. At a given point in time most inventors either have nothing to show for their efforts or they are failing to live up to the promise of past achievements. Thus an inventor’s reputation is hard to establish and harder to maintain. Some other kinds of reputation that could be brought into the equation were also fragile, for example, a reputation for loyalty to superiors. In the 1930s, all were familiar with the figure of the careerist who accumulated this reputation strategically, so as to spend it later. Winning a reputation for loyalty to vertical superiors was also a good way of making enemies out of horizontal rivals. Those with more energy than talent, who risked their credit with higher levels without the talent to back it up, invited destruction as the fate of Korneev in 1939 suggests. According to Serov (1997, p. 4), Liul0 ka was known as a ‘‘complainer’’ (zhalobshchik). Having already failed in an appeal to minister Shakhurin of the aircraft industry, he wrote to Stalin in May 1942 asking to be allowed to go back to work on the turbojet. This appeal is a rarity: unlike most, it had the desired result.36 Serov (1997, p. 3) suggests that in Liul0 ka’s case it may have suited rivals to go along with his petition in order to raise the profile of the issue in their own interest. Korolev’s similar appeals from prison met
A Soviet Quasi-Market for Inventions
37
with the more usual response: they were ignored. It is true that Korolev was subsequently transferred from the Kolyma labour camps to the NKVD sharaga at factory no. 16 in Kazan0 but this, according to Raushenbakh (1998, p. 66), was entirely the result of Tupolev’s desire to recruit him.
7. THE SECONDARY ASSET MARKET When projects are long term their need for refinancing has the necessary effect of creating a secondary asset market. This section gives a brief account of the market for R&D projects and the kinds of behaviour that can be observed. According to the principle of the command economy the ownership of each project by a ministerial fundholder could only be transferred by a centralized decree. In reality there were substantial incentives for officials to mount takeover or merger bids for projects of other fundholding authorities. The attraction of a project in progress lay in the sunk costs that had already been incurred at the expense of others to whom the new fundholder did not have to pay compensation. Even if the transfer was of people rather than physical assets, the project personnel brought with them accumulated tacit knowledge, which formed significant intangible capital. The costs of takeover were political rather than financial. First, a bid required the payment of direct lobbying costs. Second, it required the expenditure of reputation; a successful bidder made promises for which he might later be held to account. Third, it weakened the centralized enforcement of ownership rights over assets on which all fundholders ultimately relied. However, circumstances could easily arise in which it was more dangerous to abstain from the secondary market than to enter it. These can be readily translated into the costs and benefits to a proprietary dictator. In Olson’s (1993) metaphor, Stalin resembles a bandit chief who settles on a territory and monopolizes it so as to maximize the rents from it. But in secondary markets that formed under his regime, lesser bandits roved. Up to a point this could benefit the economy; it reallocated resources to those who would put them to better use. But the standard of valuation of ‘‘better use’’ was private not social. Since the losing side did not recover the social value of their assets the fear of expropriation weakened the dynamic incentive to invest rather than seek rents. Finally, even if static allocation improved, the dictator’s control over the production and sharing of rents was likely to be weakened.
38
MARK HARRISON
In the secondary market for R&D assets the transactions that we observe were of two kinds. First, there were horizontal mergers and takeovers, sometimes hostile. Second, from time to time the NKVD swooped down from above to confiscate projects using its powers of arrest and confinement. Any organization could take part in the secondary market, and small organizations could be as aggressive as large ones. There were few advantages to being small in the command system other than the chance of being overlooked. As an example of the latter, to work in a small outfit like KB-7 in 1937 meant two more years of life expectancy compared with working in the larger and more prominent RNII. Usually, however, it was better to be large. Large units could realize significant economies of scope; they were less reliant on outsiders for essential goods and services. There were also managerial economies of scale. Ministerial officials tended to promote larger units on the grounds of alleged economies of scale in research and this was reflected in the frequent calls to concentrate effort and eliminate duplication or ‘‘parallelism.’’ Whether these economies really existed is another matter. Rationalization and concentration were regarded as progressive almost beyond debate and their advantages were seldom questioned, especially when comparisons were made with the scale of R&D establishments in aeroengineering abroad.37 But as long as ministers believed in them, bigger organizations had the advantage. As a result, larger units were continually on the lookout for favourable rationalization opportunities, while smaller units also had to grow, if necessary at the expense of others, or risk being swept up by rivals at any time. The logic of the takeover bid was a call to write off the past. Consider a failing project, one that had incurred significant sunk costs without giving results on schedule. Was the project intrinsically bad, underfunded, or poorly led? If the lack of results could be reasonably attributed to lack of resources or organization, then it was efficient not only to write off the sunk costs but also to refinance the project under new management. This logic may have been stronger when the scope of activity and the number of projects were increasing because growth was likely to mean an increasing number of potentially weak projects. Thus proposals for takeovers and mergers were particularly evident in the years 1937 and 1938 during and after the purge of RNII. The purge sparked a bid to rationalize research on jet engines. Staff of the Academy of Sciences Institute of Theoretical Geophysics wrote to Prime Minister Molotov at the end of December 1937; Molotov’s secretary forwarded it to both defence minister Klim Voroshilov and deputy minister for the defence industry Kaganovich for comment.38 The authors highlighted
A Soviet Quasi-Market for Inventions
39
the unmet needs of Soviet aviation in contrast to the resources being devoted to jet aeroengine development by the ‘‘capitalist countries,’’ the designs being pursued by Breguet and Junkers in France and Germany, and the veil of military secrecy, which was hiding real progress abroad. They ascribed resistance to jet designs in the Soviet Union to a coalition of ‘‘enemies of the people’’ including designers such as Langemak of RNII and soldiers such as Efimov, chief of the Red Army artillery administration. As for the established jet engineers such as Merkulov, they charged them with ‘‘creating ‘conditions’ of work bordering on mockery’’ (the phrase ‘‘bordering on’’ could have been significant: not actually mockery, just bordering on it). They called for pure and applied research encompassing ramjets, pulse-jets, and hybrid engines to be scaled up and personnel and projects concentrated in KB-7, which the Red Army should hand over to the defence industry. Although the Institute of Theoretical Geophysics had no clear interest in the fortunes of KB-7 its intervention was probably not altruistic. The chances are that someone had put them up to it. The bid failed, however. Kaganovich called on the new NII-3 director Kostikov for comment. The latter presented a strongly argued case for his own institute to be the new centre for jet engine R&D, based on a short scientific review of jet concepts and experimental results. He concluded that it was essential to draw into this line of work people ‘‘closely involved with aviation technology’’ as opposed to those ‘‘incidentally showing an interest’’ (this was a slighting reference to KB-7); Kaganovich in turn supported the NII-3 position.39 So did the Army: the new air force chief Loktionov wrote to Voroshilov supporting the writers of the Institute of Theoretical Geophysics on the principle of giving more priority to jet engines but rejecting the case for KB-7 on grounds that the latter lacked the necessary research and production equipment. He recommended NII-3 as the new centre for jet engine development, and deputy defence minister Fed0 ko relayed these arguments to Molotov adding a proposal that NII-3 absorb relevant personnel of KB-7.40 This was the eventual outcome, although KB-7 survived until the end of 1939. As has been seen, the years 1938 and 1939 saw high mortality among aviation jet propulsion projects; in 1939/40 there were several new start-ups. Therefore it is no surprise that, in his memorandum of February 1941 to deputy prime minister Voznesenskii, minister of the aircraft industry, Shakhurin, listed the various ongoing projects and proposed ‘‘to concentrate all the work in progress in [NII-3] of the ministry for the ammunition industry [y] and transfer the institute to the ministry of the aircraft industry,’’ enclosing a draft decree to that effect.41 This particular bid failed for the moment, or was overtaken by events; after the outbreak of war, NII-3 was
40
MARK HARRISON
first subordinated directly to the central government, and handed over to the aircraft industry only in 1944. Finally, when the R&D agents were seen to have gone too far their own way, Stalin used his security agency to force them back into conformity with the priorities of state. As is well known the NKVD arrested a number of aircraft and aeroengine designers in the purges of 1936–1938 and used them to formulate proposals for implementing new designs. A number of aircraft designers from TsAGI, TsIAM, and the Tupolev design bureau, including Tupolev himself, were held at factory no. 156 (Ozerov, 1973), then reorganized as TsKB-29. Of the RNII personnel arrested in 1937, some were shot and the rest sent to labour camps; some survivors were subsequently recalled from the Kolyma and put to work in the aeroengine sharaga at factory no. 16 in Kazan0 (Golovanov, 1994, pp. 318–328). To summarize, was the secondary asset market ‘‘real’’ or just a quasimarket? The dictator did not intentionally create or authorize it; the buyers and sellers were independently self-interested; and the dictator seems to have made little attempt to align their interests with his own. But there was no equilibrating process, and enforcement rested exclusively with the dictator. For these reasons it would be a mistake to think of it as a real market, but still the lines are blurred; although only a quasi-market, the secondary asset market had some strongly ‘‘realistic’’ features.
8. CLOSING THE QUASI-MARKET FOR INVENTIONS This section describes the end of the Soviet quasi-market for inventions in jet propulsion. The market was closed down between 1944 and 1946. By 1944 Germany had revealed the breakthroughs that had been in preparation for so many years, launching jet fighters and bombers and a rocket fighter into aerial combat and firing jet powered cruise missiles and ballistic rockets at London. The fog of technological uncertainty was gradually blown away, and this gave the authorities the information they needed to recapture centralized control over aerospace innovation. The new phase began on February 18, 1944 when Stalin’s war cabinet, the GKO, resolved to give new attention to jet propulsion technology. There was a certain amount of the usual reorganization and redesignation, so GIRT, under direct control of the central government since January 1942, was renamed the Research Institute for Jet-Propelled Aviation (NIIRA) and then Research Institute no. 1 (NII-1) of the aviation industry. A further decree dated May 22, 1944 commissioned a number of aircraft designers to
A Soviet Quasi-Market for Inventions
41
fit rocket and jet boosters to existing propeller-driven aircraft; Lavochkin was to build a turbojet aircraft; Glushko, Dushkin, and Isaev were to work on new rocket motors, and Liul0 ka and Uvarov were to work on jet engines; by 1945 Liul0 ka was roughly where Whittle had been in 1939, with an engine, the S-18, ready for ground testing but not for flight (Egorov, 1994, pp. 413–424, 431–436). From this point there began a marked change in the organization of aviation R&D. Regimentation and drilling increasingly took the place of initiative and networking. This process was aided by a political shock to the designers, fundholders, and funders: the ‘‘aviators’ affair’’ of early 1946 in which leaders of the Air Force and aircraft industry were arrested and imprisoned. At the end of the war Stalin concluded that the Soviet Union was lagging in aircraft and aeroengines because relationships between the Air Force and the leading aircraft design bureaus had become too cosy, enabling favoured designers to establish monopolies and relax the pace of development (Bystrova, 2000, pp. 320–321). The purge also had wider political dimensions that do not concern us (Pikhoia, 1998, pp. 45–47). A meeting in the ministry of the aircraft industry in March 1946 whipped the design sector of the industry into line. The new minister Khrunichev promoted the solution as ‘‘raising the initiative of the chief designers’’ and ‘‘lifting the design bureaus and research institutes towards more intiative-led work [initsiativnaia rabota].’’42 What this meant in practice was a demand for fewer failures, less seeking of rents, and more accountability:43 [y] we should not take the path of adventurism, we don’t have to spend money on any project [just] in order to convince ourselves that the expenditure was pointless [y] Until now we convinced ourselves of the lack of profit in one or another project or new design [only] after the development work [y] In future we must change the system and test a person’s ability over a fixed period of time. Here I have to say that a thousand objections will be raised, they’ll say ‘‘They’re hindering us, they’re not giving us the chance to work.’’ In the context of a rational state approach we must sweep aside all such discussions so that this business isn’t taken over by demagogy. We have to find the grain that will give the state the necessary yield.
The aircraft designer Il0 iushin, first in discussion, developed the point:44 Our resources are limited. We must review all the aircraft and engine types, throw all our resources into them, and move the matter forward. We shouldn’t throw a single extra kopeck where it isn’t needed. For this, our aviation industry has to stop looking at design bureaus as ‘‘free’’ organizations in the spirit of: ‘‘if something works out that’s good, and if it doesn’t no one has to account for it.’’ The ministry of the aviation industry is accountable for the work of the design bureau. That’s clear to all.
42
MARK HARRISON
The outcome was that during 1945 the German jet engines and aircraft were inventorized and allocated for further investigation and development to research institutes and design bureaus grouped under a new chief administration for jet-propelled aviation within the ministry of the aircraft industry.45 As the scope of research and design work widened, a relentless process of regular target-setting and monitoring set in. Government decrees set targets. The industry’s progress towards each target set by the decree of April 22, 1944 was rigorously monitored.46 By 1947 the industry plan for experimental work on aeroengines for 1947 listed 48 separate projects of which 38 were for jet or rocket engines; against every project was noted the government decree that authorized it, the technical parameters set for it, and the deadlines for completion and handover for external assessment.47 There was still rivalry, but it is hard to imagine anything more different from the designer-led, uncoordinated rivalry of the 1930s. It took nearly 2 years to close the quasi-market completely; the length of time required may seem surprising. The reasons are to be found abroad: just at the time the authorities were trying to close down the market at home, two completely new markets for inventions sprang up outside the country. To add to the confusion, each market was specialized in a different, competing version of the turbojet. One market was in the Soviet zone of occupied Germany and this market continued to grow actively for more than a year after the end of the war. On offer in this market were BMW and Junkers turbojets with axial-flow compressors on the same scheme as Liul0 ka. The German designs were low powered and unreliable, however. The German market was not closed until October 1946, when Stalin authorized the wholesale deportation of nearly 3,000 rocket and aviation specialists from eastern Germany. Once on Soviet territory they came under the direct control of the MVD, which accommodated them for several years and managed their work in a number of specialized sharashki (Sobolev, 1996, pp. 58–118; Harrison, 2000). The other market was a real international one. The original Heinkel and Whittle jets had used a radial compressor, which gave superior reliability although with a large front profile and limited scope for scaling upwards. Further down the same road, Rolls Royce was now manufacturing reliable engines that were efficient for their size. In 1946 the British government agreed to the sale of 25 Rolls Royce Nene engines to the Soviet Union, followed by 30 Derwent V engines (Shavrov, 1988; Egorov, 1994). At the end of the process, the Soviet Union had a jet aircraft industry with engines that came from three genetic branches: German, Soviet, and British. In 1946 the BMW and Junkers axial-flow engines were copied and installed
A Soviet Quasi-Market for Inventions
43
in the Soviet Union’s first jet aircraft, the experimental Iak-15 and twinengined MiG-9, but were not developed further. In 1947 a Soviet jet engine, Liul0 ka’s axial-flow TR-1, was at last developed for the twin-engined Su-11 fighter. Also in 1947 the Rolls Royce radial-flow Derwents were rapidly assimilated and developed for the Iak-19 fighter, while in 1948 the Nenes were installed in the Soviet Union’s first serial production jet fighter, the MiG-15.
9. PAYOFFS, REPUTATION, AND ENFORCEMENT To summarize: for a decade and a half the field of jet propulsion was marked by technological uncertainty and information bias; specialists knew more than the officials who funded them. Under these conditions centralization was out of the question. Instead, a quasi-market for inventions evolved with a secondary quasi-market for research assets alongside it. Both markets had significant ‘‘realistic’’ features, and rivalry in both was regulated with difficulty. When the uncertainty was resolved, however, the market was closed down. In Section 9, I address the issue that remains: how, in the period of uncertainty, could this rivalry have produced any successes at all? The rewards that funding principals offered to inventors were small, distributed as much for time serving and loyalty as for achievement, and could be taken away at any time. In such a setting we would expect agents to continue to invest their efforts in seeking to share the available rents rather than in making profits from inventing something new and useful. As a first pass at this problem, it would seem that successful designers were strongly motivated by two things that were very valuable and could not be confiscated. One was intellectual curiosity and a desire to solve the intrinsically interesting problem at hand, strengthened in some cases no doubt by the desire to escape from a crazy world into a bubble of scientific rationality. The other was the reputational gain attached to priority in invention. A significant aspect of this reputation is that it was not the gift of the Soviet state, although we shall find circumstances in which the state could steer its allocation; rather, it was awarded by the ‘‘Republic of Science.’’ I borrow this phrase from Dasgupta and David (1994) but the context suggests a departure from their framework. Dasgupta and David contrast motivations and incentives in Science with those in Technology: in their terminology scientists do research so as to gain priority in discovery, and technologists engage in innovation, which is replicative of scientific
44
MARK HARRISON
discovery, so as to make profits. In the present case it is obvious that technologists could achieve priority not by discovery but by invention. Where invention is concerned, therefore, Technology is like Science. I do not mean that the position and salary that the state provided were completely unimportant. Specialists at the global frontier of military technology had material aspirations for themselves and their families just like other people, and research and design work could help them towards such goals. The cash value of the potential rewards was substantial. To give a rough idea, the average monthly pay of specialist (‘‘engineering and technical’’) workers at NII-3 in the first quarter of 1941 was 818 rubles, roughly two and a half times the average industrial wage of 1940.48 More detail is available for the 250 ‘‘management and administrative’’ staff of Liul0 ka’s aircraft factory no. 165 in August 1946; this category included everyone from the chief designer (6000 rubles a month) to the floor sweepers in the labs (200 rubles). The median monthly wage was 875 rubles, compared with 626 rubles for the average industrial wage in December of that year (Filtzer, 2002, p. 235). It was enough to staff the factory fairly fully; there were only eight vacant posts, of which seven offered less than 400 rubles.49 Nominal pay was just the start. Work in military R&D also gave access to plentiful bonus payments and awards. Director Slonimer of NII-3, for example, is quoted as having received 19,250 rubles on top of his salary in 1939 and the first half of 1940.50 The evidence is fragmentary and we have no clear picture of how such sums were fixed or allocated. On the face of it officials made recommendations to mark significant achievements, transitions, and anniversaries.51 To put it another way, it seems possible that any excuse would have done. This is certainly the impression given by prewar investigations, which threw up many alleged cases of unjustified side payments and awards. A finance ministry audit of defence industry research establishments in 1938 found that TsIAM, the lead organization for aeroengine development, was running no less than 19 separate incentive schemes on which it had spent 1.2 million rubles in 1937 along with another 200,000 rubles on rest cures and sickness benefits.52 An audit of NII-3 2 years later not only threw up Slonimer as a case of unjustified side payments but also alleged that Slonimer had used incentive schemes to pay off his colleagues and bosses.53 Whether we should fully trust such accusations is another matter; a lot of unfounded allegations were flying around at that time. Also of significance equal to nominal pay in a shortage economy was the privileged consumer provisioning available to those whose jobs gave them the right to a Moscow residence permit. Mukhin (2004) has shown that in the 1930s Moscow-based aviation specialists could be persuaded to relocate
A Soviet Quasi-Market for Inventions
45
to the provinces only with great difficulty, and then only when the ministry guaranteed their future right of return. This applied whether ‘‘the provinces’’ were hundreds of kilometres away or only just outside the city limits. In short there is no doubt that working in aviation R&D offered good pay and a reasonable life style for the time. But that is not the point. The point is: was this the mechanism that motivated inventive effort rather than time serving? This seems unlikely for three reasons. First, Frey (1997) has analysed labour markets where agents are strongly motivated by intrinsic incentives such as satisfaction or reputation. He concludes that, where intrinsic incentives are strong, they are weakened by external attempts to control motivation by substituting cash. In short, when morale matters, manipulation is demoralizing. Thus, supposing the Soviet state had been capable of targetting monetary incentives accurately on inventive effort, the result was likely to have been counterproductive. Second, as a matter of historical fact the state does not seem to have had this capability. Rather, the incomes and rights of residence of aviation designers went with position rather than success. Third, it is clear that the state could and did confiscate income, rights of residence, and position at any time. There were always safer ways of earning a living than by designing military equipment. If that was what one had to do, then it was safer to be paid less, not more. In 1950, for example, Stalin suddenly accused his favourite aircraft designer Iakovlev (2000, p. 395) of diverting state funds into excessive salary and bonus payments: ‘‘Do you know what they say about you behind your back? They tell me you’re a thief.’’ What saved Iakovlev was the support of his boss, minister of the aircraft industry Khrunichev, who proved to Stalin that Iakovlev’s design team and production workers were fewer in number, lower paid, and less well equipped than those of the other designers. In short, while position and pay should not be neglected, they can hardly be regarded as credible payoffs for an inventor’s lifetime effort. Something else that was very valuable had to be available to overcome the fear of expropriation. This is why I look for another mechanism to explain their motivation, and I find it in the one thing that could be credibly offered, since it was in the gift of the Soviet Union’s rudimentary civil society rather than of the Soviet state: reputation for priority. To demonstrate the high value and decisive importance of reputation for priority in practice, I offer four kinds of evidence. First, when it turned out that a given problem had already been solved abroad we find that Soviet designers were reluctant to put effort into replicating foreign experience or to accept foreign advice. When Stalin ordered the deportation of thousands
46
MARK HARRISON
of German specialists in atomic science and aviation, to the Soviet Union in 1946 it appeared that his country was gaining priceless human capital, but in fact it was difficult and at times impossible to persuade Soviet designers to collaborate effectively even when the Germans were willing (Harrison, 2000). In the nuclear industry, Soviet designers were reluctant to copy from American designs without introducing their own trademark innovations, not all of which were successful (Lebina, 2000). Notably, Beriia tricked the designers of the first Soviet atomic bomb into replicating the work done at Los Alamos by making them believe that they had the priority; he had their leader Kurchatov guide their work on the basis of intelligence reports in such a way that they thought they were breaking all the ground themselves and did not know that they were merely following in Oppenheimer’s footsteps (Holloway, 1994). The low status of replication generally (Wible, 1998, pp. 23–42) then helps to explain why a country so rich in invention could be so poor at innovation (Berliner, 1976): the same mechanism could not promote both. Second, we find that Soviet designers feared plagiarism before the event. Harrison (2003a) provides two cases: in one, a steam turbine designer defied a direct order from a ministerial superior to share progress with rivals and so accelerate progress; those involved were clear that the desire to protect his personal priority was the motive. In another case, the gas turbine designer, Uvarov put up a barrage of excuses to shut interested observers out of his work; he argued that it was too secret to share with naval designers who wanted to find out what he was up to. A cynic might wonder whether these specialists just wanted to hide their own lack of progress, but Uvarov at least was a serious pioneer who had real claims to protect. Third, it is evident that nothing caused more lasting personal bitterness in the ‘‘Republic of Science’’ or gave rise to a deeper desire for personal vindication than the exceptional cases in which the temporal state intervened in the process of attributing priority and caused a reputation for priority to be suppressed or enabled it to be stolen. A claim that was suppressed was to the Soviet Union’s first and the world’s second rocket aircraft, which the future chief missile designer Korolev developed in the late 1930s; he was arrested and imprisoned in 1937, while it was in development, leaving it to others to fly it for the first time in 1940. A claim that was stolen was to the Katiusha, the famous rocket mortar of World War II; Langemak developed it but the new NII-3 director Kostikov took the public credit for it after Langemak was arrested and shot in 1937 (Siddiqi, 2000, p. 25). Fourth, the high value of reputation for priority is confirmed by the fact that, when it ceased to be available as an incentive for the aviation designers,
A Soviet Quasi-Market for Inventions
47
the government had to replace it with some other inducement and this took for the form of side payments that were extravagantly large by comparison with anything previously offered. By 1944, it was clear that the Germans and British had solved the principal problems of jet reaction. Thus, Soviet designers could no longer win a worldwide reputation for priority. Korolev and Liul0 ka could be the fathers of Soviet rocketry and the Soviet turbojet but no more than that. As far as Stalin was concerned the priority was now to copy the west; replication had to come before invention. Government decrees of April and June 1946 and May 1947 now set the designers in an organized competition with each other, setting design targets and offering enormous cash prizes to the design teams and chief designers that met them. First prize in the aircraft competition included 700 thousand rubles for the chief designer together with an Order of Lenin, a Stalin Prize, and a luxury ZIS-110 private car, and many more hundreds of thousands of rubles, apartments, cars, and medals to be shared among his deputies and design staff.54 Something similar was also on offer to the aeroengine designers; in the spring of 1948 Liul0 ka was given a Stalin Prize (third class) and was personally awarded 600,000 rubles, or 100 times his monthly pay in 1946, with a further 800,000 rubles for his design team; he got another Stalin Prize the next year, upgraded to first class.55 In short, the designers themselves regarded a reputation for priority in their field as extremely valuable, were ready to take considerable risks to establish and protect it, and regarded attempts to infringe on it as one of the most heinous crimes, equivalent to a deadly physical assault. To obliterate a person’s achievement was as bad as to destroy them physically. When money took the place of reputation, enormous sums were required. Finally, markets for inventions had some common features in all countries. Fears of plagiarism and confiscation were ever present in Whittle’s calculations. Because he was a serving officer, the British government as funding principal held ‘‘Free Crown User’’ rights over his patents (Whittle, 1953, p. 47) and could and eventually did do with them as it wished. Whittle had hoped (1953, pp. 102–103) to profit from the development and manufacturing. He also feared theft by Rover, an interim collaborator (1953, pp. 115–116, 205–206), but in the end the government handed the business to Rolls Royce. For his ultimate reward Whittle had to make do with £100,000 from Parliament, a knighthood from George VI, and an FRS from the Republic of Science. As for Germany, according to Ohain (Ermenc, 1990, pp. 30, 40–41), Heinkel saw no protection in patents and defended himself by keeping the German air ministry in ignorance until the first test flight; Ohain also regarded patents as worthless. In the end the big German
48
MARK HARRISON
contracts for wartime mass production of jet engines went to BMW and Junkers, not Heinkel; this was less an act of confiscation than a punishment for pursuing a radial-flow concept when German aviation officials had decided, correctly, that the future lay with the axial compressor (Kappus in Ermenc, 1990, pp. 72–75).
CONCLUSIONS To conclude: Jet propulsion R&D was carried out in the context of a vertically organized command system. In the interwar period technological uncertainty was so great that it could be processed only in a market-like context. There was a quasi-market for inventions with horizontal rivalry, competitive rent-seeking, and attempts to bar entry and create monopolies. A secondary quasi-market in research assets also sprang up that involved takeover and merger activity. Designers created the quasi-market for inventions and held the initiative in it. There were more initiatives than the authorities were willing to fund. Inventiveness was not in short supply. The authorities’ main problem was to control it, not promote it. In the Soviet Union, jet propulsion R&D was an artisan industry. The resources available to fund research were extremely limited and funding was rationed. Budget constraints on individual projects in progress tended to become soft, however. Once a project had been selected for funding it had a good chance of its funding being continued until aggregate limits on the funding principals’ resources and patience were breached. It was difficult or impossible for the authorities to know whether they were getting value for money. Designers who succeeded in getting initial funding and subsequent refinancing were ‘‘heterogeneous engineers.’’ They invested resources in lobbying and political reputation to ensure that their projects were selected for funding and, once selected, to protect them against termination from above or takeover by rivals in the name of rationalization. When faced with adverse funding or career decisions or takeover threats designers retained the option of appealing to higher instances in the vertical hierarchy. The success of such appeals rested in part on technological reputation, but an inventor’s reputation was difficult to establish and appeals were rarely successful. A reputation for priority was one reward for successful invention and it was the only reward that the state could not easily steal or confiscate after the event. This is because reputation for priority was bestowed by the ‘‘Republic
A Soviet Quasi-Market for Inventions
49
of Science’’ (or Technology), not by the state. Some evidence of the value of a reputation for priority is that, once it could no longer be won, the state had to replace it with individual incentives valued at hundreds of monthly paychecks. The state’s exploitation of the value of reputation for priority explains why the market for inventions worked, despite the fact that it was only a quasi-market, not a proper market. Selective decentralization was effective under these conditions partly because the incentive mechanism on which it relied was assured and enforced by an external party: the Republic of Science. By definition, however, the quasi-market for inventions was the only one that could be made to work this way; its particular reputation mechanism would be powerless in any activity that involved replication, that is, in most areas of economic life.
NOTES 1. RGAE, fond 7516, opis0 1, delo 324, folios 6–11 (hereafter 7516/1/324, 6–11) (no date but 1939). 2. RGAE, 8328/1/995, 111 (December 9, 1938): emphasis added. The prototype was an Uvarov gas turbine (see Table 1). 3. RGAE, 8044/1/1182, 147 (July 26, 1944). 4. RGAE, 8159/1/6, 74 (December 1936). 5. RGAE, 8162/1/300, 65–66, 80–81 (November 17, 1940). 6. RGAE, 8044/1/460, 59 (February 5, 1941) 7. RGAE, 8162/1/574, 101 (May 7, 1942). 8. In November 1929 the post of chief of armament of the Red Army was created to help carry through its equipment modernization. The first chief of armament was Army Commander Uborevich, followed in 1931 by Army Commander, later Marshal Tukhachevskii. Among the departments reporting to the chief of armament was an administration for military inventions. In 1936 the post of chief of armament was abolished, its place taken by a chief administration for supply of weapons and equipment, and under the latter a department for inventions (see Holloway, 1982a, p. 321). On Tukhachevskii and Red Army rearmament generally see Samuelson (1996, 2000) and Stoecker (1998). 9. GARF, 8418/6/243, 35–37 (May 14, 1933). 10. RGVA, 4/14/1171, 33 (January 23, 1934); also Siddiqi (2000a, pp. 4–7). 11. RGVA, 34272/1/146, 134 (October 31, 1933). 12. RGVA, 34272/1/177, 5–10 (May 27, 1934), 17–19 (May 29, 1934), 20–21 (June 1934), 1–2 (July 26, 1934), and 33 (September 13, 1934); also Siddiqi (2000a, pp. 7–9, 2000b). 13. RGAE, 8044/1/984, 264–275 (October 22, 1943). 14. RGVA, 29/56/349 (1932–1936). 15. RGVA, 29/56/354 (1937).
50
MARK HARRISON
16. RGVA, 34272/1/146, 145 (November 16, 1933). 17. GARF, 8418/6/243, 42 (August 22, 1933). 18. RGVA, 34272/1/146, 28–39 (no date but 1932). 19. RGVA, 34272/1/105, 91–94ob (May 20, 1931). 20. RGVA, 34272/1/105, 118–120 (May 1931). 21. RGAE, 8159/1/149, 220 (July 26, 1936), 219 (September 29, 1936) and 218 (October 13, 1936). 22. The reference was to Revista Maritima, 1936, 6, 421–439; see RGAE, 7516/1/ 324, 12–42, for the translation. The article was mainly about rocket artillery; it raised possible applications to aviation on the last page. 23. RGAE, 7516/1/324, 1–4 (April 9, 1939). 24. RGAE, 8162/1/305, 30 (April 16, 1940). 25. RGAE, 7516/1/324, 10 (no date but 1939). 26. RGAE, 7516/1/323, 1–18 (no date but 1939), translates ‘‘What Can We Expect of Rockets?’’ by Major James Randolph of the US Army artillery reserve, published in Army Ordnance 19(112), January–February 1939, 225 ff. 27. RGAE, 8044/1/1182, 123–147 (July 28, 1944). 28. RGVA, 29/38/96, 1–479 (June 10, 1936). 29. RGAE, 8044/1/359, 1–187 (September 27, 1940); 8044/1/358, 1–9 (September 29, 1940). 30. For example RGAE, 8159/1/137, 2–28 (no date but 1937), 8162/1/240, 9–63 (January 9, 1940), and 8162/1/449, 2–61 (January 14, 1941) for the annual reports of RNII-NII-3 in 1936, 1939, and 1940, respectively. In 1967 the annual reports for 1937 and 1938 were transferred from RGAE to the archive of the USSR Academy of Sciences where they can no longer be traced. I thank Leonid Borodkin for looking. 31. RGVA, 29/56/361 (1937–1940). 32. RGVA, 4/14/1628, 123–128 (June 15, 1937). Kleimenov had a history of personal conflict with Tukhachevskii’s subordinates, which might have helped him; see for examples RGVA, 34272/1/177, 1–2 (July 26, 1934) and 33 (September 13, 1934). But he was isolated by Ordzhonikidze’s suicide. 33. RGAE, 8162/1/89, 125 (no date but 1939). 34. RGAE, 8162/1/299, 36–54 (March–April 1939). Commissar Sergeev was unconnected with engineer Sergeev mentioned above. 35. RGVA, 34272/1/177, 5–10 (Korneev to the Okt0 iabrskii party raikom, May 27, 1934), and 17–19 (Korolev to Tukhachevskii, May 29, 1934). 36. RGAE, 8044/1/817, 19–25 (Liul0 ka to Stalin, May 18, 1942). Stalin took it up with Malenkov, who took it up with Shakhurin (Berne & Perov, 1998, p. 86). Liul0 ka’s group was absorbed into OKB-293 in July 1942 (RGAE, 8044/1/817, 18 (August 10, 1942). 37. RGAE, 8044/1/460, 49–51 (December 31, 1940): an explanatory memorandum by People’s Commissar for the Aircraft Industry, A.I. Shakhurin, on the 1941 plan for aeroengineering research and experimentation. 38. RGVA, 4/14/1925, 16–18 and RGAE, 7515/1/378, 304–306 (both December 31, 1937). 39. RGAE, 7515/1/378, 298–303 (no date). 40. RGVA, 4/14/1925, 21–21ob (February 4, 1938) and 4/14/1925, 22–22ob (February 15, 1938).
A Soviet Quasi-Market for Inventions
51
41. RGAE, 8044/1/460, 60–57 (February 5, 1941). 42. RGAE, 8044/1/1342, 17 and 21 (March 1, 1946). 43. RGAE, 8044/1/1342, 7 (March 1, 1946). 44. RGAE, 8044/1/1342, 31 (March 1, 1946). 45. RGAE, 8044/1/1318, 21, 22 (1945). This file contains the materials for a report by the Commission for Study and Assimilation of German Jet Propulsion Technology headed by minister for the aircraft industry Shakhurin. 46. RGAE, 8044/1/1496, 317–323 (December 1945) and 274–284 (January 5, 1946). 47. RGAE, 8044/1/1637, 230–235 (July 15, 1947). 48. RGAE, 8162/1/449, 87 (April 10, 1941). 49. RGAE, 8044/1/3079, 82–91 (August 27, 1946). 50. RGAE, 7516/1/692, 3 (November 21, 1940). 51. Transitions: to mark the transformation of GIRD into RNII, and in light of their achievements including the first Soviet liquid-fuelled rocket, chief of the Red Army administration for military inventions Terent0 ev asked Tukhachevskii to set aside 2,500 rubles to be distributed among the GIRDers as bonuses (RGVA, 34272/ 1/146, 145: November 16, 1933). Achievements: to mark the successful exploitation of its rocket shells in combat against Japan and Finland, NII-3 director Slonimer asked ammunition minister Sergeev to decorate his most outstanding staff, not named (RGAE, 8162/1/306, 186–187: July 22, 1940). This request then became evidence in the charges subsequently levelled against him that I describe in the text. Anniversaries: in relation to the tenth anniversary of the prison design bureau OKB172, armament minister Ustinov and NKVD chief Kruglov wrote to Stalin to request that he award commemorative decorations to the former ‘‘enemies of the people’’ working in it (GARF, 9401/2/170, 213–228: July 13, 1947). 52. RGAE, 7515/1/379, 134–137 (April 19, 1938). 53. RGAE, 7516/1/692, 1–7 (November 21, 1940). 54. RGAE, 8044/1/1795, 94 (March 26, 1948). 55. Stalin Prizes for 1947: RGAE, 8044/1/1962, 94 (March 31, 1948). Prize money, RGAE, 8044/1/1795, 79 (no date but April 1948). Stalin Prizes for 1948: RGAE, 8044/1/1965, 7 (no date but March 1949).
ACKNOWLEDGMENTS I thank the Leverhulme Trust, the British Academy, and the University of Warwick Research & Innovations Fund for financial support of my research on ‘‘Invention, Imitation, and the Birth of Soviet Aerospace’’; the University of Warwick for study leave; the staff of the State Archive of the Russian Federation (GARF), the Russian State Economics Archive (RGAE), and the Russian State Military Archive (RGVA) for access to documents; and Mike Berry, Alex Boyd, Leonid Borodkin, Keith Dexter, Jari Eloranta, Avner Greif, Malcolm Hill, Valery Lazarev, Vera Mikhaleva, Andrei Miniuk,
52
MARK HARRISON
Liudmila Selivanova, Asif Siddiqi, Paul Stoneman, Elena Tiurina, and Ivan Rodionov and the referee for valuable comments, advice, exchange of ideas, and other helpful assistance and collaboration.
REFERENCES Official Papers GARF (Gosudarstvennyi Arkhiv Rossiiskoi Federatsii): State Archive of the Russian Federation, Moscow. RGAE (Rossiiskii Gosudarstvennyi Arkhiv Ekonomiki): Russian State Economics Archive, Moscow. RGVA (Rossiiskii Gosudarstvennyi Voennyi Arkhiv): Russian State Military Archive, Moscow.
Publications Acemoglu, D., & Robinson, J. (2000). Political losers as barriers to economic development. American Economic Review, 90(2), 126–130 Papers and Proceedings. Albrecht, U. (1993). The Soviet armaments industry. Chur, Switzerland: Harwood Academic Publishers. Almquist, P. (1990). Red Forge: Soviet military industry since 1965. New York: Columbia University Press. Bartlett, W., Roberts, J. A., & Le Grand, J. (Eds) (1998). A revolution in social policy : Quasimarket reforms in the 1990s. Bristol, England: Policy Press. Bergson, A. (1961). The real national income of Soviet Russia since 1928. Cambridge, MA: Harvard University Press. Berliner, J. S. (1976). The innovation decision in Soviet industry. Cambridge, MA: MIT Press. Berne, L.P., & Perov, V.I. (1998). Istoriia sozdaniia pervogo otechestvennogo turboreaktivnogo dvigatelia (K 90-letiiu so dnia rozhdeniia A.M. Liul0 ki). In: Iz istoriia aviatsii i kosmonavtiki (Vol. 72, pp. 77–94). Moscow: Institut istorii estestvoznanii i tekhniki RAN. Bystrova, I. (2000). Voenno-promyshlennyk kompleks SSSR v gody kholodnoi voiny. (Vtoraia polovina 40-kh—nachalo 60-kh godov). Moscow: Institut Rossiiskoi istorii Rossiiskoi Akademii nauk. Danilov, B. (1981). Iz istorii sozdaniia reaktivnoi aviatsii. Voenno-istoricheskii zhurnal, 3, 70–75. Dasgupta, P., & David, P. (1994). Towards a new economics of science. Research Policy, 23(5), 487–521. Davies, R. W. (1996). The industrialisation of Soviet Russia. Vol. 4, crisis and progress in the Soviet economy, 1931–1933. Basingstoke, England: Macmillan. Davies, R. W., & Harrison, M. (1997). The Soviet military-economic effort under the second five-year plan 1933–1937. Europe-Asia Studies, 49(3), 369–406. Dewatripont, M., & Maskin, E. (1995). Credit and efficiency in centralized and decentralized economies. Review of Economic Studies, 62(4), 541–555.
A Soviet Quasi-Market for Inventions
53
Egorov, Iu. A. (1994). Zarozhdenie reaktivnoi aviatsii v SSSR. In: G. S. Biushgens (Ed.), Samoletostroenie v SSSR (1917–1945), (Vol. 2, pp. 394–436). Moscow: TsGAI. Ermenc, J. J. (1990). Interviews with German contributors to aviation history. Westport, CT: Meckler. Filtzer, D. (2002). Soviet workers and late Stalinism: Labour and the restoration of the Stalinist system after World War II. Cambridge, England: Cambridge University Press. Frey, B. S. (1997). On the relationship between intrinsic and extrinsic work motivation. International Journal of Industrial Organization, 15, 427–439. Golley, J. (1987). Whittle: The true story. Shrewsbury, England: Airlife. Golovanov, Ia. (1994). Korolev: Fakty i mify. Moscow: Nauka. Gordon, Ye., & Dexter, A. (1999). Beyond the frontiers: Soviet mixed-power fighters, 1939–46. In: Wings of fame (Vol. 15, pp. 148–157). London: Aerospace Publishing. Grigor0 ev, N.V. (1994). Aviatsionnoe motorostroenie v Velikoi Otechestvennoi voiny 1941–1945 gg. In: G.S. Biushgens (Ed.), Samoletostroenie v SSSR (1917–1945) (Vol. 2, 164–196). Moscow: TsGAI. Harrison, M. (1996). Accounting for war: Soviet production, employment, and the defence burden, 1940–1945. Cambridge, England: Cambridge University Press. Harrison, M. (1998). The economics of world war II: An overview. In: M. Harrison (Ed.), The economics of world war II: Six great powers in international comparison (pp. 1–42). Cambridge, England: Cambridge University Press. Harrison, M. (2000). New postwar branches (1): Rocketry. In: J. Barber & M. Harrison (Eds), The Soviet defence-industry complex from Stalin to Khrushchev (pp. 118–149). Basingstoke, England: Macmillan. Harrison, M. (2003a). The political economy of a Soviet military R&D failure: Steam power for aviation, 1932 to 1939. Journal of Economic History, 63(1), 178–212. Harrison, M. (2003b). Soviet industry and the Red Army under Stalin: A military-industrial complex? Les Cahiers du Monde russe, 44(2–3), 323–342. Hayek, F. A. (1945). The use of knowledge in society. American Economic Review, 35(4), 519–530. Holloway, D. (1982a). Innovation in the defence sector. In: R. Amann & J. Cooper (Eds), Industrial innovation in the Soviet Union (pp. 276–367). New Haven, CT: Yale University Press. Holloway, D. (1982b). Innovation in the defence sector: Battle tanks and ICBMs. In: R. Amann & J. Cooper (Eds), Industrial innovation in the Soviet Union (pp. 368–414). New Haven, CT: Yale University Press. Holloway, D. (1994). Stalin and the bomb: The Soviet Union and atomic energy, 1939–1956. New Haven, CT: Yale University Press. Iakovlev, A. S. (2000). Tsel0 zhizni. Zapiski aviakonstruktora (6th ed.). Moscow: Respublika. Khlevniuk, O. V. (1996). Politburo: Mekhanizmy politchecheskoi vlasti v 1930-e gody. Moscow: Rosspen. Khlevniuk, O. V., Kvashonkin, A. V., Kosheleva, L. P., & Rogovaia, L. A. (Eds) (1995). Stalinskoe Politbiuro v 30-e gody. Sbornik dokumentov. Moscow: AIRO-XX. Le Grand, J. (1991). Quasi-markets and social policy. Economic Journal, 101(408), 1256–1267. Le Grand, J., & Bartlett, W. (Eds) (1993). Quasi-markets and social policy. Basingstoke, England: Macmillan. Lebina, N. (2000). The defence-industry complex in Leningrad (2): The postwar uranium industry. In: J. Barber & M. Harrison (Eds), The Soviet defence-industry complex from Stalin to Khrushchev (pp. 184–194). Basingstoke, England: Macmillan.
54
MARK HARRISON
Litwack, J. M. (1991). Discretionary behaviour and Soviet economic reform. Soviet Studies, 43(2), 255–279. Liul0 ka, A. M., & Kuvshinnikov, S. P. (1981). K istorii sozdaniia pervogo sovetskogo turboreaktivnogo dvigatelia. Voprosy istorii estestvoznania i tekhniki, 2, 86–96. MacKenzie, D. (1996). Knowing machines. Cambridge MA: MIT Press. Markevich, A. & Harrison, M. (2004). Quality, experience, and monopoly: Regulating the Soviet seller’s market for military goods. PERSA Working Paper no. 35. Department of Economics, University of Warwick. Internet address: http://www.warwick.ac.uk/go/persa. Medvedev, Z. A. (1978). Soviet science. New York: W.W. Norton. Mises, L. von (19491998). Human action: A treatise on economics. Auburn, AL: Ludwig von Mises Institute. Mokyr, J. (1990). The lever of riches: Technological creativity and economic progress. Oxford: Oxford University Press. Morishima, M. (1984). The economics of industrial society. Cambridge, England: Cambridge University Press. Mukhin, M. (2004). Employment in the Soviet aircraft industry, 1918 to 1940: Work culture, organization, and incentives. PERSA Working Paper no. 36. Department of Economics, University of Warwick. Internet address: http://www.warwick.ac.uk/go/persa. Olson, M. (1993). Dictatorship, democracy, and development. American Political Science Review, 87(3), 567–576. Ordway, F. I., & Sharpe, M. R. (1979). The rocket team. London: Heinemann. Ozerov, G. (1973). Tupolevskaia sharaga (2nd ed.). Frankfurt am Main: Posev. Perakh, M. (1998). Laughing under the covers (Russian oral jokes). Internet address: http://www.nctimes.net/mark/htmjokes/. Pikhoia, R. G. (1998). Sovetskii soiuz. Istoriia vlasti. 1945–1991. Moscow: RAGS pri Presidente RF. Raushenbakh, B. V. (Ed.) (1998). S. P. Korolev i ego delo. Svet i teni v istorii kosmonavtika. Moscow: Nauka. Samuelson, L. (1996). Soviet defence industry planning: Tukhachevskii and military-industrial mobilisation. Stockholm: Stockholm School of Economics. Samuelson, L. (2000). Plans for Stalin’s war machine: Tukhachevskii and military-economic planning, 1925–41. London and Basingstoke: Macmillan. Serov, G. (1997). V nachale reaktivnoi ery. Samolety mira, 3–4, 2–7. Shavrov, V. B. (1988). Istoriia konstruktsii samoletov v SSSR 1938 1950 gg (2nd ed.). Moscow: Mashinostroenie. Shleifer, A. (1985). A theory of yardstick competition. RAND Journal of Economics, 16(3), 319–327. Siddiqi, A. A. (2000). Challenge to Apollo: The Soviet Union and the space race, 1945–1974. Washington, DC: NASA History Division (NASA SP-2000-4408). Siddiqi, A. A. (2003). The rockets’ red glare: Technology, conflict, and terror in the Soviet Union. Technology and Culture, 44(3), 470–501. Simonov, N. S. (2000). New postwar branches (2): The nuclear industry. In: J. Barber & M. Harrison (Eds), The Soviet defence-industry complex from Stalin to Khrushchev (pp. 150–172). Basingstoke, England: Macmillan. Smith, G.G. (no date). Gas turbines and jet propulsion for aircraft. London: Flight Publishing Co. Sobolev, D. A. (1996). Nemetskii sled v istorii Sovetskoi aviatsii. Ob uchastii nemetskikh spetsialistov v rasvitii aviastroenii v SSSR. Moscow: Aviantik.
55
A Soviet Quasi-Market for Inventions
Starkov, B. (2000). The security organs and the defence-industry complex. In: J. Barber & M. Harrison (Eds), The Soviet defence-industry complex from Stalin to Khrushchev (pp. 246–268). Basingstoke, England: Macmillan. Stoecker, S. W. (1998). Forging Stalin’s army: Marshal Tukhachevsky and the politics of military innovation. Boulder, CO: Westview. Sultanov, I. G. (1998). Istoriia sozdaniia pervykh otechestvennykh turboreaktivnykh samoletov. Mosow: Vusovskaia kniga. Voronkov, Iu.S. (1984). Teoriia i konstruktsiia: istoriia vzaimodeistviia pri sozdanii pervykh aviatsionnykh reaktivnykh dvigatelei. In: Issledovaniia po istorii i teorii razvitiia aviatsionnoi i raketo-kosmocheskoi nauki i tekhniki (Vol. 3, pp.113–122). Moscow: Nauka. Whittle, F. (1953). Jet: The story of a pioneer. London: Frederick Muller. Wible, J. R. (1998). The economics of science. Methodology and epistemology as if economics really mattered. London and New York: Routledge. Williamson, O. E. (1996). The mechanisms of governance. Oxford: Oxford University Press.
APPENDIX Soviet R&D Projects for Jet Propulsion in Plans and Reports, 1932–1944. Year
Date
R&D Organization
Designer
Design Object
1932 1932 1934 1934 1934 1936 1936 1936 1936 1936
July 4 July 4 January 23 January 23 May 10 December January 8 April 28 December December
GDL VTI RNII RNII MAI VTI VTI RNII RNII RNII
Glushko — — — — — — — — —
1936 1937 1937
December — February 28
RNII RNII VTI
— — Uvarov
1938 1938 1938 1938 1939
February February 4 February 4 December 9 —
VTI KB-7 NII-3 VTI KB-7
— — — Uvarov —
1939
—
NII-3
—
1939
April 9
NII-3
—
Rocket motor Gas turbine Aviation boosters Rocket motors GT-1 gas turbine GT-1 gas turbine GT-1 gas turbine Rocket glider Ramjet Liquid-fuelled aviation rocket motor Rocket glider Rocket glider Gas turbines GTU-3, GTU-5 Gas turbine Ramjet Ramjet Gas turbine GTU-3 Liquid-fuelled rocket motor Rocket glider RP318 Ramjet
56
MARK HARRISON
APPENDIX
(Continued )
Year
Date
R&D Organization
Designer
Design Object
1939
June 4
KB-7
—
1939
June 4
NII-3
—
1939 1939 1939 1939
June 4 October 20 December 10 December 10
VTI NII-3 y TsKTI
—
Liquid-fuelled rocket motor Liquid-fuelled rocket motor Hybrid jet engine Gas turbine Rocket glider Gas turbine Air jets
1940
January 13
NII-3
1940
September 17
NII-3
1941
January 14
NII-3
—
1941 1941
TsIAM Z-1
Uvarov Merkulov
Z-28
—
NII-3
Bas-Dubov Zaslavskii —
SKB-1
Liul0 ka
Axial turbojet
1941
March 20 February 5; April 5 February 5; April 5 February 5; April 5 February 5; April 5 April 12
NII-3
—
1941
July 30
NII-3
—
1941
August 7
NII-3
—
Ramjet Hybrid jet engine Liquid-fuelled rocket motor for interceptor aircraft Ramjet Hybrid jet engine Liquid-fuelled rocket motor for interceptor aircraft Ramjet Hybrid jet engine Liquid-fuelled rocket motor for interceptor aircraft
1941 1941 1941
Uvarov Liul0 ka Merkulov —
Rocket glider Ramjet Hybrid jet engine Rocket glider Ramjet Hybrid jet engine Ramjet Hybrid jet engine Gas turbine GTU-3 —
Hybrid jet engine
57
A Soviet Quasi-Market for Inventions
APPENDIX
(Continued )
Year
Date
R&D Organization
Designer
Design Object
1941
December 30
NII-3
—
Ramjet Hybrid jet engine Liquid-fuelled rocket motor for interceptor aircraft Rocket booster
1942 1942
January 5 May 4
NII-3 NII-3
— —
1942
May 7
NII-3
—
1942
May 29
NII-3
—
1942
June 8
OKB-293
1942 1943 1943 1943
August 10 May 20 October 22 October 22
OKB-293 TsIAM Z-84 GIRT
Bolkhovitinov Dushkin Liul0 ka Liul0 ka Liul0 ka Merkulov —
1943
October 22
KB Z-16
Glushko
1943
October 22
OKB-293
Bolkhovitinov
1943 1943
October 22 October 22
TsAGI TsIAM
1944 1944
— —
KB Z-16 NII-3
1944 1944
— —
NIIRA TsIAM
Abramovich Liul0 ka Uvarov Glushko Dushkin Isaev Liul0 ka Fadeev Kholshchevnikov Tolstov
Ramjet Hybrid jet engine Liquid-fuelled rocket motor for interceptor aircraft Rocket booster Ramjet Hybrid jet engine Ramjet Hybrid jet engine Rocket interceptor aircraft BI fighter with liquid-fuelled rocket motor Turbojet Turbojet Ramjet Liquid-fuelled rocket motor for BI Liquid-fuelled rocket motor BI fighter with liquid-fuelled rocket motor Hybrid jet engine Turbojets Rocket motor Rocket motors Turbojet Hybrid jet engine
58
MARK HARRISON
APPENDIX
(Continued )
Year
Date
R&D Organization
Designer
Design Object
1944 1944
— July 16
TsIAM KB Z-16
Uvarov Glushko
1945
March 20
Z-16
Glushko
1945
March 20
NII-1
Dushkin
1945
March 20
TsIAM
1945 1945
March 20 March 20
TsIAM NII-1
Kholshchevnikov and Fadeev Tolstov Isaev
Turboprop Rocket motor RD-1 Single chamber liquid-fuel rocket Two chamber liquid-fuel rocket Hybrid jet engine
1945
March 20
NII-1
Dushkin
1945 1945 1945
March 20 March 20 April 4
TsIAM NII-1 Z-16
Uvarov Liul0 ka Glushko
1945
April 4
NII-1
Isaev
1945
April 4
NII-1
Dushkin
1945
April 4
TsIAM
1945 1945
April 4 November 10
NII-1 —
1945
November 10
—
Kholshchevnikov and Fadeev Liul0 ka Kholshchevnikov Fadeev Glushko
1945 1945 1945
November 10 November 10 November 10
— — —
Uvarov Tolstov Isaev
1945 1945
November 10 December
— —
Liul0 ka Glushko
Hybrid jet engine Single chamber liquid-fuel rocket Single chamber liquid-fuel rocket Turboprop Turbojet Single-chamber liquid-fuel rocket Single-chamber liquid-fuel rocket Single-chamber liquid-fuel rocket Hybrid VK-107A Turbojet Hybrid jet engine Three rocket motors E-3080 turboprop Hybrid jet engine Liquid-fuelled rocket S-18 turbojet Three-chamber liquid-fuel rocket
59
A Soviet Quasi-Market for Inventions
APPENDIX
(Continued )
Year
Date
R&D Organization
Designer
Design Object
1945
December
NII-1
Isaev
1945
December
NII-1
Dushkin
1945 1945
December December
NII-1 TsIAM
1945 1945 1946
December December January 5
TsIAM TsIAM y
Liul0 ka Kholshchevnikov Fadeev Tolstov Uvarov Glushko
Single-chamber liquid-fuel rocket Single-chamber liquid-fuel rocket Two-chamber liquid-fuel rocket Turbojet Hybrid jet engine
1946
January 5
NII-1
Isaev
1946
January 5
NII-1
Dushkin
1946 1946
January 5 January 5
NII-1 TsIAM
1946 1946
January 5 January 5
TsIAM TsIAM
Liul0 ka Kholshchevnikov Fadeev Tolstov Uvarov
Hybrid jet engine Turboprop Three-chamber liquid-fuel rocket Single-chamber liquid-fuel rocket Single-chamber liquid-fuel rocket Two-chamber liquid-fuel rocket Turbojet Hybrid jet engine Hybrid jet engine Turboprop
Sources: GARF, 9401/2/65, 385; RGAE, 7516/1/309, 15; RGAE, 7516/1/318, 42–56; RGAE, 7516/1/319, 1–36; RGAE, 8044/1/460, 59, 104; RGAE, 8044/1/613, 172; RGAE, 8044/1/817, 18; RGAE, 8044/1/829, 235–242; RGAE, 8044/1/984, 253–258; RGAE, 8044/1/985, 73–76; RGAE, 8044/1/994, 21–23; RGAE, 8044/1/1182, 77–78, 81–84; RGAE, 8044/1/1321, 59–66, 233–235, 228–230; RGAE, 8044/1/1496, 317–323; RGAE, 8044/1/1496, 274–284; RGAE, 8159/1/6, 74; RGAE, 8159/1/137, 2–28; RGAE, 8159/1/140, 12–15; RGAE, 8162/1/89, 101; RGAE, 8162/1/ 240, 55–58; RGAE, 8162/1/300, 65–66, 80–81; RGAE, 8162/1/448, 7, 9; RGAE, 8162/1/449, 16–20, 9697, 180–181; RGAE, 8162/1/574, 20, 24–26, 38–40, 85, 101; RGAE, 8328/1/696, 25, 133; RGAE, 8328/1/824, 1–50; RGAE, 8328/1/919, 84; RGAE, 8328/1/992, 6–7; RGAE, 8328/ 1/995, 106; RGAE, 8328/1/996, 16–18, 22–23ob; RGVA, 34272/1/167, 23–24, 47–55, 102–119; RGVA, 4/14/1171, 33, 36; RGVA, 4/14/1925, 21; RGVA, 4/14/2800, 4.
This page intentionally left blank
60
NETWORK QUALITY IN THE EARLY TELEGRAPH INDUSTRY Tomas Nonnenmacher ABSTRACT During the early development of the telegraph industry, the network consisted of many interconnected firms that were often local monopolists. This market structure gave firms an incentive to supply a lower quality of service and charge a higher price than an integrated monopolist. Telegraph entrepreneurs attempted to contract with each other in order to provide better quality service throughout the network. However, the high costs of monitoring and enforcing these agreements made them untenable and ultimately contributed to the integration of the industry.
Perhaps the greatest evil existing under the present system, is the absence of due responsibility on account of messages sent over the lines of two or more companies, which are unreasonably delayed, or never delivered at all. General Committee, The American Telegraph Confederation 18541
The telegraph industry in the United States was in a state of disarray in 1852. Dozens of firms, many organized into loose confederations, stretched across the country. Some firms provided excellent service, but high prices, poorly constructed infrastructure, and frequent errors were the rule rather than the exception. The poor quality of service was partially due to the industry’s market structure, which inexorably moved toward consolidation. Research in Economic History, Volume 23, 61–82 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23002-1
61
62
TOMAS NONNENMACHER
Regional monopolists controlled most of the market by 1857. Within a decade a single firm, Western Union, emerged that would dominate the industry for over 100 years. During the early years of the telegraph industry, a single message would often travel over the lines of many firms in order to reach its final destination. Each firm’s demand therefore was partially determined by the price and quality of other firms in the network. Because of this interdependence, firms in the telegraph industry experienced a collective action problem. Each individual firm had an incentive to charge a higher price and produce a lower quality service than collectively optimal. Either introducing competition or unifying the network would have increased the social surplus. Because of their interdependence, connecting firms banded together, seeking to improve the network’s overall quality. Entrepreneurs in the telegraph industry contracted with each other and organized umbrella associations, but the high transaction costs of monitoring connecting firms and enforcing agreements made these attempts to improve service unworkable.
A BRIEF HISTORY OF THE TELEGRAPH Smoke signals, mirrors, and towers with moveable arms all preceded the use of electricity to transmit messages.2 It was not until the discoveries of Andre Marie Ampere, Joseph Henry, and many others that science was advanced enough to support the use of electromagnetic communication. Samuel Morse owned the first telegraph patent in the United States and is commonly referred to as the ‘‘father’’ of electric telegraphy. However, he was not a scientist, nor did he develop the technology of the telegraph. He was a portrait painter of some repute prior to embarking into the field of telegraphy. He became fascinated with the prospect of the telegraph after discussing it with his fellow passengers on a transatlantic voyage in 1832. His principal contributions were to combine previous theories and to persevere in promoting the technology when others would not.3 Morse submitted a patent for the electromagnetic telegraph in 1838, which was granted in 1840. Unable to handle the telegraph’s technology or marketing on his own, he split the patent with Leonard Gale, Alfred Vail, and F.O.J. Smith, a scientist, a mechanic, and a Congressman from Maine. With their help, he obtained a $30,000 grant from Congress in 1843 to build an experimental line between Washington and Baltimore. Morse also hired Amos Kendall, a former Postmaster General, to manage the marketing of the patent. Morse tried to sell the patent outright to Kendall, Smith, the
Network Quality in the Early Telegraph Industry
63
Post Office, and several groups of businessmen. After failing to do so, and due to increasing tensions between Morse and Smith, the patent right was split geographically. Smith controlled New England, New York and the upper Midwest, while Kendall managed the patent for the remaining patentees in the rest of the country. The patentees were left to market the telegraph piecemeal across the country. Firms typically took on the names of the cities they connected, such as the New York and Boston or the Pittsburgh, Cincinnati and Louisville Telegraph Company. The earliest lines emanated out of New York, connecting that city with Washington, Boston, and Buffalo. Rival patents were introduced in 1846 and 1849. One of these rivals, Bain’s Electrochemical Telegraph, was declared an infringement of Morse’s patent in 1851. The other, House’s Printing Telegraph, needed high quality lines and initially did not provide much competition to Morse. An additional source of competition for the Morse patentees came from one of their own construction contracts. Their patent contract with Henry O’Reilly was not clearly written, allowing him to build lines that competed with other Morse lines. O’Reilly’s contract granted him the right to build lines connecting Philadelphia to the ‘‘principal towns on the Lakes.’’ The wording of the contract was vague, and O’Reilly took it to mean he had free reign to build lines anywhere, while the patentees wanted a larger portion of the firms’ capitalization as payment for the patent. O’Reilly began a blitzkrieg of building, spanning from the East Coast through the Midwest. Morse, Kendall, and Smith called O’Reilly a pirate, while O’Reilly claimed he was merely building lines according to his original contract. In 1852, the Census reported 75 telegraph companies with 21,147 miles of wire (United States Census Bureau, 1853). Ten firms ran lines into New York City. Three lines operated between New York and Philadelphia, three between New York and Boston, and four between New York and Buffalo. In addition, two lines operated between Philadelphia and Pittsburgh, three from the Midwest to New Orleans, and lines connected most midwestern cities.4 Rates on routes typically halved when a competitor on a route entered the market, and fell again when a third competitor appeared.5 Quality competition was also fierce, with the line that erected the best infrastructure and supplied the fastest service usually dominating less capable firms. Integration of the telegraph system began in the late 1840s and was composed of system integration – mergers between two connecting firms – and horizontal integration – mergers between two firms competing on the same route. The earliest mergers were between main and feeder lines. After the Bain patent was declared an infringement, those lines merged with Morse’s.
64
TOMAS NONNENMACHER
By 1857, regional monopolies formed through a flurry of mergers.6 They signed the ‘‘Treaty of Six Nations,’’ a pooling agreement between the six largest regional firms, excluding those eastern lines still under the direct control of Kendall and Smith. The American Telegraph Company was the largest company in the pooling agreement, with lines stretching from Nova Scotia to New Orleans. The agreement lasted through the beginning of the Civil War, when the lines connecting the North and South were severed. Western Union, with its northern and midwestern routes, remained intact, and ultimately merged with the remaining major players in the industry. After 1866, Western Union was the nation’s most wide-ranging firm with the dominant telegraph network.7 Even with this dominant status, 133 firms operated in 1878 (Reid, 1886, pp. 813–815). However, while Western Union had roughly 200,000 miles of wire, the rest of the companies operated only 87,000 miles and the median company operated only 52 miles of wire.
Telegraph Quality Many factors contributed to the quality of telegraphic service. Storms could delay messages by grounding the wire or knocking the poles over. Inferior lines, mediocre receiving and transmitting equipment, and faulty batteries caused delays. Inexperienced operators made errors in sending and transcribing messages. Fraudulent operators pocketed receipts and failed to send messages. To some degree, these factors were interchangeable; for instance, a good battery might compensate for a poorly constructed line. The quality of telegraph infrastructure was mostly a function of how little contact existed between the ground and the wire. The more ‘‘leakage’’ of the electric current, the more likely the message would be garbled. Low-quality wire was more likely to stretch to the ground or become brittle and snap. Line builders steadily progressed from thin copper wire to thicker treated iron wire during the early 1850s. Where the wire came in contact with the poles, some form of insulation was necessary. Originally this insulation was merely a bit of tar pasted between two blocks of wood. Ceramic and iron insulators were also used, but the final method was a glass insulator attached to a wooden crossbar at the top of the pole. Telegraph lines were erected at costs from as little as $50 to as much as $500 per mile. Shaffner (1859, p. 747) estimated that a good quality line could be built for $300 per mile. One hundred and fifty dollars went to the patentees for the patent license. Thirty dollars went to the construction company as profits, and $120 was the actual cost of constructing a line.
Network Quality in the Early Telegraph Industry
65
Due to problems with contractors, experimentation with new materials, or just bad luck, some lines ended up poorly built while others were more permanent.8 Telegraph managers needed to decide what improvements were necessary. Continuing with inferior infrastructure was possible in most cases, as labor could substitute for capital: a message sent once over a properly constructed line might take three or four attempts on a poorly constructed one. This process would ensure that the message was sent, but it diminished the speed and accuracy with which the message reached its destination. Telegraph Demand Unlike the current telephone system, which is used for both business and personal purposes, the telegraph was primarily a tool of commerce. Although scant information remains concerning the messages sent over the antebellum system, the existing information points to pricing and quality decisions by telegraph managers that targeted businesspeople as the main clientele.9 One snapshot of the patronage of the telegraph is given in Table 1, which lists the receipts of one of the two lines operating between New York and Boston in November 1856. This list does not include the business of the press, which sometimes accounted for close to 50 percent of the traffic volume. Although some categories include social and business messages, well over 60 percent of the 20,400 messages were commercial in nature. The
Table 1.
Messages Sent between NYC and Boston November, 1856.
Total
Total (%)
1608 1050 1930 374 228 1692 644 2020 1320 3130 750 2140 3514
7.9 5.1 9.5 1.8 1.1 8.3 3.2 9.9 6.5 15.3 3.7 10.5 17.2
Source: Lefferts (1857, p. 259).
Messages to buy goods Messages to sell goods Messages making appointments Messages relating to sickness Messages relating to death Instructions to pay money and notes Messages for railroads Reports of markets Social messages Messages respecting freighting and shipping Messages in cipher On general mercantile matters Miscellaneous
66
TOMAS NONNENMACHER
emphasis on business demand made it important that the system was able to provide reliable service.
