No title

Journal of International Economics 83 (2011) 1–13 Contents lists available at ScienceDirect Journal of International E...

Author: C. Engel | R.W. Staiger

20 downloads 577 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Journal of International Economics 83 (2011) 1–13

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

Trade, markup heterogeneity and misallocations Paolo Epifani a, Gino Gancia b,⁎ a b

Department of Economics and KITeS, Università Bocconi, via Sarfatti 25, 20136, Milano, Italy CREI, Universitat Pompeu Fabra and CEPR, Ramon Trias Fargas 25-27, 08005, Barcelona, Spain

a r t i c l e

i n f o

Article history: Received 15 December 2008 Received in revised form 6 August 2010 Accepted 11 October 2010 Available online 17 October 2010 JEL classiﬁcation: F12 F15 Keywords: Markups Dispersion of market power Procompetitive effect Trade and welfare

a b s t r a c t Markups vary widely across industries and countries, their heterogeneity has increased overtime and asymmetric exposure to international trade seems partly responsible for this phenomenon. In this paper, we study how the entire distribution of markups affects resource misallocation and welfare in a general equilibrium framework encompassing a large class of models with imperfect competition. We then identify conditions under which trade opening, by changing the distribution of markups, may reduce welfare. Our approach is novel both in its generality and in the emphasis on the second moment of the markup distribution. Two broad policy recommendations stand out from the analysis. First, whenever there is heterogeneity in markups, be it due to trade or other distortions, there is also an intersectoral misallocation, so that the equilibrium can be improved upon with an appropriate intervention. This suggests that trade liberalization and domestic industrial policy are complementary. Second, ensuring free entry is a crucial precondition to prevent adverse effects from asymmetric trade opening. © 2010 Elsevier B.V. All rights reserved.

“What is relevant for the general analysis is not the sum of individual degrees of monopoly but their deviations ” A.P. Lerner

1. Introduction Monopoly power varies widely across industries and ﬁrms. Data on 4-digit US manufacturing industries show that Price-Cost Margins (PCMs), a common measure of markups, range from 1% in the ﬁrst percentile of the distribution, to 60% in the 99th percentile. Crosscountry evidence suggests these asymmetries to be even larger in less developed economies. Moreover, monopoly power varies systematically with exposure to international competition. For instance, the average PCM is a slim 13% in US manufacturing, producing goods that are typically traded, versus a fat 33% in nontraded business sector services. The conventional wisdom is that, as the process of globalization continues, competition among ﬁrms participating in international markets will intensify, thereby alleviating the distortions associated with monopoly power. The latter presumption is not however granted, because the exact mapping between the economywide distribution of markups and the extent of misallocations is not an obvious one. The textbook partial-equilibrium analysis of the deadweight loss from monopoly seems to imply that market power, by raising prices ⁎ Corresponding author. E-mail addresses: [email protected] (P. Epifani), [email protected] (G. Gancia). 0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.10.005

above marginal costs, is always distortionary. Yet, this reasoning neglects the fact that, as pointed out by Lerner (1934) and Samuleson (1947), in general equilibrium misallocations depend on relative rather than absolute prices. If all prices incorporate the same markup, Lerner noted that relative prices would signal relative costs correctly and, absent other sources of inefﬁciencies, would lead to the optimal allocation. This suggests that distortions only come from the dispersion of market power across ﬁrms and industries. The implications of this principle are far-reaching. For instance, it implies the seemingly paradoxical result that an increase in competition in industries with below-average markups, such as those producing tradeable goods, is deemed to amplify monopoly distortions. It also raises warnings that the increased heterogeneity in observed measures of market power across industries may indicate growing misallocations. Yet, when market power is coupled with free entry, so that markups affect the equilibrium number of ﬁrms, new welfare effects arise. For instance, when ﬁrms produce differentiated products and consumers like variety, it is desirable that markups be high enough to induce the socially optimal level of entry. The aim of this paper is to study how the economy-wide distribution of markups distorts the allocation of resources in a general framework that encompasses a large class of models of imperfect competition. In doing so, we revisit and qualify the Lerner condition that markups should be uniform across industries and illustrate when and how the level and dispersion of markups matter. Our second goal is to relate the degree of monopoly power in an industry to the presence of foreign competition and to study how trade can affect welfare by changing the dispersion of market power.

2

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

In particular, we are interested in ﬁnding under what circumstances asymmetric trade liberalization may turn out to be welfare reducing. To this end, we build a model with a continuum of industries that are heterogeneous in both costs and demand conditions. Firms may produce homogenous or differentiated goods and entry may or may not be restricted. By comparing the market equilibrium with the one chosen by a benevolent social planner, we identify the misallocations due to the entire markup distribution. The results crucially depend on the assumptions about entry. If entry is restricted, we conﬁrm Lerner's principle that markup symmetry across industries and ﬁrms is a necessary and sufﬁcient condition for efﬁciency. Whenever this condition is violated, there is an intersectoral misallocation whereby relatively less competitive industries underproduce and relatively more competitive ones overproduce relative to the socially optimal quantity. Perhaps surprisingly, we ﬁnd the extent of this misallocation to be stronger when the elasticity of substitution between industries is high and we show that its welfare cost may be quantitatively signiﬁcant. We also show that trade liberalization affecting only some industries may have adverse welfare effects when it raises markup heterogeneity. In other words, contrary to the conventional wisdom, procompetitive losses from trade are possible. With free entry, the level of markups matters too and the Lerner condition about symmetry turns out to be necessary, but not sufﬁcient, for efﬁciency. Moreover, contrary to the previous case, we show that in general there exists no markup distribution capable of replicating the ﬁrst best allocation. This means that policy interventions aimed at controlling prices only are not enough to correct all the distortions and other instruments, such as subsidies, are needed. Moreover, we ﬁnd that some heterogeneity in markups, despite the misallocation it induces, may well be welfare improving. Finally, we show that free entry makes procompetitive losses from trade unlikely, even when trade increases markup dispersion. Two general policy recommendations stand out from our analysis. First, whenever there is heterogeneity in markups, be it due to trade, regulations or differential ability to collude across sectors, there is an intersectoral misallocation. Industries with above-average markups always underproduce (either in terms of output per ﬁrm or of product variety), so that the equilibrium can be improved upon with an appropriate intervention. This also suggests that trade liberalization and domestic industrial policy are complementary. Second, free entry is an important condition to prevent asymmetric trade liberalization from having possibly adverse welfare effects. A novel implication of our ﬁndings is that the relative competitiveness of the industries affected by trade liberalization matters. This observation should be taken into account when designing trade policy. Interestingly, our analysis may also help rationalize the often heard concern that trade may be detrimental in countries (especially the less developed ones) where domestic markets are not competitive enough. The reason is not that domestic ﬁrms are unable to survive foreign competition (as emphasized by the infant-industry theory), but rather that international competition may inefﬁciently increase asymmetries across industries in the economy. This paper makes contact with three strands of literature. The ﬁrst studies monopoly distortions in general equilibrium and includes classics such as Lerner (1934), Samuleson (1947), Dixit and Stiglitz (1977), but also more recent works by Neary (2003), Koeniger and Licandro (2006), Bilbiie et al. (2006) and others. Although all these papers made important contributions, they all present a collection of special cases. Our approach is more general in modelling preferences, imperfect competition (with and without free entry) and sectorial asymmetries. We believe that such a uniﬁed framework is key to understanding how monopoly distortions interact with more speciﬁc modelling assumptions. Second, this paper is related to the literature studying the welfare effects of trade in models with imperfect competition, including the works of Brander and Krugman (1983), Helpman and Krugman

(1985) and more recently Eckel (2008). The observation that, in the presence of distortions, trade might have adverse welfare effects is an application of the second-best theory and goes back to Bhagwati (1971) and Johnson (1965).1 Yet, what we ﬁnd more interesting is the more speciﬁc insight, so far neglected, that trade can affect welfare by changing the cross-sectoral dispersion of market power. Third, this paper relates to a recent literature on the macroeconomic effects of misallocations. Noteworthy contributions by Banerjee and Duﬂo (2005), Restuccia and Rogerson (2008) Hsieh and Klenow (2009), Song et al. (forthcoming) and Jones (2009) provide striking evidence that wedges distorting the allocation of resources between ﬁrms and industries within a country are quantitatively very important in explaining low aggregate productivity, particularly in less developed economies. Yet, they leave to future research the task of identifying the origin of such wedges. Our paper contributes to this line of investigation by studying how asymmetries in market power, that appear to be especially large in poor countries, may be one source of misallocations.2 The paper is organized as follows. Section 2 documents a number of little known stylized facts that motivate our analysis. Section 3 builds a general theoretical framework that encompasses the most popular models of imperfect competition. Section 4 studies the welfare effects of markup dispersion and trade when entry is restricted. Section 5 extends the analysis to the case of free entry. Section 6 concludes. 2. Motivating evidence In this section, we document a number of little known stylized facts that motivate our theoretical investigations: (1) markups vary widely across sectors and their dispersion has increased overtime; (2) asymmetric exposure to trade seems to be a likely explanation for the rise in markup heterogeneity; (3) markup asymmetries are systematically related to the level of economic development, with less asymmetries in wealthier countries. Following a vast empirical literature (see, e.g., Roberts and Tybout, 1996; Tybout, 2003; Aghion et al., 2005), we use price-cost margins (PCMs) as a proxy for market power.3 To compute them, we draw production data from the OECD STAN database,4 the CEPII ‘Trade, Production and Bilateral Protection’ database5 and the NBER Productivity database by Bartelsman and Gray. The latter is the most comprehensive and highest quality database on industry-level inputs and outputs, covering roughly 450 US manufacturing (4-digit SIC) industries for the period 1958–1996. Price-cost margins are computed as the value of shipments (adjusted for inventory change) less the cost of labor, capital, materials and energy, divided by the value of

1 Bhagwati (1971) and Johnson (1965) where the ﬁrst to argue that trade can lower welfare if it exacerbates an existing distortion. Later studies have examined sufﬁcient conditions for positive gains from trade in the presence of various distortions. See, for example, Eaton and Panagariya (1979) and references therein. 2 Basu and Fernald (2002) stress the distinction between aggregate productivity and aggregate technology and recognize that technology improvements could reduce welfare if they lead sectors with smaller-than-average markups to increase input use. Yet, they do not pursue this idea, nor they relate it to trade liberalization. Gancia and Zilibotti (2009) shows how markup asymmetries may also distort the development of technology in dynamic models with innovation. 3 An important advantage of PCMs is that they can vary both across industries and overtime. An alternative approach would be to estimate markups from a structural regression a là Hall (1988). One problem with this approach is that, to estimate markups across industries or over time, either the time or industry dimension is to be sacriﬁced, implying that markups have to be assumed constant overtime or across industries. 4 This database allows to compute PCMs for broad aggregates of traded and nontraded industries for a sample of OECD countries. 5 This dataset is based on information from the World Bank, the OECD and the UNIDO. It allows to compute PCMs across broad manufacturing aggregates for a sample of developed and developing countries.

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

3

Fig. 1. Trade openness across US industries.

shipments.6 Capital expenditures are computed as (rt + δ)Kit − 1, where Kit − 1 is the capital stock, rt is the real interest rate and δ is the depreciation rate.7 As a proxy for trade exposure at the industry level, we use the openness ratio, deﬁned as the value imports plus exports, taken from the NBER Trade database by Feenstra, divided by the value of shipments. 2.1. Markup heterogeneity across sectors and overtime We start by showing asymmetries in markups across broad sectorial aggregates. Using economy-wide data for the US in 2003 (from the OECD dataset), we ﬁnd that the average PCM equals 33% in the business sector services (mostly nontraded industries), 28% in agriculture (a heavily protected industry) and 13% in manufacturing. As for services, the average PCM equals 24% in the transport and storage industry, 28% in post and telecommunication, 38% in ﬁnance and insurance, 48% in electricity, and reaches a peak of 66% in real estate activities. Interestingly, in the renting of machinery and equipment industry, selling nontraded services, the average PCM equals 41.5%, whereas in the machinery and equipment industry, producing traded manufacturing goods, the average PCM is 9.5%. These huge asymmetries in price-cost margins between traded and nontraded industries immediately suggest that markups may crucially depend on the degree of tradeabililty of an industry's output, and hence that asymmetric exposure to international competition may be an important determinant of markup heterogeneity across industries. Perhaps surprisingly, exposure to international competition varies dramatically also among manufacturing industries. Fig. 1 reports the time evolution of the openness ratio for selected 2-digit SIC industries within US manufacturing. Note that, at one extreme, the leather industry has increased its trade share from 4% in the late 50s to 230% in the mid 90s. Other industries, such as miscellaneous products or apparel, show a similar upward trend in the trade share. At the other extreme, however, there are industries, such as printing, fabricated 6 Due to data availability, we do not net out capital expenditures and inventory change when using the OECD and CEPII datasets. 7 The US real interest rate, drawn from the World Bank-World Development Indicators, has a mean value of 3.75% (with a standard deviation of 2.5%) over the period of analysis. As for the depreciation rate, the values for δ used in the empirical studies generally vary from 5% for buildings to 10% for machinery. We choose a value of 7%, implying that capital expenditures equal, on average, roughly 10% of the capital stock.

metal products or food, whose openness ratio has increased by only a few percentage points over the past 40 years. More generally, when considering the entire distribution of the trade share across 4-digit manufacturing industries, we ﬁnd that the openness ratio increased by only 6 percentage points in the ﬁrst quintile of the distribution (from 1.2% in 1958 to 7.2% in 1994), and by more than 47 percentage points in the fourth quintile of the distribution (from 10.2% to 57.7%). These ﬁgures suggest that, by affecting some industries more than others, trade opening may have increased asymmetries in market power. Figs. 2 and 3 provide suggestive evidence consistent with this conjecture. The former reports the evolution of the standard deviation of PCMs across 450 US manufacturing industries (broken line) and the average trade openness of the same industries (solid line) over the periods 1959–96 and 1958–94, respectively. It is immediate to see that, starting in the mid 70s, the dispersion of PCMs shows a relentless increase. Moreover, the standard deviation of PCMs and the average openness chase each other closely. The simple correlation between the two series equals 0.90 (0.40 after removing a linear trend). In Fig. 3, we replace the ﬁrst moment of the openness ratio with its second moment, again across 4-digit industries. Note that the standard deviation of trade openness closely follows the standard deviation of PCMs; the simple correlation between the two series is again very high, as it equals 0.88 (0.45 for the detrended series). Thus, a ﬁrst look at the data suggests that markup heterogeneity has increased overtime and that growing asymmetries in trade exposure may be partly responsible for it.

2.2. Trade and markup heterogeneity We now look for more systematic evidence on the link between trade openness and the dispersion of market power. To this purpose, we exploit information contained in the cross-sectional and temporal variation in the NBER datasets 8 to construct the following timevarying, industry-level proxies for the dispersion of markups and trade openness: for each 3-digit SIC industry, we compute the standard deviation of PCMs and the standard deviation of the 8 We focus on US manufacturing because of the high level of disaggregation of industry data on sales and costs provided by the NBER dataset. This should provide a lower bound for the effects we aim to illustrate, as international trade may raise the dispersion of markups also by increasing asymmetries between manufacturing and service industries.

4

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

Fig. 2. Openness and cross-industry markup heterogeneity.

Fig. 3. Asymmetries in trade exposure and markup heterogeneity.

openness ratio among the 4-digit industries belonging to it. Next, we run Fixed-Effects regressions of the former on the latter to test whether markup heterogeneity increases systematically in those 3digit industries where trade exposure becomes more asymmetric. The main results are reported in Table 1. In column 1, we run a univariate regression of the standard deviation of PCMs on the standard deviation of the openness ratio and ﬁnd that the two variables are strongly positively correlated, with a t-statistic around 8. In column 2, we add a full set of time dummies to control for spurious correlation due to time effects, e.g., the deregulation of the US economy initiated by the Carter Administration in the mid '70s. The coefﬁcient of the standard deviation of openness is somewhat reduced but is still very precisely estimated, with a t-statistic of 5. The cross-industry dispersion of price-cost margins may also be affected by technological characteristics. Although Fixed-Effects estimates implicitly account for time-invariant technological heterogeneity, our results may still be driven by asymmetric technical change. Hence, to control for variation in industry technology, in

column 3 we add the standard deviation (again within 3-digit SIC industries) of total factor productivity (TFP), skill-intensity (H/L) and capital-intensity (K/Y).9 While these controls are generally signiﬁcant, they leave the sign and signiﬁcance of our coefﬁcient of interest unaffected. In column 4, we run a most severe test by adding a full set of sector-speciﬁc linear trends. Notwithstanding the loss of identifying variance, the results are unchanged. In column 5, we add the average value of all RHS variables to check whether the correlation between the second moments of PCMs and openness is driven by variation in the ﬁrst moment of our covariates. It is not. Interestingly, the coefﬁcient of average openness is small and insigniﬁcant, which suggests that the strong positive correlation between average openness and the standard deviation of PCMs illustrated in Fig. 2 was mediated by the induced increase in the 9 Our measure of TFP is TFP5, from the Bartelsman and Gray's database. Skillintensity is proxied by the ratio of non-production to production workers, and capitalintensity by plant and equipment per unit of output.

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

5

Table 1 Fixed-effects regressions for the standard deviation of PCMs. Dependent variable: standard deviation of PCMs within 3-digit SIC industries.

St. dev. openness

1

2

3

4

5

6

7

0.016*** [0.002]

0.010*** [0.002]

0.010*** [0.002] 0.041*** [0.008] 0.008 [0.007] 0.001*** [0.000]

0.012*** [0.002] 0.063*** [0.009] 0.031*** [0.008] −0.000* [0.000]

0.012*** [0.003] 0.072*** [0.010] 0.032*** [0.010] −0.000** [0.000] 0.000 [0.011] −0.033*** [0.010] −0.000 [0.013] 0.000** [0.000]

0.012** [0.005] 0.072*** [0.009] 0.032*** [0.010] −0.000** [0.000]

YES YES 3456 96 0.39

YES YES 3456 96 0.39

0.011*** [0.003] 0.072*** [0.010] 0.035*** [0.010] −0.000*** [0.000] −0.002 [0.011] −0.005 [0.011] −0.006 [0.013] 0.000 [0.000] −0.136*** [0.024] YES YES 3456 96 0.41

St. dev. TFP St. dev. H/L St. dev. K/Y Average openness Average TFP Average H/L Average K/Y

−0.033*** [0.010] −0.000 [0.013] 0.000** [0.000]

Average PCM Time dummies Industry-speciﬁc trends Observations # 3-digit SIC industries R-squared (within)

3456 96 0.03

YES

YES

3456 96 0.11

3456 96 0.14

YES YES 3456 96 0.39

Notes: Fixed-Effects (within) estimates with robust standard errors in parentheses. ***,**,* = signiﬁcant at the 1, 5 and 10-percent levels, respectively. The mean and standard deviation of all variables is computed within 3-digit SIC industries. In column 6, estimation is by Instrumental Variables, with the second moment of openness instrumented with its ﬁrst moment. Data sources: NBER Productivity Database (by Bartelsman and Gray) and NBER Trade Database (by Feenstra).

standard deviation of openness. In column 6, we therefore instrument the standard deviation of the openness ratio with its mean value to see how the rise in the second moment of openness attributable to a rise in its ﬁrst moment affects the dispersion of markups. Estimation is by Two-Stage Least Squares. In the ﬁrst stage regression for the standard deviation of openness, not reported to save space, average openness is found to be a strong instrument for its standard deviation, with a coefﬁcient of 1.65 and a t-statistic of 11.5. In the second stage regression, the coefﬁcient of the standard deviation of openness is instead the same as in the OLS regression, with a slightly larger standard error. These results are consistent with the idea that globalization increases asymmetries in trade exposure, which in turn increase markup heterogeneity across industries. Finally, in column 7 we also control for the average PCM. Although this variable is obviously endogenous, including it ensures that the

second moment of the distribution of PCMs is not mechanically driven by variation in the ﬁrst moment. The size and signiﬁcance of the coefﬁcient of interest are unaffected. Moreover, the coefﬁcient of the average PCM is negative and signiﬁcant, consistent with the idea that procompetitive forces may have induced a simultaneous fall in average markups and an increase in their dispersion across industries. 2.3. Economic development and markup heterogeneity Finally, we show how the dispersion of markups is correlated with the level of economic development. To this purpose, we have computed the standard deviation of PCMs across three-digit ISIC manufacturing industries for a sample of 49 countries in the year 2001 (from the CEPII dataset). In Fig. 4, we plot the log standard deviation of PCMs against the log of real per capita GDP. Note that higher-income

Fig. 4. Economic development and cross-industry markup heterogeneity.

6

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

Fig. 5. Economic development and markup asymmetries between manufacturing and services.

countries are characterized by a signiﬁcantly lower dispersion of PCMs. This stylized fact is even stronger when considering asymmetries in the PCMs between manufacturing and services. In Fig. 5, we plot the log difference between the average PCM in services and manufacturing for a sample of 22 OECD countries in the year 2002. Note, again, that more developed countries are characterized by much lower asymmetries in the PCMs. We thus conclude that misallocations due to asymmetries in market power seem potentially relevant for understanding economic performance. 3. General model of imperfect competition A preliminary step for a comprehensive analysis of the distortions caused by an entire markup distribution is to build a tractable multisector model of imperfect competition that is general enough. This is the goal of the present section. We start by presenting a convenient representation of preferences, technology and market structure that encompasses as special cases a large class of models used in the literature. Next, we will use this model as a workhorse to study three issues: (1) the misallocation arising in a market equilibrium, (2) when and how regulations affecting markups can replicate the ﬁrst best equilibrium and, ﬁnally, (3) the welfare effects of asymmetric exposure to international trade. In modeling trade, we will focus on a symmetric-country case that will allow us to discuss the procompetitive effect of liberalization. In the interest of clarity, however, we begin our investigation with the closed economy. 3.1. Preferences and technology We focus on economies that admit a representative agent whose utility function can be used for normative purposes. We assume that there is a unit measure of agents (implying that averages coincide with aggregates), each supplying one unit of labor inelastically.10 Preferences are given by the following CES utility function: h 1 i1 = α α W = ∫ Ci di ; 0

10

α∈ð−∞; 1;

where Ci is the sub-utility derived from consumption of possibly differentiated varieties produced in industry i ∈ [0, 1], and α governs the elasticity of substitution between industries, σ = 1/(1 − α). Maximization of (1) subject to a budget constraint yields relative demand: Pi = Pj

1−α Cj ; Ci

ð2Þ

where Pi and Pj denote the cost of one unit of consumption baskets Ci and Cj, respectively. To preserve tractability, we assume that varieties within a given industry are symmetric, so that in equilibrium each will be consumed in the same amount. This assumption is in line with our focus on betweenindustry rather than within-industry heterogeneity and is not essential.11 It is particularly useful in that it allows us to use a simple and general reduced-form representation for the sub-utility derived from consumption in a given industry. Speciﬁcally, Ci is given by: νi + 1

Ci = ðNi Þ

ci ;

ð3Þ

where ci is consumption of the typical variety in industry i and Ni is the number of available varieties, equal to the number of ﬁrms in industry i. The parameter νi in Eq. (3) captures the preference for variety and is allowed to vary across industries. From Eq. (3), a greater variety Ni is associated with higher utility whenever νi N 0. Given the price of a single variety pi, the industry price index Pi , equal to the minimum cost of one unit of Ci, can be found setting expenditure in industry i equal to the value of demand, PiCi =Nipici. Substituting (3) yields: −ν i

Pi = Ni

pi :

ð4Þ

Each variety is produced by a single ﬁrm. Firms are owned by the totality of consumers so that any positive proﬁts or losses are rebated, but the exact form of redistribution is irrelevant in our representative-agent

ð1Þ

We take labor supply as inelastic for simplicity. The effects of competition and trade when labor supply is elastic are extensively discussed in Bilbiie et al. (2006) Corsetti et al. (2007). In these models, imperfect competition also distorts the trade-off between work and leisure.

11 Extending our results to models featuring ﬁrm heterogeneity, such as Melitz (2003) and Melitz and Ottaviano (2008), would be an interesting exercise that we leave for future work. Unfortunately, combining between and within sector heterogeneity complicates substantially the analysis, making it convenient to study them separately.

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

economy. Production requires a ﬁxed cost fi and a marginal cost 1/φi in units of labor. Firms charge a price equal to a markup over the marginal cost: pi =

μ i ð·Þw ; φi

ð5Þ

where w is the wage rate and μi(d ) ≥ 1 denotes the markup function. In general, the equilibrium markup is a function of the price elasticity of demand perceived by each ﬁrm, i: μi =

1 −1 1− i

with i ≡−

∂lnyi ; ∂lnpi

where yi is production by a given ﬁrm. The perceived elasticity may in turn depend on the number of ﬁrms in an industry and/or the elasticity of substitution in consumption across goods. We impose the restriction i N σ, implying that goods are more substitutable within industries than between industries. The markup may also depend on regulations that affect market contestability. For example, there might be a competitive fringe of ﬁrms that can copy and produce any variety without incurring the ﬁxed cost, but at a higher marginal cost. The higher marginal cost may capture the lower expertise of outsiders, but can also depend on entry regulations that make production more costly for external competitors. In this case, ﬁrms may be forced to charge a limit price below Eq. (5) and equal to the marginal cost of the external competitors, in order to keep them out of the market. In what follows, we do not impose any restriction on the markup function so as to preserve generality. Moreover, to ease notation, we will denote the markup simply as μi, with the understanding that it represents a function rather than a parameter. Rather than providing a list of examples of markup functions, we just recall some of the most common reasons for markup heterogeneity. In models with differentiated products (νi N 0), these include: (1) cross-industry differences in the within-industry elasticity of substitution among varieties, as in monopolistic competition a lá Dixit–Stiglitz with a continuum of ﬁrms; (2) cross-industry differences in the (low) integer number of ﬁrms and a common elasticity of substitution, as in Dixit–Stiglitz with a discrete number of ﬁrms or in Atkeson and Burstein (2008). With homogeneous products (νi = 0), variable markups may instead result from: (3) Cournot or Bertrand competition, as in Epifani and Gancia (2006) and Bernard et al. (2003). In cases (2) and (3), more ﬁrms leads to higher competitive pressure and lower markups, i.e., ∂ μi/∂ Ni b 0. In sum, our framework can encompass the most common models of imperfect competition, describing environments where ﬁrms produce homogeneous or differentiated goods and compete in quantity or price.12 4. Restricted entry We consider now the case in which the number of ﬁrms per industry Ni is exogenously given, so that entry and exit are not allowed. Although free entry might be a reasonable assumption in many industries, entry restrictions are fairly common too, particularly in less developed countries. For instance, the number of active ﬁrms may depend on the presence of government regulations, such as licences. Restricted entry may also provide an adequate description of a short-run equilibrium in which entry has not taken place yet and ﬁxed costs are sunk, making exit never optimal. Be as it may, this case 12 See the working paper version, Epifani and Gancia (2009), for speciﬁc examples. Other models of imperfect competition that can be represented within our approach include monopolistic competition with translog demand in Feenstra (2003), the generalization of Dixit–Stiglitz preferences by Benassy (1998), Melitz and Ottaviano (2008), and models of price competition with differentiated products that use the “ideal variety” approach, such as Salop (1979), Lancaster (1979) and Epifani and Gancia (2006).

7

is useful to understand the effects of trade and monopoly power in a situation when ﬁrms make pure proﬁts. Note that when the number of ﬁrms is not a choice variable, the ﬁxed cost in production has no bearings on the efﬁciency property of the equilibrium. Therefore, without loss of generality, we simplify the exposition by setting fi =0. Recall also that proﬁts are rebated to consumers. 4.1. Market equilibrium We start by characterizing the laissez-faire equilibrium. Denote Li as the number of workers employed in industry i. By virtue of symmetry and the absence of ﬁxed costs (fi =0), production by a given ﬁrm is yi =φiLi/Ni. Then, imposing market clearing (yi =ci) into Eq. (3) we obtain: ν

Ci = φi Li ðNi Þ i :

ð6Þ

The allocation of labor across sectors can be found using Eqs. (2), (4), (5) and (6): Li = Lj

!α 1 μ j 1−α Φi 1−α ; μi Φj

ð7Þ

where Φi ≡ φiNνi i is a measure of aggregate “productivity” at the industry level, taking into account that consumption delivers a higher utility in industries where there are many ﬁrms and a strong preference for variety. As expected, whenever goods are grosssubstitutes (α N 0), more productive industries hire more workers. Further, for any ﬁnite value of α, more competitive industries (low μi) also employ more workers. Integrating Eq. (7) and imposing labor market clearing yields: 1

Li =

α

ðμ i Þα−1 ðΦi Þ1−α α : 1 1 α−1 1−α ∫ μj Φj dj

ð8Þ

0

Finally, substituting Eq. (8) into Eq. (6) and then into preferences Eq. (1), we obtain the utility of the representative agent: α α1 1 1 ∫ μ −1 Φi 1−α di i α 0 1 α : W = ∫ ðLi Φi Þ di = 1 1 0 α 1−α ∫ μ −1 di i Φi

ð9Þ

0

This is our welfare measure. Inspection of Eq. (9) immediately reveals that utility is homogeneous of degree zero in markups: multiplying all μ i by any positive constant leaves welfare unaffected. In other words, as originally argued by Lerner (1934), in this economy welfare is independent of the average markup.13 4.2. Social planner solution To illustrate the distortions that may arise in the market equilibrium, we now solve for the allocation that maximizes the utility of the representative agent, subject to the resource constraint 13 An important assumption behind this result is that labor supply is inelastic. The reason is that markups lower wages below the marginal product of labor (MPL) and thus distort the work-leisure decision. For example, in the case Φi = 1, ∀ i ∈ [0, 1] we

can show that: ð1−αÞ = α 1 −α = ð1−αÞ di b MPL = 1; w = ∫0 μ ðiÞ where the latter equality follows from the fact that, with Φi = 1, labor productivity is equal to one. The strength of this distortion depends upon the elasticity of labor supply. However, as already noted by Lerner, even in this setting a markup on leisure (or leisure goods) would restore his principle that only dispersion matters.

8

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

of the economy. This is equivalent to solving the following planning problem: 1 α 1 α max W = ∫ ðLi Φi Þ di ; 0

Li

ð10Þ

subject to the resource constraint: 1

∫ Li di = 1: 0

Taking the ratio of any two ﬁrst order conditions yields the optimal labor allocation: Li = Lj

Φi Φj

!

α 1−α

1−α α α 1 W = ∫ ðΦi Þ1−α di : 0

:

Proposition 1. When the number of ﬁrms is exogenous, welfare is homogeneous of degree zero in markups. A necessary and sufﬁcient condition to replicate the ﬁrst best allocation is that markups be identical across all industries. 4.3. The cost of heterogeneity: intersectoral misallocations If a uniform markup is sufﬁcient to replicate the optimal allocation, what is then the cost of asymmetric market power? From Eq. (8), it can be shown that the market equilibrium entails underproduction in industries where the markup is above the following productivityweighted average: 2 1 α 3α−1 1 ∫ μ j α−1 Φj 1−α dj7 6 0 μ ≡4 ; 5 α 1 1−α dj ∫ Φj and overproduction in industries where μ i b μ .14 Thus, the problem is one of an intersectoral misallocation whereby less competitive industries attract a sub-optimally low number of workers. This happens because high markups compress the wage bill. The welfare cost of the misallocation due to markup heterogeneity depends in an interesting way on the curvature of the utility function Eq. (1). To see this, suppose that the μi can be approximated by a lognormal distribution and, to isolate the effect of markup heterogeneity, consider the case Φi = Φ. Then, Eq. (9) can be rewritten as15: ð12Þ

showing that markup dispersion is more costly when goods are highly substitutable (high α). The effect of substitutability across goods on the monopoly distortion induced by asymmetric markups is not an obvious one. On the one hand, a high substitutability means that the cost of overproduction in some industries is small: indeed, this cost goes to zero as goods become perfect substitutes. On the other hand, Eq. (7) 14 Interestingly, this means that monopoly power is associated to overproduction in industries where 1 b μi b μ . Thus, the conventional wisdom that a monopolist always produces less than the socially optimal quantity turns out to be wrong. 15 To derive Eq. (12), recall that, if x~ log Normal, then:

n n2 varðlnxÞ : lnE x = nEðlnxÞ + 2

Assuming Φi to be log-normal and using the properties of lognormal distributions (twice), welfare becomes: lnW = lnEðΦÞ +

2α−1 varðlnΦÞ : 1−α 2

ð14Þ

Intuitively, welfare is an increasing function of average productivity, lnE(Φ). Perhaps more surprisingly, Eq. (14) shows that, despite the symmetry in preferences, the variance of productivity Φi becomes welfare increasing when α N 0.5, i.e., when the elasticity of substitution between goods is greater than two. This happens because, when α is sufﬁciently high, agents can easily substitute consumption from unproductive industries to high Φi ones.16 In the general case, welfare is a complex function of the entire distributions of both μi and Φi. Although it is difﬁcult to make more precise statements regarding the impact of a particular change in those distribution and their correlation, from Eq. (9) we can derive a simple formula that can be used to measure welfare given data on μi and Φi: ̲ α̲ ̅ + cov μ α ; Φ ̅ E Φ E μ α ̲ iα ; W = h ̲ E μ E Φ ̅ + cov μ ; Φ ̅

0

varðlnμ Þ ; 2ð1−α Þ

ð13Þ

ð11Þ

Comparing Eq. (11) with Eq. (7) we see immediately that, for any ﬁnite α, the decentralized equilibrium is Pareto-efﬁcient if and only if μi = μj, ∀ i, j ∈ [0, 1]. We summarize these results in the following proposition:

lnW = lnΦ−

shows that, for a given asymmetry in markups μi/μj, a high substitutability magniﬁes the misallocation of labor towards the more competitive industries. It turns out that the latter effect dominates, so that perhaps counter-intuitively a ﬂatter curvature of the utility function leads to a higher cost of markup dispersion. On the contrary, Eqs. (7), (11), and (12) show that, as we approach the Leontief case (α → − ∞), the intersectoral misallocation disappears. It is also interesting to observe that the distortion induced by markups differs fundamentally from distortions that manifest themselves as higher production costs. To see this, consider the case of no markup dispersion. When μi = μ, the welfare function becomes:

ð15Þ

̲ where Φ ̅ = ðΦi Þα = ð1−αÞ and μ = μ ðiÞ1 = ðα−1Þ . We summarize the main ﬁndings of this section in the following proposition: Proposition 2. Markup heterogeneity introduces an intersectoral misallocation, whereby industries with below-average markups overproduce, and industries with above-average markups underproduce. The extent and the cost of this misallocation are proportional to the elasticity of substitution between industries. 4.4. Procompetitive losses from trade We now open the model to international trade to show that, when entry is restricted, trade integration tightening competition in some industries may amplify monopoly distortions. Moreover, the effect can be so strong that an equilibrium with trade may be Pareto-inferior to autarky for all the countries. Although rather extreme, this example is illustrative of the neglected principle that trade can affect welfare by changing the cross-sectoral dispersion of market power. A noteworthy corollary is that the characteristics of industries affected by trade liberalization and particularly their competitiveness relative to the rest of the economy are important factors to correctly foresee the effects of globalization. In turn, the result that welfare may fall with trade liberalization is an application of second-best theory. As pointed 16 Interestingly, this also suggests that price dispersion may be beneﬁcial when it originates from technology and the elasticity of substitution is high enough. See also Jones (2009) on the role of complementarity in amplifying industry-level distortions.

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

h W=

ν α = ð1−α Þ i1 = α 1−τ + τ xM 1−τ + τ ðxMνα Þ1 = ð1−αÞ

:

ð16Þ

This expression shows that welfare is a function of the measure of traded industries, τ, the number of trading countries, M, and the markup asymmetry x between open and closed industries. Fig. 6 plots welfare as a function of τ for the case ν=0 (solid line) and the case νN 0 (broken line). In the ﬁrst case (corresponding for example to Cournot competition), the economy attains the same level of welfare in autarky (τ=0) and when trade is free in all industries (τ=1). For any intermediate case, an equilibrium with trade is Pareto inferior to autarky. The intuition for this result should be by now clear. When ν=0, there is no gain from consuming foreign varieties and the only effect of trade is to lower markups in industries exposed to foreign competition. In both extreme cases, τ=0 and τ=1, markups are uniform across industries and there is no distortion. As τ moves from zero to one, trade breaks this symmetry: it increases markup dispersion as long as open industries are a minority and lowers it afterwards. Moreover, when ν=0 and τ∈(0,1) it is easy to see that welfare declines with an increase in the number of trading countries, M. The reason, again, is that a larger number of international competitors increases the markup asymmetry between traded and nontraded industries. When ν N 0, consumers derive a higher utility from the possibility to buy foreign varieties. In our model, we can think of this variety effect as capturing any source of gains from trade that is independent of the procompetitive effect. As the ﬁgure shows, in this case an equilibrium with some trade might still be Pareto inferior to autarky when τ is low, for the gains from small volumes of trade might be too low to dominate the price distortion (this happening for low enough ν). However, when τ is large enough, the gains from variety will eventually dominate the (falling) cost of misallocations. With gains 17 The procompetitive effect of trade, whereby exposure to international competition reduces markups features prominently in Krugman (1979) and Melitz and Ottaviano (2008), among others. See Chen et al. (2009) for recent evidence.