INCENTIVES AND NETWORK QUALITY Telegraph firms were often local monopolists connected to a larger network. This section examines the price and quality incentives that such firms faced. It also examines how those incentives, when wealth reducing, could have been overcome by contract. This explanation begins with the efficiency rationale for vertical integration of bilateral monopolists and is then expanded to networks with decisions about price and quality. The efficiency argument for the vertical integration of bilateral monopolists has a long theoretical history. When there are successive monopolists in a production process, the upstream firm has the incentive to charge the downstream firm a price above marginal cost. This higher price restricts the quantity of the input utilized by the downstream firm. The incentive of both firms to charge a price above marginal cost is called ‘‘double marginalization’’ or ‘‘double markup.’’ Either integrating the bilateral monopolists or introducing competition would eliminate the incentive to set the price of the intermediate good above marginal cost. An integrated monopolist would transfer the good internally at a price equal to marginal cost, thereby increasing overall profits and consumer surplus.10 The logic of double marginalization can be applied to network industries and has been used to examine the behavior of market participants in other sectors of the telecommunications industry. Haan (2001) explains free Internet access in Europe as a consequence of the double marginalization problem between Internet service providers and telephone companies. Park and Lee (2002) provide evidence of double marginalization between fixed and mobile telephone networks in Korea. Economides and Lehr (1994) expand the theory of double marginalization in two ways important to the telegraph. First, they examine the effects of double marginalization in a network. Second, they introduce an additional choice variable, quality. Their results are similar to the traditional findings that only use price as the choice variable. They predict that an integrated monopolist will charge a lower price and provide a higher quality network than bilateral monopolists. Although their models are restricted to two firms in a network, adding additional links to the system would only amplify the double marginalization effect, essentially creating triple, quadruple, etc. marginalization.
Network Quality in the Early Telegraph Industry
67
An implication of Coase’s (1960) landmark work is that market structure will not affect efficiency if transaction costs are zero. Even under conditions of monopoly, market participants will continue to exploit gains from trade until an efficient outcome is attained. Although the incentive to doublemarginalize was present in the telegraph network, Coase would argue that this problem might have had a contractual solution. If firms could have collectively agreed to lower price and raise quality, all parties, including consumers, would have been better off. In this case, it should not matter from an efficiency standpoint whether the telegraph industry was made up of 100 interconnected firms or one integrated firm. An efficient solution could have been attained by contractual means. An implicit assumption therefore of all double marginalization arguments is that the costs of transacting around the wealth-reducing incentives are too high to overcome. The sources of these high transaction costs are easy to identify in the telegraph industry. Since individuals are boundedly rational and opportunistic, contingencies can arise or be created that allow contracts to be voided. The most obvious loophole in a contract between connecting firms was the high cost of determining the source of errors in the system. Alchian and Demsetz (1972) use measurement costs to explain the emergence of firms. They argue that production processes that involve team production should be carried out within a firm, where a central contractual agent has the ability and incentive to monitor work effort. A team production process is one where it is costly to use output to measure an input’s marginal productivity. In a firm, the central contractual agent is the residual claimant and measures effort directly rather than measuring the quality or quantity of output. Sending a telegraph message can be thought of as a team production process. It was possible to measure the output of each firm in the production process, but the costs of doing so were very high. Barzel (1982) uses redundant measurement costs as a motivation for integration. He argues that those products that are difficult to measure at intermediate stages of production are more likely to be integrated into a single firm. ‘‘If a separate organization performs this function [monitoring quality] for all steps, the conservation of information is clear: There is no longer a need for each firm to monitor the inputs in all prior steps. It is hypothesized that this is a function of the firm’’ (p. 41). Firms could navigate their way around Barzel’s redundant measurement problem in one of the two ways. Either they could integrate, thereby centralizing management, or they could jointly hire a monitor to measure quality and output at each stage of production. Centralized monitoring need not imply unified ownership of assets, although enforcing a stipulated level
68
TOMAS NONNENMACHER
of quality may be easier within a firm. Internalizing the contract replaces an agreement between the monitor and a monopolist with a contract between the monitor and an employee. The difference between the two contracts is that the employee, unlike the monopolist, can be fired should she become recalcitrant. To summarize, in a network with multilateral monopolists, each firm has an incentive to lower quality and raise price as compared to either a competitive industry or an integrated monopolist. This incentive could be theoretically overcome by contractual means. However, if transaction costs, such as measurement costs, are high, these contracts will be difficult to enforce. Instead, a centralized monitor may appear, either in the form of a supervisory organization or as the owner of a vertically integrated firm. A final implication is that when competition on a particular sub-network exists, prices should fall and overall quality should increase.
THE TELEGRAPH NETWORK Even though some figures are available, most of the information concerning the infrastructure quality, the number of errors, the problems in contracting, and the improvements after integration is anecdotal. The preponderance of the evidence consists of statements and actions of telegraph entrepreneurs during the period when the industry was highly fragmented. These observations are not inconsequential, as they point to the motivations of telegraph entrepreneurs in their quest to bring order to the industry.
Double Marginalization in Price and Quality The predictions made by Economides and Lehr concerning the behavior of bilateral monopolists are substantiated in the writings of telegraph entrepreneurs. Kendall (1848) complained of monopoly pricing when he wrote, ‘‘[T]he controllers of each section [of the network] will make their charges as high as possible without regard to the interests of other companies on the same line or of the public’’ (p. 8). Kendall recognized that each firm acted like a local monopolist and would charge lower prices if control over pricing were unified into a single organization. Kendall also complained that each firm had an incentive to provide service quality lower than that which an integrated monopolist would offer.
Network Quality in the Early Telegraph Industry
69
[T]hree companies [are to be arranged] between St. Louis and Philadelphia, making five between St. Louis and Boston. Each of the companies is responsible for its own acts or omissions, but there is no joint responsibilityy . If the lines were one, there would be but one responsibility; and in all such cases the money [for errors in transmission] would be readily refunded (p. 8).
Besides arguing that quality would be higher in an integrated network, Kendall also alludes to the high cost of locating the sources of errors in a fragmented network. Firms could merely pass the blame on to another link, thereby eliminating their responsibility to refund errors. The number of errors made by telegraph companies was never recorded exactly, but some figures are available. For instance, many firms reported the amount of money refunded to customers. Between 1848 and 1855, the Washington and New Orleans Telegraph Company refunded 2.2 percent of its total receipts. These refunds ranged from 1.8 percent in 1851 to 4.8 percent in 1848 (Washington and New Orleans Telegraph Company, 1848–1855). The Erie and Michigan Telegraph Company (1855) refunded 1.3 percent of receipts in 1855. The People’s Telegraph Company (between New Orleans and Louisville) refunded 1 percent in 1852, and the Pittsburgh, Cincinnati and Louisville Telegraph Company refunded 0.07 percent in 1850 (People’s Telegraph Company, 1852; Pittsburgh, Cincinnati and Louisville Telegraph Company, 1850). These figures underestimate the impact of errors for several reasons. Firms did not offer refunds for all errors, especially when more than one firm was involved in transmission. The first firm to receive the message would hand over a portion of the receipts to the downstream firm. Recovering these receipts was often difficult. Additional time was spent investigating the lost message and communicating with the dissatisfied customer. Finally, the numbers do not measure the fall in demand due to the low quality service. Marshall Lefferts, the President of the Merchant’s Line between Boston and New York, complained about the inability to refund customers completely. We found it necessary to issue a notice to merchants, stating that we would use our best endeavors to forward dispatches beyond the terminus of our own line, but could in no way guarantee their transmission beyondy . [A customer] presents himself at our Boston office to send a message (for instance) to New Orleans. We receive and send it to New York, and there hand it over to one of the Southern linesy . And so the message is passed on, either to stop on the way, or by good luck to reach its destination. If it does not reach its destination, and which is of such frequent occurrence, the sender of the dispatch presents himself at the counter of our office and demands the return of his moneyy . We make inquiry, and when I tell you we can get no satisfaction, it is almost the universal answer; for they all insist on having sent the message throughy . Of
70
TOMAS NONNENMACHER the large number of such mistakes and miscarriages which have come under my notice, we have not been able to refund one out of ten (Jones, 1852, pp. 142 143).11
The response of Lefferts to the contractual difficulties was a common one. Since he could not easily determine the source of the error or delay, he refused all responsibility for a message beyond the terminus of his line. But as Shaffner (1859) noted, ‘‘few lines in America can pay any interest on its capital, out of its revenue from local business’’ (p. 758). Through business was critical, but the incentives to transmit it properly were misaligned. Poor quality led to a reduction in demand and lower profits. Rather than merely refuse responsibility, firms could have contracted with and monitored each other more explicitly, sued their connections for negligence or breach of contract, joined an organization that assumed oversight responsibilities for all the firms in the network, or integrated with connecting firms in the network. Contracting and using the legal system was difficult, as firms and courts had a difficult time determining the sources of errors (Nonnenmacher, 1996, pp. 112–122). Integration or oversight organizations were the remaining alternatives. Inter-Firm Monitoring As predicted by Barzel, firms joined organizations whose main purpose was to ensure quality between stages of production. At least five formal organizations operated in the 15 years prior to the Civil War. Membership, the approximate dates of operation, and the approximate miles of wire owned by each firm are listed in Appendix B. Cooperative agreements began with Samuel Morse, who envisioned that the firms operating under his patent, specifically the first eight major firms, would cooperate fully with each other. However, squabbles between himself and F.O.J. Smith led to a breakdown in this system. Between 1845 and 1852, the ‘‘Morse Lines’’ were a loose confederation, but the personality conflicts exacerbated by the network economics fostered dissent. The other factor leading to a lack of cooperation was the contract that the patentees had signed with Henry O’Reilly. O’Reilly created a supervisory corporation, the Atlantic, Lake and Mississippi Telegraph Company, to coordinate transactions between the six lines he planned to build between Philadelphia, St. Louis, and New Orleans. These lines were generally poorly constructed, leading to immediate difficulties and errors in transmission. O’Reilly’s approach was to build lines as quickly as possible. To fund each new line, he sold all the stock of the previous line once in operation. Unfortunately, his
Network Quality in the Early Telegraph Industry
71
strategy did not place control of the lines in the hands of a single manager, leading to the lack of coordination between firms. When O’Reilly became financially insolvent, the network lost its minimal effectiveness. J. D. Reid attempted to coordinate the actions of O’Reilly’s firms by creating the National Lines System, consisting of several major lines connecting Philadelphia, Buffalo, St. Louis, and New Orleans. Reid was the superintendent of these lines and continually urged the stockholders to merge into a unified firm (Reid, 1886, p. 228). Dissension prevailed and Reid’s vision of a unified midwestern system was never realized. Individual companies defected or were eventually bought out by Western Union, leaving the remaining firms weak and vulnerable. Reid’s National Lines was a component of the larger American Telegraph Confederation (ATC). The ATC was organized in 1853 and attempted to foster cooperation between Morse-patented lines. Even before its first meeting, the industry insiders recognized that the ability to track messages that traveled over several lines would be a top concern. The American Telegraph Magazine predicted that, ‘‘The proposed Telegraphic Convention in Washington, next March, will probably find one of its most important duties to consist in providing for systematic and prompt investigation of complaints connected with the receipt and delivery of dispatches, especially in cases where the dispatches pass over several lines to reach their destination’’ (December 1852, p. 67). Sixteen firms joined the organization, consisting of many of the best-run lines in the nation. The quote with which this paper begins indicates that the members of the convention viewed problems of interconnection as a major issue. The directors passed many resolutions regarding how firms should cooperate, but, as noted in Shaffner’s Telegraph Companion, these rules were not enforceable (January 1854, pp. 29–35). When the midwestern lines began to merge with Western Union, the volume of traffic on the remaining lines fell. Beginning in the early 1850s, Ezra Cornell and his partners J.J. Speed and J.H. Wade organized a rival to the National Lines in the Midwest. The partners owned a controlling share in 14 telegraph companies with a total of 3,420 miles of line. Each partner was superintendent of a major line, but even in this partnership, the inability to coordinate the actions of the lines led to financial weakness. Speed and Wade sold their interests in 1854 to the New York and Mississippi Valley (which later became Western Union). After a bitter struggle, Cornell followed them in 1856, using a portion of the proceeds from the sale to found Cornell University. Oversight organizations began to resemble firms as defined by Alchian and Demsetz. Production was done by a team, and there were several input
72
TOMAS NONNENMACHER
owners. A central agent monitored the actions of all of the inputs. Two final conditions were not met. The central contractual agent could not effectively renegotiate an input’s contract, and the central agent was not the residual claimant. The failure to meet the last two conditions made the confederations ineffective. The ATC, for instance, passed many resolutions, but without any power to punish noncompliant firms, it was ineffective in increasing the quality of service. With the signing of the Treaty of Six Nations in 1857, a formal organization appeared in which the member organizations were large regional providers with considerable market power. The contract specifically addressed how tariffs for through messages would be set, how disputes over tariffs and mistransmitted messages would be settled, and what the consequences of refusing to comply with that settlement would be. Overlapping ownership of several firms and the penalty of expulsion from the organization were the devices that smoothed over relations between connecting firms.
Competition Evidence from the industry supports the hypothesis that competition along a single route raised quality and lowered prices. This was especially true when connecting lines were willing to switch to the firm offering the best service. Judge J.D. Caton of the Illinois and Mississippi Telegraph Company (IMTC) was perhaps the most successful at playing his connecting firms off of each other to achieve the highest quality service. The IMTC ran between St. Louis and Chicago, where it connected with firms that ultimately reached New York, the final destination of many of its messages. The connecting route was broken down into Chicago to Buffalo and Buffalo to New York segments. Two firms existed between Chicago and Buffalo in 1852, while three firms operated between Buffalo and New York. The IMTC was a ‘‘Morse’’ line, built by O’Reilly, and completely rebuilt by Caton in the early 1850s. Caton felt no obligation to Morse, and used whatever connection offered him the best service. The two firms connecting Chicago to Buffalo competed vigorously for his traffic. In a letter to George Curtiss, the president of the ‘‘Morse’’ line between Buffalo and New York, Caton wrote, ‘‘Mr. Mason our manager at Chicago says that he does not get answers sent by Cornell’s Line [Chicago to Buffalo] frequently for one, two and even three days while they were rarely over 6 hours when he sent by the House line’’ (16 January 1854, Cornell, E., Papers (1746–1888)).12 Caton’s motives were clarified in a letter from Curtiss to Cornell on 15 April 1855
Network Quality in the Early Telegraph Industry
73
(Cornell, E., Papers (1746–1888)) when he wrote, ‘‘The idea conveyed was that unless they [the House line] were careful to have theirs in good order and were able to do up his business promptly, they would lose ity . [Caton] said it was not his business to know anything about any particular line House or Morse, but he must take care of the interests of his Co. and give the business to the line that done [sic] it the most prompt.’’ Competition reduced the double-marginalization incentive. Since firms were no longer monopolists, they increased the quality of their service and lowered their prices. Some additional competition was offered via alternative routes, such as sending messages to New York through Pennsylvania instead of upstate New York. Such instances, however, were rare. In many cases, only one firm operated on a particular route. When more than one firm existed, they typically dealt only with other firms using the same patent, i.e. Morse patented firms usually sent messages only over other Morse lines.
Intra-Firm Monitoring and Integration Within a firm, the telegraph entrepreneur had to design contracts that minimized the costs of monitoring his agents. The type of contract used to employ the operators ranged widely. Sometimes a flat wage was paid to the operator. In many cases a percentage of receipts, ranging from 50 to 100 percent, plus a flat wage were paid. In many smaller stations, the operator received 100 percent of the receipts. This wage indicates that other functions that the operator performed, such as line maintenance and retransmitting messages, were just as important as generating revenues. The use of high powered incentives based solely on messages sent could and did lead to distortions of the agents’ behavior, such as a failure to receive messages or retransmit through messages.13 Two factors, both related to the overall quality of a firm’s network, reduced the costs of monitoring a particular agent. First, as the quality of the infrastructure increased, the cost of monitoring an agent decreased. If the infrastructure was good, agents could not claim that errors or delays were caused by breaks in the line. Good infrastructure also reduced the amount of time the agent spent surveying and repairing the line. Ceteris paribus, the cost of monitoring an operator on the line between New York City and Philadelphia, where the infrastructure was very good, was much lower than the cost of monitoring an operator on the line between Cincinnati and Cleveland, where the infrastructure was of lesser quality.
74
TOMAS NONNENMACHER
Second, additional information concerning the actions of other agents reduced the costs of monitoring a particular agent. For example, if agents A, B, and C were station operators on a line, information from B and C allowed the manager to monitor A. The bookkeeping practices of the more diligent managers required that operators keep track of the messages sent to and received from all other stations. If the number of messages received by agent A did not match the number of messages sent from B and C, then shirking on the part of agent A was a likely possibility. Operators on the Ohio, Indiana, and Illinois submitted a monthly ledger recording the value of messages sent to and received from the other 40 plus stations. Ezra Cornell, the manager of the line, sent dozens of letters demanding that operators fill the forms out diligently.14 This information allowed him to identify weak links in the chain and discipline or remove them. Integrated firms such as Western Union faced lower transaction costs than their nonintegrated counterparts. First, by incorporating nodes into their network, they eliminated double marginalization. Two firms that were previously connected and had an incentive to undersupply quality became an integrated firm, with a higher incentive to monitor and invest. Second, as the number of firms in a network decreased, the costs of monitoring fell. There were economies of scale in monitoring. More information concerning the number and destination of messages sent necessarily meant better information about the number of messages received. This reduced monitoring costs. Finally, the firm had the ability to fire operators who shirked. In the nonintegrated system, it was difficult to remove a downstream firm that was providing poor quality service. An oft-told story of a meeting between Hiram Sibley, an early leader of Western Union, and a group of potential Rochester investors provides an account of the collective action problem facing the industry and the logic of consolidation. Sibley presented his ‘‘crazy scheme – the buying up of broken down telegraph companies, accumulating their worthless stocks, assuming their liabilities, etc.’’ to the investors. In response, one investor asked, ‘‘You admit, Mr. Sibley, that the telegraph is a failure?yAnd you further admit that each company is a failure – that in particular and in general the whole thing is a failure?yThen how is the consolidation of failures to escape failure? If there is nothing in the result which is not in the cause, where is the element of success to come in?’’ (Thompson, 1947, p. 273). The logic of Sibley’s vision of a single firm solving the collective action problem in the industry was difficult to sell, but ultimately correct. Did integration improve the quality of the telegraph network? Evidence of superior infrastructure is difficult to quantify, as the accounting methods of
Network Quality in the Early Telegraph Industry
75
the time did not record investments and repairs clearly. The onset of the Civil War and the great increase in demand associated with it also clouds the effects of integration. However, network quality was cited ex post as a leading cause of integration. In his 1869 Annual Report, William Orton cites the lack of unity between connecting firms as an important impetus for integration.15 An integrated system provided better service, generated more revenues, and allowed for greater investments in human and physical capital. The monopoly invested in infrastructure and maintenance and also standardized rates and service across the country. Western Union had 104,584 miles of wire, 412 repairmen, and 3,469 stations in 1869 (Western Union Telegraph Company, 1869, pp. 22–23). The company quickly moved to organize its tariffs into zones, allowing customers to transmit messages at an easily discovered cost to any other part of the country. For instance in 1851, the rate for a 10 word message from Washington, DC to Chicago was $1.25 (United States Census Bureau, 1853, p. 109). In 1867, the rate from Washington, DC to various cities in Illinois ranged from $2.20 (Chicago) to $3.25 (Cairo). Five years later, the rate to Chicago had fallen to $1.25. By 1890, the rate to Chicago had fallen to $0.40, while the rate to any other location in Illinois had fallen to $0.50 (Western Union Internal Memo, 1890, Western Union Collection).16
CONCLUSION The industrial organization of the telegraph industry completed a full circle between 1845 and 1866 as it evolved from monopoly, to strong competition, to cartel, and back again to monopoly. During the 1850s, firms began to organize networks of lines in order to set prices and control quality. The members of the Treaty of Six Nations consolidated lines within large territories. No longer would new companies be formed to extend the telegraph into new territories. All major expansion occurred under the supervision of an existing firm. The days of telegraph confederations were numbered, and the era of integrated systems had arrived. A retrospective article in The Evening Post, 27 February 1885, took the position that after a period of ‘‘enthusiasm,’’ a period of ‘‘administration – of the saviors of the invention’’ was necessary. The problems of double-marginalization in price and quality were not the only forces behind vertical integration. While messages in the early system generally were handed over by hand from one firm to the next for manual retransmission, the technology allowing for automatic retransmission
76
TOMAS NONNENMACHER
improved steadily from the 1860s onward, allowing for longer continuous circuits. Entrepreneurs sought to gain market power by creating powerful, incumbent networks. Strategic considerations, such as cutting a competitor off from its connections, led to some mergers. Firms that controlled the patent rights in an area could effectively bar entry. However, the constant complaints of both telegraph entrepreneurs and their customers indicate that the chaos and low quality service that the early telegraph network provided was an important impetus for the system integration that ultimately resulted in the emergence of Western Union as the market leader with a national scope.
ACKNOWLEDGMENTS Many thanks to Lee Alston, Shane Greenstein, Larry Neal, Pablo Spiller, Richard Sicotte, David Gerard, Alexander Field, and an anonymous referee. Participants at the 1996 ASSA Cliometrics session, 1996 NBER DAE, and the economic history seminars at the universities of Illinois, Michigan, Northwestern, Indiana, and Chicago made helpful comments on earlier drafts. All remaining errors are my own.
NOTES 1. Shaffner’s Telegraph Companion (January 1854, p. 31). 2. For the history of early French optical systems, see Field (1994). 3. See Thompson (1947) and Reid (1886) for additional industrial history. 4. Information on the most important entrants into the industry, the location of their wires, and their eventual fate is given in Appendix A. 5. For instance, on the New York to Boston route, the rate fell from five cents a word to two cents and finally to one cent, with the arrival of two competitors. 6. See Appendix A and Thompson (1947) ‘‘Western Union Family Tree,’’ facing p. 424, for the major mergers of the period. The dates provided in Appendix A were gleaned from Thompson (1947) and documents in the Western Union Telegraph Company Collection (1848–1963) and Cornell, E., Papers (1746–1888). 7. In 1869, William Orton, the president of Western Union, wrote that ‘‘[Western Union’s lines] reach across the continent, from the Atlantic to the Pacific Ocean, and embrace every State and Territory in the Union but Minnesota, New Mexico and Arizona, and include the British Provinces of Nova Scotia and New Brunswick’’ (Western Union Telegraph Company, 1869, p. 8). 8. For instance, Ezra Cornell used an iron and brimstone cap to insulate the New York and Erie’s line. These caps allowed the electric current to escape every time they became wet.
Network Quality in the Early Telegraph Industry
77
9. See DuBoff (1980) for a study of business demand for the telegraph and Field (1992) for the impact of the telegraph on the broader economy. 10. Perry (1989, p. 199) provides an overview of the double marginalization literature. 11. Many other telegraphers made similar complaints. See Shaffner’s Telegraph Companion (January 1854, p. 31; April 1854, p. 210). 12. Curtiss expands in a subsequent letter, ‘‘Now I do not know how much you delay his business on your line, nor how much it is delayed in Buffaloynor whether any one is bribed to delay businessybut this much I know that the messages are not delayed after they get into our hands’’ (18 December 1854. Cornell, E., Papers (1746–1888)). 13. An operator in Meadville, PA complained of his counterpart in Erie, ‘‘There is not a day passes, but some one or more offices lose business, because another office can’t be raised in time’’ (Pew to Cornell, 4 November 1854, Cornell, E., Papers (1746–1888)). 14. Many of these letters are included in the Cornell Papers. 15. Orton wrote, ‘‘[In 1851] the great number of separate lines in operation prevented that unity and dispatch in conducting the business so essential to its [the telegraph’s] success, and the public failed to secure everywhere the benefits of direct and reliable communication’’ (Western Union Telegraph Company, 1869, p. 6). 16. In constant 1867 dollars (deflated by the CPI), the rates are 1852: $2.42; 1872: $1.51; 1890: $0.65.
REFERENCES Alchian, A., & Demsetz, H. (1972). Production, information, and economic organization. American Economic Review, 62, 777–795. Barzel, Y. (1982). Measurement cost and the organization of markets. Journal of Law and Economics, 25, 27–48. Coase, R. (1960). The problem of social cost. Journal of Law and Economics, 3, 1–44. Cornell, E., Papers, (1746–1888). Division of rare and manuscript collections, Cornell University Library. Ithaca, NY. DuBoff, R. (1980). Business demand and the development of the telegraph in the United States, 1844–1860. Business History Review, 54, 459–479. Economides, N., & Lehr, W. (1994). The quality of complex systems and industry structure. In: W. Lehr (Ed.), Quality and reliability of telecommunications infrastructure (pp. 17–42). Hillsdale: Lawrence Erlbaum. Erie and Michigan Telegraph Company (1855). Report of the president to the stockholders. Detroit, R.F. Johnstone: Cornell Papers. Field, A. (1992). The magnetic telegraph, price and quantity data, and the new management of capital. The Journal of Economic History, 52, 401–413. Field, A. (1994). French optical telegraphy, 1793–1855: Hardware, software, administration. Technology and Culture, 35, 315–347. Haan, M. (2001). The economics of free Internet access. Journal of Institutional and Theoretical Economics, 157, 359–379. Jones, A. (1852). Historical sketch of the electric telegraph. New York: Putnam.
78
TOMAS NONNENMACHER
Kendall, A. (1848). Morse’s telegraph and the O’Reilly contract: The violations of the contract exposed and the conduct of the patentees vindicated (Pamphlet). Louisville: Prentice and Weissinger. Lefferts, M. (1857). The electric telegraph: Its influence and geographical distribution. American Geographical and Statistical Society, Bulletin II, New York. Nonnenmacher, T. (1996). Law, emerging technology, and market structure: The development of the telegraph industry, 1838–1868, Doctoral Dissertation University of Illinois. Park, M.-C., & Lee, S.-W. (2002). Double marginalization problems: Evidence from the Korean fixed-to-mobile service market. Telecommunications Policy, 26, 607–621. People’s Telegraph Company (1852). Stockholder meeting minutes, Western Union Collection. Perry, M. (1989). Vertical integration: Determinants and effects. In: R. Schmalensee & R. Willig (Eds), Handbook of industrial organization (pp. 183–255). New York: North-Holland. Pittsburgh, Cincinnati and Louisville Telegraph Company. (1850). Proceedings of the annual meeting of the stockholders. Western Union Collection. Reid, J. D. (1886). The telegraph in America. New York: Polhemus. Shaffner, T. (1859). The telegraph manual: A complete history and description of the semaphoric, electric and magnetic telegraphs of Europe, Asia, Africa and America, ancient and modern. New York: Pudney and Russell. Thompson, R. (1947). Wiring a continent. Princeton: Princeton University Press. United States Census Bureau. (1853). Report of the superintendent of the census for December 1, 1852: To which is appended the report for December 1, 1851. Washington, DC: Robert Armstrong. Washington and New Orleans Telegraph Company. (1848–1855). Annual reports. Western Union Collection. Western Union Telegraph Company. (1869). Annual report of the president to the stockholders. Western Union Collection. Western Union Telegraph Company Collection. (1848–1963). Western union telegraph company collection. Washington, DC: Smithsonian Institute.
79
Network Quality in the Early Telegraph Industry
APPENDIX A. MAJOR TELEGRAPH COMPANIES Date Inc.
Name
Location
Patent
Sold/Leased/Merged
Lines along the Eastern Seaboard from Washington to Maine 1846
Magnetic
New York, Philadelphia, and Washington New York and Boston
Morse
Merged with American in 1859
1846
New York and Boston Magnetic
Morse
New Jersey
New York and Philadelphia
House
1849
Boston and New York Printing
New York and Boston
House
1849
Maine
Morse
1849
North American
1849
New York and New England
Portland and Calais New York and Washington New York and Boston
1855
American
East Coast
Several
1857
United States
East Coast
Farmer
Merged with New York and New England to form New York and New England Union in 1852. Merged with American in 1859 Reorganized as New York and Washington Printing in 1851. Merged with American in 1859 Went bankrupt and reorganized as Commercial in 1852. Leased by American in 1856 Leased by American in 1856 Merged with Magnetic in 1851 Merged with New York and Boston to form New York and New England Union in 1852. Merged with American in 1859 Merged with Western Union in 1866 Merged Western Union in 1866
1848
Bain Bain
Lines connecting the East Cost with Buffalo and Pittsburgh 1846 1846
New York, Albany and Buffalo Atlantic and Ohio
New York and Buffalo Philadelphia and Pittsburgh
Morse Morse
Merged with Western Union in 1863 Became a satellite of Western Union after it merged with the Pennsylvania, 1857
80
TOMAS NONNENMACHER
APPENDIX A. (Continued ) Date Inc.
Name
Location
Patent
Sold/Leased/Merged
1848
New York and Erie
New York and Buffalo
Morse
1848
Western
Morse
1850
New York State
Baltimore, Wheeling, and Pittsburgh New York and Buffalo
Went bankrupt 1852 and was sold to Cornell and reestablished as New York and Western Union. Dissolved in 1855 Leased by Magnetic in 1858
1850
New York State Printing
New York and Buffalo
House
1856
Pennsylvania
Philadelphia and Pittsburgh
House
Pittsburgh and Louisville
Bain, Morse
Leased to Western Union in 1856
Louisville and St. Louis Buffalo, Detroit, Pittsburgh, and Milwaukee Buffalo and Milwaukee
Bain, Morse Bain, Morse
Dayton, Toledo, Indianapolis, and Chicago St. Louis, Chicago, Iowa, and Indiana Buffalo, Cleveland, Louisville, Columbus and Cincinnati Cincinnati and St. Louis
Morse
Leased to Western Union in 1856 Leased to New York and Mississippi Valley in 1854 Merged with New York and Mississippi Valley in 1855 Leased to Western Union in 1856
Bain
Merged with New York, Albany and Buffalo in 1852 Leased by New York, Albany and Buffalo in 1856 Merged with Atlantic and Ohio in 1857
Lines through the Midwest 1847
1847
Pittsburgh, Cincinnati and Louisville Ohio and Mississippi Lake Erie
1848
Erie and Michigan
1849
Ohio, Indiana, and Illinois
1849
Illinois and Mississippi
1851
New York and Mississippi Valley
1851
Cincinnati and St. Louis
1847
Morse
Bain, Morse
Leased to Western Union in 1866
House
Reorganized as Western Union in 1856
Morse
Merged with New York and Mississippi Valley in 1855
81
Network Quality in the Early Telegraph Industry
APPENDIX A. (Continued ) Date Inc.
Name
Location
Patent
Sold/Leased/Merged
1853
Ohio
Pittsburgh to Fort Wayne
Morse
1848
Cleveland and Cincinnati
Cleveland, Columbus and Cincinnati
Morse
Merged with New York and Mississippi Valley in 1854 Merged with New York and Mississippi Valley in 1854
Washington and New Orleans People’s
New York and New Orleans Cincinnati and New Orleans
Morse
1851
New Orleans and Ohio
Pittsburgh and New Orleans
1860
Southwestern
Pittsburgh and New Orleans
Lines connecting to New Orleans 1846 1849
Barnes & Zook, Bain Morse
Several
Leased to Magnetic in 1856 Merged with New Orleans and Ohio in 1852 Leased by New Orleans and Ohio Telegraph Lessees in 1854. Reorganized as Southwestern in 1860 Merged with American in 1865
APPENDIX B. TELEGRAPH ORGANIZATION MEMBERSHIP, APPROXIMATE DATES OF OPERATION, AND APPROXIMATE MILES OF WIRE Telegraph Companies Proposed for O’Reilly’s Atlantic, Lake and Mississippi Company: 1846–1849 Atlantic and Ohio (310), Pittsburgh, Cincinnati, and Louisville (370), Ohio and Mississippi (260), Ohio, Indiana, and Illinois (795), Lake Erie (550), Illinois and Mississippi (330). Telegraph Companies in Reid’s National System: 1849–1856 Atlantic and Ohio (310), Pittsburgh, Cincinnati, and Louisville (370), New Orleans and Ohio (1,200), Ohio, Indiana, and Illinois (795), Ohio and Mississippi (260), Illinois and Mississippi (330).
82
TOMAS NONNENMACHER
Telegraph Companies in the Cornell/Speed/Wade System: 1853–1856 New York & Western Union (440), Erie and Michigan (800), Cleveland and Cincinnati (250), Cincinnati and St. Louis (400), Cleveland and Louisville (175), Cleveland and Pittsburgh (150), Allegheny and Pittsburgh (200), Erie and Waterford (15), Ohio, Indiana, and Illinois (795). Telegraph Companies in The American Telegraph Confederation: 1853–1856 New York and New England Union (250), New Orleans and Ohio (1,200), Atlantic and Ohio (310), Pittsburgh, Cincinnati and Louisville (370), Illinois and Mississippi (330), Ohio and Mississippi (260), St. Louis and Missouri River (306), Magnetic (260), Washington and New Orleans (1,700), Western (324), Philadelphia and Wilkes-Barre (115), Susquehanna River and North and West Branch (unknown), Maine (350), St. Louis and New Orleans (690), American (Baltimore to Harrisburg) (72), New York, Albany and Buffalo (513). Telegraph Companies in the Six Party Contract or Treaty of Six Nations: 1857 American (Newfoundland, Nova Scotia, New Brunswick, All New England States, New Jersey, Delaware, North Carolina, South Carolina, Georgia, and Florida), New York, Albany and Buffalo (New York), Atlantic and Ohio (Pennsylvania), Western Union (Ohio, Indiana, Michigan, Illinois, Wisconsin), New Orleans and Ohio Telegraph Lessees (Kentucky, Tennessee, Arkansas, Mississippi, Louisiana), Illinois and Mississippi & Chicago and Mississippi (Iowa, Illinois, Minnesota, Missouri).
THE SPANISH INFRASTRUCTURE STOCK, 1844–1935 Alfonso Herranz-Lonca´n ABSTRACT This paper presents the first estimates of Spanish infrastructure stock and investment for the period 1845–1935. Several sources and techniques have been used in the estimation, and the new series are reasonably reliable to the standards of historical statistics. Two distinct periods may be distinguished in the series: the years before 1895 (characterized by the prominence of railroads) and the period 1895–1935 (when most investment was addressed to other assets). The new series allow a preliminary comparison of the Spanish infrastructure endowment with that of the most advanced countries, showing a gradual process of convergence before 1936.
1. INTRODUCTION By the end of 1844, after half a century of successive wars and constant political turmoil and fiscal crisis, the Spanish infrastructure stock was meager and in very bad condition. It barely consisted of a network of paths that were unevenly distributed across the country. Of these paths only a small part was adapted to cart traffic, and an even smaller proportion actually deserved to be called roads. In addition, some small-scale ports and a few Research in Economic History, Volume 23, 83–126 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23003-3
83
84
ALFONSO HERRANZ-LONCA´N
dams, canals and irrigation ditches completed the country’s infrastructure endowment. Many of those structures were several centuries old by 1844, and some of them had their origin in former Roman works.1 Between the mid-1840s and Franco’s coup d’E´tat of July 18, 1936, the scenery was completely altered. The old paths gradually disappeared under the surface of a real network of macadam roads, which, by the eve of the Civil War of 1936–1939, connected most population centers of the country and had started to be covered with the so-called ‘‘special’’ surface treatment. At the same time, the railroads, which would perform the principal role in the Spanish transport revolution, spread all over the territory. Finally, the most important ports were endowed with complex and large-scale structures, the number of dams and irrigation canals increased substantially, and cities were equipped with modern networks of lighting, water distribution and sewage systems. Over 90 years, infrastructure had changed the face of the country and had become one of the most visible aspects of Spanish modernization. In fact, it constituted one of the factors that made economic growth possible. The slow but continuous rise of income per capita in Spain between the midnineteenth century and 1936 would have been stopped, or been constrained to some coastal areas, without the spectacular reduction in transport, communication and energy costs brought about by the new infrastructure. As economists have long insisted, this sort of cost reduction, its dynamic long-term consequences and, specifically, the changes in the structure of location incentives that it produces, are absolutely crucial for growth. Thanks to the increase in infrastructure endowments, economies are allowed to exploit fixed resources that would otherwise remain idle due to the high transport costs. In addition, infrastructure allows a large share of nonagrarian production activities to concentrate in a few industrial centers, in a process that is absolutely crucial for the growth of developing economies and the rise of technologically advanced sectors.2 It is not surprising, therefore, that infrastructure has always received the attention of economic historians and, more concretely, has always been included among the factors that may explain the evolution of the Spanish economy between 1850 and 1936. During those years, Spain experienced a sustained process of growth, but was unable to converge with the most industrialized European economies,3 and researchers have attempted to find out to what extent the shortage or inefficiency of infrastructure was one of the factors to blame for that relative failure. Research on the role of infrastructure in Spanish economic growth before the Civil War of 1936–1939 has been characterized by two features. On the one hand, as is usual in international historiography, analyses of the subject
The Spanish Infrastructure Stock
85
have focused on the railroad system. On the other, opinions on the role that railroads performed in Spanish economic growth have been divergent, provoking an intense debate that still remains unresolved. Some historians have indicated that the State’s mistakes in the regulation of railroad construction and operation prevented an intensive use of the railroad network and, as a consequence, substantially reduced the economic impact of the railroad system.4 By contrast, other researchers have insisted that railroads had an enormous growth impact in Spain, which was in fact much higher than in other European economies due, among other reasons, to the lack of opportunities for the development of inland waterways in the country.5 After those initial interpretations, research devoted to the analysis of different aspects of the railroad system has been abundant.6 In addition, during the last few years, some studies on other parts of Spain’s infrastructure have been published.7 However, in spite of these recent research efforts, an aggregate and systematic approach to the whole Spanish infrastructure stock is still missing. Such an approach would allow a better knowledge of some essential aspects of the history of Spanish infrastructure, such as the distribution of investment efforts among different assets, the importance of infrastructure capital formation within the whole economy, or the relative degree of infrastructure shortage in Spain compared with other countries. In addition, it would provide a quantitative basis for the solution of the old debate on the economic impact of Spanish infrastructure. These pages are aimed precisely at filling that gap, by providing estimates of Spanish infrastructure investment and stock between 1845 and 1935. The text is organized in four parts. The next section offers a brief account of the methods that have been followed for the construction of the new series. Section 3 presents the main results of the estimation, and performs some sensitivity analyses. Finally, on the basis of the new figures, Section 4 describes the main features of Spanish infrastructure investment and stock during the period under study. It is followed by a summary of the main conclusions, and an appendix containing the complete series and the main sources that have been used to obtain them.
2. THE NEW SERIES OF SPANISH INFRASTRUCTURE STOCK AND INVESTMENT: ESTIMATION METHOD Infrastructure is quite a tricky concept, since most people know what it is, but it is difficult to find a widely accepted definition in the literature.8 From
86
ALFONSO HERRANZ-LONCA´N
a very broad point of view, infrastructure might be described as the stock of structures and support services that are necessary for the economic development of an area. This preliminary definition, however, admits many possible interpretations, from very narrow views of the concept, such as Hirschman’s (who limited the hard core of infrastructure to transport and energy distribution), to very wide perspectives, such as those which include within the concept, the so-called ‘‘civic’’ infrastructure, i.e. ‘‘the way in which business is done’’.9 Most empirical studies, however, tend to fall somewhere between these two extreme interpretations, and consider infrastructure as the stock of physical capital goods that are organized in networks and fixed to the territory, and provide services which show some of the typical features of public goods. In Diewert’s words, infrastructure services are similar to public goods as far as: ‘‘(i) there is a substantial cost of providing the service to an area and a small marginal cost of adding an extra customer to the area service grid (y) and (ii) charging customers the marginal cost of providing the service once the area service has been established will not lead to an efficient allocation of resources’’.10 These features explain why Public Sector intervention is so frequent in infrastructure construction and management. However, infrastructure services are not pure public goods, as far as their use is subject to congestion and excludable to some extent.11 Infrastructure is usually divided into economic infrastructure (i.e. those assets that provide direct services to production, such as transport, communication or energy distribution networks) and social infrastructure (i.e. those assets that enhance social welfare, such as education and health structures).12 This is not a strict division, because some elements, such as universities, perform both economic and social functions. However, it is a useful distinction because the impact of each of those two categories on productivity is very different. Several empirical studies have shown that economic infrastructure is much more conducive to increases in output than social infrastructure.13 The differences between economic and social infrastructure make a separate study of each of those two categories advisable. In this context, this research focuses on economic infrastructure due, among other reasons, to the slow development of Spanish social infrastructure during the period under analysis, and also due to the fact that most information about its historical evolution is only available at a local level, and thus beyond the possibilities of an individual researcher. Accordingly, the assets that are considered here are the following:
87
The Spanish Infrastructure Stock
transport infrastructure (roads, railroads, urban transport, canals and ports); communication infrastructure (telegraph and telephone networks); energy distribution networks (gas and electricity); water infrastructure (dams and irrigation canals); and urban and suburban infrastructure. Within these sectors, only the assets that are usually considered as infrastructure are covered by the research. Therefore, the estimates do not measure the value of the total capital stock of sectors such as transport or electricity, but include only those elements that can be considered as infrastructure strictu sensu, i.e. which are fixed to the territory and have certain public character. For instance, within transport, railroad rolling stock, merchant ships or motorcars are not encompassed by the series, since they are not fixed to the territory. Similarly, the value of land has also been excluded from the estimation, because it does not belong in the capital stock. The starting point of the estimation is 1844. This was determined by the availability of information. In fact, a more comprehensive analysis of the Spanish industrialization process would have required starting the research 15 or 20 years earlier. However, the available data on investment in infrastructure before 1844 is too scarce to allow the estimation of yearly series. The estimates are net capital stock figures, which are intended to reflect the productive capacity of each asset, as well as its gradual decay over time. Several procedures are available to estimate net capital stock figures. Obviously, the optimal method is the direct measurement of the replacement cost of the stock in a series of benchmark years. However, the amount of information required to apply this estimation technique is enormous, making its use difficult even when analyzing economies today, and virtually impossible in historical research. As a second best choice, most researchers use the perpetual inventory method, according to which the stock is estimated through the accumulation of investment flows over time, after establishing a number of assumptions about the pattern of survival and efficiency decay of the assets. Accordingly, the net capital stock at time t is estimated as K t ¼ F0 I t þ F1 I t
1
þ . . . þ FT I t
T
(1)
where I is the level of gross fixed capital formation in each year, the series of coefficients f reflects the efficiency decay of capital over time, and T the useful life of the capital. In order to apply expression (1), it is necessary,
88
ALFONSO HERRANZ-LONCA´N
therefore, to obtain sufficiently long gross investment series, and to make likely assumptions about the useful life and the efficiency decay of each asset. The next few paragraphs describe the way in which these two aspects have been dealt with in this research.
2.1. The Gross Investment Series As far as investment data are concerned, most estimates of infrastructure stock for current economies are based on the use of public gross capital formation as a proxy for gross investment in infrastructure. This approach does not provide exact figures, because public investment includes some assets that do not belong in infrastructure and, at the same time, excludes a certain proportion of infrastructure because it is privately owned. However, the biases associated with these two problems are usually not very serious for recent times, and the use of public capital figures as representative of infrastructure is, to some extent, justified by the ease of their estimation.14 For instance, in the case of Spain, the evolution of the infrastructure stock from 1955 onwards has been recently estimated by the Instituto Valenciano de Investigaciones Econo´micas (IVIE) on the basis of public investment data.15 With the objective of carrying the analysis back in time, the IVIE has also produced figures of the State’s net capital stock for the years 1900–1990.16 However, in the Spanish case, the use of public capital figures as representative for infrastructure is not appropriate for the pre-1936 period because a large number of assets, such as railroads, tramways, the telephone system, energy distribution networks and some hydraulic works (i.e., more than 50% of the stock of infrastructure) were privately owned at the time. As a consequence, the estimation of infrastructure series for the period before 1936 needs to complement the available information about the State’s capital stock with data on private infrastructure investment. In fact, this task has been recently initiated by the IVIE itself, with an estimation of the stock of Spanish railroad infrastructure since 1844.17 Unlike this sectoral approach, in this research I have adopted a wider perspective and present estimates for all (public and private) Spanish infrastructure during the period 1844–1935. Therefore, in order to carry out the estimation, I had to retrieve investment data for each type of infrastructure, regardless of its public or private character. To that purpose, I followed different procedures. In a few cases (broad gauge railroads, State roads and ports), I was able to find some information on the amounts that
The Spanish Infrastructure Stock
89
were invested each year, and had only to make some statistical or accounting adjustments. By contrast, for other types of infrastructure, I had to estimate the value of the ‘‘gross’’ capital stock (i.e., without accounting for the efficiency decay of the assets) in one or several benchmark years, on the basis of the available technical and accounting information, and had to use physical indicators of the evolution of the stock to transform the benchmark estimates into yearly series of gross stock. These were then first-differenced in order to get ‘‘new’’ yearly investment figures, i.e. data on the annual excess investment after replacing the assets that were retired each year. Each series of ‘‘new’’ investment was then added to a hypothetical series of replacement investment, which was estimated according to the assumptions on the useful life of each asset. Finally, I took the sum of ‘‘new’’ and replacement investment as a gross investment series for each type of infrastructure.18 All investment series are presented in Table A1 of the appendix.19 Figures are expressed in constant pesetas of the mid-point year of the period under analysis (1890). The price index that has been used to deflate the available investment series (in the cases of broad gauge railroads, State roads and ports), or to express in 1890 pesetas the stock and investment indices estimated on the basis of physical indicators (in all other cases) is the deflator for ‘‘Other construction’’ investment, which has recently been estimated by Leandro Prados de la Escosura.20
2.2. Assumptions on the Useful Life and Efficiency Decay of Infrastructure In the context of capital stock estimates, the absence of adequate information often makes it necessary to establish assumptions about the useful life and the efficiency decay patterns of the assets. These assumptions use to be too simple, especially when compared with the complexity of the technological and structural changes of the economic system, which have a direct influence on the process of effective capital destruction.21 As a consequence, assumptions on these two issues tend to introduce biases in the final series that are very difficult to make up for, especially in a historical estimation exercise for which data is much scarcer than for present times. In this research I tried to choose those assumptions that seemed most likely, on the basis of the available information, but in the next section I make an approach to estimating the potential size of the biases associated with these aspects by analyzing the sensitivity of the estimates to the establishment of different assumptions.
ALFONSO HERRANZ-LONCA´N
90
The useful life figures that I finally adopted in the estimation are shown in Table 1, and are similar to those applied in other estimations of capital stock for the late nineteenth and early twentieth century. In a few cases (narrow gauge railroads and tramways) I was able to get information on assets that were retired before the end of their useful life due to their obsolescence or lack of profitability, and I subtracted their value from the net stock at the time of their retirement, and not at the end of their useful life. Assuming the useful life figures in Table 1 made it necessary to exclude the oldest infrastructure from the estimation, i.e. those tracks that were in use before the construction of the road network, and a number of dams, canals and irrigation ditches that had been built before the eighteenth century. By 1844, all of them had long reached the end of their useful life, and their valuation in net terms at the end of that year would have required information on their state of repair. According to some contemporary indications, they seem to have been in a very precarious condition, especially in the case of tracks.22 As a consequence, they would have amounted to a very tiny share of the total stock in 1844, and this would have gradually become negligible with the construction of new assets. This is shown in more detail in the next section, which includes an analysis of the sensitivity of the estimates to the exclusion of the oldest infrastructure. As far as the efficiency decay of the assets is concerned, the most usual patterns in the literature are the one-hoss shay, the straight-line and the geometric decay. According to the first one, assets would maintain full
Table 1. Useful Life Assumptions. Assets
Years of Useful Life
Railroads: Grading, works and stations Track and accessories
100 18/30a
Tramways Roads Ports Telegraph and telephone Energy distribution networks Urban infrastructure Hydraulic works
25 80 80 30 25 25 80
Sources: Feinstein (1988); Groote (1996) and, for railroad track; Go´mez Mendoza (1982) and (1989a). a Change in 1872.
The Spanish Infrastructure Stock
91
efficiency until their retirement. In other words, it would be assumed that f0 ¼ f1 ¼ . . . ¼fT 1 ¼ 1 and fT ¼ 0 in expression (1). The straightline pattern involves an identical loss of efficiency during the whole useful life of assets, i.e. it is assumed that f0 ¼ 1; f1 ¼ 1 1=T; f2 ¼ 1 2=T; . . . ; fT 1 ¼ 1 ðT 1Þ=T and fT ¼ 0: Finally, the geometric decay pattern assumes that the productive capacity of assets decreases at a constant rate d or, in other words, that ðft 1 ft Þ=ft 1 ¼ d for all toT: The geometric decay pattern has received more empirical support than the rest, the depreciation rate usually being calculated as d ¼ 2=T: On the basis of a detailed econometric analysis of second-hand asset prices, Hulten and Wykoff have suggested a modification of this expression for different types of capital. In the case of non-residential structures, they have estimated a depreciation rate of 0:91=T; which has been used here to obtain the infrastructure stock series.23 The final figures of net stock for each type of infrastructure are presented in Table A2 of the appendix. Figures always refer to the end of the year and are expressed in constant pesetas of the mid-point year of the period under analysis (1890).
3. ESTIMATION OUTCOMES AND SENSITIVITY ANALYSIS Figures 1 and 2 show Spanish infrastructure gross investment and net stock series during the period 1844–1935. In order to provide a more complete picture of the historical process of Spanish infrastructure construction, they also include the IVIE estimates for the years 1955–1994.24 Despite the gap of the Civil War and the post-war, and the coverage differences between figures for the period before 1936 and after 1954, the graphs offer a consistent picture of the evolution of Spanish infrastructure from the mid-nineteenth century onwards. They clearly show the four periods of maximum growth of the stock, i.e. the railroad mania of the last years of Queen Isabel II’s reign (1855–1866), the 1920s, the end of the Francoist dictatorship (late 1960s and early 1970s) and the period 1986–1992.25 Between those conjunctures, a series of periods of stagnation appear in the graphs. These correspond to the depression of the late nineteenth century, the Civil War and the beginning of the Francoist rule, and the crisis of 1973. Despite the likelihood of the global trends, the accuracy of the new estimates is only relative, as is usual with historical statistics, and, although
ALFONSO HERRANZ-LONCA´N
92
Million of 1890 pesetas
10000
1000
100
10
1985
1975
1965
1955
1945
1935
1925
1915
1905
1895
1885
1875
1865
1855
1845
1
Fig. 1. Spanish Gross Infrastructure Investment (1845–1994). Source: for 1845–1935, my own figures; for 1955–1994, Mas, Pe´rez, and Uriel (1998).
Million of 1890 pesetas
100000
10000
1000
Fig. 2.
1994
1984
1974
1964
1954
1944
1934
1924
1914
1904
1894
1884
1874
1864
1854
1844
100
Spanish Net Infrastructure Stock (1844–1994). Source: for 1844–1935, my own figures and, for 1955–1994, Mas et al. (1998).
The Spanish Infrastructure Stock
93
the overall process of Spanish infrastructure construction seems to be adequately described by the series, inferences drawn from minor details or from short-term variations must be taken with extreme caution. The series have been estimated on the basis of limited information, and it has been necessary to establish a number of assumptions to make up for the lack of data. This may have introduced several biases in the final figures, which may be assumed to be larger at the beginning of the period than at the end of it, in the partial series than in the aggregate, and in the short term rather than in the long term. The potential biases may have different origins. Firstly, some assets have been excluded from the series due to the absence of information. For instance, it has not been possible to include airport infrastructure in the estimates due to the lack of adequate investment data, although this exclusion seems to be of minor importance in the period before 1936.26 As has already been indicated, the assets that were constructed before the nineteenth century have also been excluded from the series, due to the lack of information on their construction costs and state of repair by 1844. As far as they were still being used at that time, the stock estimates of the first decades would contain a downward bias. In order to approach the maximum size of that bias, I have made an optimistic valuation of those assets, by assuming for them similar construction costs to those of the eighteenth and nineteenth century, and a ratio between net and gross stock of 50% by 1844. Under these assumptions, ancient infrastructure would amount to 9% of the stock in 1844, and just 1% 20 years later. A second source of potential bias is the poor quality of the empirical information that has been used to estimate some of the sectoral series. The most serious problems arise in the case of urban infrastructure, for which I have been unable to obtain reliable systematic data, and I merely suggest a correction of the aggregate figures on the basis of very insufficient evidence. But empirical information is also scarce and of very poor quality in other sectors, such as non-public railroads, local roads, the telegraph network, gas distribution infrastructure and canals. Unlike the rest of the series, which are based on reliable data on the invested resources and the physical characteristics of the stock, information regarding these five sectors is indirect and incomplete, and estimates are a mere proxy of the actual figures and may contain a relatively high margin of error. Nonetheless, the importance of these assets within the whole infrastructure is rather small, as they only amount to approximately 6% of the total stock, and 8% of the total investment during the period under study.
94
ALFONSO HERRANZ-LONCA´N
By contrast, for all other infrastructure the available information is acceptable, according to the usual standards of historical statistics, and the most important potential biases would be associated with the assumptions that have been applied to carry out the estimation process. Potential biases of this kind may be classified into three categories: (i) those that result from the deflation procedure; (ii) those that are a consequence of the individual assumptions that are applied in the estimation of sectoral series; and (iii) those that are associated with the general assumptions that were established in the application of the perpetual inventory method. Regarding the first of these three potential sources of bias, the use of an average deflator that is the same for all series removes the ‘‘index number problem’’ associated with the combination of several sectoral series into an aggregate one.27 However, as single price indices for each type of infrastructure are missing, the mismatch between the average deflator and the evolution of the price of each individual asset may have introduced biases into the final figures of unknown magnitude and sign. In fact, as far as the deflator that has been used is an average, biases may be assumed to cancel each other in the aggregate series. However, they may have reduced the accuracy of the individual estimates. The second category of problems is associated with the assumptions that have been applied to estimate each individual series. For example, in those cases in which no investment data were available, physical indicators have been accepted as proxies for the evolution of the gross stock and, as a consequence, the potential increases in the technical complexity of the assets or in the quality of materials that were used in its construction may have introduced downward biases in the growth rate of the series. By contrast, upward biases may result from increases in the productivity of construction of each network, and can also arise as a consequence of the poor conservation of the assets. Although changes in quality and in the technical characteristics of infrastructure have been allowed for as far as possible, sometimes the lack of information has prevented adjustments, and some individual series may contain biases of unknown sign and magnitude. Nevertheless, as in the case of the deflation process, although those biases may be important for the sectoral series, they are probably much less relevant in the aggregate figures. Finally, it is also necessary to analyze the consequences of the basic assumptions associated to the application of the perpetual inventory method and, more concretely, those regarding the useful lives and the efficiency
95
The Spanish Infrastructure Stock
decay of the assets. As far as the first of these two aspects is concerned, I have tried to keep the same useful lives as in other historical analyses of infrastructure. However, the accuracy of this choice is uncertain, since there is a total lack of information about the retirement and renewal practices that were followed at the time. In order to illustrate the size of the potential biases that would be associated with this problem, Fig. 3 shows the difference between my infrastructure stock series and the estimates that would result from applying a homogeneous useful life of 50 years for all assets. This useful life figure has often been assumed for buildings and structures in capital stock estimates.28 The gap between both series is rather small, amounting, on average, to 3% in 1844–1890 and to 12% in 1890–1935. Accordingly, the influence of the useful life assumptions on the final series seems to be rather low, something that may be explained by the fact that estimates in Fig. 3 are not based upon the pure application of the perpetual inventory method to investment series, but on the combination of investment data with physical indicators of the evolution of the stock.
10000 Useful lives of Table 1
Million of 1890 pesetas
Useful life of 50 years
1000
1929
1934
1919
1924
1914
1904
1909
1899
1894
1884
1889
1879
1869
1874
1864
1859
1854
1849
1844
100
Fig. 3. The Spanish Infrastructure Stock: Alternative Estimates under Different Assumptions on the Useful Life of the Assets. Source: My own figures.
ALFONSO HERRANZ-LONCA´N
96
Figure 4 compares the infrastructure stock series that has been estimated under the assumption of a geometric depreciation pattern (with d ¼ 0:91=T), against the series that would result from the assumption of a straight-line process. Gaps in the series are negligible, due to the similarity of these two depreciation patterns in the case of non-residential structures. The average difference is 1.4% and the maximum gap is only 2.9%. To summarize, it may be useful to remember Charles Feinstein’s comments on his own estimation of the British capital stock in the late nineteenth century. He suggested that a 10% average margin of error might be assumed in his aggregate series. In the case of the sectoral estimates, this bias should be increased to 25%.29 Given the differences between the UK and Spain as far as the quality of the available information is concerned, and given the outcomes of the different sensitivity analyses that have been carried out in this section, the infrastructure series that are presented here should be allowed, at least, the same margin of error as the British ones. In fact, it would be reasonable to assume larger biases in the five series for which information is scarcer and of poorer quality (non-public railroads, 10000 Modified geometric depreciation
Million of 1890 pesetas
Linear depreciation
1000
1844 1849 1854 1859 1864 1869 1874 1879 1884 1889 1894 1899 1904 1909 1914 1919 1924 1929 1934
100
Fig. 4. The Spanish Infrastructure Stock: Alternative Estimates under Different Assumptions on the Efficiency Decay Pattern. Source: My own figures.
The Spanish Infrastructure Stock
97
local roads, telegraph networks, gas distribution infrastructure and canals). It is necessary, therefore, to insist that the figures that are reproduced in the appendix must be considered carefully. As Feinstein himself stated, it is necessary to warn the researcher against the spurious air of precision conveyed by long runs of estimates systematically arrayed in neat tables.30 Keeping this caveat in mind, the next section describes the main features of the process of construction of Spanish infrastructure before the Civil War that may be inferred from the new series.
4. THE EVOLUTION OF SPANISH INFRASTRUCTURE BEFORE THE CIVIL WAR 4.1. Infrastructure Investment, 1845–1935 As may be observed in Fig. 1, Spanish investment in infrastructure showed an underlying positive growth trend, as well as very intense fluctuations throughout the period under study. A complete understanding of the dynamics of the series requires an analysis of any possible structural breaks that could have provoked a movement or alteration in the growth trend. For that purpose, the Vogelsang test has been applied to the series. This aims to contrast the existence of one-time breaks in time series in the presence of serial correlation, regardless of whether a unit root or a linear trend is present or not.31 I have applied the version of the test that was developed by Ben-David and Papell (2000) for finite series that are not I(1), because it allows testing for the presence of more than one break in the variables. The use of that version is justified in this case because, according to the ADF and the Phillips-Perron tests, the infrastructure investment series does not contain a unit root (see Table 2). The presence of a structural break has been contrasted for all years between 1850 and 1930, after excluding the extreme points of the series according to the test specification. The test statistic for the rejection of the null hypothesis of no structural break reaches its maximum F-statistic in 1854, at a level of 16.60, which is lower than the critical value of the test (17.85 at the 5% significance level).32 Therefore, Spanish investment in infrastructure can be accepted to have followed a constant structural growth trend with no significant structural breaks during the country’s first long wave of industrialization. The lack of structural breaks is a common characteristic of most Spanish economic variables between the middle of the nineteenth century and the Civil War. In fact, the structural stability of investment in infrastructure
ALFONSO HERRANZ-LONCA´N
98
Table 2.
Unit Root Tests. Gross Investment in Infrastructure (1845–1935).
ADF t-stat. PP t-stat.
4.21 3.00
H0: Presence of a unit root. Rejection of the null hypothesis at the 5% significance level. Rejection of the null hypothesis at the 1% significance level.
reinforces the impression of continuity of that long historical period, in spite of the violent fluctuations that took place in the meantime.33 In the case of the new infrastructure series, neither the profound changes in its composition that are described below, nor the substantial cyclical instability that may be seen in Fig. 1, were able to alter its structural growth trend. Table 3 presents the composition of infrastructure investment, and Figs. 5 and 6 show the percentage that gross investment in infrastructure accounted for within the GDP and total investment. This information provides a complete characterization of the process of infrastructure construction in Spain during the years under analysis. The most striking feature of the graphs is the abnormally high investment rates of the years 1855–1866, during which railroads were predominant within total investment. Those years correspond to the earliest and most intense Spanish railroad construction mania, which coincided in Spain with a substantial growth of the State’s capital formation in other infrastructure and, particularly, in the road network, ports and the telegraph system. The violence of the cycle of 1855–1866 is illustrated by the fact that the maximum level of investment, which was reached in 1862, was not regained in absolute terms until six decades later, and the maximum ratios between infrastructure investment and the GDP (3.9%), and between infrastructure investment and total investment (42.2%) were probably never reached again. Nevertheless, those percentages were not abnormal in the international context. Situations in which railroad investment accounted for 3% or more of GDP and 30% or more of total investment may be found, for instance, in Britain in 1847, the US in 1854, Germany in the 1850s and the 1860s, Hungary in the late 1860s and early 1870s or Sweden in the 1870s.34 Railroad investment manias were a widespread phenomenon in the Western economies at the time, and the Spanish case fits perfectly with other such experiences.35 The railroad mania ended in a deep investment crisis that started in 1866. Infrastructure construction resumed by the mid-1870s at much lower rates.
Railroads
Spanish Gross Investment in Infrastructure (1845–1935).
Urban Transport
Roads
Ports
Telecommunications
Energy distribution
Hydraulic Works
Total
(A) Composition of investment (annual average, %) 1845/55 1856/65 1866/75 1876/85 1886/95 1896/1905 1906/15 1916/25 1926/35
43.71 74.92 54.72 65.39 63.08 44.49 31.67 21.05 15.88
0.00 0.00 0.00 0.33 0.98 7.27 3.73 11.70 4.56
40.42 17.23 32.73 28.05 21.45 21.09 21.46 18.54 30.92
3.51 4.00 9.11 2.00 7.82 15.26 16.90 7.17 8.74
0.16 0.52 0.27 0.49 1.48 0.54 2.12 3.52 9.53
0.57 0.20 1.10 1.81 2.33 5.86 14.45 29.64 18.38
11.63 3.14 2.07 1.93 2.85 5.49 9.66 8.38 12.00
100 100 100 100 100 100 100 100 100
1845/1895 1895/1935
64.67 28.57
0.15 5.90
25.76 24.05
4.20 10.50
0.67 5.09
1.06 16.85
3.50 9.02
100 100
1845/1935
39.77
4.10
24.65
8.54
3.71
11.92
7.31
100
2.75 3.59
2.78 1.12
8.46
8.27 8.00
1.31 6.17
3.33 3.36
1.81
3.70
7.94
3.25
2.33
The Spanish Infrastructure Stock
Table 3.
(B) Yearly growth rate of investment (%) 1845/1890 1890/1935
4.23 1.64
1845/1935
0.68
5.20
99
Note: Investment growth is expressed in annual accumulative rates, adjusted to a log trend. As a consequence, rates for the whole period are not the average of those for shorter periods. Sources: see the appendix.
ALFONSO HERRANZ-LONCA´N
100 4 3.5 3
%
2.5 2 1.5 1 0.5
Fig. 5.
1935
1930
1925
1920
1915
1910
1905
1900
1895
1890
1885
1880
1875
1870
1865
1860
1855
1850
0
Spanish Gross Infrastructure Investment/GDP (1850–1935). Source: Prados de la Escosura (2003) and my own estimates.
Between 1875 and 1935, the average ratios between infrastructure investment and GDP (1.2%) and total investment (15.8%), were similar or slightly lower than figures available for other economies during the period. For instance, investment in infrastructure was 2.7% of GDP in Italy in 1890–1935, 2.2% in the UK in 1830–1913, 2.0% in Germany in 1850– 1913, 1.9% in France in 1848–1913 and 1.2% in the Netherlands in 1800– 1913.36 As for investment in infrastructure as a percentage of total investment, it was, for instance, 15.2% in Germany in 1850–1913, 14.4% in France in 1848–1913 and 22.1% in Italy in 1890–1935.37 Several periods may be distinguished in the evolution of investment in infrastructure between 1875 and the outbreak of the Spanish Civil War. As may be observed in Table 3, the most important threshold is to be found around 1895. Before that year, the main feature of Spanish investment in infrastructure was the absolute prominence of railroads, since they absorbed nearly two-thirds of the resources that were invested in the improvement and growth of Spanish infrastructure stock. By contrast, with the exception of the road network, investment in other types of infrastructure received extremely small proportions of the total resources. This situation gradually changed after the mid-1890s, when a process of intense diversification of investment started. The reduction in the importance of railroads firstly
101
The Spanish Infrastructure Stock 45 40 35 30
%
25 20 15 10 5
Fig.
1935
1930
1925
1920
1915
1910
1905
1900
1895
1890
1885
1880
1875
1870
1865
1860
1855
1850
0
6. Spanish Gross Infrastructure Investment/Gross Total Investment (1850–1935). Source: Prados de la Escosura (2003) and my own figures.
benefited ports and urban and suburban transport, although the shares of those assets in total investment were later taken over by irrigation infrastructure, the telephone system and, especially, energy distribution networks.38 These changes in composition had a substantial impact on the fluctuations of investment in infrastructure. This impact has been analyzed by isolating the cyclical component of the series through the application of the Hodrick–Prescott Filter to its logarithm, and by comparing it with the cyclical component of Spanish GDP and industrial production during the period 1850–1935. Figures 7 and 8 show the 20-years moving correlation coefficients between the cyclical components of the two output variables and that of investment in infrastructure one year before, over the same year and one year later. According to the graphs, the association between the cyclical component of investment in infrastructure and production is always higher when the infrastructure variable is taken one year later than the output variable. The strong association between the main variables of the economy and gross investment in infrastructure in the following year indicates that fluctuations of the latter followed the movements of the rest of the economy with some time lag. Apparently, during the period of study, episodes of high economic growth fostered investment in infrastructure. This situation might have been
ALFONSO HERRANZ-LONCA´N
102
Comov. with inf.(-1) Comov. with inf. Comov. with inf.(+1) 0.8
0.6
0.4
0.2
0
Fig. 7.