1.10 1.05 1.00

W

out by Bhagwati (1971) and Johnson (1965), if trade induces a contraction of industries that were already underproducing compared to the optimum, it exacerbates an existing distortion and may thus lower welfare. To isolate the point we want to make, we adopt the following simplifying assumptions. First, we consider a world populated by M N 1 identical countries so as to abstract from specialization effects. Second, to remove any unnecessary heterogeneity, we normalize the number of ﬁrms in each country to one (Ni = 1), and set φi = 1 for all i . Third, we assume that in some industries goods can be freely traded, while in others trade costs are prohibitive. Accordingly, the unit measure of sectors is partitioned into two subsets of traded and nontraded industries, ordered such that industries with an index i ≤ τ ∈ [0, 1] are subject to negligible trade costs while the others, with an index i N τ, face prohibitive trade costs. This simple description of imperfect trade integration accords well with the evidence that trade volumes are high in some industries and very low in others. We consider two complementary aspects of international integration: (1) an increase in the range τ of traded industries and (2) an increase in the number M of trading partners. Finally, we assume the markup to be a negative function of the number of competing ﬁrms in a given industry. This immediately delivers the procompetitive effect of trade, as the number of ﬁrms is one in nontraded industries and M N 1 in the others.17 For convenience, we denote the relative markup in nontraded industries as x ≡ μ(1)/μ(M). Under our assumptions, x is greater than one and increasing in M. After some straightforward substitutions into Eq. (9), we obtain:

9

0.95 0.90 0.85 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

tau Fig. 6. Trade and welfare: solid line ν = 0, broken line ν N 0.

from trade of any sort, the equilibrium with full integration (τ = 1) must necessarily dominate autarky. Even when a high ν N 0 assures positive gains from trade, when liberalization increases markup dispersion, it generates or exacerbate the intersectoral misallocation discussed above. What can then be done to counteract this negative effect of market integration? We have seen that the ﬁrst-best solution is attained with a uniform markup. Thus, if trade lowers markups in some sectors, competition policy might be used to match the change in market power in nontraded sectors too. If competition policy cannot be used, the ﬁrst best solution can still be achieved by giving an appropriate subsidy to industries producing nontraded goods. Note also that the likelihood that trade is harmful increases with x and that positive gains from trade will surely materialize if an economy is perfectly competitive (x = 1). In other words, the potential for welfare losses is higher when domestic markets are not competitive enough and trade brings large asymmetries between industries selling in world markets and the rest of the economy. These considerations may be particularly relevant for less developed countries, suggesting that in some cases promoting competition may be a prerequisite to ensuring positive gains from trade. In sum: Proposition 3. With an exogenous number of ﬁrms, procompetitive welfare losses from trade are possible when trade increases markup dispersion. 4.5. Simple quantitative exercise The example discussed in the previous section was admittedly provocative. We now show, however, that a simple quantitative exercise suggests the welfare cost of markup heterogeneity to be potentially large when entry is restricted. To this end, we use our model, together with the evidence and the data discussed in Section 2, to compute the cost of markup dispersion across US manufacturing industries, relative to a ﬁrstbest allocation in which markups are instead uniform. Using Eq. (9), and denoting by WFB welfare in the ﬁrst-best allocation, we obtain: α 1 = α 1 −1 ∫ μ i Φi 1−α di 0 W = 1−α : α α 1 W FB 1 1 −1 α 1−α ∫ μ i Φi di⋅ ∫ Φi1−α di 0

ð17Þ

0

Computing Eq. (17) requires an empirical measure of the productivity index Φi = φiNvi i for each industry i. To build such measure, we proceed as follows. Eqs. (4) and (5) imply: Φi =

μiw : Pi

ð18Þ

10

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

From utility maximization, we obtain: Ci =

1 1 Pi α−1 W = Piα−1 E; P

Table 2 Welfare cost of markup dispersion.

α−1 α α 1 is the ideal price index where E is total expenditure, P = ∫ Piα−1 di 0

associated to Eq. (1), and the latter equality follows from choosing P as the numeraire. Eq. (19) allows us to express the unobserved industry price index Pi as a function of the observed expenditure share on an industry's products, θi ≡ PiCi/E: α−1

Pi = θi α :

ð20Þ

Finally, using Eq. (20) into Eq. (18) gives: 1−α

Φi = wμ i θi α :

Constant Φi

ð19Þ

ð21Þ

Note that W/W FB is homogeneous of degree zero with respect to all the Φi. Hence, without any loss of generality, we can disregard the factor w, which is constant across sectors, and calibrate Φi using data on markups, μ i, and expenditure shares, θi, for the US 4-digit SIC industries over the period 1959–1994. In particular, to compute μi, we use the deﬁnition of price-cost margins (PCMs) in Section 2 and set μi = (1− PCMi)− 1.18 θi, is instead computed as the value of an industry's shipments plus net imports, divided by the total expenditure on manufacturing goods, using again the NBER datasets. Calibrated in this way, Φi accounts for factors, such as parameters of technology but also preferences, that affect expenditure shares other than relative markups. Computing Φi and W/W FB also requires choosing a value for the elasticity of substitution in consumption among manufacturing goods, σ = 1/(1 − α). Although available estimates of σ vary widely across studies, most of them are in the range [2, 10] and a value around σ = 5 is most frequently used in quantitative exercises. We therefore set σ = 2, 5 and 10 (implying α = 0.5, 0.8 and 0.9) as benchmark cases. We then perform two distinct exercises. First, we assume the Φi to be constant overtime and compute their cross-section from Eq. (21) by taking, for each 4-digit industry, the mean value of μi and θi overtime. Then, we compute W/W FB over the period 1959–94 using Eq. (17). This exercise allows us to isolate the welfare loss induced by the change in the distribution of markups, holding constant preferences and other technological factors. The results are reported in the left panel of Table 2. For an intermediate value of the elasticity of substitution (σ = 5), utility falls by 3.6% in the period of analysis. For σ = 2, the welfare cost of the observed rise in markup dispersion is lower (1.3%), whereas it is much larger (8.1%) for σ = 10. Second, we allow the Φi to be time-varying and compute them year by year. We use these time-varying coefﬁcients again into Eq. (17) to obtain the value of W/W FB. This exercise provides information on the overall welfare cost of changes in the markup distribution taking into account that other parameters (such as tastes and technology) have also affected expenditure shares simultaneously. The results are in the right panel of Table 2, showing that in this case the welfare cost of markup dispersion is even larger. This indicates that the evolution of the exogenous parameters contained in the Φi

σ = 2 (α = 0.5) σ = 5 (α = 0.8) σ = 10 (α = 0.9)

Time-varying Φi

W/WFB (1959)

W/WFB (1994)

ΔW/WFB (94–59)

W/WFB (1959)

W/WFB (1994)

ΔW/WFB (94–59)

0.983 0.947 0.854

0.97 0.913 0.785

−1.3% −3.6% −8.1%

0.986 0.957 0.88

0.961 0.863 0.666

−2.5% −9.8% −24.3%

has contributed to amplify monopoly distortions. In particular, over the period of analysis, relative utility falls by 2.5% for σ = 2, by 9.8% for σ = 5, and by as much as 24.3% for σ = 10. In sum, these numerical exercises suggest that, abstracting from any effect that changes in the average PCMs may have had, the observed increase in markup heterogeneity can entail substantial welfare costs. How much of these costs can then be attributed to trade integration? Given that the impact of trade on markup dispersion can be ambiguous, the answer to this question is ultimately an empirical one. Although identifying and quantifying how trade has affected the markup distribution goes beyond the scope of the current paper, the evidence discussed in Section 2 does suggest that trade may be a major determinant of observed asymmetries in market power.19 We therefore conclude that markup heterogeneity matters for misallocations and that procompetitive losses from trade may be more than a theoretical curiosum in this class of models. 5. Free entry So far, ﬁrms are making positive proﬁts and barriers to entry prevent potential competitors from challenging incumbent ﬁrms and sharing the rents. Without those barriers, entry will take place until pure proﬁts are driven to zero. We now allow for this possibility with the aim of extending our results to a widely-used class of models where entry is free. This exercise will lead to remarkably different conclusions, which qualify some of Lerner's original statements. We start by presenting the market equilibrium and show that, contrary to the previous case, welfare is a function of the average markup too. Next, we compare it with the social planner solution and discuss the inefﬁciencies that arise in the decentralized equilibrium. Finally, we study the effect of trade between identical countries and argue that, while procompetitive losses are now unlikely, asymmetric trade liberalization may still exacerbate misallocations, thereby providing a rationale for policy intervention. 5.1. Market equilibrium We now reintroduce the ﬁxed cost of production, fi, deﬁned in units of labor, and let the number of ﬁrms vary so as to guarantee that each breaks even. In this way, in equilibrium all operating proﬁts are used to cover the ﬁxed cost: pi yi −

yi w = fi wi : φi

Substituting pi from Eq. (5) and rearranging gives: 18

Price-cost margins are deﬁned as the value of shipments minus costs divided by the value of shipments. Thus: PCMi =

pi yi −

yi =

φi f i : μ i −1

ð22Þ

w y = pi yi = 1−1 = μi ; φi i

where piyi is the value of a ﬁrm's shipments in industry i, and (w/φi)yi is its variable cost. Although in our simple model labor is the only variable cost, we also net out materials and capital expenditures in our empirical deﬁnition of price-cost margins. This avoids spurious variation in the PCMs due to variation in materials intensity and capital intensity across industries.

19 In particular, if we rerun the regressions in Table 1 computing the standard deviation of markups and other variables across all the 4-digit SIC industries, we obtain that the increase in trade openness (or its standard deviation) over the period 1959–94 can explain the entire observed increase in markup dispersion across US manufacturing industries. The results are available upon request.

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

As is well-known, the free-entry condition pins down uniquely ﬁrm size. Given ﬁrm size, the number of active ﬁrms must be proportional to the amount of labor employed in each industry. More precisely, the demand for labor in industry i is: Li = Ni

yi + fi : φi

Ni =

5.2. Social planner What are the distortions in the market equilibrium? Is markup asymmetry desirable or does it impose any welfare costs? To answer these questions, we now compute the allocation that maximizes the utility of the representative agent. A benevolent social planner chooses Ni and yi = ci so as to solve: 1 ν max W = ∫ Ni i

Substituting Eq. (22) we obtain: ð23Þ

Ci Pi = Cj Pj

−α Cj L = i; Ci Lj

0 1α 1−α ν φi Ni i μ j Li A : =@ νj Lj φj N μ i

ð24Þ

j

Note that Eq. (24) differs from the analogous condition in the model without free entry Eq. (7). The key reason is that, in the model with free entry, the entire industry revenue is used up to pay workers (both for the ﬁxed and variable production costs), while in the other case a fraction of revenue is captured by proﬁts.20 Combining Eqs. (22), (23) and ( 24) we obtain an equation linking the number of ﬁrms in any two industries to relative ﬁrm size and other exogenous parameters:

1−α−αν j Nj

ð25Þ

variety

0

+

0

yi + fi di = 1: φi

1 ν L = ∫ Ni i 0

f −saving zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ}|ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ ! ﬄ{ zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ}|ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{ 1 1 1−μ −1 1 α −1 i yi di = ∫ ν i ln di + ∫ lnφi μ i di: 0 0 fi

ð26Þ This equation shows that the level of markups has now both positive and negative direct effects on welfare. The ﬁrst term in Eq. (26) is increasing in μi and captures the fact that a high proﬁt margin stimulates entry, thereby raising the number of ﬁrms and welfare so long as variety has value (νi N 0). The second term in Eq. (26) is instead decreasing in μi and captures the fact that entry implies that more productive resources are taken by the ﬁxed costs. Thus, contrary to the restricted-entry case, when νi N 0 markups now pose a trade-off between diversity and ﬁxed costs at the industry level. 20 A second, less important, difference is that, in the model without entry we set the ﬁxed cost to zero. Of course, the equilibrium allocation of labor would coincide in the two models if we had the same ﬁxed cost and if the exogenous number of ﬁrms without entry happened to be equal to the equilibrium number of ﬁrms with entry.

+ 1

yi

α 1 = α 1 y di −λ ∫ Ni i + fi di−1 ; 0 φi

and the ﬁrst order conditions for an optimum are: ∂L 1−α α ðν = 0→W ðν i + 1Þyi Ni i ∂Ni ∂L 1−α ðνi = 0→W Ni ∂yi

+ 1Þα α−1 yi

+ 1Þα−1

=

=λ

yi + fi ; φi

λNi : φi

Substituting the second ﬁrst order condition into the ﬁrst yields: yi =

φi fi : νi

ð27Þ

Taking the ratio of the second ﬁrst order condition in industries i and j delivers: 1−α−ανi

This condition will turn out to be useful below. To understand the role of markups in the model with free entry, consider for the moment the simpler Cobb–Douglas case corresponding to α = 0. Then, using Eqs. (22), (23) and (24), we can write welfare as:

1 ν lnW = ∫ ln Ni i

∫ Ni

Ni

1−α yj φi μ j = : yi φj μ i

1−α−αν i

α 1 = α di ;

The Lagrangian for the above program is:

where the latter equality follows from the fact that, with a ﬁxed cost in units of labor and without extra-proﬁts, industry revenue equals the wage bill. Using Eq. (3), the market clearing condition ci = yi, Eqs. (22) and (23), we obtain:

Ni

yi

subject to the resource constraint: 1

Finally, to solve for Li, we manipulate the demand Eq. (2 ) to yield:

+ 1

0

Ni ;yi

μ i −1 L: fi μ i i

11

1−α−ανj

=

Nj

1−α yj φi : yi φj

ð28Þ

Comparing the optimal ﬁrm scale Eq. (27) to the market outcome Eq. (22), we see that the two coincide when: μ i −1 = ν i ;

ð29Þ

that is, when the markup is equal to the preference for variety, as in the Dixit–Stiglitz case. However, comparing Eq. (28) with Eq. (25), we also see that the optimal allocation of resources across industries requires μ i =μ j, that is, a uniform markup. When the preference for variety is unequal across industries (the most realistic case) these two requirements are incompatible and we thus have the following impossibility result21: Proposition 4. When entry is free and the preference for variety is heterogeneous across industries (νi ≠ νj for at least two i, j ∈ [0, 1]), there exists no markup distribution such that the market equilibrium replicates the ﬁrst-best allocation. These results show that, in general, a uniform markup across industries and ﬁrms is still a necessary condition for efﬁciency but, contrary to Lerner's original claim, it is not sufﬁcient anymore. The reason is that proﬁts now have a dual role: they affect the allocation of resources across industries and the equilibrium number of ﬁrms per industry. As we know from the previous section, avoiding intersectoral misallocations requires μi = μj. However, markups should also 21 Epifani and Gancia (2008) review some evidence suggesting that external economies due to love for variety differ across industries.

12

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

correctly signal the social value of entry and this requires higher proﬁt margins in industries where variety is more valuable. Finally, recall that the intersectoral misallocation tends to disappear as preferences approach the Leontief case. This is true in general and the model with free entry makes no exception. In fact, taking the limit of Eqs. (28) and (25) for α → − ∞ reveals that the two equations converge to the same condition: 1 + νi yj N = i1 + ν ; j yi Nj

which is of course equivalent to Ci = Cj. In this case, the market equilibrium converges to the ﬁrst-best allocation when μ i − 1 = νi . 5.3. Optimal competition policy, markup heterogeneity and welfare We now ask what is the markup distribution that maximizes the utility of the representative agent. In other words, we are interested in ﬁnding the constrained efﬁcient allocation that a planner can achieve by controlling markups and without using lump-sum transfers. To ﬁnd it, we use Eqs. (22) and (23) to rewrite the welfare function as: ν + 1 α 1 = α h iα 1=α 1 1 Li i μ i −1 νi ν +1 W= ∫ φi di = ∫ ðLi Þ i Φi di ; 0 0 fi μi ð30Þ νi + 1 νi 1 μ i −1 φi . Note that W is increasing in Φi. where now Φi ≡ μi fi Thus, maximizing Eq. (30) is equivalent to maximizing Φi industry by industry. Then, the ﬁrst order condition is: ∂Φi νi ν +1 = i = 0→ → μ i −1 = ν i ; μi μ i −1 ∂μ i

ð31Þ

22

which is identical to Eq. (29). Thus, it is optimal to let the markup reﬂect the social value of entry, irrespective of the intersectoral inefﬁciency. This means that, so long as νi ≠ νj, an increase in markup dispersion may be welfare improving if the resulting markup distribution gets closer to the one implied by condition (29). It also implies that computing the welfare cost of a given markup distribution becomes much harder, as it now requires data on νi. Finally, it is instructive to consider the welfare costs of markup heterogeneity in the special case of νi = 0, that is, when entry has no social value per se. This case is of particular interest because it serves as a metaphor for all models where proﬁts are dissipated in equilibrium through rent-seeking activities that are socially wasteful. When νi = 0, it is of course optimal to have μi = 1; yet, this may not be feasible. What is less obvious, instead, is the cost of markup dispersion. From Eq. (24) and ∫10 Li di = 1, we obtain: α 1−α φi μ −1 i Li = 1 α : 1−α ∫ φj μ −1 dj j 0

Substituting this into the welfare function (30) yields: W=

1

∫

0

α 1−α α −1 1−α φi μ i di ;

which takes the same form as Eq. (13). Remarkably, this means that markup heterogeneity is welfare improving when α N 0.5, precisely as we found for productivity heterogeneity. This result can be understood by noting that, in an equilibrium with free entry, the markup is 22 It follows immediately that there is no room for welfare-improving intervention on markups in the Dixit–Stiglitz case. Yet, this is admittedly not the most interesting case to study procompetitive effects.

nothing but the per unit equivalent of the ﬁxed cost fi. Thus, when entry is free but has no social value, the markup acts a pure cost for the economy and it affects welfare just as the marginal cost (1/φ) in the model of Section 4 did. We summarize the main ﬁndings in the following proposition: Proposition 5. With free entry, markup symmetry is a necessary, but not a sufﬁcient condition for efﬁciency. Markup heterogeneity always leads to an intersectoral misallocation, but does not necessarily lower welfare: it may be welfare improving when the preference for variety is heterogeneous across industries, or when variety has no value (νi = 0, i ∈ [0, 1]) and the elasticity of substitution is high (α N 0.5). 5.4. The procompetitive effect of trade, welfare and misallocations Are procompetitive losses from trade possible when entry is free? Does asymmetric trade liberalization introduce distortions that may be corrected by policy makers? To brieﬂy address these questions we now open the model to trade, as in Section 4.4. For simplicity, we focus on the Cobb–Douglas case, i.e., α = 0. We consider a world of M symmetric countries and we denote by Ni the number of ﬁrms per country in industry i. Consumption of a given traded variety in a given country becomes ci = yi/Mi, so that the industry consumption basket can be written as: νi + 1

Ci = ðNi Þ

ν

ðMi Þ i yi :

Substituting this into Eq. (1) and using Eqs. (22) and (23), we obtain our welfare measure: h i 1 1 ν −1 lnW = ∫ lnCi di = ∫ ln ðNi ⋅Mi Þ i φi μ i di: 0

0

ð32Þ

As before, we model imperfect market integration by allowing some industries to be closed to trade. In other words, we set Mi = 1 in the subset of nontraded industries. When the markup function is such that μi = f(Ni ⋅ Mi) with f′(⋅) b 0, as in Krugman (1979), then the procompetitive effect of trade is always welfare improving. To see this, note that if trade has to lower the markup in an industry, it should also increase the equilibrium number of ﬁrms. Both the fall in μi and the rise in Ni ⋅ Mi increase W. Interestingly, this is true even when variety has no value, νi = 0! Procompetitive losses from trade are still possible, but only when the fall in markups due to foreign competition is so strong as to reduce the equilibrium number of ﬁrms in an industry below the optimal level.23 While this is impossible when μi = f(Ni ⋅ Mi) with f′(⋅) b 0, this outcome is conceivable if trade liberalization affects the markup not through the number of ﬁrms. In conclusion, when there is free entry, procompetitive losses from trade seem much more unlikely. Thus, an important beneﬁt of entry is that it prevents some of the possibly large costs identiﬁed in Section 4. Yet, asymmetric trade liberalization that increases markup heterogeneity exacerbates the intersectoral misallocation of resources and opens the way to Pareto improving intervention. In particular, there might be excessive product diversity or ﬁrm output in traded industries that can be corrected with an appropriate subsidy to nontraded industries. We therefore conclude with the following proposition: Proposition 6. When entry is free and markups are a negative function of the number of ﬁrms in an industry (μi = f(Ni ⋅ Mi), f′(⋅) b 0, i ∈ [0, 1]), the procompetitive effect of trade is welfare improving, even when trade ampliﬁes misallocations due to markup dispersion and variety has no value. 23 Eckel (2008) provides conditions for this outcome. In other models, such as Melitz (2003) and Corsetti et al. (2007), trade liberalization increases welfare even when the number of varieties falls.

P. Epifani, G. Gancia / Journal of International Economics 83 (2011) 1–13

6. Conclusions Competition is imperfect in most sectors of economic activity. By exposing ﬁrms to foreign competition, trade is widely believed to help alleviate the distortions stemming from monopolistic pricing. While this argument is often well-grounded, it neglects that in general equilibrium pricing distortions depend on both absolute and relative market power, and that a trade-induced fall in markups brings about misallocations when it raises their variance. The latter event is more than a theoretical curiosity, for market globalization affects predominantly industries that are already relatively more competitive. At the same time, misallocations across ﬁrms and industries have recently been identiﬁed as a strikingly important factor behind cross-country differences in economic performance. Studying how such misallocations may be rooted in the economy-wide markup distribution and how this may interact with trade liberalization is therefore crucial for the design of optimal trade and competition policies and for a better understanding of the welfare effects of trade opening in the presence of market power. This was the aim of our paper. We now summarize what we view as the main results. When ﬁrm entry is restricted, we ﬁnd that markup heterogeneity entails signiﬁcant costs and that asymmetric trade liberalization may reduce welfare. With free entry of ﬁrms, instead, markup heterogeneity is not necessarily welfare reducing, although it generates an intersectoral misallocation that policy makers can correct. In this case, we also ﬁnd that a tradeinduced increase in competition is typically welfare increasing. Yet, if trade integration raises markup dispersion, the allocation of resources can be improved upon by subsidizing production in industries that remain relatively more protected. In this sense, trade liberalization and domestic industrial policy complement each other. More in general, our analysis has emphasized the neglected principle that, in order to correctly foresee the effects of trade and competition policy, the evolution of the economy-wide markup distribution has to be taken into account, and that whether entry is restricted or not makes an important difference. By focusing on special cases, the existing literature on the topic offers a partial view only. One goal of this paper was precisely to clarify the misconceptions that may arise when restricting the analysis to special cases. A major beneﬁt of our general framework is that it illustrates the exact role of alternative assumptions in shaping the relationship between competition, misallocations and welfare. We hope that such a uniﬁed framework may prove useful in studying other issues, such as the effects of competition on growth, and in guiding future empirical and quantitative work. In particular, while we have emphasized markup heterogeneity across industries, we think that extending the analysis to heterogeneity across ﬁrms may be at least as much important. Yet, accounting for it poses a new difﬁculty, in that it requires disentangling ﬁrm-level estimates of productivity from markups. We view this as a key challenge for future research. Acknowledgements We would like to thank Giancarlo Corsetti, Jonathan Eaton, Nezih Guner, Winfried Koeniger, Omar Licandro, Enrico Sette, Guido Tabellini, Jaume Ventura, Fabrizio Zilibotti and an anonymous referee for the helpful comments and discussion. We also thank participants at the EEA Annual Congress (Barcelona, 2009), ITSG Meeting (Cagliari, 2009), RIEF Meeting (Aix-an-Provence, 2009), ESSIM (Tarragona, 2008), XXXII Symposium of Economic Analysis (Granada, 2007), Universidad Carlos III de Madrid, Banco de España, the UPF Faculty Lunch, the Macro Workshop at IEW, Zurich University and the IADBELSNIT (Paris, 2006). All errors are our own. Gino Gancia acknowledges ﬁnancial support from the Barcelona GSE, the Government of Catalonia (2009SGR1157) and the ERC Grant GOPG 240989. Part of

13

this research was done while Gino Gancia was visiting as Research Associate the Chair of Political Economy and Macroeconomics of the IEW-University of Zurich. We thank the IEW for the hospitality. References Aghion, P., Bloom, N., Blundell, R., Grifﬁth, R., Howitt, P., 2005. Competition and innovation: an inverted u relationship. Quarterly Journal of Economics 120, 701–728. Atkeson, A., Burstein, A., 2008. Pricing-to-market, trade costs, and international relative prices. American Economic Review 98, 1998–2031. Banerjee, A., Duﬂo, E., 2005. Growth theory through the lens of development economics. In: Aghion, P., Durlauf, S. (Eds.), Handbook of Economic Growth (Amsterdam: North-Holland). Basu, S., Fernald, J.G., 2002. Aggregate productivity and aggregate technology. European Economic Review 46, 963–991. Benassy, J.P., 1998. Is there always too little research in endogenous growth with expanding product variety? European Economic Review 42, 61–69. Bernard, A., Eaton, J., Jensen, B., Kortum, S., 2003. Plants and productivity in international trade. American Economic Review 93, 1268–1290. Bhagwati, J., 1971. “The generalized theory of distortions and welfare. In: Bhagwati, J., et al. (Ed.), Trade balance of payments and growth (Amsterdam: North-Holland). Bilbiie, F., F. Ghironi and M. Melitz (2006). “Monopoly Power and Endogenous Variety in Dynamic Stochastic General Equilibrium: Distortions and Remedies,” mimeo. Brander, J., Krugman, P., 1983. A reciprocal dumping model of international trade. Journal of International Economics 15, 313–321. Chen, N., Imbs, J., Scott, A., 2009. The dynamics of trade and competition. Journal of International Economics 77, 50–62. Corsetti, G., Martin, P., Pesenti, P., 2007. Productivity, terms of trade and the home market effect. Journal of International Economics 73, 99–127. Dixit, A., Stiglitz, J., 1977. Monopolistic competition and optimum product diversity. American Economic Review 67, 297–308. Eaton, J., Panagariya, A., 1979. Gains from trade under variables returns to scale, commodity taxation, tariffs and factor market distortions. Journal of International Economics 9, 481–501. Eckel, C., 2008. Globalization and specialization. Journal of International Economics 75, 219–228. Epifani, P., Gancia, G., 2006. Increasing returns, imperfect competition and factor prices. Review of Economics and Statistics 88, 583–598. Epifani, P., Gancia, G., 2008. The skill bias of world trade. Economic Journal 118, 927–960. Epifani, P., Gancia, G., 2009. “Trade, markup heterogeneity and misallocations,” CEPR DP 7217. Feenstra, R., 2003. A homothetic utility function for monopolistic competition models, without constant price elasticity. Economics Letters 78, 79–86. Gancia, G., Zilibotti, F., 2009. Technological change and the wealth of nations. Annual Review of Economics 1, 93–120. Hall, R.E., 1988. The relations between price and marginal cost in U.S. industry. Journal of Political Economy 96, 921–947. Helpman, E., Krugman, P., 1985. Market structure and foreign trade. (Cambridge MIT Press), MA. Hsieh, C.T., Klenow, P., 2009. Misallocation and manufacturing TFP in China and India. Quarterly Journal of Economics 124, 1403–1448. Johnson, H.G., 1965. Optimal trade intervention in the presence of domestic distortions. In: Baldwin, R.E., et al. (Ed.), Trade, Growth and the Balance of Payments. RandMcNally, Chicago. Jones, C. (2009). “Intermediate Goods and Weak Links: A Theory of Economic Development,” mimeo Stanford. Koeniger, W., Licandro, O., 2006. On the use of substitutability as a measure of competition. Topics in Macroeconomics 6, Issue 1, Article 2. Krugman, P., 1979. Increasing returns, monopolistic competition and international trade. Journal of International Economics 9, 469–479. Lancaster, K., 1979. Variety, equity and efﬁciency. Columbia University, New York. Lerner, A.P., 1934. The concept of monopoly and the measurement of monopoly power. Review of Economic Studies 1, 157–175. Melitz, M., 2003. The impact of trade on intra-industry reallocations and aggregate industry productivity. Econometrica 71, 1695–1725. Melitz, M., Ottaviano, G., 2008. Market size, trade and productivity. Review of Economic Studies 75, 295–316. Neary, J.P., 2003. Globalization and market structure. Journal of the European Economic Association 1, 245–271. Restuccia, D., Rogerson, R., 2008. Policy distortions and aggregate productivity with heterogeneous establishments. Review of Economic Dynamics 11, 707–720. Roberts, M.J., Tybout, J.R., 1996. Industrial evolution in developing countries. Oxford University Press, Oxford. Salop, S.C., 1979. Monopolistic competition with outside goods. Bell Journal of Economics 10, 141–156. Samuleson, P., 1947. Foundations of Economic Analysis. (Cambridge Harvard University Press), MA. Song, Z., Storesletten, K., Zilibotti, F., forthcoming. Growing Like China. American Economic Review. Tybout, J.R., 2003. “Plant-and ﬁrm-level evidence on ‘new’ trade theories,”. In: Choi, E.K., Harrigan, J. (Eds.), Handbook of International Economics. Basil-Blackwell, Oxford.

Journal of International Economics 83 (2011) 14–26

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

Trade barriers and trade ﬂows with product heterogeneity: An application to US motion picture exports Gordon Hanson a,c,⁎, Chong Xiang b,c a b c

UC San Diego, United States Purdue University, United States NBER, United States

a r t i c l e

i n f o

Article history: Received 18 November 2009 Received in revised form 22 October 2010 Accepted 27 October 2010 Available online 3 November 2010 Keywords: Product heterogeneity Motion pictures Fixed export costs

a b s t r a c t We extend Melitz (2003) to allow for both global and bilateral ﬁxed export costs. If global (bilateral) export costs dominate, the average sales ratio (import sales per product variety/domestic sales per variety), decreases (increases) in variable (ﬁxed) trade barriers, due to adjustment along the intensive (extensive) margin of trade. Using novel data on bilateral US movie exports we ﬁnd that (i) variation in box-ofﬁce revenues per movie is much larger than in the number of movies exported, and (ii) the average sales ratio decreases in geographic and linguistic distance. These ﬁndings suggest that global ﬁxed export costs dominate. © 2010 Elsevier B.V. All rights reserved.

1. Introduction Fixed trade costs have become a common feature of international trade models. To explain why most manufacturing plants do not export any output and why those that do are relatively large and productive (Bernard and Jensen, 1999, 2004), Melitz (2003), in a widely cited work, develops a model with ﬁrm heterogeneity and ﬁxed export costs. Because of ﬁxed trade charges, only more productive plants ﬁnd it proﬁtable to sell goods abroad.1 Fixed trade costs imply that adjustment in trade volumes may occur along both the intensive margin (value of trade per product) and extensive margin (number of products traded). In the standard monopolistic competition model, consumer love of variety leads all products to be exported, meaning trade varies at the intensive margin only (Helpman and Krugman, 1985). A fall in transport costs causes exports of all products to increase, consistent with the robust negative coefﬁcient on distance in the gravity model of trade (Anderson and van Wincoop, 2004). With ﬁxed export costs and ﬁrm heterogeneity, a fall in transport costs may cause trade volumes to increase both through existing exporters exporting more and new ﬁrms beginning to export. Bernard et al. (2007) ﬁnd that most variation in US merchandise exports is at the extensive margin, with smaller countries importing fewer US products.

⁎ Corresponding author. UC San Diego, United States. E-mail addresses: [email protected] (G. Hanson), [email protected] (C. Xiang). 1 When applied to aggregate data, this framework yields a gravity speciﬁcation that can account for why many country pairs do not trade (Helpman et al., 2007; Baldwin and Harrigan, 2007). 0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.10.007

Despite abundant indirect evidence that ﬁxed trade costs exist, we know little about their nature. While we can measure variable trade barriers in the form of tariffs or transport fees, no similar data exist for expenses that are ﬁxed. If ﬁxed export costs are bilateral, such that ﬁrms incur a charge to enter each new foreign market, small countries will be disadvantaged in global trade. However, if ﬁxed export costs are largely global in nature, such that once ﬁrms establish an international distribution network they face only variable charges in adding new markets, small countries are not at a disadvantage and it is only less productive ﬁrms that fail to participate in trade. In this paper, we develop a version of the Melitz (2003) model with both global and bilateral ﬁxed export costs. For countries where global ﬁxed export costs are large relative to bilateral costs, import sales per product variety (relative to domestic sales per variety) are decreasing in variable trade barriers, as adjustment occurs along the intensive margin of trade. For countries where bilateral ﬁxed export costs are relatively large, imports per product variety are increasing in ﬁxed trade barriers, due to adjustment occurring along the extensive margin. When global ﬁxed export costs are sufﬁciently large, they dominate for all countries, such that bilateral ﬁxed export costs have no signiﬁcant effects on export decisions. In contrast, when global export costs are sufﬁciently small, bilateral ﬁxed export costs dominate for all countries. Whether or not global ﬁxed export costs dominate, and for which countries they dominate, depend on the magnitudes of ﬁxed export charges and are therefore empirical questions. In US manufacturing, most adjustment to trade barriers is at the extensive margin, suggesting that bilateral ﬁxed export costs dominate. This conclusion, however, is impressionistic, as the literature lacks formal analysis of the relative importance of global versus bilateral ﬁxed export costs or whether results for manufacturing generalize to other sectors.

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

We apply our framework to data on exports of US motion pictures. An advantage of our approach is that we need not take a stand on which trade barriers represent ﬁxed impediments and which variable ones. The empirical literature offers little guidance on the issue. Standard gravity variables – distance, language, colonial history – are likely correlated with variable and ﬁxed barriers. We exploit the divergent predictions of the two regimes for whether imports per variety have positive or negative correlation with trade barriers, making our approach applicable in a wide variety of settings. The characteristics of the ﬁlm industry are consistent with the assumptions of the Melitz framework. Fixed costs are important in movie production and studios differentiate their ﬁlm product (De Vany, 2004). There is heterogeneity in movie performance, with boxofﬁce revenues for US ﬁlms exhibiting heavy right tails (De Vany and Walls, 1997, 2004). Not all US movies are exported, with the typical country importing half of the movies the US produces in a given year. Given the absence of published data on bilateral trade in motion pictures,2 we construct a new data set using information from ScreenDigest.com, an entertainment industry consultancy, which includes box-ofﬁce revenues for domestic and US movies in 46 countries over 1995–2006. We also make use of data on trade barriers in motion pictures from other sources. A broader motivation for studying motion pictures is that the vast majority of empirical work on trade is for manufacturing, with relatively little work on services. Exports of services such as motion pictures are distinct from exports of manufactures in that variable production costs (e.g., exhibiting movies to consumers) are incurred in the country of consumption, rather than the country of production, and the physical cost of transporting goods abroad (e.g., shipping master ﬁlm prints) is close to nil. For motion pictures and other cultural goods, cross-country differences in language, social mores, or religion may be the signiﬁcant barriers to trade (Rauch and Trindade, 2009). Movies are representative of other information services, including music, publishing, software, television, and video games, which are responsible for a growing share of US trade. These “copyright” industries tend to have large ﬁxed production costs (associated with creating an initial ﬁlm print, music recording, or software program) and small marginal production costs (associated with producing additional copies). In 2007, copyright industries accounted for 7% of US GDP; US movies, TV, and videos had foreign sales plus exports of $20 billion, just below that for US pharmaceuticals. Our main ﬁnding is that a model in which global ﬁxed export costs dominate for all countries in the sample best explains US exports of motion pictures. Imports per product variety are decreasing in geographic distance, linguistic distance, and other measures of trade barriers, in a manner consistent with adjustment to these barriers occurring along the intensive rather than the extensive margin. There is relatively little variation in the number of US movies that countries import but wide variation in the box-ofﬁce revenues per movie, with countries more distant from the US spending less on the US movies that they see. Argentina, for instance, imports roughly the same number of US movies each year as Germany, though the box-ofﬁce revenues per US ﬁlm there are far lower. These results hold across groups of importers separated by size or distance from the United States, suggesting that in our sample, bilateral ﬁxed export costs dominate for few, if any, countries. Whereas the gravity-equation literature consistently ﬁnds a negative correlation between log sales and geographic and linguistic distance, our speciﬁcation is based on a double difference in the form of log sales per product among imports minus log sales per product among domestic goods. Our ﬁnding is, 2 For exports of US movies, McCalman (2005) has foreign release dates but not foreign revenues. The US government does not publish trade ﬂows for movies. UN Comtrade lists motion pictures as a commodity; its trade data understate foreign boxofﬁce revenue by several orders of magnitude (see note 21).