1935
1930
1925
1920
1915
1910
1905
1900
1895
1890
1885
1880
1875
1870
1865
1860
1855
1850
-0.2
Comovements of GDP and Infrastructure Investment (1850–1935). Source: Prados de la Escosura (2003) and my own figures.
the result of Wagner’s Law, i.e. the impact on infrastructure of the increase in the financial capacity of investors (and, especially, of the public sector) due to economic growth, and also might reflect reactions to the upsurge of bottlenecks, resulting as well from the growth of the economy. The described relationship was especially intense from the 1890s onwards. By contrast, investment in infrastructure was much more independent of the fluctuations of the rest of the economy between the late 1870s and the early 1890s. This would be consistent with the extreme importance of railroad investment during those years. Unlike other sorts of infrastructure, railroads were large-scale projects of inter-regional scope and, therefore, their degree of accommodation to the short-term fluctuations of the economy was relatively low. The construction of railroads took a relatively long time, and
103
The Spanish Infrastructure Stock
Comov with inf.(-1) Comov. with inf. Comov. with inf.(+1) 0.8
0.6
0.4
0.2
0
1935
1930
1925
1920
1915
1910
1905
1900
1895
1885
1890
1880
1875
1870
1865
1860
1855
1850
-0.2
Fig. 8. Comovements of Industrial Production and Infrastructure Investment (1850–1935). Source: Prados de la Escosura (2003) and my own figures.
the bottlenecks that they were intended to break up had a much more structural and long-term nature than in the case of other types of infrastructure. As a consequence, railroad construction cycles were to some extent self-sustained and did not adapt well to the global fluctuations of the economy. This peculiar behavior of railroad investment fluctuations has been pointed out for numerous countries, and has been explained on the basis of technology, but also because of political factors and the railroad companies’ strategic behavior within the railroad oligopoly.39 By contrast, during the first-third of the twentieth century, the change in the composition of investment in infrastructure and the decline in the importance of the railroads meant a better adaptation of infrastructure capital formation to the immediate needs of the economy. After the mid-1890s, the amounts involved in each project were much smaller, and the link between
104
ALFONSO HERRANZ-LONCA´N
infrastructure assets and production activities was much closer than in the previous period. The diversification of investment from the 1890s onwards was the consequence of several factors, such as: the virtual completion of the railroad network;40 the Spanish State’s growing economic activity as reflected in the increase of public work construction; and a number of technological changes, such as technical advances in the long-distance transmission of electricity, or in telecommunications. In fact, the maximum development of these three aspects arrived in the last investment cycle of the period under study, i.e. under Primo de Rivera’s Dictatorship and the Second Republic (1923– 1936). During those years, diversification of investment reached its zenith, both in the case of public capital formation (with increasing resources devoted to roads, ports and hydraulic works) and private investment (in particular energy distribution, urban transport and telecommunications). The coincidence of the State’s activism with the dynamism of private investment resulted in very high percentages of the GDP being spent on infrastructure capital formation, up to a level of 2.5% in 1929. The exceedingly high volume of resources invested in infrastructure during the 1920s may explain a unique feature of the cyclical behavior of gross investment in infrastructure during those decades. As may be seen in Figs. 7 and 8, the correlation between production and lagged investment in infrastructure before the 1920s was rarely significant, which might be interpreted as evidence of the minor influence of the elimination of bottlenecks by infrastructure, and of the irrelevance of its short-term ‘‘backward effects’’ on the evolution of the Spanish economy. However, during the 1920s and 1930s, fluctuations of infrastructure preceded fluctuations of industrial production. This change would be consistent with the emphasis of some historians on the importance of public investment in the evolution of the Spanish industrial sector during those years. Jordi Palafox, for instance, has indicated that the military regime established in the country in 1923 was very sensitive to the interests of heavy industry, which had suffered an intense oligopolization process in the previous period. As a consequence, during the dictatorship of 1923–1930, public investment in certain sorts of infrastructure (such as the railroad and road systems and hydraulic works) was four times as large as in the years between 1917 and 1923, and became essential for the growth of the machinery, steel and concrete industries, for example. These would otherwise have stagnated during the period, since the deep structural problems of the Spanish agrarian sector substantially reduced the growth prospects of domestic markets.41
The Spanish Infrastructure Stock
105
4.2. The Spanish Infrastructure Stock As a consequence of the process of investment that has just been described, the Spanish endowment of infrastructure experienced a substantial growth during the period under analysis, which can be seen in Fig. 2. The infrastructure stock grew at a yearly rate of 3.3% and, as a result, in 1935 it was 32 times larger than it had been in 1845. Figure 2 also shows that periods of intense growth of the stock, which were associated with episodes of high investment (such as the railroad mania of 1855–1866, or the 1920s), alternated with long periods of slow growth. Table 4 shows the changes in the composition of Spanish infrastructure stock between 1845 and 1935. These changes are the direct result of the evolution of investment. Concretely, they reflect the concentration of investment efforts in railroads and roads up to 1895. By contrast, after that year, and due to the diversification of investment in infrastructure, railroads gradually lost importance within the total stock, to the advantage of ports, energy distribution networks, hydraulic works, telecommunications and urban transport, which had absorbed negligible amounts of resources during the second half of the nineteenth century. Figure 9 shows the evolution of the ratio between infrastructure stock and the total Spanish capital stock during the period under study. It clearly shows the effects of the huge investment efforts of the years between 1855 and 1866, which increased the percentage from around 18% to more than 25%. Later on, this percentage started to decrease, in a process that would continue until the Civil War. At the beginning, the reduction in the importance of infrastructure within the whole Spanish capital stock that took place after 1866 just reflected the return to normality after the excess investment of the railroad mania. However, once the stock reached the percentage level of the years before 1855 the decline did not stop. In fact, from the final years of the nineteenth century onwards, the reduction in the ratio seems to be a consequence of the diversification of the Spanish total investment and the increasing use of machinery and equipment in a growing number of sectors. From a comparative point of view, the percentages that are reproduced in Fig. 9 are similar to those of other economies during the period, such as Germany (between 1850 and 1930) or Japan and the Soviet Union (in the Inter-war period), where they fluctuated between 14 and 20%. By contrast, the Spanish percentages were lower than those of the UK (20–25%) or Italy (26–29%) during the same period.42
106
Table 4. Railroads
Urban Transport
Spanish Net Infrastructure Stock (1845–1935).
Roads
Ports
Telecommunications
Energy Distribution
Hydraulic Works
Total
(A) Composition of the stock (%) 1.16 26.75 63.18 60.53 62.71 63.02 58.58 51.36 42.98 33.41
0.00 0.00 0.00 0.00 0.08 0.35 1.68 1.77 4.10 3.63
82.17 57.20 26.92 28.76 28.22 26.09 25.65 25.51 24.81 28.31
5.79 4.40 4.20 5.47 4.28 5.52 7.89 10.51 9.98 10.04
0.00 0.10 0.38 0.31 0.32 0.55 0.47 0.78 1.31 4.18
0.10 0.34 0.20 0.35 0.73 1.08 1.80 4.50 10.37 11.40
10.78 11.21 5.13 4.58 3.66 3.38 3.93 5.57 6.46 9.02
100 100 100 100 100 100 100 100 100 100
3.35 1.81
5.66 3.38
6.80
9.35 8.67
2.60 3.97
5.82 1.87
2.32
4.39
8.45
2.73
3.25
(B) Yearly growth rates of the stock (%) 1845/95 1895/1935
11.14 0.37
1845/1935
4.25
7.91
Note: Stock growth is expressed in annual accumulative rates, adjusted to a log trend. As a consequence, rates for the whole period are not the same as the average of those for shorter periods. Sources: see the appendix.
ALFONSO HERRANZ-LONCA´N
1845 1855 1865 1875 1885 1895 1905 1915 1925 1935
107
The Spanish Infrastructure Stock 28 26 24 22
%
20 18 16 14 12
Fig. 9.
1934
1930
1922
1926
1914
1918
1906
1910
1902
1894
1898
1890
1886
1878
1882
1870
1874
1866
1862
1858
1854
1850
10
Spanish Infrastructure Stock/Total Capital Stock (1850–1935). Source: Prados de la Escosura and Rose´s (2003) and my own figures.
The relationship between infrastructure stock and GDP provides a first approximation to the relative abundance or shortage of infrastructure in Spain during the period under study. The evolution of this percentage is reported in Fig. 10, which shows its gradual growth from levels of about 10% in the mid-nineteenth century to levels above 40% on the eve of the Civil War of 1936. However, this increase was not a steady one. As could be expected, growth was steep during the years of the railroad mania, reaching levels close to 35%, but it was abruptly interrupted in 1866, and from this date to the mid-1890s, percentages fluctuated between levels of 20–35%. Growth only resumed in the last decades of the period, and during the 1930s the ratio reached levels of over 40%. How do these percentages look in an international context? It is difficult to answer this question, since the comparison among historical infrastructure endowments of different countries is challenging. This is due to the extreme paucity of infrastructure stock estimates, and the fact that the available figures are hardly comparable because of differences in definitions and in estimation techniques. In this context, Table 5 is just a preliminary attempt to carry out that comparison for a few economies for which appropriate information is available.
ALFONSO HERRANZ-LONCA´N
108 50 45 40 35
%
30 25 20 15 10 5
Fig. 10.
1935
1930
1925
1920
1915
1910
1905
1900
1895
1890
1885
1880
1875
1870
1865
1860
1855
1850
0
Spanish Net Infrastructure Stock/GDP (1850–1935). Source: Prados de la Escosura and Rose´s (2003) and my own figures.
According to the table, the endowment of infrastructure in industrialized countries stood between 30 and 50% of GDP during the second half of the nineteenth century and the first decades of the twentieth century, although, in a few cases, such as Italy and US, it might have reached even higher levels. In this context, the growth of the Spanish infrastructure as a percentage of GDP constituted a process of convergence with the situation in more advanced economies. However, that convergence process, as has been said, was not continuous, but experienced a one-time boost in 1855–1866 and, later on, a long stagnation until the last few years of the nineteenth century, when it gradually resumed. As a result of this convergence process, by the end of 1935 Spain had an infrastructure endowment that was acceptable from a comparative point of view, and had been obtained through several decades of substantial investment efforts. The country, however, still suffered from serious shortages in certain areas. For instance, it still lacked a national electricity network, suffered from a serious scarcity of secondary roads, and the density of its railroad network was rather low in the European context.43 However, its endowment of social fixed capital was relatively acceptable in many other areas. Unfortunately, the Civil War and the post-war period introduced a violent break in the previous convergence trend, which would open up a
109
The Spanish Infrastructure Stock
Table 5.
Spain Netherlands UK Japan Italy USSR US
Net Infrastructure Stock/GDP in Different Countries (1850–1930, %).
1850
1860
1870
1880
1890
1900
1910
1920
1930
10.33 29.11 43.50 na na na na
18.65 29.34 44.90 na na na na
34.07 37.86 46.30 na na na na
24.75 38.72 47.94 na na na na
26.61 39.32 49.33 19.41 71.63 na na
38.31 31.72 46.74 25.26 74.68 na na
36.06 30.27 48.21 30.51 72.21 na na
33.68 na 44.89 32.50 64.40 na na
37.80 na 38.93 42.75 65.55 17.15a 68.62b
Notes: na: not available. Sources: Spain: GDP from Prados de la Escosura (2003) and my own infrastructure stock figures. Netherlands: Groote (1996), Maddison (1995) and Centraal Bureau voor de Statistiek (1994). UK: Capital stock from Feinstein (1965), (1972) and (1988), and GDP from Deane (1968) and Maddison (1995). Japan: Capital stock from Ohkawa et al. (1966) and GDP from Ohkawa et al. (1974). Italy: Rossi et al. (1993). US: Bureau of Economic Analysis, in http:// www.bea.doc.gov. USSR: Moorsteen and Powell (1966, p. 50). a In 1928 b In 1929
huge gap between Spain and the advanced economies. Afterwards, several decades would again be necessary to reduce the distance.
5. CONCLUSIONS This paper presents new estimates of Spanish infrastructure stock and investment for the period 1845–1935. The new series are intended to provide quantitative information about one of the key factors in Spanish economic growth during the first stages of industrialization. Although some measurement of the public capital stock had already been undertaken by other scholars, this research is the first systematic attempt to estimate the value of the entire Spanish infrastructure. A variety of sources and techniques has been used in the estimation, and the new series may be accepted as reasonably reliable to the standards of historical statistics, with the possible exceptions of some minor infrastructure sectors, such as non-public railroads, the telegraph and gas distribution networks, the stock of Spanish canals and urban infrastructure. Thanks to the new series, a nearly complete picture of the process of development of Spanish infrastructure between the mid-nineteenth century and the present is available.
ALFONSO HERRANZ-LONCA´N
110
The new quantitative evidence reflects a constant and sustained effort of improvement and enlargement of the Spanish infrastructure endowment throughout the 1845–1935 period. Although investment in infrastructure was characterized by intense fluctuations, especially during the early and final years of the period under study, it is not possible to confirm the presence of structural breaks in the series that could have interrupted or substantially altered its long-term growth. Broadly speaking, two distinct periods may be distinguished in the evolution of Spanish infrastructure. On the one hand, the second half of the nineteenth century was characterized by the prominent role of railroads, which accounted for exceedingly high shares of total infrastructure investment. On the other, from the last few years of the nineteenth century onwards, railroad capital formation stagnated due to the virtual completion of the main network, and the majority of investment in infrastructure was then addressed to other assets such as roads, ports, electricity distribution and hydraulic works. The new series allow a very preliminary comparison of Spanish infrastructure endowment with that of other advanced economies of the time. Between the mid-nineteenth century and the eve of Spain’s Civil War, the ratio between Spanish infrastructure stock and GDP undertook a substantial growth process, and converged with the levels experienced in other industrialized countries. That convergence, however, was highly concentrated in certain periods, such as the railroad mania of 1855–1866, and the Inter-war years. The next question to answer in this context is to what extent the relative shortage of infrastructure during most of the period under study might have been one of the factors to blame for Spain’s lack of convergence before the Civil War. Although this problem has been considered by historians for decades, the new series allow for a more systematic approach and for the application of econometric techniques to address the issue. This task will indeed be undertaken in further research.
NOTES 1. The only serious attempt to endow the country with an extensive network of highways and canals took place in the last half of the eighteenth century. However, by 1840 the projected radial highway system was still incomplete and none of the critical junctions of the projected canals had been finished. On this subject, see Ringrose (1970, pp. 14–17) and also Madrazo (1984, pp. 162–167). 2. See, for example, Diamond and Spence (1989); Batten and Karlsson (1996); Holtz-Eakin and Lovely (1996); and Fujita, Krugman, and Venables (2000).
The Spanish Infrastructure Stock
111
3. According to Maddison’s international database, the Spanish GDP per capita was around 60% of the average income per capita of the UK, France and Germany, both by 1850 and 1929; see Maddison (1995). Recently, Prados de la Escosura has offered an image of the Spanish relative income per capita, which is slightly different than Maddison’s, showing a slow long-term decline from a higher starting point; see Prados de la Escosura (2003, pp. 179–181). On the high degree of uncertainty regarding the level of Spanish income per capita in the mid-nineteenth century, see Reis (2000). Nevertheless, this author’s recent alternative estimate for 1850 would not change the picture that is offered by Maddison’s data very much. 4. See the classical works by Tortella (1972, pp 118–121) and Nadal (1976, pp. 551–553) and recent research by Comı´ n, Martı´ n Acen˜a, Mun˜oz, and Vidal (1998, Vol. 1, pp. 140–141). 5. See, especially, Go´mez Mendoza (1983). 6. See, for instance, the proceedings of the three Conferences on Spanish Railroad History that have taken place in the last few years (Alicante, 1998; Aranjuez, 2001; & Gijo´n, 2003). A survey of the research that was carried out on the Spanish railway system up to 1998 can be seen in Go´mez Mendoza (1998). 7. The main references are shortly surveyed in Herranz-Lonca´n (2002, Chapter 1). 8. See, for instance Groote (1996, pp. 22–26). As a consequence of these definition problems, some authors renounce to define infrastructure, and describe it as, ‘‘what most people consider it to be’’; Button (1996, p. 148). Definition problems are probably associated with the fact that infrastructure does not correspond to any of the usual national account categories; within the UN System of National Accounts, the closest category would be ‘‘Other buildings and structures’’, but this also includes the buildings devoted to industrial and commercial uses, which cannot be considered as infrastructure; see United Nations (1993). 9. Hirschman (1958, p. 83); Stern (1991, p. 128). 10. Diewert (1986, pp. 3–4). 11. See a discussion on this issue in Button (1996, pp. 148–151). 12. See, for instance, Batten (1990, p. 88). 13. See, for example, Hansen (1965, pp. 7–12) or Aschauer (1989, pp. 193–194) and, for the Spanish case during the last few decades, Mas, Maudos, Pe´rez, and Uriel (1996, p. 647). 14. Hulten and Schwab (1993, pp. 271–272); Gramlich (1994, p. 1177). 15. Mas Pe´rez and Uriel (1998). 16. Mas et al. (1995, vol. 4). 17. Cucarella (1999). 18. The process of estimation of gross investment figures on the basis of physical indicators is described by Ohkawa, Ishiwate, Yamada, and Ishi (1966, p. 135) and Groote (1996, p. 95). A similar procedure for machinery is applied by De Long and Summers (1994, pp. 13–14). The definition of ‘‘new’’ investment that is reproduced in the text comes from Feinstein and Pollard (1988, p. 2). 19. The data sources can also be seen in the Appendix. Actually, in a lot of cases data search was facilitated to a great extent by the pioneering research efforts carried out by Go´mez Mendoza in the field of transport and communications; see Go´mez Mendoza (1989b). 20. Prados de la Escosura (2003).
112
ALFONSO HERRANZ-LONCA´N
21. Escriba´-Pe´rez and Ruiz-Tamarit (1995). Actually, as has been stressed by Hulten (1990, p. 127), a more rigorous analysis of the capital stock should consider the coefficients f of efficiency decay as endogenous. 22. See, for instance, Madrazo (1984, p. 235), or Go´mez Mendoza (1989a, p. 35). 23. Hulten (1990, pp. 124–125, 142). 24. The IVIE estimates are available in Mas et al. (1998). They include the following infrastructure: railroads, roads and highways, ports, airports, hydraulic works and urban infrastructure. Their coverage is therefore different from my figures, which also include energy distribution networks and telecommunications, but exclude airports. The IVIE figures have been expressed in pesetas of 1890 by applying Prados de la Escosura’s deflator for ‘‘Other construction’’ investment, which has been taken from Prados de la Escosura (2003). 25. The importance of the periods 1855–1866 and 1922–1929 was already stressed, from the viewpoint of total capital formation, by Carreras (1990, pp. 124–126). 26. Yearly figures of State non-military investment in air transport since the beginning of the twentieth century are available in Mas et al. (1995). However, two reasons prevent us from using this data to estimate infrastructure stock figures. Firstly, a large share of the Spanish airport investment was not financed by the State before 1936. And, secondly, the available data includes investment in land, which accounted for a very large share of total airport investment during the first stages of the history of air transport, and must not be included in infrastructure. On these issues, see AENA (1996). Nevertheless, the exclusion of those assets from the series is not very important. For instance, for the sake of illustration, if the value of the Stateowned airport assets (including land) between 1928 and 1935 is calculated, it amounts to just 0.15% of the total infrastructure stock. 27. As has already been indicated, the deflator that has been used in all cases is the series that was recently estimated by Prados de la Escosura (2003) for investment in ‘‘Other construction’’. 28. For instance, it has been used to obtain Spanish capital stock estimates by Cubel and Palafox (1997) and Prados de la Escosura and Rose´s (2003). 29. Feinstein (1988, p. 264). Similar margins of error are assumed by Groote in his estimation of the Dutch infrastructure stock between 1800 and 1913; see Groote (1996, p. 49). 30. Ibidem. 31. Vogelsang (1997). 32. The absence of a structural break around 1854 may be surprising, given the intensity of the fluctuations of the railroad era, which are dealt with in more detail below. However, if Fig. 1 is examined, the overall long-term structural trend of investment does not seem to have been altered by the railroad mania. Nevertheless, the proximity of the starting point of the series prevents from drawing a final conclusion on this issue. 33. Cubel and Palafox (1998) have searched for the presence of structural breaks before 1936 in the series of Spanish GDP, industrial production and investment, with no positive results. Pons and Tirado (2001) have analyzed Spanish GDP and GDP per capita in 1870–1994, and the earliest structural break they have found is in 1935, which is obviously associated with the impact of the Civil War. Finally, the lack of
The Spanish Infrastructure Stock
113
structural breaks in the series of Spanish GDP and GDP per capita before 1936 has recently been confirmed by Prados de la Escosura (2003, pp. 145–146). 34. For the UK and the US, see Mitchell (1964), Feinstein (1972, p. 40), and O’Brien (1977, p. 55); for Germany, Fremdling (1983, p. 124), and Tilly (1978, p. 414); for Hungary, Katus (1983, p. 191); for Sweden, Hedin (1967, p. 11), and Holgersson and Nicander (1968, p. 5). 35. See, for instance, Fishlow (1965, pp. 105–106), or O’Brien (1977, p. 57). 36. Groote (1996, pp. 76, 85), except for the Italian figure, which has been calculated on the basis of Rossi, Sorgato, and Toniolo (1993). 37. These percentages have been calculated from Hoffmann (1965), Le´vy-Leboyer and Bourguignon (1990) and Rossi et al. (1993). 38. Actually, production and distribution of electricity constituted the most important destination of capital in Spain during the first-third of the twentieth century, and the paid-up capital of the electricity companies reached the same level as the capital of the railroad companies by 1921; see Bartolome´ (1995, p. 109). 39. On the independent character of railroad investment fluctuations see, for Britain, Kenwood (1965, pp. 314–319), and Hawke (1970, pp. 363–379), for the US, Fishlow (1965, p. 179), for France, Caron (1983, p. 35), for Sweden, Hedin (1967, pp. 10–11), and, for Hungary, Katus (1983, p. 191). The role of political factors on this independent behavior has been stressed by Fenoaltea (1983, pp. 53–54), and Le´vy-Leboyer (1978, pp. 249–250), and the importance of the railroad companies’ strategic behavior within the railroad oligopoly, has been highlighted by Harley (1982, p. 797). 40. In 1895, 86% of the pre-Civil War length of the main (broad gauge) railroad network had already been open to public service. 41. Palafox (1980, pp. 23–33); see also Palafox (1991). However, unlike this author, Comı´ n and Martı´ n Acen˜a (1984, pp. 249–258), have pointed out that the small size of the public sector during the 1920s and 1930s prevented it from being decisive in the evolution of the Spanish industry. 42. These percentages have been calculated from the following sources: for Germany, Hoffmann (1965), for Japan, Ohkawa et al. (1966), for the Soviet Union, Moorsteen and Powell (1966, p. 50), for UK, Feinstein (1965), (1972),(1988), and, for Italy, Rossi et al. (1993). The German percentage refers to capital figures in gross terms. 43. Herranz-Lonca´n (2002), Chapter 4 and, on the absence of a national electricity distribution network, Bartolome´ (2003).
ACKNOWLEDGMENTS I wish to thank Dudley Baines, Isabel Bartolome´, A´ngel Calvo, Albert Carreras, Nicholas F. R. Crafts, Eloy Ferna´ndez Clemente, Pedro Pablo Ortu´n˜ez, Pere Pascual, Leandro Prados de la Escosura, Joan Ramon Rose´s, Carles Sudria` and an anonymous referee for their generous help and
114
ALFONSO HERRANZ-LONCA´N
valuable contributions to this research. I gratefully acknowledge financial support by the British Council, the Economic and Social Research Council and the Spanish Banco de Espan˜a, Ministerio de Educacio´n y Cultura (BEC2002–00423) and Generalitat de Catalunya.
REFERENCE AENA. (1996). Los aeropuertos espan˜oles. Su historia, 1911–1996. Madrid: AENA. Alzola, P. (1884). Ferro-carriles de vı´ a ancha y de vı´ a estrecha. Revista de Obras Pu´blicas, 32–33. Alzola, P. (1979). Las obras pu´blicas en Espan˜a. Estudio histo´rico. Madrid: Colegio de Ingenieros de Caminos, Canales y Puertos. Aschauer, D. A. (1989). Is public expenditure productive? Journal of Monetary Economics, 23, 177–200. Bartolome´, I. (1995). Los lı´ mites de la hulla blanca en vı´ speras de la Guerra Civil: un ensayo de interpretacio´n. Revista de Historia Industrial, 7, 109–139. Bartolome´, I. (1999). La industria ele´ctrica espan˜ola antes de la guerra civil: Reconstruccio´n cuantitativa. Revista de Historia Industrial, 15, 139–160. Bartolome´, I. (2003). La industria ele´ctrica en Espan˜a (1880–1936): Tecnologı´a, recursos e instituciones. Doctoral thesis, European University Institute, Florence. Batten, D. F. (1990). Infrastructure as a network system: Mera revisited. In: L. Anselin & M. Madden (Eds), New directions in regional analysis. Integrated and multi-regional approaches (pp. 76–89). London: Belhaven. Batten, D. F., & Karlsson, C. (Eds) (1996). Infrastructure and the complexity of economic development. Berlin: Springer. Becerril, A. (1946). El proceso de amortizacio´n en la industria ele´ctrica espan˜ola. Moneda y cre´dito, 18, 36–46. Bello, S. (1914). Coste de las obras hidra´ulicas en Espan˜a. Zaragoza: G. Casan˜al. Ben-David, D., & Papell, D. H. (2000). Some evidence on the continuity of the growth process among the G7 countries. Economic Inquiry, 38(2), 320–330. Button, K. (1996). Ownership, investment and pricing of transport and communication infrastructure, In: D.F. Batten & C. Karlsson (Eds), Infrastructure and the complexity of economic development (pp. 147–165). Berlin: Springer. Caron, F. (1983). France, In: P. O’Brien, (Ed.), Railways and the economic growth of western Europe (pp. 28–48). London: McMillan. Carreras, A. (1983). La produccio´ industrial espanyola i italiana des de mitjan segle XIX fins a l’actualitat. Doctoral thesis, Universitat Auto`noma de Barcelona, Barcelona. Carreras, A. (1990). Industrializacio´n espan˜ola: estudios de historia cuantitativa. Madrid: EspasaCalpe. Ceballos, J. G. (1932). Economı´a, finanzas, cambios. La realidad econo´mica y financiera de Espan˜a en los 30 an˜os del presente siglo. Madrid: El Financiero. Centraal Bureau Voor de Statistiek (1994). 1899–1994. Vijfennegentig Jaren Statistiek in Tijdreeksen. CBS.
The Spanish Infrastructure Stock
115
Cercos, R. (1968). La riqueza portuaria. In: Universidad Comercial de Deusto (Ed.), Riqueza Nacional de Espan˜a (Vol. 2, pp. 559–624). Bilbao: Universidad Comercial de Deusto. Comı´ n, F., & Martı´ n Acen˜a, P. (1984). La polı´ tica monetaria y fiscal durante la Dictadura y la Segunda Repu´blica. Papeles de Economı´a Espan˜ola, 20, 236–265. Comı´ n, F., Martı´ n Acen˜a, P., Mun˜oz, M., & Vidal, J. (1998). 150 An˜os de Historia de los Ferrocarriles Espan˜oles. Madrid: Fundacio´n de los Ferrocarriles Espan˜oles. Costa, M. T. (1981). Iniciativas empresariales y capitales extranjeros en el sector servicios de la economı´ a espan˜ola durante la segunda mitad del siglo XIX. Investigaciones Econo´micas, 14, 45–83. Cubel, A., & Palafox, J. (1997). El stock de capital en la economı´ a espan˜ola, 1900–1958. Revista de Historia Industrial, 12, 113–126. Cubel, A., & Palafox, J. (1998). La continuidad del crecimiento econo´mico de Espan˜a. 1850–1936. Revista de Historia Econo´mica, 16(3), 619–643. Cucarella, V. (1999). El stock de capital ferroviario en Espan˜a y sus provincias, 1845–1997. In: M. Mas, F. Pe´rez & E. Uriel (Eds), El ‘‘stock’’ de capital ferroviario en Espan˜a y sus provincias: 1845–1997 (pp. 61–172). Bilbao: Fundacio´n BBV. Cunningham, B. (1914). The dock and harbour engineers reference book. London: Charles Griffin & Co. Ltd. Deane, P. (1968). New estimates of gross national product for the United Kingdom. Review of Income and Wealth, 14(2), 95–112. De Long, B., & Summers, L. H. (1994). How robust is the growth-machinery nexus? In: M. Baldassari, L. Paganetto & E. S. Phelps (Eds), International differences in growth rates, Market globalization and economic areas (pp. 5–54). New York: St Martin’s Press. De Urquijo, J. L. (1968). Patrimonio de la Direccio´n General de Correos y Telecomunicacio´n, In: Universidad Comercial de Deusto. (Ed.), Riqueza Nacional de Espan˜a (Vol. 2, pp. 689–722). Bilbao: Universidad Comercial de Deusto. Diamond, D., & Spence, N. (1989). Infrastructure and industrial costs in British industry. London: Her Majesty’s Stationery Office. Diewert, W. E. (1986). The measurement of the economic benefits of infrastructure services. Berlin: Springer. Echaide, I. M. (1929). Red telefo´nica de la Diputacio´n de Guipu´zcoa. Resumen descriptivo, histo´rico y estadı´stico para la Exposicio´n Ibero-Americana de Sevilla. San Sebastia´n: Diputacio´n de Guipu´zcoa. Escriba´-Pe´rez, F. J., & Ruiz-Tamarit, J. R. (1995). Economic measurement of capital and profitability. Recherches Economiques de Louvain, 61(4), 433–459. Feinstein, C. H. (1965). Domestic capital formation in the United Kingdom 1920–1938. Cambridge: Cambridge University Press. Feinstein, C. H. (1972). National income, expenditure and output of the United Kingdom 1855–1965. Cambridge: Cambridge University Press. Feinstein, C. H. (1988). Part II. National Statistics, 1760–1920. In: C.H. Feinstein, S. Pollard (Eds), Studies in capital formation in the United Kingdom (pp. 258–401). Oxford: Clarendon press. Feinstein, C. H., & Pollard, S. (Eds) (1988). Studies in capital formation in the United Kingdom (pp. 1750–1920). Oxford: Clarendon Press. Fenoaltea, S. (1983). Italy, In: P. O’Brien (Ed.), Railways and the economic growth of western Europe (pp. 48–120). London: McMillan.
116
ALFONSO HERRANZ-LONCA´N
Ferna´ndez Ordo´n˜ez, J. A. (1986). Cata´logo de treinta canales espan˜oles anteriores a 1900. Madrid: Colegio de Ingenieros de Caminos, Canales y Puertos. Fishlow, A. (1965). American railroads and the transformation of the ante-bellum economy. Cambridge (MA): Harvard University Press. Fremdling, R. (1983). Germany, In: P. O’Brien (Ed.), Railways and the economic growth of western Europe (pp. 121–147). London: McMillan. Fujita, M., Krugman, P., & Venables, A. J. (2000). The spatial economy: Cities, regions and international trade. Cambridge, MA: The MIT Press. Garrido, M. (1968). Embalses. In: Universidad Comercial de Deusto (Ed.), Riqueza Nacional de Espan˜a (Vol. 4, pp. 683–742). Bilbao: Universidad Comercial de Deusto. Gil Carretero, S. (1968). Los transportes urbanos colectivos. In: Universidad Comercial de Deusto (Ed.), Riqueza Nacional de Espan˜a (Vol. 2, pp. 453–477). Bilbao: Universidad Comercial de Deusto. Go´mez Mendoza, A. (1982). Ferrocarriles y cambio econo´mico en Espan˜a (1855–1913). Un enfoque de nueva historia econo´mica. Madrid: Alianza. Go´mez Mendoza, A. (1983). Spain, In: P. O’Brien (Ed.), Railways and the economic growth of western Europe (pp. 148–169). London: McMillan. Go´mez Mendoza, A. (1989a). Ferrocarril, industria y mercado en la modernizacio´n de Espan˜a. Madrid: Espasa Calpe. Go´mez Mendoza, A. (1989b). Transportes y comunicaciones. In: A. Carreras (Ed.), Estadı´sticas Histo´ricas de Espan˜a, siglos XIX-XX (pp. 269–323). Madrid: Fundacio´n Banco Exterior. Go´mez Mendoza, A. (1991). Las obras pu´blicas, 1850–1935. In: F. Comı´ n & P. Martı´ n Acen˜a (Eds), Historia de la empresa pu´blica en Espan˜a (pp. 177–204). Madrid: Espasa Calpe. Go´mez Mendoza, A. (1998). Along broad lines. The economic history of Spanish railways, 1973–96. Journal of Transport History, 19(1), 1–17. Go´mez-Santos, M. (1969). El metro de Madrid. Medio siglo al servicio de la ciudad 1919–1969. Madrid: ESCELICER. Gramlich, E. M. (1994). Infrastructure investment: A review essay. Journal of Economic Literature, 32(3), 1176–1196. Groote, P. (1996). Infrastructure and Dutch economic development. A new long run data set for the Netherlands 1800-1913. Utrecht: Koninklijk Nederlands Aardrijkskundig Genootschap. Hansen, N. M. (1965). Unbalanced growth and regional development. Western Economic Journal, 4, 3–14. Harley, C. K. (1982). Oligopoly agreement and the timing of American railroad construction. Journal of Economic History, 42(2), 797–823. Hawke, G. R. (1970). Railways and economic growth in England and Wales. Oxford: Clarendon Press. Hedin, L. E. (1967). Some notes on the financing of the Swedish railroads 1860–1914. Economy and History, 10, 3–37. Herna´ndez, T. M. (1983). Ferrocarriles y capitalismo en el Paı´s Valenciano. Valencia: Ayuntamiento de Valencia. Herranz-Lonca´n, A. (2002). Infrastructure and economic growth in Spain, 1845–1935. PhD thesis, London School of Economics and Political Science, London. Hirschman, A. O. (1958). The strategy of economic development. New Haven: Yale University Press. Hoffmann, W. G. (1965). Das Wachstum del Deutschen Wirtschaft seit der mitte des 19. Jahrhunderts. Berlin: Springer.
The Spanish Infrastructure Stock
117
Holgersson, B., & Nicander, E. (1968). The railroads and the economic development in Sweden during the 1870s. Economy and History, 11, 3–51. Holtz-Eakin, D., & Lovely, M. E. (1996). Scale economies, returns to variety and the productivity of public infrastructure. Regional Science and Urban Economics, 26(2), 105–123. Hulten, C. R. (1990). The measurement of capital. In: E. R. Berndt & J. E. Triplett (Eds), Fifty years of economic measurement. The jubilee of the conference on research in income and wealth (pp. 119–152). Chicago: University of Chicago. Hulten, C. R., & Schwab, R. M. (1993). Infrastructure spending: Where do we go from here? National Tax Journal, 46(3), 261–273. Junta Central de Puertos (n.d.). Libro de puertos. Madrid: Ministerio de Fomento. Katus, L. (1983). Transport revolution and economic growth in Hungary. In: J. Komlos (Ed.), Economic development in the Habsburgh Monarchy in the nineteenth century. Essays (pp. 183–204). New York: Columbia University Press. Kenwood, A. G. (1965). Railway investment in Britain, 1825–1875. Economica, 32(127), 313–322. Le´vy-Leboyer, M. (1978). Capital Investment and Economic Growth in France, 1820–1930, In: P. Mathias & M. M. Postan (Eds), The Cambridge economic history of Europe. The industrial economies: Capital, labour and enterprise. Part I. Britain, France, Germany and Scandinavia (pp. 231–295). Cambridge University Press: Cambridge. Le´vy-Leboyer, M., & Bourguignon, F. (1990). The French economy in the nineteenth century. An essay in econometric analysis. Cambridge: Cambridge University Press. Lo´pez Herna´ndez, E. (1968). Patrimonio de la Compan˜ı´ a Telefo´nica Nacional de Espan˜a. In: Universidad Comercial de Deusto (Ed.), Riqueza Nacional de Espan˜a (Vol. 2, pp. 723–741). Bilbao: Universidad Comercial de Deusto. Maddison, A. (1995). Monitoring the World economy. Paris: OECD. Madrazo, S. (1984). El sistema de comunicaciones en Espan˜a, 1750–1850. Madrid: Turner. Mas, M., Maudos, J., Pe´rez, F., & Uriel, E. (1996). Infrastructures and productivity in the Spanish regions. Regional Studies, 30(7), 641–649. Mas, M., Pe´rez, F., & Uriel, E. (Eds) (1995). El ‘‘stock’’ de capital en Espan˜a y sus comunidades auto´nomas. Bilbao: Fundacio´n BBV. Mas, M., Pe´rez, F., & Uriel, E. (Eds) (1998). El ‘‘stock’’ de capital en Espan˜a y su distribucio´n territorial. Bilbao: Fundacio´n BBV. Mitchell, B. R. (1964). The coming of the railway and United Kingdom economic growth. Journal of Economic History, 24(3), 315–336. Moorsteen, R., & Powell, R. P. (1966). The Soviet capital stock, 1928–1962. Homewood: Yale University. Nadal, J. (1976). The failure of the industrial revolution in Spain 1830–1914. In: C. M. Cipolla, (Ed.). The Fontana economic history of Europe (Vol. 4). The Emergence of Industrial Societies. Part Two (pp. 532–626). New York: Harper & Row. Nu´n˜ez, G. (1996). Servicios urbanos colectivos en Espan˜a durante la industrializacio´n: entre la empresa privada y la gestio´n pu´blica. In: F. Comı´ n & P. Martı´ n Acen˜a (Eds), La empresa en la historia de Espan˜a (pp. 399–419). Madrid: Civitas. O’Brien, P. (1977). The new economic history of railways. London: Croom Helm. Ohkawa, K., Ishiwate, S., Yamada, S., & Ishi, H. (1966). Estimates of long-term economic statistics of Japan since 1868 (Vol. 3). Capital Stock. Tokyo: Toyo Keizai Shinpo Sha. Ohkawa, K., Takamatsu, N., & Yamamoto, Y. (1974). Estimates of long-term economic statistics of Japan since 1868. 1. National Income. Tokyo: Toyo Keizai Shinponsha.
118
ALFONSO HERRANZ-LONCA´N
Olaizola, J. (2003). La traccio´n ele´ctrica en la vı´a estrecha espan˜ola (1902–2003). III Congreso de Historia Ferroviaria, Gijo´n. Ormaechea, A´. M. (1989). Ferrocarriles en Euskadi 1855–1936. Bilbao: Eusko Trenbideak. Ortu´n˜ez, P.P. (1999). El proceso de nacionalizacio´n de los ferrocarriles en Espan˜a. Doctoral thesis, University of Valladolid, Valladolid. Palafox, J. (1980). La gran depresio´n de los an˜os treinta y la crisis industrial espan˜ola. Investigaciones Econo´micas, 11, 5–46. Palafox, J. (1991). Atraso econo´mico y democracia. La Segunda Repu´blica y la economı´a espan˜ola, 1892–1936. Barcelona: Crı´ tica. Pascual, P. (1999). Los caminos de la era industrial. La construccio´n y financiacio´n de la Red Ferroviaria Catalana (1843–1898). Barcelona: Edicions Universitat de Barcelona. Pons, J., & Tirado, D.A. (2001). Discontinuidades en el crecimiento econo´mico en el periodo 1870–1994: Espan˜a en perspectiva comparada. FEDEA Working Paper EEE 98. Prados de la Escosura, L. (2003). El progreso econo´mico de Espan˜a, 1850–2000. Madrid: Fundacio´n BBVA. Prados de la Escosura, L., & Rose´s, J.R. (2003). El capital en Espan˜a, 1850–2000. Unpublished research paper. Reis, J. (2000). How poor was the European periphery before 1850? The Mediterranean vs. Scandinavia. In: S. Pamuk & J. G. Williamson (Eds), The Mediterranean response to globalization before 1950 (pp. 17–44). London: Routledge. RENFE. (1958). Los ferrocarriles en Espan˜a, 1848–1958. Madrid: RENFE. Ringrose, D. R. (1970). Transportation and economic stagnation in Spain, 1750–1850. Durham D.C: Duke University Press. Rossi, N., Sorgato, A., & Toniolo, G. (1993). I conti economici italiani: una ricostruzione statistica 1890–1990. Rivista di Storia Economica, 10(1), 1–47. Stern, N. (1991). The determinants of growth. Economic Journal, 101, 122–133. Sua´rez de Tangil, F. (1954). Las obras pu´blicas en Espan˜a y los gobiernos de autoridad. Madrid: Real Academia de Ciencias Morales y Polı´ ticas. Tedde, P. (1978). Las compan˜ı´ as ferroviarias en Espan˜a (1855–1935). In: M. Artola (Ed.), Los ferrocarriles en Espan˜a, 1844–1943 (Vol. 2, pp.9–354). Madrid: Banco de Espan˜a. Tilly, R. H. (1978). Capital formation in Germany in the nineteenth century, In: P. Mathias, M. M. Postan (Eds), The Cambridge economic history of Europe. The industrial economies: capital, labour and enterprise. Part I. Britain, France, Germany and Scandinavia (pp. 382–441). Cambridge University Press: Cambridge. Tortella, G. (1972). Spain 1829–1874. In: R. Cameron (Ed.), Banking and economic development: Some lessons of history (pp. 91–121). New York: Oxford University Press. United Nations. (1993). System of National Accounts 1993. New York: United Nations. Uriol, J. I. (1968). La red de carreteras del Estado. Evaluacio´n por el me´todo de formacio´n histo´rica del capital. In: Universidad Comercial de Deusto (Ed.), Riqueza Nacional de Espan˜a (Vol. 2, pp. 393–426). Bilbao: Universidad Comercial de Deusto. Uriol, J. I. (1992). Historia de los caminos de Espan˜a. Madrid: Colegio de Ingenieros de Caminos, Canales y Puertos. Vogelsang, T. J. (1997). Wald-type tests for detecting breaks in the trend function of a dynamic time series. Econometric Theory, 13, 818–849. Wais, F. (1987). Historia de los ferrocarriles espan˜oles. Madrid: Fundacio´n de los Ferrocarriles Espan˜oles.
Table A1.
Spanish Gross Infrastructure Investment, 1845–1935 (millions of 1890 pesetas).
Total Electric. Gas Broad Narrow Non-Public Tramways Subway State’s Provincial Local Ports Telegraph Telephone Reservoirs Other Gauge Gauge Hydraulic Railroad Roads Roads Roads Network Network Distribution Distribution Railroad Railroad Works Network Network
1.41 1.95 3.12 4.92 7.57 10.36 12.73 14.11 14.85 16.68 18.72 19.96 32.68 48.99 87.88 113.53 131.05 133.48 118.79 97.30 60.72 30.01 13.84 5.97 9.11 9.57 10.59 11.40 20.58
0.25 0.57 0.64 0.65 0.67 0.41 0.09 0.03 0.01 0.04 0.38 0.69 0.69 0.69 0.96 1.33 1.42 1.45 1.49 1.19 1.20 1.80 2.06 2.13 2.68
0.12 0.24
2.40 2.40 3.17 5.13 7.69 9.46 11.86 14.13 11.93 9.80 12.75 9.35 11.31 14.25 8.70 13.43 26.53 27.93 24.07 23.02 17.70 11.57 12.27 14.92 11.35 8.15 10.23 9.37 7.78
0.33 0.33 0.39 0.55 0.76 0.90 1.09 1.28 1.09 0.93 1.16 0.99 1.02 1.06 1.10 1.14 1.18 1.23 2.82 3.08 3.38 3.70 1.84 1.92 1.99 0.42 0.42 0.42 2.66
0.07 0.07 0.09 0.12 0.17 0.20 0.24 0.28 0.24 0.20 0.26 0.21 0.23 0.26 0.30 0.21 0.21 0.22 0.22 0.23 0.24 0.26 0.23 0.24 0.25 0.25 0.26 0.27 0.33
0.40 0.34 0.36 0.44 0.56 0.75 0.85 1.10 1.54 1.22 1.27 2.12 1.27 1.63 2.42 2.80 3.73 4.66 9.70 9.74 7.36 4.87 2.98 3.47 2.42 1.73 3.29 3.58 1.13
0.08 0.08 0.08 0.08 0.08 0.77 1.08 1.09 0.46 0.31 0.46 0.59 0.58 0.39 0.17 0.01 0.04 0.06 0.14 0.11 0.10 0.02 0.01
0.10 0.09 0.07 0.06 0.01 0.07 0.13 0.20 0.21 0.23 0.25 0.23 0.23 0.19 0.29 0.28 0.21 0.13 0.14 0.27 0.25 0.29 0.21 0.31 0.32 0.33 0.52 0.59 0.54
0.60 0.11 0.11 0.15 0.29 0.29 0.22 0.22 0.22 0.40 0.40 0.40 0.40 0.40 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.19
1.06 1.06 1.06 1.06 3.13 3.13 3.13 3.13 3.13 3.13 3.13 3.17 5.52 5.52 2.64 2.64 2.92 2.92 2.92 2.92 2.92 2.88 0.53 0.53 0.53 0.53 0.26 0.26 0.26
6.37 6.35 8.37 12.68 20.74 25.79 30.98 35.21 33.71 32.78 38.06 37.22 53.73 73.44 104.21 135.06 167.02 171.90 160.24 138.32 94.18 55.08 33.49 28.66 27.37 22.96 27.79 28.23 36.40
6.47 6.43 8.44 12.74 20.74 25.86 31.11 35.40 33.91 33.00 38.30 37.44 53.95 73.62 104.49 135.33 167.22 172.02 160.37 138.58 94.42 55.36 33.69 28.95 27.68 23.27 28.29 28.79 36.91
119
1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873
Total Corrected (Inc. Urban Infrastructure)
The Spanish Infrastructure Stock
APPENDIX
120
APPENDIX (Continued ) Total Reservoirs Other Gas Broad Narrow Non-Public Tramways Subway State’s Provincial Local Ports Telegraph Telephone Electric. Hydraulic Railroad Roads Roads Roads Network Network Distribution Distribution Gauge Gauge Network Works Network Railroad Railroad
33.72 37.04 29.14 25.20 50.53 61.97 69.70 57.48 58.66 51.65 39.24 30.59 23.89 25.27 30.88 39.97 49.83 64.76 69.07 65.25 68.92 59.14 44.59 16.47 19.41 25.01 25.99 23.18 23.65 23.42 14.70 8.94
2.60 1.78 1.50 1.73 1.51 1.73 2.84 3.84 5.32 6.67 6.74 6.32 6.77 7.23 8.53 11.52 13.99 15.87 15.97 13.62 10.35 7.82 5.58 4.41 5.04 7.54 11.99 14.89 15.27 15.02 12.89 8.36
0.36 0.47 0.24 0.01 0.09 0.28 0.27 0.09 0.14 0.28 0.15 0.05 0.41 0.76 0.60 0.61 0.89 1.08 1.27 1.49 1.76 2.07 2.44 2.28 1.46 1.10 1.26 1.33 1.33 1.33 1.33 1.33
0.36 1.12 0.36 0.76 0.03 0.14 0.16 0.27 0.39 0.76 2.38 2.74 0.32 0.50 1.97 2.97 4.11 2.27 2.65 7.72 18.11 6.86 3.76 2.45 4.91
5.52 7.40 11.66 13.60 14.47 17.05 14.80 16.69 33.09 19.09 19.30 17.20 21.29 25.28 23.21 19.28 17.32 15.43 15.03 15.96 14.88 17.13 18.71 18.39 14.72 12.98 9.21 9.09 13.04 14.82 10.06 16.12
2.80 2.94 3.10 3.26 3.44 3.62 3.81 4.01 4.83 3.47 2.52 7.95 6.87 2.07 0.13 0.13 0.13 4.19 4.35 1.13 1.14 1.16 1.17 0.60 0.60 2.52 2.57 2.62 2.67 2.72 2.78 2.83
0.34 0.35 0.36 0.37 0.39 0.40 0.41 0.43 0.28 0.02 0.22 0.64 1.40 0.26 0.03 0.03 0.03 0.43 0.44 0.72 0.75 0.50 0.51 0.06 0.06 0.06 0.06 0.35 0.36 0.73 0.75 0.78
5.39 4.64 0.60 0.64 0.64 0.81 1.20 1.06 0.89 1.14 2.74 6.01 3.47 4.18 4.02 3.73 4.43 13.64 10.94 9.54 10.61 12.26 12.26 16.01 14.25 11.00 9.33 8.88 10.30 9.89 11.69 13.53
0.15 0.33 0.72 0.67 0.51 0.23 0.16 0.24 0.25 0.29 0.33 0.37 1.75 2.04 2.00 0.83 1.04 1.15 1.25 0.91 0.75 0.36 0.12 0.09 0.11 0.24 0.21 0.23 0.13 0.09 0.22 0.42
0.05 0.05 0.21 0.26 0.21 0.09 0.07 0.40 0.33 0.35 0.44 0.07 0.08 0.12 0.16 0.25 0.17 0.20 0.20 0.41 0.66
0.43 0.52 0.91 1.35 1.15 0.83 0.36 0.76 1.20 1.48 1.62 1.40 0.97 1.15 1.91 1.99 1.57 1.36 1.63 1.63 0.91 0.98 1.24 1.30 0.84 0.43 0.52 0.91 1.35 1.33 1.73 1.28
0.70 0.76 0.83 0.20 0.21 0.23 0.25 0.27 0.30 0.32 0.35 0.35 0.52 0.61 1.24 2.00 2.87 3.18 2.99 2.65 2.86 4.49 4.31 3.14 1.72 3.68 5.04
0.27 0.47 0.47 0.47 0.61 0.68 0.87 0.87 0.87 0.62 0.46 0.10 0.11 0.18 0.19 0.19 0.22 0.25 0.33 0.36 0.39 0.37 0.40 0.25 0.18 0.23 0.34 0.41 0.44 0.56 1.00 1.06
0.26 0.26 0.26 0.26 0.26 0.26 0.26 0.82 0.82 0.82 2.62 2.72 2.89 2.81 2.95 2.85 2.60 2.05 1.76 1.91 2.47 3.08 3.19 3.25 2.66 1.64 2.81 2.96 4.47 4.69 5.77 5.82
51.83 56.21 48.96 47.56 73.61 88.55 95.46 87.48 107.67 86.10 76.93 73.69 70.26 71.91 75.31 82.09 93.25 123.18 125.78 114.42 115.79 110.15 96.42 70.29 64.37 68.42 76.75 87.44 83.21 80.29 69.47 71.11
52.24 56.71 49.83 48.85 74.70 89.34 95.80 88.20 108.82 87.51 78.47 75.02 71.19 73.01 77.12 83.98 94.75 124.48 127.33 115.97 116.27 112.02 99.25 74.20 66.53 70.95 84.09 104.69 89.75 83.87 71.81 75.79
ALFONSO HERRANZ-LONCA´N
1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905
Total Corrected (Inc. Urban Infrastructure)
6.92 9.30 14.00 20.58 22.81 26.11 36.13 30.02 27.63 11.08 21.54 20.50 2.47 0.88 2.04 20.31 27.63 31.69 37.52 39.47 43.80 52.55 62.32 57.69 50.24 38.03 30.52 18.52 8.77 4.71
5.38 6.29 7.42 7.29 11.43 14.63 12.61 10.92 9.88 7.76 5.61 5.49 6.38 7.34 9.42 8.89 6.51 6.85 7.45 6.75 4.87 8.18 6.74 1.24 2.02 2.38 2.51 2.44 2.17 1.56
1.39 1.52 1.66 1.79 1.53 0.84 0.49 0.50 0.50 0.45 0.33 0.27 0.27 0.27 0.27 0.20 0.12 0.10 0.15 0.20 0.10 0.01 0.04 0.12 0.12 0.04 0.06 0.12 0.06 0.02
3.46 6.82 8.71 1.51 2.51 2.31 4.61 0.74 1.72 3.97 11.84 5.93 1.71 1.48 3.79 10.41 12.91 10.94 8.00 23.77 19.27 11.30 6.60 2.76 6.36 5.63 8.56 11.78 9.89 2.51
2.70 5.41 4.06 2.72 3.08 8.40 14.55 13.64 8.61 2.84 2.41 4.82 2.41 2.97 6.48 4.30 1.06 1.36 2.20
12.16 14.14 15.32 18.01 21.21 20.65 20.12 21.78 21.09 19.97 19.45 16.26 11.46 11.80 13.84 16.70 19.59 21.91 19.36 43.37 84.47 76.78 75.66 74.65 79.52 72.82 41.84 47.53 43.78 44.36
2.89 2.95 3.00 3.06 0.92 0.93 0.63 0.63 0.63 0.64 0.64 1.78 4.39 1.47 1.48 1.49 1.50 1.51 1.53 5.99 6.19 6.46 6.83 7.26 7.63 8.06 1.60 16.45 0.93 2.76
0.81 0.84 0.76 0.79 0.81 0.84 0.87 0.90 0.93 0.96 0.99 4.02 1.83 2.36 2.51 2.68 2.85 3.04 3.24 4.89 5.34 5.85 6.42 7.06 7.73 8.48 9.96 9.49 8.31 8.71
16.39 17.96 20.67 18.16 16.26 16.23 16.95 14.08 15.75 12.42 7.83 7.96 6.97 8.48 8.93 10.45 9.67 10.96 11.35 11.72 9.70 28.26 28.33 30.25 8.69 8.23 8.84 30.45 38.21 30.27
0.87 0.78 0.77 1.02 0.98 1.06 0.64 0.94 0.89 0.87 2.08 2.33 2.33 1.20 1.57 1.78 1.87 1.30 0.96 0.36 0.12 0.24 0.36 0.64 0.47 0.41 0.19 0.18 0.41 0.66
0.52 0.77 1.00 1.03 1.04 1.36 2.01 1.24 1.32 1.61 4.37 2.80 3.16 3.34 1.38 0.99 1.37 1.47 0.97 10.71 20.54 18.72 11.46 73.77 39.56 50.14 6.11 1.70 5.04 10.44
1.96 2.20 2.81 2.49 1.75 0.97 1.31 2.19 2.98 2.57 2.23 1.78 1.63 0.91 0.98 1.24 1.30 0.84 0.43 0.52 0.91 1.35 1.33 2.06 2.61 4.24 4.61 4.60 3.83 2.32
4.83 3.64 3.16 7.40 6.95 8.26 5.43 24.47 25.71 29.84 25.63 30.06 26.81 25.71 28.12 36.18 39.64 45.27 48.73 71.97 70.54 67.19 62.70 51.81 47.55 12.73 21.11 31.49 41.78 30.63
1.64 2.67 2.86 3.65 3.54 4.20 5.42 6.40 5.72 6.07 5.85 4.49 5.60 5.44 6.27 6.32 7.46 7.21 13.02 10.85 11.83 12.83 14.63 15.23 16.63 15.80 13.29 11.13 9.89 6.92
4.54 3.33 3.80 4.28 3.58 3.30 6.72 7.94 7.41 7.17 5.39 2.74 0.76 5.01 5.49 8.29 4.35 2.88 0.91 1.85 3.60 4.84 6.11 6.61 13.12 16.52 10.25 36.06 42.09 36.34
63.78 73.21 85.95 91.06 95.33 101.68 113.93 122.76 122.16 105.37 113.77 109.12 81.17 79.74 88.79 129.00 145.18 160.51 167.28 241.04 284.13 296.97 294.34 333.54 285.22 249.99 163.73 222.99 216.53 184.42
67.07 79.70 94.25 92.50 97.72 103.88 118.32 123.46 123.79 109.15 125.05 117.34 87.95 85.02 94.99 141.85 165.48 184.78 187.89 271.88 305.18 310.02 305.21 338.46 294.10 261.53 175.98 235.21 227.25 188.91
121
Note: As is indicated in the text, for broad gauge railroads, State’s roads and ports, investment figures have been estimated directly from information on capital formation flows. In all other cases, they have been calculated on the basis of gross stock estimates (see the estimation method in Section 2 above). The price index that has been used to deflate the available investment series (in the cases of broad gauge railroads, State’s roads and ports), or to express in 1890 pesetas the stock and investment indices estimated on the basis of physical indicators (in all other cases) is the deflator for ‘‘Other construction’’ investment, which has recently been estimated by Leandro Prados de la Escosura (2003). Sources: (a) For broad gauge railroads, the main source of information is a sample of capital accounts of railroad companies. Investment data has been collected for Compan˜ı´a de Caminos de Hierro del Norte de Espan˜a, from Tedde (1978, pp. 264–290); Compan˜ı´a de los Ferrocarriles de Madrid a
The Spanish Infrastructure Stock
1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935
122 ALFONSO HERRANZ-LONCA´N
Zaragoza y a Alicante (MZA), from the company’s yearly accounts; most broad gauge Catalan companies (before their absorption by MZA), from Pascual (1999); Compan˜ı´a del Ferrocarril de Tudela a Bilbao, from Ormaechea (1989, p. 18); and Compan˜ı´a de los Ferrocarriles Andaluces, Compan˜ı´a de los Caminos de Hierro del Sur de Espan˜a and Compan˜ı´a de los Ferrocarriles de Madrid a Ca´ceres y Portugal (y del Oeste de Espan˜a), from the companies’ yearly accounts and Ortu´n˜ez (1999). The share of land and rolling stock within investment has been calculated from information in MZA’s yearly accounts, Alzola (1884/1885, Vol. 33, p. 228), Herna´ndez (1983), Tedde (1978, pp. 264–290), and Pascual (1999). The average unit cost coming from this sample of companies (which accounted for more than 80% of the broad gauge network during most of the period under study) has been applied to the rest of the network, whose mileage has been taken from Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1860–1924) and Wais (1987). (b) For narrow gauge railroads, average unit cost per km of electrified and non-electrified line has been estimated on the basis of a large sample of companies’ capital accounts in 1922 (corresponding to 89% of the network), which is available in Spain, Ministerio de Fomento, Estadı´stica de Obras Pu´blicas (1922). The share of land and rolling stock within investment has been estimated on the basis of Alzola (1884/1885). Unit cost figures have been applied to each year (electrified and non-electrified) mileage, which has been estimated from Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1860–1924), Anuario de Ferrocarriles de D. Enrique de la Torre (1922–1935), and Olaizola (2003, pp. 16–18). (c) For non-public railroads, yearly mileage in operation has been estimated on the basis of Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1874–1924) and Anuario de Ferrocarriles de D. Enrique de la Torre (1922–1935), and average unit cost has been taken from Alzola (1884/1885, Vol. 33, p. 228). (d) For tramways, average unit cost per km of horse, steam and electricity-drawn lines has been estimated on the basis of a large sample companies’ capital accounts in 1907, which is available in Spain, Ministerio de Fomento, Estadı´stica de Obras Pu´blicas (1907). The share of rolling stock has been taken from Gil Carretero (1968, p. 462). Unit cost figures have been applied to the yearly mileage of each type of traction, which has been estimated from Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1882–1924), Anuario de Ferrocarriles de D. Enrique de la Torre (1893–1935), and Ceballos (1932, Vol. 7, p. 381). (e) For the subway, average unit cost per km of line comes from Go´mez-Santos (1969, p. 40), and has been applied to each year’s mileage, which comes from RENFE (1958, p. 122), and Comı´ n, Martı´ n Acen˜a, Mun˜oz, and Vidal (1998, Vol. 2, p. 307). (f) For State’s roads, the investment series is based on the State’s expenditure on the road network from 1859 onwards, which is available in Uriol (1968). Before 1859, the average construction cost, coming from Uriol (1968), p. 414, has been applied to each year’s network mileage, which has been estimated from Uriol (1992, pp. 15–16, 25,67). (g) For provincial roads, the average construction cost has been estimated on the basis of information on bids for road construction at the expense of the provincial institutions in 1896–1899, which comes from the Revista de Obras Pu´blicas (various years). The resulting figure has been applied to each year’s network mileage, which comes from Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1862–1924), and Anuario Estadı´stico de Espan˜a (1931–1935). (h) For local roads, the average construction cost has been estimated on the basis of information on the State’s expenses in local roads in 1911– 1924, which has been taken from Go´mez Mendoza (1991, p. 192). The resulting figure has been applied to each year network mileage, which comes
The Spanish Infrastructure Stock 123
from Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1866–1924) and Anuario Estadı´stico de Espan˜a (1931–1935). (i) For ports, the basis of the investment series until 1908 is the series of port construction expenditure that is provided by Cercos (1968). For the period 1908–1935, a comparison of the physical characteristics of the Spanish ports in 1908, 1925 and 1965, taken from Spain, Ministerio de Fomento, Estadı´stica de Obras Pu´blicas (1908), Junta Central de Puertos (n.d., pp. 204–213), and Cercos (1968, p. 624), has been complemented with information on the evolution of total port expenditure of the port Juntas in 1908–1925, taken from Junta Central de Puertos (n.d., pp. 214– 222), and the amounts allocated to port investment in the State’s budget in 1925–1965, coming from Sua´rez de Tangil (1954, pp. 50–51), and Cercos (1968, p. 605). (j) For the telegraph system, the value of the network in 1896 has been estimated on the basis of its physical description, which comes from the Estadı´stica Telegra´fica de Espan˜a (1896) and the unit cost of each of its components, which has been taken from De Urquijo (1968, p. 694), and Lo´pez Herna´ndez (1968). The resulting gross value has been brought backward and forward according to the network mileage, which comes from the Estadı´stica Telegra´fica de Espan˜a (1864–1934). (k) For the telephone system from 1924 onwards, data on the nearly monopolistic Compan˜ı´a Telefo´nica Nacional de Espan˜a’s assets, has been taken from its yearly accounts, and this has been applied unit value figures coming from Lo´pez Herna´ndez (1968). For small companies, information has been taken from the Estadı´stica Telegra´fica de Espan˜a (1924–1934), the yearly accounts of the company Red Provincial de Guipu´zcoa and Echaide (1929). Before 1924, the value of the gross stock in that year has been brought backward according to the evolution of the number of networks and their subscribers, coming from the Estadı´stica Telegra´fica de Espan˜a (1885–1924). (l) For the gas distribution network, the gross value of the stock in 1900 has been estimated from information about the companies’ capital accounts in Costa (1981, pp. 49–57), and brought forward and backward according to the evolution of production, which comes from Carreras (1983, Vol. 1, pp. 72–73). (m) For the electricity distribution network, the gross value of the stock in 1943 has been estimated from the information on the companies’ capital accounts that is provided in Becerril (1946), and brought forward and backward according to the evolution of production, which comes from Bartolome´ (1999). (n) For hydraulic works, the construction cost of each individual reservoir has been estimated from information in Garrido (1968). For other hydraulic infrastructure after 1883, the value of the State’s investment in hydraulic works (net of the value of State reservoirs) has been used as a lower bound of total investment. This information is available in Mas, Pe´rez, and Uriel (1995, Vol. 4). Before 1883, the State’s role in the construction of hydraulic works was marginal and a valuation of those canals for which data are available has been used as a lower bound of the gross stock. Information about individual canals has been obtained from Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1862–1924), Bello (1914), Alzola (1979), Ceballos (1932) and Ferna´ndez Ordo´n˜ez (1986). (o) No systematic information is available on urban infrastructure. The last column of the table is just a preliminary correction of the sum of all other infrastructure, in an attempt to reduce the potential bias associated with the exclusion of urban infrastructure from the series. A lower bound of the bias in 1900–1905 has been calculated from information on local public works in Nu´n˜ez (1996, p. 400), and carried forward and backward according to the evolution of urban transport investment (between 1894 and 1935) and investment in gas distribution infrastructure (between 1845 and 1894).
124
Table A2.
Spanish Net Infrastructure Stock, 1844–1935 (millions of 1890 pesetas).
Broad Narrow Non-Public Tramways Subway State’s Provincial Local Ports Telegraph Telephone Gas Electric. Reservoirs Other Gauge Gauge Railroad Roads Roads roads Network Network Distribution Distribution Hydraulic Railroad Railroad Network Works Network 0.81 2.21 4.11 7.16 11.94 19.28 29.27 41.44 54.77 68.58 83.98 101.14 119.23 149.71 195.95 280.21 388.54 512.37 636.24 743.10 826.52 871.82 885.57 882.86 872.28 865.01 858.42 853.09 848.66 855.11 875.02
0.25 0.82 1.44 2.07 2.69 3.05 3.09 3.06 3.02 2.96 2.95 3.28 3.92 4.54 5.15 6.02 7.24 8.53 9.81 11.07 12.00 12.92 14.43 16.19 18.03 20.41 22.69
0.12 0.30 0.61
147.51 147.12 146.73 147.11 149.44 154.32 160.90 169.80 180.89 189.64 196.16 205.56 211.45 219.24 229.87 234.83 244.47 267.09 290.86 310.50 328.87 341.71 348.27 355.46 365.22 371.30 327.50 332.41 330.17 332.63 333.22
7.10 7.29 7.48 7.74 8.14 8.75 9.50 10.43 11.53 12.44 13.18 14.14 14.91 15.71 16.54 17.40 18.29 19.21 20.16 22.69 25.47 28.50 31.82 33.24 34.73 36.27 71.48 71.44 80.64 91.95 94.26
1.56 1.60 1.64 1.70 1.79 1.92 2.08 2.29 2.53 2.73 2.89 3.10 3.26 3.44 3.65 3.90 4.05 4.20 4.36 4.52 4.69 4.86 5.05 5.21 5.38 5.55 5.73 5.92 6.10 6.35 6.60
10.80 10.99 11.12 11.28 11.50 11.85 12.38 13.00 13.88 15.17 16.14 17.14 18.98 19.95 21.28 23.37 25.82 29.17 33.42 42.66 51.83 58.51 62.64 64.82 67.47 69.04 69.91 72.32 75.00 75.20 79.65
0.08 0.16 0.23 0.30 0.38 1.14 2.18 3.20 3.56 3.76 4.10 4.57 5.01 5.25 5.26 5.10 4.99 4.90 4.89 4.85 4.81 4.69 4.56 4.57
0.09 0.19 0.27 0.33 0.38 0.36 0.42 0.53 0.72 0.90 1.10 1.31 1.50 1.67 1.80 2.03 2.23 2.36 2.41 2.45 2.64 2.79 2.98 3.08 3.28 3.44 3.61 3.97 4.39 4.74 5.00
2.41 2.97 3.03 3.08 3.19 3.42 3.65 3.81 3.98 4.14 4.48 4.81 5.14 5.47 5.79 5.76 5.72 5.68 5.64 5.61 5.57 5.53 5.50 5.46 5.43 5.40 5.37 5.35 5.32 5.43 5.62
16.72 17.50 18.26 19.02 19.77 22.59 25.38 28.13 30.86 33.55 36.21 38.85 41.48 46.43 51.33 53.30 55.25 57.45 59.62 61.76 63.89 65.99 68.03 67.70 67.37 67.05 66.73 66.14 65.56 64.99 64.42
Total Corrected (Inc. Urban Infrastructure)
187.00 189.86 192.65 197.41 206.41 223.31 245.02 271.60 302.00 330.44 357.53 389.48 420.11 466.77 532.36 627.63 752.03 906.16 1,062.42 1,204.33 1,321.95 1,393.49 1,424.76 1,433.89 1,438.05 1,440.89 1,428.04 1,431.63 1,438.68 1,461.67 1,491.64
187.00 189.96 192.83 197.65 206.69 223.58 245.35 272.04 302.61 331.24 358.52 390.68 421.48 468.30 534.02 629.51 754.10 908.36 1,064.67 1,206.62 1,324.42 1,396.11 1,427.56 1,436.79 1,441.13 1,444.17 1,431.48 1,435.41 1,442.86 1,466.18 1,496.40
ALFONSO HERRANZ-LONCA´N
1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874
Total
24.14 25.31 26.68 27.83 29.23 31.73 35.19 40.09 46.33 52.59 58.39 64.64 71.29 79.13 89.95 102.65 117.05 131.36 143.11 151.44 157.12 160.49 162.66 165.44 170.69 180.33 192.74 205.14 217.07 226.71 231.78 233.84 236.73 240.75 244.57 252.38 263.15 271.57 278.05 283.40 286.61
1.02 1.35 1.43 1.47 1.65 1.90 2.04 2.13 2.32 2.50 2.56 2.79 3.36 3.99 4.54 5.25 6.18 7.29 8.59 10.13 11.94 14.07 16.24 17.85 18.87 19.82 20.87 21.94 22.99 24.03 25.06 26.11 27.24 28.49 29.87 31.13 31.87 32.11 32.19 32.29 32.35
0.36 1.46 1.77 2.46 1.65 1.72 1.83 2.03 2.35 3.02 5.29 7.84 7.87 8.09 9.75 12.36 15.43 16.87 18.64 25.41 42.00 47.34 49.37 50.03 52.48 53.45 57.29 63.46 62.36 62.48 62.43 64.36 62.65 61.84 63.24
335.38 343.45 350.23 358.63 369.51 384.92 397.55 438.94 458.83 480.16 493.88 508.50 527.21 542.98 555.51 566.41 576.54 585.63 595.90 605.11 618.57 632.49 645.77 654.29 661.67 664.87 665.54 670.14 676.46 679.60 687.72 691.96 698.14 705.07 714.61 726.63 737.96 749.90 763.36 775.95 787.28
96.01 97.64 101.37 104.87 108.36 105.28 104.85 92.47 86.67 77.94 82.19 87.24 87.81 87.16 86.30 85.02 86.92 89.23 89.32 89.06 88.79 89.39 88.94 88.49 89.97 91.48 93.02 94.60 96.82 96.07 97.75 99.47 101.23 103.03 104.87 104.55 104.23 103.62 103.02 102.43 101.85
6.87 7.14 7.42 7.71 8.01 8.32 8.64 8.81 8.72 8.83 9.36 10.64 10.77 10.66 10.56 10.46 10.75 11.06 11.64 12.24 12.59 12.95 12.85 12.75 12.66 12.56 12.76 12.96 13.53 14.12 14.73 15.36 16.02 16.59 17.18 17.78 18.41 19.05 19.72 20.41 21.12
83.30 82.87 82.49 82.10 81.90 82.09 82.14 82.01 82.13 83.85 88.82 91.20 94.27 97.13 99.67 102.88 115.27 124.82 132.86 141.87 152.44 162.89 176.97 189.13 197.90 204.89 211.36 219.17 226.49 235.52 246.29 259.80 274.72 292.19 306.94 319.62 332.13 345.22 355.29 366.91 375.08
4.76 5.34 5.85 6.18 6.22 6.20 6.22 6.25 6.32 6.42 6.56 7.80 9.19 10.47 10.81 11.40 12.02 12.67 12.97 13.17 13.07 12.79 12.48 12.18 12.00 11.80 11.63 11.40 11.13 10.96 10.92 11.17 11.34 11.57 12.15 12.69 13.27 13.40 13.82 14.16 14.45
0.05 0.09 0.30 0.55 0.75 0.82 0.86 1.23 1.52 1.83 2.22 2.22 2.23 2.28 2.38 2.55 2.64 2.76 2.88 3.20 3.76 4.17 4.81 5.67 6.53 7.37 8.51 10.26 11.19 12.17 13.39
5.31 5.98 7.03 7.85 8.30 8.25 8.62 9.42 10.49 11.61 12.48 12.91 13.54 14.90 16.23 17.12 17.75 18.64 19.47 19.54 19.67 19.99 20.32 20.21 19.73 19.33 19.18 19.30 19.47 20.16 20.57 21.48 22.42 23.82 24.80 25.10 24.77 24.72 25.26 26.54 27.51
0.70 1.44 2.21 2.33 2.46 2.60 2.76 2.93 3.12 3.33 3.56 3.78 4.16 4.62 5.69 7.49 10.08 12.89 15.42 17.51 19.73 23.50 26.96 29.12 29.78 32.08 35.63 39.08 41.22 42.78 48.53 53.61 59.80 62.92 84.96 107.45 133.17
6.00 6.38 6.75 7.27 7.85 8.62 9.37 10.12 10.61 10.94 10.90 10.86 10.90 10.95 11.00 11.07 11.18 11.37 11.58 11.82 12.04 12.28 12.37 12.39 12.46 12.64 12.88 13.16 13.55 14.38 15.26 16.71 19.17 21.80 25.17 28.41 32.27 37.31 43.27 48.48 53.97
63.86 63.31 62.76 62.22 61.68 61.15 61.19 61.22 61.25 63.09 65.00 67.05 69.01 71.09 73.04 74.72 75.83 76.64 77.58 79.08 81.17 83.35 85.56 87.15 87.71 89.44 91.29 94.63 98.16 102.72 107.29 110.52 112.50 114.94 117.82 119.97 121.81 127.05 133.46 139.26 144.75
1,523.87 1,550.05 1,571.46 1,620.56 1,684.86 1,757.48 1,821.42 1,905.71 1,970.20 2,024.24 2,074.88 2,120.37 2,166.76 2,216.71 2,273.87 2,335.91 2,427.41 2,519.99 2,601.28 2,682.73 2,759.59 2,821.58 2,855.59 2,882.94 2,914.58 2,953.60 2,999.85 3,039.42 3,075.14 3,096.73 3,120.22 3,136.57 3,160.95 3,194.50 3,231.44 3,271.11 3,318.43 3,378.25 3,447.34 3,516.42 3,568.61
1,528.92 1,555.75 1,578.16 1,628.04 1,692.76 1,765.34 1,829.63 1,914.68 1,980.19 2,035.30 2,086.76 2,132.67 2,179.65 2,230.89 2,289.33 2,352.21 2,444.31 2,537.74 2,619.82 2,700.95 2,778.90 2,842.82 2,879.74 2,908.17 2,941.26 2,986.45 3,048.41 3,092.23 3,129.17 3,150.83 3,176.88 3,194.18 3,222.50 3,261.55 3,296.88 3,336.03 3,382.82 3,444.25 3,510.93 3,578.57 3,631.69
125
897.22 911.29 919.44 954.43 1,001.46 1,057.59 1,103.03 1,150.45 1,192.30 1,221.25 1,240.29 1,251.98 1,264.17 1,282.35 1,309.59 1,341.31 1,387.59 1,437.61 1,483.18 1,531.85 1,570.14 1,593.43 1,588.37 1,586.41 1,590.18 1,594.97 1,596.97 1,597.77 1,597.45 1,587.16 1,570.99 1,553.46 1,538.11 1,524.36 1,516.06 1,509.39 1,507.82 1,516.74 1,521.10 1,525.12 1,513.82
The Spanish Infrastructure Stock
1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915
126
APPENDIX (Continued ) Electric. Gas Broad Narrow Non-Public Tramways Subway State’s Provincial Local Ports Telegraph Telephone Total Total Corrected Reservoirs Other Gauge Railroad Roads Roads Roads Network Network Distribution Distribution Gauge (Inc. Urban Hydraulic Railroad Railroad Infrastructure) Network Network Works
1,514.05 1,513.03 1,493.50 1,472.00 1,451.25 1,447.87 1,451.53 1,459.67 1,473.25 1,489.61 1,511.55 1,544.77 1,587.01 1,623.49 1,651.93 1,667.86 1,675.83 1,671.39 1,657.93 1,641.25
287.59 288.39 289.93 292.11 296.18 299.46 300.30 301.70 303.99 305.78 305.87 309.37 311.24 305.14 301.57 298.06 294.63 291.27 284.99 281.42
32.31 32.20 32.06 31.92 31.79 31.62 31.35 31.01 30.68 30.39 30.13 29.81 29.43 29.09 28.79 28.50 28.17 27.85 27.57 27.27
71.22 73.40 72.29 70.92 71.35 77.86 86.17 92.03 95.52 112.61 120.62 124.76 125.29 120.90 120.91 120.41 121.57 125.48 129.39 126.20
2.70 8.09 12.08 14.68 17.63 25.87 40.18 53.46 61.58 63.86 65.69 69.91 71.68 74.00 79.80 83.38 83.68 84.29 85.72
797.96 807.77 812.96 815.94 820.93 828.73 839.33 852.12 862.24 896.64 970.77 1,036.36 1,099.80 1,160.75 1,224.86 1,280.85 1,303.64 1,332.28 1,356.28 1,381.63
101.27 101.85 105.03 105.25 105.48 105.71 105.96 106.21 106.53 111.18 115.98 120.96 126.19 131.71 137.48 143.54 143.00 157.38 156.15 156.67
21.86 25.62 27.14 29.17 31.34 33.65 36.11 38.72 41.51 45.90 50.69 55.92 61.66 67.95 74.83 82.37 91.27 99.63 106.72 114.11
378.56 382.13 384.67 388.69 393.11 399.01 404.05 410.34 416.94 423.75 428.50 451.74 474.75 499.38 502.09 504.27 506.93 531.00 562.68 586.04
15.40 16.46 17.50 17.83 18.45 19.20 20.00 20.33 20.37 19.44 18.92 18.55 18.31 18.30 18.13 17.90 17.50 17.11 16.92 16.90
17.34 19.54 22.00 24.59 25.19 25.39 25.83 26.39 26.42 36.15 55.57 72.57 81.79 153.01 187.83 232.20 231.19 225.79 223.82 227.21
28.20 28.31 28.27 27.79 27.36 27.12 26.91 26.44 25.74 25.11 24.75 24.67 24.57 25.06 26.25 28.76 31.45 33.79 35.41 35.75
153.71 177.68 197.23 214.62 233.67 260.16 289.28 322.88 358.08 415.31 469.49 518.91 561.26 590.64 614.78 603.69 601.57 608.24 625.12 629.73
59.20 63.00 67.86 72.52 77.94 83.36 89.85 96.02 103.49 112.92 123.42 134.81 147.84 161.28 175.96 189.67 200.71 209.46 216.81 221.11
148.40 149.37 148.34 151.57 155.24 160.70 162.80 163.41 162.04 161.63 162.97 165.53 169.33 172.76 182.66 195.85 202.61 235.11 273.27 305.25
3,627.08 3,681.44 3,706.85 3,726.99 3,753.96 3,817.46 3,895.34 3,987.45 4,080.28 4,248.01 4,453.07 4,674.40 4,888.39 5,131.13 5,322.08 5,473.73 5,533.46 5,649.46 5,757.36 5,836.24
3,698.63 3,757.99 3,786.78 3,809.11 3,838.54 3,910.70 4,003.93 4,115.51 4,223.28 4,413.74 4,627.00 4,852.47 5,069.42 5,309.57 5,501.06 5,656.43 5,719.18 5,837.36 5,948.57 6,024.04
Sources: The stock figures are the result of applying the perpetual inventory method to the investment figures in Table A1. The value of the net stock at the end of 1844 in the cases of roads, ports and hydraulic works has been estimated on the basis of Uriol (1992), Spain, Ministerio de Fomento, Memoria(s), Anuario(s) and Estadı´stica(s) de Obras Pu´blicas (1862–1924), Bello (1914), Alzola (1979), Ceballos (1932), Ferna´ndez Ordo´n˜ez (1986) and Cunningham (1914).
ALFONSO HERRANZ-LONCA´N
1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935
HAVE AMERICAN WORKERS ALWAYS BEEN LOW SAVERS? PATTERNS OF ACCUMULATION AMONG WORKING HOUSEHOLDS, 1885–1910 John A. James, Michael G. Palumbo and Mark Thomas ABSTRACT Based on empirical patterns of annual earnings and saving from new micro-data covering a large sample of American workers around a hundred years ago, we develop a model for simulating the cross-section distribution of wealth at the turn of the twentieth century. Our methodology allows for a direct comparison with the wealth distribution from a sample of families in a comparable part of the contemporary income distribution. Our primary finding is that patterns of wealth accumulation among American workers at the turn of the century bear a striking resemblance to contemporary profiles.
Research in Economic History, Volume 23, 127–175 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23004-5
127
128
JOHN A. JAMES ET AL.
1. INTRODUCTION It is well known that the aggregate personal saving rate in the U.S. peaked in the early decades of the twentieth century and has been at an historical trough for much of the past quarter century. This paper offers a fresh perspective on this transformation by asking a deceptively simple question: how does the saving behavior of American workers in the era before social insurance compare to modern standards? Did workers contribute to the historically high rates of personal saving in the Gilded Age? Was wealth accumulation widely dispersed among American households? Or did American workers at the end of the nineteenth century save as little and accumulate as little as their counterparts at the end of the twentieth? How did household behavior differ in an era in which both demographic and marketplace risks were higher than today, and in which public mechanisms to ameliorate their consequences were less developed than today? It is not possible to address these issues satisfactorily with current evidence. There were no systematic surveys of working-class saving behavior before 1917/1918 (during which saving was stimulated by the sale of war bonds and a shortage of consumer goods). There are no comprehensive wealth estimates for U.S. households between the Federal Census of 1870 and the post-1945 Surveys of Consumer Finance. The pattern of asset ownership and its distribution by income group during this period must therefore be inferred from such fragmentary data as do exist, in particular the data on savings behavior reported in the various worker surveys produced by state and federal government after 1875. The most common method used to generate the pattern of lifetime asset-holding from cross-section data on annual savings is to cumulate the mean savings level by age (Rotella & Alter, 1993; Gratton & Rotondo, 1991; Haber & Gratton, 1994). This procedure has two major disadvantages: it assumes that the pattern of savings revealed in a single snapshot is indicative of lifetime accumulation patterns, and it provides no evidence on the distribution of savings among the working class, despite considerable heterogeneity in their economic circumstances.1 This paper produces estimates of lifetime asset accumulation by working households that incorporate evidence on the distribution of saving and accumulation, and which also allow for variability in income levels and saving rates by the same household over time. We have created a new data set of more than 27,000 families for the period 1885–1908, drawn from the records of the numerous household surveys conducted by state Bureaus of Labor Statistics (BLS). These data provide the raw material for a simulation model of life-cycle patterns of saving and wealth accumulation in the pre-New Deal
Low Saving Americans?
129
era, which explicitly allows for variance in income and savings by each household over time, and which permits us to report the distribution of wealth accumulation by age. The simulation model is constructed such that we can readily compare our results to the profile of asset ownership for contemporary working-class American families, as revealed by the Panel Study of Income Dynamics (PSID) (Hubbard, Skinner, & Zeldes, 1995). What do we find? Most strikingly, it appears that working-class families at the end of the nineteenth century saved at similar rates to working households today; they do not appear to have been stimulated to higher saving rates by precautionary or life-cycle considerations. Then, as now, the typical middle-income household did little or no saving in any given year. Not that thrift was entirely absent. When they did save, family saved aggressively, putting aside large proportions of their current income. But saving levels were not consistent enough over time to result in greater assetholding among the American working-class a century ago than now, despite the absence of social insurance. Indeed, we document distributions for total net worth and financial wealth relative to income in 1900 that are strikingly similar to those evident in 1984. In particular, our results indicate that the bottom 20 percent of households had almost no wealth at age 65 at the end of the nineteenth century – much the same as today. Our findings raise some profound questions about the nature of individual and social welfare in late nineteenth-century America, as well as its transformation over the following century. Both the saving data and the simulated wealth distribution are at odds with the predictions of the lifecycle model of saving often used to describe wealth accumulation before 1935; neither show the classic hump-shaped profile when plotted against the age of the household head. The low rate of wealth accumulation also seems at odds with the precautionary model, especially given the higher degree of economic risk and uncertainty a century ago (indeed, one of the arguments used to explain the low personal savings rate in the late twentieth century has been the relative absence of income risk for modern working households given the protections of unemployment insurance, Medicare, and other social welfare programs). How can we reconcile our findings with the received understanding of the social histories of the urban working class before World War I? We cannot, in this paper, offer categorical answers, but we develop some suggestions, which speak both to our understanding of the nature of the risks faced and the means and motives used to counter them in the absence of social insurance. The structure of the paper is as follows: We describe the database on household incomes and saving in Section 2. Section 3 describes our
130
JOHN A. JAMES ET AL.
simulation model for life-cycle saving and wealth accumulation in detail, while our primary results appear in Section 4. Section 5 compares the simulation results to similar data for the U.S. economy a century later. Section 6 concludes the paper by returning to the questions raised in the introduction – were American workers relatively weak accumulators a century ago, and how should we explain their behavior in the face of considerable risk and uncertainty?
2. THE STATE BLS SURVEY DATA ON FAMILY SAVING 2.1. State BLS Survey Design Our data set is derived from surveys of American workers undertaken by state-administered BLS during the late nineteenth and early twentieth centuries. More than a hundred surveys were published from 1873 through 1911, primarily during the late 1880s and early 1890s (Carter, Ransom, & Sutch, 1991b). Coverage varied from state-to-state and from survey-tosurvey, both in terms of the range of questions asked of respondents and the range of individuals surveyed. Some states aimed their surveys at workers employed in a specific industry, with saturated coverage (the Michigan model), while others preferred to poll workers sampling randomly from many different industries (the Kansas model). In general, the surveys focused on wage-earning households, excluding workers in agriculture.2 Almost all surveys followed a common structure, asking workers about their occupational and personal histories, their current employment and family conditions, and, in certain cases, their incomes, expenditures, and saving during the past year. Each BLS survey covered a cross-section of workers – none cataloged experiences of individual workers across years. Our database therefore pools unrelated cross-section surveys, which record information pertaining to household saving behavior and labor earnings. The savings data set used those BLS surveys that provide information on the age, sex, nativity, and marital status of the household head, as well as the size, income, and reported saving of the family. Table A1 sets out the surveys used, identified with an asterisk, including information on the year of the survey,3 its industrial coverage, and the number of workers sampled. The final data set took the surveys that met our requirements, ensuring that
Low Saving Americans?
131
coding was consistent across surveys and that variables, such as income and savings, were defined consistently.4 Savings in the BLS surveys were either recorded directly (in response to questions such as ‘‘how much did you save last year?’’), or may be calculated as a residual from reported incomes and expenditures (calculated saving). Our analysis concentrates on reported saving, given the well-known susceptibility of calculated saving to errors in the measurement of both income and expenditure.5 The text of the survey questions makes it clear that respondents were asked about the annual flow of financial saving out of current income, rather than the stock of accumulated savings or total assets. Some forms of saving are likely to have been excluded from the BLS surveys – including asset accumulation via capital gains and funded contributions to pension plans (we do, however, explicitly model capital gains in residential equity), and the purchase of stocks, bonds or other financial instruments.6 However, given the time period covered, and the economic status of the households included in our sample, these gaps are unlikely to be either large or serious.7 State BLS surveys generally recorded the level of saving only when it was positive. This raises the question of how we should interpret the blank cells – as missing variables or as indicating zero savings over the survey period? We take as our guide the results of the Michigan survey of vehicle-workers in 1896, which allowed three responses to the question ‘‘amount saved in the past year?’’ – a positive amount, ‘‘none,’’ or no response. Of the 2,787 survey respondents who did not report positive saving, 2,576 (92.4 percent) explicitly answered ‘‘none,’’ while only 211 reported no response. It seems reasonable, on this evidence, to interpret ‘‘no response’’ as indicating no saving on the part of the household. At the same time, it seems implausible that all zero-savers precisely matched expenditures to income without borrowing against past or future incomes. While the great majority of worker households were doubtless liquidity constrained (DeLong & Summers, 1986), some families must have dissaved in any given year, either out of accumulated assets or by borrowing against future income (e.g., by running up store credits).8 We model dissaving using data from five Kansas surveys, which explicitly solicited and recorded responses when survey households ran ‘‘deficits’’ (negative saving values) during the previous year (see Appendix A for further discussion). Two final comments are necessary. First, although some surveys reported information on annual family income – income from all sources combined – most did not.9 We have therefore modeled the income process in the simulation using the earnings of the household head alone. Finally, because
132
JOHN A. JAMES ET AL.
we merge information covering different states over different years, we used an intertemporal–interstate price index to convert both incomes and savings to a common base (U.S. average 1900). Our price index adjusts the DavidSolar consumer price index for the U.S. as a whole for relative interstate variation using Michael Haines’s state consumer price index for 1890 (David & Solar, 1977, p. 16; Haines, 1989). This method of deflating annual earnings and saving does not qualitatively alter the empirical patterns compared with the original, current-dollar data.10
2.2. Saving behavior by BLS households Savings averaged 9.1 percent of annual (household head) income across all families in our pooled sample. Panel A of Fig. 1 shows rapidly increasing saving rates among young families and sizable saving rates, on average, through prime age. Such a finding would suggest potentially substantial wealth-holdings among American workers a hundred years ago, except that
percent of households
percent of income
0.2 0.15 0.1 0.05 0 15
25
35
45
55
0.5 0.4 0.3 0.2 0.1 0 15
65
25
35
age A. Average Annual Household Saving Rates
percent of income
percent of income
0.3 0.2 0.1 0 25
35
45 age
55
C. Average Saving Rates for Positive Savers
Fig. 1.
55
65
B. Frequency of Positive Annual Household Saving
0.4
15
45 age
65
0.45
60th percentile
70th percentile
80th percentile
90th percentile
0.3 0.15 0 15
25
35
45
55
65
age D. Distribution of Saving Rates, Deciles 60%-90%
Distributions of Annual Saving Rates by Age.
Low Saving Americans?
133
the median saving rate at all ages in the data is zero. Fig. 1, panel B, shows that fewer than half of families at every age reported positive saving during the previous year; across the entire pooled sample the proportion was 41.2 percent. The proportion of savers increased substantially during the early working ages of the household head, peaking in his/her late-twenties and declining thereafter. Fig. 1, panel D, provides more detail on the distribution of saving among the upper half of the distribution, presenting saving rate deciles by age. This picture not only reinforces the point that the majority of working-class families at every age did not save; but also reveals substantial dispersion in the distribution of positive saving across families at each age. As panel C of Fig. 1 shows, those families that did save exhibited commendable thriftiness, saving more than a fifth of their annual incomes (22.8 percent) on average. Indeed, thrift appears to have been the principal architect of saving among working-class households. Differentiating between savers and non-savers at each age reveals small differences in average annual expenditures, but large differences in family income. For example, across the entire sample, non-saving families spent their entire incomes of $429 a year, while families which saved spent, on average, $442, despite their much larger incomes of $575. Among families headed by a 30 year old, non-savers spent $482, while savers spent $462 (out of $600); for families headed by 40 yearolds, non-savers spent $501 and savers, $501 (out of $648). On average, savers at all ages tended to have larger incomes, but chose remarkably similar levels of annual consumption expenditures. The finding that the majority of families at the end of the nineteenth century did not save, while a small proportion saved a great deal, may suggest that the wealth distribution among working-class households was very unequal. However, this inference is warranted if (and only if) families maintained the same relative positions in the saving rate distribution over their lifetimes. Only then does it follow that if the median family did not save in any given year, then the median family accumulated no wealth over its lifetime; only then is a highly skewed wealth distribution determined by the empirical finding that the top 10 percent of working-class families saved 32 percent of their annual incomes in any given year. On the other hand, a high degree of mobility across the saving rate distribution for each workingclass family from year-to-year is capable of generating a more equitable distribution and modest, positive wealth-holdings at all ages. This could occur if incomes were highly variable from year-to-year, such that households would accumulate assets in good years, while saving nothing (and decumulating) in lean times.11
134
JOHN A. JAMES ET AL.
Thus, two opposing inferences (and many in-between) are consistent with the cross-section evidence of Fig. 1. First, many households might never have saved at all during their entire lifetimes, while many others saved frequently (and, on average, a lot every year). Such strong persistence in household saving would generate a highly skewed cross-sectional wealth distribution. Alternatively, many (or all) households might have been episodic savers. They might have saved infrequently, but saved substantial amounts when they did. In this case, saving levels (or rates) at the household level would show little persistence from year-to-year and the overall wealth distribution would be more equitable. In principle, econometric models estimated using true panel data containing many years worth of saving and income data would allow these possibilities to be distinguished, but such data do not exist for the late nineteenth century. Nor is there a direct source for measuring a point-in-time wealth distribution for this period. Our methodology, therefore, involves the application of simulation techniques to build up a wealth distribution from the household survey data, calibrated to fit the microeconomic evidence on household structure and incomes and incorporating appropriate estimates for the persistence of incomes and saving rates.
2.3. A Representative Sample? Before describing the simulation model in detail, we should discuss the issue of representativeness. Our data are necessarily fragmentary – there were no systematic surveys of worker saving in this period, which provide the sort of microeconomic data we require. The nearest equivalents are the federal BLS survey of 1889/1890, which was limited to high-income households in a few selected industries for a single year, and the 1900/1902 survey, designed to be ‘‘more comprehensive’’(US Commissioner of Labor, 1904, p. 11), but which did not report household-level information (see note 5, above). Similarly, while many states participated in the worker survey movement, only a handful produced useable data on saving. We therefore readily acknowledge the regional limitations of our data set, but argue that it is an occupational hazard of working with micro-economic data in this period. We can, however, test the extent of bias inherent in the regional limitation by using data from the federal Cost of Living Survey of 1900/1902, which covered 33 states and which collected schedules ‘‘in almost exact proportion to the number of industrial employees’’ in each geographical division. The survey reported state means of household incomes and ‘‘calculated savings,’’
135
Low Saving Americans?
constructed from the difference between reported income and expenditure. In order to determine whether Kansas and Michigan, which dominate our data set, were unusual states among the 33 covered, we ran weighted OLS regressions of saving levels on the log of income (y) and dummies for the two states. It is possible to run the regression for all households and for renting households only. The results were substantially same (t-statistics in parentheses): All households :
Saving ¼
769:56 þ 127:67 y þ 2:45 Kansas ð 2:85Þ ð0:04Þ ð3:04Þ
8:59 Michigan ð 0:32Þ
R2 ¼ 0:17; n ¼ 33 Renters only :
Saving ¼
613:55 þ 99:91 y þ 5:73 Kansas þ 0:16 Michigan ð 2:46Þ ð2:59Þ ð0:04Þ ð0:01Þ
R2 ¼ 0:11; n ¼ 33 Note that the state dummies are not statistically (or numerically) significantly different from zero.12 On this basis, neither Kansas nor Michigan were unusual savings locations.13 Worker surveys by their very nature are not pure random samples. The Michigan BLS selected specific industries for analysis; workers in other sectors were not included in the sample. The other state surveys in our data set were answered by mail; respondents are unlikely to be a random sample of the entire population. As the discussion in Appendix A makes clear, comparison with the public use sample of the 1910 Census (the first that permits us to distinguish employees in the labor force) indicates that our data set oversamples workers in middle age, workers in manufacturing, and undersamples unskilled workers. The age distribution is of limited importance, since our simulation procedure explicitly models age. The undersampling of unskilled workers produces an upward bias in the savings rates in our data set, although we should not exaggerate its significance, given the collinearity of skill and income.14 There is no a priori expectation of bias in the over-sampling of manufacturing workers, however; this is a purely empirical issue. Evidence internal to our data set suggests, again, that our data set overstates modestly the level of savings among working-class households – regressions of savings levels on a range of variables including industrial dummies reveal that manufacturing-workers saved relatively more than other groups.15 This might well be explained by the tendency of workers in manufacturing to undertake higher levels of precautionary saving in response to higher levels of income instability caused by more prevalent unemployment than in the transportation, trade, and service sectors.
136
JOHN A. JAMES ET AL.
The overall conclusion is that there is a small but discernable upward bias in saving by the households in our data set. The simulation results are therefore likely to produce levels of wealth accumulation that are somewhat too high at the mean, as well as understating the number of households at the bottom of the wealth distribution (thus, the median is likely to be biased upward by somewhat more than the mean). However, we should not exaggerate the extent of such upward bias – a priori considerations, as well as evidence both internal and external to the data set, indicate that the divergences in saving levels, after taking age, income, the number of children, and other modeled variables into consideration are likely to be small.
3. A SIMULATION MODEL FOR HOUSEHOLD SAVING AND WEALTH ACCUMULATION 3.1. General Simulation Strategy This paper applies a simulation model to estimate the cross-section distribution of wealth by age based on the income and saving information contained in our database (see Chart 1). We simulate life histories of annual saving for a sample of hypothetical families and then use these simulated life-cycle saving patterns to generate life-cycle wealth profiles. Our procedure generates several variables for 10,000 hypothetical families at each age of the household head (from 15 through 65 years) designed to be representative of working-class families in the U.S. at the end of the nineteenth century. The simulated variables include family structures (marital status and family size), family incomes, saving rates, housing equity, and financial wealth. At each of the five steps, we calibrate the simulation model to match observed patterns in the household data as closely as we can. 3.2. Simulating Family Structures The first part of the simulation model generates a distribution of marital status and numbers of children consistent with the historical survey data. These particular family characteristics are useful because they provide strong explanatory power for family income and saving levels in the survey data. For each (hypothetical) family in each year, we determine whether unmarried heads of household marry; whether an additional child is born to married couples; and, among relatively older households, whether a child
Age=15 Marital status= single Income= $120
If head is not a homeowner and financial wealth accumulation> 1.25 times annual income, random draw to determine whether house is purchased this year
Family Structure If head is single, random draw to determine whether marries this year If head is married, random draw to determine whether has a child this year Or, random draw to determine whether child leaves the household this year
Income Moffitt model generates household income, based on household structure, lagged income, birth year polynomial, and random error term draw
Chart 1.
Update age by one year
Low Saving Americans?
Housing Wealth If head is homeowner, home equity increases by amount of mortgage payment. Equity appreciates by 1.5% per year.
Initial Conditions
Financial Wealth Saving, net of mortgage payment if homeowner, is added to accumulated past saving
Saving Within cell defined by age and income, random draw to determine position in intra-cell saving rate distribution If draw<λ, household remains in same relative position in saving rate distribution as last year If draw>λ, then a new random draw to determine position in intra-cell saving rate distribution
Distributions of Annual Saving Rates by Age. 137
138
JOHN A. JAMES ET AL.
leaves the household. Age-specific marriage and fertility probabilities are calculated from our survey data. These relatively simple modeling procedures allow the simulation sample to match family characteristics of the survey data set almost precisely. Thus, a typical simulation generates a marriage probability of 0.9018 for a 50 year old householder, compared to 0.9020 in the BLS data set; a typical simulation produces an average family size of 3.130 for a household headed by a 30 year old, compared to 3.128 in the BLS data set.
3.3. Simulating Persistent Family Incomes over the Life Cycle Secondly, we generate annual household incomes for each family in the simulation. Incomes are likely to exhibit substantial autocorrelation over the life cycle (see, inter alia, Lillard & Willis, 1978; MaCurdy, 1982; Abowd & Card, 1989; and Carroll, 1992 for evidence relating to contemporary American workers and families). Our simulation procedure explicitly captures persistence in income levels over time, as well as the effects of family structure on income levels. The relationship to be estimated is a first-order autoregressive model for the natural log of income (y) using data for families indexed by i and survey years indexed by t: yiðtÞt ¼ ayiðtÞt
1
þ f X0 iðtÞt b þ Z0 iðtÞ g þ iðtÞt
(1)
The lagged endogenous variable creates a problem both econometrically and practically (since we do not have regularly spaced surveys). We therefore follow the technique described by Moffitt (1993) for estimating restricted dynamic models using only repeated cross-section data. The twostep method begins by estimating income in the current period using a mix of time-invariant and time-varying explanators, which take on known values in the past, most notably higher-order polynomials in age; this regression is then back dated and used to predict income in the previous period. The second stage estimates the degree of persistence in incomes over time (given by the parameter a) by regressing current income against predicted past income as well as time-varying (X) and time-invariant (Z) variables as in (1). This technique is spelled out in more detail in Appendix B. We ran the Moffitt model over two data sets derived from the BLS surveys. The first included only those data sets for which evidence was available on both earnings and saving (designated with an asterisk in Table A1); the second included all those surveys for which incomes were reported (this
Low Saving Americans?
139
increases the sample size to over 43,000 observations; see the complete list in Table A1). Table B1 contains parameter estimates for each data set. In the first-stage regression, age terms and sex are statistically significant predictors of earnings; the coefficients are stable across the two data sets. In the second stage regression, we estimate a to be 0.835 in the smaller data set and 0.834 in the larger. The finding of strong first-order autocorrelation in household incomes over the life cycle is consistent with historical evidence derived from identical units in successive years. Thus, Friedman (1957, p.190) reports ‘‘0.83 as a reasonably typical value’’ for non-farm consumer units, on the basis of data for various consecutive years between 1929 and 1948; Parsons (1978, p. 555) reports a slightly higher figure (0.907) for earnings autocorrelation for 45–54 year-old high school graduates in the mid-1960s.16 To simulate current incomes for each family during each year, we use the econometric estimates of a, b, and g, simulated family structures, and lagged simulated incomes and add random errors for e. We draw the errors from a normal distribution with constant variance across age. The pattern of simulated incomes, by quintiles at each age, is shown in panel B of Fig. 2; the distribution of actual income in panel A. It is apparent that the simulation module effectively tracks not only the means and standard deviations at each age, for which regression models typically work quite well, but the entire distribution of age-specific income. Family incomes in our data are, therefore, well-described by a first-order autoregressive process with independent and conditionally homoskedastic, normally distributed errors.
3.4. Simulating Persistent Rates of Saving over the Life Cycle The third step, simulating saving rates – saving as a fraction of income – for each family in the simulated economy for each year proved to be rather more difficult, for two reasons. First, we needed to ensure that the simulated economy inherited the most important empirical features of the historical household data. However, regression-based simulation models (similar to (1)) were unable to generate skew in the age-specific distributions of annual saving rates reasonably close to what appears in the survey data. The simulations based on estimated autoregressive models match average and standard deviations of household saving rates by age well, but they cannot simultaneously replicate a distribution in which almost 60 percent of families save nothing, while many others save more than 20 percent of their current incomes.17 We experimented with quite complex polynomial
140
JOHN A. JAMES ET AL. 20th percentile 60th percentile
40th percentile 80th percentile
800 600 400 200 0 15
25
35
45 55 age A. Quintiles of Actual Income
20th percentile 60th percentile
65
40th percentile 80th percentile
800 600 400 200 0 15
Fig. 2.
25
35
45 55 age B. Quintiles of Simulated Income
65
Distributions of Actual and Simulated Annual Household Incomes by Age.
functions in age and income, as well as spline functions, but evidently all were too parsimoniously specified to explain the distribution of saving in the base data set. We develop instead a non-parametric procedure to ensure that the joint distribution of age, income, and household saving exhibited in the data is captured by the simulation. Appendix C describes the method in more detail; an overview is offered here. The simulation procedure for annual saving rates involves randomly drawing a position in the saving rate distribution within a ‘‘cell,’’ determined
Low Saving Americans?
141
each year by the intersection of the family’s age bracket and income bracket. Relatively strong persistence in income levels at the household level through simulation of Eq. (1) implies some degree of persistence in income brackets from year-to-year. This, in turn, implies some indirect persistence in saving rates from year-to-year, as long as saving rates respond positively to current income, which is the case in the underlying data. Our non-parametric simulation procedure for annual saving rates, conditional on age and income, allows a second, direct manner in which household saving rates persist over the life cycle. To incorporate direct persistence in household saving rates over time, we introduce a parameter, l, taking a value between 0 and 1 (constant across families), which represents the degree to which this year’s position in the saving rate distribution depends on last year’s. The model is constructed such that if l ¼ 0:0; household saving rates are independent over time, given current age and income. Higher values of l imply greater degrees of persistence in families’ positions in the saving rate distribution, conditional on their ages and income drawn, each year. When l ¼ 1:00; each simulation family occupies the same relative position in the saving rate distribution, conditional on its age and income, each year. Our non-parametric model tracks age-specific average savings behavior quite accurately, and is also very successful at fitting the skewed crosssection distributions apparent in the raw data (this is true no matter what level of l is selected). The non-parametric simulation model accurately predicts that more than half of families at each age report zero saving each year, while a minority of families at each age save substantial amounts of their annual incomes and only a few run deficits each year (detailed results available on request). Simulations based on different values of l imply different cross-section distributions of accumulated wealth even though all will match the same cross-section distribution of saving rates in the historical data (subject to small approximation errors). This is most evident at the extremes. Thus, when annual saving rates are completely independent across years ðl ¼ 0:0Þ given age and family income, simulations yield measurable wealth accumulations at the lower end of the distribution, compared to l ¼ 1:0; when each family effectively maintains a fixed position in the savings hierarchy (subject to random shocks to income) and the bottom half of the distribution accumulates no wealth. However, what is true at the extremes is much diluted at reasonable intermediate rates of saving persistence, as Fig. 3, which shows age-specific cross-section distributions of wealth at different values of l, demonstrates. Based on the graphs of age-specific quintiles (20th, 40th, 60th, and 80th percentiles), dispersion in accumulated wealth across families of given ages
142
JOHN A. JAMES ET AL. 20th percentile 60th percentile
40th percentile 80th percentile
5000
4000
4000
3000
3000
Dollars
Dollars
5000
2000
20th percentile
40th percentile
60th percentile
80th percentile
2000 1000
1000 0
0 15
25
35
45
55
15
65
25
35
45 age B. Lambda= 0.25
age A. Lambda= 0.10 20th percentile 60th percentile
5000
40th percentile 80th percentile
4000
65
40th percentile 80th percentile
4000
3000
Dollars
Dollars
20th percentile 60th percentile
5000
55
2000 1000
3000 2000 1000
0
0 15
25
35
45 age C. Lambda= 0.50
55
15
65
20th percentile 60th percentile
25
35 45 age D. Lambda= 0.75
55
65
40th percentile 80th percentile
5000
Dollars
4000 3000 2000 1000 0 15
Fig. 3.
25
35
45 age E. Lambda= 0.90
55
65
Comparisons of Simulated Wealth Distributions by Age Across Various Values of the Parameter Generating Saving Rate Persistence.
clearly increases as l rises; it is, however, only as l approaches 0.9 that wealth-holding by the lowest quintile falls toward zero. Further, even at very low levels of l, households at the 20th percentile are weak accumulators, with small wealth-holdings by age 65. To determine the value of l that best characterizes saving behavior in our historical survey data, we compare estimated autoregressive regression models based on the simulation saving rate data with estimated (Moffitttype) models based on the actual data. We first estimate an equation like (1)
Low Saving Americans?
143
for saving rates using Moffitt’s estimation strategy on the actual repeated cross-section survey data. Results in column 3 of Table B1 show the estimated autocorrelation coefficient for current and lagged annual household saving rates to have been 0.865. Since our simulation program generates savings and income histories for the hypothetical households, we can estimate first-order autoregressive models on simulated saving rates in ‘‘true’’ panel data sets for selected values of l.18 We choose a value of l that implies a level of autocorrelation close to that estimated in the raw data (0.865); the best choice is l ¼ 0:80: Fig. 4 compares actual and simulated savings levels for the 70th and 90th percentiles of the savings distribution at each age (for l ¼ 0:8). These are remarkably similar (of course, for all deciles at and below the median, the actual and simulated savings levels will be identical, at zero). To summarize, the simulation procedure incorporates two sources of persistence in annual saving rates. First, current saving rates depend on current income in the model, which itself is modeled using a persistent process for each family. Second, conditional on current (random) income and age, annual saving rates are autocorrelated through an additional stochastic component for each family.
3.5. Home Equity and Wealth Accumulation over the Life Cycle The worker surveys only reported saving out of current income; they do not explicitly include increases in wealth due to accumulation of equity among homeowners or other unrealized capital gains. It was therefore necessary to incorporate corrections in our simulation for wealth accumulation from these additional sources. The key issue here is homeownership, which was a primary form of wealth accumulation in the late nineteenth century, just as it is today. We therefore modeled equity accumulation explicitly to consider the relative importance of home equity and non-housing (or, for simplicity, financial) wealth in household net worth. Typical home mortgages around 1900 consisted of short-term contracts in which interest only was paid over the duration of the loan, with balloon payments of principal due at maturity (Snowden, 1995; Rotella & Alter, 1995). The accumulation of home equity or the paying down of principal at maturity was therefore accomplished only through prior saving. Our measure of saving already includes the build-up of home equity through the payment of principal. However, in view of the evidence on the real appreciation of house values (Hoover, 1960; Rees, 1961; Long, 1960), we modeled
144
JOHN A. JAMES ET AL. actual
250
simulated
Dollars
200 150 100 50 0 15
25
actual
250
35
45 age A. 70th percentile
55
65
55
65
simulated
Dollars
200 150 100 50 0 15
Fig. 4.
25
35
45 age B. 90th percentile
Comparison of Distributions of Actual and Simulated Annual Saving Levels by Age, Deciles 70% and 90%.
home equity to appreciate more rapidly than financial wealth. From 1865 through 1890, the relative price of housing rose at a stable 1.5 percent annual rate (compared to non-housing consumer prices). Several state BLS reports published information on home ownership, and we use this information in turn to build a module to simulate home equity over the life cycle.19 Workers in our model consider purchasing houses that cost 2.5 times their annual incomes, a figure consistent with the BLS reports. At this time mortgages required sizable down payments – usually one-half or more of property value. In our model, therefore, workers become eligible
Low Saving Americans?
145
to purchase a house after accumulating sufficient savings for a 50 percent down payment (or 1.25 times their annual incomes). Not all eligible families, however, actually purchase a home; rather, the purchase of a home by an eligible family in a given year is a random event.20 We choose age-specific probabilities of home purchase among eligible families to replicate the observed proportion of homeownership in the data: a third of all working-class families owned their homes; by age 55, two-thirds were homeowners.21 Mortgages typically ran between 3 and 5 years, usually being renewed or rolled over at maturity. We assume mortgage holders to reduce principal at a constant rate over 20 years; this is consistent with repayment terms calculated by Rotella and Alter (1995). Financial wealth distributions were generated by simply summing past saving levels (the product of saving rates and income levels). We do not compound past saving levels when computing accumulated financial wealth levels. Only a small proportion of households owned accounts in mutual savings banks or other financial institutions (e.g., fewer than 10 percent of Michigan’s furniture-workers in 1889 ) – most savings must have been held as cash in this period. Compounding past saving, moreover, would be appropriate only for assets yielding unrealized capital gains. Our home equity simulation module explicitly incorporates the most important case. Finally, as described thus far, nothing in the simulation model prohibits families from accumulating unbounded negative levels of wealth through frequent dissaving. Although actual families might have been able to run temporary deficits, financed by store credit, loans from relatives, or pawnbroking (see Rotella & Alter, 1993), there must have been limits to the extent of household deficit finance. We do not therefore allow simulated families to have overall negative levels of wealth.22 Prohibiting negative total net worth (home equity plus financial wealth) in the simulations affects the location of only the very bottom (less than 5 percent) of families in the wealth distribution by age.
4. SIMULATION RESULTS FOR WEALTH ACCUMULATION Panel A of Fig. 5 reproduces the age-specific distribution of accumulated wealth for households near the end of the nineteenth century, as derived by our simulation model using l ¼ 0:80: The figure shows the simulated accumulation profiles for the 20th, 40th, 60th, and 80th percentiles. In panel B
146
JOHN A. JAMES ET AL. 20th percentile 60th percentile
5000
40th percentile 80th percentile
4000 3000 2000 1000 0 15
25
35
45 55 age A. Distribution of Total Wealth
20th percentile 60th percentile
10
65
40th percentile 80th percentile
8 6 4 2 0 15
25
35
45 55 age B. Distribution of Wealth-Income Ratio 20th percentile 60th percentile
6
65
40th percentile 80th percentile
5 4 3 2 1 0 15
Fig. 5.
25
35
45 55 65 age C. Distribution of Financial Wealth-Income Ratio
Simulated Distributions of Household Wealth by Age by Quintile.
Low Saving Americans?
147
the same distribution is graphed as a proportion of average income for the entire sample; these calculations are reported to facilitate comparison with data on household accumulation for the late twentieth century (see Fig. 8). Panel C of Fig. 5 displays simulated ratios for financial wealth (i.e. net worth less than home equity) to income, an appropriate reference if illiquidity mitigates the use of housing wealth as a means to finance old age. Our analysis reveals some wealth accumulation by nearly all workingclass American families a hundred years ago, particularly in the form of home equity. Median wealth families accumulated about 3 years worth of income in net worth by age 55; about half that amount was in the form of housing wealth (note, however, that home equity fell as a proportion of total wealth among the highest accumulators). Below median wealth, however, families accumulated quite modest amounts of financial assets during their working lives, often less than an average year’s worth of income. Are these results realistic for this period? The only benchmark that we have found for wealth-holdings among working-class households at the end of the nineteenth century are figures for present worth reported by the sample of Michigan iron-workers in 1890.23 This sample is too small to construct reliable estimates of the wealth distribution by age (which would have precluded the need for simulation), but age-specific means can be usefully compared with our simulated data, as Fig. 6 shows. Allowance was made in the construction of this figure for the differences in family structures and incomes between the two samples by using regression techniques to reshape the Michigan data to mimic the demographic structure of the simulated families.24 The figure shows that these two series generate similar wealth levels among the youngest and oldest families, but that the ironworkers reported somewhat larger wealth values during middle age.25 We can also compare our results to other historical reconstructions. The estimate at l ¼ 0:8 of mean wealth accumulation of $2,925 at 65 years of age is reasonably close to Rotella and Alter’s (1993) figure of $3,211, derived from the 1889–1990 Federal BLS survey, after correcting for inflation (prices in 1900 were about 8 percent higher than in 1890) and given the upward bias in their base data. Our figure also comports well with Gratton and Rotondo’s mean figure of $5,000 in 1918 prices ($2,772 in 1900 prices) for the 30 years centered on 1900 (1991). The crucial parameter in the transformation from cross-section saving distribution to cumulative wealth distribution is the value of the saving persistence parameter, l. We have already suggested that the shape of the wealth distribution is quite insensitive to the precise choice of l within reasonable ranges. Table 1 supplements the evidence of Fig. 3 on this point,
148
JOHN A. JAMES ET AL. simulated families
Michigan families
2000
Dollars
1500
1000
500
0 15
Fig. 6.
35 Age
45
55
Comparison of Average Actual and Simulated Wealth from the Sample of Michigan Iron-workers.
Table 1. Savings Persistence Parameter (l) 0.10 0.25 0.50 0.75 0.90
25
Wealth Accumulation by American Working-Class Households (1900 prices). Wealth of Mean Household ($)
Wealth of Median Household ($)
2,816 2,825 2,829 2,910 2,956
2,601 2,603 2,563 2,553 2,274
Source: Estimated from simulation procedures described in the text.
by showing how little both mean wealth and median wealth at age 65 vary with different savings persistence parameters, from 0.1 to 0.9. Mean wealth varies by less than 5 percent between these extremes; the wealth of the median household is more sensitive to differences in the persistence parameter, but it is only at very high levels of l that median wealth begins to fall
Low Saving Americans?
149
significantly. The more persistent are saving rates, ceteris paribus, the smaller is the proportion of families that accumulate, and the more skewed the wealth distribution among working-class households. Not that all low wealth-holders were necessarily inconsistent savers. Many families approached old age with modest amounts of wealth because they offset past saving with occasional dissaving. Evidence for such patterns appears in Fig. 7. We depict lifetime saving histories for four randomly selected families from the bottom of the wealth distribution at age 55 (two drawn randomly from the 10th percentile and two from the 25th) and four from the top (two from around the 75th percentile and two from around the 90th). None of the households selected from the bottom part of the wealth distribution experienced non-positive saving rates uniformly across years. The low wealth-holders here all show some years in which saving was above zero. Symmetrically, none of the selected high wealth-holders saved consistently in every year: in the simulation, all of these families experienced several years in which spending equaled or exceeded current earnings.26 Thus, these simulated profiles reinforce the argument that cross-section evidence on the savings behavior of a particular household, as derived from a single worker survey, provides an imperfect guide to its lifetime accumulation pattern.
5. COMPARING HISTORICAL AND CONTEMPORARY DISTRIBUTIONS OF WEALTH How do our results for the end of the nineteenth century compare to the modern profile of saving rates and wealth accumulation among workingclass American households? Comparative evidence for saving behavior is provided in Table 2. For the modern period, we have constructed distributions of annualized saving rates from the Survey of Consumer Finances panel data covering 1983–1989 for a group of households that conform to our historical data.27 Saving rates were calculated as the annualized change in net worth between 1983 and 1989 (less estimated capital gains), divided by average family income (in 1982 and 1988). The average saving rate for contemporary households is somewhat below the rate of 9.1 percent in the historical data set. However, the greater access to borrowing against future income for today’s families generates a bigger and longer left-tail in the distribution, creating a downward bias relative to the credit-constrained families of the Gilded Age. Limiting the
200
100
100
150
200
0
100 dollars
dollars
dollars
200
0
-100
-100
-200
-200 15
25
35 45 age 10th percentile person
55
0 -100 -200
15
25
35 45 age 10th percentile person
55
15
25 35 45 age 25th percentile person
55
15
25 35 45 age 75th percentile person
55
200
0 -100
dollars
200 dollars
dollars
100
100 0 -100 -200
-200 25
35 45 age 25th percentile person
55
15
25 35 45 age 75th percentile person
dollars
200 100
200 100
0
0
-100
-100 -200
-200 15
Fig. 7.
55
25
35 45 age 90th percentile person
55
15
25 35 45 age 90th percentile person
55
Lifetime Saving Outcomes for Some Randomly Selected Families from the Simulation Across Positions in the Wealth Distribution.
JOHN A. JAMES ET AL.
dollars
15
200 100 0 -100 -200
151
Low Saving Americans?
Table 2. Household Saving Rates in the Historical Survey Data and in the 1983/1989 Survey of Consumer Finances Panel Data. Proportion of Families in Various Saving Rate Intervals Age group
Saving rate range(%) o0.02
0.02–0.10
0.10–0.25
40.25
(A) Historical Household Survey Data on Working Americans (c. 1890) 20–29 30–39 40–49 50–59 All
57 54 58 60 59
7 7 7 6 7
20 22 19 17 19
16 17 16 17 15
(B) 1983/1989 Survey of Consumer Finances Working-Class Families 20–29 30–39 40–49 50–59 All
61 54 39 47 51
19 23 22 27 22
4 14 19 14 12
16 9 20 12 15
Note: For the SCF, families were selected if respondent has a high school degree; at least twothirds of family income comes from wages and salaries; average family income less than $100,000 per year. Saving is computed as the change in net worth between 1983 and 1989 less capital gains. Total family income is averaged between 1982 and 1988 values.
comparison to non-negative savers produces almost identical saving rates in the two periods. Table 2 further shows that the distribution of saving was very similar in the two periods. Panel A shows the proportion of families in the BLS data set for each of four saving rate intervals at each age; panel B reproduces the same data for the SCF panel. The table indicates that the median saving rate for all families in this population (high school graduates only) was zero in both periods, although the proportion of zero savers was higher at middle ages in the historical data than today. Similarly, the proportion of high savers (425 percent) was much the same overall in each of the samples, although the specific distribution by age shows more variability among recent households. The most notable difference between these two series is in the proportion of small savers (measured as 42 and o10 percent), which is considerably lower in the historical data set than in the SCF data. This may indicate that very small amounts of saving (o $50) went underreported in the BLS surveys, which may, in turn, suggest that the extent of zero savings
152
JOHN A. JAMES ET AL.
is (somewhat) over-reported. However, the most remarkable aspect of the table is the considerable similarity between the historical and contemporary saving profiles among working-class households. Our contemporary reference point for wealth-holding and its distribution among working-class households is the sample constructed by Hubbard, Skinner, and Zeldes (1994, 1995), subsequently HSZ, from the 1984 wave of the PSID.28 HSZ split their PSID sample into three groups based on the educational attainment of the household head to control for differences in wealth accumulation across different levels of lifetime income. They use three education categories for household heads: without a high school diploma (28 percent of their sample); with a high school diploma (52 percent); and with a college degree (20 percent). We compare our simulated distributions for late nineteenth-century workers to HSZ’s 1984 sample of families headed by a person with a high school degree. Self-employed persons and employers, who resided in the upper end of the lifetime income distribution a century ago, represent approximately the same proportion of all households (17.4 percent of the workforce in the 1910 public use manuscript census sample), as do college graduates today (20 percent of HSZ’s PSID sample). Further, our data tend to underrepresent unskilled wage-workers relative to skilled workers. Unskilled workers comprise only 15 percent of our sample, but about 33 percent of the workforce according to the public use manuscript census samples of 1900 and 1910. By excluding those who left high school without graduating, we are able to balance the undersampling of unskilled workers in the simulations when making comparisons to contemporary working families. Overall, our simulation model yields age-specific distributions of wealthto-income remarkably similar to those obtained by HSZ (1995) among respondent families with a high school degree in the 1984 PSID.29 In particular, comparisons between our Fig. 5B and their Fig. 1 (Panel b, 1995, p. 367) do not suggest uniformly greater wealth accumulation across the distribution before the advent of federal social insurance programs. At the bottom of the savings distribution, they identify a significant proportion of households who accumulated almost nothing before retirement. Among high school graduates, 15.6 percent of households whose head was aged 50–59 held wealth less than their annual income. Our data find that households in the bottom quintile of the late nineteenth century income distribution also accumulated very modest amounts of wealth; the wealthholdings of a household at the 20th percentile was about 70 percent of income at age 60. Unsurprisingly, perhaps, given the significantly higher income levels and the expansion of saving opportunities, families at the
Low Saving Americans?
153
upper end of the distribution tend to accumulate at a faster rate now than then. Thus, in 1984 the ratio of wealth to income at the 80th percentile at age 60 was about 8; in our simulation, it was 6.75.30 Within these limits, the similarities between the two data sets are remarkable. Thus, in 1984, the ratio of median wealth to median income for households headed by a high school graduate aged 50–59 was 3.4 – precisely the same figure produced by our simulation procedure for that age group a century earlier. Conceivably, different results might follow from excluding home equity from the comparison of historical and contemporary wealth accumulation among working-class American families. This might occur if retirement, social insurance and welfare expenditure programs initiated by the federal government during this century affected families’ decisions about financial wealth accumulation, but not home ownership or equity. However, we find no dramatic differences between the two distributions of non-housing wealth by age (we compare our Fig. 5C to one generated by HSZ shown in Fig. 8).31 Among contemporary working-class American families, approximately one-fifth accumulate zero financial assets during their prime working lives; much the same result obtains for our simulation model. At the upper end of the distribution, working families accumulated non-housing wealth to about 5 times their (permanent) income, then and now.32
Fig. 8. Non-housing Wealth by Age, as a Fraction of Permanent Income: High School Graduates in 1984. Lines indicate 20th, 40th, 60th, and 80th quantiles, by age. Source: PSID extract constructed by Hubbard et al. (1995), courtesy of authors.
154
JOHN A. JAMES ET AL.
Thus, saving patterns among working-class American families around the turn of the twentieth century do not suggest substantially greater wealth accumulation relative to contemporary families similarly positioned in the lifetime income distribution. Moreover, sensitivity analysis indicates that this similarity is relatively insensitive to changes in the simulation parameters – even to quite radical changes, as in the case of the value of l, the measure of saving persistence through time, the most important parameter in the model.
6. CONCLUSION How should we characterize working-class American savings behavior a century ago given these findings? The answer, unsurprisingly, is complex. On the one hand, a majority of working-class families in any given year at the end of the last century saved none of their incomes at all. On the other, households who did not save in the current year had probably saved previously and would quite likely save in the future. Furthermore, when families did save, they tended to set aside substantial amounts of income. Nonetheless, our calculations reveal that a significant proportion of the working-class population (over 20 percent) accumulated almost no financial wealth during their lifetimes – almost the same proportion as observed in the contemporary U.S. Moreover, at the upper levels of the (middle-income) wealth distribution, saving appears to have been sufficiently prevalent, frequent, and sizeable to generate very similar levels of accumulation to those of the upper echelon of working-class families today, at roughly all parts of the distribution by age.33 Indeed, the most striking aspect of our comparison between wealth distributions at the beginning and end of the twentieth century is the remarkable similarity they show among families in the broad center of the income distribution. This finding of continuity should be placed in the context of interpretations that attribute low savings rates and wealth-holdings by families at the bottom of the income (and wealth) distribution in contemporary society to the deleterious impact of social insurance programs on the incentive to save. The policies most frequently indicted as responsible for low household saving rates include: 1. universal Social Security retirement benefits, considered to crowd out private saving and wealth accumulation otherwise needed to finance expenditures during old-age (e.g., Feldstein, 1974, 1996);
Low Saving Americans?
155
2. mandated social insurance programs, such as Medicare and Unemployment Insurance, which reduce the exposure of American families to economic risks and, thus, may reduce precautionary incentives to save (e.g., among others, Carroll and Summers, 1987; Kotlikoff, 1988; Hubbard et al., 1995; Engen & Gruber, 2001); 3. resource-tested welfare programs, such as AFDC, Food Stamps, and Medicaid, which imply effective tax rates up to 100 percent on previously accumulated wealth for potentially low-income families due to spenddown eligibility provisions (Hubbard et al., 1995); 4. high effective marginal tax rates on capital gains and income under the progressive corporate and personal income tax statutes (e.g., Poterba, 1994). None of these public policies existed before 1917. Yet even in their absence, low savings and minimal wealth accumulations were the norm for much of the American working class: low-income households saved just as little in an era before any of these programs and policies were in mind, let alone in place. If it is true that American workers have always been low savers (further evidence from the 1920s and 1930s is provided in Thomas, James, & Palumbo, 1999), a further irony emerges – namely that the declining personal savings rate over the twentieth century originated in the upper echelons of the income distribution – among professionals, the selfemployed, stock-holders, and capitalists – those generally ignored in the blame for the ‘‘American savings crisis,’’ and whose behavior is difficult to link to the disincentive effects of social programs that do not benefit them proportionately.34 Obviously, it is not enough to contrast two moments in time, point to similarity of outcomes, and assert that American savings behavior has been unaffected by policy. Policy responses posit counterfactual, rather than historical, comparisons. We would need to know what working-class savings would have been in the absence of social security receipts, for example, to be able to determine the effect of the New Deal program on thrift. But the historical perspective does raise important questions. Most obviously, why do we not observe more widespread wealth accumulation in the era before social insurance? It is not enough to suggest that savings are income elastic and that households then saved less because of lower household incomes. We also need to understand how families at the bottom of the income ladder could prepare financially for old age and retirement and manage through volatile circumstances in the absence of private funds or public support.
156
JOHN A. JAMES ET AL.
One response to these skeptical inquiries would be to argue that workingclass households had fewer incentives to save at the end of the nineteenth century. Within the life-cycle model, the most obvious elements influencing wealth accumulation are: (i) life expectancy; (ii) retirement patterns; (iii) intergenerational transfers (reverse bequests). Savings will be lower, ceteris paribus, the lower is life expectancy; the less frequent is retirement from wage-earning; and the larger are transfers of income from young to old. Within the precautionary model, the key elements are: (i) probability of permanent disablement from income earning (accident, morbidity); (ii) probability of extended life beyond normal expectation. The higher these probabilities, the more incentive to save. What does the historical record indicate? Life expectancy was certainly lower at the end of the nineteenth century than today. Life expectancy at birth among males averaged 47.3 years in 1900 compared to 70.1 in 1980.35 However, most of this difference was due to infant and child mortality; life expectancies for adult males have increased much less over the twentieth century. Thus, for Massachusetts residents in 1890, the average 20 year-old male could expect to live another 40.7 years; the average 40 year-old, another 27.4 years; the average 60 year-old, another 14.7 years.36 These do not, in our judgement, constitute low life expectancies. Moreover, if families in our historical sample were life-cycle savers facing short life expectancy after 55 years of age, then our analysis ought to reveal hump-shaped accumulation patterns before that age.37 However, neither the primary data, nor our simulation results, reveal tendencies for reduced saving, let alone prevalent dissaving, before 55 years of age. Thus, it seems unlikely that low saving at the end of the nineteenth century was a response to low expectation of life. Perhaps workers did not save for old age because they expected to continue in the workforce. Certainly, the traditional account of retirement, in which workers were presumed unable to afford retirement before the introduction of social security in 1935, is consistent with this argument. However, recent analysis of employment transitions using the 1900 public use sample of the U.S. Census has challenged this interpretation. Carter and Sutch (1996, p.17) find that almost 20 percent of wage and salary workers over the age of 55 retired before death; the singulate mean age of those retiring was 67.5 years, affording an average 10.3 years of life without work. But this turns out to be only a minor qualification, since, according to Carter and Sutch, over 80 percent of older workers did not retire at all; moreover, they acknowledge that many of those who did retire may have done so temporarily, only to reenter the labor force at a later date. Lee
Low Saving Americans?
157
(1996) derives an even lower estimated retirement rate – about 4.75 percent for all workers aged 55 years and older in 1900.38 Putting all of these results together would suggest that most workers did not save for retirement because they never expected to retire; rather, an extended expectation of life was accompanied by an extended expectation of work. Further, many who did accumulate with retirement in mind would have been forced to change their minds, as negative income shocks depleted their savings. Thus, low savings could be attributed to a mixture of ex ante and ex post factors – the relatively small number of workers who planned retirement saved at a higher rate than those who did not; while those who accumulated fewer savings had less opportunity to give up work. However, even this is too cut and dried. First, labor force participation data from the BLS surveys indicate that days worked declined throughout the employment distribution after age 55, suggesting widespread ‘‘partial’’ retirement. Second, the probability of enforced retirement because of sickness or disability was significant in this period, implying that there ought to have been a strong precautionary motive for saving even in the absence of planned retirement. In conjunction with relatively high life expectancy (and its inherent unpredictability), labor force patterns of older Americans should have generated higher savings than our simulations indicate. Why then did Americans not save more? One possible explanation may be found in the pattern of incomes of older Americans. The figures produced by the 1889–1890 federal Cost of Living Survey suggests that Americans did not experience dramatic income decline in old-age (Haines, 1979). Indeed, households headed by older workers (55 years and above) had the highest average per capita income of any age group in the survey (at $216 compared to $174.2 for all other households).39 Partly, this was due to the smaller family size of older households; but it also reflects the changing composition of family incomes. Almost half of the income of households headed by 60+ year-olds came from children or boarders; in contrast, over 90 percent of incomes for younger households (20–39 years old heads) was earned by the household head (Rotella & Alter, 1993, p. 115).40 Adult children who remained in the home subsidized their aged parents; the rooms of the children who left were rented out. Child rearing constituted a form of asset accumulation in response to life-cycle and precautionary (income insecurity) concerns; nearly half of retired men aged 64 years or older lived with their children in 1880 (Costa, 1997). By 1950, this cohabitation rate had fallen to 20 percent. Moreover, given the economies of scale of household management and the relative risks inherent in raising a family and maintaining a monetary ‘‘nest-egg’’ in an era of substantial economic instability, it may
158
JOHN A. JAMES ET AL.
have been a more efficient method of providing for old age and its uncertainties than financial accumulation.41 If a substantial fraction of working-class families did count on private intergenerational transfers in the past, this raises important questions about the impact of social programs on savings. Most significantly, it suggests that public social security may simply have replaced private social security. What is particularly important about this possibility is that both private and public retirement programs are financed on a pay-as-you-go basis. Thus, one would not expect public social security to ‘‘crowd out’’ private saving, if it in fact simply replaced a private unfunded retirement plan among many working-class Americans.42 In addition, the implication that public social security caused greater reductions in labor supply and major changes in intra-family relations, compared with modest reductions in private saving, seems consistent with, for example, Costa’s (1997) conclusions about how turn-of-the-century workers responded to Union Army Pensions. But these are topics for future research and further modeling.
NOTES 1. A further disadvantage of these studies resides in the source data they use, namely the federal BLS Cost of Living survey of 1889–1890. We argue below that this source is biased toward high-wage industries, therefore overstating the extent of saving by working households; moreover, it relies on a definition of saving that is likely to produce biased estimates. 2. Note that during 1897–1908 non-farm households undertook more than twothirds of all saving (and 93 percent of private sector saving); agricultural households contributed just 1.1 percent to aggregate savings over the same period (Goldsmith, 1955, I, p. 267). 3. Note that year refers to the period of the survey itself, rather than the date of publication; months are expressed as fractions (thus 1884.5 refers to a survey for the year beginning in June 1884, while 1887 refers to a survey begun in January 1887). All the surveys report annual information, collected for the previous 12 months. 4. The chosen surveys were coded either by the Historical Labor Statistics Project at the University of California or by ourselves. 5. In contrast, Rotella and Alter (1993) and Gratton and Rotondo (1991) base their wealth accumulation estimates on the Federal BLS inquiry of 1889–1890, in which the ‘‘surplus’’ was calculated as the simple difference between income and expenditure. As Higgs (1893, p. 263) asked of the ‘‘United States returns, how far are we justified in regarding the frequent ‘surplus on hand’ as an indication of annual saving? y is not the excess of income over expenditure frequently rather ‘unaccounted for’ than a surplus?’’ Under-reporting of some expenses (notably tobacco and alcohol) are a common weakness of budget surveys. In addition, Gratton and Rotondo (1991, p. 343) note that ‘‘interest charges on real estate loans y and certain
Low Saving Americans?