15

therefore, not a corollary of results in the gravity literature. We are unaware of previous literature that uses such a speciﬁcation in testing trade models. In Section 2, we present the model and estimation strategy. In Section 3, we discuss data and preliminary data analysis. In Section 4, we present empirical results and in Section 5, we conclude. 2. Theory In this section, we extend the Melitz model and allow ﬁxed export costs to have both bilateral and global components. The bilateral components are incurred each time a producer enters a new export market; the global components are incurred once, when a producer starts exporting. The model predicts that the correlation between trade costs and revenues per movie depends on whether the global components of ﬁxed export costs dominate the bilateral ones or viceversa. 2.1. Model setup There are many sectors, over which consumers have Cobb– Douglas preferences, and T + 1 countries, where u indexes the exporter and k = 1…T indexes the importers. We focus on the movie industry and leave other sectors in the background. Movies are subject to cultural discount (Waterman, 2005). For a consumer in country k, a movie from country u (the US) reduces utility by δuk as compared with a domestic movie, where 0 b δuk b 1. Similar to a pure iceberg trade cost, δuk is the portion of a movie's value that is not “lost in translation” in moving from one country to another. We expect δuk to be higher the more similar are two countries' culture and language. Movies are also subject to ﬁxed and variable trade costs. Variable ad valorem trade fees, deﬁned as tuk N 1, include tariffs and surcharges on foreign movie revenues (Marvasti and Canterbery, 2005), transaction costs in negotiating contracts between US movie distributors and foreign movie exhibitors (Gil and LaFontaine, 2009),3 advertising expenses (which depend on the length of time a movie spends in theatres), and ﬁlm printing expenses (which depend on the number of times a movie is shown and the number of theatres in which it appears). Fixed export costs are associated with allocating the right to distribute a movie in different countries, creating an international marketing campaign, editing movies for foreign audiences, and adding subtitles or dubbing movie dialogue. Fixed costs may be market speciﬁc (if each country requires its own marketing campaign) or global in nature (if one marketing campaign serves multiple countries). Fixed and variable trade charges may have common underlying determinants, with distance, language, and the costs of enforcing contracts possibly affecting both. The subutility for movies for the representative consumer in country k is CES,4 n h iσ−1 o σ σ−1 uk = ∑j θj ckj σ + ∑u≠k ∑j θj δuk cuj σ σ −1 ;

ð1Þ

where σ N 1 is the substitution elasticity. A demand shifter, θj, captures heterogeneity in movies. A movie with a high θj is popular (E.T., Titanic); one with a low θj is unpopular (Ishtar, Gigli). In line with previous literature on ﬁrm heterogeneity, we assume the popularity of a movie does not depend on the country in which it is shown and 3 Gil and LaFontaine (2009) ﬁnd that in the Spanish movie industry distributors sign an initial contract for sharing revenues with movie exhibitors, which may be renegotiated multiple times, as a movie reveals itself to be more or less popular than expected. Contracting costs are thus a function of the level of revenues and not simply the discrete outcome of whether or not a movie is shown. 4 We specify consumer preferences using the discrete choice framework in Hanson and Xiang (2008) and obtain similar results to those reported here.

16

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

that θj is drawn from a common distribution, G(θ). Consistent with this assumption, top grossing movies tend to be the same across countries. For individual US-made movies in 2000, the correlation between domestic and foreign box-ofﬁce revenues is 0.81 (Scott, 2004). Heterogeneity in demand is isomorphic to heterogeneity in marginal costs (Melitz, 2003). We introduce heterogeneity on the demand side as it compresses variation in admission prices for movies within a country, consistent with the data (De Vany, 2004). From (1), box-ofﬁce revenues (total sales) of a country-u movie in country k are 1−σ 1−σ sukj = θσj −1 δσuk−1 tuk pukj Ak ;

Ak ≡

αYk ; Pk1−σ

ð2Þ

where j indexes the movie, Yk and Pk are income and the CES price index in country k, α is the expenditure share for the movie industry, and pukj is the price of movie j (not including the policy trade barrier). In Eq. (2), both the cultural discount and the policy trade barrier appear as variable trade costs and have similar effects on the sale of movie j. Box-ofﬁce sales of domestically produced movie h in country k equal, σ −1 1−σ pkkh Ak :

skkh = θh

ð3Þ

We assume movie production occurs in ﬁve steps. (i) A producer in country u hires fE units of country-u labor to produce a master ﬁlm print, which is a sunk labor input. (ii) The producer draws a θ from the distribution G(θ). (iii) The producer incurs a ﬁxed production cost of b units of country-u labor. (iv) The producer uses a variable labor input to exhibit the movie to an audience, with input costs incurred in the country where the audience is located. For each unit of the movie shown in country k, the producer hires one unit of country-k labor. And (v) the producer collects proﬁts. In our model, each ﬁrm produces a single variety, as in Melitz (2003). However, our setting also has a multi-product-ﬁrm interpretation, consistent with Bernard et al. (2006). Most movies are distributed by large studios, which may be involved in production as well.5 Even though there are a large number of studios worldwide, a half dozen distribute most high-grossing ﬁlms (Columbia/MGM, Paramount, Disney, Warner Brothers, 20th Century Fox, and Universal Pictures). One can think of a studio as a multi-product ﬁrm. Within a year, a studio will release many movies, where each is a product variety. The short lived nature of movies allows studios to differentiate movies in time,6 avoiding the simultaneous release of two or more ﬁlms. In this setting, for each product variety we can still apply assumptions (i)–(v) and proceed with the analysis.7 5

Making a movie involves development (securing rights, screen writing, casting, ﬁnancing), production (ﬁlming, special effects, music, editing), distribution (marketing, negotiating with theatres), and exhibition. While studios are usually involved in distribution, they have varying roles in earlier stages, sometimes handling a movie from start to ﬁnish (Paramount and Mission: Impossible II) and sometimes buying distribution rights to an already ﬁnished movie (20th Century Fox and Little Miss Sunshine). US antitrust rulings prevent movie studios from owning theatres, keeping studios out of exhibition. 6 By the end of three weeks, the average movie has earned 66% of its total box-ofﬁce revenues (De Vany and Walls, 1997). 7 Multi-product ﬁrms bring two additional elements into consideration. First, the popularity of a movie may depend on the “ability” of its studio. Following Bernard et al. (2006), we assume the distribution of studio abilities is independent of θ and that the ability of a given studio is common for all products. With each studio randomly drawing its ability when it enters the market, studio abilities are given for each individual movie. Second, the entry decisions and numbers of studios in the market may depend on studio-level sunk entry costs and per period studio-level ﬁxed costs, as well as studio-level ability. Since we lack data on the distribution of movies by studio, we leave studios in the background of the analysis. Some portion of our product-level sunk entry cost fE might be indistinguishable from the studio-level ﬁxed cost, but fE also remains in the background.

Following Melitz (2003), we assume the movie industry is monopolistically competitive.8 By assumption (iii), for a movie created in country u its showings are provided using labor in the country where consumers watch the movie. Since price is a constant markup over marginal cost, the price of a movie shown in country k is the same for all movies, regardless of where they are produced: pkkj = pukj =

σ w for all u; k; j; σ −1 k

ð4Þ

where wk is the wage in country k. Because the cultural discount is a source of home bias in demand, it does not affect prices (Anderson and van Wincoop, 2004). Similarly, θj affects the quantity demanded but not prices. Eq. (4) implies that in any market k, the prices of domestic and foreign movies are the same. 2.2. Decision for exports In this sub-section we consider the export decisions by the producers of country-u movies. We show how the pattern of exports depends on bilateral and global ﬁxed export costs, and derive the expressions for the number of exported movies and their total sales abroad. We allow the producers to observe their type before making the export decision, consistent with the movie industry where producers release ﬁlms on the domestic market ﬁrst, and then, if they are successful, abroad.9 The export decision for a given country-u movie j involves the comparison between its variable proﬁts abroad and ﬁxed export costs. Eqs. (2) and (4) imply that the variable proﬁts for movie j in country k equal σ −1

πukj = θj

DQ k ; D =

ðσ −1Þσ−1 σ −1 1−σ 1−σ ; Q k = δuk tuk wk Ak ; σσ

ð5Þ

where Ak is deﬁned in Eq. (2) and D is a constant. Intuitively, the Q-index, Q k, represents access to the country-k market from country u. Market access is relatively strong for high Q-index countries, which have a relatively large GDP, Yk, and/or relatively low variable trade costs, δuk and tuk. Fixed export costs for movie j consist of the global component of fG units of country-u labor, and the bilateral components of fuk units of country-k labor.10 This means that the total ﬁxed export cost to show movie j in countries k = 1…K equals wu fG + ∑Kk = 1 wk fuk . Consider the bilateral components ﬁrst. Conditional on showing movie j abroad, showing it in an additional country, k, requires the additional ﬁxed export cost wk fuk, and brings in the additional variable proﬁt πukj. Therefore, conditional on showing movie j abroad, it is shown in country k if and only if πukj − wk fuk ≥ 0, or θj ≥ θ uk =

σ −1 σ fuk δ1−σ uk tuk wk DAk

!

1 σ −1

:

ð6Þ

8 While the number of major movie studios is small, the number of top talents involved in a movie (e.g., actors, directors, screenwriters, technicians, and producers) is large. Competition between the top talents is important for the competition between movies. To secure the star power of these talents, studios bid for their services (Waterman, 2005), driving expected proﬁts toward zero. Alternatively, suppose the movie producers have more market power than in the monopolistic competition setting. This implies a higher markup in Eq. (4), but otherwise does not affect our results qualitatively. 9 McCalman (2005) shows that movies released simultaneously in the US and a few foreign markets are released in other foreign markets with a lag. 10 We make this assumption to keep our exposition concise. If, instead, both global and bilateral ﬁxed export costs are in units of country-u labor, the only noticeable change to our results is that Eq. (16) will have the additional term, wk. This suggests augmenting our main estimation Eq. (17), with importer characteristics. We experimented with these additional controls in Hanson and Xiang (2008); they were statistically insigniﬁcant and had little effect on our results.

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

Eq. (6) is the export constraint imposed by the bilateral ﬁxed export costs, and the expression for θ uk does not depend on the global ﬁxed export costs, fG. Rank the θ uk 's for all the T importing countries from low to high and re-label the countries according to the rank order of the θ uk 's; i.e., θ u1 b θ u2 b … θ uT . This ranking of the θ uk 's is illustrated in Fig. 1. Country 1 is the most accessible destination market for country-u movies (e.g., the U.K.) and country T is the least accessible destination market for country-u movies (e.g., Latvia). To determine which movies are exported and which are not, start from the movies with the lowest θ's. First, the movies with θj ≤ θ u1 (i.e., those to the left of θ u1 in Fig. 1) do not satisfy Eq. (6) for any importer and so they are not exported. Intuitively, the variable proﬁts for these movies from the most accessible destination market, country 1, are insufﬁcient to pay for the ﬁxed export cost speciﬁc to that country, w1 fu1, let alone the additional global ﬁxed export cost, wu fG. Next, the movies with θj ∈ (θ u1 ,θ u2 do not satisfy Eq. (6) for the importers 2…T and so they are not exported to those countries. However, these movies satisfy Eq. (6) for country 1, and they will be shown in country 1 if they satisfy the additional condition that πu1j −w1 fu1 ≥ wu fG :

ð7Þ

Eq. (7) says that the variable proﬁts from country 1 can cover both the bilateral and global components of ﬁxed export costs. Since fG N 0, the movie with θ u1 does not satisfy Eq. (7). Suppose Eq. (7) holds for some movie with θ0 ∈ (θ u1 ,θ u2 . Then Eq. (7) also holds for all the movies with θj ≥ θ0, because these movies bring in more variable proﬁts than the one with θ0 but face the same ﬁxed export costs. Therefore, for these movies with θj ≥ θ0 the pattern of exports is the same as in the standard Melitz model, with bilateral ﬁxed export costs only. That is, the movies with θj ∈ [θ0,θ u2 ) are exported to country 1, those with θj ∈ [θ u2 ,θ u3 Þ are exported to countries 1 and 2, etc. Now suppose Eq. (7) does not hold for any θj ∈ (θ u1 ,θ u2 . Then we can examine whether some movie in the range (θ u2 ,θ u3 ] satisﬁes the condition ∑2k = 1 πukj −wk fuk ≥wu fG . If it does, then we can pin down the pattern of exports as discussed above. If it does not, then we examine whether in the range (θ u3 ,θ u4 ] satisﬁes the some movie condition ∑3k = 1 πukj −wk fuk ≥wu fG , etc.

a Exported to K+1 countries …

θG

θu1

θu 2 ….

θu , K +1

θuK

θu , K + 2 …

Domestic movies Exported to K countries

b

θG' G’ = {1,…,L}

θu1 …

θuK

θu , K +1 θu , K + 2 …

B’ = {L+1,…,T}

θuL

θu , L +1 …

B = {K+1,…,T}

G = {1,…,K}

θG Fig. 1. a: Pattern of exports. b: Increase in global ﬁxed export cost.

θuT

17

Summarizing this algorithm, for movie j to be exported, n o K θj ≥ θ G = minθj ∑k = 1 πukj −wk fuk ≥ wu fG ;

ð8Þ

where K is such that θ uk b θ G for all k ≤ K and θ u;K + 1 N θ G . Eq. (8) is the export constraint imposed by the global ﬁxed export costs. It says that the variable proﬁts from the inframarginal importers 1…K, net of the bilateral ﬁxed export costs speciﬁc to these countries, remain sufﬁcient to cover the global ﬁxed export cost. A country-u movie is exported if and only if it satisﬁes both Eqs. (6) and (8). Fig. 1a illustrates the pattern of exports. The value of θ uk lies between θ uK and θ u;K + 1 . None of the movies with θj b θ G gets exported since they do not satisfy Eq. (8). All the movies with θj ≥ θ G are exported to at least the inframarginal K countries 1…K since they also satisfy Eq. (6) for these countries. In addition, the movies with θj ∈ [θ G ,θ u;K + 1 ) are exported to the inframarginal importers 1…K, those with θj ∈ (θ u;K + 1 , θ u;K + 2 ] are exported to importers 1…K + 1, etc. This pattern of exports suggests the following partition of importers. Let G (for Global) denote the set of inframarginal importers k = 1…K, for which θ uk b θ G . These countries have high Q-indices (i.e., large market size and/or low variable trade costs with country u) and are the most accessible destination markets for country-u movies. For these countries, the global ﬁxed export cost dominates the bilateral ones, and the global export constraint of Eq. (8) is more binding. Let B (for bilateral) denote the set of importers k = K + 1, …,T, for which θ uk N θ G . These countries have low Q-indices (i.e., small market size and/or high variable trade costs) and are the least accessible markets for country u. For them, bilateral ﬁxed export costs dominate global ones, and the bilateral export constraint of Eq. (6) is more binding. The relative size of the sets G and B depends on the size of the global ﬁxed export costs relative to bilateral ﬁxed export costs.11 As the global ﬁxed export cost, fG, increases, θ G increases by Eq. (8) but θ uk remains unchanged by Eq. (6). This implies that the set G expands to encompass more countries, ceteris paribus. We illustrate these changes using Fig. 1b. The increase in fG increases θ G from a value in the range (θ uK , θ u;K + 1 ) to a higher value in the range (θ uL , θ u;L + 1 ), where L N K.12 As the line corresponding to θ G shifts to the right, the set G expands to incorporate the additional countries {K + 1,…,L}, and the set B contracts accordingly. In addition, if the increase in fG is sufﬁciently large, the line θ G could shift to the right of θ uT , in which case the set B is empty and all importers are in the set G. In this case, the global ﬁxed export cost dominates for every importer and the bilateral ﬁxed export costs play no role in the decision for exports. The other polar case is when global ﬁxed export cost is very small so that the set G is empty.13 Then, bilateral ﬁxed export costs dominate for every importer, as if the global ﬁxed export cost did not exist. We now derive (a) the number of country-u movies exported to country k, and (b) total sales of country-u movies in country k, for the G and B sets of importers. Following the literature on the Melitz model we assume that the distribution function G(∙) is Pareto with G(θ) = 1 − aς/θς, where a, ς N 0 and θ ∈ [a, + ∞). A large ς means thin tails for G(θ). We assume that ς N σ − 1, such that total movie sales are ﬁnite. Let Nu denote the number of country-u movies that draw from the distribution G(θ).14 11 The analytical solution for θ G is difﬁcult to obtain, since θ G depends on the levels and distributions of fuk and Qk. 12 This also shows that the increase in fG has a discrete effect on the composition of the set G, which changes only if θ G rises above θ u;K + 1 . 13 Theoretically, this case obtains when Eq. (7) holds for some movie with θ0 ∈ (θ u1 ,θ u2 . 14 As in Melitz (2003), Nu is pinned down by the sunk entry cost, fE. In Hanson and Xiang (2008) we derive Nu assuming T + 1 identical countries and each country having one sector, as in Melitz (2003). Nu and its counterparts in other countries are jointly determined.

18

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

First, by Eq. (8) and Fig. 1, the G set of importers share the common export cut-off of θ G . Using the value of θ G we have: ∞ ς −ς ðaÞnuk = Nu ∫θ dG θuj = Nu a ðθ G Þ ; k ∈ G; G ∞ 1−σ ðbÞSuk = Nu ∫θ sukj dG θuj = Cu nuk δσuk−1 tuk Ak w1−σ ; k G

Cu =

Dσ ςθ σG −1 ς−ðσ −1Þ

ð9Þ

; k ∈ G;

To see the intuition behind Eq. (9), consider the total sales of country-u movies in country k, Suk. Eq. (9b) is a gravity-like prediction in which Suk responds to country-k characteristics, such as income, and variable trade costs between u and k. This variation consists of an extensive margin – the number of country-u movies exported to k – and an intensive margin — the average sale per country-u movie. In Eq. (9a), the extensive margin is exporting-country speciﬁc and does not vary with importing-country characteristics. As a result, all variation in Suk occurs along the intensive margin. The ﬁxed export cost does not affect the intensive margin because it does not vary across the G set of importers. Next, for the B set of importers, we can use the values of θ uk to derive ∞ ðaÞnuk = Nu ∫θ dG θuj = Nu aς ðθ uk Þ−ς ; k ∈ B; uk

ðbÞSuk = Nu ∫θ sukj dG θuj = ∞

uk

= nuk fuk wk

ς

Bσ ςa σ−1 1−σ 1−σ σ−ς−1 t Ak wk ðθ uk Þ N δ ς−ðσ −1Þ u uk uk

σς ; k ∈ B; ς−ðσ −1Þ

ð10Þ

In contrast to Eq. (9), Eq. (10) says that variation in Suk across the B set of importers k occurs primarily along the extensive margin, nuk, and that the intensive margin, Suk/nuk, does not depend on variable trade costs, δuk and tuk, or expenditure on movies by k.15 To see the basis for this result, compare importing country l, which has low variable trade costs with exporting country u, to importing country m, which has high variable trade costs with u. Higher variable trade costs in m mean that total sales of country-u movies in m are lower than in l. They also imply a higher cut-off level of θ for country-u movies shown in m, such that m imports a smaller number of movies from u. Given the assumption that the distribution of movie types is Pareto,16 these two effects exactly offset each other, leaving sales per movie unaffected by variable trade costs.17 2.3. Decisions for domestic production We now consider domestic movie production in the importers k = 1…T. The decision for domestic production involves the comparison between the variable proﬁts from the domestic market and the

ﬁxed production cost parameter, b. To minimize notation, we assume that b is invariant across importers. Since the distribution of θ has support [a, +∞), the lowest variable proﬁt a country-k movie can reap in the domestic market is aσ − 1AkDw1k − σ by Eqs. (3) and (4), where D is a constant deﬁned in Eq. (5). If this lower bound for variable proﬁt exceeds the ﬁxed production cost, bwk, then the number of country-k movies produced, nkk, equals the number of country-k movies that draw from G(θ), Nk. Intuitively, the ﬁxed production cost is relatively small and so does not affect the decision for domestic production. However, if the lower bound for variable proﬁt is less than bwk, then of the Nk domestic movies that could be made, only those that can at least break even in the domestic market are actually made; i.e., nkk b Nk. The intuition is that the ﬁxed production cost b is relatively large and so it affects which movies are produced domestically. Summarizing nkk b Nk if b N b k ; b k = a

σ−1

−σ

Ak Dwk ; otherwise nkk = Nk :

Eq. (11) is the constraint for domestic production imposed by the ﬁxed production cost. In the importers with a large domestic market, Ak, domestic movies collect high variable proﬁts; as a result, b k is large and the constraint for domestic production is less likely to be binding. Rank the importers in descending order in terms of their b k 's so that b 1 N b 2 N … b T . For ease of exposition suppose that b M N b N b M + 1 . Then we can partition the importers into two sets. Let GD denote the set of importers k = 1…M, for which b k N b. These countries have large domestic markets and so for them the constraint for domestic production is not binding. Let BD denote the set of importers k = M + 1…T, for which b k b b. These countries have small domestic markets and so for them the constraint for domestic production is binding. Fig. 2 illustrates the sets GD and BD. We now derive (a) the number of country-k movies actually made, and (b) the total sales of domestic movies. For the GD set of importers, ðaÞnkk = Nk : ∞

1−σ

ðbÞSkk = Nk ∫a skkh dGðθkh Þ = C0 nkk wk 1−σ σ ς : C0 = aðσ −1Þ ς−ðσ −1Þ

Ak ;

An importing country with a large GDP, Yk, may also have a large number of domestic movies and so a low CES price index, Pk. This tends to reduce the demand for foreign movies and may dampen the effect of Yk. In a stylized model, Helpman et al. (2004) show that the competition effect of Pk may completely offset the country-size effect of Yk in general equilibrium such that nuk is the same across all importing countries k. This result is derived under factor-price equalization and identical trade costs for every country pair, assumptions we do not impose. 16 In Hanson and Xiang (2008) we show that our results are robust to alternative assumptions. In particular, we show that under a two-segment Pareto distribution Eqs. (14) and (16), our main estimating equations, need to be augmented by 2ndorder terms of importer characteristics, such as GDP and population. We then estimate the augmented regressions and ﬁnd that these 2nd-order terms are jointly insigniﬁcant. 17 Suppose that there is convex market access cost as in Arkolakis (2007). To derive analytical solutions for Suk we consider the Arkolakis setting with his parameters α, β and γ set to 0. Then the only noticeable changes to our results are (a) the expression for sukj is multiplied by the fraction of consumers that the producer of movie j reaches in country k, (b) Eq. (10b) becomes Suk = nuk fuk(wk − 1)c1, where c1 is a constant, and (c) wk enters into Eq. (16). These changes do not affect our results qualitatively, as we discuss in note 10.

ð12Þ

For the BD set of importers, a given domestic movie h is made if and only if the variable proﬁts from the domestic market exceeds the ﬁxed production cost; i.e., θσh − 1AkBw1k − σ ≥ bwk. This is equivalent to 1 σ σ −1 bwk . Using the value of θ kk we have θh ≥ θ kk = BAk ∞

ðaÞnkk = Nk ∫θ dGðθkh Þ = Nk aς ðθ kk Þ−ς ; θ kk = kk

15

ð11Þ

∞

ðbÞSkk = Nk ∫θ skkh dGðθkh Þ = kk

= nkk bwk

1 σ σ−1 bwk BAk

Bσςaς 1−σ σ−ς−1 N A w ðθ kk Þ ς−ðσ −1Þ k k k

σς : ς−ðσ −1Þ

b

… b K +2

b K +1

constraint binding

bK

… b2 constraint not binding

Fig. 2. Constraint for domestic production.

b1

ð13Þ

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

2.4. Main estimating equations We now have two ways of partitioning the importers, ﬁrst using the global and bilateral ﬁxed export costs and then using the ﬁxed production cost. In general, these two approaches produce different partitions. However, country size determines both the ranking of θ uk and the ranking of b k (e.g., Japan is a large country for U.S. movies and also a large country for Japanese movies), and so the ranking of θ uk and the ranking of b k could be highly correlated. For ease of exposition we assume that18 Assumption 1. G = GD and B = BD. Assumption 1 implies that for the G set of importers where the global ﬁxed export cost dominates, the constraint for domestic production is not binding. Eqs. (9) and (12) imply that ln

σ −1 Suk = nuk t Bσςθ u : = ð1−σ Þln uk + lnCu ; Cu = Skk = nkk δuk ς−ðσ−1Þ

ð14Þ

In Eq. (14), Suk/nuk and Skk/nkk are the average sales in country k of movies produced in country u and of movies produced domestically. On the left of Eq. (14) are average sales in relative terms, which we refer to as the average sales ratio. By expressing average sales as a log double difference, the competitiveness of market k, Pk, and domestic expenditure on movies, αYk, drop out. With global ﬁxed export costs and pure sunk costs, the average sales ratio is negatively correlated with variable trade barriers between an importer and an exporter. A result similar to Eq. (14) holds for the standard monopolistic competition model, with no ﬁrm heterogeneity or ﬁxed export costs, where the average sales ratio is, ln

! sukj Suk = nuk t = ln = ð1−σ Þln uk : Skk = nkk skkj δuk

ð15Þ

In the monopolistic competition model, the variation of Suk occurs along the intensive margin, as in Eq. (14). However, the standard monopolistic competition model also predicts all movies are exported, contrary to Melitz-type models. Assumption 1 also implies that for the B set of importers where the bilateral ﬁxed export costs dominate, the constraint for domestic production is binding. Eqs. (10) and (13) imply that ln

Suk = nuk f = ln uk : Skk = nkk b

ð16Þ

19

B set of importers. Our theory also predicts that the G (B) set of importers tend to have high (low) Q-indices; i.e., large market sizes and/or low variable trade costs with respect to the U.S. We thus carry out our estimation in two steps. Let Xuk be a vector of covariates for trade barriers between the US and country k. We ﬁrst pool across all importers in our data and estimate, 19 S = nukt ln ukt = α u + α t + βXuk + εukt ; Skkt = nkkt

ð17Þ

where αt represents year ﬁxed effects and αu is a constant (the U.S. is the only exporting country in our data). The ﬁrst step reveals whether global or bilateral ﬁxed export costs are the salient feature of our data. If most importers in our data are in the G set, then by Eq. (14) we have that β ≤ 0; if most of our importers are in the B set, then by Eq. (16) β ≥ 0. To identify which importers are in the G and B sets, we partition the importers in our data into two sub-samples, those with high apparent Q-indices (i.e., large market sizes and/or low variable trade costs with respect to the U.S.) and those with low Q-indices. According to our theory, the high (low) Q-index sub-sample should correspond to the G (B) set. We then estimate regression (17) separately for the two sub-samples. If global (bilateral) ﬁxed export costs dominate for the large majority of importers in the sample and the set B (G) is empty (or consists of only a handful of countries), then β ≤ 0 (β ≥ 0) for both sub-samples. If global ﬁxed export cost dominates for some importers but bilateral ones dominate for the others, then β ≤ 0 for the high Q-index sub-sample but β ≥ 0 for the low Q-index sub-sample. In practice, Xuk will include proxies rather than direct measures of bilateral trade barriers (distance, common language, etc.). We do not know whether these factors are associated with ﬁxed or variable impediments. An advantage of the speciﬁcation in Eq. (17) is that we need not resolve the ﬁxed-versus-variable-trade-barrier dilemma. Since global and bilateral ﬁxed export costs give opposite sign predictions for the correlation between the average sales ratio and trade barriers, testing one against the other simply involves seeing whether the parameter vector β is positive or negative. Also, the double differencing implicit in the average sales ratio in Eq. (17) sweeps out of the estimation market competitiveness, the consumption share of movies, and number of movies that could be made for country k, all of which are hard to measure. Finally, although the gravity literature consistently shows a negative correlation between ln(Sukt) and Xuk, ln(Sukt) is but one of the four terms entering into the average sales ratio. Thus, the ﬁndings of the gravity literature do not necessarily imply the sign of β ex ante. 3. Data

Eq. (16) says that the average sales ratio is positively correlated with ﬁxed trade costs, and it is similar to the predictions under the original setting of Melitz (2003). 2.5. Empirical speciﬁcations We use Eqs. (14) and (16) for our empirical speciﬁcations. Let country u be the US, Sukt and nukt be total box-ofﬁce revenue for US ﬁlms and the total number of US ﬁlms shown in country k in year t, and Skkt and nkkt be the total box-ofﬁce revenue and total number of domestically produced ﬁlms shown in k in year t. Both Eqs. (14) and (16) predict that the average sales ratio is correlated with trade barriers. Eq. (14) predicts a negative correlation between the average sales ratio and variable trade barriers for the G set of importers, whereas Eq. (16) predicts a positive correlation between the average sales ratio and ﬁxed trade costs for the 18 Without Assumption 1, we need to derive the estimating equations for the countries in both G and BD, and for those in both B and GD. We derive these additional equations in Hanson and Xiang (2008), and show that they have little impact on the estimation results, suggesting that the data are consistent with Assumption 1, at least for the countries in our sample.

3.1. Exports of US motion pictures We evaluate the demand for US ﬁlms and domestically made ﬁlms using data on box-ofﬁce revenues by country and year.20 Box-ofﬁce revenues are equivalent to the c.i.f. (customs, insurance, freight) value of motion picture services consumed in cinemas, plus retail markups. These revenues include import duties, transport costs, and other trade costs incurred in delivering the service to the consumer, as well as sales taxes and exhibition fees collected by cinemas. 19 Recent work examines the correlation between the normalized number of ﬁrms exporting to a country and the size of the importing country (e.g., Eaton et al., 2004; Arkolakis, 2007). In Hanson and Xiang (2008) we show this correlation does not help us distinguish global versus bilateral ﬁxed cost. 20 Individuals consume services of new movie releases through cinemas and previous movie releases through TV and video rentals or purchases. During our sample period, distributors tend to release movies to cinemas ﬁrst, then to the pay per view, home video, cable TV, and broadcast TV markets in sequence (e.g., for the typical movie there is a 150–200 day lag between the date of release in theatres on the US market and the date of release on the home video market) (Waterman, 2005). Data on ﬁlm revenues by origin country from non-cinema sources are unavailable.