159
other expenses for homeowners’’ are not included in the 1889–1890 survey, such that ‘‘expenditure y is underestimated and surplus is overestimated.’’ For renters alone, the average savings rate in the 1889–1890 survey is 8.8 percent out of head of household income, slightly lower than in our sample; the median savings rate was zero. The average surplus of income over consumption in a similar BLS survey in 1900/1902 of 25,440 households (for which no individual household data are available) was 8.1 percent out of ‘‘husband’s income’’ (the median savings rate cannot be calculated); the figure was substantially lower for renting households (below 6 percent). Both of these surveys were biased toward high-income households. The 1889–1890 survey focused on ‘‘high-tariff’’ industries: average annual earnings (in 1900 prices) was $603, compared to full-time annual earnings of all non-farm employees of $434 (Lebergott, 1964, p. 528). Lebergott (1976, pp. 310–325) argues that the 1900/1902 survey, with average earnings of $613 in 1900 prices, also drew from the upper echelons of non-farm families. Thus, 37 percent of surveyed families had members in labor organizations, compared to less than 3 percent across the entire labor force; while 66 percent of households had life insurance, and 31 percent owned property, which Lebergott considers, ‘‘extremely high proportions even in much later years’’ (1976, p. 321). By contrast, our data indicate much lower rates of unionism (12.2 percent) and life insurance coverage (26 percent); real average earnings in our sample was $491. As noted in Appendix A, it is likely that our data set also exclude those at the very bottom of the distribution – most notably, the casual workers who were unlikely to be captured by on-job interviews. The share of unskilled workers in our data set is 15 percent, compared to 33 percent at the Census of 1900. To that extent, our findings are likely to overstate the extent of worker saving, thereby reinforcing our argument that the median working-class household in any given year was a non-saver. 6. We argue in Section 4 that because of the nature of home mortgages at that time reported savings generally included payments into home equity. One case in which this was not true was the Eighth Annual Report of the Michigan Bureau of Labor and Industrial Statistics (1890), which asked for ‘‘amount saved (aside from payments on home)’’ and also for ‘‘amount paid on home during year.’’ To ensure consistency, in this case we calculated mortgage interest paid from the mortgage remaining at the beginning of the year and subtracted it from the amount paid on the home to get payments into equity over the course of the year. That amount was then added to the other savings figure to give us reported savings. 7. In the BLS survey of 1900/1902, only five out of 1,480 households providing information on ‘‘disposition of surplus’’ reported the purchase of stocks; a further three ‘‘loaned money.’’ 8. The Federal Cost of Living Survey of 1900/1902 (U.S. Commissioner of Labor, 1904, p. 513) reported ‘‘the manner of meeting deficit’’ for 507 households; 244 obtained credit, 94 applied former savings, and 13 borrowed money (150 did not respond, and 6 sold or mortgaged property or employed other methods). 9. This provides another reason not to use data on ‘‘calculated savings,’’ since these will be biased downward by the omission of additional income. As mentioned in Appendix A, the two saving variables are however strongly linearly related among the c. 8,000 families reporting both. One source of income that was never mentioned in the BLS reports was Civil War pensions. Although originally limited to
160
JOHN A. JAMES ET AL.
war-related conditions, in 1890 a new pension law was passed providing pensions to all who served in the Union army for more than 90 days; were honorably discharged; and were disabled for any cause whatsoever. With these relaxed eligibility requirements, the number of men receiving pensions rose sharply after 1890 (by 1900 disability pensions began to be treated as general old age assistance to Union veterans). By 1900, 30 percent of all white males aged 55–59 and 18 percent of those aged 60–64 were receiving federal pensions. The average recipient received $139 annually in 1900, compared with a mean earnings of $491 for the household heads in our data set (Vinovskis 1990, pp. 24–25; Costa, 1995). Because of the irregular and unpredictable manner in which eligibility conditions and payments were extended, however, it seems unlikely that persons not qualified for a pension at the time of their BLS survey (usually in the 1890’s) would have been counting on a future pension benefit and, thus, have had a major impact on their current savings decisions. 10. Adjusting by Williamson and Lindert’s (1980, pp. 323–325) state cost-of-living series, to permit some variation in relative regional prices over time, made no essential difference to the results reported. 11. However, given the historical experience reflected in our data set, it would require considerable independence of household savings decisions over time to generate a wealth distribution that exhibits substantial asset accumulation for the median family. 12. The results for saving rates out of income were essentially the same; in neither case were the state dummies statistically significantly different from zero. 13. Table B1 indicates that the crucial demographic variable influencing savings rates was the number of children. We therefore undertook a formal statistical test of the demographic structure of our data set against Census data (the number of children for married male non-farm heads of households) for the nearest relevant year, retrieved from the IPUMS files. This provided a slight problem – while the Kansas and Oklahoma BLS samples could be related to the 1900 and 1910 Census returns respectively, the appropriate point of comparison for Michigan and Missouri is 1890 – for which no Census schedules survive. We therefore recovered 1890 data by interpolating between the 1880 and 1900 data. We first compared the state samples to their respective Census equivalents, finding that in all cases there was no statistically significant difference in family size on average. We then undertook the more stringent test of disaggregating family size by (5 year) age category; in all but one case (Michiganders over 60, for whom the BLS reported more children than the Census), difference of means tests indicated that there was no difference between the two data sets. We then compared the state Census data to the national average to test regional representativeness. The difference of means tests rejected dissimilarity in all the states except Kansas, which was revealed to have fewer children on average than the U.S. as a whole. The stricter test of family size by age category revealed no statistically significant differences for Kansas, Missouri or Oklahoma; Michigan families in their 30s were, however, shown to have smaller families than the average U.S. household of the same age. Given the negative sign on family size in Table B1, this is consistent with a small upward bias in measured savings rates in our data set. 14. As a test of this hypothesis, we ran a regression of the level of savings on income, skill levels, age, and a number of other variables. The results were as follows (skilled workers are the omitted dummy variable; certain variables omitted,
161
Low Saving Americans? t-statistics in parentheses): Savings ¼
0:10 Income ð15:05Þ
5:12 Children ð 12:23Þ
5:78 Unskilled ð 3:18Þ
6:62 Semi-skilled ð 4:96Þ
3:15 Professional . . . ð 1:74Þ
R2 ¼ 0:32; n ¼ 23; 200
15. A regression of the level of savings on income, industrial sector, age, and a number of other variables produced the following results (transportation is the omitted dummy variable; certain variables omitted, t-statistics in parentheses) Savings ¼ 0:12 Income ð17:08Þ
5:12 Children þ 19:64 Manuf ð 12:30Þ ð13:43Þ
46:26 Construction ð 3:93Þ
83:98 Trade=services . . . ð 9:08Þ
R2 ¼ 0:31; n ¼ 23; 200
16. Friedman (1957, Table 19) reports autocorrelation coefficients of 0.85 for professional workers (1929–1934); 0.83 for Wisconsin taxpayers (1929–1935); and 0.83 for urban families (1947–1948). We also estimated a from our data sets by combining individual observations into birth cohorts; this provides us with annual data for a large number of years. GLS regressions of the log of mean income of cohort j in year t on the log of income in year t 1 produce an estimate of 0.837 ðz statistic ¼ 17:91Þ for pairwise combinations derived from the saving subset of the data; and 0.832 ðz statistic ¼ 7:77Þ for cohorts derived from the income data set. 17. This raises the question of why households may have saved only periodically. One possibility is that savings only took place during periods in which positive net income shocks created transitory income that was out aside against a rainy day; another is that negative net income shocks (due to unemployment, wage cuts, or higher than expected expenditures) depleted savings in one year, generating positive savings in a subsequent year to recover the target or buffer-stock level of accumulated wealth. It is not possible to address these issues within the scope of this paper; they are dealt in some detail in the context of consumption smoothing strategies in James, Palumbo, and Thomas (2004). 18. Durbin-h tests reveal weak, but statistically significant, autocorrelation in the residuals, in which case OLS estimates with a lagged dependent variable are inconsistent. Therefore we use the technique suggested by Hatanaka (1974) when both autocorrelation and a lagged dependent variable are present. An initial estimate of r is constructed from OLS estimates, the variables are transformed and then the equation is reestimated including the lagged residuals from the first stage on the right-hand side. 19. Rotella and Alter (1995) discuss these reports in detail. Although several of the surveys covered only specialized populations, such as Michigan iron-workers, they find the patterns therein generally to be representative to those based on more comprehensive, but more aggregated, data on home ownership and mortgage holding reported in the 1890 census. 20. For simplicity, we assume that families purchase at most a single home during their lives. 21. Based on 1900 Census data, Collins and Margo (2001, p. 70) find 46 percent of white household heads to have been home owners.
162
JOHN A. JAMES ET AL.
22. Households in the simulation may have negative financial wealth at any age only up to the positive value of home equity holdings (implicitly being able to borrow against equity). The financial wealth of non-homeowners then is bounded from below at zero. 23. Data on individual wealth-holding are available for earlier in the nineteenth century – notably from the manuscript censuses of 1850, 1860, and 1870 (e.g. Galenson, 1991; Galenson & Pope,1992; Kearl, Pope, & Wimmer, 1980; Pope, 1989). It is not possible to make direct comparisons of these data with our results since they refer to non-farm males, without distinguishing wage-earners from the self-employed (Soltow, 1975). However, only about 60 percent of all non-farm males over age 30 held positive wealth in 1870 (Soltow, 1975, p. 38; Galenson, 1991, p. 597); the comparable proportion among wage earners should have been even lower. 24. A regression of present worth on age and income polynomials and family composition variables was run on the Michigan data; then fitted values were constructed based on the estimated Michigan coefficients and the age, income, and family characteristics of the simulated families. 25. Note, moreover, that the wording of the worth question in the Michigan survey implies that the stock value of durable consumption goods (e.g., furniture) may have been included in reported present worth. Durable stocks are excluded from our figures; this could also help explain the discrepancy between the two wealth profiles shown in Fig. 6. At the same time, the diminishing discrepancy above the age of 45 indicates a substantial cohort effect in wealth accumulation among Michigan iron-workers, associated perhaps with the Civil War. 26. This emphasis on unstable savings behavior over time is consistent with evidence drawn from the records of the Philadelphia Saving Fund Society after 1850 (Alter, Goldin, & Rotella, 1994). Over half of the accounts opened by working-class males in 1850 were closed within two years, while ‘‘many that survived the first five years [had] subsequent deposit activity only once every three years’’ (p. 763). About 35 percent of male accounts remained inactive once opened (calculated from p. 763). Note that these calculations refer to that small proportion of American households that had savings accounts, who were in turn likely to be among the most committed and consistent savers. 27. Families were drawn from the SCF according to the following criteria: head of household has a high school degree (but no more than 12 years of education); wages and salaries contribute to no less than 67 percent of family income during 1982 and 1988; average annual family income was below $100,000; households exhibiting implicit saving rates of greater than 250 percent were excluded. 28. Hubbard, Skinner, and Zeldes (1995) exclude employer pensions and social security benefits in their measures of net worth and financial wealth, as do we. 29. Since our data are limited to workers, we have no information on those households that accumulated sufficient funds to retire. Moreover, our data also exclude those workers who accumulate enough to make the transition to selfemployment. But we do not believe these biases to be significant, either in scale or in impact on our analysis. To begin with, as Lee (1996) has observed, the rate of retirement remained low at the end of the 19th century – perhaps no more than 5 percent of the entire sample withdrew entirely from the labor market after age 60 or 65; it is probable that the bulk of those who did retire were professionals and the
Low Saving Americans?
163
self-employed. The scale of self-employment was greater for our period; in the public use sample of the 1910 Census, 4.3 percent of the non-farm workforce worked for their own account at age 20, growing to 20.4 percent by age 50. Let us assume that this increase reflects a high transition to self-employment, rather than a cohort effect originating in the shift from workshop to factory over the late 19th century. Let us also assume that such transitions were based on high rates of accumulation, rather than unanticipated (or anticipated) windfall gains. In that case, our sample will exclude a cadre of high savers – but only after they have transferred out of the sample. Note that our conclusions regarding low saving hold throughout the age distribution, including the younger age groups when such aggressive savers should be in the data set. Moreover, if we assume that the top quintile of savers is excluded from our population, we can examine the 70th quantile as if it were the median; shifting our focus to this extent clearly raises the levels of accumulation and the ratios of wealth to income, but by relatively modest amounts. Moreover, it is no longer clear what reference point in the modern data should be taken. Are the selfemployed closer in character to high school graduates or the college educated? 30. Our quintiles all show wealth-holding rising steadily through middle age. In contrast, the quintiles of real estate and personal wealth holding calculated by Conley and Galenson (1994) for Chicago residents from the 1860 manuscript census show earlier peaks, generally coming in the 40’s. 31. This figure obtained through private correspondence with the authors. 32. A 5 percent random sample of simulated households indicates that renters accumulate substantially less non-housing wealth than homeowners before 1910. At age 35, renters (with an average income of $612, compared to $600 for owners), have accumulated $281 of non-housing wealth, compared to $701 for owners (whose house was worth $814); at 45, renters (earning $571) have $370 of non-housing wealth, owners (with income of $594) have $1112 of non-housing wealth and $1020 of housing wealth; at 55, renters (earning $526) own $461 of non-housing wealth, owners (earning $531) have $1030 of non-housing wealth and $1059 of housing wealth; at age 65, renters (earning $518) have accumulated $645 of financial wealth, compared to $1592 for owners (earning $533); the value of housing for owners at age 65 is $1347. 33. Note that the comparison of wealth-lifetime income ratios then and now is a legitimate exercise even though real incomes have risen substantially over the twentieth century. Davis and Gallman (1978, p. 50) note that even though higher savings rates appear to be associated with higher incomes in the cross section, savings do not appear to have been income-elastic historically over time. 34. Evidence from the 1990s similarly suggests that falling savings rates was concentrated among the top income groups (Maki & Palumbo, 2001). 35. The 1890 figures come from Historical Statistics: Colonial Times to 1970 (US Bureau of the Census, 1975); the 1980 figures from U.S. Department of Health and Human Services (1985). 36. The average 60 year old in 1980 could expect to survive another 17.5 years – an increase of only 19 percent on 1890 levels or, to put it more concretely, an increase of only 5.1 percent in the length of adult life. Haines (1998) estimates e20 for all U.S. males in 1890 at 40.96 years. Note also the additional demands on wealth of lengthier female lives in combination with lower female labor force participation and age-gaps at marriage.
164
JOHN A. JAMES ET AL.
37. For recent estimates of expected time to be spent in retirement for this period, see Lee (2001). 38. A primary reason for the discrepancy between these estimated retirement rates is that Lee (1996) heeds Margo’s (1993) advice and does not exclude the long-term unemployed (6 months and longer) from the labor force. Margo (1993) argues that by treating the long-term unemployed as out of the labor force, Ransom and Sutch (1986) overestimate retirement rates in 1900. 39. These figures are calculated from Grattan and Rotondo (1991, p. 351), after converting from constant (1917/1918) to current prices using their preferred price deflator. 40. These figures refer to renters only; owner-occupiers are excluded from the analysis. 41. The idea that intergenerational transfers sustained those workers who did survive to old age is supported by indirect evidence from our data set. In particular, ‘‘static’’ regression models of the data show a negative correlation between the number of children in a family and its current saving rate. Since nearly 13 percent of married, prime-aged workers in our historical database have five or more children (compared to fewer than 3 percent in a similar sample from the 1980 CPS, and only 1.5 percent in the 1989 CPS), this might account for a significant proportion of the distribution. To address this question fully would require a structural model of fertility, savings and income. However, a quick back of the envelope calculation using the parameters reported in Table B1 suggests that a decline in the average number of children per family from the levels recorded in the historical data set to the levels reported at the 2000 Census of Population would, ceteris paribus, have raised savings rates by some 1.45 percent. 42. An interpretation consistent with Diamond and Hausman’s (1984) analysis of the impact of social security benefits on savings behavior using panel data derived from the NLS for 1966–1976. 43. Autoregressive models are the only class in which dynamic properties can be identified using repeated cross-section data (Moffitt, 1993). In particular, without actual panel data, one cannot generally separate the total error variance into between-family and within-family components. 44. Of course, had they worked, parametric (regression-based) simulation models, such as that successfully utilized to simulate family incomes over the life cycle, would have allowed the inclusion of several additional variables (like marital status and number of children at home, which were included in the income regression model) to help explain saving decisions. 45. A less restrictive characterization, allowing the year’s saving bracket to be in the neighborhood of, rather than exactly equal to, last year’s produces very similar results. So the precise specification here is not crucial.
ACKNOWLEDGMENTS The authors have benefitted from many comments from participants at seminars and presentations at the American Economic Association meetings;
165
Low Saving Americans?
Meiji University, Tokyo; Tobingen University; Universit, Louis Pasteur, Strasbourg; University of California, Davis; University of Oxford; the Von Gremp Workshop on the History of Entrepreneurship in the U.S. Economy at UCLA; and the Washington Area Economic History Seminar. We are particularly indebted to Jonathan Skinner for his surprisingly useful advice and suggestions. All responsibility for errors, as usual, rests with the authors.
REFERENCES Abowd, J. M., & Card, D. (1989). On the covariance structure of earnings and hours changes. Econometrica, 57, 411–445. Alter, G., Goldin, C., & Rotella, E. (1994). The savings of ordinary Americans: The Philadelphia saving fund society in the mid-nineteenth century. Journal of Economic History, 54, 735–767. Carroll, C. D. (1992). The buffer-stock theory of saving: Some macroeconomic evidence. Brookings Papers on Economic Activity, 61–135. Carroll, C. D., & Summers, L. (1987). Why is the U.S. national saving so low? Brookings Papers on Economic Activity, 607–635. Carter, S. B., Ransom, R. L., & Sutch, R. (1991a). Those ‘cold, enduring Arabic characters’: The quality of late nineteenth-century state Labor Bureau Statistics. Unpublished manuscript, University of California. Carter, S. B., Ransom, R. L., & Sutch, R. (1991b). The historical statistics project at the University of California. Historical Methods, 24, 52–65. Carter, S. B., Ransom, R. L., Sutch, R., & Zhao, H. (1992). Codebook and user’s manual: A survey of 514 wage-earners in Kansas, 1895; Reported in the tenth annual report of the Kansas Bureau of Labor and Industrial Statistics. Institute of Business and Economic Research, Berkeley. Carter, S. B., Ransom, R. L., Sutch, R., & Zhao, H. (1993). Codebook and user’s manual: A survey of 719 male stone workers in Michigan, 1888; Reported in the sixth annual report of the Michigan Bureau of Labor and Industrial Statistics. Institute of Business and Economic Research, Berkeley. Carter, S. B., & Sutch, R. (1996). Myth of the industrial scrap heap: A revisionist view of turnof-the-century American retirement. Journal of Economic History, 56, 5–38. Collins, W. J., & Margo, R. (2001). Race and home ownership: A century-long view. Explorations in Economic History, 38, 68–92. Conley, T. G., & Galenson, D. W. (1994). Quantile regression analysis of censored wealth data. Historical Methods, 27, 149–165. Costa, D. L. (1995). Pensions and retirement: Evidence from union army veterans. Quarterly Journal of Economics, 297–319. Costa, D. L. (1997). Displacing the family: Union army pensions and elderly living arrangements. Journal of Political Economy, 105, 1269–1292. David, P. A., & Solar, P. (1977). A bicentenary contribution to the history of the cost of living in America. In: P. Uselding (Ed.), Research in economic history (Vol. 2, pp. 1–80). Greenwich, CT: JAI Series Press.
166
JOHN A. JAMES ET AL.
Davis, L. E., & Gallman, R. E. (1978). Capital formation in the United States during the nineteenth century. In: P. Mathias, M.M. Postan, (Eds.), Cambridge Economic History of Europe (Vol. VII, Part 2, pp. 1–69). Cambridge: Cambridge University Press. DeLong, J. B., & Summers, L. H. (1986). The changing cyclical variability of economic activity in the United States. In: R. J. Gordon (Ed.), The American business cycle (pp. 679–734). Chicago: University of Chicago Press. Diamond, P. A., & Hausman, J. A. (1984). Individual retirement and savings behavior. Journal of Public Economics, 23, 81–114. Engen, E., & Gruber, J. (2001). Unemployment insurance and precautionary saving. Journal of Monetary Economics, 47, 545–579. Feldstein, M. (1974). Social security, induced retirement and aggregate capital accumulation. Journal of Political Economy, 82, 905–926. Feldstein, M. (1996). Social security and saving: New time series evidence. National Tax Journal, 90, 51–164. Friedman, M. (1957). A theory of the consumption function. Princeton, NJ: Princeton University Press. Galenson, D. W. (1991). Economic opportunity on the urban frontier: Nativity, work, and wealth in early Chicago. Journal of Economic History, 51, 581–603. Galenson, D. W., & Pope, C. (1992). Precedence and wealth: Evidence from nineteenth-century Utah. In: C. Goldin & H. Rockoff (Eds), Strategic factors in nineteenth-century American economic history (pp. 225–241). Chicago: University of Chicago Press. Goldsmith, R. (1955). A study of saving in the United States, Vols 3. Princeton: Princeton University Press. Gratton, B., & Rotondo, F. M. (1991). Industrialization, the family economy, and the economic status of the American elderly. Social Science History, 15, 337–362. Haber, C., & Gratton, B. (1994). Old age and the search for security: A social history. Bloomington: Indiana University Press. Haines, M. (1979). Industrial work and the family life cycle, 1889–1890. In: P. Uselding (Ed.), Research in economic history (Vol. 4, 289–356). Greenwich CT: JAI Press. Haines, M. (1989). A state and local consumer price index for the United States in 1890. Historical Methods, 22, 97–105. Haines, M. (1998). Estimated life tables for the United States, 1850–1910. Historical Methods, 31, 149–169. Hatanaka, M. (1974). An efficient estimator for the dynamic adjustment model with autocorrelated errors. Journal of Econometrics, 2, 199–220. Higgs, H. (1893). Workmen’s budgets. Journal of the Royal Statistical Society, 56, 255–294. Hoover, E. (1960). Retail prices after 1850’. In: Conference on research in Income and Wealth (Ed.), Trends in the American economy in the nineteenth century (pp.141–186). Princeton, NJ: Princeton University Press. Hubbard, R. G., Skinner, J., & Zeldes, S. (1994). The importance of precautionary motives in explaining individual and aggregate saving. Carnegie-Rochester Conference Series on Public Policy, 40, 59–126. Hubbard, R. G., Skinner, J., & Zeldes, S. (1995). Precautionary saving and social insurance. Journal of Political Economy, 103, 360–399. James, J. A., Palumbo, M., & Thomas, M. (2004). Consumption smoothing among workingclass American families before social insurance. Unpublished manuscript, University of Virginia.
Low Saving Americans?
167
Kearl, J. R., Pope, C., & Wimmer, L. (1980). Household wealth in a settlement economy: Utah, 1850–1870. Journal of Economic History, 40, 477–496. Kotlikoff, L. J. (1988). Health expenditures and precautionary savings. In: L. J. Kotlikoff (Ed.), What determines savings? Cambridge, MA: MIT Press. Lebergott, S. (1964). Manpower in economic growth. New York: McGraw-Hill. Lebergott, S. (1976). The American economy. Princeton, NJ: Princeton University Press. Lee, C. (1996). The expected length of retirement and life-cycle savings, 1850–1990. Unpublished manuscript, University of Chicago. Lee, C. (2001). The expected length of male retirement in the United States, 1850–1990. Journal of Population Economics, 14, 641–650. Lillard, L. A., & Willis, R. J. (1978). Dynamic aspects of earnings mobility. Econometrica, 46, 985–1012. Long, C. D. (1960). Wages and earnings in the United States, 1860–1890. Princeton, NJ: Princeton University Press. MaCurdy, T. (1982). The use of time series processes to model the error structure of earnings in a longitudinal data analysis. Journal of Econometrics, 18, 83–114. Maki, D. & Palumbo, M. (2001). Disentangling the wealth effect: A cohort analysis of household saving in the 1990s. Federal Reserve Board of Governors Finance and Economics, Discussion Paper 2001–21. Margo, R. A. (1993). The labor force participation of older Americans in 1900: Further results. Explorations in Economic History, 30, 409–423. Michigan Bureau of Labor and Industrial Statistics (1890). Eighth Annual Report. Lansing. Moffitt, R. (1993). Identification and estimation of dynamic models with a time series of repeated cross-sections. Journal of Econometrics, 59, 99–123. Parsons, D. O. (1978). The autocorrelation of earnings, human wealth inequality, and income contingent loans. Quarterly Journal of Economics, 92, 551–569. Pope, C. (1989). Households on the American frontier: The distribution of income and wealth in Utah, 1850–1900. In: D. Galenson (Ed.), Markets in history (pp. 148–189). Cambridge: Cambridge University Press. Poterba, J. (1994). Government saving incentives in the United States. In: J. Poterba (Ed.), Public policies and household saving (pp. 1–18). Chicago: University of Chicago Press. Ransom, R., & Sutch, R. (1986). The labor of older Americans: Retirement of men on and off the job, 1870–1937. Journal of Economic History, 46, 1–30. Rees, A. (1961). Real wages in manufacturing, 1890–1914. Princeton, NJ: Princeton University Press. Rotella, E., & Alter, G. (1993). Working class debt in the late nineteenth century United States. Journal of Family History, 18, 111–134. Rotella, E. & Alter, G. (1995). Buying homes with borrowed money: Worker’s use of mortgage credit in the late 19th century. Unpublished manuscript, Indiana University. Snowden, K. A. (1995). The evolution of interregional mortgage lending channels, 1870–1940: The life insurance-mortgage company connection. In: N. R. Lamoreaux & D. G. Raff (Eds), Coordination and information: Historical perspectives on the organization of enterprise (pp. 209–247). Chicago: University of Chicago Press. Soltow, L. (1975). Men and wealth in the United States, 1850–1870. New Haven: Yale University Press. Thomas, M., James, J. A. & Palumbo, M. G. (1999). Retirement saving before social security. National Tax Journal Papers and Proceedings, Austin, Texas, 361–370.
168
JOHN A. JAMES ET AL.
US Bureau of the Census. (1975). Historical statistics of the United States: Colonial times to 1970. Washington: Government Printing Office. US Commissioner of Labor. (1904). Eighteenth annual report of the commissioner of labor: Cost of living and retail prices of food. Washington: Government Printing Office. US Department of Health and Human Services. (1985). US decennial life tables for 1979–1981. Volume 1, number 1, August. DHSS Publication no. 85-1150-1. Vinovskis, M. A. (1990). Have social historians lost the Civil War? Some preliminary demographic speculations. In: M. A. Vinovskis (Ed.), Toward a social history of the American Civil War (pp. 1–30). Cambridge: Cambridge University Press. Williamson, J. G., & Lindert, P. H. (1980). American inequality. New York: Academic Press.
APPENDIX A. THE DATA 1. The BLS Surveys The savings data used in this paper were compiled from surveys of workers produced by the State Labor Bureaus of Kansas, Michigan, Missouri, and Oklahoma between 1885 and 1908. Michigan undertook intensive surveys of workers in particular industries for each report, while the other states sampled broadly from the entire non-farm wage-earning population. The difference in coverage reflected different sampling philosophies. The Michigan Bureau commented, ‘‘It was deemed of more value to make a canvass of a few industries and do the work thoroughly and systematically than to spread the work over many occupations and only obtain partial reports’’ (Carter, Ransom, Sutch, & Zhao, 1993). Data were gathered through personal interviews. The interviewers, either special agents (as in the case of the furniture- and railway-workers) or regular bureau employees (as in the case of the iron-workers) visited the factories or work sites, questioned the workers, and recorded their answers themselves. Coverage was intensive: the survey of furniture-workers in 1889, for example, included 70 percent of employees cited in the federal census of manufactures for the same year. The Kansas model (followed by most states) was designed to be representative of the working population. The 1895 Kansas report noted that ‘‘the department was also careful that no particular branch of labor should receive especial notice in distribution of the forms.’’ The report continues: ‘‘In compiling these returns and placing these views and opinions before the public, a rigid adherence to the law of impartiality and disinterested personal opinion had been steadfastly followed, and no pet theory allowed to bear on the conscientious efforts to report exactly the language and figures given’’ (Carter, Ransom, Sutch, & Zhao, 1992; see also Carter, Ransom, & Sutch, 1991a, pp. 12–14). The Kansas, Missouri, and Oklahoma bureaus
169
Low Saving Americans?
carried out mail surveys. Questionnaires were sent to workers across the state accompanied by a stamped envelope and a pledge of confidentiality. In 1897, for example, the Kansas bureau distributed 5,000 schedules; 1,755 were returned, and 1,204 of these were judged sufficiently complete to be reported; the bureau judged the responses to have been quite representative of the occupational (among non-farm wage earners) and geographical structure of the state (see Table A1). We can test the representativeness of the savings data set by making comparisons of the demography of the sample to the Census of Population. Worker characteristics in the BLS data set are compared faute de mieux to wage-workers in the public use sample of the 1910 manuscript census (there are no micro-level data for 1890, while the 1900 Census does not provide the necessary information on employment status). The findings are as follows: First of all, the BLS surveys appear to underrepresent somewhat older workers. Those aged 56–65 constitute only 9.6 percent of the total sample in the merged BLS data, but 12.2 percent of non-farm workers in the 1910 PUMS. Conversely, middle-aged workers were slightly overrepresented in the BLS data – those aged 36–45, for example, making up 21.6 percent compared to 19.6 percent in the 1910 PUMS. The proportions of younger workers in the two samples are quite comparable. These small differences in age structure, however, should not influence our results since we control for age throughout the analysis and focus on age-specific distributions of savings and wealth. The industrial distribution of BLS workers, concentrating on manufacturing, construction, and transportation, while relatively neglecting retail trade and domestic service, ensures that women are underrepresented in our data set – in our data 5.2 percent of workers between ages 16–25 were female, compared to 35.5 percent in the 1910 PUMS. The BLS surveys also overstate skilled workers and understate unskilled workers (the proportion of semiskilled workers by age being quite comparable). Between the ages of 31 and 35, for instance, in the BLS data, there are 13.1 percent unskilled workers and 33.6 percent skilled, while in the 1910 PUMS the figures are 26.4 and 26.0 percent respectively. The consequences of such sampling issues for measured savings are discussed in the text. 2. Measuring Saving We have made some minor adjustments to reported saving in the raw data to deal with the fact that it is generally bounded from below by zero, even though it seems likely for some families to have spent more than their incomes during the previous year. To adjust for this censoring problem, we
170
JOHN A. JAMES ET AL.
calculated some negative saving values and some small positive saving values for families reporting exactly zero saving by using the saving distributions from five Kansas surveys. In those surveys reported savings could be negative since respondents were asked if they ran a surplus or deficit over the year and if so how much. This procedure did not dramatically change the distribution of saving in our data.
Table A1. State
Composition of the Data Sets.
Survey Number of Year Observations
Specific Groups Covered
Kansas 1884.5 1885.5* 1886.5* 1895* 1896* 1897 1899 1903* 1904* 1905.5* 1906.5*
346 401 392 414 447 552 836 544 371 328 393
Maine 1886.5 1887.5 1888 1890 1894 1900 Michigan 1888* 1889* 1890* 1892.5* 1895* 1896*
71 102 86 1,073 552 99 713 5,171 7,794 5,924 3,128 3,972
Stone workers Furniture workers Iron workers Railway workers Hack, bus line, and street railway workers Vehicle workers
171
Low Saving Americans?
Table A1. State
Survey Number of Year Observations
Missouri 1890.75*
258
New Hampshire 1886 1887 1893
49 283 279
Oklahoma 1908* 1909
242 117
Pennsylvania 1879 1880
310 232
West Virginia 1893.3
175
Wisconsin 1895
1,367
Federal cost of living survey 1889 6,465 Total observations on saving 27,466 Total observations on income 43,245
(Continued). Specific Groups Covered
172
JOHN A. JAMES ET AL.
APPENDIX B. ESTIMATING AUTOREGRESSIVE MODELS WITH REPEATED CROSS-SECTION DATA Following Moffitt’s (1993) notation, consider a first-order autoregressive model for the natural log of income (y) for a family i surveyed during year t: yiðtÞt ¼ ayiðtÞt
1
þ X0 iðtÞt b þ Z0 iðtÞ g þ iðtÞt
(B.1)
X denotes a vector of time-varying variables, such as marital status and number of children, both assumed exogenous to the iid error term, e; Z is a vector of time-invariant characteristics such as the sex and year of birth for the household head. The parameter a captures the degree of persistence in incomes over time.43 What hinders estimation of Eq. (B.1) for repeated cross-section data is the absence of information on lagged values of the X’s, rather than absence of information on lagged values of the dependent variable. With information on lagged X variables, Eq. (B.1) can be estimated by instrumental variables procedures, even if lagged y’s are unobserved (Moffitt, 1993). Moffitt’s estimation strategy involves searching the database for exogenous variables taking known values in the past. Clearly, time-invariant variables, denoted by Z, are candidates, but polynomial terms in age work as well, since knowledge of one’s current age implies knowledge of age in any past year. Thus, the idea is to estimate (B.1) using two-stage least squares based on the following first stage regression for predicting lagged incomes for families in the cross-section survey data: yiðtÞt ¼ W0 iðtÞt d1 þ Z0 iðtÞ d2 þ oiðtÞt ,
(B.2)
where W is a high-order polynomial in the age of the respondent for family i surveyed during year t. Given current information, each respondent’s age and all time-invariant variant variables (Z0 s) relevant for preceding years are known. Therefore, given estimates of the first stage regression coefficients in (B.2), polynomials in the respondents’ ages last year, W0 iðtÞt 1 can be constructed and the time-invariant variables, Z0 iðtÞ are also known. A fitted value y^ iðtÞt 1 can then be constructed and substituted for the actual lagged value of household income on the right-hand side of Eq. (B.1). Applying least squares to (B.1) in a second-stage regression results in consistently estimated parameters for a, b, and g. Identification of the slope coefficients in (B.1) using Moffitt’s two-step procedure typically requires that the age polynomial term, W, provide significant explanatory power for annual family incomes through (B.2), while being excluded from direct estimation of equation (B.1). It follows that this
173
Low Saving Americans?
particular estimation strategy for repeated cross-section data applies only to restricted autoregressive models relative to those identified using true panel data (see Table B1). Table B1.
Parameter Estimates from First-Order Autoregressive Modela.
Explanatory Variable
Dependent Variable (A)
(B)
Log of income
Saving rate
Saving subset
Earnings dataset
Lagged-dependent variable
0.8348 (0.015)
0.8341 (0.015)
0.865 (0.051)
Marital status
0.0661 (0.007)
0.0726 (0.006)
0.0028 (0.002)
– –
– –
0.108 (0.001)
Michigan resident
0.0256 (0.025)
0.0304 (0.010)
0.014 (0.011)
Female respondent
0.0877 (0.033)
0.0561 (0.019)
0.0005 (0.009)
Age polynomial? Birth year polynomial?
No Yes
No Yes
No Yes
Adjusted R2 Adjusted R2, 1st stage
0.324 0.311
0.250 0.240
0.075 0.065
30,234
43,245
26,983
Number of children
Number of observations a
All models estimated using Moffitt’s (1993) 2SLS Estimator with fourth-order age polynomial in first stage regression (all model specifications include survey year dummy variables).
174
JOHN A. JAMES ET AL.
APPENDIX C. A NON-PARAMETRIC SIMULATION MODEL FOR PERSISTENT HOUSEHOLD SAVING RATES As the text describes, regression-based simulation models (similar to (1)) were unable to generate skew in the age-specific distributions of annual saving rates reasonably close to the raw data. To solve this problem, we employ a non-parametric simulation procedure for annual saving rates, given each family’s age, current income, and past saving behavior. Using the actual survey data, we compute saving rates at 21 quantiles in the distributions for 64 ‘‘cells’’ defined by eight 5-year age brackets and eight income brackets. The 21 positions refer to the lowest and highest 2.5 percentile brackets and the 19, 5.0 percentile brackets in between. Having 27,000-plus observations in our data set allows us to confidently approximate the distribution of saving rates, conditional on age and income, in this non-parametric manner. Intuitively, the simulation procedure for annual saving rates involves randomly drawing a position in the saving rate distribution within a ‘‘cell,’’ determining each year by the intersection of the family’s age bracket and income bracket. Furthermore, each family’s position in the saving rate distribution this year depends on its position in the distribution last year. Thus, our non-parametric procedure incorporates three pieces of information likely to be most important for explaining household saving decisions: the head’s age, current income, and past saving behavior.44 Note the recursive structure implicit in the non-parametric simulation model for annual saving rates: income directly affects current saving rates, but not vice versa, and the random components of income and saving rates are not correlated in a given year. Our non-parametric simulation procedure accurately matches average saving rates for each age group in the actual data and closely matches standard deviations and distribution deciles as well. Relatively strong persistence in income levels at the household level through simulation of Eq. (1) implies some degree of persistence in income brackets from year-to-year. This, in turn, implies some indirect persistence in saving rates from yearto-year, as long as saving rates respond positively to current income, which is the case in the underlying data. However, our non-parametric simulation procedure for annual saving rates, conditional on age and income, also allows a second, direct manner in which household saving rates persist over the life cycle. This specification allows us to simulate versions of the model in which annual saving rates are
175
Low Saving Americans?
determined to some degree by household-specific (time-invariant) saving propensities. To incorporate direct persistence in household saving rates over time, we introduce a parameter, l, ranging between 0 and 1 (constant across families), which represents the probability that each family’s saving rate cell this year is the same as last year’s saving cell outcome. With probability 1-l, on the other hand, this year’s saving rate cell is drawn independently of last year’s outcome. Specifically, let SRBRACKt stand for this year’s position in the saving rate distribution, conditional on a family’s age and income brackets in a given year. SRBRACKt is a random variable taking an integer value 1 through 21, as described above. Our non-parametric simulation program assumes: Pr ðSRBRACKt equals SRBRACKt 1 Þ
¼l
Pr ðSRBRACKt is independent of SRBRACKt 1 Þ ¼ 1
(Event A)(C.1) l
(Event B)
We randomly draw a uniform (0,1) deviate to determine whether Event A (saving rate persistence this year) or Event B (independence this year) occurs. If Event B occurs, we then draw a second uniform (0,1) deviate for this family to represent its position in the saving rate distribution, conditional on its age and income cell, this year. If Event A occurs, then the relative position of the simulated household in the saving rate distribution, conditional on age and income, is the same as last year.45 Thus, if l ¼ 0:0; household saving rates are independent over time, given current age and income. However, larger values of l imply greater degrees of persistence in families’ positions in the saving rate distribution, conditional on their ages and income drawn, each year. When l ¼ 1:00 each simulation family occupies the same relative position in the saving rate distribution, conditional on its age and income, each year. This case represents the least (year-to-year) mobility we consider in the conditional saving rate distribution. While arbitrary in functional form, the benefits of this procedure are its (computational) simplicity to program, its ability to capture persistence in household saving rates (heterogeneous saving behavior across families), and assurance that the density function for saving rates in the non-parametric simulation model integrates to one for each family each year.
This page intentionally left blank
176
WORKER ABSENTEEISM UNDER VOLUNTARY AND COMPULSORY SICKNESS INSURANCE: CONTINENTAL EUROPE, 1885–1908 John E. Murray ABSTRACT Prior to widespread social insurance, European governments experimented with a variety of programs to protect workers from income loss due to illness. This paper examines the consequences for worker absenteeism of making sickness insurance coverage voluntary or compulsory. Medical benefits appear to have reduced absenteeism for all workers. The effect of paid sick leave depended on insurance fund membership status. Betterpaid workers found it easier to take time off in compulsory than in voluntary funds. Distinctive information problems plagued voluntary systems, and eventually were resolved by rejecting the voluntary ideal and forcing all workers into a single risk pool.
Research in Economic History, Volume 23, 177–207 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23005-7
177
178
JOHN E. MURRAY
1. INTRODUCTION Absenteeism is costly. Lost work time due to absence has been estimated at about 3 percent in the UK, 6 percent in Sweden, and 2 percent in the USA (Barmby, Ercolani, & Treble, 2002; Gilleskie, 1998). Even so, the economic literature on the topic has been quite small; Henrekson and Persson (2004) note that the several thousand pages of Handbook of labor economics do not refer to ‘‘sick leave’’ or ‘‘absenteeism’’ at all. The worker’s decision whether to come to work is the visible outcome of a complex process of sickness (real or pretended), assessment of health that may include medical attention, and assessment of costs and benefits of time off from work. Studies of European workers generally do not account for out-of-pocket health care costs in this calculation, since they are small or non-existent; studies of American workers should include the costs of medical attention, but with the exception of Gilleskie (1998) they typically do not. This dichotomy between European and American systems suggests that the institutions created to enable sick workers to stay home and recover are important factors in any study of absenteeism, but cross-country studies thus far have not explicitly modeled basic differences between systems such as whether workers are required to carry insurance against sickness and absenteeism versus being able to optout voluntarily (Drago and Wooden, 1992; Barmby, et al., 2002). This paper examines absenteeism rates among workers covered by a variety of sickness insurance programs in the European past. Sickness insurance, a combination of medical insurance and paid sick leave, was among the first types of social insurance advanced by European governments over a century ago. By the turn of the century, government mandated or voluntarily purchased sickness insurance programs covered millions of workers. Even then absenteeism was costly. The present sample indicates that up to 6 percent of covered work days were lost to worker absence annually in Denmark, 2–3 percent in Austria and Germany, and 1–2 percent in France and Belgium. Some European workers voluntarily obtained insurance coverage, and legal requirements compelled other workers to join their sick funds. Workers absented themselves differently in each type of fund. The initial decision to make coverage compulsory or voluntary imposed costly information asymmetries on the voluntary funds that hobbled them financially. The actuarially sound compulsory funds were able to pay for ever more days off. This paper suggests that the best explanation of these differing trends in absenteeism rates originates not in worker health or workplace characteristics but in institutional differences in the provision of social insurance.
Worker Absenteeism under Voluntary and Compulsory Sickness
179
2. SICKNESS INSURANCE AND ABSENTEEISM TRENDS Historians have closely examined the rise of government affiliated, organized, or funded social insurance programs (Mitchell, 1991b; Ko¨hler, Zacher, and Partington, 1982; Baldwin, 1990; Dutton, 2002; Huberman and Lewchuk, 2003; Murray, 2003). There are several reasons to expect that a historical perspective can shed considerable light on issues of short-term labor supply. Since workers’ real incomes were so much smaller during the time of this study, their savings would have been much smaller as well. Without a buffer of savings, worker absenteeism when ill should have been more responsive to incentives provided by sickness insurance benefits. Since present day studies tend to consider single social insurance programs with relatively homogeneous benefits, such as the UK (Barmby, Orme, and Treble, 1991; Doherty, 1979) or Sweden (Johansson & Palme, 1996), structural variation in the present programs is especially valuable. Sickness insurance funds influenced mortality rates. Using a panel that consisted of Belgium, Denmark, France, Germany, and Sweden, Winegarden and Murray (1998) showed that the elasticity of mortality rates with respect to the share of covered population was on the order of -0.5. While they concluded that this was implausibly large, it still indicated that sickness insurance led to a steeper decline in mortality rates than would otherwise have occurred. However, understanding of mortality influences may not lead to better understanding of influences on morbidity. Table 1 shows shares of deaths due to various causes and shares of absenteeism days due to various causes in two locations, Austria in 1890, in which about a half million workers were covered, and in Leipzig over a two decade period, which covered about a million man-years at risk. Some highly lethal ailments were responsible for relatively little absenteeism (e.g., pneumonia, tuberculosis, and other infectious diseases), while some chronic problems that induced high rates of absenteeism killed relatively few workers (bronchitis and digestive and skin problems). Because factors that influenced sickness and absenteeism differed from those that affected mortality, they are best analyzed in their own right. The countries included in the present study, Austria, Belgium, Denmark, France, and Germany, organized their programs according to three fundamentally different principles (Table 2). Classical liberal ideology led France, which heavily influenced Belgium, to emphasize the right of workers to choose whether or not to join sick funds, and so nearly all French and Belgian funds were composed of workers who had joined voluntarily.
180
JOHN E. MURRAY
Table 1.
Shares of Days Lost and Deaths due to Selected Causes. Austria 1890
Cause Pneumonia Tuberculosis Venereal disease Neuralgia Acute bronchitis Chronic bronchitis Heart disease Digestive diseases Skin disease Extremities Accidental injuries
% days lost 2.8 7.5 0.4 2.3 6.3 2.5 1.9 12.9 3.2 2.7 11.1
Leipzig. 1887–1905, Males Only % deaths 11.3 37.2 0.0 0.2 1.6 1.5 3.6 5.7 0.2 0.5 4.4
Diseases Infectious
% days lost 16.7
% deaths 36.0
Genito-urinary Nervous system Respiratory
1.5 5.3 17.4
2.1 6.1 19.7
Circulation Digestion Skin Extremities Injuries
3.7 10.7 6.8 19.8
9.3 7.7 0.3 6.8
Note: Columns do not add to 100 percent because source document did not report all causes of illness and death. Source: US Commissioner of Labor (1911, pp. 346–347, 1265, 1270).
Sickness insurance in the Scandinavian tradition began in Denmark. Here, membership was also voluntary, but state provision of subsidies to encourage workers to join funds gave it a greater role than in the other voluntary programs. A third system arose in German-speaking central Europe, where Germany and then Austria established compulsory insurance for a wide range of workers. Compulsion was enforced by monitoring employer records and then imposing fines on firms that left employees covered. Altogether, compulsory funds in the present study included those in Austria, most German funds, and funds for French miners. Voluntary funds in the present study included all those in Belgium and Denmark, all French funds other than those for miners, and mutual aid funds in Germany. In the remainder of this essay I discuss these funds in terms of their compulsory or voluntary membership.1 Compulsory and voluntary sickness funds shared many characteristics.2 Covered workers paid dues to belong to a sick fund. In compulsory funds dues were set at a proportion of the worker’s wage, commonly 1–3 percent. Voluntary funds were more likely to charge the same flat rate per month to each member in order to emphasize equality of costs and benefits for all members. Upon falling ill, a worker filed a claim by notifying the fund, verifying his condition with a physician, and, in Germany and Austria, waiting a few days to draw sick pay (US Commissioner of Labor, 1911, pp. 232, 1189). He received a payment from his fund to replace his usual wage. Compulsory funds paid half to two-thirds the going wage. French workers typically received a flat rate such as 2 francs per day of sickness (US
Types of Funds in this Study.
Name of Funds
Years in Sample
Voluntary or Compulsory
Description of Members or Fund
Austria Austria Austria Belgium Belgium Denmark France
General Apprentice Miners Recognized Unrecognized All Adult
1890–1907 1890–1907 1892–1903 1885–1904 1885–1895 1893–1907 1886–1905
Mostly compulsory Compulsory Compulsory Voluntary Voluntary Voluntary Voluntary
France France Germany
Free Miners Parish or Communal
1886–1905 1896–1907 1885–1908
Voluntary Compulsory Compulsory
Germany
Local
1885–1908
Compulsory
Germany
Establishment
1885–1908
Compulsory
Germany
Building
1885–1908
Compulsory
Germany Germany
Guild Registered aid
1885–1908 1885–1908
Compulsory Voluntary
Germany
State registered aid
1885–1908
Voluntary
Germany
Miners
1885–1908
Compulsory
Workers in enumerated classes For apprentices For miners Regulated closely by government Relatively unregulated by government All funds Relatively closely regulated; did not cover children of members Relatively unregulated For miners Workers required to be insured but unable to obtain insurance through any of the other types of funds Workers employed in one occupation or one industry in a specific geographic area Predated Bismarckian law; sponsored by particular firms Building tradesmen; typically formed for duration of particular construction project Predated Bismarckian law; for guild members Predated Bismarckian law; registered with national government Predated Bismarckian law; registered with state government For miners
181
Country
Worker Absenteeism under Voluntary and Compulsory Sickness
Table 2.
182
JOHN E. MURRAY
Commissioner of Labor, 1911; Frankel & Dawson with the co-operation of Dublin, 1911). An incapacitated worker could collect sick pay for several months, after which either liability might fall to a separately organized longterm disability insurance fund or his benefits would end altogether. All funds provided for free or very low cost attendance by a physician and provision of medicines and equipment. Absenteeism recorded in the statistics analyzed in this paper probably involved some degree of ill-health, because workers needed the approval of a physician for their claim to be approved; that is, claims of illness were not totally fabricated. Different parties to the insurance contract might disagree on whether a worker was too sick to perform his duties, however. All systems relied upon outside sources of income to reduce the outof-pocket price of insurance. Employers contributed to nearly all compulsory funds, at a ratio of one to every two marks paid by workers. Among German miners’ funds, the employers and workers paid equally (US Commissioner of Labor, 1911, pp. 233, 826, 1179, 1210). The Danish government provided generous subsidies, to the tune of about 30 percent of sickness society income by 1910 (Gibbon, 1913). Voluntary funds in France and Belgium received little public money, but did rely on ‘‘honorary’’ fund memberships purchased by the more civic-minded bourgeoisie. Honorary members agreed to pay dues and not to make any claims. These payments and other donations and subsidies accounted for perhaps 12 percent of French fund income toward the end of this period.3 Internal policy differences distinguished voluntary from compulsory funds. Among the most important was free choice of personal physician. Compulsory German and Austrian funds gave their members wide latitude in choosing their doctor (US Commissioner of Labor, 1911, pp. 232, 1186). Voluntary funds in Belgium, Denmark, and France did not typically require members to see a fund-employed physician (Schepers, 1993; Gibbon, 1913, p. 40; US Commissioner of Labor, 1911, p. 620; Herzlich, 1982, p. 245; Frankel and Dawson with the co-operation of Dublin, 1911, p. 208; Mitchell, 1991a, p.176). Methods of compensation for health care providers differed as well. Most German physicians were paid by the procedure according to a schedule negotiated by local physician groups with the insurance funds (US Commissioner of Labor, 1911, p. 1186; Gibbon, 1913, p. 54). Physicians employed by voluntary funds were likely to be on a salary or capitation payments (Schepers, 1993; US Commissioner of Labor, 1911, pp. 619, 620, 809). Consequences of incentives for physicians who had to compete for patients and were paid by the procedure, compared to salaried physicians who were assigned patients by the insurer, are discussed below.
183
Worker Absenteeism under Voluntary and Compulsory Sickness
3. FACTORS IN WORKER ABSENTEEISM To examine influences on absenteeism, several fixed-effects models were estimated based on fund level data as found in turn-of-the-century publications (see Appendix). Three measures of absenteeism were available for analysis. Prevalence was defined as the average number of days absent from work and claimed against sick funds per worker-year. Incidence was defined as the average number of claims per worker-year, and duration the average number of days per claim.4 Prevalence is identically equal to the product of incidence and duration. The availability of all three measures is useful to understand potential mechanisms through which various factors influenced days missed per worker. Benefits of sickness funds were reported in terms of sick pay and medical expenses per claimed day. Medical expenditures included physician services, medicines, and equipments but not hospitalization. Real values were estimated by deflating by country-specific price indices (Maddison, 1995) and converting to 1,900 US$ at the rate reported by US Commissioner of Labor (1911). Other independent variables included real GDP per capita, assumed to correlate with real wages (Maddison, 1995) and the mean number of members per fund and its square. Other researchers have suggested a link between unemployment and sickness absence (Whiteside, 1987), so a dummy variable for recessions was included that was set equal to one for years in which, according to Maddison (1995), real GDP per capita had declined from the previous year.5 Mean values for all these variables can be found in Table 3. The quality of the data seems quite good, although a rank-ordering of that quality might be in order. The most reliable data came from the Table 3.
Means and Standard Deviations of Fund Characteristics. Compulsory Funds
Sick days per worker Claims or claimants per worker Sick days per claim or claimant Sick pay per sick day Medical benefits per sick day Average membership GDP per capita Sources: See text for details.
Voluntary Funds
Mean
SD
Mean
SD
7.25 0.41 17.59 0.25 0.22 803 2945
2.99 0.25 2.48 0.09 0.06 928 308
4.90 0.27 18.92 0.30 0.23 272 2770
1.28 0.08 2.31 0.08 0.10 139 399
184
JOHN E. MURRAY
German and Austrian funds, since the compulsory nature of the system required extensive government monitoring of the funds, which led to the collection of these statistics on a national scale in each case. Since the French miners’ funds were old and compulsory, their statistics would also have been carefully collected for similar reasons. The Danish data are probably reliable as well for similar reasons; although membership was voluntary, the political decision had been made to subsidize the funds relatively heavily so as to draw in as many workers as possible. Again, government intervention was extensive and included collection and publication of statistics, although not as fully as in the German and Austrian cases. French adult funds and Belgian recognized funds were likely to have been somewhat less carefully collected. These were voluntary funds that had agreed to submit to a greater degree of government regulation than other funds in those countries. Their regulators and other contemporary observers reported that these funds were not professionally operated, and we might guess that some of the carelessness in operations would have filtered down to data collecting as well. The least trustworthy data were the free French funds and unrecognized Belgian funds. Whereas statistics for the other funds covered all or nearly all such funds, and so can be presumed representative, the free and unrecognized funds conducted their operations with little oversight, although this changed over time. In Belgium in 1894 and France in 1898, the state aimed to bring previously ‘‘free’’ or ‘‘unrecognized’’ benefit societies under closer regulation, in part to address the greater risks that these unregulated societies were exposing themselves to (Verbruggen, 1996, p. 423; Frankel & Dawson with the co-operation of Dublin, 1911, p. 203). Which firms reported data to government statistical bureaus, and how representative they might have been simply cannot be answered. What we can say is that the reporting funds were in fact quite large. In 1895 the free French funds had 315,000 members, while adult funds had just over a million, and in Belgium the unrecognized funds at 36,000 members were somewhat less than half as large as the recognized funds. Some simple economic theory and results from earlier studies can guide expectations of regression results. First, consider sick pay. A worker who desired to take time off due to sickness may have been ill, but no bright line divided sick-but-able-to-work from too-sick-to-work. In this sense, absenteeism was a choice influenced by incentives, so it is reasonable to suppose that a sick worker who faced a choice to stay home or work considered the cost of each alternative: continuing to work may have degraded his health, but without paid sick leave staying home reduced his income. Paid sick
Worker Absenteeism under Voluntary and Compulsory Sickness
185
leave, then, should have increased absenteeism by reducing its cost. Present day studies have found the value or availability of paid sick leave to be positively related to days missed per worker in the UK (Doherty, 1979) and Sweden (Johansson & Palme, 1996), to sickness events per worker in a panel of firms in four English-speaking countries and Austria (Drago & Wooden, 1992), and to average duration of recovery from sickness in the US (Butler & Worrall, 1985) and the UK (Doherty, 1979). Ideally the best measure of sick pay would be a relative one: the replacement rate, calculated as sick pay as a proportion of wage. Unfortunately there were no available data on wage rates of covered workers that could be used to estimate replacement rates, so the results presented here were based on the value of paid sick leave per claimed sick day. Whether medical benefits reduced absenteeism by providing better access to medical care depended on whether health care of a century ago was efficacious. There is good reason to believe that it was. By the late nineteenth century, heroic regimes of rigorous bleedings and purgings had been abandoned. Surgical and medical care most applicable to working age members of sick funds had reached a high state of development. By the turn of the twentieth century technological advances such as X-rays, which many funds paid for, allowed accurate diagnosis of many ailments. Therapeutic efficacy was catching up to advances in diagnostics. Lister’s antiseptic techniques, particularly applicable to orthopedic surgery that followed from workplace accidents, prevented many iatrogenic infections (Bynum, 1994, pp. 140, 223). Cheap anti-inflammatory drugs such as aspirin and analgesics such as morphine came into common use. Management of cardiac symptoms such as angina with amyl nitrate and nitroglycerin improved quality of life. Equipment such as trusses for workers who suffered from inguinal hernias enabled them to get back to work sooner (Song & Nguyen, 2003). The net effect of sick pay and medical insurance together is ambiguous a priori. Economic studies occasionally fail to link either to absenteeism or reach counterintuitive conclusions. In one of the earliest and most widely cited papers on the topic, the availability of sick leave reduced days missed per worker in the United States (Allen, 1981), and a similar negative effect on incidence and duration was found more recently (Gilleskie, 1998). After correcting for selection biases in an international panel, sick leave had no significant effect on absenteeism (Drago & Wooden, 1992). In other studies its availability did not directly affect cases per worker (Barmby, Orme, & Treble, 1991), nor did its value affect days per case (Fenn, 1981). The weight of the evidence appears to favor the disincentive to work effect of social insurance (Barmby, Ercolani, & Treble, 2002).
186
JOHN E. MURRAY
An urgent and ongoing concern of insurance managers, government officials, and social scientists was that many claims were fraudulent, and that many funds were being asked to fund malingering. Fraud was seen as an important threat to sickness fund viability. The most effective response was thought to be the peer pressure that only small funds could sustain (Gibbon, 1913, p. 121). If this conventional wisdom were correct, then additional numbers of fund members would be associated positively with absenteeism, as long as larger funds lacked sufficient peer pressure to discourage frivolous or fraudulent claims. At the same time, funds acknowledged that spurious claims were a potential problem, and they took steps to protect themselves. To compensate for the lack of peer pressure, larger funds employed spies to maintain surveillance on claimants.6 The Leipzig funds used 21 paid and 250 volunteer ‘‘sickness controllers’’ who visited claimants and reported back to the funds’ central office on their status (US Commissioner of Labor, 1911, p. 1199). Danish funds also employed special inspectors who visited claimants (Frankel & Dawson with the co-operation of Dublin, 1911, p. 191). The square of the fund size was included in regressions, to test for a non-linear relationship that would reflect the potential effectiveness of such monitoring by the larger funds that could afford monitoring. Economic downturns may have affected sickness claims. In 1920s, Britain the general slump was associated with more absenteeism (Whiteside, 1987). By contrast, present day studies suggest a negative relation between unemployment and absenteeism (Doherty, 1979; Johansson & Palme, 1996). Higher real wages today may induce an efficiency wage-type effect that makes workers anxious to keep their jobs, or it may have been that early twentieth-century workers were so poorly fed and sheltered that the effect of economic downturns was to induce greater sickness among them. Thus, whether this coefficient’s sign should be positive or negative is indeterminate a priori. The results of the regression estimates appear in Table 4. The reported estimates come from weighted least squares regressions in which the weights were set equal to the number of members in each fund group in each year, so as to prevent small funds from skewing the results. Inclusion of year-fund interactions did not change the present results, but did increase R2 estimates. In addition, results reported here were robust to other specifications including semilog, inclusion of lagged dependent variables, random effects, and the use of differences as dependent variables. A potential estimation problem was endogeneity of health, demand for insurance, and worker search for jobs that would provide such insurance. Given the breadth of coverage required in the Austrian and German compulsory funds,
187
Worker Absenteeism under Voluntary and Compulsory Sickness
Table 4.
Effects of Insurance Funds on Absenteeism Measures, by Membership Status. Compulsory Membership
Voluntary Membership
Days/worker Claims/worker Days/claim Days/worker Claims/worker Days/claim Intercept
0.42 (2.19) 0.18 (0.06) 0.26 (0.08) 0.10 (0.03) 0.03 (0.01)
5.19 (2.07) 0.50 (0.06) 0.12 (0.02) 1.13 (0.13) 0.46 (0.04)
0.25 (0.28)
GDP/1,000 population
0.24 (0.33)
1.12 (2.40) 0.20 (0.06) 0.61 (0.07) 0.05 (0.03) 0.06 (0.01) 0.45 (0.07) 0.02 (0.30)
Recession
0.02 (0.02)
0.01 (0.01)
0.02 (0.01)
0.98
0.92
Sick pay Medical Members Members2
0.20 (2.55) 0.29 (0.07) 0.59 (0.08) 0.13 (0.04) 0.005 (0.007)
Claims/worker
Adj R2 N
0.98
204
0.90 (0.26)
22.78 (6.59) 0.02 (0.14) 0.13 (0.05) 1.50 (0.30) 0.69 (0.09) 0.74 (0.04) 2.85 (0.83)
0.29 (0.34)
0.06 (0.02)
0.08 (0.04)
0.05 (0.02)
0.98 126
0.93
0.30 (2.72) 0.48 (0.06) 0.08 (0.02) 0.71 (0.13) 0.27 (0.05)
0.94 116
Notes: Each regression included dummy variables for fund type and year. All continuous variables in logs. Standard errors in parentheses. White tests indicated homoskedastic residuals. Regressions weighted by number of members in each observation. Significant at 0.10 level, 0.05 level, 0.01 level.
endogeneity could not have been a problem there. All German workers earning less than 2,000 marks had to obtain insurance, as did Austrian workers in virtually every industry except agriculture, forestry, and domestic service. The voluntary funds are potentially a different story, since workers in ill health may well have sought jobs that offered insurance. However, endogeneity is not addressed further in this paper. First, the empirical evidence for it is lacking. Whaples and Buffum (1991) found a very weak relationship between ill health and sickness and accident insurance purchase. Further, in recent studies that used individual level American data, the prospect of endogeneity was acknowledged but no econometric adaptations were used (Vistnes, 1997, p. 307; Gilleskie, 1998, p. 20;
188
JOHN E. MURRAY
Fishback & Kantor, 1992). Finally, in practice, voluntary funds generally required a medical examination as a condition of membership, which should have helped limit the extent of such problems. Some influences on worker absenteeism had the same effects on both types of funds, and other factors had different effects. I consider the effect of recessionary conditions first. In voluntary funds, workers missed 6 percent more days during recessions than during growth periods. Whiteside (1987) found a similar relationship between unemployment and absenteeism in interwar Britain, which she attributed to health problems that were made worse by the lack of work. Unemployed workers may also have used sickness benefits as a substitute for unemployment benefits, at least among those whose sick fund membership continued after a layoff. Absenteeism by compulsory fund workers did not significantly change during recessions. Regarding peer pressure, coefficients on fund size and its square tell approximately the same story across all regressions. There was more absenteeism in larger funds than in smaller funds. This was especially pronounced in the voluntary funds, where absenteeism increased with the square of membership and not just proportionally. It seems more likely that members of larger funds, as hypothesized at the time, lacked the peer pressure that would have discouraged them from taking unnecessary absences. Monitoring of claimants by very large compulsory funds may have discouraged long claims. Up to about 1,850 members, or one standard deviation beyond the mean, larger funds experienced longer average durations of claims, but beyond that fund size the duration decreased with number of members, perhaps because it was those larger funds that employed sickness controllers. Fund characteristics and the state of the economy appear to have influenced worker absence rates as expected; the effect of individual level incentives was subtler. In both types of funds, daily expenditures on medical benefits were associated with significantly less absenteeism by all three measures. This is strong indirect evidence for the efficacy of medicine of the day, although two additional factors came into play as well here, the effects of waiting periods and of rest. To tease out a more precise relationship between medical benefits and absenteeism, Table 5 presents parameter estimates from a subsample in which payments for physician attendance could be distinguished from those for medicines and equipment. In voluntary funds, physician benefits, but not drug benefits, were associated with fewer missed work days. A possible relationship comes into clearer focus in the compulsory fund case. Both physician benefits and drug benefits significantly reduced days missed, but in different ways, physician benefits through fewer claims per
189
Worker Absenteeism under Voluntary and Compulsory Sickness
Table 5.
Influence of Specific Medical Benefits on Absenteeism. Compulsory Funds
Physician benefits Drug benefits N
Days/ member
Cases/ member
0.37 (0.087)
0.47 (0.092)
0.27 (0.087) 182
0.071 (0.093) 182
Voluntary Funds
Days/case
Days/ member
Cases/ member
Days/case
0.11 (0.075)
0.22 (0.12)
0.24 (0.25)
0.13 (0.19)
0.20 (0.075) 182
0.0031 (0.14) 123
0.066 (0.28) 107
0.041 (0.21) 107
Notes: Each coefficient was taken from a regression with the same regressors as in Tables 2 and 3, except that spending on medical benefits per sick day was divided into spending on physician benefits and on drug and medical equipment spending per sick day. Standard errors are in parentheses. All continuous variables in logs so that parameter estimates give elasticities. White tests indicated homoskedastic residuals. Regressions weighted by number of members in each observation. Significant at 0.10 level, 0.01 level.
member and drug benefits through shorter sickness episodes. It seems reasonable to suppose that the drug and equipment payments provided diagnostic services, analgesics, trusses, and the like that actually improved worker health and brought them back to health and back to work sooner. A similar logic may apply to physician attendance. If a German or Austrian worker visited a physician, the cost was billed to the fund, but if the physician cured or at least improved the worker’s state before the end of the waiting period, the member would have returned to work and would not have filed a claim for sick pay. An important role for rest is suggested by the coefficients of incidence, which were negative and significant in both duration regressions. They implied that a 10 percent increase in claims per worker would reduce the average duration of sickness by 4.5 percent in compulsory funds and 7.4 percent in voluntary funds. Voluntary and compulsory funds that enabled a greater share of their members to be absent then found those more frequent absence spells to be shorter in duration. The alternative to taking several brief spells off work was for a worker to continue at his post until he was so incapacitated that work was impossible for a matter of months. When workers were able to rest when needed, the ultimate result was fewer days missed. Sick pay had different effects on absenteeism in compulsory funds from those effects on voluntary fund members. This contrast suggests deeper
190
JOHN E. MURRAY
structural differences between the two types of funds, and these differences appeared in both the contemporary literature and in subsequent historical analysis. In compulsory funds, more valuable paid sick leave was associated with greater absenteeism, an effect that operated on claims per worker and days per claim approximately equally. This reflects the disincentive effects of sick pay. The reduction in cost of missing work increased the demand for sick days. The paradox lies in the equally significant effect of sick pay in the voluntary funds, but here in the opposite direction. A 10 percent increase in sick pay for voluntary fund members was associated with a 5 percent reduction in absenteeism, all of which appears to have occurred through shorter durations of sickness. To relate the elasticities given in the regressions to the differing trends in absenteeism – upward in compulsory funds and downward in voluntary funds – we need to know about trends in the independent variables. Medical benefits cannot explain the difference in absenteeism trends. For both sets of funds more medical benefits reduced absenteeism, according to the regression results in Table 4. Now for most of the compulsory funds, the value of medical benefits trended upward over the period with available data. For example, medical benefits increased in value in German funds by 15 percent and in Austrian funds by 7 percent. The negative coefficients for medical benefits imply that this would have reduced absenteeism in these funds, ceteris paribus, but that did not happen. Other things that were not held constant – sick pay and fund size dominated the influence of medical benefits, and absenteeism trended up. In voluntary funds, medical benefits trended downward in value, in free French funds by 39 percent and in recognized Belgian funds by 19 percent. This alone would have increased absenteeism according to the negative medical benefit coefficient, but the effect of spending more on medical care was dominated by other factors, and absenteeism among these workers declined. So while the coefficients of medical benefits were signed so as to suggest efficacious medical treatment, medical benefits did not cause the divergent trends in absenteeism. Growth in average membership size cannot explain the divergence either. In nearly all fund categories, membership size was trending upward over time. For example, the average free French fund had 132 members in 1886 and 143 members in 1905, while the average German establishment fund had 229 members in 1885 and 399 members in 1908. In each case, the effect of increasing membership would have been to increase absence per member, according to the positive elasticities. Thus, part of the upward trend in compulsory fund absenteeism can be attributed to increasing fund size, but this same variable cannot explain the downward trend in absence of
191
Worker Absenteeism under Voluntary and Compulsory Sickness
voluntary fund members, where it must have been overwhelmed by other influences. What seems to explain the divergence best was the effect of paid sick leave. Table 6 analyzes how increasingly valuable daily sick pay affected sick days missed per worker year. Most funds increased the value of their sick pay over this period. It increased slightly for French miners and members of adult French funds, but by a quarter for unrecognized Belgian fund members, a third for most German fund members, and over half for Austrian apprentices. The elasticities from Table 4 imply estimated changes in sick leave that could be attributed to the changes in sick pay, ceteris paribus: increases for the compulsory funds, and decreases for the voluntary funds. Comparing these estimated changes to the actual changes in absenteeism suggests that changes in sick pay alone accounted for a fourth to a third of the changes in absenteeism rates for German and Austrian funds as well as free French funds, and nearly half of that in unrecognized Belgian funds. The estimated effects of changes in paid sick leave were in the same direction as the overall trends in absenteeism and large in magnitude, and thus economically meaningful. Table 6. Country
Magnitude of Sick Pay Effect on Absenteeism in Selected Funds.
Type of Fund
% D Sick Pay Per Sick Day (1)
All funds All funds Miners Miners Apprentices
+31 +17 +4 +32 +57
+9 +5 +1 +9 +17
+35 +18 +24 +31 +62
26 28 5 30 27
Adult Free Unrecognized
+8 +19 +23
4 10 11
39 27 24
10 35 48
% D Sick Total % D Days due to Sick Days(3) D Sick Pay(2)
Share of % D Sick Days due to D Sick Pay(4)
Compulsory funds Germany Austria France Germany Austria Voluntary funds France France Belgium
Notes: % change measured from first year with data available to last year. % D sick days due to D sick pay was estimated by multiplying column (1) times the elasticity as given in Table 4 (0.29 for compulsory funds; and 0.50 for voluntary funds). Column (4) was estimated by dividing column (2) by column (3).
192
JOHN E. MURRAY
Why would sick pay have had distinctly different effects on absenteeism in the two types of funds, and how might sick pay have been related to the different trends in absenteeism over time? The result in compulsory funds is as expected, but to expect that more valuable sick leave induced more sick leave is to view the question only from the worker’s perspective. Presumably, workers did want to take more days off when they were sick – if they could afford to. But fund management had to address whether it could afford to pay for days off. When the answer was no, one solution was to reduce or eliminate payments for longer claims. Hence, more sick pay for voluntary fund members required fund management to reduce the average duration of a claim. As a contemporary observer noted of Austria, ‘‘The average duration of cases of disability is a matter in which the administration of the fund has great influence’’ and that appears to have been the case among the voluntary continental funds as well (US Commissioner of Labor, 1911, p. 278; cf. Butler & Worrall, 1985). This relationship between sick pay and absenteeism was a symptom of a set of problems facing the voluntary funds. The sources of the difficulties were the classic information problems of moral hazard and adverse selection, which burdened voluntary funds more severely than compulsory funds.
4. INFORMATION DIFFERENTIALS AND FINANCIAL CONSEQUENCES The case presented here for linkage of greater sick pay to less absenteeism is circumstantial. The published documents I have relied upon for statistics and commentary do not report on internal fund communications. However, statements and statistics from contemporary observers as well as later analysis by historians suggest that a story based on greater information problems of voluntary funds has substantial explanatory power. Some problems such as moral hazard and principals versus agents were common to both types of funds, but others such as adverse selection were specific to voluntary funds. Insurers did not fail to understand the problems; they recognized them and reacted in ways familiar to present-day students of risk and insurance. Even so, the net effect of moral hazard, adverse selection, and agency conflict was greater financial difficulties for the voluntary funds.
Worker Absenteeism under Voluntary and Compulsory Sickness
193
4.1. Moral Hazard and Principal–Agent Problems in All Funds A goal of sickness insurance was to circumvent the price mechanism and make certain goods deemed socially important, such as medical care and rest, available to insured workers more or less without regard to cost. To the extent that sickness insurance offered reductions in the price of medical care and of absenteeism from work, we would expect that insured workers would consume both more medical care and more days of rest. This is really just an increase in quantity demanded that follows from a decline in price, but economists sometimes refer to it as ‘‘moral hazard.’’7 Both voluntary and compulsory funds induced a medical care moral hazard. Initially, in Denmark government subsidies were to pay the entire cost of physician visits. However, the demand for such ‘‘free’’ medical care overwhelmed the system, causing subsidies to rise faster than expected. In response, funds required workers to buy discounted ‘‘sick tickets’’ for admission to a physician’s office, which resolved the problem of excess demand for physician services (Ito, 1980; Gibbon, 1913). In Germany, insured persons sought medical care for ever more trivial complaints, for which, fund officials complained, they would not have sought help if insurance had not been available to pay the bills (Gibbon, 1913, p. 64). The peer-pressure argument that hypothesized a positive correlation between fund size and absenteeism also suggests a concern for moral hazard problems at that time. All funds recognized the absenteeism moral hazard, but attempts to fix it often created further informational problems. To prevent workers from taking an unnecessary day off, many funds required a physician to certify that they were incapacitated.8 Introducing physician approval, however, replaced a moral hazard with a principal–agent conflict: did physicians work for the member or the fund? If the member had free choice, it was the former, but if the fund employed the physician, it was the latter. In France after the Pension Act of 1910, mutual aid society members could see their own ‘‘treating physician’’ only after getting approval from a fund-employed ‘‘controlling physician’’ (Mitchell, 1991b, p. 246). In German funds in which workers enjoyed free choice of physicians, fund-employed doctors also monitored independent physicians, in this case with second examinations. The records of such examinations suggest the magnitude of such principal– agent conflicts. German funds and members both enjoyed the right to demand a second opinion from a variety of ‘‘confidential medical advisors,’’ either fundemployed physicians or committees composed of physician and insurer
194
JOHN E. MURRAY
representatives.9 Here, to a great extent the physician–agent’s diagnosis depended upon the identity of the principal. Given free choice of physician, as in most compulsory funds, patients were principals, and personal physicians who gave the initial diagnoses of incapacitation were their agents. Physicians who monitored the primary physicians were agents of the insurers. Probability of claim approval followed accordingly. Table 7 shows the results of such second examinations in three regional fund groups in northern Germany, 1909–1910. While initial consultations tended to end in favor of the worker, second examinations often favored the fund. The large share of workers who had obtained statements from their own physician that they were incapacitated and who then returned to work rather than be examined by a fund doctor suggests a widespread problem with dubious claims, the essence of the absenteeism moral hazard (Gibbon, 1913, pp. 111–118). Participants in this system understood implications of agency: German physicians wanted to retain even those annoying patients who presented with dubious symptoms, in order to keep the capitation fees they brought with them. To keep the patients and their payments, contemporary
Table 7. Fund Group, Year
Results of Second Opinions in Germany, 1909 1910. Leipzig, 1910 N
%
Ko¨nigsberg, 1909 N
%
N
%
N
%
8,497
Returned to work before second exam
1,300
Second examinations
5,827
Able to return to work immediately
2,739
47
828
31
841
37
346
37
Able to return to work within a week
699
14
545
21
498
22
361
39
Other (further examination, sent to hospital, etc.)
466
8
659
25
414
18
0
0
1,806
31
619
23
527
23
219
23
15
544
2,657
Kiel, 1910
Referrals
Unable to work
3,179
Ko¨nigsberg, 1910
17
2,635
355
1,415 13
2,302
480
34
935
Notes: Recovery as % of referrals; others as % of examinations. Source: (Gibbon (1913), pp. 117–118). In Leipzig, among those workers who were not examined a second time were 1,259 who did not come for the examination and 111 who were excused.
Worker Absenteeism under Voluntary and Compulsory Sickness
195
observers asserted, physicians gamed the system by approving claims that they understood to be of doubtful veracity. The fund’s medical advisors then routinely rejected them, thereby keeping the fund healthy and the initial physician’s pay intact (Gibbon, 1913, p. 112).
4.2. Adverse Selection in Voluntary Funds If information problems were pervasive, burdens of adverse selection into membership were unique to voluntary funds. Compulsory funds had to accept all comers. Among prospective members of voluntary funds, those workers most interested in buying coverage were the ones who believed they were most likely to make a claim. Magnifying this tendency was the practice among voluntary French funds of charging a flat rate for all workers, known today as community rating (Thomasson, 2004). To discourage older applicants who were poorer risks, the voluntary funds often rejected applicants over the age of 40 because, according to a contemporary advice manual for fund managers, ‘‘the risk of illness is considerably augmented after that age’’ (Frankel & Dawson with the co-operation of Dublin, 1911, p. 208; Mitchell, 1991a, p. 174).10 But that was not enough. Adverse selection by poor health can be established in German voluntary funds independent of age effects. Since compulsory and voluntary insurance operated side by side in Germany, member age distributions and claim rates in each pool can be compared. Figures 1 and 2 illustrate male enrollment and absenteeism by age group in the best-documented area of German funds, Leipzig (US Commissioner of Labor, 1911, pp. 1259–1263). Figure 1 shows that the age structure of compulsory funds was skewed toward younger members and that in voluntary funds toward older members, suggesting adverse selection of older members into voluntary funds.11 Stronger evidence of selection problems, holding age constant, appears in Fig. 2. Within each age group, an average voluntary fund member claimed far more sick days per year than did a typical compulsory fund member. According to the German source, the peak of claims among voluntary fund members aged in their early twenties was in fact due to differential expected probabilities of making claims: Practically all the male population, including the weaker and those who are physically less valuable, are sent to work in the earlier ages; in a few years, however, the weaker persons must give up the occupations in which they are engaged, but realizing their need for insurance, continue their membership as voluntary members (US Commissioner of Labor, 1911, p. 1263; emphasis added).
196
JOHN E. MURRAY
% of membership in age group
25 20
compulsory
voluntary
15 10 5
un
de r 15 15 to 20 19 to 25 24 to 30 29 to 35 34 to 40 39 to 45 44 to 5 0 49 to 5 5 54 to 6 0 59 to 65 64 to 70 69 75 to an 74 d ov er
0
age group
Fig. 1. Age Distribution of Male Members of Sickness Funds in Leipzig, 1887–1905. N ¼ 952,674 Compulsory Members and N ¼ 43,771 Voluntary Members. Source: US Commissioner of Labor (1911, pp. 1257, 1259).
50
days missed per year
45
compulsory
voluntary
40 35 30 25 20 15 10 5
15
un d
er 15 to 1 20 9 to 2 25 4 to 2 30 9 to 3 35 4 to 3 40 9 to 4 45 4 to 4 50 9 to 5 55 4 to 5 60 9 to 6 65 4 to 6 70 9 t 75 o an 74 d ov er
0
age group
Fig. 2. Sick Days by Age Group, Male Members of Sickness Funds in Leipzig, 1887–1905. N ¼ 952,674 Compulsory Members and N ¼ 43,771 Voluntary Members. Source: US Commissioner of Labor (1911, pp. 1257, 1263).
Worker Absenteeism under Voluntary and Compulsory Sickness
197
Voluntary funds recognized that their lack of information on applicant health was a problem, and they strategically compensated for it. They required medical examinations conducted by physicians whom they employed directly (Mitchell, 1991a, p. 174; Frankel & Dawson with the co-operation of Dublin, 1911, p. 208). French mutual aid societies rejected applicants whose examination revealed any kind of chronic lung problem for fear it would develop into tuberculosis, treatment for which represented an expensive, long-term obligation (Mitchell, 1991b, p. 269). Some voluntary funds even imposed what are now called ‘‘pre-existing condition’’ clauses. Danish funds admitted ill applicants into funds as long as they agreed never to make a claim based on current illnesses (US Commissioner of Labor, 1911, p. 611).12 Compulsory funds did not routinely examine applicants (US Commissioner of Labor, 1911, p. 1199).
4.3. Consequences for Financial Stability Ultimately, strategies to relieve pressure on voluntary fund finances failed, and these funds endured constant financial struggles. Belgian funds, desperate to enroll more dues-paying members, offered more and more valuable benefits, such as a 23 percent increase in daily sick pay over the period of this study. This only exacerbated their chronic financial problems (Schepers, 1993; see also Verbruggen, 1996, p. 425). A Belgian government official charged with overseeing the mutual aid societies warned about their insurance activities (US Commissioner of Labor, 1911, p. 489): ‘‘In general, the mutual sick-benefit societies do not fulfill the necessary requirements of a safe and rational organization,’’ and in particular they granted benefits ‘‘without taking into consideration whether or not their assets are adequate’’ to pay for them. Raising premium payments was not politically feasible. As it was, reported an observer, ‘‘It is the elite of the working class alone that can stand the cost of sick insurance’’ (US Commissioner of Labor, 1911, p. 494). Exclusion of the needy from coverage, and lack of interest among the able-bodied in obtaining insurance, represented a serious political problem. The debates over compulsory insurance that began in the 1890s continued for half a century. In the end, compulsory coverage was found to be the only way to maintain the viability of Belgian funds (Schepers, 1993). Here, then, is an example of an adverse selection-driven ‘‘death spiral,’’ which ends in the dissolution of voluntary health insurance programs (Buchmueller & DiNardo, 2002; Thomasson, 2004).
198
JOHN E. MURRAY
Similar problems plagued the French system. Funds found that they were in over their heads when they expanded the range of benefits to include pensions and unemployment insurance, while their original efforts at sickness insurance remained actuarially unsound. French establishment funds became ever more dependent upon subsidies from sponsoring firms (Dreyfus, 1996). Among all French funds, the value of assets per participating member fell steadily from 1898 to 1905, a drop of about 28 percent. In Germany over the same period, this measure rose about 10 percent (US Commissioner of Labor, 1911, pp. 814, 1227). Zeldin (1979, p. 299) summarized the failures of the French societies: ‘‘Ignorance of the principles governing insurance was common, methods of administration amateur in the extremeisy. The most serious omission was that the whole movement was never established on an actuarial basis.’’ Observers made similar criticisms about the heavily subsidized Danish system. Due to their ‘‘unscientific and unsound’’ methods of setting membership fees, ‘‘practically all [sickness insurance societies] are insolvent from an actuarial standpoint with insufficient funds accumulated to meet their claims permanently’’ (Frankel & Dawson with the co-operation of Dublin, 1911, p. 190). One solution adopted by several funds was to pool their risks in a reinsurance scheme; another was to dissolve. Both were signs of another death spiral. As early as 1908, Denmark was beginning to require employers to enroll certain classes of workers in compulsory insurance schemes; in this case it was several thousand seasonal agricultural workers from Russia and Poland (Gibbon, 1913, p. 14). By 1933, sickness insurance coverage was compulsory for Danes over 16 years of age. A related problem was a vicious cycle of negative expectations. French aid societies tended to be small; nearly three-fourths of them in 1902 had fewer than 100 members (Zeldin, 1979, p. 299). A single gravely ill member could bankrupt a typical society (Mitchell, 1991a, p. 179). Younger workers who feared that their fund would have dissolved by the time they needed to make a claim stayed away because they understood that benefits of membership, from their perspective, were backloaded (Frankel & Dawson with the cooperation of Dublin, 1911, p. 156). Consequently, membership in French funds aged over time and claim rates rose beyond the ability of fund assets to service (Mitchell, 1991a, pp. 174–179). Here too, the instability of voluntary insurance was not resolved until after World War II when compulsory social insurance was imposed (Ko¨hler, Zacher, & Partington, 1982, p. 121). The relationship of the value of sick pay to briefer sickness spells, and hence lower overall absenteeism rates, followed from voluntary fund
Worker Absenteeism under Voluntary and Compulsory Sickness
199
financial straits. To deal with financial crises voluntary funds tried to limit expenditures on benefits. Rather than indemnifying members who paid their own physicians out-of-pocket, most French and Danish funds and Belgian socialist funds employed physicians directly, paying them a salary or with capitation fees (Schepers, 1993; US Commissioner of Labor, 1911, pp. 620, 809). In extreme times, such as during the Great Depression, denial of benefits was an explicit strategy of Belgian funds (Schepers, 1993). Their hope was to relieve financial difficulties, but the effect was to discourage membership. Most important for the present purposes, voluntary funds ordered their physicians to limit expenses, which entailed briefer spells of paid-for absence (Mitchell, 1991b, pp. 141, 243–246, 284). Members could have thwarted this strategy by seeking physicians who were willing to bend the rules on spending, but precisely to prevent shopping for physicians who would provide desired diagnoses, these funds typically denied members the choice of their own doctor (Schepers, 1993; Gibbon, 1913, p. 40; US Commissioner of Labor, 1911, p. 620; Herzlich, 1982, p. 245; Frankel & Dawson with the co-operation of Dublin, 1911, p. 208; Mitchell, 1991a, p. 176). The contrast with the compulsory German funds could not have been starker. As enrollment was required for a broad class of workers, they did not suffer from the problems generated by a steadily aging membership (Frankel & Dawson with the co-operation of Dublin, 1911, pp. 55–156). Members were generally allowed their choice of physician from a list of doctors provided by the local medical association (Gibbon, 1913, pp. 31–32). These doctors were not usually on salary but were paid by the procedure, according to a fee schedule negotiated by the funds and medical associations. Federations of funds paid capitation fees to local medical associations, which then divided this income among its members based on the fee schedule (Gibbon, 1913, p. 54). Thus, while German fund physicians were urged to be as economical as possible (Frevert, 1985, p. 647), benefit spending constraints were not as binding as upon French physicians. Perhaps the clearest indication of how the two kinds of funds differed lies in the chief complaint lodged against each system’s physicians. Several sources describe French physicians as performing too little of their therapeutic duties due to insurance constraints (Mitchell, 1991b, p. 141). German funds, on the other hand, feared that physicians judged workers to be incapacitated too readily, resulting in too many payments of sick leave and medical benefits. The results of second examinations confirmed the fears of German insurers. There was little they could do about generous granting of benefits, as the primary sanction imposed on malingering workers was not denial of benefits, but orders for bed rest (Gibbon, 1913, p.112).
200
JOHN E. MURRAY
Compounding the difficulties on the benefit side were structural differences in income sources. Austrian and German funds charged their members dues that were set in terms of a small share of the worker’s pay. In Austria, workers paid 2 percent and their employers paid 1 percent to the funds. As a result, as real wages rose, so did insurer income. This was especially pronounced in Germany, where funds not only set their dues in percentage terms, but also raised that share over time. For example, 42.3 percent of establishment funds charged members 2 percent or less of earnings in 1888, but by 1908 less than 20 percent did; the share of firms charging over 3 percent rose from 3.3 to 26.6 percent. Voluntary funds in Belgium and France were already thought to be charging ‘‘too much’’; that is, so much as to reduce the demand for coverage to less than politically desirable levels. Thus political and economic constraints prevented them from raising their premium levels. From 1895 to 1907, receipts per member among Belgian recognized funds were constant. In Austria over this time they rose by a third and in Germany they nearly doubled (US Commissioner of Labor, 1911, pp. 280, 488, 1218, 1223, 1238). These sources of financial difficulties on the income side were independent of information asymmetries, but mattered nonetheless. Financial concerns that were ultimately due to selection problems heavily influenced voluntary fund clinical decisions. For workers in all funds, higher levels of sick pay decreased the cost of absenteeism and so increased days missed from work. This is the absenteeism moral hazard of sick pay, which influenced the demand for paid days off by all covered workers. The source of the different effects of sick pay must then lie in the funds, not the workers. Fund finances were critical in determining whether the supply of sick benefits would meet demand, as summarized at the time by the French statistician and reformer Jacques Bertillon (1892, p. 561): The fact is that when these societies grant compensation they attach less importance to their regulations than to the state of their till. A rich society gives its help more liberally than a poor one; and this is absolutely the sole cause of the large English societies, which are often very old and generally rich, granting more daily indemnities than the French (for instance), who are obliged to exercise the strictest economy.
Read ‘‘German’’ for ‘‘English’’ and here is the crux of the matter. The more expensive the sick leave, the harder it was for ill workers to extract additional days of absenteeism from French and Belgian fund officials. To paraphrase Bertillon, however, the more stable German funds were not nearly as constrained in granting their benefits.
Worker Absenteeism under Voluntary and Compulsory Sickness
201
5. CONCLUSIONS Social insurance affects labor supply. Economists have primarily been concerned with long-term declines in labor force participation rates associated with disability insurance (Parsons, 1980) and relatively short-term declines in labor demand associated with tax increases to pay for such insurance (Gruber & Hanratty, 1995). Very short-term effects of sick pay or medical insurance on absenteeism have been examined, but rarely in tandem (Gilleskie, 1998 is an exception). In the United States, for as long as the health care system is funded by employer-provided insurance for workers, the debates of the early 1990s on greater government intervention in health care will likely continue, and the European case will serve as a benchmark for comparison (Cutler, 2002). How the earliest social insurance programs influenced worker absenteeism is an important aspect of social insurance that is not yet fully understood. The role of adverse selection in the transition from voluntary to compulsory insurance recalls the Akerlof ‘‘lemons’’ model (Akerlof, 1970). A variety of funding arrangements to provide social insurance was possible: compulsory insurance funded by worker and employer, voluntary insurance funded by worker and civic-minded bourgeoisie, and voluntary insurance heavily subsidized by the government. Each drew a different proportion of bad risks into its pool of coverage, consequences of which appear in the historical record. The fundamental initial decision that most heavily influenced these consequences, in terms of worker absenteeism, was whether to require universal coverage (Newhouse, 1994, p. 3). Financial health of compulsory funds mattered as much as workers’ physical health in determining whether they could stay home from work when sick. Informational asymmetries ultimately overwhelmed the counterefforts of voluntary funds, such as sick tickets, exclusion of older workers, and organized donations from the middle class. Voluntary sick funds maintained their delicate equilibrium by rejecting sick workers’ absenteeism requests, which defeated the purpose of social insurance, and ultimately led to the wider imposition of compulsory insurance throughout Europe.
NOTES 1. Austrian association and registered aid funds were voluntary, as in Germany. Because Austrian data was reported separately for miners, apprentices, and all other funds, the ‘‘all other’’ category for Austria contains mostly but not completely
202
JOHN E. MURRAY
compulsory fund data. In 1906, the voluntary funds accounted for 18 percent of all insured workers; their mortality rate was much higher than that of other funds (US Commissioner of Labor, 1911, pp. 273, 277). 2. For more details see Murray, (2003). See also Spree, (1988). 3. Honorary membership was recorded separately from ‘‘ordinary’’ membership, so that data analyzed below covered only ordinary members. 4. The latter two measures depend on the number of claims filed, and here the data are not strictly comparable. German and Austrian funds reported the number of claims. Danish funds did not report any claim information. French and Belgian funds did not report the number of claims but did report the number of members who filed claims. A few members who filed many claims could have caused the incidence rates as measured by either claims or claimants per member to differ from each other. However, because Austrian funds reported both claims and claimants, it is possible to estimate correlation coefficient of claims per member and claimants per member by year, which turned out to be r ¼ .99. Thus, the difference in definitions of incidence seem do not seem critical. 5. Recession years included Austria, 1893, 1900, 1901, and 1903; Belgium, 1888, 1891, and 1901; France, 1895, 1897, and 1900–1902; Germany, 1886, 1891, and 1901. 6. See also Riley (1997, pp. 99–104) on similar practices of surveillance among English-friendly societies. 7. In a stricter sense, moral hazard properly refers to behavioral changes by covered persons that increase the likelihood of the insured-against event. 8. Present day effectiveness of medical approval is questionable. Fenn (1981) showed that requiring a medical certificate had no significant effect on duration of absence among modern British workers. 9. Few workers exercised this right. For example, in Ko¨nigsberg in 1909, 15 of 2,635 second examinations were performed at the worker’s request. 10. By contrast, some compulsory funds attempted to charge experience-rated premiums. In Austrian funds, the worker’s age when he joined the fund determined his dues for the duration of his membership, with older entrants paying more (US Commissioner of Labor, 1911, pp. 233–234). 11. The same pattern of older membership can be seen in the voluntary association funds in Austria, 1891–1895, relative to all Austrian males (US Commissioner of Labor, 1911, p. 362). 12. Danish funds recognized that a particularly complicated example arose when workers claimed to be incapacitated due to hernia, since this condition arose over a long period of exertion which may have begun before the worker was a member of a particular sick fund (US Commissioner of Labor, 1911, p. 594).
ACKNOWLEDGMENTS George Alter, Phil Coelho, Lee Craig, Bob Drago, Herb Emery, Martin Gorsky, Bernard Harris, Sheila Ryan Johansson, Kate Lynch, Deirdre McCloskey, Allan Mitchell, Don Parsons, Chen Song, Werner Troesken, seminar participants at Carnegie Mellon University, the Cliometrics,
Worker Absenteeism under Voluntary and Compulsory Sickness
203
Economics and Human Biology, and Social Science History Association conferences, and a referee provided helpful comments on earlier drafts. Alex Field and the anonymous reader for REH were especially helpful. Jo¨rg Baten and Deni Franjkovic kindly provided some otherwise hard-to-find German data.
REFERENCES Akerlof, G. A. (1970). The market for ‘lemons’: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 84, 488–500. Allen, S. G. (1981). An empirical model of work attendance. Review of Economics and Statistics, 63, 77–87. Baldwin, P. (1990). The politics of social solidarity: Class bases of the European welfare state, 1875–1975. New York: Cambridge University Press. Barmby, T. A., Ercolani, M. G., & Treble, J. G. (2002). Sickness absence: An international comparison. Economic Journal, 112, F315–F331. Barmby, T. A., Orme, C. D., & Treble, J. G. (1991). Worker absenteeism: An analysis using microdata. Economic Journal, 101, 214–229. Bertillon, J. (1892). Morbidity and mortality according to occupation. Journal of the Royal Statistical Society, 55, 559–600. Buchmueller, T., & DiNardo, J. (2002). Did community rating induce an adverse selection death spiral? Evidence from New York, Pennsylvania, and Connecticut. American Economic Review, 92, 280–294. Butler, R. J., & Worrall, J. D. (1985). Work injury compensation and the duration of nonwork spells. Economic Journal, 95, 714–724. Bynum, W. F. (1994). Science and the practice of medicine in the nineteenth century. New York: Cambridge University Press. Cutler, D. (2002). Equality, efficiency, and market fundamentals: The dynamics of international medical-care reform. Journal of Economic Literature, 45, 881–906. Doherty, N. A. (1979). National insurance and absence from work. Economic Journal, 89, 50–65. Drago, R., & Wooden, M. (1992). Determinants of labor absence: Economic factors and workgroup norms across countries. Industrial and Labor Relations Review, 45, 764–778. Dreyfus, M. (1996). Mutual benefit societies in France: A complex endeavour. In: M. van der Linden (Ed.), Social security mutualism: The comparative history of mutual benefit societies (pp. 209–224). Bern: Peter Lang. Dutton, P. V. (2002). Origins of the French welfare state: The struggle for social reform in France, 1914–1947. New York: Cambridge University Press. Fenn, P. (1981). Sickness duration, residual disability, and income replacement: An empirical analysis. Economic Journal, 91, 158–173. Fishback, P. V., & Kantor, S. E. (1992). ‘Square Deal’or raw deal? Market compensation for workplace disamenities, 1884–1903. Journal of Economic History, 52, 826–848. Frankel, L. K., Dawson, M. M. with the co-operation of Dublin, L. I. (1911). Workingmen’s insurance in Europe. New York: Charities Publication Committee.
204
JOHN E. MURRAY
Frevert, U. (1985). Professional medicine and the working classes in imperial Germany. Journal of Contemporary History, 20, 637–658. Gibbon, I. G. (1913). Medical benefit: A study of the experience of Germany and Denmark. New York: E. P. Dutton. Gilleskie, D. B. (1998). A dynamic stochastic model of medical care use and work absence. Econometrica, 66, 1–45. Gruber, J., & Hanratty, M. (1995). The labor–market effects of introducing national health insurance: Evidence from Canada. Journal of Business and Economic Statistics, 13, 163–173. Henrekson, M., & Persson, M. (2004). The effects on sick leave of changes in the sickness insurance system. Journal of Labor Economics, 22, 87–114. Herzlich, C. (1982). Evolution of relations between French physicians and the state, 1880–1980. Sociology of Health and Illness, 4, 241–253. Huberman, M., & Lewchuk, W. (2003). European economic integration and the labour compact, 1850–1913. European Review of Economic History, 7, 3–41. Ito, H. (1980). Health insurance and medical services in Sweden and Denmark, 1850–1950. In: A. J. Heidenheimer & N. Elvander (Eds), The shaping of the Swedish health system (pp. 44–67). New York: St. Martin’s Press. Johansson, P., & Palme, M. (1996). Do economic incentives affect work absence? Empirical evidence using Swedish micro data. Journal of Public Economics, 59, 195–218. Ko¨hler, P. A., Zacher, F., & Partington, M. (1982). The evolution of social insurance, 1881–1981: Studies of Germany, France, Great Britain, Austria, and Switzerland. London: Frances Pinter for the Max-Planck-Institut. Maddison, A. (1995). Monitoring the world economy. Paris: OECD. Mitchell, A. (1991a). The function and malfunction of mutual aid societies in nineteenthcentury France. In: J. Barry & C. Jones (Eds), Medicine and charity before the welfare state (pp. 172–189). London: Routledge. Mitchell, A. (1991b). The divided path: The German influence on social reform in France after 1870. Chapel Hill: University of North Carolina Press. Murray, J. E. (2003). Social insurance claims as morbidity estimates: Sickness or absence? Social History of Medicine, 17, 225–245. Newhouse, J. P. (1994). Symposium on health care reform. Journal of Economic Perspectives, 8, 3–11. Parsons, D. O. (1980). Decline in male labor force participation. Journal of Political Economy, 88, 117–134. Riley, J. C. (1997). Sick, not dead: The health of British workingmen during the mortality decline. Baltimore: Johns Hopkins University Press. Schepers, R. (1993). The Belgian medical profession and the sickness funds. Sociology of Illness and Health, 15, 375–392. Song, C., & Nguyen, L. (2003). The effect of hernias on the labor force participation of union army civil war veterans. In: D. L. Costa (Ed.), Health and labor force participation over the life cycle: Evidence from the past (pp. 253–310). Chicago: University of Chicago Press for NBER. Spree, R. (1988). Health and social class in imperial Germany: A social history of mortality, morbidity, and inequality. Oxford: Berg. Thomasson, M. (2004). Early evidence of an adverse selection death spiral: The case of Blue Cross and Blue Shield. Explorations in Economic History, 41, 313–328.
Worker Absenteeism under Voluntary and Compulsory Sickness
205
US Commissioner of Labor. (1911). Twenty fourth annual report of the Commissioner of Labor: Workmen’s insurance and compensation systems in Europe. Washington, DC: Government Printing Office. Verbruggen, P. (1996). The mutualist movement in Belgium. In: M. van der Linden (Ed.), Social security mutualism: The comparative history of mutual benefit societies (pp. 419–429). Bern: Peter Lang. Vistnes, J. P. (1997). Gender differences in days lost due to illness. Industrial and Labor Relations Review, 50, 304–323. Whaples, R., & Buffum, D. (1991). Fraternalism, paternalism, the family, and the market: Insurance a century ago. Social Science History, 15, 97–122. Whiteside, N. (1987). Counting the cost: Sickness and disability among working people in an era of industrial recession, 1920–1939. Economic History Review, 40, 228–246. Winegarden, C. R., & Murray, J. E. (1998). The contributions of early health-insurance programs to mortality declines in pre-World War I Europe: Evidence from fixed-effects models. Explorations in Economic History, 35, 431–446. Wright, C. D. (1905). Coal mine labor in Europe: Twelfth special report of the Commissioner of Labor. Washington, DC: Government Printing Office. Zacher, G., (1898–1908). Die arbeiterversicherung im Ausland. Berlin: Verlag der ArbeiterVersorgung. Zeldin, T. (1979). France 1848–1945: Politics and anger. New York: Oxford University Press.
206
JOHN E. MURRAY
APPENDIX This paper analyzed an unbalanced panel of claim, benefit, and economic data that was compiled from a variety of sources: public and private, government and university, and continental and American. The panel consisted of 16 series that ranged in time from 11 years to 24 years. The sources for each series are listed here by country. Germany Communal, local, establishment, building, guild, registered aid, and state registered aid funds: Kaiserlichen Statistischen Amt, Statistik der Krankenversicherung der Arbeiter im Jahre xxxx, Statistik des Deutschen Reichs, Neue Folge. Volume 38, 1889; volume 90, 1897; volume 96, 1898; volume 121, 1900; volume 170, 1904; and volume 229, 1908. Miners Wright (1905). Twelfth special report of the Commissioner of Labor: Coal mine labor in Europe (p. 339). Washington: GPO; US Commissioner of Labor. (1911). Twenty fourth annual report of the Commissioner of Labor: Workmen’s insurance and compensation systems in Europe (pp. 1252–1254). Washington: GPO. Belgium Recognized and Unrecognized Funds Dr [Georg] Zacher. (1899). Die Arbeiter-Versicherung in Belgien, 1. Band, Heft XII. In: Dr [Georg] Zacher (Ed.), Die Arbeiter-Versicherung im Auslande (pp. 9–11). Berlin: Verlag der Arbeiter-Versorgung. Joseph Begasse. (1906). Die Arbeiterversicherung in Belgien, 3. Band, Heft XIIa: 1. Nachtrag zu Heft XII. In: Dr [Georg] Zacher (Ed.), Die Arbeiter-Versicherung im Auslande (pp. 8, 9). Berlin: Verlag der ArbeiterVersorgung. France Adult and Free Funds Dr [Georg] Zacher. (1898–1908). Die Arbeiter-Versicherung in Frankreich, 1. Band, Heft IV. In: Dr [Georg] Zacher (Ed.), Die Arbeiter-Versicherung im Auslande (pp. 12–14, 17, 18). Berlin: Verlag der Arbeiter-Versorgung.
Worker Absenteeism under Voluntary and Compulsory Sickness
207
Dr [Georg] Zacher. (1902). Die Arbeiter-Versicherung in Frankreich, 2. Band, Heft IVa: 1. Nachtrag zu Heft IV. In: Dr [Georg] Zacher (Ed.), Die Arbeiter-Versicherung im Auslande (pp. 4, 6, 7, 9–11). Berlin: Verlag der Arbeiter-Versorgung. US Commissioner of Labor. (1911). Twenty fourth annual report of the Commissioner of Labor: Workmen’s insurance and compensation systems in Europe (pp. 813, 816, 820). Washington, DC: GPO. Miners: Wright (1905). Twelfth special report of the Commissioner of Labor: Coal mine labor in Europe (pp. 227, 228, 231). Washington, DC: GPO. Denmark Dr. [Georg] Zacher. (1900). Die Arbeiterversicherung in Da¨nemark,1. Band, Heft I. In: Dr [Georg] Zacher (Ed.), Die Arbeiter-Versicherung im Auslande (pp. 6, 7). Berlin: Verlag der Arbeiter-Versorgung. Dr. [Georg] Zacher. (1903). Die Arbeiterversicherung in Da¨nemark, 2. Band, Heft Ia: Nachtrag zu Heft I. In: Dr [Georg] Zacher (Ed.), Die Arbeiter-Versicherung im Auslande (pp. 8, 9). Berlin: Verlag der ArbeiterVersorgung. Aage So¨rensen. (1908). Die Arbeiterversicherung in Da¨nemark, 5. Band, Heft Ib: 2. Nachtrag zu Heft I. In: Dr. [Georg] Zacher (Ed.), Die ArbeiterVersicherung im Ausland (pp. 2, 3). Berlin: Verlag der Arbeiter-Versorgung. Austria Miners: Wright, (1905). Twelfth special report of the Commissioner of Labor: Coal mine labor in Europe (pp. 71, 75). Washington, DC: GPO. Apprentices: US Commissioner of Labor. (1911). Apprentices’ funds. Twenty fourth annual report of the Commissioner of Labor: Workmen’s insurance and compensation systems in Europe (pp. 288–291). Washington, DC: GPO. All other funds: US Commissioner of Labor. (1911). Operations of the sickness insurance system. Twenty fourth annual report of the Commissioner of Labor: Workmen’s insurance and compensation systems in Europe (pp. 270–281). Washington, DC: GPO.
This page intentionally left blank
208
URBAN REAL WAGES AROUND THE EASTERN MEDITERRANEAN IN COMPARATIVE PERSPECTIVE, 1100–2000 S- evket Pamuk ABSTRACT This study examines the long-term trends in wages of skilled and unskilled construction workers in Constantinople-Istanbul, and to a lesser extent in other urban centers in the Near East and the Balkans from about 1100 until the present. It also compares long-term trends in eastern Mediterranean wages with those elsewhere in Europe. Two events had significant and long-lasting impacts on urban real wages around the eastern Mediterranean during the last millennium: the Black Death and modern economic growth. The available price and wage data also point to the existence of a gap in urban real wages between northwestern Europe and the eastern Mediterranean during the first half of the sixteenth century.
INTRODUCTION During the last two decades economists and economic historians have paid a good deal of attention to the estimation of the per capita real product of Research in Economic History, Volume 23, 209–228 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23006-9
209
210
S - EVKET PAMUK
different countries and the analysis of what happened over time to the gap between the leaders and followers. One of the most interesting questions in this respect concerns the emergence and evolution of the gap in levels of real income between today’s developed and developing countries. We know that the gap is large today and can infer from the growth record of both groups of countries that it was smaller or did not exist at all prior to the onset of modern economic growth. Recent research by Angus Maddison and others has confirmed the existence of a gap in 1820 (Maddison, 2001, 2003). There is little information, however, about the period before 1820. How large was this gap in 1750, and was there a gap in 1600 or in 1500? Were levels of income in Europe and Asia comparable before the Industrial Revolution? These inquiries inevitably give rise to questions about the prevailing trends in per capita incomes, productivity, and institutions during the early modern era not only in Western Europe but also in today’s developing regions. Unfortunately, with the exception perhaps of a handful of developed countries, estimates for per capita GDP for the period before 1820 are difficult to construct and not sufficiently reliable. Moreover, it has not been possible to construct detailed estimates for any of the developing countries for the period before 1820 or even 1870. An alternative approach for studying the gap in levels of per capita income or the standards of living has been to compare real wages of specific occupations, most often of skilled and unskilled construction workers in urban areas. Real wage data are of far better quality than per capita GDP estimates for the period before World War I for all of the developing countries and available for a wider sample. In fact, real wage series are virtually the only solid piece of information we have for the standards of living in the developing countries for the period before 1870 if not 1914. In short, real wages continue to be the most reliable source of information about the standards of living of at least part of the population. They also provide the most convenient vehicle for international comparisons of standards of living. Although they cannot be claimed to be ‘‘national’’ in any sense, urban real wage series exist for many regions and large inter-regional differences within the same country are not apparent in these series. Nonetheless, real wage series are open to valid objections. Even if we accept the representative wage as an adequate proxy for the annual per capita earnings of labor, this does not mean that it should be a good proxy for income per capita. The latter depends on the further assumption that factor shares across countries are similar. In many parts of Europe and Asia during the early modern era and until World War I, incomes of households were often determined by
Urban Real Wages around the Eastern Mediterranean
211
changes in employment levels, participation ratios of men, women, and children, and above all, by non-market incomes. Despite these qualifications, the link between wages and the standards of living remains. A decline in real wages did result in a decline in the standards of living or welfare of the household because more labor had to be supplied to buy the same amount of goods, thereby leading either to a decline in other types of income, or in the case where the household responded to the decline in real wages by working harder or longer, a decline in leisure time. (Van Zanden, 1999; De Vries, 1993). Among urban workers, construction workers were a relatively homogeneous category of labor over time and space. Moreover, in contrast to the payments made to other employees, urban construction workers received a high proportion if not all of their pay in cash rather than in kind or in the form of shelter, food, and clothing. As a result, their wages allow for useful inter-country comparisons between preindustrial societies. Utilizing a large volume of archival documents and other sources, this study examines the long-term trends in the wages of skilled and unskilled construction workers in Constantinople-Istanbul, and to a lesser extent in other urban centers in the Near East and the Balkans, from about 1100 until the present. Data from three different sources are used for the study. For the period before 1450, a recently gathered set of wage and price data from Byzantine sources are employed. For the Ottoman period until World War I, we rely on the results of a recently published study which utilized detailed wage and price data collected from large numbers of account books and price lists located in the Ottoman archives in Istanbul (Ozmucur & Pamuk, 2002). Finally, for the period after World War I, real wages of manufacturing workers in Turkey available mostly from the official publications are linked to the real wage series for urban construction workers from the Ottoman era. We thus arrive at a reliable series for a large region of the Old World stretching back almost a millennium. We will then compare long-term trends in eastern Mediterranean wages with long-term wage trends in other parts of Europe. The key issue in this comparison will be the origins and evolution of a wage gap between the eastern Mediterranean and other parts of Europe. For this purpose, we will make use of the recent literature on European wages in the late Middle Ages and the early modern era, most notably the study by Bob Allen which was published at the same time as our previous study on Ottoman prices and wages (Allen, 2001). Several basic conclusions emerge from this study. First, two events had significant and long-lasting impacts on urban real wages around the eastern
212
S - EVKET PAMUK
Mediterranean during the last millennium: the Black Death and modern economic growth. The Black Death caused urban real wages to rise by as much as 100 percent and remain above their pre-Black Death levels until population growth during the sixteenth century reversed most if not all of those gains, not only in Western Europe and the western half of the Mediterranean but also around the eastern Mediterranean. Second, while local conditions always had a significant impact on real wages in specific countries, wage differentials between the eastern and western halves of the Mediterranean remained limited until the nineteenth century. Third, differences in real wages both between the eastern and western halves of the Mediterranean and also between individual countries have increased since the Industrial Revolution, during both the nineteenth and twentieth centuries. We begin below with a discussion of the data and our methods of index construction for each of the three time periods. The paper concludes with an overview and interpretation of the results.
BYZANTINE PERIOD, 1100–1453 The Byzantine Empire with its capital city at Constantinople ruled over large parts of the eastern Mediterranean for most of the Middle Ages. While archival documents are not available from the Byzantine era, a group of Byzantine historians has recently published a sizable collection of observations of incomes, wages, and prices gathered from a variety of manuscripts. (Morrison & Cheynet, 2002; Cheynet, Malamut, & Morrison, 1991; Morrison, 1989). In this collection, observations are available as daily or annual wages or incomes of different types of workers, government officials, soldiers, ecclesiastics, and professionals in different locations across the empire. These observations cover the entire Byzantine period, but they become sufficiently detailed only for the period after 1100. For that later period, observations of nominal wages for skilled and unskilled urban construction workers do not exceed two dozen in number. More than half of the available observations on wages of construction workers are for the capital city, but observations are also available for Crete, Salonica, and other locations. Moreover, the absolute levels as well as changes over time and spatial changes in the levels of other income and wage observations in this data set can be utilized to obtain additional information about and to increase our confidence in the available observations of the wages of urban construction workers.
213
Urban Real Wages around the Eastern Mediterranean
As for prices, the data set provides sufficiently large number of observations only for two standard commodities, namely wheat and olive oil. In addition, the limited number of observations for other cereals, especially barley, meat, other animal products and livestock can be used for obtaining additional information about prices and the aggregate price level at different points in time. In addition, long-term changes in slave prices were used to gain additional insights into trends in wages. While we were able to deflate nominal wages in the Ottoman period by the price index of a detailed basket of consumer goods, for the Byzantine period nominal wages were deflated by an index consisting of the prices of wheat and olive oil. The Byzantine real wage indices for skilled and unskilled construction workers were then linked to the corresponding indices for the Ottoman period. Our calculations based on the Byzantine wage and price observations indicate that prices and nominal wages began to rise during the eleventh century due to fiscally motivated debasements (Kaplanis, 2003). However, real wages remained roughly unchanged until mid-14th century as summarized in Graph 1. While the available data are not as detailed as one would like, there can be no doubt that the Black Death led to a large long-term increase in nominal and real wage levels around the eastern Mediterranean. Annual wages of skilled workers jumped from less than 20 gold hyperpyra to more than 50 hyperpyra while food prices showed only modest increase after mid-century. Wheat prices showed large fluctuations between good 6.0
5.0
Skilled Unskilled Avg. Manuf. Wages, 1914-2000 (1914 = 1,28)
4.0
30-Year Moving Averages, Skilled 30-Year Moving Averages, Unskilled
3.0
2.0
1.0
0.0 1000
Graph 1.
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
Real Daily Wages of Construction Workers in Constantinople-Istanbul, 1100–2000 (unskilled in 1500 ¼ 1,0).
S - EVKET PAMUK
214
and poor harvest years, but normal prices increased from about 1/6 or 1/5 hyperpyra per modios (12.8 kg or 17 l) to about 1/4 hyperpyra at the end of the century. Urban real wages at the end of the fourteenth century were above their pre-Black Death levels by as much as 100 percent and they remained well above their pre-Black Death levels until the end of the sixteenth century, 150 years into the Ottoman era. This large jump in the urban wage levels was paralleled by and confirmed further by the doubling of slave prices across the Byzantine territories during the second half of the fourteenth century (Morrison & Cheynet, 2002, pp. 847–850). In order to facilitate comparisons with the Ottoman period for which more detailed price and wage data are available, we have converted the Byzantine and Ottoman wage and price observations for the fifteenth century into a common form, grams of silver per metric units. These observations, which are summarized in Table 1, reflect the continuity between the two eras despite the radical break in economic policy, monetary units, and metrology. Such continuity and comparability increase our confidence in the available observations on both sides of the year 1453. While the price and wage series expressed in common units are comparable before and after 1453, it is also clear that the prices were higher and real wages were lower during the Byzantine period. The most important explanation for this pattern was the deterioration of the Byzantine economy during its last century. As the territory under the control of the Byzantine state shrank, Constantinople Table 1. Comparisons between Byzantine and Ottoman Prices and Wages in Constantinople-Istanbul in Constantinople-Istanbul. Averages
Byzantine, 1400–1450
Ottoman, 1460–1500
Wheat prices in own units Wheat prices in common units Olive oil prices in own units Olive oil prices in common units Daily wages in own units skilled construction workers Daily wages in common units skilled construction workers Daily wages in own units unskilled construction workers Daily wages in common units unskilled construction workers
0.4 hyperpyra/modios 31.3 g silver/100 l 2 1/3 hyperpyra/10 l 3.10 g silver/l
12.8 akches/kile 23.5 g silver/100 l 4.8 akches/okka 2.82 g silver/l 8.75 akches/day
4.28 g silver/day
5.95 g silver/day
75 hyperpyra/year
4.77 akches/day
2.86 g silver/day
3.24 g silver/day
Sources: Morrison and Cheynet (2002) and Pamuk (2001). The gold:silver ratio was approximately 10 for this period. Annual wages are converted into daily wages at 180 days/year.
Urban Real Wages around the Eastern Mediterranean
215
often had difficulties in securing its food and raw materials from the surrounding regions. In the aftermath of the Black Death, urban real wages also registered large increase elsewhere around the eastern Mediterranean, well beyond Byzantine territories. Ashtor (1976) provides detailed evidence that urban real wages roughly doubled after the Black Death and stayed high during the fifteenth century in Egypt and Syria. Similar increase in urban real wages occurred in the Balkans as well (Morrison & Cheynet, 2002). It also appears on the basis of the Byzantine evidence that the skilled- and unskilled-wage differentials declined after the Black Death. However, more detailed data would be necessary to establish this latter trend with greater certainty (see Graph 1).
OTTOMAN PERIOD, 1489–1914 For our price series in the Ottoman period, we utilized data on the prices of standard commodities (food and non-food items) collected from more than 6000 account books and price lists located in the Ottoman archives in Istanbul. The food indices included the prices of ten leading items of consumption, namely flour, rice, honey, cooking oil, mutton, chickpeas, lentils, onions, eggs, and olive oil for burning. Among these, flour, rice, cooking oil, mutton, olive oil and honey provided the most reliable long-term series and carried the highest weights in our food budget. In cases where the prices of one or more of these items were not available for a given year, missing values were estimated by an algorithm that applied regression techniques to the available values (Ozmucur & Pamuk, 2002; Pamuk, 2001). Based on the available evidence regarding the budget of an average urban consumer, the weight of food items in the overall indices was fixed between 75 and 80 percent. The weight of each commodity in the overall index was then based on the shares of each in total expenditures of the respective institutions. To cite two prominent examples, in the absence of long series on bread prices, the weight of flour, mostly wheat flour, varies mostly between 32 and 40 percent of food expenditures and 24–32 percent of overall expenditures, depending on the fluctuations in prices. Similarly, the weight of meat (mutton) varies between 5 and 8 percent of the overall budget. Prices of non-food items obtained from a variety of sources, most importantly the palace account books, were then added to the indices. These commodities are soap, wood, coal, and nails by weight (used in construction and repairs). Cloth was not included in our indices because we were unable
216
S - EVKET PAMUK
to ensure that the long-term price series obtained from the archival documents belonged to cloth of the same quality. For the period 1860–1914, data from archival sources are limited. For this reason, the detailed quarterly wholesale prices of the Commodity Exchange of Istanbul covering about two-dozen commodities and a separate series for imported cotton textiles were used. Indices based on these prices were then linked to those for the earlier period with the help of detailed data for both retail prices of individual commodities and prices at the Commodity Exchange for 1860–1862 and 1913–1914. Daily wage data were gathered from more than five thousand account books of the construction and repair sites in Istanbul and other cities. These account books contain daily wages for both unskilled and a variety of skilled construction workers. Wages for unskilled workers referred mostly to one type of worker, called irgad in the early period and renc- ber after about 1700. In contrast, daily wage rates could be found in for more than half a dozen categories of skilled construction workers in these account books. In order to utilize the additional information, an index was constructed for skilled wages that included the wages of carpenters, masons, stonecutters, ditchdiggers, plasterers, and others. Based on the relative frequency with which they appeared in the account books, the greatest weight in this index was given to the category of neccar, specialists who built wooden houses and the wooden parts of buildings. There also existed a separate category of carpenter (marangoz), which apparently referred more to makers of furniture. The share of neccar fluctuated between 50 and 60 percent in our skilled wage index. Istanbul was chosen primarily because the data were most detailed for the capital city. However, price and wage data from the account books of the pious foundations is available for other cities of the empire as well. Price observations for a shorter list of commodities similarly obtained from the account books of pious foundations in the Ottoman archives in Istanbul were used to construct separate series for the cities of Edirne, Bursa, Konya, Trabzon, Damascus, and Jerusalem. In these Ottoman cities, the overall change in the price level from 1490 to 1860 and the medium term trends were quite similar to the price trends in Istanbul (Pamuk, 2001, Graph 3.1 and Appendix Tables 5.1 through 5.6). Price data gathered by Ljuben Berov suggest that the Balkans experienced similar increases in prices during the sixteenth and seventeenth centuries (Berov, 1976; a summary is available in Berov, 1974). The evidence, thus points to similar price trends for the region stretching from the Balkans through Anatolia to Syria. In Egypt, the local currency was the para or medin whose silver content and rate of debasement
Urban Real Wages around the Eastern Mediterranean
217
differed from those of the akc- e. Nonetheless, it is possible to construct price indices in grams of silver for Cairo on the bases of data supplied by Andre Raymond from the court records of that city (Pamuk, 2001, Appendix, Table 5.7; Raymond, 1973–1974, Vol. I, pp. 17–80). These indices indicate that prices in Cairo expressed in grams of silver moved together with those in Istanbul and other Ottoman cities in the akc- e region until 1800. Observations on the daily wages of skilled and unskilled construction workers also available from the account books of the pious foundations and collected by Andre Raymond for Cairo show clearly that nominal wages in other Ottoman cities showed similar trends during these four centuries (Raymond, 1973–1974, Vol. II, pp. 383–386; Hanna, 1984, pp. 43–46). In other words, although we are unable to offer the same details for other cities around the eastern Mediterranean, we are confident that the long-term trends we established for the city of Istanbul also closely reflect the patterns in other urban areas. Real wages of construction workers in Istanbul and other urban centers across the Ottoman Empire during the century after 1453 were higher than corresponding real wages in Constantinople and other Byzantine urban centers during the century before 1453, especially for skilled construction workers. This difference is in large part due to the difficulties of the Byzantine society and economy during its last century of existence. Constantinople often experienced problems in securing its food supply during this period. From their relatively high levels at the end of the fifteenth century, Ottoman urban real wages experienced a steady and large decline during the sixteenth century, by as much as 40 percent. This trend was due, at least in part, to the increases in population around the eastern Mediterranean during the sixteenth century. With a long-term perspective, one may thus interpret the sixteenth century decline in real wages around the Mediterranean and in many parts of Europe as the second and reverse leg of a movement that began with the sharp increases after the Black Death during the second half of the fourteenth century. An inverse-V-shaped pattern in urban real wages was thus completed by the end of the sixteenth century and urban real wages in the eastern Mediterranean stood close to their levels before the Black Death (for Italy, see Malanima, 2004; see Fig. 4). After remaining roughly unchanged until the middle of the eighteenth century, the Ottoman urban real wages increased by about 20–30 percent from the late eighteenth until the mid-nineteenth century and then by another 40 percent during the late nineteenth and early twentieth centuries. On the eve of World War I, real wages of unskilled construction workers were
218
S - EVKET PAMUK
about 10–20 percent above their levels in 1500. Because relative prices shifted in favor of goods consumed by higher-income consumers during these centuries and because the skill premium began to rise late in the nineteenth century, real wages of skilled workers in 1914 stood at more than 50 percent above their levels in 1500 (Ozmucur & Pamuk, 2002, pp. 303–304). In comparison to the first half of the fourteenth century, that is, the era before the Black Death, both unskilled and skilled urban wages stood in on the eve of World War I at least 100 percent higher. Urban wages were undoubtedly a limited category in the Ottoman economy. Nonetheless, there can be no doubt that these very long-term trends in urban wages reflect changes in the productivity and income levels of the underlying economy, most importantly in the agricultural sector where more than three-fourths of the labor force were still employed on the eve of World War I. In the light of this new evidence, we now need to consider the possibility of a slow and modest rise in labor productivity around the Eastern Mediterranean in the era before the Industrial Revolution.
URBAN REAL WAGES SINCE WORLD WAR I With the spread of industrialization, wages of urban construction workers have declined in importance as a wage category. For the same reason, it has become increasingly difficult to obtain regular observations on this category from the published statistics of individual countries. In contrast, information on the wages of manufacturing workers is much more readily available. As a result, we decided to link our wage series for the construction workers until World War I with the more readily available series of national average wages for manufacturing workers for the period after 1914. Taking into account the relative levels of both construction and manufacturing wages in the period 1900–1930 and especially the sectoral wages provided in the Ottoman Industrial Census of 1913–1915, we decided to link the national average daily manufacturing wage to the Istanbul daily construction wages at 16 percent above the wage of an unskilled construction worker and at 41 percent below the wage of a skilled construction worker in 1913–1914. Wages of skilled construction workers at Istanbul were 118 percent higher than those of unskilled construction workers during 1913–1914. For our national average manufacturing wage series, we utilized the Ottoman Industrial Census of 1913–1915, the nominal manufacturing wage series prepared by Tuncer Bulutay until the early 1990s which revises and utilizes various official series and the average manufacturing wage indices prepared
Urban Real Wages around the Eastern Mediterranean
219
by the State Institute of Statistics for the most recent period.1 For the deflator, we used the consumer price indices for Istanbul prepared by the Istanbul Chamber of Commerce, for Ankara prepared by the Under secretariat for the Treasury and Foreign Trade until 1960 and the urban areas consumer price index prepared by the State Institute of Statistics for the period since 1960 (Pamuk, 2001, pp. 19, 53). After remaining below their pre-World War I levels until after World War II, real manufacturing wages in Turkey increased by 300 percent from 1950 until the end of the century. Urban real wages, thus exhibited the largest increases during the last millennium in the half century since 1950 (Graph 1). This pattern applies to all countries and regions around the eastern Mediterranean from the Balkans to Syria and Egypt. For most countries around the eastern Mediterranean, real wages did not show significant increases during the period 1914–1950. The notable exception was that part of Mandate Palestine, which ended up as the state of Israel where real wages more than doubled from 1914 to 1950. In the period since 1950, the largest increases in urban real wages around the eastern Mediterranean occurred in Greece and Israel (3–5-fold), while real wage increases were most limited in Egypt, Syria, and in former Yugoslavia (less than 2-fold or 200 percent increase). Real wage increases in Turkey fall between these two groups of countries. Even in the slow-growing countries, in Egypt, Syria, and the former Yugoslavia, however, real wage increases since 1950 were greater than those in any other century or any other two-century period during the last millennium.2
A COMPARATIVE PERSPECTIVE, 1300–2000 In this section, we will examine long-term trends of urban real wages around the eastern Mediterranean within a Mediterranean and European comparative perspective. For this purpose, we will again adopt a three-period approach. For the late medieval period before and after the Black Death, all available evidence points to a similar pattern in urban real wages on both sides of the Mediterranean, east and west. The Black Death led to a large long-term increase in nominal and real wage levels. Urban real wages at the end of the fourteenth century were above their pre-Black Death levels by as much as 100 percent in the eastern and western halves of the Mediterranean in Italy, France, and Spain, as well the Balkans, the Byzantine Empire, Syria, and Egypt.3 They declined somewhat but remained well above their pre-Black
220
S - EVKET PAMUK
Death levels through the fifteenth century. During the fifteenth century, nominal and real wages of Constantinople and Cairo construction workers were close to but appear to be lower than those in Italy (Ashtor, 1969, pp. 511–524). Real wages of urban construction workers in England exhibited a similar pattern during the fourteenth and fifteenth centuries. However, due to the limitations of the data, it is not easy to compare the levels of Eastern Mediterranean and English real wage levels for this early period. (Clark, 2005.) For a comparison of the urban real wages around the eastern Mediterranean with those in other parts of Europe from the sixteenth century until World War I, we make use of the recent study of prices and wages in European cities by Bob Allen. Allen utilized a large body of data most of which were compiled during the early part of this century by studies commissioned by the International Scientific Committee on Price History founded in 1929. In order to facilitate comparisons, he converted all price and wage series into grams of silver and chose as a base to the index of average consumer prices prevailing in Strasbourg during 1700–1749 (Allen, 2001). Allen has argued that even though wages in a single city may be accepted as a barometer of wages in the whole economy, international comparisons need to be made between the cities at similar levels in the urban hierarchy. Since his study uses data from cities at the top of their respective urban hierarchies such as London, Antwerp, Amsterdam, Milan, Vienna, Leipzig, and Warsaw, it would make sense to insert Istanbul, another city at the top of the urban hierarchy of its region, into this framework. It is not very difficult to do so since prices and wages are already expressed in grams of silver in the present study. However, it was still necessary to express Istanbul prices in terms of the Allen base of Strasbourg 1700–1749 ¼ 1.0. For this purpose, Ottoman commodity prices for the interval 1700–1749 were applied to Allen’s consumer basket with fixed weights. A second and equally useful method of linking Istanbul’s price level to those of other European cities in the Allen set was to employ the detailed annual commodity price series gathered by Earl Hamilton for Valencia and Madrid for 1500–1800 and compare them with the Istanbul prices for the same commodities (Hamilton, 1934, 1947, Appendices). Since Valencia and Madrid prices were already calibrated into the Allen set, it was then possible to determine the Istanbul price level vis-a`-vis European cities for each interval. The price series for flour, mutton, olive oil, cooking oil, onions, chickpeas, pepper, sugar, and wood were used in these calculations. The two procedures produced results that were quite similar.
221
Urban Real Wages around the Eastern Mediterranean
Our indices show that daily wages in Istanbul and other eastern Mediterranean cities expressed in grams of silver were comparable to many other locations in northern and southern Europe in the early part of the sixteenth century. However, since Istanbul prices were higher than those in other cities in Allen’s sample, real wages in Istanbul varied between 60 and 90 percent of real wages in other cities during that period (Graph 2). It is interesting that while real wages continued to decline after 1600 in southern and many other parts of Europe, they remained little changed in Istanbul during the seventeenth and until late in the eighteenth century leading to greater convergence with other parts of Europe except the northwest. A wage gap of one-third to one-half between Istanbul and the leading cities in northwestern Europe continued until the Industrial Revolution. In Table 2, we offer another comparison of the real wages in eastern Mediterranean with those in northwestern and southwestern Europe before the Industrial Revolution. For this purpose we present and make use of more detailed price data as well as wage indices for western Netherlands, London, Valencia, and Madrid collected by Jan Luiten Van Zanden, Bob Allen, and Earl Hamilton respectively in addition to our own for Istanbul.4 Table 2 shows that nominal wages of skilled and unskilled construction 16 14 12 10
Istanbul
Antwerp+Amsterdam
London
Paris
Valencia
Leipzig
Vienna
Warsaw
8 6 4 2 0 14501499
Graph 2.
15001549
15501599
16001649
16501699
17001749
17501799
18001849
18501899
19001913
Real Wages of Unskilled Construction Workers in European Cities, 1450–1913 (Wages in Grams of Silver/CPI).
222
Table 2.
Real Wage Comparisons between Istanbul and Other European Urban Centers, 1500–1750. 1500–1550 Istanbul
Western Netherlands
1700–1750 London
Valencia
Istanbul
Western Netherland
London
Madrid
3.30 5.17
8.60 13.36
10.50 14.70
5.70 11.60
0.36 0.41 1.30 3.89 2.51 0.80 0.94 1.70 12.37
0.87 0.62 4.07 5.35
0.59 3.12 5.63
Wages, nominal, in grams silver Unskilled Skilled
3.37 5.94
3.45 5.66
3.20 5.00
4.20 6.50
0.28 0.33 1.23 1.79
0.19 0.76 1.85
0.54 2.10 3.74 1.95
0.55
0.27
Prices, in grams silver per kilogram Bread Wheat Mutton Butter Olive Oil Chick Peas Rice Honey Sugar Black Pepper Soap Coal
0.38 0.42 1.10 4.22 2.67 0.59 0.87 2.67 15.95 17.93 2.67
0.47 4.38 15.86 1.09
3.50 0.06
0.72 1.79 7.95 33.52 1.64
0.58 4.62
7.42
8.10
3.00 0.22
2.28
7.00 0.17
3.04 0.20
0.90
1.58
1.63
1.30
Price indices (Istanbul 1500–1550 ¼ 1,0) 1.00
0.68
0.63
0.92
Real wages Istanbul (Unskilled, 1500–1550 ¼ 100) Unskilled Skilled
100 176
151 247
151 236
135 209
109 170
162 251
191 268
130 265
Sources: Pamuk (2001), Allen (2001), van Zanden at http://www.iisg.nl/hwp, De Vries and Van der Woude (1997), and Hamilton (1934, 1947).
S - EVKET PAMUK
1.57
0.47 3.66 4.21 2.73 1.56 2.49 2.96 6.08
Urban Real Wages around the Eastern Mediterranean
223
workers measured in grams of silver were quite comparable in these four regions during the first half of the sixteenth century. However, our direct comparison of the prices of more than half a dozen commodities which are available for both regions indicate that, in grams of silver terms, Istanbul and Madrid prices were higher by about 50 percent than those in western Netherlands and London during the same period. By the first half of the eighteenth century, prices and nominal wages in western Netherlands and London measured in grams of silver had risen by about two and a half times, with the nominal wages lagging somewhat behind prices. Prices and nominal wages in Spain rose to a lesser extent during the same period. In Istanbul, on the other hand, prices and nominal wages during the first half of the eighteenth century, measured in grams of silver, were not very different from those of two centuries earlier. In other words, detailed price and wage data presented in Table 2 indicate that real wages in Istanbul were below those in western Netherlands and England by one-third to one-half during the first half of the sixteenth century and this gap remained roughly unchanged until the era of the Industrial Revolution. This more direct and more detailed comparison between Istanbul and the western Netherlands, London, Madrid, and Valencia price and wage series is thus consistent with the real wage trends outlined earlier in Graph 2. Both Graph 2 and Table 2 suggest strongly that we need to look at the period before the sixteenth century for the origins of the wage gap between the eastern Mediterranean and northwestern Europe. It has been frequently argued that in the absence of detailed price series, it would still be useful to deflate nominal wages by wheat or grain prices and arrive at ‘‘wheat or grain wages’’ as a reasonable approximation of the purchasing power of wages, especially in view of the large share of cereals in the average consumer basket during the late medieval and early modern periods. As Jan Luiten Van Zanden has warned recently, however, wheat or grain wages may at times provide a misleading picture because of the wide variations in grain prices (Van Zanden, 1999). In Table 3 we present nominal wages, grain prices and wages for the two leading urban centers in the eastern Mediterranean, Istanbul, and Cairo, and compare them with grain wages in other European urban centers in the early modern era. The barley prices presented in this table for Poland are unusually low which gives us a distorted picture of the standards of living in this part of the European periphery. For this reason, it may be useful to investigate in greater detail cereal prices in Poland in the future. It is also remarkable that the range of cereal prices around Europe was just as wide around 1800 as it had been around 1600. This pattern is consistent with arguments that commodity
224
Table 3.