20

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

Data on box-ofﬁce revenues for 46 countries over the period 1995– 2006 are available from ScreenDigest.com.21 In each country, ScreenDigest reports the number of ﬁlms screened, total ﬁlm attendance, and total box-ofﬁce revenues for ﬁlms imported from the US and ﬁlms produced domestically.22 Data cover ﬁrst-run theatrical releases, which exclude older ﬁlms, pornographic ﬁlms, and movies shown only on TV or through the home video market. For Europe, coverage begins in 1995, while for other regions it begins later. Data are compiled from government agencies, national ﬁlm bodies, ﬁlm exhibitor and distributor associations, and company spokespeople.23 One concern about the ScreenDigest data is it may represent a sample of countries selected for having large ﬁlm markets. ScreenDigest selects countries for inclusion in the data based on the markets in which it perceives its clients to have an interest (because of actual or potential movie revenues). Were the company to avoid countries that import few foreign ﬁlms, our results on the extensive and intensive margins of trade would not be globally representative. The ScreenDigest data include all of Europe and North America, and most of Asia and South America (see Table 1).24 Missing are Africa, the Caribbean and Central America, Central Asia, most Islamic republics, and Paciﬁc island nations, which are regions dominated by small developing countries. To determine whether ﬁlm imports in these economies differ from the ScreenDigest sample, we examined ﬁlm listings in major newspapers and internet sites covering 25 countries in the missing regions during 2008 and early 2009.25 In these countries 1.8 new US movies are released on average each week, equivalent to 93.2 US movies a year, a level of imports slightly higher than those of the Baltic countries, which are among the smaller markets in the ScreenDigest sample. Although we do not have information on box-ofﬁce revenues in these countries, they are importing large numbers of US ﬁlms, suggesting the markets excluded from ScreenDigest are not ones with minimal US movie imports. 3.2. Trade barriers in motion picture trade The method for testing the Melitz model that we develop in Section 2 calls for all relevant trade barriers to be included in the estimation. We include measures of geographic distance, cultural

Table 1 Data coverage. Country

No. of domestic ﬁlms

No. of US ﬁlms

Rev. per domestic ﬁlm

Rev. per US ﬁlm

Lithuania Latvia Estonia Romania Slovak Republic Iceland Malaysia Slovenia Czech Republic Philippines Singapore Hungary Finland Argentina Thailand Belgium Turkey New Zealand Portugal Indonesia Poland Ireland Norway Denmark Austria Sweden Netherlands Switzerland Brazil Russia Spain Italy South Korea Mexico Australia Canada China France Germany UK USA

4 2 8 9 4 5 12 7 19 103 6 22 14 53 26 44 15 5 15 11 22 4 15 24 22 27 31 47 42 59 120 106 77 22 24 77 90 223 127 89 327

134 92 85 114 99 128 159 114 101 186 162 106 106 152 137 232 125 152 125 114 122 102 118 112 124 115 132 122 143 143 223 168 124 163 179 193 20 162 153 169 327

0.04 0.04 0.08 0.05 0.02 0.18 0.45 0.04 0.38 0.39 0.51 0.19 0.79 0.17 1.06 0.10 1.21 0.84 0.18 1.80 0.78 0.33 1.20 1.34 0.14 1.22 0.67 0.20 0.80 0.84 1.02 1.39 5.17 1.70 0.99 0.36 1.73 2.25 1.52 3.65 27.83

0.05 0.05 0.06 0.06 0.06 0.09 0.10 0.10 0.26 0.26 0.33 0.38 0.39 0.41 0.41 0.42 0.43 0.49 0.50 0.52 0.64 0.65 0.67 0.73 0.75 0.97 1.00 1.03 1.39 1.50 2.35 2.37 2.49 2.65 2.88 3.38 3.50 3.88 5.22 6.30 27.83

Revenue is in millions of 2007 US dollars. 21

The only public data on global ﬁlm trade is UN Comtrade, which lists motion pictures as a commodity, Cinematographic Film Exposed or Developed (SITC 883), capturing the reported value of physical shipments of ﬁlm prints across borders. Physical ﬁlm shipments vastly understate ﬁlm revenues. Comtrade reports 2000 US ﬁlm exports of $0.5 million to France, $0.5 million to Germany, and $6.5 million to the UK, while ScreenDigest reports 2000 box-ofﬁce revenues for US ﬁlms of $513 million in France, $615 million in Germany, and $429 million in the UK (Hancock and Jones, 2003). 22 Most box-ofﬁce revenues are earned shortly after a ﬁlm is released (De Vany and Walls, 1997), suggesting that revenues reported in a given year match the movies released in that year. Some revenue data are available for ﬁlms imported from countries other than the US, but the countries covered vary across destinations (e.g., while the UK is a major importer of movies from India, other countries are not). 23 ScreenDigest deﬁnes the origin country for a ﬁlm by the location of the company that produces the ﬁlm. Filming may occur in multiple locations. Titanic (1997), for instance, was shot in Canada, Mexico and the United States, with most other production activities occurring in Los Angeles. ScreenDigest considers the movie to be US in origin because the production companies, 20th Century Fox and Paramount, are US based. Despite Titanic's ﬁlming locations, it is clearly a US movie. The dialogue is in English, it was ﬁrst released in the US market, and its cultural themes were targeted to a US audience. The cultural discount involved in exporting Titanic to, say, Italy would logically have the US as the reference point. 24 ScreenDigest has box-ofﬁce revenues for nine additional countries (Bulgaria, Chile, Colombia, Egypt, Iceland, India, Israel, the UAE, and Venezuela), but is missing data on the number of movies exhibited for US ﬁlms, domestic ﬁlms, or both, which precludes including these countries in the analysis. 25 These countries are the Dominican Republic and Jamaica in the Caribbean; Costa Rica, El Salvador, Honduras, Nicaragua, and Panama in Central America; Bolivia, Ecuador, Peru and Uruguay in the Andes and Southern Cone; Malaysia and Vietnam in Southeast Asia; Bangladesh, Kazakhstan, Pakistan, and Sri Lanka in Central and South Asia; Jordan, Lebanon, Qatar, and the United Arab Emirates in the Middle East; and Ethiopia, Ghana, Kenya, Nigeria, and Uganda in Sub-Saharan Africa.

distance, levies on ﬁlm imports, quantitative restrictions on ﬁlm imports, and the protection of intellectual property rights. For cultural trade barriers between the United States and its trading partners, we use indicators of linguistic dissimilarity between countries (Melitz, 2002). Following Fearon (2003) and Desmet, Ortuno-Ortin, and Wacziarg (2009), we calculate linguistic distance as 1 minus the expected value of a linguistic similarity factor between a person randomly drawn from the United States and one randomly drawn from country k: LDuk = 1−∑l ∑o plu pok

pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ Glo = 15;

ð18Þ

where l indexes the ethnic groups that speak different languages in the US, o indexes those in country k and plu and pok are the population shares of language groups lpand o in the US and country k. The ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ linguistic similarity factor is Glo = 15, where Glo is the number of branches of the language tree that groups l and o share and 15 is the maximum number of branches. We make linguistic similarity concave with respect to Glo because early divergence in the language tree (e.g., Indo-European vs. Japanese language families) is likely to signify greater cultural difference than later divergence (e.g., Romance vs. Germanic languages). In Section 4, we compare linguistic distance to other language variables. As another measure of culture dissimilarity, we use a related metric of religious distance. Data on the global language tree is from Fearon (2003) and on the global religion tree is

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

from Fearon and Mecham (2007). For additional measures of cultural distance, we use three indices of national values from Hofstede (2001): an individualism index (intensity of perception that social ties between individuals are loose), a masculinity index (strength of belief that men should have assertive roles in society), and a power distance index (willingness to accept an unequal distribution of power).26 In unreported results, we utilize trade barriers facing US movie exports measured using annual reports by the Motion Pictures Association of America (MPAA) to the US Trade Representative. The MPAA report covers over 100 countries, listing for each the policies its members claim adversely affect their business. We include measures of policies for levies and tariffs (whether a country applies tariffs on ﬁlm imports or taxes on royalties for foreign ﬁlms), quantitative restrictions (whether a country applies import quotas on foreign ﬁlms or sets minimum screen time for domestic ﬁlms), the extent of piracy of ﬁlm-related intellectual property (whether estimated revenues from pirated ﬁlms exceed 70% of box-ofﬁce revenues in the country), and restrictions on language (whether a country mandates that foreign-language movies be dubbed locally). We also utilize data on the protection of intellectual property rights (IPRs). McCalman (2005) ﬁnds that whereas moderate IPR protection encourages the spread of US movies, either very weak or very strong IPR protections decrease the speed with which US movies are released abroad.27 The measures of IPR protection we use are the Ginarte and Park (1997) index of patent protection in 2000, the Global Competitiveness Report measure of the strength of IPR protection, averaged over 2003 and 2004, an indicator for whether the US Trade Representative has placed a country on the Priority Watch List for inadequate protection of intellectual property rights in a given year, and ) an indicator for whether a country has entered into force the World Copyright Treaty of the World Intellectual Property Organization. 3.3. Distribution and ﬁxed export costs in motion pictures Estimates of the ﬁxed costs of exporting motion pictures are difﬁcult to obtain. Case studies of how US movies are distributed are informative about the nature and magnitude of these costs. In movie production, the major studios discussed in Section 2 dominate, accounting for 83% of US box-ofﬁce revenues in 2003. Foreign movie distribution is similarly concentrated. The major studios operate ﬁve foreign distribution companies, which distribute nearly all US movies abroad (Epstein, 2006). Distribution companies handle not just movies produced by their own studios but movies from other studios, as well. For instance, United International Pictures, a joint venture of Universal and Paramount, has distribution facilities in over 35 countries, handling ﬁlms produced by its principals, other major studies, and independent studios (Scott, 2004). McCalman (2004, 2005) provides contractual details on foreign movie distribution. The Motion Picture Association (2003) reports that for the average US movie in 2003 65% of total costs were due to ﬁlm production and 35% were due to marketing, which includes making ﬁlm prints; advertising on radio, TV, newspapers and other media; and promotional activities. These costs encompass just domestic marketing. Foreign marketing brings additional expenses, which average 60% of domestic marketing costs. Key components of a marketing campaign include the trailer, used to advertise in theatres, on television, and via 26 These indices are based on surveys IBM conducted of its global employees in 70 countries in the 1960s and 1970s. US values (country averages) are 91 (44) for the individualism index, 62 (51) for the male dominance index, and 40 (59) for the power distance. Relative to other countries, US respondents tend to be more likely to perceive social ties as being weak, to have stronger beliefs that men should have assertive roles in society, and to be more willing to accept an unequal distribution of power. 27 In related work, McCalman (2004) ﬁnds that while Hollywood studios are more likely to use licensing arrangements in countries with moderate IPR protection, they tend to use more integrated governance structures in countries with either high or low IPR protection.

21

the internet, and the “hook,” a phrase or image that summarizes the marketing pitch (e.g., Arnold Schwarzenegger declaring, “Hasta la vista, baby,” in Terminator 2) (Epstein, 2006). Developing the trailer and the hook, to which studios devote considerable resources, are ﬁxed distribution costs. Promotional materials can be used across national markets by translating phrases and bits of dialogue. In the United States, marketing includes intensive television and radio advertising leading up to a ﬁlm's release. Abroad, studios tend to eschew such saturation ads and use other forms of publicity instead, such as pre-release appearances by a ﬁlm's stars at international venues (Seagrave, 1997). These appearances, often stipulated in actors' contracts, generate signiﬁcant global press coverage. The trailer, the hook, actor appearances, and other promotional devices represent global distribution costs. Other distribution expenses are speciﬁc to individual markets. These include negotiating with local theatre chains over the number of cinemas in which a ﬁlm will appear, how long a ﬁlm will run, and how ﬁlm revenues will be shared between the exhibitor and the distributor (Gil and Lafontaine, 2009); coordinating release dates across regions; producing foreignlanguage subtitles or dubbing movie dialogue; and custom editing ﬁlms for speciﬁc countries (Scott, 2004). The literature thus contains evidence of both global and bilateral ﬁxed export costs. We leave it to our empirical approach to determine the relative importance of these in determining how trade ﬂows adjust to trade barriers.

3.4. Preliminary data analysis For the countries in our sample, the US is the largest source for movie imports. Fig. 3 shows that US movies account for over 60% of total box-ofﬁce revenues (domestic plus foreign movie sales) in all countries except China, France, the Philippines, and Korea.28 In all but six countries, the US accounts for over 40% of the number of movies exhibited. China is clearly an outlier. Its government permits no more than 20 foreign ﬁlms to be released in the country each year.29 Other countries also place limits on ﬁlm imports, but they tend to be much less restrictive. Between 2001 and 2006, 12 countries set minimum requirements for the amount of screen time devoted to domestic ﬁlms. In only four countries are time limits in excess of 30% and even in these cases theatres can often circumvent limits by showing domestic features in the afternoon, on weekdays, or other times when demand is slack (Chung and Song, 2008). An alternative strategy is to take on a passive domestic partner for the distribution of a movie, which often allows a ﬁlm to qualify for domestic treatment (Waterman, 2005). Table 2 gives summary statistics for the key variables used in the analysis. During this period, 327 new domestically produced movies were shown on average in the US each year. Consistent with the presence of ﬁxed exports costs of some kind, the typical country in our sample imports less than half of US movies produced annually, with the mean number of US movies exhibited equal to 142. Most countries are clustered around this mean, with the country at the 20th percentile (Hungary) importing 106 US movies annually and the country at the 80th percentile (Singapore) importing 162 movies. In contrast, box-ofﬁce revenues per movie show wide variation. Mean revenues per movie are $1.24 million (in 2007 US dollars), with the country occupying the 20th percentile (Slovenia) at $0.10 million and the country occupying the 80th percentile (Italy) at $2.37 million. The ratio of the 80th to the 20th percentile for the number of US movies imported is 1.5, compared to 23.7 for box-ofﬁce revenues per US ﬁlm. 28 In the ﬁgures, we report data for 2001–2006, the years for which we have complete data for most countries. For the estimation results reported in the next section, we use the full sample period. 29 All regression results in Section 4 are robust to excluding China from the estimation sample.

France

.4

Philippines South Korea

2

Belgium Spain Canada Philippines Australia Italy UK Mexico Singapore France Germany Argentina New Zealand Brazil RussiaSouth Thailand Netherlands Portugal Austria Korea Turkey Poland Switzerland Norway Sweden Indonesia Denmark Finland Hungary Ireland Czech Republic

China

-4

0

.2

China

Malaysia Lithuania Iceland Romania Slovenia Slovak Republic Latvia Estonia

0

.6

.8

Slovenia AustraliaRomania Russia Iceland ArgentinaMexico Lithuania Hungary Brazil Republic Latvia Slovak Singapore UK Germany Portugal Estonia Austria Czech Republic Turkey Netherlands Norway Indonesia Ireland Poland Finland Spain Denmark Sweden Switzerland Malaysia Thailand Belgium Italy

-2

New Zealand Canada

Log number of US films imported

1

4

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

Share of US in Box Office Revenues

22

0

.2

.4

.6

.8

1

Share of US in No. of Films Exhibited

-2

0

2

4

Log box-office revenues per US film

Fig. 3. Share of US ﬁlms in national movie expenditure, 2001–2006.

Though the data reveal variation at both the intensive margin (revenues per ﬁlm) and extensive margin (number of ﬁlms) of trade, it is clear the former dwarfs the latter. The wider variation in the intensive over the extensive margin is also seen in Fig. 4, which plots log revenues per US movie against log number of US movies (each expressed as the deviation from sample means), averaged by country over 2001–2006. The standard deviation for revenues per movie is 1.39 compared to 0.38 for the number of movies. Fig. 4 gives evidence consistent with global ﬁxed export costs mattering more than bilateral ﬁxed export costs. More direct evidence comes from how intensive and extensive margins of trade vary with importing-country characteristics. When global ﬁxed export costs dominate, the number of US movies imported should be uncorrelated with importer characteristics. Fig. 5a and b plot the number of US movies imported and box-ofﬁce revenues per US movie (relative to cross-country means) against importing-country GDP and linguistic distance from the US. Whereas revenues per US ﬁlm have a positive correlation with importer GDP and a negative correlation with importer linguistic distance (Fig. 5b), the correlation between the number of US ﬁlms imported and these characteristics is weak and Table 2 Summary statistics. Variable

N

Mean

St. Dev.

Number of US ﬁlms Number of domestic ﬁlms Revenue per US ﬁlm Revenue per domestic ﬁlm Log average sales ratio Linguistic dissimilarity index Log linguistic dissimilarity index English ofﬁcial language Island Latitude difference with US Longitude difference with US Log distance to US Log GDP Log population Ginarte–Park index Intellectual property protection index Super 301 action Ever sign WCT Has levy or tariff on foreign ﬁlms Has quantitative restrictions on foreign ﬁlms Has high level of ﬁlm piracy Imposes restrictions on language dubbing Individualism index Masculinity index Power distance index

284 284 284 284 284 284 284 284 284 284 284 284 284 284 271 284 284 284 284 284 284 284 273 273 273

141.366 46.345 1.293 0.859 0.386 0.751 − 0.304 0.095 0.074 16.609 104.237 − 0.068 26.146 16.767 0.723 0.695 0.141 0.345 0.095 0.282 0.116 0.092 0.592 0.498 0.516

41.483 52.865 1.506 0.965 0.958 0.143 0.185 0.294 0.262 16.714 47.48 0.414 1.328 1.215 0.156 0.161 0.348 0.476 0.294 0.451 0.332 0.289 0.208 0.234 0.217

Sample is 284 observations on 46 countries over period 1995–2006.

-4

Fig. 4. Intensive versus extensive margin of movie imports, 2001–2006.

statistically insigniﬁcant (Fig. 5a). Similar results hold for geographic distance. Importer characteristics appear to be associated with trade through the intensive rather than extensive margin. To examine the intensive and extensive margins more formally, we follow Eaton et al. (2004) and use the identity, Nuit × (Ruit/Nuit) ≡ (Ruit/Rit) × Rit, where Ruit is box-ofﬁce revenues for US movies in country i at time t, Nuit is the number of US movies shown in i at t, and Rit is aggregate box-ofﬁce revenues in i at t. We then estimate the following two regressions (with robust t statistics in parentheses): ln Nuit = 0:32 ln ðRuit = Rit Þ + 0:12 ln Rit + εit ð1:74Þ ð12:60Þ ln ðRuit = Nuit Þ = 0:68ln ðRuit = Rit Þ + 0:88 ln Rit −εit ð3:67Þ ð95:06Þ where we express variables as deviations from the mean to eliminate the constant. By the logic of least squares, across the two regressions the constant and error terms sum to zero and the coefﬁcients on each variable sum to one. The relative magnitude of the coefﬁcients describes how the number of movies (the extensive margin) and revenues per movie (the intensive margin) vary across countries according to market size. A 10% increase in US market share in a country is associated with an increase in revenue per US movie of 6.8% and an increase in the number of movies of 3.2%; a 10% increase in total market size in a country is associated with an increase in revenues per US movie of 8.8% and an increase in the number of movies of only 1.2%. This is further evidence most adjustment in US movie exports occurs at the intensive margin. The theoretical results presented in Section 2 suggest a simple way to identify the nature of ﬁxed trade costs is to examine the sign of the correlation between trade barriers and the average sales ratio (average revenue per US movie/average revenue per domestically made movie). Fig. 6a and b plot the average sales ratio against geographic distance to the US and linguistic dissimilarity with the US. Geographic and linguistic distance each has a negative correlation with the average sales ratio, which is consistent with global ﬁxed export costs being dominant (Eq. (14)) and inconsistent with bilateral ﬁxed export costs being dominant (Eq. (16)). We do not know whether geographic and linguistic distance affect variable or ﬁxed trade barriers. Yet, from our theoretical predictions, it appears distance affects bilateral trade along the intensive margin, such that in motion pictures the relevance of distance for trade is in how it affects variable trade barriers. There is some variation at the extensive margin, which, strictly speaking, is inconsistent with Eq. (14), but it is quite small by comparison to the intensive margin.

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

-2

China

-4

-2

0

2

4 2 0

Iceland

-2

0

Belgium Spain Canada Philippines Australia Italy UKGermany Mexico Singapore France Malaysia Argentina New Zealand BrazilKorea Russia Thailand Netherlands IcelandLithuania Portugal Austria South Turkey Poland Switzerland Norway Sweden Romania Slovenia Indonesia Denmark Finland Hungary Czech Ireland Republic Slovak Republic Latvia Estonia

UK Germany China CanadaFrance Australia Mexico South SpainKorea Italy Russia Brazil Switzerland Netherlands Sweden Austria Denmark Norway Ireland Poland Indonesia Portugal New Zealand Turkey B elgium Thailand Argentina Finland Hungary Singapore Philippines Czech Republic Slovenia

4

-4

-2

Log GDP Fitted values

0

Actual values

4

.2

.4

Fitted values

4

Fitted values

2

UK Australia

Germany Canada

France Mexico

China South Korea

Spain Italy

0

Russia Brazil Switzerland Netherlands Sweden Austria Denmark Ireland Norway Poland Portugal New Zealand Belgium Argentina Czech Republic

Indonesia Turkey Thailand Finland Hungary Singapore Philippines

-2

Slovenia Romania Slovak Republic Latvia Lithuania

Malaysia Estonia

-4

-2

-.2

Log linguistic dissimilarity with US

Log box-office revenues per US film

2 0

Philippines Singapore Malaysia Thailand South Korea Turkey Indonesia Finland Hungary Estonia

China

-.4

2

Actual values

-4

Log number of US films imported

Belgium Spain Canada Italy FranceMexico Germany New Zealand Argentina Brazil Russia Netherlands PortugalLithuania Austria Switzerland Poland Norway Sweden Romania Slovenia Denmark Ireland Czech Republic Slovak Republic Latvia

UK Australia

0

Log GDP

4

Actual values

Malaysia

Romania Estonia Slovak Republic LatviaLithuania

-4

2

Log box-office revenues per US film

4

(b)

-4

Log number of US films imported

(a)

23

-.4

-.2

0

.2

.4

Log linguistic dissimilarity with US Actual values

Fitted values

Fig. 5. Movie imports and importing-country characteristics, 2001–2006. (a) Number of US movies imported. (b) Box-ofﬁce revenues per US movie.

4. Estimation results 4.1. Basic results Table 3 presents estimation results for Eq. (17). The dependent variable is the log average sales ratio (average box-ofﬁce revenue per US movie relative to average box-ofﬁce revenue per domestic movie). Standard gravity trade barriers appear as regressors. The sample consists of 46 countries for the years 1995–2006, with data for some countries not beginning until after 2000. We ﬁrst examine the role of linguistic distance. In column 1, the log linguistic dissimilarity index (from Eq. (18)), is negative and precisely estimated,30 indicating that the more linguistically different a country is from the United States the less it spends per US movie it imports, relative to domestic movies. Simply by adding linguistic distance to a regression with year dummies, the explained variation rises from 5% to 27%. The coefﬁcient estimates imply that a country with linguistic dissimilarity one standard deviation below the mean (Switzerland) would have sales per US movie that are 66 log points higher than a country at one standard deviation above the mean (Estonia). To see whether this result is driven by use of English, column 2 includes a dummy variable that equals 1 if a country has English as its primary language. The English dummy is positive and

30

Results are similar when we measure the index in levels rather than logs.

statistically signiﬁcant at the 10% level, suggesting that Englishspeaking countries spend more per US movie than non-English speaking countries. In column 3, which includes both language variables, the English dummy loses signiﬁcance while linguistic dissimilarity remains precisely estimated. It appears linguistic distance captures more than whether two countries speak the same language, also picking up other aspects of cultural dissimilarity and divergent national experience. In subsequent speciﬁcations, we use linguistic dissimilarity to measure linguistic distance. Columns 4–6 of Table 3 examine the role of geographic distance. Coefﬁcients on linguistic dissimilarity become smaller in magnitude with the inclusion of geographic distance but remain highly signiﬁcant. The ﬁrst geographic distance variable is a dummy that equals 1 if a country is an island (and so is likely to have a history of trade with culturally distinct nations); its coefﬁcient is positive and statistically signiﬁcant in all speciﬁcations.31 We then consider two sets of variables, log great circle distance to the US and the absolute values of longitudinal and latitudinal differences with the US. For both sets of variables, the coefﬁcients are negative and precisely estimated, meaning countries farther away from the US spend less per US movie. In column 6, which includes both sets of distance variables together, great circle distance loses signiﬁcance while longitude and latitude 31 See Feyrer and Sacerdote (2009) on how the exposure of islands to international shipping affects their involvement in commerce and Rose (2005) on how being an island promotes international trade.

24

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

2 1

Switzerland Austria Belgium Portugal Slovak Republic Germany Slovenia RussiaArgentina Spain Hungary Ireland UKFrance Italy Brazil Netherlands LatviaRomania

0

Mexico

Australia China

Lithuania Poland Estonia Czech Republic Sweden Norway Denmark Finland

Iceland

-1

Turkey

Singapore Philippines New Zealand South Korea Thailand Indonesia Malaysia

-2

Log relative box office revenue per film

(a)

-1

-.5

0

.5

1

Log distance to US (mean) lsd

Fitted values

2

Canada

4.2. Global versus bilateral ﬁxed export costs

AustriaSwitzerland Belgium Slovak Republic Portugal

Germany

1

Australia UK

0

Netherlands

Slovenia Russia Argentina Spain Ireland BrazilItaly FranceMexico

Hungary China

Latvia Romania

Lithuania PolandRepublic Czech Sweden

Estonia

Singapore Philippines Finland South Korea Thailand Turkey

Norway Denmark

-1

New Zealand

Indonesia Malaysia

-2

Log relative box office revenue per film

(b)

latitudes). Demand for US movies thus appears to be stronger in countries that are climatically more similar to the US. One interpretation of this result is that movie exports are larger to regions that are on the same seasonal cycle as the United States. For instance, US studios tend to release blockbuster action ﬁlms in the early US summer (to attract youth and families on summer vacation), which may align with consumer patterns elsewhere in the northern hemisphere. In subsequent speciﬁcations, we use longitude and latitude to measure geographic distance. It may be surprising that the average sales ratio for US movie exports is negatively correlated with geographic distance. After all, the physical cost of shipping ﬁlms across borders is limited to the cost of transporting a few master ﬁlm prints, suggesting much less role for distance in trade than for commodities or manufactured goods. Yet, our results are consistent with Blum and Goldfarb (2006), who ﬁnd that trade in internet services is also negatively correlated with distance. Even in services with apparently low physical transport costs, geographic distance depresses trade. This may reﬂect the importance of transaction costs in negotiating contracts between US movie distributors and foreign movie exhibitors, as noted by Gil and LaFontaine (2009).

-.4

-.2

0

.2

.4

Log linguistic dissimilarity with US (mean) lsd

Fitted values

Fig. 6. Relative average revenue per US movie and trade barriers, 2001–2006. (a) Distance to the US. (b) Linguistic dissimilarity with the US.

remain statistically signiﬁcant. The coefﬁcient on distance in latitude is about twice as large as that on distance in longitude, suggesting barriers to trade in US movies are greater in going from the northern to southern hemisphere (i.e., going from temperate latitudes to the tropics) than in crossing a major ocean (i.e., staying within temperate

In accordance with Fig. 6, the regression results in Table 3 show that the average sales ratio is negatively correlated with common measures of trade barriers. Countries more distant from the US – in terms of geography or language – have lower sales per US movie (relative to sales of domestic movies). This suggests geographic and linguistic distance matter for trade through variable trade barriers, rather than through ﬁxed trade costs. These results conﬁrm that adjustment in motion picture trade primarily occurs along the intensive margin. With respect to the theoretical model presented in Section 2, the data suggest that global ﬁxed export costs dominate for most countries in our data. However, these results do not rule out the possibility that bilateral ﬁxed export costs dominate for a subsample of countries. It could be that, on average, adjustment to trade costs occurs along the intensive margin, but that for a subset of countries bilateral ﬁxed export costs dominate and adjustment occurs along the extensive margin. To examine this possibility, we partition our sample into a high Q-index sub-sample and a low Q-index sub-sample. Our theory suggests that the countries with a relatively low Q index are ones for which bilateral ﬁxed export costs are mostly likely to dominate. These are countries that have relatively small markets and/or relatively high variable trade costs vis-à-vis the United States. To aide in constructing

Table 3 Average sales ratio and gravity trade barriers. (1) Linguistic dissimilarity

(2)

− 2.448⁎⁎ (0.540)

English ofﬁcial language

0.543 (0.279)

(3)

(4)

(5)

(6)

− 2.725⁎⁎ (0.666) − 0.341 (0.409)

− 1.922⁎⁎ (0.545)

− 1.831⁎⁎ (0.518)

− 1.879⁎⁎ (0.555)

1.524⁎⁎ (0.306) − 0.015⁎⁎

0.654⁎ (0.269)

1.468⁎⁎ (0.300) − 0.014⁎⁎

Island Latitude diff. with US

(0.004) − 0.008⁎⁎ (0.002)

Longitude diff. with US Log distance to US R2 N

0.269 284

0.073 284

0.276 284

0.381 284

− 0.788⁎⁎ (0.150) 0.340 284

(0.005) − 0.007⁎⁎ (0.002) − 0.179 (0.229) 0.383 284

The speciﬁcation is that in Eq. (17). The dependent variable is log sales per US movie minus log sales per domestic movie. Coefﬁcient estimates for year dummies are not shown. Standard errors (clustered by importing country) are in parentheses. * (**) indicates statistical signiﬁcance at the 5% (1%) level.

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26 Table 4 Results partitioning countries by market size and distance to US.

30

(a)

Q index 0 0 0

28

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0

26

log GDP

25

0 0 00 0 0 0 0 0 0

0 0 0 0 0 0 0 00 0 0 0 0 00 0 0

0 0 0

24

0 0

0 0 0 0 0 0

0 0 0 0 0 1 1 1

0 0 00 0 0 0 1 1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

1 1 1 1 1 1

22

1 1 1

-.6

-.8

Low

High

Low

(2a)

(2b)

(3a)

(3b)

− 1.922** − 1.884* − 2.920 − 1.670* − 2.587* (0.545) (0.816) (1.895) (0.625) (1.079) Island 1.524** 2.111** 1.160** 2.162** 1.073** (0.306) (0.498) (0.251) (0.519) (0.327) Latitude diff. with US − 0.015** − 0.018** − 0.013 − 0.016** − 0.022 (0.004) (0.007) (0.019) (0.005) (0.017) Longitude diff. with US − 0.008** − 0.009** − 0.010 − 0.009** − 0.006 (0.002) (0.002) (0.006) (0.002) (0.006) R-squared 0.381 0.263 0.576 0.288 0.703 N 284 213 71 238 46 Linguistic dissimilarity

0 0 0 0 0 0 0 0 0 0 0

(1)

0

0 0 0 00 0 0

Q index

High

-.4

-.2

0

log linguistic dissimilarity

30

(b)

The dependent variable is log sales per US movie minus log sales per domestic movie. Column 1 shows results from the full sample (identical to column 4 of Table 3); column 2a (2b) shows results for countries with a high (low) Q index based on the partition of countries according to size and distance in Fig. 7a; and column 3a (3b) shows results for countries with a high (low) Q index based on the partition of countries according to size and distance in Fig. 7b. Standard errors (clustered by importing country) are in parentheses. * (**) indicates statistical signiﬁcance at the 5% (1%) level.

0

28

0 0 0 0 0 0 0 0 0 0 0

0 0

0 0 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0

26

0 0

0 0 0

0 0 0 0 0 0 0 1 1 11 1 1 1 1 1 1 1 1 1 1 1 1 1 1

0 0 0 0 0 0

0 0 0

0 0 0 0 0 0 0 0

1 1

1 1 1 1 1 1 1

0 0 0 0

11 1 1 1 1

22

24

log GDP

0 0 0

-1

-.5

0

.5

1

log geographic distance Fig. 7. a: Country size and linguistic distance from the US. b: Country size and geographic distance from the US.

the partition, Fig. 7 plots log GDP, an obvious measure of market size, against either log linguistic distance from the United States (panel a) or log geographic distance from the United States (panel b). Points in the graphs indicated by a one are those that have either a relatively small GDP (deﬁned as a value below the 20th percentile for the sample), or a combination of high distance (deﬁned as above the 80th percentile for the sample) and moderately low GDP (deﬁned as below the 40th percentile for the sample). The sets of observations deﬁned by these cutoffs represent clear breaks in the bivariate distribution of size and distance, and they form our low Q-index sub-sample.32 The other observations are our high Q-index sub-sample. In unreported results, we examined alternative data partitions and found very similar results to those reported below.33 Table 4 reports the regression from column (4) of Table 3 for ﬁve samples of countries: the full sample (column 1); countries with a low or high Q index based on linguistic distance, as seen in Fig. 7a 32 Based on linguistic distance (as in Fig. 5a), the low Q-index sub-sample includes Estonia, Finland, Hong Kong, Hungary, Indonesia, Latvia, Lithuania, Malaysia, Philippines, Singapore, Slovak Republic, Slovenia, Thailand and Turkey. The low Q-index sub-sample based on geographic distance (as in Fig. 5b) is the same set of countries minus Finland, Hungary and Turkey, and plus New Zealand. 33 In this approach we select the two sub-samples ex ante. In unreported results we also estimate a switching regression model (Kiefer 1978; Hartley 1978), where the partition of the sub-samples is estimated by the estimation procedure. The results are similar to Table 4.

(columns 2a, 2b); and countries with a low or high Q index based on geographic distance, as seen in Fig. 7b (columns 3a, 3b). In all columns, the average sales ratio is negatively correlated with trade barriers. For countries with a high Q index (columns 2a, 3a), all coefﬁcients are precisely estimated; for countries with a low Q index (columns 2b, 3b), some coefﬁcients are imprecisely estimated but the sample sizes for these groups are much smaller, producing larger standard errors for the coefﬁcient estimates. There is no evidence that the sign of the correlation between distance and the average sales ratio differs between countries according to their size and distance from the United States. In fact, the negative quantitative impact of linguistic distance on the average sales ratio is larger for countries with a lower Q index, the opposite of what Eqs. (14) and (16) predict. These results suggest that global ﬁxed export costs dominate for even the smaller and more distant countries in our data, and that bilateral ﬁxed export costs do not have signiﬁcant effects on export decisions.34 In unreported results, we examined the relationship between the average sales ratio and other trade barriers. We found no relationship between the average sales ratio and indicators of the protection of intellectual property rights (the Ginarte–Park patent protection index, the Global Competitiveness Report measure of the strength of IPR protection, an indicator for whether a country is on the US Priority Watch List, or an indicator for whether a country has signed the World Copyright treaty) or indicator variables based on MPAA reports for whether a country imposes barriers on ﬁlm imports (has levies or tariffs on movies, imposes quantitative restrictions on foreign ﬁlms, is subject to high levels of piracy in movie related intellectual property, or imposes language restrictions on the dubbing of foreign ﬁlms). We did ﬁnd evidence that box-ofﬁce revenues for US movies are stronger in countries in which beliefs favor men having assertive roles in society, based on measures from Hofstede (2001).35

5. Discussion Fixed export costs ﬁgure prominently in recent theoretical trade models. Although current empirical research using manufacturing data suggests that these costs exist, it has little to say about their 34 These results are consistent with either (a) the set B is empty, or (b) the set B has one or a handful of small and distant countries. Since case (b) provides only a handful of data points it is hard to empirically distinguish from case (a). 35 Action ﬁlms with dominant male characters (e.g., Dirty Harry, Rambo, The Terminator, Gladiator) have long been a staple of US movie studios, consistent with this ﬁnding.

26

G. Hanson, C. Xiang / Journal of International Economics 83 (2011) 14–26

nature. In services, which are undergoing rapid growth in trade, we know little about the impediments to global commerce. We extend the Melitz (2003) model to allow for both global and bilateral ﬁxed export costs. Data on US ﬁlm exports support global ﬁxed export costs over bilateral ﬁxed export costs. Trade in movies adjusts primarily along the intensive margin. Even small countries import large numbers of US ﬁlms, leaving only modest variation in the extensive margin of trade. Along the intensive margin, average revenues per US ﬁlm vary widely across countries and are negatively correlated with geographic distance, linguistic distance, and other measures of trade barriers. Our results depart from ﬁndings in the literature for trade in manufacturing, in which adjustment in trade occurs primarily at the extensive margin. One explanation for the difference in results between sectors is that ﬁxed trade barriers for information services (as characterized by motion pictures) are dissimilar to those for manufacturing. For information services, products are delivered through devices consumers already own (TVs, radios, computers, cell phones) or an existing infrastructure that can be shared by producers (theatres, telephone lines, satellite networks), with products requiring little in the way of post-sale service. Fixed barriers to trading information services include developing a marketing strategy to attract consumers and negotiating contracts over the delivery of intellectual property. Once a marketing strategy and standard contract have been developed for one market, they may be easy to replicate in other markets, making the ﬁxed costs of delivering information services primarily global in nature. Acknowledgments We thank David Hummels, Marc Melitz, Phillip McCalman, Nina Pavcnik, Stephen Redding, and seminar participants at UCSD, the NBER, the Princeton University IES Workshop and the 2008 EIIT Meetings for helpful comments; Adina Ardelean, Sirsha Chatterjee, Anca Cristea, Michael Ewens, and Jeffrey Lin for excellent research assistance; and the National Science Foundation for ﬁnancial support. References Anderson, James E., van Wincoop, Eric, 2004. Trade costs. Journal of Economic Literature 42 (3), 691–751. Arkolakis, Costas, 2007. “Market Penetration Costs and the New Consumers Margin in International Trade.” Mimeo, Yale University Baldwin, Richard E., Harrigan, James, 2007. Zeroes, Quality, and Space: Trade Theory and Trade Evidence: NBER Working Paper, 11471. Bernard, Andrew B., Jensen, J. Bradford, 1999. Exceptional exporter performance: cause, effect, or both. Journal of International Economics 47, 1–25. Bernard, Andrew B., Jensen, J. Bradford, 2004. Why some ﬁrms export. Review of Economics and Statistics 86, 561–569. Bernard, Andrew B., Redding, Stephen, Schott, Peter K., 2006. “Multiproduct Firms and Trade Liberalization.” Mimeo, Yale University Bernard, Andrew B., Jensen, J. Bradford, Redding, Stephen, Schott, Peter K., 2007. Firms in International Trade: NBER Working Paper, 13054.

Blum, Bernardo, Goldfarb, Avi, 2006. Does the internet defy the law of gravity? Journal of International Economics 70, 384–405. Chung, Chul, and Minjae Song. 2008. “Preference for Cultural Goods: Demand and Welfare in the Korean Film Market.” Mimeo, KIEP. De Vany, Arthur, 2004. Hollywood Economics. Routledge, London. De Vany, Arthur, Walls, David W., 1997. The market for motion pictures: rank, revenue, and survival. Economic Inquiry 35 (4), 783–797 October. De Vany, Arthur, Walls, W. David, 2004. Motion picture proﬁt, the stable Paretian hypothesis, and the curse of the superstar. Journal of Economic Dynamics and Control 28 (6), 1035–1057. Desmet, Klaus, Ignacio Ortuno-Ortin, Romain Wacziarg, 2009. "The Political Economy of Ethnolinguistic Cleavages." NBER Working Paper No. 15095. Eaton, Jonathan, Kortum, Samuel, Kramarz, Francis, 2004. Dissecting Trade: Firms, Industries, and Export Destinations: NBER Working Paper, 10344. Epstein, Edward Jay, 2006. The Big Picture: Money and Power in Hollywood. Random House, New York. Fearon, James D., 2003. Ethnic and cultural diversity by country. Journal of Economic Growth 8 (2), 195–222. Fearon, James, Robert Q. Mecham. 2007. "Religious Classiﬁcation and Data on Shares of Major World Religions." Mimeo, Stanford University. Feyrer, James, Sacerdote, Bruce, 2009. Colonialism and modern income: islands as natural experiments. Review of Economics and Statistics 91 (2), 245–262. Gil, Ricard, and Francine LaFontaine. 2009. “Using Revenue Sharing to Implement Flexible Pricing: Evidence from Movie Exhibition Contracts.” Mimeo, UC Santa Cruz Ginarte, Juan C., Park, Walter G., 1997. Determinants of patent rights: a cross-national study. Research Policy 26, 283–301. Hancock, David, Jones, Charlotte, 2003. Cinema Distribution and Exhibition in Europe, 2nd Edition. ScreenDigest, London. Hanson, Gordon H., Xiang, Chong, 2008. Testing the Melitz Model of Trade: An Application to US Motion Picture Exports: NBER Working Paper, 14461. Hartley, Michael J., 1978. Estimating mixtures of normal distributions and switching regressions: comment. Journal of the American Statistical Association 73, 738–741. Helpman, Elhanan, Krugman, Paul, 1985. Market Structure and Foreign Trade. MIT Press, Cambridge, MA. Helpman, Elhanan, Melitz, Marc J., Rubinstein, Yona, 2007. Estimating Trade Flows: Trading Partners and Trading Volumes: NBER Working Paper, 12927. Helpman, Elhanan, Melitz, Marc J., Yeaple, Stephen R., 2004. Export versus FDI with heterogeneous ﬁrms. American Economic Review 94, 300–316. Hofstede, Geert, 2001. Culture's Consequences, Comparing Values, Behaviors, Institutions, and Organizations Across Nations. Sage, Thousand Oaks CA. Kiefer, Nicholas M., 1978. Discrete parameter variation: efﬁcient estimation of a switching regression model. Econometrica 46, 427–434. Marvasti, Akbar, Canterbery, E. Ray, 2005. Cultural and other barriers to motion pictures trade. Economic Inquiry 43 (1), 39–54. Motion Picture Association, 2003. “US Entertainment Industry: 2003 Market Statistics.” Mimeo, MPA. McCalman, Philip, 2004. Foreign direct investment and intellectual property rights: evidence from Hollywood's global distribution of movies and videos. Journal of International Economics 62 (1), 107–123. McCalman, Phillip, 2005. International diffusion and intellectual property rights: an empirical analysis. Journal of International Economics 67 (2), 353–372. Melitz, Marc J., 2003. The impact of trade on intra-industry reallocations and aggregate industry productivity. Econometrica 71 (6), 1695–1725. Rauch, James E., Trindade, Vitor, 2009. Neckties in the tropics: a model of international trade and cultural diversity. Canadian Journal of Economics 42, 809–843. Rose, Andrew, 2005. Which international institutions promote international trade? Review of International Economics 13 (4), 682–698. Seagrave, Kerry, 1997. American Films Abroad: Hollywood's Domination of the World's Movie Screens. McFarland, Jefferson, NC. Scott, Allen J., 2004. Hollywood and the world: the geography of motion-picture distribution and marketing. Review of International Political Economy 11 (1), 33–61. Waterman, David, 2005. Hollywood's Road to Riches. Harvard University Press, Cambridge, MA.