S - EVKET PAMUK
‘‘Grain Wages’’ of Unskilled Construction Workers, 1400– 1800. 1400–1420
1500–1520
1600–1620
1680–1700
1780–1800
Daily wages in grams silver South. England Holland Florence-Milan Warsaw-Kracow Istanbul Cairo
2.6
2.9 3.0 2.6 1.0 3.3 2.6
3.7 6.5 5.4 2.8 4.1 2.5
5.7 7.7 3.9 1.8 4.3 2.0
9.2 8.7 3.2 2.9 4.4 1.7
22.2 21.5 39.1 2.3 35.6 28.0
61.5 57.2 103.6 9.0 50.0 38.0
70.6 46.5 46.8 8.0 37.4 26.0
113.2 87.5 81.1 10.3 42.0 19.0
13.2 14.0 6.6 45.1 9.3 9.3
6.0 11.4 5.3 34.8 8.2 6.6
8.1 16.6 9.3 23.2 11.5 7.7
8.1 9.9 6.0 20.6 10.5 9.0
4.7 2.8 2.7
Wheat or rye prices in grams silver/100 l South. England Holland Florence-Milan Warsaw-Kracow Istanbul Cairo
26.5 21.0 22.0 4.0 24.0 30.0
Daily wages in liters of wheat or rye South. England Holland Florence-Milan Warsaw-Kracow Istanbul Cairo
10.0 21.5 11.6 9.0
Sources: Europe: Van Zanden (1999), Allen-Unger (2004); Istanbul: Pamuk (2001), Morrison and Cheynet (2002); Cairo: based on Ashtor (1969), Raymond (1973-1974).
price convergence around Europe remained limited until the institutional changes and the transportation revolution of the nineteenth century. It is also worth noting that because wheat prices did not deviate significantly from wheat prices elsewhere in southern and western Europe, levels of wheat wages in Istanbul and Cairo are in line with what we already know about the purchasing power of wages and the standards of living around the eastern Mediterranean. It thus appears that real wage levels in Cairo were comparable to those elsewhere around the eastern Mediterranean before the Industrial Revolution. However, nominal and real wages in Cairo appear to lag behind those elsewhere in the region during the nineteenth century. Real wage differences between the eastern Mediterranean and western Europe continued to widen during the nineteenth century and until 1950.5
Urban Real Wages around the Eastern Mediterranean
225
Real wage differentiation within the eastern Mediterranean has accelerated since 1950. Greece and Israel have experienced increases in real wages that brought their wage levels closer to those in western Europe. At the other end of the spectrum, wage increases in Yugoslavia, Syria, and Egypt continued to lag behind those in western Europe, widening the wage gap between these two groups of countries even further. On the other hand, urban wage increases in Turkey since 1950 have been roughly comparable to those in western Europe.
CONCLUSION One of the more important questions regarding the world economy in the early modern era concerns the emergence and evolution of the gap in levels of real income between today’s developed and developing countries. With the exception perhaps of a handful of countries, however, estimates for per capita GDP for the period before 1820 are difficult to construct and not sufficiently reliable. An alternative approach for studying the differences in levels of per capita income or the standards of living has been to compare real wages of skilled and unskilled construction workers in urban areas. One needs to be cautious about using daily wages of urban construction workers as indicators of the standards of living for an entire country. Nonetheless, in the absence of reliable information about production and income, real wage series still serve as the best indicator available for long-term trends in standards of living. Utilizing a large volume of archival documents, this study established for the first time the long-term trends in wages of skilled and unskilled construction workers in Constantinople-Istanbul and other urban centers around the Eastern Mediterranean from the twelfth century until World War I. These price and wage series were then inserted into a larger framework of price and wage trends in European cities during the same period. Several basic conclusions emerge from this study. First, two events had significant and long lasting impacts on urban real wages around the Eastern Mediterranean during the last millennium: the Black Death and modern economic growth. The Black Death caused urban real wages to rise by as much as 100 percent and remain above their pre-Black Death levels until population growth during the sixteenth century reversed most, if not all, of those gains not only in Western Europe and the western half of the Mediterranean but also around the eastern Mediterranean.
S - EVKET PAMUK
226
Our real wage series point to the existence of a modest, but statistically significant upward trend in urban wages around the eastern Mediterranean dating back to the seventeenth century. It is thus possible that there occurred a slow rise in productivity across the region in the era before the Industrial Revolution. Increases in urban real wages continued at a higher pace after the Industrial Revolution, during the century before World War I. However, the largest real wage increases to occur around the eastern Mediterranean during the last millennium took place in the half-century since 1950. We may thus conclude that the most significant impact of modern economic growth on the standards of living in this region occurred during the last half century. With the arrival of more rapid economic growth after 1950, significant real wage differences began to emerge within the region as well, between economies with higher and lower rates of economic growth. This study has also provided new insights into the origins and evolution of the wage gap between this region and elsewhere in Europe. The available price and wage data point to the existence of a gap in urban real wages between northwestern Europe and the eastern Mediterranean during the first half of the sixteenth century that persisted until the Industrial Revolution. The available evidence thus suggests strongly that we need to look at the period before the sixteenth century for the origins of the wage gap between the eastern Mediterranean and northwestern Europe. On the other hand, a significant wage gap between the eastern Mediterranean and other regions of Europe cannot be observed before the Industrial Revolution. While local conditions always had significant impacts on real wages in specific countries, wage differentials between the eastern and western halves of the Mediterranean remained limited until the nineteenth century. In contrast, differences in real wages between the eastern and western halves of the Mediterranean have increased substantially since the Industrial Revolution, during both the nineteenth and twentieth centuries.
NOTES 1. Further details on manufacturing wage series can be obtained from Pamuk (2001, pp. 77–78). 2. Maddison (2003); compare with Berov (1979). 3. Ashtor (1969, pp. 517–524). For Italy, trends in urban real wages in the aftermath of the Black Death can be followed in detail from Malanima (2004). 4. The van Zanden indices are available from http://www.iisg.nl/hwp.
Urban Real Wages around the Eastern Mediterranean
227
5. For further evidence on real wage trends in the eastern and western halves of the Mediterranean from the Industrial Revolution until 1950, see Williamson (2000).
ACKNOWLEDGMENTS An earlier version of this paper was presented at the Utrecht conference, ‘‘Towards a Global History of Prices and Wages,’’ in August 2004. The author would like to thank Yigit Akin for research assistance, Gregory Clark, Peter Temin, and the participants of the Utrecht conference for many helpful comments and suggestions.
REFERENCES Allen, R. C. (2001). The great divergence in European wages and prices from the middle ages to the first world war. Explorations in Economic History, 38, 411–447. Ashtor, E. (1969). Histoire des Prix et des Salaries dans l’Orient Medieval. Paris: SEVPEN. Ashtor, E. (1976). A social and economic history of the near east in the middle ages. Berkeley: University of California Press. Berov, L. (1974). Changes in price conditions in trade between Turkey and Europe in the 16th-19th century. Etudes Balkaniques, 3, 168–178. Berov, L. (1976). Prices in the Balkans during the 16th-19th centuries and the European revolution of prices. Sofia: Publishing House of the Bulgarian Academy of Sciences (in Bulgarian). Berov, L. (1979). Wages in the Balkans during the period of manufacturing capitalism and the Industrial revolution. Bulgarian Historical Review, 1, 91–115. Cheynet, J.-C., Malamut, E. & Morrison, C. (1991). Prix et Salaries a Byzance (Xe-XVe Siecle). Hommes et Richesses dans l’Empire Byzantin (Vol. 11, pp. 339–374). Paris. Clark, G. (2005). The condition of the working-class in England, 1200–2004. Journal of Political Economy, 113. De Vries, J. (1993). Between purchasing power and the world of goods: Understanding the household economy in early modern Europe. In: J. Brewer & R. Porter (Eds), Consumption and the world of goods (pp. 85–132). London & New York: Routledge. De Vries, J., & Van der Woude, A. (1997). The first modern economy: Success, failure, and perseverance of the Dutch economy, 1500–1815. Cambridge: Cambridge University Press. Hamilton, E. J. (1934). American treasure and the price revolution in Spain, 1501–1650. Cambridge, MA: Harvard University Press. Hamilton, E. J. (1947). War and prices in Spain, 1650–1800. Cambridge: Harvard University Press. Hanna, N. (1984). Construction work in Ottoman Cairo, 1517–1798. Supplement aux Annales Islamologiques, Cahiers No. 4. Cairo. Kaplanis, C. (2003). The debasement of the ‘dollar of the middle ages. The Journal of Economic History, 63, 768–801. Maddison, A. (2001). The world economy: A millennial perspective. Paris: OECD Development Centre.
228
S - EVKET PAMUK
Maddison, A. (2003). The world economy: Historical statistics. Paris: OECD Development Centre. Malanima, P. (2004). Labor, productivity, wages in Italy, 1270–1913. Paper presented to the conference, Towards a global history of prices and wages, Utrecht, August. Morrison, C. (1989). Prix et Salaries a Byzance (Ve-VIIe Siecle). In Hommes et Richesses dans l’Empire Byzantin (Vol. I, pp. 239–260). Paris. Morrison, C., & Cheynet, J.-C. (2002). Prices and wages in the Byzantine world. In: A. E. Laiou (Ed.), The economic history of Byzantium: From the seventh through the fifteenth century (pp. 815–878). Washington, DC: Dumbarton Oaks. O¨zmucur, S., & Pamuk, S. (2002). Real wages and standards of living in the Ottoman empire, 1489–1914. The Journal of Economic History, 62, 292–321. Pamuk, S. (2001). 500 Years of prices and wages in Istanbul and other cities. Ankara: State Institute of Statistics (in Turkish and English). Raymond, A.(1973–1974). Artisans et Commerc- ants au Caire au XVIIIe Siecle (Vol. 2). Damascus: Institut Francais de Damas. Van Zanden, J. L. (1999). Wages and standards of living in Europe, 1500–1800. European Review of Economic History, 2, 175–195. Williamson, J. G. (2000). Real wages and relative factor prices around the Mediterranean, 1500–1940. In: S. Pamuk & J. G. Williamson (Eds), The Mediterranean response to globalization before 1950 (pp. 45–75). London: Routledge.
JAPANESE UNSKILLED WAGES IN INTERNATIONAL PERSPECTIVE, 1741–1913 Jean-Pascal Bassino and Debin Ma ABSTRACT Constructing consumption baskets for the benchmark periods 1745–1754 and 1882–1886, and price indices, we calculate real wages for Japanese unskilled daily laborers in 1741–1913. Matching caloric content and protein contents in our Japanese consumption baskets with those for Europe, we compare Japanese and European urban real wages. Real wages in Kyoto and later Tokyo are about a third London wages but comparable to wages in major Southern and Central European cities for 1700–1900. In Japan, wages are substantially higher in the Meiji period than in the Tokugawa period. These findings have implications for the debate on conditions in Europe and Asia on the eve of the Industrial Revolution.
1. INTRODUCTION How rich or poor was Japan before she embarked on the path of modern economic growth following the Meiji Restoration (1868)? Intellectual interest in this question, often in curious synchronism with the tempo of the Research in Economic History, Volume 23, 229–248 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23007-0
229
230
JEAN-PASCAL BASSINO AND DEBIN MA
post-War Japanese economic miracle, surged up to the 1980s, and since then, has eased into rational retrenchment. The pre-War pessimist consensus that Tokugawa Japan was an extremely backward society has been dispelled – a victory for the so-called optimists. But the optimists’ subsequent claim that 19th century Japanese living standards could have been comparable to or higher than contemporaneous Northwestern Europe may represent an overreach.1 Two recent assessments of this debate have been far more cautious, giving the 1700–1870 Tokugawa economy a slow but positive 0.1 percent and 0.15 percent growth rate in real wages and per capita GDP respectively, yielding a purchasing power parity (PPP) adjusted to 1870 Japanese per capita income at slightly less than a quarter of that of the British level.2 Per capita GDP estimates for Tokugawa Japan have been, as most scholars agree, highly tentative.3 Yet, studies compiled in the past few decades utilizing the far richer Tokugawa prices and wages data have focused almost exclusively on constructing real growth trends over time, rather than absolute levels across nations. This is not surprising in view of the formidable methodological issues that confound international comparisons even for the contemporary period. Recent path-breaking works by Robert Allen (2001, 2005), which use standard caloric and protein intake from consumption baskets as a benchmark for inter-regional and international comparison, have charted the trends and levels of five centuries of real wages across Europe. Ozmucur and Pamuk (2002) and van Zanden (2003) have extended this line of work beyond Europe. Allen (2005) makes a preliminary attempt to extend his real wage comparisons to Japan, India, and China. Given data and other constraints, however, his calculation of the purchasing power of Japanese real wages (1741–1913) relies on backward projection from the 1882 benchmark year and the use of an Indian consumption basket, both of which, we will show, incur serious index number problems. This article adopts Allen’s methodology for calculating real wages but utilizes actual 18th and 19th century Japanese wage and price data compiled by generations of Japanese scholars and reconstructed consumption baskets based on historical information on 18th and 19th century Japanese consumption patterns. As our Japan–Europe real wage comparison hinges on the caloric and protein contents in their respective consumption baskets, this current study largely focuses on the comparative purchasing powers of unskilled laborers, leaving the comparative studies for skilled laborers or high-income groups to future research that would need to confront more sophisticated issues of cross-national utility and welfare comparison.
Japanese Real Wages
231
Our study reveals that the real wages for Japanese unskilled laborers in Kyoto and Tokyo for the Tokugawa period are roughly a third of the London level but comparable to cities in contemporaneous Southern and Central Europe. We also show that our 1882–1886 benchmark Meiji real wage series is nearly double that of the 1745–1754 benchmark Tokugawa series. With warnings on the limitations of real wage comparisons, this current study seems to lend some tentative support to the view that the preMeiji Japanese initial conditions, while not reflecting abject poverty, may not have been much more favorable than other developing countries such as Turkey or Java (Indonesia) as revealed in the studies by Ozmucur and Pamuk (2002) and van Zanden (2003). On the other hand, our finding of a discontinuous jump in real wages between late Tokugawa and Meiji periods raises new doubts about the prevailing view of a stagnant real wage profile for this transition period. The rest of the paper is organized into four sections. Section 2 briefly discusses the results obtained in previous studies; Section 3 constructs the Japanese consumption baskets; Section 4 presents the main result of our international comparison; and Section 5 summarizes the main findings with a discussion.
2. PRICE AND WAGE SERIES DATA Research on wage and price series for Tokugawa Japan has been a thriving and productive enterprise in the last several decades.4 For the 17–19th centuries, there are several series of nominal wages (Umemura, 1961; Saito, 1978) summarized in a recent book and article by Saito (1998, 2005). One series is for the unskilled day laborers for the Kyoto and Tokyo area for 1741–1867, contained in the Mitsui company records (Mitsui Bunko, 1989). Another series, originally constructed by Sano (1962), is the craftsmen’s wages for the Kanto area for 1818–1894. Then there is the more comprehensive wage series compiled by the Long-Term Economic Statistics (LTES) project. This LTES series starts at 1880 and links to the present (Ohkawa et al., 1967). There are also available several consumer price indices and other deflators that have been used to derive real wage indices. Clearly, these are disparate real wage series differing in labor quality, definitions, and regions. The real wage series from Mitsui Bunko ends in the 1860s while the LTES series begins in 1885, leaving the 1868–1884 period – the period that saw Japan’s historically unprecedented economic and political revolution – in a statistical abyss.
232
JEAN-PASCAL BASSINO AND DEBIN MA
300
4 Saito Index
Williamson Index
Rice wages (right scale)
250 3 200 150
2
100 1 50 0 0 1730 1740 1750 1760 1770 1780 1790 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910
Fig. 1. Real Wage Indices and Rice Wages in Japan, 1741–1913. Notes: Saito Index and Williamson Index (1831 ¼ 100, left scale); rice wages in kilograms of rice per day (right scale). The missing data for the period of 1762–1790 in the Mitsui Bunko records could possibly be due to the loss of archival material. Nominal wages data, which are missing for 1881, 1888–1891, and 1893 in the LTES series, have been interpolated. Sources: Rice price and nominal wages in Kyoto up to 1871 (Mitsui Bunko (1989), Tables 6 and 7); in Tokyo from 1880 (LTES volume 8, Ohkawa et al. (1967), p. 245 for wages, p. 153 for rice).
To make up for this gap and arrive at a single continuous series for the Tokugawa–Meiji period, Williamson (1998) ‘‘patched’’ together the Sano and LTES series of real wages. In Fig. 1, we present this ‘‘patched’’ series – the Williamson Index – rescaled to fit alongside the Saito Index (Saito, 1998), which is a real wage index for unskilled laborers in Kyoto.5 The Williamson Index indicates overall stagnant Japanese real wages from the late-Tokugawa to the Meiji period, a trend also confirmed by Saito’s own reconstruction of the Sano’s series (Saito, 2005, Appendix Table 2B). Unfortunately, this often-quoted thesis of Tokugawa–Meiji stagnancy in real wages, despite its problematic methodology, has rarely been rigorously tested. In Fig. 1, we present the Mitsui Bunko and the LTES day laborers’ real wages in kilograms of rice (in the right scale), that is, nominal wages divided by the current period rice prices, alongside the Saito and Williamson Indices with the year 1741 equal to 100 for both series (see the left scale in Fig. 1). With rice having a disproportionate weight in consumption expenditure, it is not surprising that for the Tokugawa period, the rice wage index in Fig. 1 tracked fairly well the Saito Index which is deflated by a CPI where rice prices had a great share. The real discrepancy is for the Meiji period where
Japanese Real Wages
233
the level of the LTES-based rice wage rose consistently above that of the Mitsui Bunko-based rice wage for the Tokugawa period. The right scale of Fig. 1 shows that while a day’s work by an unskilled laborer could purchase about 2 kg of rice on average in the Tokugawa era, a day’s pay could buy him over 3 kg of rice in the 1880s and 1890s (both rice prices and nominal wages are presented in the appendix). Our finding of the jump in rice wages between the Tokugawa and Meiji periods sharply contradicts the stagnant real wage trend as displayed in the Williamson Index. The basis for a stagnant real wage trend in the Tokugawa–Meiji transition period stems from two questionable sources. First is the original craftsmen’s wage series compiled by Sano (1962), which as pointed out in Saito (2005), contains various data problems. Second and more fundamental is the problematic methodology of linking disparate real wages series at overlapping periods without a check on the actual levels of these wages in terms of real purchasing power. This, as we will show below, is a major problem in Allen’s Japanese real wage series, which uses backward projection from the Williamson Index.
3. CONSUMPTION BASKETS Until recently, studies of wages and prices for Tokugawa Japan have not seriously attempted direct level comparison across countries, obviously hampered by the observed disparities in dietary and consumption patterns between Japan and the West. Our approach here is first to construct consumption baskets at both ends that would satisfy an annual per capita requirement of 1,940 cal and 80 g of protein plus fixed quantities of linen and lamp oil. We then calculate the costs of these respective consumption baskets with price data for our chosen benchmark periods. Thus, the levels of real wages defined as the ratio of daily wages to the cost of a consumption basket (annual expenditures for one adult) serve as our yardstick for the levels of living standards in Japan and Europe (Allen, 2005). As the levels of real wages adopted here are really an index of a laborer’s ability to purchase a combination of calories and proteins as well as fixed quantities of linen and lamp oil, they are similar to the concept of ‘‘welfare ratio’’ used in Allen (2001).6 Table 1 reconstructs two consumption baskets A and B based on available consumption data for staple food in the early 18th, late 19th (Kito, 1986, 1989; Umemura, Takamatsu, & Itoh, 1983), and the first surveys for non-staple consumption of the early 20th century (Toyo Keizai, 1980).
234
Table 1.
JEAN-PASCAL BASSINO AND DEBIN MA
Composition of Consumption Baskets (per capita per year).
Bread (kg) Beans (excluding soybeans) (l) Meat (kg) Butter (or ghee) (kg) Soybeans (kg) Rice (kg) Barley and wheat (kg) Fish (kg) Buckwheat and others (kg) Edible oil (l) Linen (m) Lamp oil (l)
Japan A (this study)
Japan B (this study)
4
4
Europe Allen (2005) 208 52 26 10.4
52 114 10 3.5 16 1 5 2.6
26 30 70 75 1 5 2.6
India Allen (2005)
52 26 10.4 143
5 2.6
5 2.6
Note: Soybeans are used as a proxy for various processed products: soybean paste (miso), soy sauce (shoyu), tofu, and fermented soybeans (natto). Thus, our annual amount is higher than the actual consumption reported for the 1920s in surveys by the Ministry of Agriculture and Forestry (quoted in Toyo Keizai (1980), volume 2, p. 590). The diet of the inter-war period included meat and milk, and a larger volume of fish. The average per capita consumption for 1921–1925 was 18 kg beans (of which 12.9 kg soybeans), 4.1 kg meat, 2.1 l milk, 1.5 kg eggs, and 8.1 kg fish.
Consumption basket A, with its relatively higher quantities of rice, soybeans, and fish, aims to capture the expenditure of the ‘‘normal’’ or average income group, while basket B, with a relatively higher amount of barley, but lower quantities of rice and soybeans and zero fish, represents the subsistence consumption. In Table 1, we also present the European consumption basket used in Allen (2001, 2005) alongside our Japanese baskets. The differences in their respective diets are striking – major food items in the European diet, such as bread and meat were nearly non-existent in Japanese consumption. But what is more surprising is that the so-called Indian basket compiled and used by Allen for calculating Japanese real wages is far-removed from the actual Japanese consumption pattern as revealed by our Japanese baskets. His Indian basket allocated too much rice and an implausibly large quantity of meat to Japanese consumers while completely ignoring their substantial intake of soybeans and non-rice grains. In Table 2, we follow Allen and convert these food consumption baskets into comparable units of calories and proteins. It shows that, with their
235
Japanese Real Wages
Table 2.
Calories and Proteins for Food Consumption Baskets (per capita per year). Calories
Proteins (g)
Japan A Japan B Europe India Japan A Japan B Europe India Bread (kg) Beans (l) Meat (kg) Butter (kg) Soybeans (kg) Rice (kg) Barley and wheat (kg) Fish (kg) Buckwheat and others (kg) Edible oil (l)
12
12
558 1,096 92 10 148 24
279 288 646 692 24
Total
1,940
1,941
1,396 160 178 207
160 178
1,375
1
1
49 23 2 2 5
24 6 17 22
82
70
57 10 14
10 14
29
253 1,941 1,966
81
53
Note: For rice caloric content, we used 3,510 cal/kg reported by Mosk and Pak (1978) rather than 3,573 in Allen (2005). For protein content, we used the 75 g/kg, which is the average of 80 for brown rice and 70 for white rice given by Chang (2000) instead of 100 used in Allen (2005). Mosk and Pak (1978) used 65 but it seems to be for white rice. We used 2450 calories and 100g of protein for bread, 1125 cal and 71 grams for beans (other than soybeans), 2500 cal and 200g for meat, and 7268 cal and 7g for butter, as in Allen (2001, 2005). We used the figures of caloric and protein contents reported by Mosk and Pak (1978) for most other foodstuff: 1050 cal and 181g for fish, 3920 cal and 343g for soybeans, 8800 cal and 0g for edible oil, and 3370 cal and 88g for barley, (caloric and protein contents for barley are used as proxy for barley and wheat; for buckwheat, we used 3370 cal and 108g.
drastically different composition, Japanese baskets provided daily calories and proteins broadly equivalent to Allen’s European and Indian baskets. This is indicative of the long-standing tradition that rice and soybeans rather than meat intake had been the most important and cost-effective source of protein in the Japanese diet. As we show below, the validity of our international comparison of real wages hinges on the correct design and choice of consumption baskets.
4. THE INTERNATIONAL COMPARISON Our prices and nominal wages for the Tokugawa period mainly come from the published Mitsui company account books (Mitsui Bunko, 1989). We use the benchmark year 1750 (the 10-year average of 1745–1754) for Kyoto to
236
JEAN-PASCAL BASSINO AND DEBIN MA
match with Allen’s mid-18th century benchmark for Europe.7 As Allen (2005) had no Tokugawa prices and nominal wages but instead used the 1882 benchmark data from the LTES volume based on the India basket to project backward, we also construct a 1884 benchmark (the average of 1882–1886). After consistency checks, we decided to choose 1884 over 1882 to avoid the sharp price movement in the period of the so-called Matsukata deflation. We follow Allen (2005) to add 5 m of linen and 2.6 l of lamp oil in the Japanese budgets. Our price and wage data, all converted in grams of pure silver, are presented in detail in Table 3. The cost of consumption baskets is clearly sensitive to the basket construction. For example, the costs of Japan basket B, the subsistent level basket in Table 1 that supplied roughly equivalent amount of calories and proteins as that of Japan basket A, would be roughly 40 percent cheaper and would correspondingly give a higher Japanese real wage than from basket A.8 But this would also imply that consumers with the cheaper Japan basket B would have to settle for a less desirable diet with inferior crops and no fish. To be comparable to the European baskets which are defined here more for the average income level, we use consistently the Japan basket A. In this regard, our real wage index, given a certain choice of baskets, is also a welfare measure broader than a combination of calories and proteins required for subsistence. With the information on nominal wages and total costs of budgets, we present our real wages in the last row of Table 3. The comparison shows that despite their much lower nominal silver wages for the mid-18th century, Japanese real wages for unskilled laborers, due to the relatively cheaper cost of the consumption basket, are actually on a par with Milan and at about a third of the level of London. Furthermore, while nominal wages increased from 2.82 to 4.5 g of silver between 1750 and 1884, the total cost of their consumption basket actually declined from 326 to 277 g, leading to a near doubling of real wages for unskilled laborers between these two benchmark periods. The price decline between these two periods is most pronounced for lamp oil, linen, and soybeans, all of which were possibly associated with the opening of the country to international trade since the mid-19th century. Table 4 presents our Japanese real wages data in 50-year averages for 1750–1913 alongside those for London, Amsterdam, Strasbourg, Milan, and Madrid. Table 4 gives the annual series (our annual Japanese real wages for 1750–1913 are presented in full in the Appendix Table). Both Table 4 and Fig. 2 demonstrate unequivocally that real wages for Japanese unskilled laborers were clearly in the rank of those in Central and Southern European cities, but far below those in Northwestern Europe. It was only in the Meiji
237
Japanese Real Wages
Table 3.
Unit-Prices (in grams of Pure Silver) and Expenditure Shares (Percentage, in Parentheses) in Consumption Baskets.
Bread (kg) Beans (other than soybeans) (l) Meat (kg) Butter (kg)
England 1750–1759
Northern Italy 1750–1759
Japan (Kyoto) 1745–1754
Japan (Tokyo) 1882–1886
1.37 (57.3) 0.42 (4.4) 3.33 (17.4) 6.89 (14.4)
0.91 (53.6) 0.58 (8.5) 2.32 (17.1)
0.99 (1.2)
1.01 (1.5)
4.87 (4.9) 3.19 (1.7) 498 (100.0)
2.32 (6.8) 8.63 (12.2) 2.32 (1.7) 353 (100.0)
1.49 (23.8) 1.49 (52.1) 0.76 (2.3) 2.18 (2.3) 0.61 (3.0) 5.66 (1.7) 5.83 (8.9) 5.66 (4.5) 326 (100.0)
0.97 (18.2) 1.5 (61.9) 0.77 (2.8) 2.19 (2.8) 0.62 (3.6) 5.62 (2.0) 3.07 (5.6) 1.79 (1.7) 277 (100.0)
11.14
3.44
2.82
4.5
0.01
0.009
0.016
Soybeans (kg) Rice (kg) Barley and wheat (kg) Fish (kg) Buckwheat and others (kg) Edible oil (l) Linen (m) Lamp oil (l) Total cost of basket Nominal wages (grams of silver) Real wages
0.023
Source: Data for prices of rice, soybeans, beans, edible oil, lamp oil (kerosene), and linen in 1882–1886 are retail prices in Tokyo reported in LTES (volume 8, Ohkawa et al., 1967, pp. 153–156), converted to metric units by the ratios mentioned in the source; for barley, wheat, buckwheat, and other grains, for which retail price data are not available, we first calculate the ratio of the farm gate prices of barley, naked barley, and wheat relative to paddy for available years in 1874–1901; then we multiplied that ratio to the actual average retail rice price in 1882– 1886; we adopted the same procedure for buckwheat and others (buckwheat, foxtail millet, proso millet, barnyard millet, sorghum); agricultural price data are in LTES volume 8 (Ohkawa et al., 1967, pp. 168–169); we calculated the ratio of wholesale price of fish (salmon and cod) to wholesale price of rice in 1882–1886 and multiplied this ratio to the retail price of rice (wholesale price of fish and rice are from Historical Statistics of Japan, CD-ROM, Tables 17–16). All prices are converted into silver on the basis of 25 g of pure silver per yen as in Allen (2005).
238
JEAN-PASCAL BASSINO AND DEBIN MA
Data for prices of rice and lamp oil in 1745–1754 are based on Mitsui Bunko retail prices in Kyoto (Table 6). Prices of soybeans are available for Osaka in Mitsui Bunko from 1757 (Table 3). Unit-price is slightly higher than the price of white rice in Kyoto (Table 6); considering that the soybeans traded in Osaka were of above average quality, we just assume that the soybeans price in 1745–1754 to be equal to rice prices. Conversion ratio based on the LTES (volume 9, Umemura et al., 1966, p. 250) are: 180 l/koku, 150 kg/koku for rice, 129 kg/koku for soybeans, and 3.75 kg/ken. There are no retail prices for fish, barley and wheat, buckwheat and other grains, and linen for 1745–1754. We derived them using their price ratio to rice in early Meiji (beans 0.63, barley and wheat 0.51, buckwheat and other grains 0.41). For price of cotton linen, we used price in Osaka in 1840–1849 (Miyamoto, 1963, pp. 208–210) since raw cotton prices between 1746–1750 and 1846–1850 shown in Harada and Miyamoto (1985, p. 84) remained roughly constant. As lamp oil is rapeseed oil, we used the price of lamp oil as a proxy for edible oil. Rapeseed oil in Japan is comparable to olive oil in Southern Europe used both for lighting and cooking. It is neither a cheap nor a low-quality item, as few alternatives existed (sesame oil was more a condiment than an edible oil). All prices in monme are converted into silver grams assuming 3.11 g of pure silver per monme. For England and Northern Italy, we rely on calculation by Allen (2005). For Italy, 1 kg edible oil is assumed equivalent to 1 kg butter, in order to be consistent with Allen (2001, 2005). Caloric contents are only slightly different.
Table 4.
Real Wages of Unskilled Workers in Japan and Selected European Cities.
London Amsterdam Strasbourg Milan Madrid Kyoto (BM1750) Tokyo (BM 1884)
1750–1799
1800–1849
1850–1874
1875–1899
1900–1913
0.020 0.020 0.009 0.007 0.009 0.008
0.020 0.016 0.012 0.005 0.014 0.008
0.025 0.014 0.010 0.006 0.013 0.006
0.037 0.022 na 0.009 0.013
0.040 0.031a na 0.012 0.015
0.017b
0.020
Source: The Kyoto (BM 1750) series is based on the benchmark 1745–1754 (Table 3) and the Saito Index used in Fig. 1. The Tokyo (BM 1884) series is based on the benchmark 1882–1886 (Table 1) and extrapolated using nominal wages reported in LTES volume 8 (Ohkawa et al., 1967), deflated by using the consumer price index reported in Ohkawa and Shinohara (1979, p. 387), extended backward for the period 1882–1885 by relying on the early version of the LTES CPI (Ohkawa et al., 1967, p. 135). Series for the European cities are based on the series of welfare ratio constructed by Allen (2001), adjusted for taking into account differences in basket composition; the coefficient of adjustment for London, calculated on the basis of data reported by Allen (2001, 2005), is applied to other European cities. na ¼ not applicable. a 1900–1910. b 1882–1899.
239
Japanese Real Wages
London
Oxford
Milan
Kyoto
Tokyo
0.045 0.040 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000 1740 1750 1760 1770 1780 1790 1800 1810 1820 1830 1840 1850 1860 1870 1880 1890 1900 1910 1920
Fig. 2.
Real Wages in Kyoto and Tokyo in International Comparison. Source: Same as for Table 4.
period that Japanese real wages of unskilled laborers began to rise beyond that of Southern Europe.
5. DISCUSSION AND SUMMARY It is now possible to compare our real wages with Allen’s findings. As indicated earlier, Allen’s basket, with a much higher amount of rice and meat and no soybeans, yields a total cost of 540 g of silver based on the 1882 LTES prices, nearly twice the cost of our Japan basket A (see Allen, 2005, Appendix Table 5.3). Thus, his India-basket-deflated real wage for unskilled laborers for the 1882 benchmark, 0.009, is only half of our level for 1884 but happens to be identical to our 1750 benchmark-period level (see Table 3). As Allen used his 0.009 level of real wage to project backward to the Tokugawa period using the Saito Index with the relatively trend-less Williamson index serving as the intermediate link, his back-cast Tokugawa real wages ends up about the same as our 1750 benchmark-based Tokugawa real wages, which are about 0.008 on average for 1750–1874 in Table 3. This is a pure coincidence that should not distract us from the problematic nature of the backward projection method. Furthermore, Allen’s series missed the surge in real wages since the Meiji Restoration as revealed in our study. For our Tokugawa real wages, we did some robustness checks by comparing the nominal wages and prices used in this study with other comparable
240
JEAN-PASCAL BASSINO AND DEBIN MA
series in Saito (1998, 2005). We are reasonably certain that our data, particularly the rice price, used in this study are within the normal range of other Tokugawa data. Thus, the near doubling of our real wages between the Tokugawa and Meiji periods in our data is rather puzzling. Clearly, there are major differences in the objective and methods of statistical compilation between the Mitsui Bunko series and the LTES wages series. Furthermore, our simplified consumption baskets did not take into account such important items as housing rents, burning fuel, alcohol, etc., which might have changed in the Tokugawa–Meiji transition period. Most importantly, the Tokugawa– Meiji transition may have also spelled profound changes in labor institutions and labor contracts such as food allowances or side payments to cash wages. Whether or not this jump in real wages is a statistical illusion or an indication of real economic changes in the Tokugawa–Meiji transition period as argued, for example, by Huber (1971), is an issue for future research. With these caveats in mind, we provide summary information of unskilled Japanese and U.K. laborers’ wages and income for the two benchmark periods in Table 5. The table shows that our wages expressed in silver and grain units are broadly consistent with other independent studies, particularly the recent preliminary work by Peter Lindert et al. (2004) on global price and wage comparisons. Table 5 also reveals that the Kyoto–Tokyo grain wages relative to those in London are consistently higher than their Table 5. A Summary Comparison of Japanese and U.K. Wages and Incomes (Numbers in Parentheses are Ratios over London with London ¼ 100).
Benchmark London Kyoto– Tokyo
Silver Wages (g)
Grain Wages (in kilograms of bread)
1,750 11.14 (100) 2.8 (25)
1,750 8.1 (100) 2.74 (34)
1,884 34.36 (100) 4.5 (13)
1,884 15.5 (100) 3.95 (26)
Real Wages
1,750 0.024 (100) 0.009 (38)
1,884 0.036 (100) 0.016 (44)
GDP per capita (in 1990 International dollars) 1,700 1,250 (100) 570 (46)
1,884 3,622 (100) 836 (23)
Note: Wage measured in kilograms of bread in London and Milan, the 5-year average of wage divided by the 5-year average prices of rice in kilograms in Kyoto–Tokyo. On the basis of the caloric content, adjusted for taking into account the cost of fuel for cooking rice, 1 kg of rice is regarded equivalent to 1.3 kg of bread (the ratio of the caloric contents is 1.43). GDP per capita data for Britain, Italy, and Japan are from Maddison (2001, pp. 206, 264) and Maddison (2004, pp. 60–61). Real wages are averages of 1745–1754 and 1882–1886, respectively.
241
Japanese Real Wages
relative silver wages. Furthermore, the Kyoto–Tokyo real wages relative to those of London are higher than both the relative grain and silver wages. This is consistent with the theme of the recent article by Broadberry and Gupta (2003), which viewed this as an indication of higher relative U.K. productivity in the tradable goods. In Table 5, we also present Angus Maddison’s per capita income estimates (or ‘‘guesstimates’’) for the 18th century.
6. SUMMARY To sum up, this article is a first attempt to make international comparison of 18th and 19th century Japanese real wages using current price benchmarks (rather than Meiji period back-projection). Our finding of a purchasing power of unskilled Japanese laborers being around a third of the London level seems to place Japan in the ranks with Turkey, Java, or Southern Europe in the 18th century. This finding, subject to further tests, would have important implications both for the old debate on Japanese living standards and new understandings of the initial conditions for economic take off. Since Japan is part of the East Asian civilization, we believe our study of Japanese real wages is also a contribution toward understanding the debate over the ‘‘great divergence’’ raised by Kenneth Pomeranz’s (2000) recent work on China–Europe comparison and marks a step forward toward a more rigorous comparison of global living standards for the 18th and 19th centuries. Our study also reveals an important discontinuity in real wage purchasing power between the late Tokugawa and early Meiji. This finding is a clear warning against backward projection methods based on the Meiji benchmark. While further careful research is needed to understand the cause of this discontinuity, it does serve to draw our attention to re-evaluate the economic impact of the transition from Tokugawa to Meiji.
NOTES 1. See Susan Hanley (1983, 1997) for the optimistic assessment of a high living standard in 19th century Japan. For Yasuba’s rebuttal and his own assessment, see Yasuba (1986, 1987). 2. For real wage growth, see Saito (2005). For real per capita income growth, see Maddison (2001, p. 255). On page 264, Maddison gave 1870 Japanese and British per capita incomes at 737 and 3,191 respectively, in 1990 international dollars. This is
242
JEAN-PASCAL BASSINO AND DEBIN MA
fairly drastic upward adjustment in comparison with previous exchange rate-based estimates, which would give Japanese per capita income around the 1860s and 1870s at only 10 percent of the contemporaneous British level. 3. The few quantitative GDP studies such as Maddison (2001) and Yasuba (1987) that did extend to international comparison were largely back-of-envelope type of calculations. 4. Comprehensive summary and discussion of studies on Japanese prices and wages can be found in Harada and Miyamoto (1987) and Saito (1998). 5. Real wage series calculated by using nominal wage data reported in Mitsui Bunko, deflated using the consumer price index calculated by Shimbo (1978), which is based on price data reported in the same source. 6. In Allen (2001), the welfare ratio is calculated as the ratio of annual income of a representative household of unskilled worker (daily nominal wage multiplied by 250, assuming 250 working days and no other income) to the minimal expenditures for one household measured as three times the cost of the consumption basket per adult (assuming five persons per household, or the equivalent of three adults in terms of consumption). The real wage level, as defined in Allen (2005) is therefore equivalent to a welfare ratio divided by 750 (250 3); however, it should be noted that the consumption basket used in Allen (2001) also includes fuel and soap. 7. In the Mitsui Bunko records, there is also a nominal wage series for Tokyo (Edo) for 1818–1871. Although with a different trend, the levels for the Edo series are broadly equal to the Kyoto series. Here we use the Kyoto series only for the Tokugawa period. 8. Costs of basket A for 1750 and 1884 benchmark periods would be 236 and 200 g of silver respectively. This would give levels of Japanese real wages at 0.012 and 0.023 respectively, higher than the 0.09 and 0.012 deflated by the cost of Japan basket A.
ACKNOWLEDGMENTS We want to thank encouragement and suggestions from Gregory Clark and an anonymous referee from this Journal as well as from Robert Allen, Steve Broadberry, Osamu Saito, Jeffrey Williamson, Jan Luiten van Zanden, and the participants of the conference ‘‘Towards a Global History of Prices and Wages’’ held in Utrecht (the Netherlands) in August 19–21, 2004. Our thanks also go to Kyoji Fukao, who helped us with access to library and research facilities at the Institute of Economic Research at Hitotsubashi University in Japan. Research scholarship from the research scholarship of Maison Franco-Japonaise, in Tokyo (French Ministry of Foreign Affairs) for Jean-Pascal Bassino, and funding from the Foundation for Advanced Studies on International Development (FASID) for Debin Ma are also gratefully acknowledged. We remain solely responsible for the errors.
243
Japanese Real Wages
REFERENCES Allen, R. C. (2001). The great divergence in European wages and prices from the Middle Ages to the First World War. Explorations in Economic History, 38, 411–447. Allen, R. C. (2005). Real wages in Europe and Asia: A first look at the long-term patterns. In: R. C. Allen, T. Bengtsson & M. Dribe (Eds), Living standards in the past: New perspectives on well-being in Asia and Europe. Oxford: Oxford University Press. Broadberry, S., & Gupta, B. (2003). The early modern great divergence: Wages, prices and economic development in Europe and Asia, 1500–1800. Working Papers, Department of Economics, University of Warwick. Chang, T.-T. (2000). Rice. In: K. F. Kiple & K. C. Ornelas (Eds), The Cambridge world history of food, (Vol. 1, pp. 132–149). Cambridge: Cambridge University Press. Hanley, S. B. (1983). A high standard of living in nineteenth-century Japan: Fact or fantasy? Journal of Economic History, 43, 183–192. Hanley, S. B. (1997). Everyday things in pre-modern Japan: The hidden legacy of material culture. Berkeley: University of California Press. Harada, T., & Miyamoto, M. (Eds) (1985). Reikishi no Nakano Bukka (Prices in history: A symposium). Tokyo: Toubensha. Huber, J. R. (1971). Effect on prices of Japan’s entry into world commerce after 1858. Journal of Political Economy, 79, 614–628. Kito, H. (1986). Meiji zenki no shushokuko zeito sono chiiki pattern (Food composition and its regional pattern in early Meiji). Sophia Economic Papers, 31, 30–43. Kito, H. (1989). KinseiNihon no shusho kutaikei to jino henka (Staple food and population change in early modern Japan). In: A. Hayami, O. Saito & S. Sugiyama (Eds), Tokugawa shakai kara no teibo; hatten, kozo, kokusai kankei (A perspective from the Tokugawa society: Development structure and international relations) (pp. 32–55). Tokyo: Toabunkan. Lindert, P., Allen, R. C., Devereux, J., Hellie, R., Hoffman, P. T., Jacks, D. S., Ma, D., Mironov, B. N., Pamuk, S., Van Zanden, J. L., & Ward, M. (2004). Preliminary global price comparisons, 1500–1870. At http://www.iisg.nl/hpw/conference.html Maddison, A. (2001). The world economy: A millennium perspective. Paris: OECD Development Centre. Maddison, A. (2004). The world economy: Historical statistics. Paris: OECD. Mitsui Bunko (Ed.) (1989). Kinsei Nihon goki ni okeru juyo bukka no dotai (Trends of major prices in early modern Japan). Tokyo: University of Tokyo Press. Miyamoto, M. (Ed.) (1963). Kinsei Osaka no bukka to rishi (Prices and interest rates in Osaka during the pre-Meiji period). Tokyo: Sonbun sha. Mosk, C., & Pak, S. (1978). Food consumption, physical characteristics, and population growth in Japan, 1874–1940. Working Paper no. 102, Department of Economics, University of California at Berkeley. Ohkawa, K., & Shinohara, M. (Eds) (1979). Patterns of Japanese economic development, a quantitative appraisal. New Haven: Yale University Press. Ohkawa, K., Noda, T., Takamatsu, N., Yamada, S., Kumazaki, M., Shionoya, Y., & Minami, R. (1967). Bukka (Prices), Nihon Choki Keizai Tokei (Estimates of Long-Term Economic Statistics (LTES) of Japan since 1868). (Vol. 6). Tokyo: Toyo Keizai. Ozmucur, S., & Pamuk, S. (2002). Real wages and standards of living in the Ottoman empire, 1489–1914. Journal of Economic History, 62, 293–321.
244
JEAN-PASCAL BASSINO AND DEBIN MA
Pomeranz, K. (2000). The great divergence: China, Europe, and the making of modern world economy. Princeton: Princeton University Press. Saito, O. (1978). The labor market in Tokugawa Japan: Wage differentials and the real wage level, 1727–1830. Explorations in Economic History, 15, 84–100. Saito, O. (1998). Chingin to rodo to seikatsu suijun: Nihon keizaishi ni okeru 18–20 seiki, (Wages, labor, and living standards in Japan from the 18th to the 20th century). Tokyo: Iwanami Shoten. Saito, O. (2005). Wages, inequality and pre-modern growth in Japan, 1727–1894. In: R. C. Allen, T. Bengtsson & M. Dribe (Eds), Living standards in the past: New perspectives on well-being in Asia and Europe. Oxford: Oxford University Press. Sano, Y. (1962). Kenchiku roˆdoˆsha no jisshitsu chingin: 1830–1894 nen (Real wages of construction workers). Mita gakkai zasshi, 55, 1009–1036. Shimbo, H. (1978). Kinsei no bukka to keizai hatten (Early modern prices and economic development). Tokyo: Toyo Keizai Shimposha. Toyo Keizai (1980). Showa Kokusei Soran (Retrospective statistics of the Showa era). Tokyo: Toyo Keizai Shinposha. Umemura, M. (1961). Kenchiku roˆdoˆsha no jisshitsu chingin 1726–1958 nen (Real wages of construction workers, 1726–1958). Economic Research-Keizai Kenkyu, 12, 172–176. Umemura, M., Takamatsu, N., & Itoh, S. (1983). Chiiki Keizai Tokei (Regional economic statistics), Choki Keizai Tokei (Estimates of Long Term Economic Statistics (LTES) of Japan since 1968) (Vol. 13), Tokyo: Toyo Keizai. Umemura, M., Yamada, S., Hayami, Y., Takamatsu, N., & Kumazaki, M. (1966). Nogyo (Agriculture), Choki Keizai Tokei (Estimates of Long Term Economic Statics (LTES) of Japan since 1968) (vol. 9). Tokyo: Toyo Keizai. Van Zanden, J. L. (2003). Rich and poor before the Industrial Revolution, a comparison between Java and the Netherlands at the beginning of the 19th century. Explorations in Economic History, 40, 1–23. Williamson, J.G. (1998). Real wages and relative factor prices in the Third World 1820–1940: Asia. Harvard Institute of Economic Research Discussion Paper no. 1844. Yasuba, Y. (1986). Standards of living in Japan before industrialization: From what level did Japan begin? A comment. Journal of Economic History, 46, 217–224. Yasuba, Y. (1987). The Tokugawa legacy: A survey. Economic Studies Quarterly, 38, 290–308.
Table A.1.
Daily Nominal Wages and Rice Prices in grams of Silver for 1741–1871 and 1880–1913.
Year Daily Wages White Rice (per kg) Year Daily Wages White Rice (per kg) Year Daily Wages White Rice (per kg) 2.57 2.81 2.58 2.72 2.60 2.66 2.47 2.89 2.80 3.58 3.50 2.57 2.50 2.63 2.49 2.49 2.55 2.52 2.53 2.49 2.49 2.49
1.88 1.73 1.67 1.66 1.58 1.80 1.68 1.69 1.52 1.60 1.55 1.33 1.11 1.04 1.52 1.94 1.69 1.55 1.52 1.33 1.20 1.15
1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836
3.22 3.28 2.81 2.71 2.75 2.74 2.77 2.85 2.97 4.17 3.31 3.14 3.42 3.37 3.97 3.02 2.83 2.80 3.00 2.88 3.03 2.92 3.11
1.68 1.69 1.57 1.72 1.47 1.28 1.17 1.31 1.46 1.46 1.55 1.63 1.82 1.51 1.68 2.15 1.86 2.01 1.71 2.16 2.50 1.84 2.50
1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890
3.83 4.20 3.70 3.69 4.90 4.23 4.11 4.26 4.56 5.88 4.87 5.66 5.25 5.38 5.50 4.75 4.50 4.00 3.75 4.00 4.10 4.20 4.30
3.04 4.58 3.29 3.68 4.67 8.32 19.36 18.49 9.51 13.53 13.29 7.75 1.99 2.01 1.67 1.24 1.41 1.76 1.44 1.29 1.28 1.45 2.24
245
1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
Japanese Real Wages
APPENDIX. DATA ON NOMINAL WAGES AND RICE PRICES ARE GIVEN IN TABLE A.1 AND REAL WAGE SERIES FOR KYOTO-EDO IN TABLE A.2.
246
Table A.1
(Continued)
Year Daily Wages White Rice (per kg) Year Daily Wages White Rice (per kg) Year Daily Wages White Rice (per kg) 2.55 2.86 2.80 2.57 2.75 2.52 2.55 2.63 2.57 2.74 2.57 2.52 2.49 3.37 2.36 1.93 1.90 2.13 2.75 2.74 2.75 2.75 2.91
1.47 1.91 1.89 1.52 1.65 1.82 1.77 1.77 1.68 1.90 1.77 1.64 1.52 1.25 1.26 1.37 1.45 1.72 1.72 1.37 1.43 1.36 1.40
1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859
3.02 2.95 2.97 3.08 3.45 3.36 2.80 3.59 3.61 3.30 3.16 2.81 2.99 3.00 3.37 3.00 2.91 3.14 3.48 3.34 3.33 3.37 3.45
3.70 2.51 2.31 1.71 1.65 1.79 1.70 1.73 1.91 2.10 1.99 2.07 2.15 2.76 2.77 2.00 2.22 2.30 1.88 1.87 2.22 2.91 2.91
1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913
4.40 4.50 4.88 5.25 5.50 6.50 7.25 8.25 8.50 9.25 9.75 9.75 10.00 10.00 10.25 10.50 12.25 13.25 13.00 13.25 14.00 14.50 14,75
1.82 1.91 1.97 2.01 2.03 2.18 2.72 3.29 2.21 2.75 2.82 2.89 3.31 3.11 3.01 3.47 3.79 3.51 2.98 2.89 3.82 4.48 4,82
Note: conversion units used: 1 silver monme ¼ 3.11 g of pure silver; 1 koku of rice ¼ 150 kg; 1 yen ¼ 25 g of pure silver; and 1 sho rice ¼ 1.425 kg. Sources: For 1741–1871, based on Mitsui (1989); for 1880–1913, based on Table 6 and LTES (Ohkawa, 1967, Table 5, p. 153; Table 25, column 30, p. 245; missing data for nominal wages interpolated for 1881, 1888–1893, and 1893).
JEAN-PASCAL BASSINO AND DEBIN MA
1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813
Real Wage Series for Kyoto-Edo (1741–1869) and Tokyo (1882–1913).
Real Wage
Year
Real Wage
Year
Real Wage
Year
Real Wage
1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766
0.006 0.007 0.007 0.007 0.007 0.008 0.007 0.008 0.008 0.011 0.012 0.008 0.009 0.009 0.007 0.006 0.007 0.007 0.007 0.008
1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816
0.008 0.007 0.008 0.008 0.008 0.007 0.007 0.008 0.008 0.008 0.007 0.008 0.007 0.012 0.008 0.006 0.006 0.006 0.008 0.009 0.009 0.009 0.010 0.011 0.009 0.009
1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866
0.010 0.008 0.010 0.009 0.008 0.008 0.008 0.009 0.008 0.007 0.008 0.008 0.008 0.007 0.008 0.007 0.008 0.007 0.008 0.008 0.006 0.007 0.006 0.007 0.006 0.004
1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913
0.016 0.016 0.017 0.018 0.017 0.018 0.017 0.018 0.019 0.019 0.020 0.020 0.019 0.019 0.018 0.018 0.019 0.021 0.021 0.021 0.020 0.019 0.019
247
Year
Japanese Real Wages
Table A.2.
Year
(Continued)
Real Wage
Year
Real Wage
Year
Real Wage
0.008 0.007 0.007 0.007 0.007 0.007 0.007 0.008 0.009 0.007 0.007 0.006 0.006 0.007 0.005 0.006 0.006 0.007
1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840
0.008 0.008 0.009 0.010 0.009 0.009 0.012 0.009 0.009 0.009 0.010 0.012 0.008 0.008 0.008 0.009 0.008 0.007 0.008 0.007 0.005 0.006 0.006 0.007
1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890
0.004 0.006 0.003 0.004 0.005 0.007 0.008 0.007
0.017 0.017 0.017 0.015 0.015 0.017 0.017 0.017 0.015
Year
Real Wage
JEAN-PASCAL BASSINO AND DEBIN MA
1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790
248
Table A.2
RELATIVE BRITISH AND AMERICAN INCOME LEVELS DURING THE FIRST INDUSTRIAL REVOLUTION Marianne Ward and John Devereux ABSTRACT We provide new measures of relative UK and US GDP per capita and output per worker for the crucial years between 1830 and 1870. Our estimates are current price comparisons that compare expenditure on GDP for five benchmark years using new price data. They show that the US leads in income per capita and output per worker compared to Great Britain and the United Kingdom. We check our estimates against sectoral productivity data and real wages.
1. INTRODUCTION The long debates surrounding nineteenth-century living standards, growth and structural change during the early and middle nineteenth century for the US and the UK highlight the importance of this period in each nation’s development. Our knowledge of these economies during these years comes largely from information on changes within the domestic economy. Research in Economic History, Volume 23, 249–286 Copyright r 2005 by Elsevier Ltd. All rights of reproduction in any form reserved ISSN: 0363-3268/doi:10.1016/S0363-3268(05)23008-2
249
250
MARIANNE WARD AND JOHN DEVEREUX
We know little about relative UK/US or British/US income and productivity before 1870. To fill this gap, we offer a comparison of income per capita and output per worker for the UK and Great Britain relative to the US in five benchmark years between 1830 and 1870. Our estimates cover a crucial period for both the countries and they provide a new perspective on their relative economic performance. All international GDP comparisons wrestle with the fact that exchange rates do not reflect relative purchasing power. Researchers get around this problem with projections. This works as follows. First, they establish a relative GDP benchmark for a recent year using a purchasing power parity (PPP) adjusted comparison. Then, they project the benchmark backwards in time using domestic growth rates. As Gallman (1966) pointed out 40 years ago, the projection approach suffers from a fundamental index number problem, since it compares income across economies for long periods using the relative prices of the recent year. Maddison (1995, 2001, 2003), for example, compares income over two centuries with 1990 prices. The index number problem arises because the goods and services produced in 1990 bear little resemblance to those of, say, 1820. The second problem faced by projections is that growth rates are not always comparable. This is especially problematic for long periods where small differences in growth rates cumulate into large differences in projected income levels, and where the quality of the data falls as we move back in time. In contrast to projections, our estimates rely on new price level benchmarks to directly compare UK/US GDP for five benchmark years between 1830 and 1870. These estimates are in current international prices. The advantage of this approach is that it circumvents the difficult index number problems faced by long-run projections. Since we compare income across space at a point in time, we can be sure that the UK and the US produce similar goods and services.1 We proceed as follows: Section 2 introduces our estimates of UK/US income per capita and output per worker. The results show the US with a small edge in income per capita and a larger lead for output per worker. Section 3 examines the price benchmarks underlying our GDP comparisons. Section 4 compares our benchmarks against alternative approaches. First, we compare UK/US GDP with sectoral data for 1860 and 1870. Then, we look at the implications of higher US real wages for UK/US GDP estimates. Section 5 explains differences between our results and Maddison’s (1995, 2001, 2003) GDP projections. Finally, Section 6 steps back to consider the course of relative US/UK income over the very long run by comparing US/
251
Relative Income
UK living standards and labor productivity from 1830 to 1990. Section 7 concludes with directions for further research.
2. RELATIVE UK/US GDP PER CAPITA AND OUTPUT PER WORKER This section provides our benchmark comparisons of income per capita and output per worker. We provide estimates for the UK and Great Britain. Both of these comparisons are important. Given Irish poverty and its large share in UK population before the famine, it could be argued that the more interesting comparison for questions of industrialization is between the US and Great Britain. On the other hand, the UK comparison is central with regard to living standards. We compare expenditure on GDP using as far as possible the OEEC methodology developed by Gilbert and Kravis (1954) in the early 1950s.2 Eq. (1) gives the benchmark estimates of GDP per capita at time t, yt, in current international prices where Yt is the ratio of UK/US nominal GDP per capita in US dollars and pt is the UK/US price level calculated as a Fisher Ideal index using both US and UK weights.3 yt ¼ Y t =pt
(1)
There are four points to note about the Gilbert and Kravis approach. First, the price benchmarks in (1) compare the prices of items of identical quality across space at a point in time. They should not be confused with domestic price indices, such as the CPI or the GDP deflator, which compare identical items across time within an economy. For interspatial comparisons, expenditure weights as well as the items compared change for each benchmark. In other words, the interspatial price data cannot compare price levels across time within either economy. Second, by construction, benchmark comparisons will not equal projections for either prices or quantities. As discussed, the benchmarks and domestic price indices differ in weights and good sampled. It follows that using US and UK domestic price indices to project a UK/US price benchmark to another benchmark year, will not produce a result that is equal to the interspatial price benchmark for that year. This point is fundamental to all international comparisons.4 For similar reasons, we should not expect agreement between GDP projections and GDP benchmark comparisons. Even for recent years, differences between GDP benchmarks and projections are often pronounced.5
252
MARIANNE WARD AND JOHN DEVEREUX
The third point is that the Gilbert and Kravis (1954) approach takes nominal income as given. To compare UK/US GDP, we use the interspatial price benchmarks to deflate nominal GDP obtained from standard domestic sources. All subsequent international expenditure comparisons follow this procedure. Fourth, the Gilbert and Kravis (1954) approach provides checks on the plausibility of GDP comparisons. The explicit recognition by (1) that UK/ US real GDP is the ratio of UK/US nominal GDP and the UK/US price level allows us to examine the consistency of real income estimates against relative nominal income and relative price levels. As shown in later sections, sectoral GDP comparisons provide a further powerful check on the expenditure-based benchmarks. We provide the GDP per capita comparisons in Table 2. The price level benchmarks cover 1831, 1839, 1849, 1859 and 1869. These years reflect data availability. In a later section, we outline sources and methods in more detail. For now, we concentrate on the results.6 The second column is the ratio of nominal income per capita calculated with market exchange rates. From Table 1, UK nominal income per capita for 1831 exceeds that of the US. For the other years, it is approximately equal, with the exception of 1859 where the US is ahead. As we would expect, Great Britain has higher
Table 1. Year
Per Capita Income (US ¼ 100).
Nominal GDP Per Capita
Price Level
Real GDP Per Capita
124 106 102 92 105
164 134 124 112 105
76 79 82 83 100
151 127 114 107 118
164 134 124 112 105
92 94 92 97 112
United Kingdom 1831 1839 1849 1859 1869 Great Britain 1831 1839 1849 1859 1869
Notes and Sources: Real GDP per capita is the ratio of relative nominal GDP per capita at market prices and the UK/US price level. See Appendix for sources.
253
Relative Income
nominal income. For all years, it exceeds that of the US. As the share of Irish population falls, the UK and British measures converge. The third column gives our UK/US price level benchmarks as derived from disaggregated price and expenditure data. The benchmarks cover consumption, investment and government spending. Two features of UK/US price levels stand out. First, the UK price level is higher for all years.7 Second, it falls over time. It starts in 1831 at 64 percent above US levels. By 1869, the margin is down to 5 percent. As shown later, this partly reflects a decline in relative British food prices due to greater integration of world markets. The final column in Table 2 is our real UK/US GDP per capita. It shows that US income per capita exceeds that of the UK for all years except 1869 when per capita income are equal. Thus, the benchmark comparisons show that US leadership in terms of living standards begins early. As expected, the US lead over Great Britain is smaller. It is less than 10 percent for all years and Britain leads in income per capita for 1869.8 The results for output per worker are in Table 2. The UK/US comparison is in the top panel, with the bottom panel showing the GB/US comparison. The second column is UK/US nominal output per worker. The US shows higher nominal output per worker throughout. The difference with income per capita reflects the substantially lower US labor force participation rates.
Table 2. Year
Output Per Worker (US ¼ 100).
Nominal GDP Per Worker
Price Level
Real GDP Per Worker
95 97 81 73 75
164 134 124 112 105
58 72 66 66 72
115 116 90 87 85
164 134 124 112 105
70 86 73 78 81
United Kingdom 1831 1839 1849 1859 1869 Great Britain 1831 1839 1849 1859 1869
Notes and Sources: Real GDP per worker is the ratio of UK/US nominal GDP per worker and UK/US price levels from Table 1. See Appendix for nominal GDP per worker sources.
254
MARIANNE WARD AND JOHN DEVEREUX
The fourth column is our estimate of real UK/US output per worker. It shows the US with a large lead over the UK with UK output per worker around two-thirds of US levels for most years. The US lead over Great Britain is smaller but still substantial. The UK/US GDP benchmarks in Table 1 and 2 have striking implications for important debates. They suggest that US leadership begins long before the early 1900s as suggested by Maddison (1995, 2001, 2003). Before discussing these matters, we will take a closer look at our benchmarks. We will then provide crosschecks using data on sectoral productivity and real wage rates.
3. THE PRICE BENCHMARKS To this point, we have provided few details about the benchmarks. We now turn to this task starting with prices and then moving on to expenditure patterns. In each case, we provide a short description. A complete account is in the appendix. 3.1. Prices Table 3 shows UK/US relative prices for the benchmark years. With the exception of food, we give aggregate results. For the most part, we rely on retail prices. For consumption, the Weeks Report (US Congress, 1886) provides comprehensive coverage of US retail prices for 1859 and 1869. These data are derived from retail records for a broad cross-section of American towns and cities (see Hoover, 1960). For other years, we rely on retail price data for Massachusetts from the pioneering study of Carroll D. Wright (Massachusetts Bureau of Statistics of Labor, 1885).9 We supplement these sources with Adams (1986, 1992) for Maryland and West Virginia as well as many other studies.10 The British sources are scattered and incomplete in comparison with the American data. Ashton (1949), Neale (1966) and Brassey (1873) provide retail price series for particular localities obtained from newspaper quotations or, in the case of Brassey, one retail establishment. We supplement these sources with retail prices from Caird (1878), Burnett (1966), Chadwick (1860), Dodd (1951), Edmonds (1839), MacKenzie (1921), Mitchell (1971, 1988), Mulhall (1898), Neild (1842), Porter (1850a) and Purdy (1861). These data are often for either a single year or a small number of years. They are from direct enquiries by knowledgeable observers, budget accounts or other sources.
255
Relative Income
Table 3.
Relative UK/US Prices, 1831–1869 (US ¼ 100). 1831
1839
1849
1859
1869
Food
239
191
164
144
129
Bread Wheat flour Potatoes Beef Mutton Pork Bacon Milk Butter Cheese Tea Coffee Sugar Drinks Tobacco Housing Fuel Light Soap Clothing Domestic service Transportation
174 235 348 260 312 212 264 121 210 269 280 342 212 168 718 91 319 103 142 50 58 90
165 218 185 175 275 128 166 110 104 155 260 366 203 146 460 99 179 84 139 50 70 92
93 166 85 177 319 168 233 105 121 167 249 306 160 134 385 112 161 92 154 60 76 105
95 132 111 170 187 146 187 101 140 135 159 214 132 115 350 98 124 89 84 60 83 110
93 125 116 164 160 127 179 110 127 123 88 156 88 121 486 96 95 83 106 65 67 104
Consumption Investment Government Overall
182 46 71 164
148 54 68 134
137 50 64 124
120 69 67 112
117 52 64 105
Notes and Sources: See Appendix.
For the most part, we used price data for the qualities likely to be consumed by the working class. We checked our price data against a wide variety of institutional sources used by Feinstein (1996, 1998) as well as standard wholesale series. We were also fortunate to have access to highquality unpublished data for clothing and other items furnished by Gregory Clark of the University of California at Davis. In general, there is fair agreement between sources. We should emphasize, however, that the relative scarcity of retail price data for the UK means that judgments are necessary in many cases. We will return to this issue in our concluding comments.
256
MARIANNE WARD AND JOHN DEVEREUX
These price data refer to urban areas. To compare GDP, we need national prices covering both urban and rural areas. To go from urban to national prices, we adjust food and rent prices to reflect lower prices in rural areas. For the UK, we assume that urban rents are twice rural rents based on Hunt (1973). We find little evidence to suggest differences between UK urban and rural food prices, so we make no adjustment in this case. Our US food price adjustments are from Brady (1972) and Hatton and Williamson (1991). We estimate the US urban-rural food price spread is 23 percent for 1831–1849 and 15 percent for 1859 and 1869. For rent, we use Hatton and Williamson’s (1991) estimate of an urban-rural rent spread of 113 percent for Michigan in 1891.11 National prices are a weighted average of urban and rural prices, with urbanization rates taken from Bairoch and Goertz (1986). Turning to the price level benchmarks, Table 3 shows food prices, which are consistently higher in the UK. This is plausible given the high transport costs of the period. The largest price gaps for food are early on suggesting improved market integration as the nineteenth century proceeds. UK food prices are twice American levels in 1831. Forty years later, the gap is just 30 percent.12 Moving to other items of expenditure, UK tobacco prices are consistently higher. This is due to taxes.13 Alcohol prices are also higher. While beer prices are below US levels, spirits are more expensive.14 Again, the culprit is taxes. We proxy fuel costs with coal for the UK and coal and firewood for the US. We transform wood to a coal equivalent by comparing the BTUs (British Thermal Units) of a cord of wood and a ton of coal.15 Wood is cheaper in the US while coal is more expensive. Overall, fuel prices are higher in the UK. For clothing, our comparisons show lower UK prices. This is due to US protection. Our estimates likely overstate the British advantage for clothing as they exaggerate the protection afforded to American cottons.16 From Table 3, we see higher US rents.17 Remember our estimates are for the overall economy. Our data show that US urban rents are well above UK urban levels.18 We have two service items, transportation and domestic services. Before 1849, we base transportation costs on coach rates from Hawke (1970) and Fishlow (1965, 1966). For 1859 and 1869, our estimates refer to train passenger rail rates per mile. Travel costs are similar for both countries.19 We compare costs of domestic service with wage rates including board using Lebergott (1964) for the US and Feinstein (1996) for Great Britain. As expected, US wages for domestics are higher. For investment, we use bilateral Fisher indices for construction plus equipment and machinery. We estimate the construction benchmark with
257
Relative Income
material prices and construction wages. The equipment and machinery benchmarks use wholesale price data for iron. The results show that investment prices are lower in the UK, reflecting lower wages and iron prices. For government, we follow Gilbert and Kravis (1954) in constructing UK/US relative prices for civilian and defense expenditures. The civilian price index is a weighted average of relative nominal wages and the geometric mean of rent, fuel and light prices from the consumption index. The defense price index is a weighted average of relative nominal wages and the geometric mean of the construction, equipment and machinery, fuel and clothing relative prices. Finally, Table 3 shows that overall UK price levels fall relative to the US between 1831 and 1869. The explanation for this lies with both relative price changes and expenditure weights. We therefore, postpone the discussion of overall price changes until after we present the expenditure weights.