Journal of International Economics 83 (2011) 27–36

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

Does intellectual property rights reform spur industrial development?☆ Lee Branstetter a,e,⁎, Ray Fisman b,e,1, C. Fritz Foley c,e,2, Kamal Saggi d,3 a Carnegie Mellon University, H. John Heinz III College, School of Public Policy and Management, and Department of Social and Decision Sciences, 5000 Forbes Avenue, Pittsburgh, PA 15213, United States b Columbia University, Columbia Business School, 3022 Broadway, Uris Hall 622, New York, NY 10027, United States c Harvard University, Harvard Business School, Soldiers Field Road, Boston, MA 02163, United States d Vanderbilt University, VU Station B #351828, 2301 Vanderbilt Place, Nashville, TN 37235, United States e NBER, United States

a r t i c l e

i n f o

Article history: Received 2 September 2008 Received in revised form 1 September 2010 Accepted 2 September 2010 Available online 15 September 2010 JEL classiﬁcation: F2 F1 K11 O3

a b s t r a c t An extensive theoretical literature generates ambiguous predictions concerning the effects of intellectual property rights (IPR) reform on industrial development. The impact depends on whether multinational enterprises (MNEs) expand production in reforming countries and the extent of decline in imitative activity. We examine the responses of U.S.-based MNEs and domestic industrial production to a set of intellectual property rights reforms in the 1980s and 1990s. Following reform, MNEs expand the scale of their activities. MNEs that make extensive use of intellectual property disproportionately increase their use of inputs. There is an overall expansion of industrial activity after reform, and highly disaggregated trade data indicate higher exports of new goods. These results suggest that the expansion of multinational activity more than offsets any decline in imitative activity. © 2010 Elsevier B.V. All rights reserved.

Keywords: Intellectual property rights Multinational enterprises Foreign direct investment Industrial development Production shifting

1. Introduction Do stronger intellectual property rights (IPR) spur industrial development? Over the last two and a half decades, policy makers have debated the beneﬁts of IPR reform.4 One of the central concerns raised in these debates is that stronger IPR would curtail the ability of local ﬁrms to imitate and build on the advanced technologies of foreign ﬁrms, potentially slowing economic progress. This was a common concern in the discussions of the 1995 Agreement on Trade Related Aspects of Intellectual Property Rights (TRIPS) that required members

☆ The statistical analysis of ﬁrm-level data on U.S. multinational enterprises was conducted at the Bureau of Economic Analysis, U.S. Department of Commerce, under arrangements that maintain legal conﬁdentiality arrangements. The views expressed herein are those of the authors and do not reﬂect the ofﬁcial positions of the U.S. Department of Commerce. ⁎ Corresponding author. Tel.: + 1 412 268 4649. E-mail addresses: [email protected] (L. Branstetter), [email protected] (R. Fisman), [email protected] (C.F. Foley), [email protected] (K. Saggi). 1 2 3 4

Tel.: + 1 212 854 9157. Tel.: + 1 617 495 6375. Tel.: + 1 615 322 3237. An excellent overview of the debate is provided in Maskus (2000).

0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.09.001

of the World Trade Organization to comply with a set of minimum standards of IPR.5 However, these costs could be partially offset by beneﬁts that arise from increased investment and production by multinational enterprises (MNEs). Stronger IPR could induce MNEs to expand their scale of operations, manufacture technologically sophisticated goods, and quicken the rate of shifting production of existing goods to IPR-reforming countries. In this paper, we empirically assess the effects of stronger IPR on industrial development. Our work is motivated by a rich theoretical literature on the global effects of IPR reform. Helpman (1993) develops several variants of a North–South general equilibrium product cycle model in which Northern innovation expands the range of differentiated goods produced in the world while Southern imitation leads to North–South production shifting. A robust ﬁnding of this analysis is that stronger IPR protection is never in the interest of the South. If stronger IPR in the

5 Another major area of concern focused on the high prices ﬁrms might be able to charge for patent-protected goods under strong IPR. The impact of IPR reform on prices and consumer welfare has been the focus of an extensive literature. See, for example, Maskus (2000), McCalman (2001), and Chaudhuri et al. (2006). In contrast, relatively little empirical work has focused on the potential impact of IPR reform on industrial development.

28

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

South is treated as a reduction in the rate of Southern imitation and Northern ﬁrms are assumed to not shift production to the South through foreign direct investment (FDI), Southern IPR reform lowers the rate of Northern innovation and thereby limits the portfolio of products available globally. 6 If North–South FDI is permitted, a reduction in Southern imitation leads to more FDI but hurts the South because Northern multinationals charge higher prices than Southern imitators. Lai (1998) extends Helpman (1993) to allow both the level of FDI and Northern innovation to respond endogenously to changes in the strength of Southern IPR protection, and this model is further extended in Branstetter et al. (2007), to the case where innovation, FDI, and imitation are all endogenously determined. In these extensions, unlike Helpman (1993), in any equilibrium with a positive rate of imitation, North–South FDI does not lead to factor price equalization. A lower wage in the South creates an incentive to move production of existing varieties there, but multinationals seeking to beneﬁt from this incur a higher risk of imitation when they move production to the South. In Branstetter et al. (2007), as in Grossman and Helpman (1991a), imitation is a costly activity that requires a deliberate investment on the part of Southern ﬁrms seeking to copy Northern products. Stronger IPR protection in the South increases these costs, reducing imitation and lowering the risks faced by multinationals. Multinationals that move to the South employ the labor resources freed up by the decline in imitative activity. Production shifting allows for a reallocation of Northern resources towards innovative activity. Under certain parameter assumptions, a strengthening of Southern IPR protection enhances Southern industrial development because the increase in North–South FDI more than offsets the decrease in Southern imitation. We test three hypotheses that follow from this theory. First, using detailed data on U.S. MNE activity, we seek to determine whether multinational ﬁrms respond to reforms by increasing production in reforming countries. Second, we use industry-level data to test whether growth in production by MNEs and local ﬁrms that are not engaging in imitation exceeds the decline in imitative activity. Finally, we look for evidence that the production of new goods shifts to reforming countries by analyzing initial export episodes that are identiﬁed in disaggregated U.S. trade statistics. In our analyses, we focus on the effects of well-documented discrete changes in patent regimes over the 1980s and 1990s. We follow Branstetter et al. (2006), which assemble a set of substantive IPR reforms based on a number of primary and secondary sources. These are listed in Table 1.7 Our approach of analyzing responses to well-deﬁned IPR changes has the advantage that we may use ﬁxed effects to control for features of the business environment in a country that are constant, that are hard to measure, that are correlated with the strength of IPR in the country cross-section, and that may affect ﬁrm behavior and industrial development in a manner similar to IPR.8 We provide a more detailed discussion of these reforms below. While we employ some of the same data as Branstetter et al. (2006), in this paper our goal is to formulate a broader assessment of the impact of IPR reform on the level and nature of industrial

6 Glass and Saggi (2002) obtain a result similar to that of Helpman (1993). See also Markusen (2001). 7 We include patent reforms in Japan in our sample even though it is a high income country, and it may not therefore be relevant to the models in which a Southern country reforms its IPR. Many students of the Japanese economy have pointed to the existence of a dual economy in Japan, with some industries achieving extremely high levels of productivity relative to the U.S. and other lagging far behind the U.S. productivity frontier. Given the substantial relative productivity lags that existed in some sectors, particularly at the beginning of our sample, we incorporate data from Japan in the empirical analyses described below. See McKinsey Global Institute (2000) and Porter et al. (2000) for discussions of these issues. 8 While the 16 countries in our sample are quite heterogeneous in terms of their income, location, and industrial development at the time of reform, we recognize the need to exercise caution in extrapolating these results to countries outside the sample.

Table 1 Timing of major patent reforms. Country

Year of reform

Argentina Brazil Chile China Colombia Indonesia Japan Mexico Philippines Portugal South Korea Spain Taiwan Thailand Turkey Venezuela

1996 1997 1991 1993 1994 1991 1987 1991 1997 1992 1987 1986 1986 1992 1995 1994

Notes: this table provides information about the timing of reforms in the countries that strengthen their intellectual property rights and are included in the sample.

development in reforming countries, whereas this earlier work focused solely on technology transfers within U.S. multinationals. We ﬁnd that U.S.-based multinationals expand the scale of their activities in reforming countries after IPR reform along multiple dimensions. Afﬁliates increase their assets, net property, plant and equipment (net PPE), employment compensation, transfer of technology from abroad, and research and development (R&D) expenditures. These increases are particularly large for ﬁrms that are especially likely to value reforms in the sense that they, prior to reforms, deploy high levels of technology abroad.9 This evidence is consistent with U.S. multinationals shifting production of more technologically intensive goods to afﬁliates in response to reforms. We further assess the impact of IPR reform on overall industrial activity. Our results indicate that industry-level value added increases after reforms and that this effect is concentrated in technologyintensive industries and in industries where MNE activity is concentrated. These ﬁndings suggest that declines in imitative local activity are offset by increases in the activity of multinationals and other ﬁrms that are not engaging in imitative activity. Although the theory described above stresses the direct role of changes in MNE activity, a large body of empirical work indicates that MNE expansion could generate indirect beneﬁts for the host country by fostering the growth of local input suppliers, as in Javorcik (2004a), and by transferring advanced knowledge and skills to the local workforce, as in Poole (2009). IPR reform appears to lead to an overall enhancement of Southern industrial development. We obtain further suggestive evidence on the rate at which production is transferred to reforming countries by analyzing disaggregated U.S. trade statistics. Following Feenstra and Rose (2000), we construct for each reforming country an annual count of initial export episodes, deﬁned as the number of 10-digit commodities for which recorded U.S. imports from a given country exceed zero for the ﬁrst time in our data. This measure increases sharply after IPR reform, suggesting that any decline in indigenous innovation is more than offset by an expanded range of goods being produced by MNEs and other ﬁrms. Again, the evidence suggests that IPR reform enhances, rather than retards, Southern industrial development. The rest of the paper is organized as follows. Section 2 brieﬂy discusses the sample of IPR reforms we study. Section 3 describes our data on U.S. multinational afﬁliates and parents and presents 9 The results on technology transfer and research and development expenditure are similar to those presented in Branstetter et al. (2006). We report them here to illustrate that, as the scale of MNE activity expands in IPR-reforming countries, it also becomes more technology-intensive.

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

empirical tests and results based on these data. Section 4 characterizes data from United Nations Industrial Development Organization databases and presents empirical tests and results based on these data. Section 5 describes the disaggregated U.S. import statistics and presents tests and results based on these data. Section 6 concludes.

29

details of local intellectual property law. This process led us to choose a date of reform that varied from the date identiﬁed in prior research like Maskus (2000) for Japan, Taiwan, and Brazil, and we checked the robustness of our results by using the alternative dates identiﬁed in prior research. Our main empirical results are not sensitive to these changes.

2. Patent reforms 3. Multinational ﬁrm responses Our sample of IPR reforms includes the 16 patent reform episodes identiﬁed in Branstetter et al. (2006) and listed in Table 1. These include changes in IPR that occurred between 1982 and 1999 and that occurred in countries where there was a substantial amount of U.S. MNE afﬁliate activity prior to reform.10 Each reform can be classiﬁed according to whether or not it expanded or strengthened patent rights along ﬁve dimensions: 1) an expansion in the range of goods eligible for patent protection, 2) an expansion in the effective scope of patent protection, 3) an increase in the length of patent protection, 4) an improvement in the enforcement of patent rights, and 5) an improvement in the administration of the patent system. There is a surprising degree of similarity in these reforms, with 15 out of 16 exhibiting expansion of patent rights along at least 4 of these 5 dimensions. These substantive reforms could have a material impact on industrial development and are therefore well-suited to our empirical approach. A detailed discussion of the individual patent reform episodes, their distinctive characteristics, and their common features, is provided in the Appendix to Branstetter et al. (2006), particularly Sections A.2 and A.3. This sample of 16 patent reform episodes does not include all episodes that take place over our sample period. To address the concern that our results depend on the reforms in the sample, we conduct robustness tests in which we augment our sample of reforms to include data on reforms in Austria (1984), Canada (1987), Denmark (1983), Ecuador (1996), Finland (1995), Greece (1992), Norway (1992), and Panama (1996). These reform episodes and dates are taken from Qian (2007), but they seem to be less substantive than the reforms we study. The reforms in Austria, Canada, Finland, Greece, and Norway consisted primarily of changes in the patent law's treatment of pharmaceutical products, and changes occurred in a context where national health insurance systems effectively set drug prices at the national level and limited the impact of intellectual property rights protection in this industry.11 The National Trade Estimate Reports on Foreign Trade Barriers published by the United States Trade Representative after the patent reforms in Panama and Ecuador calls into question the effectiveness of enforcement of the legal changes and provides an interpretation of the timing of reform in Ecuador that differs from Qian (2007). Finally, direct communication with the Danish Patent Ofﬁce suggested that the reforms of 1983 were quite minor and narrow in focus.12 Nevertheless, in robustness checks below, we expand our set of reforms to include these countries, and their inclusion does not qualitatively affect our results. In assigning dates to patent reform, we followed prior research like Maskus (2000) and Qian (2007), who have identiﬁed a key stage in the patent reform process that marked a clear shift in the trajectory of policy. We went to the original source documents consulted by Maskus (2000), discussed these reforms with multinational managers, and also sought the input of country-based experts in the 10 These time constraints imply that we do not evaluate Indian patent reform. While it ratiﬁed the TRIPs Agreement in 1995, India delayed the ﬁnal approval of a TRIPsconsistent patent law until 2005 and put off even the ﬁrst major step toward patent reform until 1999, the ﬁnal year of our sample period. 11 Canada's reform, reviewed in detail by McFetridge (1997), also applied to foodstuffs, and we account for this in our robustness checks. 12 We thank Annette Zerrahn of the Danish Patent Ofﬁce for providing her insights. Ms. Zerrahn explained that the changes amounted to an adjustment of the fee structure in Danish patent law and the ratiﬁcation of the Budapest Treaty regarding deposit of micro-organisms for examination.

3.1. Empirical speciﬁcation In assessing whether stronger IPR induces an expansion of multinational activity, we take a difference-in-differences approach. Individual afﬁliates are followed through time, and the basic speciﬁcation tests how MNE activity changes around the time of reform, controlling for country, parent ﬁrm, and afﬁliate characteristics that might impact the variables of interest. The basic speciﬁcation takes the form: Silt = α0 + αil + αt + β0 yjt + β1 Pit + β2 Hjt + β3 Rjt

ð1Þ

+ β4 Rjt * Techil + εit where l indexes the individual afﬁliate, i the afﬁliate's parent ﬁrm, j the afﬁliate's host country, and t the year. Several measures of the scale of multinational activity serve as dependent variables. Taken literally, the theoretical models discussed in Section 1 deﬁne the scale of multinational activity as the number of distinct products for which production has shifted to the South. While our data on multinational activity are at the afﬁliate level, they do not include product-level data. Hence, our data are not sufﬁciently disaggregated to measure production so deﬁned. However, by measuring the response of multinational afﬁliates to IPR reform along a number of dimensions, we look for evidence of a change in the scale of afﬁliate production, which is likely correlated with the notion of production shifting that is analyzed in theoretical work. Using sales to measure changes in the scale of multinational operations after IPR reform creates serious problems of inference. Afﬁliate sales increase by about 18% following reform, but this increase could reﬂect an increase in market power resulting from IPR reform rather than production shifting. The extensive literature on the imperfect correspondence between accounting measures of proﬁt and economic concepts of proﬁt suggests that there is little we can do to get around this. We tracked the ratio of sales to employee compensation as one rough measure of markups and found no evidence that this measure increases after reform.13 While reassuring, this does not conclusively prove the absence of any contamination of our sales data by changes in proﬁtability.14 As a consequence, we focus on measures of afﬁliate inputs rather than measures of sales. We use measures of an afﬁliate's assets and net PPE as proxies for the scale of physical capital. In the theoretical literature discussed in Section 1, labor is the only factor of production, so it would naturally increase as a result of reform. In practice, production shifting of more technologically sophisticated goods might have a relatively modest effect on the overall size of the workforce and much more of an impact on its composition; the ﬁrm might change the skill mix of its afﬁliate

13

We thank an anonymous referee for suggesting this test. This is only one of the many concerns around the use of sales data. Short run ﬂuctuations in sales could also be driven by demand shocks that are temporally correlated with the IPR regime change but conceptually distinct from it. In our input data, we see evidence consistent with the view that ﬁrms gradually expand their capital stock and increase the technological intensity of their activities in IPRreforming countries. Sales changes driven by this response to IPR reform would likely emerge with a lag. Given the paucity of post reform observations for many of our reforming countries, our ability to quantify these shifts convincingly is limited. 14

30

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

workforce, hiring more managers and engineers. We therefore analyze data on total employment compensation. This variable should capture changes in the size of the workforce as well as shifts in its skill composition. The production of more sophisticated products in reforming countries would likely require an increase in the use of the parent ﬁrm's technology. Following Branstetter et al. (2006), we use afﬁliate level data on the volume of intraﬁrm royalty payments for intangible assets to track changes in the licensing of technology from the parent.15 If IPR reform induces ﬁrms to shift production of more technologically intensive products to afﬁliates in reforming countries, we would expect to see those payments increase relative to afﬁliate sales.16 Finally, the inception of production of more technologically intensive products should be associated with an increase in afﬁliate level R&D spending. While U.S. MNEs undertake basic and applied research abroad, the R&D conducted by afﬁliates in developing countries, which account for most of the countries in our sample, is focused on the modiﬁcation of parent ﬁrm technology for local markets, as explained in Kuemmerle (1999) and other work. Thus, afﬁliate R&D and technology transfers from the parent should be considered complements, implying that R&D spending should increase as a result of IPR reform. Branstetter et al. (2006) also present results of tests analyzing the impact of IPR regime changes on afﬁliate R&D. The key variables of interest are the Post Reform Dummy variable Rjt, and the interaction Rjt * Techil that allows estimates of an afﬁliate's response to patent reform to differ for ﬁrms that extensively deploy intellectual property in non-reforming countries around the world prior to reforms. The High Technology Transfer Dummy, Techil, is generated as follows: afﬁliates of parents that, over the four years prior to a particular reform, receive at least as much technology licensing income from afﬁliates outside the reforming countries as the parent of the median afﬁliate in the reforming country over the same period are assigned a dummy equal to one. For the other half of the sample, Techil equals zero. This dummy variable thus captures the relative propensity of different parent ﬁrms to both create intellectual property and deploy it outside the home country.17 By comparing the responsiveness of this subset of ﬁrms where we expect a disproportionate impact of IPR reforms, we can better discriminate between the view that IPR reform leads to greater production shifting and alternative interpretations of our results.18 We include a number of controls. αil are time-invariant ﬁxed effects for the afﬁliate, αt are year ﬁxed effects for the entire sample, and yjt are country-speciﬁc linear time trends. Pit and Hjt are vectors of time-varying parent and host country characteristics respectively. We control for the total sales of the parent system as well as the level of parent ﬁrm R&D spending. Host country characteristic controls include measures of median corporate tax rates, inward FDI restrictions, capital controls, dividend withholding tax rates applying to U.S. multinational ﬁrms, trade openness, the log of per capita GDP, the log of GDP, and the log of the real exchange rate. As such, our speciﬁcation controls for the determinants of FDI that are captured by gravity

15 Our earlier paper describes at length the nature of these data and the issues that arise in using them as indicators of technology transfer. 16 Rather than using payments for technology, Eaton and Kortum (1996, 1999) have used measures of patenting by inventors outside their home countries as an indicator of technology transfer. Lerner (2002) and Branstetter et al. (2006) ﬁnd strong evidence that patenting by foreign inventors increases in developing countries after IPR reform. 17 This approach has some advantages over the approach taken in Branstetter et al. (2006), where ﬁrms were differentiated on the basis of the patents generated by their parents. This relied on the incomplete mapping of ﬁrms to U.S. patent assignees developed in the NBER Patent Citation Database (Hall et al. (2001)) and did not adjust for the different propensities to patent observed across different industries (Cohen et al. (2002)). 18 See, for example, Javorcik (2004b).

equations. Each of the MNEs we analyze is a U.S. MNE so the afﬁliate ﬁxed effect controls for the distance between the U.S. and the afﬁliate's host country. The time dummies absorb variation in the size of the U.S. economy, and, as indicated above, the speciﬁcations explicitly control for the log of GDP of the host country. We do not view this basic speciﬁcation as a structural estimating equation in any sense, nor do we impute structural interpretations to any of the regression parameters generated by it. 3.2. Data Data on U.S. multinational ﬁrms comes from the U.S. Bureau of Economic Analysis (BEA) annual Survey of U.S. Direct Investment Abroad and the quarterly Balance of Payments Survey, and our data covers the years 1982–1999. The survey forms concerning MNE activity capture extensive information on measures of parent and afﬁliate operating activity like levels of assets, net PPE, employment compensation, and R&D expenditures.19 MNEs must also report the value of royalties paid by afﬁliates to parents for the sale or use of intangible property. American tax law requires that foreign afﬁliates make these payments. The reported ﬁgures on the value of intangible property transferred include an amalgam of technology licensing fees, franchise fees, and fees for the use of trademarks. However, the aggregate data indicate that intangible property transfers are overwhelmingly dominated by licensing of technology for industrial products and processes. R&D data were not reported annually in the early years of our sample period. Regular reporting began only in 1989. This means that pre reform R&D data are limited for a number of the reforms we investigate. As a consequence, we must interpret results based on R&D data with an extra measure of caution. 3.3. Results Fig. 1 indicates the changes in the scale of afﬁliate activity across afﬁliates of ﬁrms that do and do not extensively deploy technology abroad. It illustrates median assets for afﬁliates in the Low and High Technology Transfer samples in the pre reform period and the post reform period. The clear bars present medians for the pre reform years, and the shaded bars present medians for the post reform years. There are large increases in the scale of activity for both kinds of afﬁliates, but the increases are larger for those afﬁliates that have parents that transfer technology abroad prior to the reform more aggressively. To study these patterns more rigorously, Table 2 presents results based on Eq. (1) that test whether afﬁliates expand their operations at the time of reform. As noted above, our tests focus on measures of inputs to afﬁliate production. All variables are available annually for the period 1982–1999, except afﬁliate R&D which is only available from 1989 onwards. The statistical signiﬁcance of coefﬁcients in all tables is denoted by two asterisks to indicate signiﬁcance at the 1% level and a single asterisk to indicate signiﬁcance at the 5% level. The dependent variable in column 1 is the log of afﬁliate assets. The positive coefﬁcient on the Post Reform Dummy indicates that afﬁliates of U.S. MNEs expand their assets at the time of reform. Because the dependent variable is measured in logs, this coefﬁcient has a semi-elasticity interpretation, implying an increase of about 16% following reforms. In column 2, we include the interaction term, allowing the impact of reform to vary for afﬁliates that are connected to parents that tend to extensively deploy technology abroad. The 0.1114 coefﬁcient on the Post Reform Dummy indicates that even afﬁliates of U.S. MNEs with below median levels of technology transfer

19 In order to obtain information on parent ﬁrm R&D expenditures in years in which this item was not captured in BEA surveys, the BEA data on publicly traded parents is linked to COMPUSTAT using employee identiﬁcation numbers.

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

Fig. 1. The clear bars indicate median assets (in thousands) for afﬁliates in the Low and High Technology Transfer samples over all years in the pre reform period. The shaded bars depict medians following reform. The High Technology Transfer sample includes afﬁliates of parents that over the four years prior to a reform average total royalty payment receipts from all afﬁliates that are at least as large as the receipts of the parent of the median afﬁliate in the reforming country.

31

abroad expand their capital stock at the time of reform. The Post Reform Dummy interacted with the High Technology Transfer Dummy is also positive and statistically signiﬁcant, indicating an additional 9% expansion among afﬁliates with a High Technology Transfer Dummy equal to one. Thus, IPR reforms trigger increases in afﬁliate assets, and these increases are larger among the ﬁrms most likely to beneﬁt from reform. Columns 3 and 4 present results of the same speciﬁcation using the log of net PPE as the dependent variable. In the third column, the coefﬁcient on the Post Reform Dummy is 0.1248, and it is signiﬁcant at the 1% level. In column 4, we incorporate both this dummy and an interaction term. The coefﬁcient on the Post Reform Dummy is now small and statistically insigniﬁcant at conventional levels, while the coefﬁcient on this dummy interacted with the High Technology Transfer Dummy has a coefﬁcient of 0.1882, signiﬁcant at the 1% level. The ﬁfth and sixth columns present estimates of the impact of reform on employment compensation. In the ﬁfth column, the estimated impact of reform on employment compensation is 0.1634. The sixth column also includes the interaction term. The results in this column imply that afﬁliates of ﬁrms with low transfers of technology abroad increase employment compensation by about 12%. Afﬁliates of ﬁrms that extensively transfer technology abroad increase employment compensation by an additional 8%, implying a total expansion of around 20% for afﬁliates of ﬁrms that make extensive use of intellectual property abroad. While the results of the ﬁrst six columns all imply an expansion of multinational activity in the wake of patent reform, they do not

Table 2 U.S. multinational afﬁliate responses to reform. Dependent variable:

Post reform dummy Post reform dummy * high Technology transfer dummy Host country corporate tax rate Host country inward FDI restrictions Host Country Capital Controls Host Country Withholding Tax Rate Host Country Trade Openness Log of Host Country GDP per Capita Log of Host Country GDP Log of Real Exchange Rate Log of Parent R&D Expenditures Log of Parent System Sales No. of obs. R-squared

Log of afﬁliate assets

Log of afﬁliate net PPE

Log of afﬁliate employment compensation

100 × log of intraﬁrm royalty payments/ afﬁliate sales

100 × log of R&D expenditures/afﬁliate sales

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

0.1590 (0.0140)**

0.1114 (0.0173)** 0.0912 (0.0181)** 0.1361 (0.0985) − 0.0601 (0.0386) − 0.0664 (0.0235)** 0.3365 (0.1386)* 0.0062 (0.0016)** 0.3406 (0.1518)* 0.9037 (0.1632)** − 0.3161 (0.0198)** 0.0076 (0.0035)* 0.0467 (0.0088)** 26,184 0.8884

0.1248 (0.0328)**

0.0245 (0.0430) 0.1882 (0.0443)** 0.5099 (0.2370)* − 0.0971 (0.0748) − 0.0819 (0.0555) 0.4473 (0.3135) 0.0071 (0.0035)* 0.7166 (0.2908)* − 0.1374 (0.3226) − 0.3231 (0.0483)** 0.0315 (0.0089)** 0.0555 (0.0143)** 22,342 0.8377

0.1634 (0.0157)**

0.1210 (0.0205)** 0.0790 (0.0217)** 0.4746 (0.1104)** − 0.0483 (0.0327) − 0.0376 (0.0244) − 0.4565 (0.1646)** 0.0001 (0.0016) 0.4713 (0.1900)* 0.6179 (0.1924)** − 0.3657 (0.0238)** 0.0054 (0.0040) 0.0601 (0.0093)** 24,844 0.8789

0.0787 (0.0268)**

− 0.1311 (0.0274)** 0.3985 (0.0323)** − 0.0286 (0.1680) 0.0407 (0.0512) − 0.0926 (0.0431)* − 0.5657 (0.2324)* − 0.0074 (0.0029)* 0.6963 (0.3193)* − 0.0007 (0.3339) − 0.1097 (0.0401)** 0.0072 (0.0040) 0.0058 (0.0092) 25,600 0.6651

0.0151 (0.0232)

− 0.0129 (0.0252) 0.0546 (0.0275)* 0.2966 (0.1432)* − 0.0735 (0.0480) − 0.0368 (0.0299) 0.3879 (0.1783)* 0.0016 (0.0020) 0.0196 (0.4135) − 0.0537 (0.4109) 0.0578 (0.0373) 0.0072 (0.0027)** − 0.0015 (0.0039) 16,143 0.6645

0.1315 (0.0985) − 0.0610 (0.0386) − 0.0675 (0.0235)** 0.3340 (0.1387)* 0.0063 (0.0016)** 0.3335 (0.1522)* 0.9086 (0.1635)** − 0.3179 (0.0198)** 0.0079 (0.0036)* 0.0461 (0.0089)** 26,184 0.8882

0.4965 (0.2371)* − 0.0993 (0.0749) − 0.0825 (0.0556) 0.4333 (0.3137) 0.0072 (0.0035)* 0.6986 (0.2916)* − 0.1305 (0.3234) − 0.3280 (0.0483)** 0.0322 (0.0089)** 0.0544 (0.0143)** 22,342 0.8375

0.4701 (0.1104)** − 0.0488 (0.0328) − 0.0382 (0.0244) − 0.4594 (0.1646)** 0.0001 (0.0016) 0.4639 (0.1904)* 0.6229 (0.1928)** − 0.3673 (0.0238)** 0.0056 (0.0040) 0.0596 (0.0093)** 24,844 0.8788

− 0.0567 (0.1690) 0.0373 (0.0512) − 0.0957 (0.0436)* − 0.5805 (0.2342)* − 0.0072 (0.0029)* 0.6684 (0.3208)* 0.0196 (0.3357) − 0.1181 (0.0403)** 0.0079 (0.0041) 0.0087 (0.0091) 25,600 0.6625

0.2946 (0.1433)* − 0.0737 (0.0480) − 0.0369 (0.0299) 0.3824 (0.1784)* 0.0016 (0.0020) 0.0129 (0.4142) − 0.0512 (0.4115) 0.0566 (0.0373) 0.0074 (0.0027)** − 0.0019 (0.0039) 16,143 0.6644

The dependent variables are the log of afﬁliate assets in columns (1) and (2), the log of afﬁliate net property plant and equipment in columns (3) and (4), the log of afﬁliate employment compensation in columns (5) and (6), 100 times the log of one plus the ratio of intraﬁrm royalty payments to afﬁliate sales in columns (7) and (8), and 100 times the log of the ratio of one plus the ratio of afﬁliate research and development expenditures to afﬁliate sales in columns (9) and (10). The Post Reform Dummy is a dummy equal to one in the year of reform and in the years following the reforms identiﬁed in Table 1. The High Technology Transfer Dummy is a dummy that is equal to one for afﬁliates of parents that over the four years prior to a reform average total royalty payment receipts from all afﬁliates that are at least as large as the receipts of the parent of the median afﬁliate in the reforming country. Host Country Corporate Tax Rate and Host Country Withholding Tax Rate are annual median tax rates paid by afﬁliates in a host country. Host Country Inward FDI Restrictions and Host Country Capital Controls are dummies equal to one when inward FDI restrictions and capital controls exist, and they are drawn from Brune (2004) and Shatz (2000). Host Country Trade Openness is the index of constant price openness taken from Heston et al. (2002). The Log of Host Country GDP per capita and Log of Host Country GDP are derived from data provided in the World Bank World Development Indicators. Log of Real Exchange Rate is computed using nominal exchange rates and measures of inﬂation are from the IMF's IFS database. The Log of Parent System Sales is the log of total sales of the parent and its afﬁliates. All speciﬁcations include afﬁliate and year ﬁxed effects as well as country-speciﬁc time trends. Heteroskedasticity-consistent standard errors appear in parentheses. ** and * denote signiﬁcance at the 1% and 5% levels.

32

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

necessarily imply a change in the rate at which the production of new goods is transferred to the South. Further evidence of production shifting is obtained by analyzing the transfer of technology from the parent ﬁrm and R&D performed by afﬁliates. These measures of afﬁliate activity were analyzed in Branstetter et al. (2006) in empirical work that focused speciﬁcally on these components of ﬁrms' reactions to IPR reforms. We measure transfers of technology using royalties paid by afﬁliates to parents for the sale or use of intangible assets. Because larger afﬁliate sales volumes may automatically result in higher levels of royalty payments to the parent and because many afﬁliates do not make royalty payments, we use the log of one plus the ratio of royalty to sales. For expositional purposes, we multiply this value by 100 in our reported results. Speciﬁcations explaining this variable appear in columns 7 and 8 of Table 2. As indicated in column 7, the overall impact of reform on technology transfer is positive and statistically signiﬁcant. When we include the interaction term as in column 8, the coefﬁcient on the Post Reform Dummy is now negative and statistically signiﬁcant. However, the 0.3985 coefﬁcient on the interaction term is positive, highly signiﬁcant, and its magnitude is very large. For the afﬁliates of ﬁrms that are more likely to value IPR reform because they make more extensive use of parent technology abroad, royalty payments increase substantially in response to reforms. While most R&D spending by U.S. MNEs is concentrated in the U.S., some foreign afﬁliates have substantial R&D expenditures. As noted above, the vast majority of this R&D spending is designed to modify the parent ﬁrm's technology to local circumstances and conditions. It can thus be seen as a complement to technology transfers from the parent. If the post reform increase in technology licensing payments identiﬁed in columns 7 and 8 truly represents the deployment of new technology rather than simply an increase in the price of technology, then we would expect that increase to be mirrored by an increase in afﬁliate R&D spending. In columns 9 and 10 of Table 2, we show results of a speciﬁcation using the log of one plus the ratio of afﬁliate R&D expenditures to afﬁliate sales as the dependent variable. As in columns 7 and 8, we multiply this value by 100 for expositional purposes. As indicated in column 9, the overall impact of reform on afﬁliate R&D is positive but not statistically signiﬁcant at conventional levels. When we include the interaction term, as in column 10, the coefﬁcient on the Post Reform Dummy is small and statistically indistinguishable from zero at conventional levels. However, the coefﬁcient on the interaction of the Post Reform Dummy and High Technology Transfer Dummy is positive and statistically signiﬁcant. Thus, for afﬁliates of parents that are likely to especially value strong IPR, there is a signiﬁcant post reform increase in R&D. Fig. 2 illustrates the dynamics of the increase in afﬁliate activity. To create this ﬁgure, we regress the log of afﬁliate assets on afﬁliate ﬁxed effects, country–year ﬁxed effects, the controls that are not collinear with the ﬁxed effects (namely the log of total parent system sales and the log of parent R&D) and a set of indicator variables for the years that lead and lag IPR reform for afﬁliates that have a High Technology Transfer Dummy equal to one. As such, the coefﬁcients on these timespeciﬁc dummies illustrate how the stock of assets changes for afﬁliates of ﬁrms that extensively deploy technology abroad. We plot these coefﬁcients in Fig. 2, along with 95% conﬁdence interval bounds. The coefﬁcients show variation around zero before reform and an upward trend following reform. Thus, the changes in afﬁliate scale appear to begin at the time of reform.20 Uncertainty concerning the effective enforcement of reforms could account for the lag in response. Managers might gradually update their beliefs on the efﬁcacy of reforms, leading to an increase in the scale and scope of production. 20 The coefﬁcients on the dummies for the 4, 3, and 2 years before reform have an average that is close to zero, and they are not signiﬁcantly different from zero in an F-test of joint signiﬁcance. However, the coefﬁcients on the dummies for the 4, 3, and 2 years after reform are positive and signiﬁcant at the 10% level in an F-test of joint signiﬁcance.