3.2. Weights We faced many difficulties in finding appropriate expenditure weights. For the UK, we took overall consumption, investment and government weights from Crafts (1985) and Mitchell and Deane (1962). For the US, we take shares from Gallman (2000) and Trescott (1960). The biggest obstacles lie in finding weights for individual commodities and services. Ideally, these weights should apply to years compared and they should cover overall expenditure. We were unable to meet these requirements. In general, there is more information for Britain. This is because of greater British interest in poverty. Starting from Eden (1797), each economic downturn spawned a host of budget studies for poor and working class consumers.20 For food and other weights, we relied on Horrell (1996) supplemented by Feinstein (1998) and Thomas (1995). Clearly, these are British rather than UK weights. The biggest drawback of our expenditure data is that it does not refer directly to benchmark years. For example, our weights for 1849–1869 are largely based on data collected for much later years. We return to these problems and their implications for our estimates shortly. There are no systematic investigations of US consumption patterns before the 1870s. Brady (1972) provides individual weights for 1830. We take our US consumption item weights from David and Solar’s (1977) revision of Brady (1972), Kuznets (1966) and Lebergott (1996). As is the case with the UK, the expenditure data do not relate directly to our benchmark years.
258
MARIANNE WARD AND JOHN DEVEREUX
Table 4.
US and British Consumption Weights, 1831–1869. 1831
1839
1849
1859
1869
Food Alcohol Tobacco Housing Fuel and light Light Soap Clothing Domestic service Transportation Total consumption Investment Government
0.362 0.070 0.020 0.135 0.062 0.022 0.009 0.157 0.031 0.049 0.918 0.070 0.013
0.350 0.068 0.020 0.130 0.060 0.021 0.008 0.152 0.030 0.048 0.887 0.100 0.013
0.317 0.078 0.028 0.137 0.029 0.010 0.010 0.138 0.034 0.043 0.825 0.160 0.016
0.314 0.079 0.029 0.122 0.030 0.010 0.010 0.142 0.035 0.044 0.816 0.170 0.015
0.293 0.073 0.026 0.121 0.027 0.010 0.009 0.130 0.032 0.041 0.760 0.204 0.036
Total (C þ I þ G)
1.000
1.000
1.000
1.000
1.000
Food Alcohol Tobacco Housing Fuel and light Light Soap Clothing Domestic service Transportation Total consumption Investment Government
0.438 0.099 0.007 0.088 0.022 0.010 0.010 0.073 0.025 0.039 0.812 0.120 0.068
0.439 0.099 0.007 0.088 0.022 0.010 0.010 0.073 0.025 0.039 0.814 0.106 0.080
0.458 0.104 0.008 0.092 0.023 0.010 0.010 0.077 0.026 0.041 0.849 0.096 0.055
0.342 0.134 0.017 0.103 0.048 0.012 0.005 0.115 0.027 0.042 0.844 0.095 0.061
0.347 0.136 0.018 0.105 0.049 0.012 0.005 0.117 0.027 0.042 0.858 0.091 0.051
Total (C þ I þ G)
1.000
1.000
1.000
1.000
1.000
US
Great Britain
Notes and Sources: See Appendix.
Table 4 provides the expenditure weights for the main categories of consumption and overall GDP.21 There are large differences in expenditure shares between the UK and the US. Given the dramatic differences in relative prices, some differences are to be expected. While consumption accounts for the largest portion of expenditure in both the countries, the US investment share increases dramatically
259
Relative Income
over the period. It is twice the UK share by 1869. The most obvious difference between the US and British consumption patterns is the smaller US share of food. This probably reflects lower US food prices. It may also be due to higher US living standards and lower levels of urbanization. Other notable differences in consumption patterns are the greater portion of British budgets devoted to drinks and the smaller shares for housing and clothing. Returning to the UK/US overall price levels in Table 3, we can now explain the fall in the UK’s relative price level between 1831 and 1869. The prices and expenditure weights in Tables 3 and 4 show that it is due to the fall in the UK/US consumption price level. While government prices also fall over this period, their share of total expenditure is small. The share of investment is larger than for government, but UK/US investment prices fluctuate between 1831 and 1869. The fall in consumption prices, in turn, is explained by falling food prices, presumably due to better integration of world markets. Reductions in the relative price of drinks and tobacco also play a role. As mentioned, we calculate overall UK/US price levels as Fisher indices. It is therefore instructive to examine relative UK/US price levels obtained using US and British weights respectively. Table 5 gives relative prices for overall GDP, consumption and food. A central finding of the literature on international GDP comparisons is that comparing two economies using the expenditure weights of the richer economy will produce a higher relative price level than a comparison using the weights of the poorer economy. This result arises from the fact that relative quantities consumed tend to be negatively related to relative price levels.22 Table 5.
UK/US Relative Prices Using US and British Weights. 1831
1839
1849
1859
1869
1.721 1.831 2.502
1.383 1.489 2.017
1.242 1.398 1.674
1.146 1.252 1.449
1.059 1.221 1.287
1.567 1.802 2.283
1.305 1.467 1.817
1.227 1.347 1.599
1.079 1.151 1.428
1.040 1.118 1.296
US Weights Overall GDP Consumption Food British Weights Overall GDP Consumption Food Sources: See Appendix.
260
MARIANNE WARD AND JOHN DEVEREUX
In our case, we would expect relative UK price levels to be higher with US weights. This turns out to be the case with the exception of 1869 for food where they are roughly equal. On the other hand, Table 5 shows the spreads are generally smaller than found for recent years. For post 1970 GDP comparisons, price levels calculated using UK weights are 80–85 percent of price levels obtained with US weights (see Maddison, 1995, Tables C1–C6). We interpret the low spreads in Table 5 as potentially a consequence of our weak expenditure data. As outlined earlier, these do not directly relate to comparison years. On the other hand, we do not believe this produces a systematic bias in our estimates.
4. CONFIRMATIONS AND CROSSCHECKS The benchmark estimates of UK/US GDP in previous sections are controversial since they depart from widely held views on relative UK/US living standards before 1870. This section provides supporting evidence. First, we compare UK/US GDP from the output side with sectoral productivity data for 1860 and 1870. Next, we discuss the implications of higher US real wages for relative UK/US GDP. In a sense, real wages proxy real GDP from the income side.23 4.1. Sectoral Productivity Paige and Bombach (1959) emphasize the importance of sectoral estimates as a check on expenditure comparisons. We provide our estimates of UK/ US output per worker for 1860 and 1870 derived from sectoral data in Table 6. To construct the sectoral benchmarks, we draw on Rostas (1948), Paige and Bombach (1959) and Broadberry (1997). Our estimates cover eight sectors: agriculture, mining, manufacturing, construction, trade, transportation, finance/services and government.24 For manufacturing, mining and public utilities, we take our estimates from Broadberry and Irwin (2004). We provide new estimates for agriculture, construction, transportation and trade derived from nominal value added per worker and sectoral price benchmarks. For agriculture, we compare farm gate prices. For construction, our price benchmarks rely on building wages and input prices. For transportation, we compare train costs per passenger mile and per tonmile. For trade, we estimate the quantity of goods flowing through wholesale and retail channels. For finance/services and government, we assume equal labor productivity. Complete details are in the appendix.
261
Relative Income
Table 6.
Comparative UK/US Labor Productivity (US ¼ 100). 1860
1870
Agriculture Mining Manufacturing Construction
58 165 52 66
59 98 55 86
Total Commodities
62
65
Trade Transportation Public utilities Finance/services Government
71 50 113 100 100
71 65 113 100 100
Total Services
89
79
Total GDP
75
75
Sources: See Appendix.
The results show higher US output per worker in both the years. Overall UK/US output per worker is 75 for 1860 and 1870. Moreover, the US leads in terms of output per worker for both commodities and services. The largest leads are in commodities. This is consistent with the expenditure benchmarks from Table 2, which show UK/US output per worker is 66 in 1859 and 72 in 1869. These findings differ from the sectoral comparisons of Broadberry (1997) and Broadberry and Irwin (2004), that show the UK leading in output per worker. Broadberry and Irwin (2004), for example, show the UK leading by 6 percent in 1860 and 5 percent in 1870. Fortunately, it is easy to reconcile our results with theirs. As it turns out, Broadberry (1997) and Broadberry and Irwin (2004) find higher UK output per worker primarily because of agriculture. For 1870, they show UK output per worker in agriculture as 9 percent higher than the US. This is twice our estimate. A priori, we are skeptical of this estimate. Consider the implications for British and Irish agriculture separately. Turner (1996, Table 5.2, p. 129) puts Irish agricultural output per worker at 48 percent of British levels for 1871. Given that Ireland comprised 36 percent of the UK agricultural labor force, the Broadberry and Irwin estimates imply labor productivity for British agriculture is 35 percent higher than the US They also imply that Irish output per worker is 65 percent of US levels. These estimates seem implausible.
262
MARIANNE WARD AND JOHN DEVEREUX
The second problem with their agricultural estimates is that they require UK farm-gate prices below US levels. Using standard estimates of value added in agriculture, Gallman (1960) for the US and Feinstein (1972) for the UK, their estimates yield an 1870 UK/US price level of 70.25 This is implausible given what we know about transport costs and relative UK/US wholesale prices during a period in which the US exported food to the UK. To reconcile the results for agriculture, we have to look at their procedures. Broadberry and Irwin (2004) derive their agricultural estimates by projecting a 1910 benchmark from Broadberry (1997) backwards with sectoral output indices.26 The explanation for the agricultural differences lies in this benchmark. Table 7 provides their agricultural benchmark for 1910, along with our revision. Agricultural output per worker is the ratio of value added per worker deflated by the UK/US relative price level. They show a UK/US price level of 118. Our estimate is 107. The key difference lies in different estimates of value added per worker. They appear to use value added per worker for Great Britain ($378) instead of the UK (which is $276 from Feinstein, 1972). Second, their estimate of US value added per worker ($347) is too low. Our estimate is $496. It is in line with all the US estimates we could find (see Tostlebe, 1957 and Strauss and Bean, 1940). Thus, they overstate 1910 UK agricultural productivity by almost 100 percent. This carries over to their projections to earlier years leading to biased estimates. Once we correct agriculture, our estimates are in agreement with theirs for overall GDP per worker for 1870.27
Table 7.
The 1910 UK/US Agricultural Benchmark (US ¼ 100).
Relative Nominal Farm Gate Output Per Prices PUK/ Worker UK PUS ($) Broadberry (1997) Ward and Devereux
Nominal Output Per Worker US ($)
Relative Nominal Output Per Worker YUK/YUS
Real Output Per Worker UK/US yUK/yUS
118
378
347
109
92
107
276
496
56
52
Notes and Sources: The Broadberry and Irwin (2004) agricultural benchmark is from Broadberry (1997) page 27. Our estimate uses the sources and methods outlined in the appendix as applied to 1910. Note, Broadberry refers to 1908/1909 while our estimates refer to 1910.
263
Relative Income
4.2. Real Wages There is a consensus among historians that US real wages were much higher during this period. This is potentially an important point. Could the US have higher real wages and still have lower output per worker? As it turns out, this is indeed a possibility. To show this, Eq (2) considers the relationship between economy wide real wages and relative output per worker, where wus/uk are relative real wages, yus/uk is relative output per worker, b is the share of labor income in GDP and p/pc is the relative GDP price level divided by the consumption price level. The wage here refers to the economy wide real wage and covers all labor income that is unskilled, skilled, clerical and professional workers. W uk=us ¼ yuk=us buk =bus p=pc (2) This expression shows that relative output per worker can deviate from relative real wages if there are differences in labor shares and/or differences in relative price levels for consumption relative to overall GDP. As it happens, both factors are at work. In the first place, there is evidence that US labor shares were higher. Gallman (1992), for example, assumes a labor share of 0.68. In contrast, most work for Great Britain and the UK assumes a labor share of around 0.6 (see Crafts, 2004, Table 1). The difference in labor shares will reduce relative UK real wages as compared to output per worker. Second, our price benchmarks show differences between the overall UK/US price level and for consumption, with UK consumption prices higher for all years. This will also depress UK/US real wages as compared to output per worker. Can these forces explain higher UK output but lower real wages? The calculations in Table 8 suggest that this is unlikely to be the case. The second column gives our benchmark estimates of relative output per worker. The third column shows the implied real wage rates given our relative prices and assumed higher US labor shares. As expected the US lead in real wages is higher than output per worker, reflecting the higher labor share in the US and the relatively lower cost of consumption. On the other hand, these differences are probably not enough to produce higher UK output per worker given what we know about UK/US real wage rates.28 To illustrate this point, the final column gives widely cited estimates of relative real wages from Williamson (1995). We see a commanding US lead for all years. While higher US real wages are not necessarily inconsistent with higher UK output per worker, the size of the US lead in real wages makes that unlikely. Williamson’s wages cover
264
MARIANNE WARD AND JOHN DEVEREUX
Table 8. Year
Output Per Worker and Real Wages (US ¼ 100). Benchmark Output Per Worker
Implied Real Wage
Real Wage Williamson (1995)
58 72 66 66 72
45 56 51 52 55
61 52 57 59 59
70 86 73 78 81
54 67 56 62 63
69 58 61 62 62
United Kingdom 1831 1839 1849 1859 1869 Great Britain 1831 1839 1849 1859 1869
Notes and Sources: To obtain the implied real wage, we assume that the share of labor in GDP for the US is 0.7 while it is 0.6 for the UK. We use the relative consumption and overall price levels from Table 3 to obtain implied real wage rates. The last column is from Table A2-1 of the updated data appendix to Williamson (1995). We form the UK index by adding real wages for Great Britain and Ireland using labor force weights.
unskilled workers. We might obtain different results using real wages for all workers, unskilled, skilled, clerical and professional. The results of Table 8 are therefore suggestive rather than conclusive.29
5. RECONCILING BENCHMARKS WITH PROJECTIONS The most famous UK/US projections are from Angus Maddison (1995, 2001, 2003), who provides an annual series on relative income per capita between 1870 and 2000 in 1990 prices. He also covers, 1820, 1830, 1840, 1850 and 1860.30 The second column of Table 9 shows Maddison’s projections of UK/US income per capita. Here the UK leads by approximately 25 percent for all years. This is 40 percent above our benchmarks.31 What explains the differences? As mentioned, conflicts between benchmarks and projections occur in all international comparisons. They arise because of differences in weights, goods sampled and procedures used to
265
Relative Income
Table 9.
Comparing Estimates: UK/US Real GDP Per Capita (US ¼ 100). Maddison’s Projections
1820 1830 1840 1850 1860 1870
1990 prices, 1990 borders
1990 prices, current borders
131 123 122 125 126 126
127 119 118 121 121 122
1950 prices, current borders
97 90 89 92 92 93
Benchmark
Na 76 79 82 83 100
Notes and Sources: Updated estimates of real GDP per capita in 1990 prices from http:// www.eco.rug.nl/Maddison/. We transform Maddison’s estimates to a Fisher Ideal basis in column three using the information on C-6 page 172 of Maddison (1995). Column four uses a 1950 benchmark from Ward and Devereux (2003) as derived from Gilbert and Kravis (1954), and projected using the Maddison UK/US GDP series. The benchmark estimates in column 5 are from Table 1 and refer to 1831, 1839, 1849, 1859, and 1869.
calculate growth rates.32 Maddison (1995, 2001, 2003) compares UK/US output using Geary–Khamis world prices with 1990 borders. To allow comparison with our estimates, we have to change his estimates to a Fisher Ideal basis and current borders. These estimates are provided in column three. The resulting changes are minor suggesting that the explanation for differences rests elsewhere. A closer look suggests that the main reason why his estimates differ from the benchmarks is his 1990 base. To show this, the fourth column of Table 9 projects UK/US income per capita using a 1950 price benchmark from Gilbert and Kravis (1954). These projections show the US as ahead for all years, although the US lead is smaller than the benchmarks. The result that the US leads with early prices is a fairly general one. It holds if we project UK/US GDP using 1960 prices from Dennison (1967) or 1965 prices from Maddison (1983). Only when we use price structures from recent years, such as 1990, is UK’s income higher. Which base year prices are appropriate? We prefer prices as close as possible to the period in which we are interested. In other words, we believe that 1950, 1960 or 1965 prices are superior to those of 1990. Thus for UK/US GDP projections, the important differences are not always between direct comparisons and long-span projections per se but rather the set of base year prices used to compare output.33
266
MARIANNE WARD AND JOHN DEVEREUX
6. A LONGER RUN PERSPECTIVE Our UK/US benchmark comparisons from earlier sections show the US ahead in terms of income per capita and output per worker from 1830 to 1870. We have argued that these results are broadly in line with what we know about sectoral productivity and real wages. In this section, we take a longer run perspective by integrating our estimates with data from the post1870 period to provide a long-run comparison of relative UK/US living standards and relative labor productivity. Figure 1 tracks income per capita and output per worker from 1831 to 1990 by linking our UK/US and GB/US benchmarks from Tables 1 and 2 to revised Ward and Devereux (2003) expenditure benchmarks for 1870–1990.34 These estimates show the US with a lead in income per capita and especially output per worker from 1830 to date. While the US is ahead, there is no evidence that it forges ahead before the 1900s.35 In fact, the benchmarks show the UK maintains its relative income and productivity levels until the First World War, supporting the claims of Deirdre McCloskey 1.15 UK/US Real GDP Per Capita
1.05
UK/US Real GDP Per Worker GB/US Real GDP Per Capita
Relative Real GDP
0.95 0.85 0.75 0.65 0.55 0.45
1987
1981
1975
1969
1963
1957
1951
1945
1939
1933
1927
1921
1915
1909
1903
1897
1891
1885
1879
1873
1867
1861
1855
1849
1843
1837
1831
0.35
Year
Figure 1. Relative Income, 1831–1990. Notes and Sources: UK/US and GB/US benchmarks are calculated as the ratio of relative nominal GDP to the UK/US price levels. UK/US price levels for 1831–1869 (see Appendix). UK/US price levels for 1870–1990 are updated from Ward and Devereux (2003). US, UK and GB nominal GDP, population and labor force: See Appendix.
267
Relative Income
(1970). Only then do we see relative decline. The low point is 1950 where UK income per capita reaches 45 percent of US levels. The UK recovers in the post-war period reaching 70 percent of US levels by the 1970s, a position maintained to the present. Thus, Fig. 1 suggests that American leadership in living standards and productivity is long standing and dates at least to the early part of the nineteenth century. If correct, this raises fundamental questions about the sources of American prosperity. Over the last 20 years, the literature on American primacy focuses on the period after 1870, and in particular on American industrialization and urbanization. These forces do not explain the high levels of income per capita and output per worker early on when the US is an agricultural economy. From a British perspective, its position relative to the US in Fig. 1 adds fuel to long-standing debates about British growth and living standards. Most notably, it suggests that relative British decline is largely a feature of the period after the First World War.
7. CONCLUDING COMMENTS We offer tentative estimates of UK and US living standards and output per worker for the crucial years between 1830 and 1870. During this period, the UK experienced the full force of the industrial revolution and the US became an economic powerhouse. Our estimates show the US leads in income per capita and output per worker, when compared to Great Britain and the United Kingdom, between 1830 and 1870. Our findings are consistent with evidence from other sources such as real wage comparisons and sectoral productivity data. They are also consistent with long-span projections when we use a relative price structure close in time to the period compared. Our estimates are necessarily tentative. The next step is to refine them. In the first place, the fact that we rely on price data, not collected specially to compare UK/US prices, hampers our work. One way around this problem lies in the many guides published for potential immigrants to the US during these periods. These may partially substitute for the price surveys that form the base for modern international GDP comparisons.36 In addition, we need ways to better integrate the various sources of price data that we have. One possibility is hedonic regressions.37 The key advantage of this approach is that it allows us to adjust location and quality difference. It may also help us to integrate data from retail, wholesale and institutional sources. Our rural/ urban price adjustments also need refining. Finally, an important area for future improvement is the expenditure weights.
268
MARIANNE WARD AND JOHN DEVEREUX
Can we extend our GDP benchmarks further back in time? Here we are pessimistic. The constraint we face is nominal GDP.38 We do not possess a proper US series for nominal GDP before the 1830s.39 Furthermore, the British nominal GDP estimates before 1830 remain controversial.40 In all likelihood, progress on UK/US comparisons before the 1830s will continue to depend on GDP projections complemented by real wage comparisons. Even in this case, price benchmarks can help by providing benchmarks for GDP projections or real wage comparisons that reflect patterns of relative prices relevant to the period compared.
NOTES 1. The recent interest in current price estimates begins with Prados de la Escosura (2000) and his short-cut approach. Our work builds on our earlier work in Ward and Devereux (2003) where we provide UK/US current price benchmarks from 1870 to 1990. Other current price PPP expenditure comparisons in economic history include Heston (1998), Van Zanden (2003) and Ma, Fukao, and Yuan (2004). Heston and Summers (1980) is a pioneering early study. Broadberry (2003) and Broadberry and Irwin (2004) criticize the current price estimates of Ward and Devereux and Prados de la Escosura for the UK/US. Our reply is Ward and Devereux (2004). 2. Their work forms the basis for the International Comparison Project (ICP) and the Penn World Tables. Clark (1940) is a forerunner of the OEEC methodology. He also provides one of the first UK/US sectoral comparisons. Maddison (2004) provides an appreciation of Clark’s achievement. 3. Our task is simplified by the fact that the UK and the US share similar consumption patterns. Van Zanden (2003) compares income per capita between Java and the Netherlands where this does not hold. 4. This point is enshrined in the handbook of the ICP. Consider the following passage on page 8 of the 2004 handbook:‘‘ y On the other hand, it is not possible to ensure complete consistency between PPPs and the temporal price indices in the national accounts. yHowever, the system of weights used to calculate PPPs are influenced by data from more than one country, whereas the weights used to calculate price indices within a country are not. yFor these and other kind of reasons, the index formulae used to measure price movements within a country and those used to calculate multilateral PPPs are generally not fully consistent with each other. This can create discrepancies between the two kinds of data, which may be difficult to explain. It is impossible to avoid these problemsy 5. See, for example Varjonen (2001). 6. It should be borne in mind that our price series are from British sources. The limited evidence that we could find suggests that price levels were fairly similar in Great Britain and Ireland. 7. The higher UK prices in Table 1 are consistent with contemporary views. Senior (1830, p. 2) implies a GB/US price level of 156 while Carey (1835, pp. 223–224) puts it at 132 (See also Cairnes (1874)). Adam Smith (1776/1937, p. 70),
Relative Income
269
writing about a much earlier period, held that the US price level was lower: ‘‘The prices of provisions is everywhere in North America much lower than England.’’ 8. The UK’s relative standing improves between 1859 and 1869 due to the effects of the Civil War. 9. The Wright study is less known. The price data come from a ‘‘large number of bills, day-books and ledgers containing the accounts of the daily transactions of large country stores at different intervals; of memorandum, pass and family account books; of general family expense books, farm accounts and manuscript material obtained from original sources by agents of the Bureau, in various sections of the state’’ Massachusetts Bureau of Statistics of Labor (1885, pp. 40–41). Altogether the price estimates were based on 120,000 price quotations. 10. Other US sources, for example Adams (1944), provide price indexes without providing prices for individual items. 11. Seaman (1852, p. 278), finds a spread of 50 percent between rural and urban prices. This is greater than our estimates. It seems that he refers to farm-gate/urban price differences rather than our urban/rural retail spread. In general, we suspect that our estimates may understate US urban/rural retail spreads. Evidence in support of this possibility comes from Adams (1986, 1992), who shows much lower rural prices for Maryland and West Virginia as compared to our US estimates. 12. We also compared UK/US farm gate prices from 1830 to 1870. The results are consistent with the relative food price levels in Table 3. 13. Higher British prices of tea and coffee are also due to taxes. 14. Contemporary observers note low US prices for spirits see Rorabaugh (1979). 15. We omit firewood for the UK, as we were unable to obtain a satisfactory British series. 16. Taussig (1931) suggests that the US cotton industry is competitive at least in lower quality items early on (See also Irwin and Temin, 2001). 17. Clark (2002) provides an alternative British rent series based on rent records for many properties. To using the Clark series, increases the UK rents and reinforces our results. 18. We base our housing comparisons on rents. Alternatively, we could compare construction costs for housing. While such comparisons are inherently less reliable, they provide insight into the reasons why rents were higher in the US at least for urban areas. Our investment benchmarks show higher US construction costs. These are largely due to wages. Wage costs accounted for 50 percent of US housing costs during this period (see Adams, 1975 and David & Solar, 1977). In addition to costs, Henry Carey (1835) emphasizes higher US interest rates as a second factor explaining higher US rents. A final factor is the generally greater US depreciation rates for housing. 19. Our referee points out that by basing our comparison on rates per mile, we ignore the fact that trips were longer in the US. Since all journeys have a fixed cost, our comparison is biased against the US. Unfortunately, the data do not allow us to adjust the estimates to account for this bias. 20. To compare GDP, we need GDP weights reflecting overall consumption patterns. There is little budget information on the consumption patterns of the middle class or the well to do. 21. In order to save space, we provide the individual food item weights in the appendix.
270
MARIANNE WARD AND JOHN DEVEREUX
22. More formally, this is known as the Paasche–Laspeyeres Spread. This issue is widely discussed in the literature on international comparisons (see Kravis, Heston, & Summers, 1982). Neary and Gleeson (1997) provide a recent formal account. 23. Heston (1998) suggests that wage comparisons are an important check on real income comparisons in history. 24. The 1870 sectoral comparison is a revised version of that appearing in Ward and Devereux (2004). 25. See Ward and Devereux (2004). 26. A further problem is that they project UK agricultural output using value added, while they use gross output for the US. This biases the relative position of the US downwards since gross output grows faster than value added leading Broadberry (1997) to greatly underestimate US productivity when he uses a 1937 benchmark to project backward to 1870. 27. We will supply a complete reconciliation of our results with those of Broadberry and Irwin (2004) on request. 28. The appendix provides nominal wage data for construction workers and domestics. We also looked at wage rates for agricultural workers and other unskilled occupations. Williamson (1984) for the UK and Williamson and Lindert (1980) as well as Weiss (1975) for the US provide some earnings for clericals and professionals. Comparing UK/US wage rates for a wide variety of occupations and industries is a major undertaking that is beyond the scope of this paper. A preliminary look at the data suggests that nominal wage differentials in favor of the US were substantial. They were often above 40–50 percent for unskilled and skilled workers, a little higher than found by Adams (1970) for an earlier period. The situation is more complicated for clerical and professional occupations, given the scattered and incomplete data. We suspect relative British earnings were sometimes higher in these areas. 29. This does not exhaust possible checks. In particular, we should note the ‘‘short cut’’ UK/US GDP comparisons provided by Prados de la Escosura (2000). His approach is similar to ours in that he compares GDP in current international prices. His method relies on a reduced form relationship between relative price levels and other economic variables estimated for post-1950 data. Using the results, he then compares GDP for earlier years. Transforming his estimates to a Fisher Ideal basis, we find UK income is around 90 percent of US income for our benchmark years. His approach raises some difficult issues. In particular, it assumes the relationship between relative price levels and other variables is constant over time. 30. The first systematic comparisons of British and American living standards for years before 1870 are Kuznets (1964) and Gallman (1966). We consider their work later. 31. It should be noted that Maddison’s estimates imply the US with a slight lead in output per worker owing to lower rates of labor force participation in the US. 32. There is a substantial literature on reconciling benchmarks and projections. Recent work includes Aten and Heston (2002) and Dalgaard and Serensen (2002). 33. Using 1960 prices, Kuznets (1964) shows GB/US income per capita as roughly equal in 1840 implying that US leads the UK (see also Kuznets (1977)). Gallman compares GB/US income using exchange rates, a sectoral comparison and a longspan projection with 1950 prices. His GDP projection shows GB/US income as 90 in
271
Relative Income
1840, (1966, Table 1, p. 5). The fact that these studies used pre-1970 base prices explains why they differ from Maddison (1995, 2001, 2003). 34. Our revisions to the Ward and Devereux (2003) expenditure benchmarks do not change any of the substantive results. 35. The differences between the per capita and per worker estimates before 1914 reflect increases in the US labor force participation rates. 36. Fearon (1818), for example, provides detailed price benchmarks for the major US cities of the period. 37. Clark (2002) and Margo (2000) explore the use of hedonic regressions in the case of prices and wages respectively for historical data. 38. We also calculated UK/US price benchmarks from 1801 to 1831. Without nominal GDP, we cannot compare real GDP. Our benchmarks do, however, allow us to compare real wages. 39. David (1967) and Weiss (1994) provide US GDP for earlier years but their procedures do not yield nominal GDP estimates. 40. See Clark (2001).
ACKNOWLEDGMENTS This research was made possible by financial support from Loyola College and a grant from the National Science Foundation. We thank an anonymous referee for very helpful suggestions and for drawing our attention to some fundamental methodological issues. We also appreciate the comments of Robert Allen, Peter Lindert, Jan Luiten Van Zanden, Peter Scholliers and the other participants at the Utrecht conference on the Global History of Prices and Wages. We thank Santhi Hejeebu for suggestions on an earlier draft. Finally, special thanks are due to Gregory Clark for his editorial assistance, for providing price data and for showing us how to compare the prices of coal and wood. The usual disclaimer applies.
REFERENCES Adams, D. (1970). Some evidence on English and American wage rates, 1790–1830. Journal of Economic History, 30, 499–520. Adams, D. (1975). Residential construction industry in the early nineteenth century. Journal of Economic History, 35, 794–816. Adams, D. (1986). Prices and wages in Maryland, 1750–1850. Journal of Economic History, 46, 625–645. Adams, D. (1992). Prices and wages in Antebellum America, the West Virginia experience. Journal of Economic History, 52, 206–216.
272
MARIANNE WARD AND JOHN DEVEREUX
Adams, T. M. (1944). Prices paid by Vermont farmers for goods and services received by them for farm products, 1790–1940. University of Vermont and State Agricultural Policy Bulletin no. 507. Ashton, T. S. (1949). The standard of life of the workers in England, 1790–1830. Journal of Economic History, 9, 19–38. Aten, B., & Heston, A. (2002). Benchmark reconciliations revisited. Unpublished Manuscript. University of Pennsylvania. Bairoch, P., & Goertz, G. (1986). Factors of urbanization in the nineteenth century developed economies: A descriptive and econometric analysis. Urban Studies, 23, 285–305. Barger, H. (1955). Distribution’s place in the American economy since 1869. Princeton: Princeton University Press. Bowley, A. L. (1901). The statistics of wages in the United Kingdom during the last hundred years (Part VIII) wages in the building trades—concluded. Journal of the Royal Statistical Society, 64, 102–112. Brady, D. (1964). Relative prices in the nineteenth century. Journal of Economic History, 24, 145–203. Brady, D. (1972). Consumption and the style of life. In: L. Davis, R. Easterlin & W. Parker (Eds), American economic growth. New York: Harper & Row. Brassey, T. (1873). Work and wages practically illustrated. New York: D. Appleton. Broadberry, S. (1997). Forging ahead, falling behind, and catching-up: A sectoral analysis of Anglo-American productivity differences, 1870–1990. Research in Economic History, 17, 1–37. Broadberry, S. (2003). Relative per capita income levels in the United Kingdom and the US since 1870. Journal of Economic History, 63, 852–863. Broadberry, S., & Irwin, D. (2004). Labor productivity in Britain and America during the nineteenth century. NBER Working Paper no. 10364. Burnett, J. (1966). Plenty and want. London: Nelson. Caird, J. (1878). The landed interest and the supply of food. London: Cassel. Cairnes, J. E. (1874). Some leading principles of political economy newly expounded. New York: Harper. Carey, H. C. (1835). Essay on the rate of wages. Philadelphia: Carey, Lea & Blanchard. Chadwick, D. (1860). On the rate of wages in Manchester and Salford, and the manufacturing districts of Lancashire, 1839–59. Journal of the Statistical Society of London, 23, 1–36. Clark, C. (1940). The conditions of economic progress (1st ed.). London: Macmillan. Clark, G. (2001). The secret history of the industrial revolution. Working Paper. University of California, Davis. Clark, G. (2002). Shelter from the storm: Housing and the industrial revolution, 1550–1912. Journal of Economic History, 62, 489–511. Clark, G. (2004). The price history of English agriculture. Research in Economic History, 22, 41–124. Coehlo, P., & Shepherd, R. (1976). Regional differences in real wages: The United States, 1851–1880. Explorations in Economic History, 13, 203–230. Cole, A. (1938). Wholesale commodity prices in the United States, 1700–1861: Statistical supplement. Cambridge: Harvard University Press. Crafts, N. F. R. (1985). British economic growth during the industrial revolution. New York: Oxford University Press.
Relative Income
273
Crafts, N. F. R. (2004). Productivity growth in the industrial revolution: A new growth accounting approach. Journal of Economic History, 64, 521–535. Dalgaard, E., & Serensen, H. S. (2002). Consistency between PPP benchmarks and national price and volume indices. Unpublished manuscript, OECD. David, P. (1967). The growth of real product in the United States before 1940; new evidence, controlled conjectures. Journal of Economic History, 27, 151–197. David, P., & Solar, P. (1977). A bicentenary contribution to the history of the cost of living in America. Research in Economic History, 2, 1–80. Deane, P., & Cole, W. A. (1967). British Economic Growth, 1688–1959 (2nd ed.). Cambridge: Cambridge University Press. Dennison, E. (1967). Why growth rates differ. Washington, DC: Brookings Institution. Dodd, A. H. (1951). The industrial revolution in North Wales. Cardiff: University of Wales Press. Eden, F. M. (1797). The state of the poor. London: J. Davis. Edmonds, R. (1839). A statistical account of the parish of Madron. Journal of the Statistical Society of London, 4, 198–223. Fearon, H. (1818). A narrative of a journey through America. London: Longman. Feinstein, C. H. (1972). National income, expenditure and output of the United Kingdom 1855–1965. Cambridge: Cambridge University Press. Feinstein, C. H. (1996). Conjectures and contrivances: Economic growth and the standard of living in Britain during the industrial revolution. Unpublished Working Paper, Oxford University. Feinstein, C. H. (1998). Pessimism perpetuated: Real wages and the standard of living in Britain during and after the industrial revolution. Journal of Economic History, 58, 625–658. Feinstein, C. H., & Pollard, S. (Eds) (1988). Studies in capital formation in the United Kingdom: 1750–1920. New York: Clarendon Press. Fishlow, A. (1965). American railroads and the transformation of the ante-bellum economy. Cambridge: Harvard University Press. Fishlow, A. (1966). Productivity and technological change in the railroad sector, 1840–1910. Trends in the American economy in the nineteenth Century BER Studies in Income and Wealth, 30, 583–646. Gallman, R. (1960). Commodity Output, 1839–1899. In: W. Parker (Ed.), Trends in the American Economy in the Nineteenth Century (Vol. 24). NBER Studies in Income and Wealth. Princeton: Princeton University Press. Gallman, R. (1966). Gross national product in the United States, 1834–1909. NBER Studies in Income and Wealth, 30, 3–76. Gallman, R. (1992). American economic growth before the civil war: The testimony of the capital stocks. In: R. Gallman & J. Wallis (Eds), American economic growth and standard of living in the United States before 1860. Chicago: University of Chicago Press. Gallman, R. (2000). Economic growth and structural change in the long nineteenth century. In: S. Engerman & R. Gallman (Eds), The Cambridge economic history of the United States. Cambridge: Cambridge University Press. Gallman, R., & Weiss, T. (1969). The service industries in the nineteenth century. NBER Studies in Income and Wealth, 34, 287–365. Gilbert, M., & Kravis, I. (1954). An international comparison of national products and the purchasing power of currencies. Paris: OEEC.
274
MARIANNE WARD AND JOHN DEVEREUX
Hatton, T., & Williamson, J. (1991). Wage gaps between the farm and the city: Michigan in the 1890s. Explorations in Economic History, 51, 605–632. Hawke, G. (1970). Railways and economic growth in England and Wales 1840–1870. Oxford: Oxford University Press. Heston, A. (1998). Long term growth, real wages and international real output comparisons. Unpublished, University of Pennsylvania. Heston, A., & Summers, R. (1980). Comparative Indian economic growth: 1870–1970. American Economic Review, 70, 96–101. Hoover, E. (1960). Retail Prices After 1950. In: W. Parker (Ed.), Trends in the American economy in the nineteenth century (Vol. 24). NBER Studies in Income and Wealth Princeton: Princeton University Press. Horrell, S. (1996). Home demand and British industrialization. Journal of Economic History, 56, 561–604. Hunt, E. H. (1973). Regional wage variations in Great Britain: 1850–1914. Oxford: Clarendon Press. Irwin, D., & Temin, P. (2001). The antebellum tariff on cotton textiles revisited. Journal of Economic History, 61, 777–798. Jefferys, J., & Walters, D. (1955). National income and expenditure in the United Kingdom: 1870–1950. In: S. Kuznets (Ed.), Income and wealth, series V. London: Bowes and Bowes. Kendrick, J. W. (1961). Productivity trends in the United States. Princeton: Princeton University Press. Kloft, L. (1995). Problems and issues in constructing national and regional price indices for the United States: 1851–1880. Dissertation. Iowa City, IA: University of Iowa. Kravis, I. B., Heston, A., & Summers, R. (1982). World product and income: International comparisons of real gross product. Baltimore and London: Johns Hopkins University Press. Kuznets, S. (1961). Capital in the American economy. Princeton: Princeton University Press. Kuznets, S. (1962). Quantitative aspects of the economic growth of nations: VII, the share and structure of consumption. Economic Development and Cultural Change, 10, 1–92. Kuznets, S. (1964). Notes on the pattern of US economic growth. In: E. Edwards (Ed.), The nation’s economic objectives. Chicago: University of Chicago Press. Kuznets, S. (1966). Modern economic growth: Rate, structure, and spread. New Haven: Yale University Press. Kuznets, S. (1977). Two centuries of economic growth: Reflections on US experience. American Economic Review, Papers and Proceedings, 67, 1–14. Lebergott, S. (1964). Manpower in economic growth. New York: McGraw-Hill. Lebergott, S. (1976). The American economy: Income, wealth, and want. Princeton, NJ: Princeton University Press. Lebergott, S. (1996). Consumer expenditures: New measures and old motives. Princeton: Princeton University Press. Ma, D., Fukao, K., & Yuan, J. (2004). Price level and GDP in pre-war East Asia. Working paper. Tokyo: The National Graduate Institute for Policy Studies. MacKenzie, W. A. (1921). Changes in the standard of living in the United Kingdom, 1860– 1914. Economica, 3, 211–230. Maddison, A. (1983). A comparison of level of GDP per capita in developed and developing economies, 1700–1980. Journal of Economic History, 43, 27–41.
Relative Income
275
Maddison, A. (1995). Monitoring the world economy, 1820–1992. Paris: Development Center of the Organization for Economic Co-Operation and Development. Maddison, A. (2001). The world economy: A millennial perspective. Paris: Development Center of the Organization for Economic Co-Operation and Development. Maddison, A. (2003). The world economy: Historical statistics. Paris: Development Center of the Organization for Economic Co-Operation and Development. Maddison, A. (2004). Macro-measurement before and after Colin Clark. Unpublished. Margo, R. A. (1996). The rental price of housing in New York City. Journal of Economic History, 56, 605–625. Margo, R. A. (2000). Wages and labor markets in the United States, 1820–1860. National Bureau of Economic Research Series on Long-Term Factors in Economic Development. Chicago: University of Chicago Press. Massachusetts Bureau of Statistics of Labor. (1885). Sixteenth annual report of the Massachusetts Bureau of Statistics of Labor. Boston: Wright and Potter. Matthews, R. C. O., Feinstein, C. H., & Odling-Smee, J. C. (1982). British economic growth, 1856–1973. Stanford: Stanford University Press. McCloskey, D. N. (1970). Did Victorian England fail? Economic History Review, 23, 446–459. Mitchell, B. R. (1971). Abstract of British historical statistics. Cambridge: Cambridge University Press. Mitchell, B. R. (1988). British historical statistics. Cambridge: Cambridge University Press. Mitchell, B. R. (1998). International historical statistics: Europe, 1750–1993. New York: Stockton Press. Mitchell, B. R., & Deane, P. (1962). Abstract of British historical statistics. Cambridge: Cambridge University Press. Mulhall, M. (1898). The dictionary of statistics (4th ed.). London: G. Routledge and Sons. Neale, R. S. (1966). The standard of living: A regional and class study. Economic History Review, 29, 590–606. Neary, J. P., & Gleeson, B. (1997). Measuring the wealth of nations: Reference prices and multilateral real income indexes. Economic and Social Review, 28, 401–421. Neild, W. (1842). Comparative statement of the income and expenditure of certain families of the working class in Manchester and Dukinfield, in the years 1836 and 1841. Journal of the Statistical Society of London, 4, 320–334. Ojala, E. M. (1952). Agriculture and economic progress. Oxford: Oxford University Press. Paige, D., & Bombach, G. (1959). A comparison of national output and productivity of the United Kingdom and the United States. Paris: Organization of European Economic Cooperation. Porter, G. R. (1850a). On a comparative statement of prices and wages during the years from 1842 to 1849. Journal of the Statistical Society of London, 13, 210–218. Porter, G. R. (1850b). On the self taxation of the working class in the United Kingdom. Journal of the Statistical Society of London, 13, 358–364. Prados de la Escosura, L. (2000). International comparisons of real product, 1820–1990: An alternative data set. Explorations in Economic History, 37, 1–41. Purdy, F. (1861). On the earnings of agricultural labourers in England and Wales. Journal of the Statistical Society of London, 24, 328–373. Rhode, P. (2002). Gallman’s annual output series for the United States, 1834–1909. NBER Working Paper no. 8860. Rorabaugh, W. J. (1979). The alcoholic republic, an American tradition. New York: Oxford University Press.
276
MARIANNE WARD AND JOHN DEVEREUX
Rostas, L. (1948). Productivity, prices, and distribution in selected British industries. Cambridge: Cambridge University Press. Seaman, E. (1852). Essays on the progress of nations. New York: Scribner. Senior, N. (1830). Three lectures on the cost of obtaining money. London: John Murray. Simon, M. (1960). The United States balance of payments, 1861–1900. In: W. Parker (Ed.), Trends in the American economy in the nineteenth century (Vol. 24, pp. 629–717). NBER Studies in Income and Wealth. Princeton: Princeton University Press. Smith, A. (1776 / 1937). The wealth of nations. New York: Modern Library. Strauss, F., & Bean, L. (1940). Gross farm income and indices of farm production and prices in the United States, 1867–1937. US Department of Agriculture, Technical Bulletin. 703 Taussig, F. W. (1931). The tariff history of the United States. Cambridge: Harvard University Press. Temin, P. (1964). Iron and steel in nineteenth century America. Cambridge: M.I.T. Press. Thomas, M. (1995). A substantial Australian superiority? Anglo-Australian comparisons of consumption and income in the late nineteenth century. Australian Economic History Review, 35, 10–37. Tostlebe, A. (1957). Capital in agriculture: Its formation and financing. Princeton: Princeton University Press. Towne, M.W., & Rasmussen, W. D. (1960). Farm gross product and gross investment in the nineteenth century. In: W. Parker (Ed.), Trends in the American economy in the nineteenth century (Vol. 24, pp. 255–312). NBER Studies in Income and Wealth. Princeton: Princeton University Press. Trescott, P. (1960). The United States government and national income, 1790–1860. In: W. Parker (Ed.), Trends in the American economy in the nineteenth century, NBER Studies in Income and Wealth, (Vol. 24, pp. 337–361). Princeton: Princeton University Press. Turner, M. (1996). After the famine: Irish agriculture, 1850–1914. Cambridge: Cambridge University Press. United States Bureau of the Census. (1975). Historical statistics of the United States, colonial times to 1970. Washington: Government Printing Office. United States Congress. House of Representatives (1886). Reports on the average retail prices of the necessaries of life in the United States. Report prepared by Joseph D. Weeks for the tenth census of the United States. Washington, DC: Government Printing Office. United States Congress. Senate. Committee on Finance. (1893). Wholesale prices, wages and transportation (Aldrich wholesale price report). Washington, DC: Government Printing Office. Van Zanden, J. (2003). Rich and poor before the industrial revolution: A comparison between Java and the Netherlands at the beginning of the nineteenth century. Explorations in Economic History, 40, 1–23. Varjonen, S. (2001). Consistency between GDP based on PPP’s and national accounts time series. OECD Unpublished. Ward, M., & Devereux, J. (2003). Measuring British decline: Direct versus long span income measures. Journal of Economic History, 63, 826–851. Ward, M., & Devereux, J. (2004). A reply to professor Broadberry. Journal of Economic History, 64, 879–891. Weiss, T. (1975). The service sector in the United States, 1839 through 1899. New York: Arno Press. Weiss, T. (1992). US labor force estimates and economic growth, 1800–1860. In: R. Gallman & J. Wallis (Eds), The standard of living in the United States before 1860. Chicago: University of Chicago Press.
277
Relative Income
Weiss, T. (1994). Economic growth before 1860: Revised conjectures. In: T. Weiss & D. Schaefer (Eds), American economic development in historical perspective. Stanford: Stanford University Press. Williamson, J. G. (1984). Did British capitalism breed inequality? Boston: Allen and Unwin. Williamson, J. G. (1995). The evolution of global labor markets since 1830: background evidence and hypotheses. Explorations in Economic History, 32, 141–196. Williamson, J. G., & Lindert, P. (1980). American inequality: A macroeconomic history. New York: Academic Press. Young, E. (1875). Labor in Europe and America: A special report on the rates of wages, the cost of subsistence, and the conditions of the working classes in Great Britain, France, Belgium, Germany, and other countries of Europe, also in the United States and British America. Philadelphia: S.A. George and Co.
APPENDIX. DATA SOURCES Nominal GDP UK GDP at market prices. 1831–1869: GDP at market prices from C.H. Feinstein as reported in Mitchell (1988). After 1855, Mitchell (1988, Table 5A, p. 836) provides a compromise estimate of GDP at factor cost. For earlier years, only GDP measured from the expenditure side is available. To approximate the compromise measure at factor cost, we reduce this by 8 percent, the average difference between compromise and expenditure based estimates for 1855–1885. We convert the compromise estimates of GDP at factor cost to market prices using the factor cost adjustment from Table 5, p. 832. 1870–1949: Mitchell (1988), compromise estimate at factor cost in Table 5A converted to market prices using the factor-cost adjustment in Table 4. 1950–1990: http://www.nationalstatistics.gov.uk. Expenditure at current market prices. GB GDP at market prices. 1831–1905: We assume that British GDP at market prices is 82.5 percent of UK GDP at market prices before the famine and 93 percent after. We take these ratios on the authority of Deane and Cole (1967, p. 168). The resulting British GDP estimates accord with independent estimates for Great Britain from Deane and Cole for all years except 1841. Exchange rates: http://www.eh.net/hmit/exchangerates/ pound.php. US GDP. 1831–1869: For 1831, we project the 1839 estimates using the US nominal GDP series from http://www.eh.net/hmit/gdp/. 1839–1859 GDP calculated as the sum of commodities from Gallman (1960, Table A.1) and services from Gallman and Weiss (1969), Tables A.1, A-4. For 1869, we
278
MARIANNE WARD AND JOHN DEVEREUX
use the Gallman GNP from Rhode (2002). We adjusted this estimate to include the government using Kendrick (1961, Table A-IIb) and Kuznets (1961, Table R-23). Finally, we convert GNP to GDP by adding factor payments from abroad from Simon (1960). 1870–1888: Rhode (2002) adjusted to include government expenditures and from GNP to GDP. 1889–1928: Kendrick (1961, Table A-IIb). Converted from GNP to GDP using Simon (1960) and US Bureau of the Census (1975). 1929–1990: http://www.bea.doc.gov/bea/dn/nipaweb/index.asp. Estimates dated September 29, 2004.
Population Estimates US Population. 1831–1899: US Bureau of the Census (1975), Series A7; 1900–1990: http://www.census.gov/. GB Population. 1831–1905: The sum of England, Wales and Scotland from Mitchell (1998, Table A5). UK Population. 1831–1849: Mitchell (1988). 1859–1869: Feinstein (1972); 1870–1990: http://www.eh.net/hmit/ukgdp/.
Labor Force Estimates US Labor Force. 1831–1899: Weiss (1992) provides labor force estimates by decade. We convert these estimates to our benchmark years with US population growth rates from US Bureau of the Census (1975), Series A7. 1900–1955: Lebergott (1964, Table A-3). Civilian labor force; 1960–1990: http://www.bls.gov/fls/home.htm. UK Labor Force. 1859–1955: Feinstein (1972), Table 57. Feinstein (1972), Table 11.8 provides an estimate of the UK working population in 1851. We project this estimate to 1849 using the growth rate of the UK population from Mitchell (1988). We obtain the estimates for 1839 and 1831 by assuming that the British participation rates for each year applied to the UK. 1960–1990: http://www.bls.gov/fls/home.htm. British Labor Force. 1831–1905: Deane and Cole (1967), Table 31 for 1831. Mitchell (1988), Table 2B provides an estimate of the British occupied population in 1841. We project this estimate to 1839 using the growth rate of the British population. For 1831 and 1839, we assume that the working population was 99 percent of the occupied population based on evidence for 1851–1871 from Feinstein (1972), Table 11.8. British estimates for
279
Relative Income
1849–1869 are from Feinstein, Table 11.8. Feinstein provides estimates of the British occupied population in 1851, 1861 and 1871. We apply the working/occupied population ratio for the UK to the British estimates, and convert to our benchmark years using the growth rate of the British population. The Price Benchmarks We calculate the UK/US relative price levels as Fisher Ideal indices. Exceptions are noted as they arise. Consumption Prices Food. UK: Wheat Flour Chadwick (1860) for 1839–1849 projected to other years with prices from Clark (2004) and Sauerbeck wholesale prices. Bread: Mitchell (1988) all years. Potatoes: 1839–1869 Chadwick (1860), 1839–1859 projected to other years using Clark (2004) and Sauerbeck. Beef: Chadwick (1860), 1839–1859 projected to 1831 and 1869 using Clark (2004) and Sauerbeck. Mutton: 1831 and 1839 Neale (1966); all other years Brassey (1873). Pork: 1831 Ashton (1949) projected to other years with Clark (2004). Bacon: 1831 Ashton (1949) projected to other years by Clark (2004). Milk: 1839– 1859 Chadwick (1860) projected to other years by Clark (2004) and Brassey (1873). Butter: 1839–1859 Chadwick (1860) projected to other years by Clark (2004) and Brassey (1873). Cheese: 1831 Ashton (1949). 1839 Porter (1850a), 1849–1869 Brassey (1873). Sugar: 1849–1869 Brassey (1873) projected to other years with wholesale prices from Mulhall (1898). Tea: 1839–1859 Chadwick (1860), projected to other years by prices from Mulhall (1898) and Brassey (1873). Coffee: 1839 Porter (1850a), 1849–1869 Brassey (1873) projected to other years with wholesale prices from Mulhall (1898). US: Massachusetts Bureau of Statistics of Labor (1885) for 1831– 1849 and the Weeks Report as summarized by Kloft (1995) for 1859–1869. There is one exception. We derive bread from wheat flour using the relationship between bread and flour from Ward and Devereux (2003). Drinks. UK: The evidence from many sources suggests that beer was around 18 pence per imperial gallon for our period. For spirits we used Porter (1850b) benchmark to generate a series from the wholesale price of barley. We assume one bushel of spirits yielded 4.5 imperial gallons of spirits. We further assume the mark up was constant in real terms, projecting it to other years using Feinstein’s (1998) CPI and adding duties.
280
MARIANNE WARD AND JOHN DEVEREUX
US: Beer from Massachusetts Bureau of Statistics of Labor (1885). For spirits we used the Philadelphia wholesale price of whiskey from Cole (1938). We adjusted for the assumed lower quality of relative UK whiskey by using the price relative for gin. Using these sources we still found unrealistically high US prices for spirits from 1831 to1859. We therefore assume that UK prices for spirits were twice the US prices for 1831–1859. Tobacco. UK: We adjust UK import prices from Mulhall (1898) to retail prices using the Porter (1850b) tobacco benchmark for 1850. We projected to other years by assuming a constant retail mark up in real terms and adding tobacco duties from Mulhall (1898). US: Massachusetts Bureau of Statistics of Labor (1885) for all years. We assume that 1869 is equal to 1859. Housing. UK: We projected 1872 UK rent benchmarks from Ward and Devereux (2003) backwards using Feinstein’s (1998) British rent index. US: We take urban US rents from the Weeks Report, Margo (1996) and Brady (1964). Fuel. UK: Retail coal prices provided by Gregory Clark. US: Coal: Young’s (1875) US estimate for 1869. His estimates were projected backwards using wholesale prices from Philadelphia from Cole (1938). Wood: 1869 is from Young (1875) as the average of hard and soft wood per cord projected to other years with data on wood prices from Massachusetts Bureau of Statistics of Labor (1885). We convert wood to coal equivalent using US government conversion factors from http://www.ilatitudes.com/hvac/ residential/information/DOE/doecomparfuel.html, and assuming 20 percent moisture content. Following Lebergott (1976), we assume that 16 percent of urban families and 85 percent of rural families used wood for firing, for all benchmark years. Light. Candles. UK 1839 Neild, 1839–1869 Brassey (1873). 1831 wholesale prices from Mulhall (1898). Soap. UK: 1849–1869 Brassey (1873), 1839 Porter (1850a)/Mulhall (1898), 1831 Mulhall (1898). 1841 for 1839, 1851 for 1849, 1861 for 1859. Clothing. Fearon (1818) provides a UK/US clothing benchmark for 1818. The Massachusetts Bureau of Statistics of Labor (1885) provides another benchmark for 1883. We also have estimates of retail prices in the 1830s and 1840s–1850s (provided in Table A.1). We take US prices for coats, shawls, bonnets, hats and stockings from the Massachusetts Bureau of Statistics of Labor (1885). The British prices were provided by Gregory Clark. Based on the retail price information and the independent benchmarks we estimate UK/US clothing prices at 50 for 1831 and 1839, 60 for 1849 and 1859 and 65 for 1869.
281
Relative Income
Domestic Service. We use Feinstein (1996) for Great Britain who gives wages plus board. Lebergott (1964) provides our US cash wages. For 1819–1869 we take board from Brady (1964, Table 19, p. 190). This refers to men. We adjust it to women using the information on board for men and women for 1869 and 1874 from Young (1875). We assume board is constant between 1831 and 1849. Transportation. UK: Hawke (1970) outside coach rates for 1831–1839 then average train rate per mile from Table 11.02, p. 48. US: 1831–1839 outside coach rates from Fishlow (1965); 1849–1869 average rate per train mile from Fishlow (1966, Table 1, p. 585). Investment Prices. Consists of benchmarks for construction and machinery and equipment. We estimate the construction benchmark with material prices and construction wages. The material prices are wholesale prices for pig iron, bar iron and copper. We take the UK prices from Mitchell (1971) and the Aldrich Report (United States Congress. Senate. Committee on Finance, 1893). US prices are from Cole (1938), Temin (1964) and the Aldrich Report (United States Congress. Senate. Committee on Finance, 1893). British construction wages are from Bowley (1901) and Feinstein (1996) while US wages are from Adams (1970), the Aldrich Report and Coehlo and Shepherd (1976). We base our equipment and machinery benchmark on iron prices. Government Prices. For government, we follow Gilbert and Kravis (1954) in constructing relative prices for civilian and defense expenditures. The civilian price index is a weighted average of relative nominal wages and the geometric mean of rent, fuel and light prices from the consumption index. The defense price index is a weighted average of relative nominal wages and the geometric mean of the construction, equipment and machinery, fuel and clothing relative prices. Expenditure Weights Consumption US weights. 1831 and 1839: Consumption weights for food, tobacco, fuel, light, soap and transportation from David and Solar (1977). 1830 weights used for 1831 and 1839. Alcohol, clothing and domestic service weights are from Lebergott (1996). Housing weights for 1831–1869 derived from US sectoral nominal GDP data derived above. Adjusted to housing as a share of consumption. 1839 housing weight used for 1831. 1849–1869: 1869 weights for food, tobacco and fuel from Kuznets (1962) used for 1849, 1859 and 1869. Light, clothing, domestic service and transportation are from Lebergott (1996). Housing as described above.
282
MARIANNE WARD AND JOHN DEVEREUX
British weights. 1831–1849: Individual consumption weights for food, housing, tobacco, fuel, light and soap are from Horrell (1996). Alcohol and clothing from Feinstein (1998), and domestic service and transportation are from Thomas (1995). 1859 and 1869: Consumption item weights are from Thomas (1995). Investment Weights: US: The shares of construction and equipment and machinery in total investment are from Gallman (2000, Table 13.1). UK: The shares of construction and equipment and machinery in total investment are from Feinstein and Pollard (1988, Appendix Table III3). Overall weights. US: Investment as a share of GDP from Gallman (2000, p. 39). Government spending as a share of GDP is from Trescott (1960). Consumption is calculated as a residual. UK: The shares of consumption, investment and government spending in total GDP are from Crafts (1985) and Mitchell and Deane (1962). Table 1 provides the price data that we used to calculate UK/US price levels from 1831 to 1869. Note, that the food and rent price data are adjusted for urban/rural price differences. Table 2 provides the individual food item weights. UK/US Sectoral Benchmarks Benchmark Sectoral Labor Productivity. We calculate sectoral productivities using UK/US nominal value added per worker and UK/US sectoral price levels. US Nominal GDP at Factor Cost. GDP at factor cost from Gallman (1960) and Gallman and Weiss (1969). UK Nominal GDP at Factor Cost. GDP at factor cost from Mitchell (1988). US Sectoral GDP shares. 1860 and 1870 are from Gallman (1960, p. 43) and Gallman and Weiss (1969, Table A.1, p. 306). These estimates refer to 1859 and 1869. We project them to 1860 and 1870 etc. using growth rates in nominal GDP from Rhode (2002). UK Sectoral GDP shares. Valued-added shares are constructed from Feinstein (1972), Mathews, Feinstein, and Odling-Smee (1982) and Deane and Cole (1967). US Sectoral Labor Force. Gallman (1960) and Gallman and Weiss (1969) UK Sectoral Labor Force. Feinstein (1972).
283
Relative Income
UK/US Sectoral Price Levels Agriculture. UK/US Farm Gate Prices for wheat, barley, oats, hay, hops, potatoes, milk, beef, mutton and pigmeat and UK gross output weights for 1870. US prices: Towne and Rasmussen (1960, pp. 281–312). UK prices: Ojala (1952, pp. 193–208), where prices are projected to benchmark years using Sauerbeck wholesale prices. We use 1870 weights for 1860. We use price data from Clark (2004) to project the Ojala 1870 prices to 1860. Construction. We use the price index for construction described in Table 3. Trade. Trade is measured by volume of commodities moving through wholesale and retail trade. UK margins are from Jeffreys and Walters (1955) as reported in Feinstein (1972, p. 12). For the US, we adopted a comparable procedure using the margins from Barger (1955, Table 26, p. 92). Transportation. Fisher Ideal price index for price per passenger and per ton-mile. US prices: Fishlow (1966, Table 1, p. 585). UK prices: Hawke (1970) and Mitchell and Deane (1962, Table 5, p. 225). We calculate average passenger mile and average ton-mile rates from Hawke (1970). Mining, Manufacturing and Public Utilities. Tables 1, 3 and 5 from Broadberry and Irwin (2004). Labor productivity for public utilities is set equal to transport.
284
Table A.1. Item
Units
1831
1831
1839
1839
1849
1849
1859
1859
1869
1869
in US$
GB
US
GB
US
GB
US
GB
US
GB
US
per lb per lb per 100 lbs per lb per lb per lb Per lb Per US gall Per lb Per lb Per lb Per lb Per lb (US) ton per lb per lb
0.05 0.05 1.24 0.12 0.12 0.11 0.14 0.23 0.25 0.16 1.34 0.43 0.16 8.91 0.13 0.14
0.03 0.02 0.36 0.05 0.04 0.05 0.05 0.19 0.12 0.06 0.48 0.12 0.08 2.80 0.13 0.10
0.05 0.06 1.25 0.14 0.13 0.12 0.17 0.21 0.25 0.15 1.50 0.47 0.17 6.67 0.14 0.12
0.03 0.03 0.67 0.08 0.05 0.10 0.10 0.19 0.24 0.09 0.58 0.13 0.08 3.72 0.16 0.09
0.04 0.04 1.20 0.14 0.13 0.13 0.18 0.20 0.24 0.13 1.04 0.36 0.11 4.67 0.12 0.11
0.04 0.02 1.42 0.08 0.04 0.08 0.08 0.19 0.20 0.08 0.42 0.12 0.07 2.90 0.13 0.07
0.04 0.04 1.23 0.13 0.15 0.12 0.16 0.20 0.25 0.15 0.98 0.33 0.11 4.84 0.14 0.09
0.04 0.03 1.10 0.08 0.08 0.09 0.09 0.20 0.18 0.11 0.62 0.15 0.08 3.91 0.16 0.11
0.05 0.05 1.62 0.21 0.22 0.19 0.27 0.31 0.36 0.23 0.97 0.43 0.12 5.99 0.13 0.13
0.06 0.04 1.40 0.13 0.13 0.15 0.15 0.28 0.29 0.19 1.11 0.28 0.14 6.30 0.16 0.13
MARIANNE WARD AND JOHN DEVEREUX
Bread Wheat Flour Potatoes Beef Mutton Pork Bacon Milk Butter Cheese Tea Coffee Sugar Fuel, coala Light, candles Soap
GB and US Price Data.
a
per month per month per US gall per US gall per lb per week each each each each Pair Costs per mile per UK ton per UK ton per lb per week
2.07 2.45 3.31 3.43 0.25 0.22 na Na 0.98 0.14 2.18 3.75 4.79 13.00 1.90 3.38 1.04 2.16 1.11 2.88 0.38 0.46 0.08 0.09 21.87 36.86 30.38 92.00 na 0.24 0.97b 1.98b
2.48 2.69 2.66 2.54 3.97 3.78 4.26 3.57 0.26 0.34 0.25 0.34 na na na na 1.10 0.24 0.96 0.25 2.61 3.75 2.63 3.48 na na 4.23 6.43 na na 1.68 4.37 na na 1.04 2.16 na na 0.69 1.49 na na 0.41 0.34 0.08 0.09 0.03 0.03 21.83 35.05 10.94 22.75 51.15 106.88 30.06 67.50 na 0.25 0.19 0.24 4.20 7.87 4.14 8.85
2.96 4.73 0.26 na 1.05 3.12 na na na na na 0.03 12.68 63.27 0.23 4.75
3.36 4.38 0.34 na 0.30 3.74 na na na na na 0.02 23.38 60.00 0.26 8.85
4.60 7.37 0.34 6.77 1.46 3.75 na na na na na 0.03 15.73 48.67 0.16 7.19
5.26 7.09 0.40 3.66 0.30 5.60 Na Na Na Na Na 0.03 40.63 81.66 0.28 12.64
Relative Income
Rent, four rooms Rent, six rooms Beer Spirits Tobacco Dom service wages Coats Shawls Bonnets Hats Stockings Transportation Pig iron Bar iron Copper Construction wages
US Prices are for wood and coal, quoted per coal equivalent. Details are in the appendix. Daily wages.
b
285
286
MARIANNE WARD AND JOHN DEVEREUX
Table A.2.
US and UK Individual Food Item Weights, 1831–1869. 1831
1839
1849
1859
1869
0.027 0.210 0.043 0.201 0.056 0.131 0.033 0.017 0.033 0.002 0.052 0.073 0.123
0.027 0.210 0.043 0.201 0.056 0.131 0.033 0.017 0.033 0.002 0.052 0.073 0.123
0.026 0.163 0.083 0.161 0.045 0.105 0.026 0.068 0.131 0.008 0.036 0.050 0.099
0.026 0.163 0.083 0.161 0.045 0.105 0.026 0.068 0.131 0.008 0.036 0.050 0.099
0.026 0.163 0.083 0.161 0.045 0.105 0.026 0.068 0.131 0.008 0.036 0.050 0.099
0.224 0.148 0.056 0.075 0.067 0.051 0.083 0.079 0.051 0.034 0.053 0.007 0.073
0.224 0.148 0.056 0.075 0.067 0.051 0.083 0.079 0.051 0.034 0.053 0.007 0.073
0.224 0.148 0.056 0.075 0.067 0.051 0.083 0.079 0.051 0.034 0.053 0.007 0.073
0.135 0.067 0.024 0.120 0.096 0.073 0.119 0.117 0.083 0.050 0.045 0.006 0.064
0.135 0.067 0.024 0.120 0.096 0.073 0.119 0.117 0.083 0.050 0.045 0.006 0.064
US weights Bread Wheat flour Potatoes Beef Mutton Pork Bacon Milk Butter Cheese Tea Coffee Sugar GB weights Bread Wheat flour Potatoes Beef Mutton Pork Bacon Milk Butter Cheese Tea Coffee Sugar