Fig. 2. This ﬁgure displays the dynamics of changes in afﬁliate size around the time of reform. The points are generated by regressing the log of afﬁliate assets on afﬁliate ﬁxed effects, country–year ﬁxed effects, the controls from the speciﬁcations in Table 2 that are not collinear with the ﬁxed effects (namely the log of total parent system sales and the log of parent R&D), and a set of dummies that are equal to one in years relative to the IPR reform for afﬁliates that have a High Technology Transfer Dummy equal to one. The points are the coefﬁcient estimates on these time-speciﬁc dummies for afﬁliates in the High Technology Transfer sample.

This general pattern of a gradual response that builds over several years is similar to the timing of increases in R&D spending, technology transfer, and multinational patenting reported in Branstetter et al. (2006).21 The results in Table 2 are robust to a number of considerations. As noted in Branstetter et al. (2006), while IPR-strengthening legislation was enacted in Argentina and China in the 1990s, multinational managers have called into question the effectiveness of enforcement of reform in these two countries. We therefore repeated the speciﬁcations shown in Table 2 with a restricted sample that excluded Argentina and China. We obtained results qualitatively similar to those shown here.22 Because Japan differs in important ways from the other countries that undertook signiﬁcant IPR reforms, we also repeated our analyses with Japan omitted from the sample. Our results are not qualitatively affected by this change. Our results are also robust to the inclusion of region–year ﬁxed effects. Concerns that measurement of the High Technology Transfer Dummy might cause it to proxy for ﬁrm size led us to incorporate an interaction term of a measure of ﬁrm size and the Post Reform Dummy.23 This also does not affect our results. Expanding the sample of reforms to include all afﬁliates in Austria, Denmark, Ecuador, Finland, Greece, Norway, and Panama and Canadian afﬁliates in the pharmaceutical and foodstuffs industries does not qualitatively affect our results, nor does substituting the reform dates for Japan, Taiwan, and Brazil from Maskus (2000) for our own. While the speciﬁcations in Table 2 pool observations across countries and industries, we also ran the one in column 2 country by country. Although the sizes of the subsamples used in these tests are much smaller than those used to generate the results in Table 2, the coefﬁcients on the interaction of the Post Reform Dummy and the High Technology Transfer Dummy is positive in 13 out of 16 cases, and it is at least marginally signiﬁcant in 7 out of 16 cases. We also ran industry by industry tests based on the speciﬁcations presented in columns 1 and 2 of Table 2 using the industry groups

21 The gradual increase could also reﬂect the fact that we are examining the stock of capital rather than the ﬂow. 22 These results are available from the authors upon request. The only speciﬁcations that were sensitive to the exclusion of Argentina and China were those that employed the log of the R&D to sales ratio as the dependent variable. 23 We have also checked the robustness of the results to including the High Technology Transfer Dummy with other policy variables.

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

provided in the BEA data. The results indicate that activity in nearly all industry groupings increases, with the largest estimated increase occurring for afﬁliates in the chemical manufacturing industry.24 The coefﬁcient on the interaction of the Post Reform Dummy and the High Technology Transfer Dummy in speciﬁcations like the one in column 2 of Table 2 is positive for 6 of the 10 industry groupings, and it is signiﬁcant for Chemical Manufacturing, Industry Machinery and Equipment, and an “Other Manufacturing” aggregate that includes the manufacturing of medical instruments and scientiﬁc equipment. In each of these industry domains, intellectual property protection is likely to be especially important. Taken together, these results strongly suggest that multinationals respond to IPR reform by shifting production to reforming countries. To examine whether this positive impact is sufﬁcient to have a positive effect on overall levels of industrial development, we turn to an analysis of industry-level data in reforming countries. 4. Industry-level output responses The preceding analysis documents the positive effect of IPR reform on U.S. multinational activity. However, the overall impact on output also depends on whether the growth in production by U.S. MNEs, MNEs from other countries, and domestic ﬁrms not engaging in imitation is sufﬁcient to offset any decline among local imitators. While we do not have ﬁrm-level data to examine these effects, we can use industry-wide measures of production in reforming countries to analyze broad economic changes. One important caveat to this analysis is that our data do not capture changes in imitative activity in the informal sector, so our estimates could understate the negative impact of stronger IPR on imitative ﬁrms.25 4.1. Empirical approach and data We examine the impact of IPR reform on industrial output using a speciﬁcation similar to that employed in the previous section: VAjt = α0 + αij + αt + β0 yjt + β1 Hjt + β2 Rjt

ð2Þ

+ β3 Rjt *IndTechi + εit VA measures the log of value added in industry i in country j in year t. The controls include country–industry pair ﬁxed effects, time ﬁxed effects, country-speciﬁc linear time trends, and a vector of timevarying characteristics of country j, including measures of the corporate tax rate, inward FDI restrictions, capital controls, trade openness, the log of GDP per capita, and the log of the real exchange rate. The main coefﬁcients of interest are on Rjt, the Post Reform Dummy, and its interaction with industry-level attributes that indicate the extent of potential beneﬁts from IPR reform. We consider two such industry characteristics. The ﬁrst is a measure of the importance of technological innovation to ﬁrms in an industry, which we measure using a Technology-Intensive Dummy that is equal to one for the following industries: electrical machinery, industrial chemicals, other chemicals, professional and scientiﬁc equipment, and transportation equipment. As an alternative approach, we also generate an industry-level measure of FDI intensity by looking at the cross-industry distribution of U.S. FDI in countries where intellectual property is well protected throughout our sample. The intuition is that these are the sectors where multinationals would naturally choose to invest abroad if unconstrained by IPR concerns. We generate a High FDI Dummy equal to one for industries that had 24 The coefﬁcient on the Post Reform Dummy is positive and signiﬁcant for all industry groupings except Primary and Fabricated Metal Manufacturing and Wholesale Trade. 25 We thank an anonymous referee for this observation.

33

above median levels of afﬁliate sales activity in countries with a 1980 total patent protection index above 3.57 in Ginarte and Park (1997). Data on industry value added are drawn from the United Nations Industrial Development Organization (UNIDO) database, which provides measures of value added at the ISIC 3-digit level in a common format for a large number of member states. This variable captures the activity of both multinational afﬁliates and domestic ﬁrms. While there are some gaps in the data, there is reasonably complete coverage for most countries in most years. To ensure comparability between these results and our earlier results on U.S. MNE activity, we limit our sample to the years 1982–1999. 4.2. Results Table 3 reports results of estimating Eq. (2). The positive coefﬁcient on the Post Reform Dummy in the ﬁrst column suggests that value added increases after patent reform, and this effect is statistically signiﬁcant at conventional levels.26 This ﬁnding is inconsistent with the view that IPR reform induces a collapse of indigenous industrial activity that more than offsets MNE expansion. Column 2 reports a speciﬁcation that includes the interaction of the Post Reform Dummy and the Technology-Intensive Dummy. The coefﬁcient on this interaction term is positive and statistically signiﬁcant at the 1% level, implying that output expansion is particularly pronounced in technology-intensive industries. The point estimate implies an expansion of industry-level value added of nearly 20% for these industries. Column 3 presents results of a speciﬁcation in which the Post Reform Dummy is interacted with the High FDI Dummy. The coefﬁcient on this interaction term is positive and statistically signiﬁcant. Expansion in industry value added is also especially large in industries where MNEs are active. These results are robust to a number of checks. As in our other analyses, we repeat our tests dropping China and Argentina from the sample and obtain qualitatively similar results, as indicated in columns 4–6. We consider an alternative speciﬁcation incorporating country–year ﬁxed effects. The country–year ﬁxed effects absorb the impact of all variables that are the same for all industries in a given country at a given time, so in this speciﬁcation it is no longer possible to estimate the impact of country controls or the direct effect of the Post Reform Dummy because these are the same for all industries in a given country at a given time. However, it is still possible to estimate the differential impact of IPR reform on technology-intensive and FDIintensive sectors. These effects remain strongly positive and highly signiﬁcant, even in this more demanding speciﬁcation.27 We have also expanded the sample of reforms to include those in Austria, Canada, Denmark, Ecuador, Finland, Greece, Norway, and Panama and changed the reform dates for Japan, Taiwan, and Brazil to conform to those of Maskus (2000). Neither of these adjustments has substantive effects on our results. Our use of country–industry ﬁxed effects controls for time-invariant measures of factor endowments interacted with time-invariant measures of factor demands, but we also have included time-varying measures of human capital interacted with the human capital intensity of individual industries and time-varying measures of physical capital interacted with the physical capital intensity of individual industries. Including these controls does not qualitatively alter our results. 5. Initial export episodes Interpreted literally, the models that motivate our empirical exercise focus on the initiation of production of new goods in 26 These results need to be interpreted with caution. A systematic expansion in markups in IPR-sensitive industries could be observationally equivalent to an expansion of output. 27 We thank Nick Bloom for suggesting this additional speciﬁcation.

34

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

Table 3 Impact of reform on industry value added in reforming countries. Dependent variable:

Log of industry value added

Sample:

All reforms

Post Reform Dummy Post Reform Dummy* Technology-Intensive Dummy Post Reform Dummy * High FDI Dummy Host Country Corporate Tax Rate Host Country Inward FDI Restrictions Host Country Capital Controls Host Country Trade Openness Log of Host Country GDP per Capita Log of Real Exchange Rate No. of obs. R-squared

Drop China and Argentina

(1)

(2)

(3)

(4)

(5)

(6)

0.0956 (0.0167)**

0.0731 (0.0173)** 0.1252 (0.0252)**

0.0501 (0.0203)*

0.0605 (0.0188)**

0.0383 (0.0194)* 0.1222 (0.0275)**

0.0226 (0.0226)

− 0.1503 (0.1107) − 0.2450 (0.1012)* 0.0851 (0.0329)** 0.0008 (0.0023) 2.1138 (0.1488)** − 0.3417 (0.0318)** 6884 0.9595

− 0.1506 (0.1106) − 0.2448 (0.1016)* 0.0853 (0.0328)** 0.0008 (0.0023) 2.1157 (0.1485)** − 0.3420 (0.0316)** 6884 0.9596

0.0924 (0.0211)** − 0.1140 (0.1147) − 0.2350 (0.1098)* 0.0602 (0.0336) − 0.0001 (0.0023) 2.0320 (0.1551)** − 0.3253 (0.0309)** 6183 0.9582

− 0.0937 (0.1291) − 0.2381 (0.1012)* 0.2097 (0.0336)** 0.0002 (0.0028) 2.4143 (0.2007)** − 0.4025 (0.0413)** 6069 0.9584

− 0.0933 (0.1290) − 0.2375 (0.1016)* 0.2097 (0.0334)** 0.0002 (0.0028) 2.4141 (0.2001)** − 0.4029 (0.0411)** 6069 0.9586

0.0888 (0.0234)** − 0.0748 (0.1342) − 0.2320 (0.1097)* 0.1819 (0.0338)** − 0.0012 (0.0029) 2.3596 (0.2130)** − 0.3798 (0.0399)** 5427 0.9570

The dependent variable is the log of industry value added. Columns (4)–(6) report results obtained when China and Argentina are dropped from the sample. The Post Reform Dummy is a dummy equal to one in the year of reform and in the years following the reforms identiﬁed in Table 1. The Technology-Intensive Dummy is equal to one for ISIC codes 351, 352, 383, 384, and 385. The High FDI Dummy is equal to one in industries that had above median levels of afﬁliate sales activity in countries with a 1980 total patent protection index above 3.57 in Ginarte and Park (1997). Host Country Corporate Tax Rate is the annual median tax rates paid by afﬁliates in a host country. Host Country Inward FDI Restrictions and Host Country Capital Controls are dummies equal to one when inward FDI restrictions and capital controls exist, and they are drawn from Brune (2004) and Shatz (2000), respectively. Host Country Trade Openness is the index of constant price openness taken from Heston et al. (2002). The Log of Host Country GDP per capita is derived from data provided in the World Bank World Development Indicators. Log of Real Exchange Rate is computed using nominal exchange rates and measures of inﬂation are from the IMF's IFS database. All speciﬁcations include country–industry and year ﬁxed effects as well as country-speciﬁc time trends. Heteroskedasticity-consistent standard errors appear in parentheses. ** and * denote signiﬁcance at the 1% and 5% levels.

reforming countries following IPR reform. The measures of afﬁliate and industry activity analyzed above are not sufﬁciently disaggregated to permit the tracking of afﬁliate activity at the individual product level. Therefore, to capture more directly the extent of new production in reforming countries, we build on the method of Feenstra and Rose (2000). This approach requires the use of disaggregated U.S. import statistics to obtain counts of initial export episodes. For each country– year observation, our measure of new goods production is the number of 10-digit commodities in U.S. imports from a given country that are recorded as exceeding zero for the ﬁrst time in a given year. Because the U.S. is the world's single biggest market for many commodities, looking at the date at which a particular reforming country starts exporting a particular good to the U.S. may be a reasonable indicator of production shifting for that good. This measure is imperfect in that domestic production may precede exports by several years. Furthermore, the notion of production shifting implies the initiation of production in one location and the cessation of production in another. However, it is very difﬁcult to identify the termination of production of particular goods in countries that do not undertake reform. Because of these data constraints, we are only able to examine one side of production shifting.28

28 Note, however, there is evidence that the expected cessation of production of certain goods in the U.S. is taking place. A recent study by Bernard et al. (2010) uses conﬁdential plant-level data to show that cessation of production of certain goods is occurring at a fairly rapid rate within U.S.-based manufacturing plants. These authors document a shift to the production of more capital- and skill-intensive goods for surviving plants, consistent with the evolving comparative advantage of U.S. manufacturers. Plants that do not shift their product mix in this way are less likely to survive. These patterns are broadly consistent with the view of production shifting presented in Section 2.

5.1. Empirical approach and data In order to study the initiation of the production of new goods, we use this speciﬁcation: Pjt = α0 + αj + αt + β0 Hjt + β1 Rjt + εit

ð3Þ

Our dependent variable P is a count measure that, for country j in year t, gives the number of 10-digit commodities where recorded U.S. imports exceeded zero for the ﬁrst time. Intuitively, this is a proxy for the arrival rate of new products. This count is regressed on country–year variables that control for a country's changing export capabilities, including country dummy variables, time dummy variables, a vector of time-varying characteristics of country j, and the Post Reform Dummy. The country characteristics included in the vector Hjt are the same as those used in the speciﬁcations presented in Table 3. Because the data cover only an 11 year time frame, as explained below, country-speciﬁc time trends are not included. For data, we utilize the U.S. trade database created by Feenstra et al. (2001). Annual data on U.S. imports from nearly all countries worldwide are available at the HS 10-digit level of disaggregation, which is very close to the level of individual products. One difﬁculty in using these data is that the 10-digit commodity classiﬁcation system was extensively revised in 1989. As a consequence, data before and after the revision are not comparable at the most disaggregated level. We therefore focus on the post-1988 period, where the data are measured consistently. To maintain comparability with our earlier results, we do not extend the data past 1999, so that the results in this section draw upon data from 1989 to 1999.29

29 These data contain some observations that record extremely small trade ﬂows, often followed by no activity. We are concerned that these anomalous observations might bias upward the counts of initial export episodes, so we report results obtained when we drop these questionable observations, though we obtain qualitatively similar results with the full data set.

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

35

As in our earlier speciﬁcations, we examine whether the estimated impact of IPR reform is stronger in technology-intensive product categories. In such tests, we analyze initial export episodes in HS 10digit product categories that are associated with the industry codes identiﬁed in the previous section as being technology-intensive: electrical machinery, industrial chemicals, other chemicals, professional and scientiﬁc equipment, and transportation equipment. For all country–year observations, we create separate counts of initial export episodes arising in only these product categories. We refer to these product categories as Tech Goods.

Maskus (2000) for our own. Expanding the set of reforming countries to include Austria, Canada, Denmark, Ecuador, Finland, Greece, Norway, and Panama produces results qualitatively very similar to those shown here. Finally, if we restrict our counts to include only those in commodity categories that existed at the beginning of our data, thereby excluding counts of new export episodes emerging in new commodity categories, we obtain results very similar to those employed here.

5.2. Results

Building on the work of Grossman and Helpman (1991a) and Helpman (1993), the theoretical literature on North–South product cycle models of international trade and investment shows that the effect of Southern IPR reform on the global economy hinge critically on the responses of multinational ﬁrms. In particular, this literature shows that if multinational ﬁrms based in the North are sufﬁciently responsive to the change in the Southern IPR regime, the expansion of their Southern afﬁliates could more than offset a decline in indigenous Southern imitation, thus spurring industrial development in the South. In this paper, we empirically confront hypotheses derived from these theories with a mix of evidence drawn from data on the activities of U.S. multinational afﬁliates, industry-level data from reforming countries, and disaggregated U.S. import data. We analyze changes in measures of afﬁliate activity, industry activity, and the introduction of new products in response to IPR reforms in 16 countries. All of the evidence indicates that stronger IPR in the South accelerates the transfer of production to reforming countries. We ﬁnd that U.S. MNE afﬁliate activity increases following reform and that increases are most pronounced among afﬁliates of ﬁrms that tend to deploy more technology abroad and that are therefore more likely to beneﬁt from reform. These ﬁndings are consistent with parents sharing new technology with their afﬁliates so that these afﬁliates can begin the manufacture of new, more sophisticated goods. The increase in production shifting through multinational ﬁrms could be smaller than the decrease in imitative activity by indigenous

Table 4 provides results from regressions based on Eq. (3). The dependent variable measures the count of initial export episodes at the 10-digit level. In column 1 we provide results using the Poisson ﬁxed effects regression model derived by Hausman et al. (1984) that takes into consideration the count nature of the dependent variable. In this speciﬁcation, we use data on initial export episodes in all product categories and all reforming countries. The coefﬁcient on the Post Reform Dummy is positive, signiﬁcant at the 1% level, and has a semi-elasticity interpretation; the coefﬁcient implies an increase in the production of new goods after IPR reform of about 28%. In column 2, we limit the sample to Tech Goods. For this subsample, the point estimate of the coefﬁcient on the Post Reform Dummy is slightly larger in magnitude than that in column 1, and it is signiﬁcant at the 1% level. These results are also robust to several checks. As indicated in columns 3 and 4, the coefﬁcients on the Post Reform Dummy are little changed by dropping Argentina and China from the sample. As an alternative speciﬁcation, we employ the ﬁxed effects negative binomial model of Hausman et al. (1984), and we obtain similar results, as shown in columns 5–8. Including a dummy variable measuring the onset of a free trade agreement with the U.S. does not alter the results, and our results are qualitatively unchanged when we substitute dates for the IPR reforms of Japan, Brazil, and Taiwan from

6. Conclusion

Table 4 Impact of reform on initial export episodes. Dependent variable:

Count of initial export episodes

Speciﬁcation type:

Poisson

Sample

All reforms

Goods categories

All goods

Post Reform Dummy Host Country Corporate Tax Rate Host Country Inward FDI Restrictions Host Country Capital Controls Host Country Trade Openness Log of Host Country GDP per Capita Log of Real Exchange Rate No. of obs. Log likelihood

Negative binomial

Tech goods

Drop Argentina and China

All reforms

All goods

All goods

Tech goods

Drop Argentina and China Tech goods

All goods

Tech goods

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

0.2772 (0.0129)** 0.0588 (0.0794) − 0.0673 (0.0248)** − 0.1123 (0.0198)** 0.0110 (0.0008)** 0.0173 (0.0422) 0.3140 (0.0268)** 176 − 2342

0.3182 (0.0309)** 0.0949 (0.1916) 0.1598 (0.0508)** − 0.1948 (0.0559)** 0.0115 (0.0018)** − 0.1301 (0.0944) 0.2788 (0.0634)** 176 − 1162

0.2902 (0.0148)** − 0.0363 (0.0961) − 0.0338 (0.0268) − 0.0237 (0.0226) 0.0114 (0.0008)** − 0.1746 (0.0571)** 0.4101 (0.0302)** 154 − 2147

0.3866 (0.0359)** 0.1055 (0.2315) 0.1074 (0.0540) − 0.0725 (0.0646) 0.0096 (0.0018)** 0.0964 (0.1272) 0.2597 (0.0740)** 154 − 1061

0.2340 (0.0635)** − 0.2282 (0.3653) − 0.1260 (0.1187) − 0.1287 (0.0831) 0.0057 (0.0034) − 0.1582 (0.1571) 0.4106 (0.1282)** 176 − 912

0.3420 (0.1040)** − 1.2256 (0.5526)* − 0.1055 (0.1805) − 0.1201 (0.1420) 0.0059 (0.0043) 0.0788 (0.2023) 0.2402 (0.2009) 176 − 691

0.2488 (0.0716)** − 0.1762 (0.4604) − 0.1344 (0.1246) − 0.1249 (0.0970) 0.0081 (0.0038)* − 0.1071 (0.1915) 0.4748 (0.1528)** 154 − 801

0.4165 (0.1198)** − 0.8890 (0.6793) − 0.1380 (0.1859) − 0.1310 (0.1731) 0.0082 (0.0047) 0.2670 (0.2155) 0.3065 (0.1344) 154 − 606

The dependent variable is the annual count of HS 10 digit product categories in which a reforming country reports exports to the U.S. for the ﬁrst time. Data are taken from the trade data base documented in Feenstra et al. (2001). The Tech Goods sample includes the set of 10-digit commodity categories that is associated with ISIC codes 351, 352, 383, 384, and 385. Columns (3), (4), (7) and (8) display results generated after removing Argentina and China from the sample. The Post Reform Dummy is equal to one in the year of reform and in the years following the reforms listed in Table 1. Host Country Corporate Tax Rate is the annual median tax rate paid by U.S. MNE afﬁliates in a country. Host Country Inward FDI Restrictions and Host Country Capital Controls are dummy variables drawn from Brune (2004) and Shatz (2000), respectively. Host Country Trade Openness is the index of constant price openness taken from Heston et al. (2002). The Log of Host Country GDP per Capital is derived from data provided in the World Bank World Development Indicators. Log of Real Exchange Rate is computed using nominal exchange rates and measures of inﬂation are from the IMF's IFS database. All speciﬁcations include country and year ﬁxed effects. ** and * denote signiﬁcance at the 1% and 5% levels.

36

L. Branstetter et al. / Journal of International Economics 83 (2011) 27–36

Southern producers. Evidence from industry-level value-added data suggests that this is not the case. Rather, the estimated effect of reform on industry activity implies that the increase in activity by MNEs and other ﬁrms that are not engaging in imitative activity more than compensates for any decline in imitation. Furthermore, evidence from highly disaggregated U.S. import data indicates that initial export episodes increase following reform. Analyses of changes in industrial activity after reform point out that activity in technology-intensive industries responds particularly strongly to reform. Stronger IPR in the South appears to lead to an acceleration of production shifting, enhancing Southern industrial development. Over the long run, production shifting should free up Northern resources for investment in innovative activity. In this paper, we do not attempt to estimate the magnitude or timing of this longer run, general equilibrium effect. However, other researchers have noted a robust expansion of U.S. innovative activity in the 1990s, even as manufacturing jobs have continued to move offshore. The rate of growth of real R&D spending in the U.S. accelerated signiﬁcantly in the 1980s and 1990s as the IPR reforms we study unfolded. Relative to inventors based in other countries, those in the U.S. appear to have increased their generation of new ideas.30 Along with this surge in innovative outcomes has come an acceleration in total factor productivity growth, which has persisted in recent years.31 These developments are consistent with the kind of general equilibrium resource reallocation stressed in product cycle models like the one in Grossman and Helpman (1991b). However, these are complex phenomena with multiple causes; exploring the potential link between production shifting and the apparent acceleration of innovation in the U.S. is a focus of ongoing research. Acknowledgements We wish to thank Pol Antràs, Nicholas Bloom, Paul David, Jonathan Eaton (the Editor), Amy Glass, Gene Grossman, Bronwyn Hall, Elhanan Helpman, Fuat Sener, two anonymous reviewers, and seminar participants at Columbia University, Keio University, LSE, Stanford University, the University of Adelaide, UC-Berkeley, the University of Pittsburgh, and the NBER ITI and Productivity meetings for helpful comments. We are grateful to Matej Drev, Sergei Koulayev, and Yoshiaki Ogura for excellent research assistance and to the National Science Foundation (SES grant no. 0241781) for ﬁnancial support. Foley also thanks the Division of Research at Harvard Business School for providing generous funding. References Bernard, A., Redding, S., Schott, P., 2010. Multi-product ﬁrms and product switching. The American Economic Review 100 (1), 70–97. Branstetter, L., Fisman, R., Foley, F., 2006. Does stronger intellectual property rights increase international technology transfer? Empirical evidence from U.S. ﬁrm-level data. Quarterly Journal of Economics 121 (1), 321–349. Branstetter, L., Fisman, R., Foley, F., Saggi, K., 2007. Intellectual Property Rights, Imitation, and Foreign Direct Investment: Theory and Evidence. NBER WP No. 13033. Brune, N., 2004. In Search of Credible Commitments: The IMF and Capital Account Liberalization in the Developing World. Working paper. Yale University.

30 31

See Kortum and Lerner (1999) for a discussion of evidence based on patent data. See Gordon (2003) and the studies cited therein.

Chaudhuri, S., Goldberg, P., Jia, P., 2006. Estimating the effects of global patent protection in pharmaceuticals: a case study of quinolones in India. The American Economic Review 96 (5), 1477–1514. Cohen, W., Goto, A., Nagata, A., Nelson, R., Walsh, J., 2002. R&D spillovers, patents, and the incentives to innovate in Japan and the United States. Research Policy 31 (8–9), 1349–1367. Eaton, J., Kortum, S., 1996. Trade in ideas: patenting and productivity in the OECD. Journal of International Economics 40 (3–4), 251–278. Eaton, J., Kortum, S., 1999. International technology diffusion: theory and measurement. International Economic Review 40 (3), 537–570. Feenstra, R., Rose, A., 2000. Putting things in order. The Review of Economics and Statistics 82 (3), 369–382. Feenstra, R., Romalis, J., Schott, P., 2001. U.S. Imports, Exports, and Tariff Data, 1989–2001. NBER WP No. 9387. Ginarte, J., Park, W., 1997. Determinants of patent rights: a cross-national study. Research Policy 26 (3), 283–301. Glass, A.J., Saggi, K., 2002. Intellectual property rights and foreign direct investment. Journal of International Economics 56 (2), 387–410. Gordon, R.J., 2003. Exploding productivity growth: context, causes, implications. Brookings Papers on Economic Activity 2, 207–298. Grossman, G.M., Helpman, E., 1991a. Innovation and Growth in the Global Economy. MIT Press, Cambridge. Grossman, G.M., Helpman, E., 1991b. Endogenous product cycles. The Economic Journal 101 (408), 1214–1229. Hall, B., Jaffe, A., Trajtenberg, M., 2001. The NBER Patent Citation Data File: Lessons, Insights, and Methodological Tools. NBER WP No. 8498. Hausman, J., Hall, B., Griliches, Z., 1984. Econometric models for count data with an application to the patents-R&D relationship. Econometrica 52 (4), 909–938. Helpman, E., 1993. Innovation, imitation, and intellectual property rights. Econometrica 61 (6), 1247–1280. Heston, A., Summers, R., Aten, B., 2002. Penn World Table Version 6.1. Center for International Comparisons at the University of Pennsylvania (CICUP). Javorcik, B., 2004a. Does foreign direct investment increase the productivity of domestic ﬁrms? In search of spillovers through backward linkages. The American Economic Review 94 (3), 605–627. Javorcik, B., 2004b. The composition of foreign direct investment and the protection of intellectual property rights in transition economies. European Economic Review 48 (1), 39–62. Kortum, S., Lerner, J., 1999. Stronger protection or technological revolution: what is behind the recent surge in patenting? Research Policy 28 (1), 1–22. Kuemmerle, W., 1999. The drivers of foreign direct investment into research and development: an empirical investigation. Journal of International Business Studies 30, 1–24. Lai, E.L.C., 1998. International intellectual property rights protection and the rate of product innovation. Journal of Development Economics 55 (1), 133–153. Lerner, J., 2002. 150 years of patent protection. The American Economic Review 92 (2), 221–225. Markusen, J.R., 2001. Contracts, intellectual property rights, and multinational investment in developing countries. Journal of International Economics 53 (1), 189–204. Maskus, K.E., 2000. Intellectual Property Rights in the Global Economy. Institute for International Economics, Washington DC. McCalman, P., 2001. Reaping what you sow: an empirical analysis of international patent harmonization. Journal of International Economics 55 (1), 161–186. McFetridge, D., 1997. Intellectual property rights and the location of innovative activity: the Canadian experience with compulsory licensing of patented pharmaceuticals. Working Paper. Carleton University. McKinsey Global Institute, 2000. Why the Japanese Economy is not Growing: Micro Barriers to Productivity Growth. McKinsey & Company, Inc. Poole, J., 2009. Knowledge transfers from multinational to domestic ﬁrms: evidence from worker mobility. Working Paper. University of California at Santa Cruz. Porter, M., Takeuchi, H., Sakakibara, M., 2000. Can Japan Compete? Perseus, Cambridge, MA. Qian, Y., 2007. Do national patent laws stimulate innovation in a global patenting environment? A cross-country analysis of pharmaceutical patent protection, 1978–2002. The Review of Economics and Statistics 89 (3), 436–453. Shatz, H., 2000. The location of U.S. multinational afﬁliates. Ph.D. dissertation, Harvard University. United States Trade Representative, Executive Ofﬁce of the President, Various Issues. National Trade Estimate Report on Foreign Trade Barriers.

Journal of International Economics 83 (2011) 37–52

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

International trade in durable goods: Understanding volatility, cyclicality, and elasticities Charles Engel a, Jian Wang b,⁎ a b

Department of Economics, 1180 Observatory Drive, University of Wisconsin, Madison, WI 53706, United States Research Department, Federal Reserve Bank of Dallas, 2200 N. Pearl Street, Dallas, TX 75201, United States

a r t i c l e

i n f o

Article history: Received 3 July 2008 Received in revised form 27 July 2010 Accepted 29 August 2010 Available online 9 September 2010 JEL classiﬁcation: E32 F3 F4

a b s t r a c t Data for OECD countries document: 1. imports and exports are about three times as volatile as GDP; 2. imports and exports are pro-cyclical, and positively correlated with each other; 3. net exports are counter-cyclical. Standard models fail to replicate the behavior of imports and exports, though they can match net exports relatively well. Inspired by the fact that a large fraction of international trade is in durable goods, we propose a two-country two-sector model in which durable goods are traded across countries. Our model can match the business cycle statistics on the volatility and comovement of the imports and exports relatively well. The model is able to match many dimensions of the data, which suggests that trade in durable goods may be an important element in open-economy macro models. © 2010 Elsevier B.V. All rights reserved.

Keywords: Durable goods International real business cycles Elasticity puzzle Backus–Smith puzzle

1. Introduction One of the most established empirical regularities in international real business cycle (IRBC) analysis is the counter-cyclical behavior of net exports. In contrast, the behavior of imports and exports themselves has been largely neglected in the literature. They are much more volatile than GDP and both are pro-cyclical, facts which are at odds with the predictions of standard models.1 The large drop in the volume of world trade in 2008–2009 has attracted ample notice. But the drop in international trade is generally consistent with the patterns of cyclical trade movements we have seen over the past 35 years. These data lead us to expect a large drop in the volume of trade when markets experience a steep recession, especially if it is expected to be prolonged. Inspired by the evidence that a large fraction of international trade is in durable goods, we propose a twocountry two-sector model, in which durable goods are traded across countries. Simulation results show that our model can match the trade sector data much better than standard models.

⁎ Corresponding author. Tel.: + 1 214 922 6471. E-mail addresses: [email protected] (C. Engel), [email protected] (J. Wang). 1 The only paper that examines import and export volatility to our knowledge is that of Zimmermann (1999). That paper uses exogenous exchange rate shocks to generate the volatility of imports and exports. This explanation is contradictory to the positive correlation between imports and exports. We give more details later. 0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.08.007

We ﬁrst document two empirical ﬁndings that are very robust across our 25-OECD-country data: 1. The standard deviations of real imports and exports are about two to three times as large as that of GDP.2 2. Real imports and exports are pro-cyclical and also positively correlated with each other. We label the ﬁrst ﬁnding “trade volatility”, and the second one “positive comovement”. We also conﬁrm in our dataset the well-documented negative correlation between net exports and output. In standard international business cycle models, imports and exports are far less volatile than in the data — they are even less volatile than GDP. We demonstrate this in a variety of models, both real business cycle and sticky-price dynamic models. We emphasize that the issue is not resolved by building versions of the model with high real exchange rate volatility. Although a more volatile exchange rate helps to increase the volatility of imports and exports, it generates a negative correlation between imports and exports. This is at odds with the ﬁnding of “positive comovement”. We propose a model in which countries primarily trade durable goods, inspired by the fact that a large portion of international trade is in durable goods. In OECD countries, trade in durable goods on average accounts for about 70% of imports and exports. The importance of capital goods in international trade has also been

2 Similar results are also reported in Table 11.7 of Backus et al. (1995), Heathcote and Perri (2002), and Zimmermann (1999).

38

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

documented by Eaton and Kortum (2001). Boileau (1999) examines an IRBC model with trade in capital goods. That study ﬁnds that allowing direct trade in capital goods improves the model's performance in matching the volatility of net exports and the terms of trade. Boileau's model shares some characteristics of the one we examine. It does not include consumer durable goods, which are necessary to understand some aspects of the data. Moreover, Boileau (1999) does not examine the implications of his model for imports and exports individually, which is the focus of our study. Erceg et al. (2008) also emphasize that trade in capital goods helps model to replicate trade volatility. They argue that trade balance adjustment may be triggered by investment shocks from either the home or foreign country and such adjustment may not cause substantial real exchange rate ﬂuctuations. Warner (1994) ﬁnds that global investment demand has been an important determinant of US exports since 1967. However, we ﬁnd that a model with trade in capital goods but not consumer durables is inadequate. In order to match the volatility of the trade data, a large share of traded goods must be durable. But if we take all of those traded goods to be capital, then the model would require, for example, that the US obtains almost all capital goods from imports while simultaneously exporting large quantities of capital. Our model goes further by including both capital and durable consumption goods in international trade.3 In our baseline twocountry two-sector model, nondurable goods are nontraded. Durable consumption ﬂows require both home and foreign durable goods varieties and capital goods are aggregated from home and foreign varieties of capital. Simulation results show that the benchmark model can successfully replicate “trade volatility” and “positive comovement”. In addition, net exports in our model are countercyclical and as volatile as in the data. So our model can match the trade sector data much better than the standard models. This improvement is not at the cost of other desirable features of standard models. The aggregate variables such as output, consumption, investment and labor, can also match the data well. Our ﬁnding that imports and exports are more volatile than GDP and pro-cyclical, and our model of durables in international trade, have the potential to explain a signiﬁcant portion of the drop in international trade during the global ﬁnancial crisis. Drawing on our work, and using disaggregated data on imports and exports, Levchenko et al. (2009) ﬁnd strong evidence that the compositional effect played an important role in explaining the collapse of US trade in 2008: international trade occurs disproportionately in the sectors that domestic demand and production collapsed the most. The investment and durable consumption sectors are good examples. Since investment and durables expenditure are several times more volatile than GDP and international trade is highly concentrated in these durable goods, trade would be expected to experience larger swings than GDP as well. They also ﬁnd evidence for the supply-chain/ vertical linkage effect: trade fells more in sectors that are used intensively as intermediate inputs. The positive comovement of imports and exports documented in our paper suggests that both imports and exports decline during an economic downturn. The countercyclicality of net exports is caused by a sharper decline of imports than exports rather than an increase of exports during economic downturns. We also consider a model in which both durable and nondurable goods are traded across countries. In this model, nondurable goods account for about 30% of trade as we found in OECD countries. The model generates results similar to our benchmark model. The only

3 Baxter (1992) has durable consumption in a two-sector model. The model setup is very different from ours and is used to address different issues. Sadka and Yi (1996) build a simple small-country real-business-cycle model with durable consumption goods. They use this model to demonstrate that the increase of consumption durables due to a permanent decrease in their prices may be an important element in explaining the 1980s US trade deﬁcits.

noticeable difference is that imports and exports become less volatile, but both of them are still more than twice as volatile as output. An important empirical puzzle that has confronted international trade economists is the mismatch between estimated short-run and long-run elasticities of import demand. As Ruhl (2005) and others have discussed, typically short-run elasticities are estimated to be near unity, but long-run elasticities are generally found to be considerably higher. That pattern arises naturally in any model such as ours in which durable capital and consumer goods are traded, because durable stocks cannot be adjusted quickly in response to price changes. Another interesting feature of our model is its implications for understanding comovement of relative consumption and real exchange rates, as in the Backus and Smith (1993) puzzle. Our model suggests that it may be important to distinguish carefully consumption purchases (which include purchases of consumer durable goods) and consumption ﬂows (which include the ﬂow of services delivered from previously purchased durables). The remainder of the paper is organized as follows: Section 2 displays statistics on “trade volatility” and “positive comovement”. Section 3 describes our two-country two-sector benchmark model. Section 4 explains our calibration of the model. Section 5 compares simulation results of our benchmark model and the standard models used in the literature. Section 6 concludes. 2. Empirical ﬁndings and share of trade in durable goods In this section, we ﬁrst show some facts about international real business cycles: 1. Real imports and exports are about two to three times as volatile as GDP. 2. Both real imports and exports are procyclical and positively correlated with each other. 3. Real net exports are counter-cyclical. Then we present evidence that trade in durable goods accounts for a large portion of imports and exports in OECD countries. 2.1. Trade volatility and cyclicality Our dataset includes quarterly real GDP, real imports, and real exports, of 25 OECD countries during the period between 1973Q1 and 2006Q3.4 The data are from the OECD Economic Outlook database. All variables are logged except net exports which are divided by GDP, and HP ﬁltered with a smoothing parameter of 1600. Table 1 shows the volatility of those variables and comovement of real imports and real exports with GDP. The standard deviation of GDP on average, is 1.51%. Both real imports and exports are much more volatile than GDP. On average, the imports are 3.3 times, and exports are 2.7 times as volatile as GDP. This result is not driven by outliers. The sample median is very close to the sample mean. The volatilities of imports and exports in the US are close to the sample mean. However, the ratio of net exports to GDP in the US is less volatile than it is in any other country. Two things stand out for comovement of real imports and real exports with GDP. First, both imports and exports are pro-cyclical. This result is very robust: imports are positively correlated with GDP in all 25 countries. The average correlation is 0.63. The same is true for exports except in two countries: Denmark and Mexico. The average correlation between exports and GDP is 0.39. Second, imports and exports are positively correlated in all countries except Australia, Mexico, New Zealand and Spain. The average correlation between imports and exports is 0.38. In this table, we also conﬁrm a welldocumented ﬁnding in previous studies: net exports are countercyclical. This is true in all countries except Austria and Hungary. The average correlation between net exports and GDP is −0.24. 4 Austria starts from 1988Q1, Czech Republic from 1993Q1, and Hungary from 1991Q1. The data of Germany are for West Germany only which end in 1991Q1. The data after uniﬁcation (1991Q1–2006Q3) show similar patterns.

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

39

Table 1 Volatility and comovements of international trade. Standard deviations relative to that of GDP

Correlation with GDP

Country

SD of GDP (%)

Real import

Real export

NetExport a GDP

Real import

Real export

NetExport a GDP

corr(IM,EX)b

Australia Austria Belgium Canada Czech Republic Denmark Finland France Germany Hungary Iceland Ireland Italy Japan Korea Mexico Netherlands New Zealand Norway Portugal Spain Sweden Switzerland UK United States Mean Median

1.38 0.88 1.03 1.42 1.52 1.35 2.02 0.86 1.29 0.97 2.18 1.62 1.31 1.22 2.43 2.36 1.28 2.58 1.26 1.95 1.04 1.35 1.51 1.36 1.52 1.51 1.36

4.23 2.10 2.74 3.15 2.39 2.65 2.74 3.97 2.26 4.19 3.22 3.04 3.44 4.19 3.08 5.97 2.28 2.39 4.01 2.96 4.23 3.14 2.78 2.72 3.33 3.25 3.08

2.69 2.75 2.36 2.65 2.61 2.46 2.73 3.22 2.86 6.53 1.91 1.96 3.03 3.51 2.70 2.53 1.99 1.53 3.22 2.72 2.86 2.54 2.08 2.17 2.63 2.73 2.65

0.63 0.54 0.67 0.66 1.02 0.72 0.68 0.58 0.69 2.66 1.28 0.73 0.70 0.36 0.82 0.89 0.62 0.66 1.37 0.61 0.77 0.70 0.54 0.39 0.25 0.78 0.68

0.49 0.60 0.73 0.74 0.53 0.55 0.73 0.77 0.78 0.54 0.59 0.38 0.70 0.60 0.81 0.75 0.61 0.40 0.34 0.81 0.56 0.61 0.66 0.61 0.83 0.63 0.61

0.16 0.67 0.74 0.66 0.33 − 0.09 0.22 0.68 0.52 0.53 0.45 0.50 0.38 0.16 0.31 − 0.20 0.62 0.22 0.36 0.51 0.05 0.46 0.68 0.45 0.41 0.39 0.45

− 0.33 0.36 − 0.17 − 0.12 − 0.09 − 0.57 − 0.41 − 0.27 − 0.06 0.14 − 0.29 − 0.08 − 0.26 − 0.34 − 0.62 − 0.78 − 0.10 − 0.18 − 0.03 − 0.38 − 0.46 − 0.25 − 0.10 − 0.25 − 0.47 − 0.24 − 0.25

− 0.10 0.85 0.92 0.62 0.74 0.53 0.36 0.57 0.40 0.25 0.06 0.77 0.38 0.21 0.28 − 0.32 0.75 − 0.25 0.13 0.44 − 0.14 0.53 0.72 0.59 0.19 0.38 0.40

Note: The data are from OECD Economic Outlook database. They are quarterly data of OECD 25 countries during the period between 1973Q1 and 2006Q3. (Due to data limitation, Austria starts from 1988Q1, Czech Republic starts from 1993Q1, and Hungary starts from 1991Q1.) The data of Germany are for West Germany only which end in 1991Q1. The data after uniﬁcation (1991Q1–2006Q3) show similar patterns as reported in this table. Real imports (exports) are more than twice as volatile as GDP in 22 (19) out of 25 countries at the 5% level in a one-side test. Real imports (exports) are positively correlated with GDP in 25 (21) out of 25 countries at the 5% level in a one-side test. Under the same test, real net exports are negatively correlated with GDP in 15 out of 25 countries and real imports and exports are positively correlated in 19 out of 25 countries. These results are obtained from 1000 bootstraps with replacement. Similar volatility and cyclicality of imports and exports is also found in aggregate EU data. Results are available upon request. NetExport a All variables are logged (except for ), and HP ﬁltered with a smoothing parameter of 1600. b

GDP

corr(IM, EX) is the correlation of real imports and exports.

2.2. Trade in durable goods in OECD countries Here we present some descriptive statistics on trade ﬂows that help to motivate our model of trade in durables. We obtain our 25 OECD-country data from NBER–UN World Trade Data and use the latest available data (year 2000) to calculate the share of durable goods in international trade. The data are at the 1- or 2-digit SITC level. Table 2 shows how we divide imports and exports into categories of durable and nondurable goods. (See Appendix A for more details.) Table 3 reports the share of durable goods in international trade. On average durable goods account for about 70% of imports and exports (excluding energy products SITC 3) in these countries. Results are similar if we also exclude raw materials (right panel of Table 3). In particular, machineries and transportation equipment (SITC 7) on average account for more than 40% of trade for OECD countries.5 These ﬁndings are very similar to those reported by Baxter (1995) and Erceg et al. (2008). We also examine the volatility of durable goods trade and other categories of trade in a data set for US trade only. We use quarterly nominal US trade data at the 2-digit SITC level from the US International Trade Commission (http://www.usitc.gov/). Import and export price indexes at the 2-digit SITC level are obtained from the Bureau of the Census through Haver Analytics. Nominal trade data are deﬂated by corresponding price indexes to calculate real imports and exports. In the end, we have real import and export data at the 25 We note two outliers for exports. Exports of New Zealand and Iceland are mainly in category zero (FOOD AND LIVE ANIMALS).

digit SITC level for 1997Q1–2006Q2. Imports and exports are classiﬁed into three categories: raw materials, durable goods and nondurable goods according to the standard described above. Real imports and exports are logged and HP ﬁltered with a smoothing parameter of 1600. We calculate the standard deviation for each category. In exports, raw materials and durable goods are much more volatile than nondurable goods: the standard deviations of raw materials and durable goods are respectively 7.78% and 6.54%, but only 2.86% for nondurable goods. Imports show less dispersion in volatility: the standard deviations of raw materials and durable goods are 5.02% and 5.00% respectively. It is 4.89% for nondurable goods. We note that these statistics are not precise given our rough classiﬁcation of goods into the durable and nondurable categories, and given that we use only 38 observations of HP ﬁltered data. Durable goods also show stronger correlation with GDP in our data. For imports, the correlation between durable goods and GDP is 0.53. It is −0.35 for raw materials and −0.17 for nondurable goods. For exports, the correlation between durable goods and GDP is 0.82. It is − 0.02 for raw materials and 0.65 for nondurable goods. 3. A two-country benchmark model There are two symmetric countries in our model, Home and Foreign. There are two production sectors in each country: nondurable and durable goods. All ﬁrms are perfectly competitive with ﬂexible prices. Nondurable goods can only be used for domestic consumption. Durable goods are traded across countries and used for durable consumption and

40

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

Table 2 Dividing SITC categories into different sectors. SITC

Description

Sector

0 1 2 3 4 5 6 61 62 63 64 65 66 67 68 69 7 8 81 82 83 84 85 87 88 89 9 91 93 95 96 97

Food and live animals Beverages and tobacco Crude materials, inedible, except fuels Mineral fuels, lubricants and related materials Animal and vegetable oils, fats and waxes Chemicals and related products, n.e.s. Manufactured goods classiﬁed chieﬂy by material Leather, leather manufactures, n.e.s., and dressed furskins Rubber manufactures, n.e.s. Cork and wood manufactures (excluding furniture) Paper, paperboard and articles of paper pulp, of paper or of paperboard Textile yarn, fabrics, made-up articles, n.e.s., and related products Non-metallic mineral manufactures, n.e.s. Iron and steel Non-ferrous metals Manufactures of metals, n.e.s. Machinery and transport equipment Miscellaneous manufactured articles Prefabricated buildings; sanitary, plumbing, heating and lighting ﬁxtures and ﬁttings, n.e.s. Furniture, and parts thereof; bedding, mattresses, mattress supports, cushions and similar stuffed furnishings Travel goods, handbags and similar containers Articles of apparel and clothing accessories Footwear Professional, scientiﬁc and controlling instruments and apparatus, n.e.s. Photographic apparatus, equipment and supplies and optical goods, n.e.s.; watches and clocks Miscellaneous manufactured articles, n.e.s. Commodities and transactions not classiﬁed elsewhere in the SITC Postal packages not classiﬁed according to kind Special transactions and commodities not classiﬁed according to kind Coin, including gold coin; proof and presentation sets and current coin Coin (other than gold coin), not being legal tender Gold, non-monetary (excluding gold ores and concentrates)

Nondurable Nondurable Raw materials Energy products Nondurable Nondurable Durable Durable Nondurable Nondurable Nondurable Durable Durable Durable Durable Durable Durable Durable Nondurable Nondurable Nondurable Durable Durable Nondurable Nondurable Nondurable Durable Durable Durable

Note: See Appendix A for more details.

capital accumulation. Because of the symmetry between these two countries, we describe our model focusing on the Home country.6 Trade in capital goods and consumer durables would introduce too much volatility in trade, so we allow for installation costs. This is a wellknown feature of international RBC models, but this also allows us to build a model consistent with another widely-recognized fact: trade elasticities are higher in the long run in response to persistent shocks than they are in the short run. In addition, we introduce an iceberg cost of trade. Here, we want to capture the idea that there is a “home bias” in the consumption of durables, as well as in the use of capital goods in production. Especially for large economic areas such as the US or the European Union, imports are a relatively small component of the overall consumption basket, or mix of inputs used in production. Because we model traded goods as being highly substitutable in the long-run, it does not seem natural to simultaneously introduce home bias directly into the utility function or production function. Instead, and consistent with much of the recent literature in trade, we posit that there are costs to trade which lead to this home bias even in the long run. We note that there is a tension in modeling the behavior of trade volumes over the business cycle. Imports and exports are pro-cyclical and their standard deviation (in logs) is much larger than that of GDP. At the same time, they are apparently not very responsive in the short run to price changes. The model of consumer durables and investment goods captures these features for reasonable parameter values. We discuss the calibration in Section 4, after the presentation of the model. 3.1. Firms There are two production sectors in each country: the nondurable goods sector and the durable goods sector. Nondurable and durable 6

We list all equilibrium conditions for both countries in an appendix posted on the authors' websites.

goods in the Home country are produced from capital and labor according to χ 1−χ j j j j LHt ; Y Ht = AHt K Ht

ð1Þ

where j ∈ {N, D} denotes nondurable (N) and durable (D) goods j j sectors. AHt and LHt are respectively the TFP shock and labor in sector j. j Capital KHt is a CES composite of Home- and Foreign-goods capital j

K Ht =

γ−1 γ−1 1 1 jH jF αγ K Ht γ + ð1−αÞγ K Ht γ

!

γ γ−1

;

ð2Þ

jk

where in the notation such as Kit , we use the subscript i to denote the country in which the capital is used, the ﬁrst superscript j to denote the sector (nondurable or durable) and the second superscript k to denote the origin of the goods. For instance, K NH Ht is the Home country produced durable good that is used in the nondurable goods sector of the Home country. The ﬁrm buys labor and rents capital from households in competitive markets. For given wage (WHt) and rental price of capital jH jF (RHt and RHt), the ﬁrm chooses capital and labor to minimize the cost of production. Capital is not mobile across sectors though we assume that labor can move freely from one sector to another. The nondurable and durable goods markets are also competitive, so the price of j nondurable and durable goods P Ht is equal to the marginal cost −1 χ j j j 1−χ −χ χ−1 PHt = AHt RHt W Ht χ ð1−χÞ :

ð3Þ

From the ﬁrm's cost minimization problem, we can ﬁnd the standard demand function for capital and labor by equating the marginal productivity to the real factor cost.

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

where ψ is the weight of Home durable goods in the durable consumption stock and θ is the elasticity of substitution between the Home and Foreign durable goods. The law of motion for durable consumption is

Table 3 Share of durable goods in trade. Exclude energy products

Exclude materials and energy

Country

Import

Export

Import

Export

Australia Austria Belgium Canada Czech Rep Denmark Finland France Germany Hungary Iceland Ireland Italy Japan Korea Mexico Netherland New Zealand Norway Portugal Spain Sweden Switzerland UK US Mean Median

0.70 0.69 0.66 0.77 0.72 0.60 0.72 0.67 0.69 0.74 0.55 0.73 0.65 0.57 0.76 0.73 0.68 0.66 0.70 0.65 0.68 0.68 0.65 0.69 0.69 0.68 0.69

0.56 0.67 0.66 0.64 0.75 0.47 0.61 0.68 0.71 0.77 0.28 0.59 0.64 0.89 0.78 0.78 0.60 0.26 0.59 0.53 0.65 0.73 0.69 0.74 0.75 0.64 0.66

0.71 0.70 0.67 0.77 0.73 0.61 0.73 0.68 0.70 0.75 0.56 0.73 0.66 0.58 0.77 0.74 0.69 0.66 0.71 0.66 0.69 0.69 0.66 0.70 0.70 0.69 0.70

0.45 0.69 0.66 0.69 0.77 0.48 0.65 0.68 0.71 0.78 0.28 0.59 0.64 0.89 0.78 0.78 0.61 0.26 0.61 0.54 0.66 0.76 0.69 0.74 0.77 0.65 0.68

k

k

Δ Ht =

∞ j Et ∑ β u DHt +j ; CHt +j ; LHt +j ; j=0

where the period utility u(DHt + j, CHt + j, LHt + j) is a function of durable consumption (DHt + j), nondurable consumption (CHt + j), and labor supply (LHt + j). The period utility function takes the form of 2 31−σ 1 ζ−1 ζ 1 ζ−1 ζ−1 ν ζ ζ 4 μ ζ D + ð1−μ Þζ C −ρLHt 5 Ht Ht

jk

DHt = ψ

θ−1 θ

+ ð1−ψÞ

F DHt

θ−1 θ

ð4Þ

#

θ θ−1

;

jk

ð8Þ

where j ∈ {D, N} and k ∈ {H, F}. We follow the literature to include capital adjustment costs in our model. In the Home country, it takes the following form 1 jk jk 2 j ϕ I −δK Ht = K Ht ; 2 2 Ht

ð9Þ

where j ∈ {D, N} and k ∈ {H, F}. Symmetric adjustment costs exist in the Foreign country. The Home and Foreign countries can only trade real bonds, which are in terms of the Home durable goods. It is well-known that transient shocks have a permanent wealth effect in a linearized openeconomy model with incomplete international ﬁnancial markets. To make our model stationary, we (2004) to introduce a follow Kollmann 1 2 . Φ is very close to zero and quadratic bond holding cost ΦBHt +1 2 the cost does not affect any results in our model.8 For the given production structure, the household's budget constraint is N

:

1 θ

jk

K Ht +1 = ð1−δÞK Ht + IHt ;

B 1 2 H H NH NH DH DH dHt + ΔHt + IHt + ΛHt + IHt + ΛHt + Ht +1 + ΦBHt +1 2 1 + it DF NF DF DF dFHt + ΔFHt + IHt + ΛNF + PHt Ht + IHt + ΛHt DH

PHt CHt + PHt

It is an augmented Greenwood et al. (1988), (GHH henceforth) utility function with consumption as a CES composite of durable and nondurable consumption. The stock of durable consumption is a H F function of the Home (DHt) and Foreign (DHt) durable consumption stocks

H DHt

ð7Þ

k

In the Home country, the representative household supplies labor, accumulates and rents capital to ﬁrms, chooses nondurable consumption and accumulates durable consumption stock to maximize expected lifetime utility

1 k k 2 ϕ1 dHt −δD DHt = DHt ; 2

where Δ Ht is the cost of changing durables produced by country k.7 If there were no adjustment costs to durables, durable consumption purchases would be very volatile in response to shocks. Empirical work (see for example, Mankiw, 1982 and Gali, 1993), ﬁnds that durable consumption adjusts more smoothly and is less volatile than a model with no adjustment costs would imply. Gali (1993) suggests that adjustment costs may account for the excess smoothness of durable consumption, and indeed Startz (1989) ﬁnds that adjustment costs can account for the behavior of durable consumption in a permanent income model. Bertola et al. (2005) ﬁnd support on micro level data for a model with a ﬁxed cost of adjustment. Aggregate consumption is not likely to exhibit the same lumpiness as micro data, so we adopt the standard quadratic adjustment cost formulation (as in Startz, 1989). The law of motion for capital stocks in the durable and nondurable sectors is given by

jk

1 θ

ð6Þ k

ΛHt =

"

k

where k ∈ {H, F} denotes the Home and Foreign countries. dHt is the kcountry durable consumption goods purchased by the household at time t. As in Erceg and Levin (2006) and Whelan (2003), the household also has to pay a cost to adjust the durable consumption stock

3.2. Households

1−σ

k

DHt +1 = ð1−δD ÞDHt + dHt ;

Note: Data are from international trade data, NBER–United Nations World Trade Data (http:// cid.econ.ucdavis.edu). Entries are shares of durable goods in imports and exports (year 2000). Left panel of the table reports results for imports and exports excluding energy products (SITC 3). Raw materials (SITC 2) and energy products (SITC 3) are excluded from imports and exports in the right panel. Share of durable goods in bilateral trade among Canada, EU, Japan and US is similar to the results reported in this table. Results are available upon request.

ut =

41

ð5Þ

DH NH NF NF DH DH DF DF ≤WHt LHt + PHt BHt + RNH Ht KHt + RHt KHt + RHt KHt + RHt KHt ;

ð10Þ

7 Adjustment costs are scaled by the total durable consumption stock (DHt) so that H H F F the cost of adding new durable consumption (dHt − δD DHt and dHt − δD DHt) is the same for both types of durable consumption. The same format is also used in the capital adjustment cost functions. 8 There are several other techniques used in the literature to deal with this nonstationarity problem. See Schmitt-Grohé and Uribe (2003) for more discussion.

42

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

Fig. 1. Structure of benchmark model. Note: Numbers in this ﬁgure are percentage of total output.

DF where PHt is the price of Foreign country produced durable goods, which is in terms of the Home country's currency. it is the return to the real bond BHt + 1. Subject to this budget constraint, the household maximizes expected lifetime utility.

3.3. Other equilibrium conditions Nondurable goods can only be used for domestic nondurable consumption. So the market clearing condition for Home nondurable goods is N

YHt = CHt :

ð11Þ

Durable goods are used for durable consumption and capital investment in both countries. We also assume there is an iceberg trade cost for international trade. Only a fraction 1 − τ of goods arrives in the destination country, so the market clearing condition for Home durable goods is D NH DH DH = dHHt + ΔHHt + IHt + ΛNH YHt Ht + IHt + ΛHt +

+

1 2 ΦBHt +1 2 1 2

1−τ

:

LHt = LHt + LHt

D

ð13Þ

BHt + BFt = 0:

ð14Þ

We assume that after taking into account the trade cost, the law of one price holds DF

PHt =

St PFtDF 1−τ

DH PHt DH = PFt ; St ð1−τÞ DF PHt

ð17Þ

where ω1 is the steady-state expenditure share of nondurable consumption. ω2 and ω3 are respectively the steady-state expenditure shares of Home and Foreign durable consumption. This is not the same as the utility-based CPI, but is closer to the CPI measure used in national accounts. The CPI deﬂated real exchange rate is deﬁned by Qt =

St PFt : PHt

ð18Þ

To solve our model, we divide all nominal prices in the Home country by the price of nondurable goods (PN Ht). That is, we use the nondurable goods as numeraire. In the Foreign country, all nominal prices are divided by the price of Foreign nondurable goods (PN Ft). 4. Calibration

The labor and bond markets clearing conditions are N

ω ω 1 3 N DH ω2 DF PHt = PHt PHt PHt ;

ð12Þ

2

NH DH DH dHFt + ΔHFt + IFt + ΛNH Ft + IFt + ΛFt + ΦBFt +1

In Section 5, we report real exchange rates based on the consumer price index (CPI). In the Home country, the CPI is deﬁned by:

ð15Þ

ð16Þ

is the price of Foreign durable goods in the Home country. where St is the nominal exchange rate deﬁned as the value of one unit of Foreign currency in terms of the Home currency.

We calibrate our model such that in the steady state, the structure of the economy is the same as in Fig. 1.9 In our benchmark economy, durable goods account for 40% of output. Among durable goods, half are used for consumption (equivalent to 20% of total output) and the other half are used for investment (20% of total output).10 Among durable consumption goods, 65% is used for domestic consumption (13% of total output) and 35% is used for exports (7% of total output). Among durable investment goods, 70% is used for domestic investment (14% of total output) and 30% is used for exports (6% of total output). In this economy, investment accounts for 20% of total output and consumption (durable plus nondurable) accounts for the remaining 80%. The trade share of output is 13%. Those features match the US data closely. Table 4 shows parameter values that we use to match our benchmark model with the described economy structure. We set the shares of home goods in capital (α) and durable consumption (ψ) at 50%. That is, there is no home bias exogenously built into our 9 Details about how to solve the steady state can be found in an appendix posted on the authors' websites. 10 Durable expenditure in our calibration is higher than the US data, which is about 15% of output. However, many goods with characteristics of durables—such as shoes and clothing–are classiﬁed as nondurables in the data.

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52 Table 4 Calibration. Parameter Value α χ γ τ β δ δD μ ν ψ ρ σ θ ζ ϕ1 ϕ2 Φ Ξ1 Ξ2 σ(εN Ht) σ(εD Ht)

Description

0.5 0.36 9.1

Share of home goods in capital when trade cost is zero Capital share in production (Long-run) Elasticity of substitution between home and foreign capital 0.1 (Iceberg) International trade cost 0.99 Subjective discount factor 0.013 Depreciation rate of capital 0.05 Depreciation rate of durable consumption 0.23 Share of durable consumption stock in consumption bundle 1.65 Preference parameter of labor supply 0.5 Share of home goods in durable consumption when trade cost is zero 5.83 Preference parameter 2 Preference parameter 6.85 (Long-run) Elasticity of substitution b/t home and foreign durable consumption 1.1 Elasticity of substitution b/t durable and nondurable consumption 1.4a Durable consumption adjustment cost a 8.5 Capital adjustment cost 0.00001 Bond holding cost 0.87 AR(1) coefﬁcient of technology shock in nondurable goods sector 0.9 AR(1) coefﬁcient of technology shock in durable goods sector 0.0096 Standard deviation of productivity shock in nondurable goods sector 0.036 Standard deviation of productivity shock in durable goods sector

a Entries are values used in the benchmark model. In other models, they are adjusted to match the volatility of durable consumption and aggregate investment.

economy structure. Instead, we generate the observed low trade share from the iceberg trade cost τ. We will discuss this more later. As in Backus et al. (1992), the capital share in production (χ) is set to 36%, and the subjective discount factor is set to 0.99. The depreciation rate of durable consumption (δD) is set to 0.05, which implies a 20% annual depreciation rate for consumption durables. A similar depreciation rate has been used by Bernanke (1985) and Baxter (1996). Given those parameters, we choose other parameters to match the economy structure as in Fig. 1. We ﬁrst choose the preference parameter μ and the depreciation rate of capital (δ) jointly to match the relative size of durable and nondurable goods sectors, and the size of investment in durable goods. μ is set to 0.23 and δ is set to 0.013 such that 1. the durable goods sector accounts for 40% of total output and, 2. investment accounts for 50% of durable goods, or equivalently 20% of total output. Consumption durables account for the remaining 50% of durable goods, or equivalently 20% of total output. Two methods have been used in the literature to estimate the elasticity of substitution between the Home and Foreign goods. In the data, the trade share of output increases substantially over time after a small but permanent decrease in the tariff. The estimates from this strand of literature range from 6 to 15 with an average of 8.11 In another strand of literature, the same elasticity is estimated from transitory relative price changes at the business cycle frequency. Estimates found in these studies are much smaller, roughly around one.12

11 For instance, see Feenstra and Levinsohn (1995), Head and Ries (2001), and Lai and Treﬂer (2002). Yi (2003) also points out that to replicate this empirical ﬁnding in a general equilibrium model, we need an elasticity of more than 14, arising from the fact that measured trade grossly overstates the value added component of exports. 12 The cross-industry average in Reinert and Roland-Holst (1992) is 0.91 and it is 0.81 in Blonigen and Wilson (1999). In aggregate models, Heathcote and Perri (2002) ﬁnd a point estimate of 0.9. Bergin (2006) estimates a New Open Economy Macro model and obtains an estimate of 1.13. Corsetti et al. (2008a) estimate a trade elasticity that corresponds to an elasticity of substitution of 0.85.

43

Several studies have offered explanations for this puzzle with a common feature that the long-run elasticity of substitution is high, but the short-run elasticity is low due to some market frictions. Ruhl (2005) proposes a model in which ﬁrms must pay a ﬁxed cost to change their export status. The beneﬁts from changing export status are not enough to recover the ﬁxed cost under transitory shocks, so the elasticity of substitution is low when shocks are transitory. However, in the face of persistent shocks, ﬁrms will pay the ﬁxed cost and change their export status, which leads to a large increase of trade share even for a small, but permanent price change. Drozd and Nosal (2007) use the friction of international marketing to reduce the response of output to relative price changes. In Ramanarayanan's (2007) model, importers use foreign goods as intermediate inputs in production. Home and Foreign intermediate goods are perfectly substitutable in the long run, but switching between them in the short run is very costly. Following this literature, we assume that the Home and Foreign are highly substitutable in the long run, but in the short run there is a quadratic cost for adjusting the durable consumption and capital stocks. We will show later that our model can also deliver a reasonable short-run elasticity of substitution. The trade cost (τ) and the elasticity of substitution between the home and foreign goods are calibrated to match two empirical ﬁndings: 1. the trade share of total output is about 13%; 2. the longrun elasticity of substitution between the Home and Foreign goods is high. In our calibration, the long-run elasticity of substitution between the home and foreign capital (γ) is set to 9.1. The elasticity of substitution between the home and foreign durable consumption (θ) is set to 6.85. In steady state, trade in capital goods (durable consumption goods) accounts for 46% (54%) of total trade. This calibration of γ and θ implies an overall elasticity of 7.9, which is the same as in Head and Ries (2001).13 The trade cost (τ) is calibrated to 0.1, that is, 90% of goods arrive in their destination countries in international trade. For given γ and θ, this trade cost generates a trade share of 13%. We use different values for γ and θ to generate different home bias levels for capital and durable consumption. Capital is more biased towards home goods than durable consumption (70% vs. 65%). For given trade costs, the degree of home bias increases with the elasticity of substitution, so we assign a higher elasticity of substitution to capital goods. Alternatively, we can assume the same elasticity of substitution, but higher trade costs for capital goods. In either method, capital can have a higher level of home bias than durable consumption. We used the ﬁrst method because it matches a pattern observed in the data. For a given decrease in trade cost, the ﬁrst method predicts that the share of investment goods in international trade increases relative to the share of durable consumption. Intuitively, investment goods are more substitutable across countries than durable consumption under this setup. So when the trade cost decreases, there is more substitution for investment goods than for durable consumption. As a result, the share of investment goods in trade increases. The same pattern is also found in the US data: from 1994 to 2006—the share of capital goods except automotive in total export goods increased from 34.4% to 45.1%.14 The preference parameters σ and ν are set to their standard levels used in the GHH utility function. The parameter ρ is chosen such that labor supply is one third in steady state. We assume that the elasticity of substitution between durable and nondurable consumption is low (ζ = 1.1).15 The adjustment cost of durable consumption (ϕ1) is chosen to match the volatility of durable expenditure, which is about

13

9.1 × 46% + 6.85 × 54% ≈ 7.9. The data are from Haver Analytics (US International Transactions). Of course, this pattern is also consistent with another explanation: the trade cost decreases more for capital goods than for durable consumption goods. 15 Whelan (2003) calibrates this parameter to be 1. Baxter (1996) ﬁnds that a reasonable range for this variable is between 0.5 and 2.5. 14

44

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

three times as volatile as output in the data. The adjustment cost of capital stock (ϕ2) is calibrated to match the volatility of investment, which is about three times as volatile as output in the data. We follow Erceg and Levin (2006) in calibrating productivity shocks in the durable and nondurable goods sectors. However, there is no information about the cross-country spillovers of those shocks in their closed-economy model. Empirical ﬁndings usually suggest small cross-country spillovers. For instance, Baxter and Crucini (1995) ﬁnd no signiﬁcant international transmission of shocks, except for possible transmission between US and Canada. In Kollmann's (2004) estimate between the US and three EU countries, the spillover is 0.03. In Corsetti et al. (2008a), the spillover is − 0.06 for traded goods and 0.01 for nontraded goods. We will ﬁrst set those spillovers at zero and then choose some values used in the literature to check whether our results are robust under different shock structures. D Let AN it and Ait be respectively, the productivity shocks in nondurable and durable goods sectors of country i ∈ {H, F}. They follow univariate AR(1) processes in the benchmark model N

N

N

ð19Þ

D

D

D

ð20Þ

Ait +1 = Ξ1 Ait + εit +1 Ait +1 = Ξ2 Ait + εit +1 :

As in Erceg and Levin (2006), the AR(1) coefﬁcient Ξ1 is set to 0.87 and Ξ2 is set to 0.9. The variance–covariance matrix of innovations D N D [εN Ht εHt εFt εFt]′ takes the form of 2

σN2

6 6 σ DN 6 Σ=6 6 ρ × σ2 4 N N 0

σDN

ρN × σN2

2

σD

0

0

σN2

ρD × σD2

σDN

0

3

7 2 ρD × σD 7 7 7 σDN 7 5

ð21Þ

σD2

N N D D where σ2N is the variance of εHt (εFt ). σD2 is the variance of εHt (εFt ) and σDN is the covariance. As in Erceg and Levin (2006), the standard deviation of εN Ht (σN) is 0.0096 and it is 0.036 for σD. Within each country, the innovations are correlated across sectors. The correlation σDN is set to 0.29 as in Erceg and Levin (2006). The cross-country σD σN correlation of innovations in the durable goods sector (ρD) is 0.258 by following BKK and it is set to zero in nondurable goods sector (ρN = 0). (Corsetti et al. (2008a) estimate ρN to be zero.) This shock structure corresponds to the Benchmark model in Table 6. Alternative shock structures are also considered and will be discussed when we present our results.

5. Model performance The model is solved and simulated using ﬁrst-order perturbation methods. The model's artiﬁcial time series are logged (except for net exports) and Hodrick–Prescott ﬁltered with a smoothing parameter of 1600. The reported statistics in this section are averages across 100 simulations. Our benchmark model can match the observed IRBC statistics, including “trade volatility” and the “positive comovement” of imports and exports as documented in Section 2.1, and can replicate the elasticity puzzle in the trade literature. 5.1. International RBC statistics 5.1.1. Performance of standard models In this subsection, we show that the standard models in the literature and their extensions cannot replicate trade volatility and positive comovement simultaneously. We consider two types of models: the IRBC model and the DSGE model. Table 5 shows simulation results for these models. We use exactly the structure of the bond-economy model as in Heathcote and

Perri (2002) in our standard IRBC model (labeled HP in Table 5). This model has the same structure as BKK's model, but limits the ﬁnancial market to a real-bond market only. Baxter and Crucini (1995) compare this incomplete ﬁnancial market model with the model with perfect risk-sharing and ﬁnd that they behave very similarly if the productivity shock is not extremely persistent or the cross-country spillover of productivity shocks is high. Table 5 also reports results for the DSGE model. This is the extension of the IRBC model that assumes monopolistic competition, trade in nominal bonds, Calvo staggered price setting, and a monetary policy (Taylor) rule. Those models are often used in the studies of monetary policy in open economies. GHH is the DSGE model with the preference function proposed by Greenwood et al. (1988). We use the same class of utility function in our benchmark model. We include this model to show that our benchmark model results are not driven by this choice of utility function. We also report results for two more extensions of the DSGE model: the model with low intertemporal elasticity of substitution (Lo-elast) and one with an uncovered interest rate parity shock (UIP). The standard international RBC model and DSGE models cannot replicate the volatility of the real exchange rate. We use those two methods to increase this volatility to see if it helps the model's performance in matching the behavior of imports and exports. Since the model setups and calibrations are very standard in the literature (for instance, see Backus et al., 1992), we leave them in an appendix available on the authors' websites. The parameters of utility are calibrated so that the intertemporal elasticity of substitution is 0.5, the elasticity of substitution between Home and Foreign goods is 0.9 in IRBC models (following Heathcote and Perri, 2002) and 1.5 in DSGE models, and the share of Home goods in the aggregate Home consumption good is 0.85. Panel A of Table 5 reports the standard deviations of aggregate variables relative to that of GDP. In the standard IRBC model (HP), imports and exports are even less volatile than GDP. The same discrepancy has also been reported in Table 2 of Heathcote and Perri (2002).16 That study ﬁnds that the assumption of ﬁnancial autarky can improve the volatility of imports and exports in a very limited way. The added features in the DSGE model and GHH models cannot solve this problem. Imports and exports are still far less volatile than what they are in the data. However, the GHH utility function does make the volatility of net exports much closer to the data. This follows because imports and exports are more volatile (due to more variable consumption in the GHH model), and imports and exports are less correlated than what they are in the DSGE model. Panel B shows the correlations of real imports, real exports, and real net exports with GDP, as well as the correlation between real imports and exports. Imports and exports are measured at their steady-state prices (constant price). The models of HP, DSGE and GHH match the data in that real imports and exports are pro-cyclical and positively correlated with each other. Net exports are counter-cyclical in these models. That is, the standard models can replicate the “positive comovement” feature, though they fail the “trade volatility”. Panel C reports the same statistics as Panel B, but imports, exports and net exports are measured in terms of ﬁnal consumption goods, instead of constant prices. The results are similar to those in Panel B. 17 Besides the volatility of imports and exports relative to GDP, another feature missing from the standard DSGE model is the high volatility of the real exchange rate. A natural question is whether we can increase the volatility of imports and exports in a model with more volatile real exchange rates. We follow Chari et al.'s (2002) “elasticity method” to increase real exchange rate volatility by 16

Zimmermann (1999) ﬁnds similar results in a sticky-price model. Raffo (2008) ﬁnds that real net exports measured with constant prices are procyclical under a standard utility function. We ﬁnd that this conclusion may be sensitive to the volatility of investment and the elasticity of substitution between Home and Foreign goods. 17

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

45

Table 5 Performance of standard models. Panel A: Standard deviations relative to that of real GDP Consumption †

Data HP‡ DSGE‡ GHH‡ Lo-elast.‡ UIP‡ CDL‡

Investment

0.798 0.462 0.545 0.613 0.401 0.925 0.521

Real import

2.890 2.663 2.830 2.697 2.767 2.875 2.441

3.335 0.727 0.826 0.935 1.651 3.477 2.196

Real export 2.626 0.608 0.835 0.947 1.625 3.466 2.048

RealNetExport RealGDP

Real ER♯

0.250 0.087 0.077 0.173 0.467 1.016 0.636

2.432 0.385 0.375 0.284 1.216 1.458 3.092

Panel B: Correlation with real GDP

†

Data HP‡ DSGE‡ GHH‡ Lo-elast.‡ UIP‡ CDL‡

Real import

Real export

RealNetExport RealGDP

0.827 0.929 0.801 0.894 − 0.647 0.286 0.997

0.415 0.588 0.663 0.278 0.973 0.069 − 0.980

− 0.467 − 0.551 − 0.214 − 0.497 0.852 − 0.112 − 0.991

corr(RIMt, REXt)♯ 0.194 0.628 0.809 0.252 − 0.799 − 0.894 − 0.992

Panel C: Correlation with real GDP

‡

HP DSGE‡ GHH‡ Lo-elast.‡ UIP‡ CDL‡

Real import

Real export

RealNetExport RealGDP

0.999 0.988 0.985 0.369 0.569 − 0.976

0.500 0.601 0.241 0.984 0.070 − 0.973

− 0.819 − 0.552 − 0.608 0.848 − 0.181 0.984

corr(RIMt, REXt)♯ 0.491 0.634 0.152 0.212 − 0.749 0.999

♯ Real ER is the (CPI-based) real exchange rate. corr(RIMt, REXt) is the correlation of real imports and exports. In Panels A and B, the imports, exports and net exports are measured in constant (steady-state) prices. They are measured in Panel C in terms of ﬁnal consumption goods. † US data as in Table 1. ‡ HP (Heathcote and Perri, 2002) is the standard IRBC model with incomplete ﬁnancial market (real bonds only). DSGE is the standard DSGE model as described in an appendix available on the authors' websites.. GHH is the DSGE model with GHH utility function. Lo-elast is the DSGE model with low intertemporal elasticity of substitution (σ = 5), UIP is the DSGE model with the uncovered interest rate parity shock, and CDL is the HP model with low elasticity of substitution between home and foreign goods following Corsetti et al. (2008a). RealNetExport ) and HP ﬁltered data. Entries are averages over 100 simulations of length 120. Statistics are based on logged (except for RealGDP

decreasing the value of the intertemporal elasticity of substitution. Corsetti et al. (2008c) examine this approach in an incomplete markets model in which a single bond is traded. They show that this approach can generate high exchange rate volatility without exogenous unreasonable monetary noise provided the model included two sectors (nontradables and tradables). Some authors have also used an uncovered interest rate parity (UIP) shock to generate exchange rate variations in DSGE models.18 Recently, Corsetti et al. (2008a) emphasize the wealth effect of productivity shocks in driving the real exchange rate. They show in a model with low elasticity of substitution between home and foreign goods, the real exchange rate can be very volatile (as 75% volatile as in the data). In addition, they ﬁnd in their model that a country's terms of trade and the real exchange rate appreciate when its productivity increases relative to the rest of the world, which is consistent with the US data.19 In the model CDL of Table 5, we set the elasticity of substitution between home and foreign goods at a low level (0.32) and generate similar results as in Corsetti et al. (2008a). In our simulation results, we ﬁnd that the volatilities of real imports, exports and the exchange rate all increase in those three models. Under certain calibrations of the UIP shock, the model can also replicate the pro-cyclical movement of imports and exports, though the correlation between exports and output is nearly zero. However, there is a striking departure of these

18 For instance, see Kollmann (2004) and Wang (2010). This approach is similar to Zimmermann's (1999), which adds an exogenous source of exchange rate volatility. 19 Another reason that the terms of trade may improve after an increase of productivity is because the productivity increase may come in the form of lower costs for new goods. See Corsetti et al. (2007) for details.

models from the data: real imports and exports are highly negatively correlated in those models. In standard models, Home and Foreign intermediate goods are used to produce ﬁnal goods. The ﬁnal goods are used for consumption and investment. There are two factors affecting the volatility of imports: 1. the volatility of demand for ﬁnal goods and, 2. the substitution between Home and Foreign goods. Under the standard calibration, the majority (about 75%) of ﬁnal goods (and therefore imports) goes to consumption. Consumption is less volatile than GDP in the data. So if we want to match the volatility of consumption, demand for ﬁnal goods will not be very volatile. Given the low volatility of demand for ﬁnal goods, we can still have very volatile imports and exports if there is a lot of substitution between home and foreign goods. The model with low intertemporal elasticity of substitution and the UIP model increase the volatility of the terms of trade and therefore the volatility of imports and exports. Exchange rate movements induce ﬂuctuations in the relative price of imports and exports. In return, the substitution between Home and Foreign goods increases the volatility of imports and exports. But when the terms of trade changes, imports and exports move in opposite directions. So this method produces a negative correlation between imports and exports, which is contradictory to the data. Baxter and Stockman (1989) ﬁnd little evidence of systematic difference in the volatilities of real imports and exports when countries switch from ﬁxed to ﬂexible exchange rate regimes, though the real exchange rates became substantially more variable during this period. This ﬁnding also suggests that the high volatility of international trade ﬂows is unlikely to come from the exchange rate ﬂuctuations.

46

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

Table 6 Simulation results of benchmark model. Panel A: Standard deviations relative to that of real GDP

a

Data Benchmarkb High spillover Medium spillover High correlation High correlation 2 No correlation High persistence Technology costs Traded nondurable Low durable share

C

I

0.798 0.878 0.948 0.917 0.920 0.874 0.902 0.922 0.961 0.748 0.748

2.890 2.594 2.905 2.894 2.750 2.666 2.757 2.840 2.828 2.950 2.612

DC 2.983 2.473 2.738 2.754 2.680 2.381 2.658 2.473 2.551 2.892 2.628

L

RIM

REX

RNX

0.670 0.547 0.539 0.549 0.549 0.544 0.549 0.539 0.535 0.571 0.569

3.335 2.633 1.826 2.652 2.880 2.558 2.619 2.423 2.726 2.048 0.960

2.626 2.678 1.775 2.615 2.936 2.596 2.678 2.411 2.779 2.082 0.933

0.250 0.337 0.322 0.393 0.402 0.266 0.596 0.580 0.355 0.302 0.240

Q 2.432 1.262 1.297 1.271 1.058 0.435 1.470 1.282 1.041 1.113 0.803

Panel B: Correlation with GDP

a

Data Benchmarkb High spillover Medium spillover High correlation High correlation 2 No correlation High persistence Technology costs Traded nondurable Low durable share

RIM

REX

RNX

corr(RIM, REX)

Elasticityc

σY, Y

σC, C

0.827 0.606 0.576 0.599 0.630 0.801 0.564 0.618 0.560 0.714 0.828

0.415 0.411 0.405 0.324 0.337 0.554 0.375 0.333 0.232 0.388 0.220

− 0.467 − 0.187 − 0.129 − 0.228 − 0.288 −0.177 − 0.135 − 0.180 −0.292 − 0.331 −0.374

0.194 0.421 0.160 0.171 0.265 0.577 0.215 0.097 0.386 0.550 0.228

0.90 1.05 0.69 0.89 1.19 1.89 1.07 0.95 1.41 0.69 0.70

0.68 0.01 − 0.03 − 0.01 0.03 0.56 − 0.02 0.16 0.08 0.002 −0.04

0.60 − 0.17 0.23 −0.14 − 0.20 0.39 − 0.23 0.03 −0.09 − 0.08 − 0.08

(0.12) (0.20) (0.13) (0.19) (0.19) (0.27) (0.26) (0.19) (0.13) (0.11) (0.13)

Note: RealNetExport , Q—CPI-based real exchange C—consumption, I—investment, DC—durable consumption, L—labor, RIM—real imports, REX—real exports, RNX—real net exports deﬁned as RealGDP rate. corr(RIM, REX)—correlation of real imports and exports, σY, Y —cross-country correlation of output, σC, C —cross-country correlation of consumption. The cross-country correlations are between the United States and the rest of OECD countries (Corsetti et al., 2008a). a Data as in Table 1. b The standard deviation of GDP in benchmark model is 2.26%. All variables are logged (except for RNX) and HP ﬁltered with a smoothing parameter of 1600. Entries are averages over 100 simulations of length 120. c This column reports the estimates of the short-run elasticity of substitution between home and foreign goods. The estimate in the ﬁrst row is from Heathcote and Perri (2002). Other values are estimated with data simulated from our models. Entries are averages over 100 simulations. Standard errors are in parentheses.

5.1.2. Models with trade in durable goods In this subsection, we present simulation results of our model with trade in durable goods. Table 6 shows simulation results for nine models. All of these models can match the data fairly well. 5.1.2.1. Benchmark model. As in standard IRBC models, our benchmark model can replicate the volatility (relative to that of GDP) of aggregate variables such as consumption, investment, durable expenditure, and labor. In addition, our model matches the trade sector data much better than the standard models: imports and exports are about three times as volatile as GDP and both of them are pro-cyclical and positively correlated with each other. Our model can also match the well-known ﬁnding that net exports are counter-cyclical. When productivity shocks are persistent, it is well understood that investment will be volatile. Agents wish to change the capital stock quickly to take advantage of current and anticipated productivity shocks. This effect contributes to the high volatility our model produces for imports and exports, because capital goods are traded. A positive productivity shock leads to a desire to increase Home's stock of domestically produced and foreign-produced capital. This leads to the increase in demand for imports when there is a positive productivity shock. A positive productivity shock also increases the supply of Home's export good, lowering its world price, and thus increasing exports. These effects are standard in RBC models, and explain why the models can generate pro-cyclical imports and exports. However, if only investment goods are durable, and consumption goods are nondurable, the model does not produce sufﬁcient volatility in

imports and exports.20 When we introduce a consumer durable sector, there is an additional source of volatility. Demand for consumer durables, like demand for investment goods, is forwardlooking, but it is not expected productivity per se, but rather higher wealth from higher expected future income that leads to volatility in demand for durable consumer goods. Fig. 2 shows impulse response functions of our benchmark model for a one-standard-deviation shock in the home durable goods sector. In the face of a positive shock in the home durable goods sector, the price of durable goods (relative to nondurable goods) decreases, which leads to substitution from nondurables to durables. Because the shock is persistent, there is also a signiﬁcant wealth effect that pushes up demand for both home and foreign durable consumption goods. As a result, purchases of durable goods (both home- and foreignproduced durable consumption goods) in the home country increase while consumption of nondurable goods declines. Aggregate consumption in the home country rises because the increase of durable purchases exceeds the decline of nondurable consumption. The price of foreign-produced durables relative to home-produced durables also increases: the terms of trade (import prices divided by export 20 For instance, if we change the depreciation rate of durable consumption (δD) in our benchmark model to one and the adjustment cost to zero, the (relative) standard deviation of imports and exports decreases from 2.6 to 2. In this exercise, the capital adjustment cost is changed such that the (relative) standard deviation of investment is the same as in our benchmark model (2.6). We also change the elasticity of substitution between the home and foreign consumption goods to 0.9, which reﬂects the fact that the short-run elasticity of substitution is low in our benchmark model due to adjustment costs.

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

47

Fig. 2. Impulse response functions. Note: The impulse response functions are with respect to a one-standard-deviation shock in the durable goods sector of home country for the benchmark model. The vertical axis shows percentage deviation from the steady state and the horizontal axis shows the number of periods after the shock. DC is the abbreviation for Durable Consumption. XX Expenditure on YY DC means country XX's expenditure on durable consumption goods that are produced in country YY. ND is the abbreviation for Nondurable.

prices) worsen in Fig. 2. The change in the terms of trade leads to substitution toward home-produced durables. As a result, the foreign country consumes more home durable goods, though foreign aggregate consumption declines. Similarly, the home country also shifts from foreign-produced durable goods toward home-produced durable goods because of this substitution effect. But under the wealth effect, domestic consumers raise their demand for both domestic and foreign tradable (durable) goods, even if the latter are more expensive in terms of the former. Overall, the wealth effect and the effect of the decline in the price of durables relative to nondurables lead to an increase in import demand, despite the increase in the price of foreign durables relative to home durables. Indeed, total expenditure on imports increases more than the value of exports, leading to a decline in the trade balance. However, part of that increase in import expenditure comes from the increased price of imports. But overall, the model still generates pro-cyclical movements of import and export quantities. The wealth effect is an important channel for our model to generate pro-cyclical demand for foreign durable consumption goods. Following

King (1990), we ﬁrst calculate the welfare gain to home households of a positive shock in the home durable good sector. We then examine the consumption behavior of these households assuming they receive instead an initial wealth shock (in the form of an endowment of foreign bonds) that increases welfare in the same amount, holding all prices (including wages) constant at their steady-state levels. The size of the shock is calibrated to match the welfare increase after a one-standarddeviation positive shock in home durable goods sector. Fig. 3 shows the impulse response functions for various variables from this shock in our general equilibrium model and from the wealth shock to home households. Due to the wealth effect, home country's consumption of nondurable and durable goods increases while labor supply declines. The wealth effect accounts for about 25% of the increase in homeproduced durable consumption goods and about 50% of the increase in foreign-produced durable consumption goods. We also consider a case in which there is no cost for adjusting durable consumption. In this case, the wealth effect accounts for a smaller (about 30%) share of the increase in home country's consumption of foreign-produced durable consumption goods.

48

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

Fig. 3. Impulses response functions of home country variables. Note: The impulse response functions of overall effect are with respect to a one-standard-deviation shock in the durable goods sector of home country in the benchmark model. The impulse response functions of wealth effect are with respect to a positive shock to the initial bond holding of the home country. All prices are ﬁxed at their steady-state levels and the size of the initial bond holding shock is calibrated to match the welfare increase after a one-standard-deviation shock in the durable goods sector of home country in the benchmark model. The vertical axis shows percentage deviation from the steady state and the horizontal axis shows the number of periods after the shock. All impulse response functions are for variables in the home country. Home Durable Consumption is home country's expenditure of home-produced durable consumption goods. Foreign Durable Consumption is home country's expenditure of foreign-produced durable consumption goods.

Baxter and Crucini (1995) and Corsetti et al. (2008a) also discuss the impact of the wealth effect on the international transmission of shocks. They ﬁnd that the scope for self-insurance via international trade may be limited when productivity shocks are close to having a unit root. In particular, Corsetti et al. (2008a) show that the wealth effect can be strong enough to generate an appreciation of the terms of trade after a positive shock if the productivity shock is very persistent and the Home and Foreign goods are highly substitutable. The longrun elasticity of substitution is high in our model. However, the adjustment costs of investment and durable consumption effectively reduce the short-run elasticity of substitution in our model. In addition, our productivity shocks are less persistent than those in Baxter and Crucini (1995) and Corsetti et al. (2008a). As a result, our model performs similarly to other standard IRBC models and predicts a depreciation of the terms of trade after a positive productivity shock. The unconditional correlation of nondurable consumption and GDP is positive in our model though nondurable consumption is negatively correlated with GDP conditional on a shock in the durable goods sector. In our benchmark model, the correlation between nondurable consumption and GDP is 0.85 and it is 0.82 in the US data.21 This is because the shocks in the nondurable goods sector play an important role in driving the volatility of GDP and nondurable consumption and GDP is positively correlated in the case of nondurable goods sector shocks. Durable expenditure is also positively correlated with GDP in our model. The unconditional correlation is 0.44 in the simulated data and it is 0.77 in the US data. Our benchmark model can also match trade dynamics fairly well. Fig. 4 shows the correlations between GDP and real imports, exports and net exports at various leads/lags for the US data and the data simulated from our benchmark model. As noted by Ghironi and Melitz (2007), the correlation between GDP and imports exhibits a tent-shaped pattern, while the correlations of exports and net exports with GDP are S-shaped.22

Our model captures these qualitative patterns well. Note in particular that the model captures the fact that, while current imports are positively correlated with GDP, imports are negatively correlated with lagged GDP at longer horizons. However, our model's correlation of both imports and exports with lagged GDP declines quickly–too quickly–as the horizon increases. It appears especially that exports increase with a lagged response to a positive shock to GDP. It might be possible to capture this dynamic behavior by incorporating a lag between orders of durable goods and delivery. Table 6 also reports the short-run elasticity of substitution between Home and Foreign goods that is estimated from our model. The estimated short-run elasticity of substitution in our benchmark model is 1.05 with a standard error of 0.20. The short-run elasticity of substitution implied by our model is very close to what is found in the empirical studies with business-cycle-frequency data, such as Bergin (2006) and Heathcote and Perri (2002), even though our model builds in a high long-run elasticity of substitution. The lower short-run elasticity arises because of the cost of adjusting the stocks of durable consumption and capital stocks. In a model with trade in durables, the lower short-run trade elasticity arises naturally and accords with the standard practice in macroeconomic modeling of the gradual accumulation of capital. That is, the trade elasticity puzzle is easy to understand in a context in which trade is in durables which are accumulated slowly over time. 5.1.2.2. Alternative speciﬁcations. The performance of our model is robust under alternative calibrations of productivity shocks in Table 6. Models High Spillover and Medium Spillover in Table 6 release the constraint of no spillovers in the benchmark model. The process of shocks is modiﬁed to a VAR(1) format N

N

N

N

ð22Þ

D

D

D

D

ð23Þ

N

N

N

N

ð24Þ

D

D

D

D

ð25Þ

AHt +1 = Ξ1 AHt + Ξ3 AFt + εHt +1 AHt +1 = Ξ2 AHt + Ξ3 AFt + εHt +1

21

The quarterly US data is from the Bureau of Economic Analysis during 1973Q1 and 2007Q2. Both the US data and the simulated data are logged and HP ﬁltered before calculating the correlations. 22 Ghironi and Melitz (2007) have also implicitly observed the procylicality of imports and exports in their Fig. 1.

AFt +1 = Ξ1 AFt + Ξ3 AHt + εFt +1 AFt +1 = Ξ2 AFt + Ξ3 AHt + εFt +1 ;

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

Fig. 4. Cross-correlation in different lags. Note: The data are quarterly US data from OECD Economic Outlook dataset during 1973Q1–2006Q3. The model is our benchmark model. Model Statistics are averages over 100 simulations of length 120. All variables are logged (except for Net Exports/ GDP) and HP ﬁltered with a smoothing parameter of 1600.

where coefﬁcient Ξ3 is the cross-country spillover of productivity shocks.23 BKK ﬁnd a relatively large spillover of 0.088. In the model of High Spillover, Ξ3 is set to this value. It is set to a medium level of 0.044 in the model of Medium Spillover. In the model High Correlation, the cross-country correlation of productivity shocks in the durable goods sector (ρD) is set to a relatively high level, 0.468, which is found in Corsetti et al. (2008a). Cross-country output correlation is nearly zero in the standard IRBC models,24 though this correlation is usually large in the data. Our model provides little insight on this issue. We ﬁnd that the crosscountry correlation of output increases if we allow innovations to be highly correlated across countries. Our motivation here is that Solow residuals measure productivity shocks with considerable error. To the extent that the measurement errors are uncorrelated across countries, the measured cross-country correlation of productivity may be severely downward biased. In the model High Correlation 2, we set cross-country correlation of innovations to 0.7 for both durable goods and nondurable goods sectors, the cross-country output correlation increases from 0.01 to 0.56, which is very close to the data. Indeed, most aspects of the model 23 24

As in Erceg and Levin (2006), we assume the cross-sector spillover is still zero. For instance, the cross-country output correlation in Backus et al. (1992) is − 0.18.

49

hold up well under this alternative speciﬁcation, with the notable exception of real exchange rate volatility which is much lower than in other parameterizations.25 In particular, the volatility of imports and exports and their correlation with GDP do not decline. We emphasize that we have not chosen the correlation of productivity shocks under this parameterization to match data. Further examination of the measurement of productivity shocks is warranted, but we leave this to future work. Model High Persistence tries a different approach to improve the output correlation. We follow Baxter and Crucini (1995) to have highly persistent shocks. The AR(1) coefﬁcients of shocks in durable and nondurable goods sectors are set to 0.95 in the model of High Persistence. Output and consumption under this setup become more correlated across countries than in our benchmark model, as shown in Baxter and Crucini (1995). The statistics on the cross-country correlations of output and consumption can be improved further as we increase the persistence of the shocks. However, such a parameterization leads to small import and export volatilities and negative correlation of imports and exports in our model. Model Technology Costs in Table 6 considers a different model for the adjustment costs of capital and durable consumption stocks. Under this setup, it is costly to adjust the proportion of home to foreign capital (durable consumption). This is a crude way of capturing the notion that it takes time to change technologies. As a result, the proportion of home to foreign durables only changes gradually. Similar trade adjustment costs have been introduced in some recent papers. For instance, Bodenstein et al. (2007) allow for convex costs for adjusting the share of oil used in consumption and production. Erceg et al. (2009) incorporates adjustment costs that penalize rapid changes in bilateral trade shares. Such adjustment costs will dampen the short-run substitution effect between the Home and Foreign goods and induce a wedge between the short- and long-run trade elasticity. Ramanarayanan (2007) models this idea more explicitly with a putty-clay technology. In addition, we use adjustment costs for the total stock of capital (durable consumption) to match the volatility of investment in capital (durable consumption). This new setup generates results similar to our benchmark model. Cross-country output correlation in the Technology Costs model (0.08) comes slightly closer to the data than in the benchmark model (0.01). However we also note that we have more freedom in calibrating the “Technology Costs” model because it has more cost parameters than our benchmark model. In the model of Traded Nondurable (Table 6), we allow Home and Foreign countries to trade part of their nondurable consumption goods.26 This model has the same production function for the nondurable goods sector as our benchmark model. But unlike our benchmark model, a fraction of nondurable goods can be traded across countries. Home and Foreign traded nondurable consumption are aggregated into a traded nondurable consumption composite. This composite and the nontraded nondurable consumption are aggregated into nondurable consumption. The rest of the model follows the same setup of our benchmark model. We calibrate this model such that: 1. nondurable goods account for 30% of trade; 2. share of capital goods in trade is 43%; 3. trade accounts for 14% of output. The model generates results similar to our benchmark model. The only noticeable difference is that imports and exports are less volatile in the model of Traded Nondurable. This result is not surprising because nondurable consumption is less volatile than durable expenditure and investment in our model. Diverting some trade to nondurable consumption decreases the overall volatility of the trade. Even in this case, imports and exports are still more than two times as 25 Our model's ability to match the trade sector data also holds up well in the case with zero cross-country correlation of productivity shocks (Model No Correlation in Table 6). We thank an anonymous referee for suggesting us this exercise. 26 See the appendix on the authors' websites for details.

50

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

volatile as output. In this model, we have assumed that productivity shocks are the same for traded and nontraded nondurable goods. In the US data reported in Section 2.2, nondurable imports and exports are more volatile than output. So the productivity shocks for traded nondurable goods may be more volatile than nontraded nondurable goods in the data. If we allow this difference in our model, the volatility of imports and exports in our model may come closer to the data. Model Low Durable Share illustrates the importance of trade in durable goods for our model to match the trade data. An important difference between our model and the standard IRBC models is that our model has a larger share of durable goods in international trade. To highlight the importance of this structure, we modify our model such that it has the same share of durable goods in international trade as in the HP model (25.5%). The simulation results are reported in the model of Low Durable Share (Table 6). As in the HP model, both imports and exports become less volatile than output in this model. In all of our calibrations, we note the following shortcomings: as in almost all RBC models, real exchange rate volatility is still lower than in the data. However, our model does quite well relative to the literature. The standard deviation of the real exchange rate in our benchmark model is roughly 50% of the standard deviation in the data.27 Our model also generates stronger correlation between the real exchange rate and macroeconomic variables such as consumption than the data.28 Adding features such as pricing to market, distribution costs and sticky prices into our model may help to match the disconnect between the exchange rate and macro economic variables in the data. Across all speciﬁcations, our model produces somewhat lower correlations of real imports with GDP than what appears in the data. And, perhaps as a consequence, net exports are not as negatively correlated with GDP as in the data.

5.2. Backus–Smith puzzle When agents can trade a complete set of contingent claims, but face potentially different goods prices, in a variety of contexts models imply that relative cross-country consumption should be perfectly positively correlated with the real exchange rate. But, beginning with Backus and Smith (1993), several studies ﬁnd empirically that the correlation between relative consumption and the real exchange rate is generally low, even negative in many countries. Some recent papers offer models to explain this correlation when capital markets are not perfect, and only bonds are traded, for instance, Corsetti et al. (2008a), and Benigno and Thoenissen (2008). Our model shares some features of these models, but also offers some new insight: the introduction of consumer durables raises interesting issues about how consumption should be measured in tests of the Backus–Smith puzzle. The dynamics of consumption and the real exchange rate in response to a shock to productivity in the durable sector looks very much like those in Benigno and Thoenissen (2008). As shown in Fig. 2, a positive shock lowers the price of the durable export (the terms of trade deteriorate), and because of home bias, that tends to work toward a real CPI depreciation. But that effect can be more than offset by the increase in the relative price of nondurable goods, which are not traded across countries. There are two forces working to push up the price of nontraded goods: ﬁrst, there is the traditional Balassa– Samuelson effect. The increase in productivity pushes up the real wage, thus pushing up the relative price of nontradables. In addition, overall consumption in the home country increases from a wealth 27 Unlike in models with low intertemporal elasticity of substitution and exogenous exchange rate shocks (models Lo-elast. and UIP in Table 5), exchange rate volatility in our model is mainly from the ﬂuctuations of nontradable prices. 28 The correlation between the real exchange rate and nondurable consumption (durable expenditure) is 0.28 (− 0.64) in our model. The corresponding correlation in the US data (1973Q1–2007Q2) is 0.10 (− 0.01).

effect, because higher productivity increases lifetime income for the home country. Even if there were no factors mobile between sectors, that would tend to push up the price of the nontradable goods, and help foster a real appreciation. We have that aggregate consumption is increasing, and under our calibrations, a real appreciation. The model's ability to explain the positive correlation between the real exchange rate and relative consumption hinges critically on the variability of nontradable prices, but empirical work has found a negligible role of nontradable prices movements in driving observed ﬂuctuations in the real exchange rates.29 In addition, the terms of trade and the real exchange rate are negatively correlated in our model though the opposite is true in the data. In response to a positive productivity shock, the terms of trade depreciates and therefore is negatively correlated with relative consumption in our model, which is at odds with the data. Obstfeld and Rogoff (2000) emphasize that the terms of trade and the real exchange rates are generally positively correlated in the data. Corsetti et al. (2008b) and Enders et al. (2008) also ﬁnd in their empirically based VAR exercises that both the terms of trade and the real exchange rates appreciate after a positive productivity shock. Our model fails to replicate these ﬁndings and we leave this for our future research. However, our model also offers some new insight into measurement issues that may ultimately shed light on this puzzle. Durable consumption measured in national accounts data is expenditures on new durable consumption goods. However, it is the service ﬂow from the stock of durable consumption that enters the utility function. As emphasized by Obstfeld and Rogoff (1996, page 98), the consumer smoothes the service ﬂow from the stock of durable consumption, instead of the path of expenditures on durables. We measure total consumption as in the data by the sum of nondurable consumption and the investment in durable consumption TCHt = CHt + DCHt ;

ð26Þ

where TCHt is the total consumption in Home country. CHt is nondurable consumption and DCHt is durable consumption expenditure. Durable expenditure is deﬁned as the sum of Home- and Foreign-good durable expenditure DH DF H H F F DCHt = Pˆ Ht dHt + ΔHt + Pˆ Ht dHt + ΔHt :

ð27Þ

From the utility function, we know the “true” consumption (utility consumption) is : UCHt = μ DHt + ð1−μ Þ CHt 1 ζ

ζ−1 ζ

1 ζ

ζ−1 ζ

ζ ζ−1

ð28Þ

We calculate the correlation of 1. the (log) CPI-based real exchange rate and the (log) relative total consumption (log(TCHt) − log(TCFt)), and 2. the utility-based real exchange rate and the (log) relative utility consumption (log(UCHt) − log(UCFt)).30 The correlation between the real exchange rate and total consumption differential (log(TCHt) − log(TCHt)) is −0.23. So our model can replicate the negative correlation between the real exchange rate and relative consumption documented in the data. However, the correlation between the utility-based real exchange rate and the utility consumption differential (log(UCHt) − log (UCHt)) is 0.26. Based on the fact that the ﬁnancial market is limited to trade in non-state-contingent real bonds and leisure is nonseparable in the utility function, a correlation of 0.26 still implies a relatively good amount of risk-sharing between the Home and Foreign countries. 29

See Engel (1999) and Chari et al. (2002). Please see an appendix posted on the authors' websites for details about how to calculate the utility-based real exchange rate. 30

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

6. Conclusion The behavior of imports and exports is, of course, a key component of the linkages among economies. We document two important empirical regularities for these two variables that have been largely neglected in the literature: both imports and exports are much more volatile than GDP and both are pro-cyclical. Our model confronts and, to a degree, successfully explains these empirical regularities. By modeling trade in durables, we can understand the high volatility of imports and exports relative to output. Trade volatility has stimulated much discussion recently after the collapse of global trade in 2008 and 2009. Levchenko et al. (2009) ﬁnd empirical evidence that the mechanism emphasized in our paper – trade in durable goods – has played an important role in the recent collapse of US trade. Trade in durables also offers a natural explanation for the trade elasticity puzzle — that the response of imports to changes in the terms of trade is low at business cycle frequencies, but is high when considering the long-run effect of permanent price changes. Our model performs well compared to other models, because it offers an explanation that is also consistent with the observation that imports and exports are both pro-cyclical, and positively correlated with each other, even when the terms of trade and real exchange rate are as volatile as in the data. We believe that the forward-looking nature of investment decisions and decisions to purchase consumer durables are a key feature of trade behavior. Our model noticeably fails to account for the high correlation of output across countries, which is a failure shared by essentially all rational expectation equilibrium models. However, we think that modeling trade as durables may still be a promising avenue for dealing with this puzzle as well, through channels that are not explored in this paper. One possibility is that while the common (across countries) component of productivity shocks may account for a small share of the variance of productivity, it may be that agents typically receive strong signals about the future common component. If news helps to drive business cycles (as in Beaudry and Portier, 2007), then perhaps news about the common component of productivity shocks helps contribute to the high correlation of business cycles across countries. News about future productivity is especially important for durables, so the impact of news may be especially strong on the investment and consumer durables sectors. Another avenue that may deserve further exploration is a model with nominal price stickiness, as in DSGE models. Our model of durable trade creates large swings in demand for imports, which indeed is what allows it to account for trade volatility. But an increase in Home demand for Foreign output has only a small effect on Foreign's output level. Instead, in our model, prices adjust so that more of Foreign's output is channeled toward Home. In a model with sticky prices, changes in demand may lead to changes in aggregate output, and so create a channel for international spillovers. While these channels do exist in current DSGE models, they are not strong because the models do not account for large pro-cyclical movements in imports and exports. It is an empirical fact that a large fraction of trade is in durables. Indeed, we view explaining this phenomenon–rather than assuming it, as we do in this study–to be another interesting topic for future research. What we have accomplished here is to demonstrate that trade in durables signiﬁcantly alters the behavior of imports and exports in an RBC model in a way that can account for some striking empirical facts. Acknowledgments We thank David Backus, Paul Bergin, Martin Boileau, Lukasz Drozd, Jonathan Heathcote, Enrique Martinez-Garcia, Mark Wynne, Kei-Mu Yi and seminar participants at Boston College, Chicago GSB, Dartmouth College, Tufts University, the Federal Reserve Bank of Dallas, NBER IFM Meeting, 2008 Midwest Macro Meetings, and HEC/CIRPEE Real Business Cycle Conference for helpful comments.

51

We also would like to thank Co-Editor Giancarlo Corsetti and two anonymous referees for constructive comments. All views are those of the authors and do not necessarily reﬂect the views of the Federal Reserve Bank of Dallas or the Federal Reserve System. Engel acknowledges support for this research from a grant from the National Science Foundation. Appendix A. Dividing imports and exports into categories of durable and nondurable goods SITC categories 0 (FOOD AND LIVE ANIMALS), 1 (BEVERAGES AND TOBACCO), and 4 (ANIMAL AND VEGETABLE OILS, FATS AND WAXES) are nondurable goods. Category 7 (MACHINERY AND TRANSPORT EQUIPMENT) belongs to durable goods. Category 2 is raw materials that exclude fuels such as petroleum. Category 3 contains energy products such as coal, petroleum, gas, etc. The remaining categories are more difﬁcult to classify. This is particularly true for category 5 (CHEMICALS AND RELATED PRODUCTS, N.E.S.). Even if we go down to the 3-digit level, it is still unclear which categories belong to durable goods. We ﬁnd that this category includes many nondurable goods, such as fertilizers, medicines, cleaning products, etc. To avoid exaggerating the share of durable goods, we put the whole category 5 into nondurable goods. But we note that this category does include some durable goods, such as plastic tubes, pipes, etc. For categories 6, 8 and 9, we go down to the SITC 2-digit levels for more information about the durability of goods. Category 6 (MANUFACTURED GOODS CLASSIFIED CHIEFLY BY MATERIALS) classiﬁes goods according to their materials. We assume that goods produced from leather, rubber, or metals are durables (61–62 and 66–69). Goods produced from wood (other than furniture), paper, or textile (63–65) are nondurables. Category 8 includes other manufactured products that are not listed in categories 6 and 7. We assume that construction goods (81), furniture (82), professional instruments (87), photographic equipments (88) are durable goods. Travel goods (83), clothing (84), footwear (85) and remaining goods (89) are classiﬁed as nondurables. Category 9 includes products that are not classiﬁed elsewhere. In this category, we assume that coins and gold (95–97) are durables. All remaining products are classiﬁed as nondurables. References Backus, D., Smith, G., 1993. Consumption and real exchange rates in dynamic economies with non-traded goods. Journal of International Economics 35, 297–316. Backus, D., Kehoe, P., Kydland, F., 1992. International real business cycles. Journal of Political Economy 100, 745–775. Backus, D., Kehoe, P., Kydland, F., 1995. International business cycles: theory and evidence. In: Cooley, T. (Ed.), Frontiers of Business Cycle Research. Princeton University Press. Baxter, M., 1992. Fiscal policy, specialization, and trade in the two-sector model: the return of Ricardo? Journal of Political Economy 100 (4), 713–744. Baxter, M., 1995. International trade and business cycles. In: Grossman, G., Rogoff, K. (Eds.), Handbook of International Economics, 3. Baxter, M., 1996. Are consumer durables important for business cycles? The Review of Economics and Statistics 78 (1), 147–155. Baxter, M., Crucini, M., 1995. Business cycles and the asset structure of foreign trade. International Economic Review 36, 821–854. Baxter, M., Stockman, A., 1989. Business cycles and the exchange-rate regime: some international evidence. Journal of Monetary Economics 23, 377–400. Beaudry, P., Portier, F., 2007. When can changes in expectations cause business cycle ﬂuctuations in neo-classical settings? Journal of Economic Theory 135 (1), 458–477. Benigno, G., Thoenissen, C., 2008. Consumption and real exchange rates with incomplete markets and non-traded goods. Journal of International Money and Finance 27 (6), 926–948. Bergin, P., 2006. How well can the new open economy macroeconomics explain the exchange rate and current account? Journal of International Money and Finance 25 (5), 675–701. Bernanke, B., 1985. Adjustment costs, durables, and aggregate consumption. Journal of Monetary Economics 15, 41–58. Bertola, G., Guiso, L., Pistaferri, L., 2005. Uncertainty and consumer durables adjustment. Review of Economic Studies 72 (4), 973–1007. Blonigen, B., Wilson, W., 1999. Explaining Armington: what determines substitutability between home and foreign goods? Candian Journal of Economics 32 (1), 1–21.

52

C. Engel, J. Wang / Journal of International Economics 83 (2011) 37–52

Bodenstein, M., Erceg, C., Guerrieri, L., 2007. Oil Shocks and External Adjustment. International Finance Discussion Papers, Number 897. Boileau, M., 1999. Trade in capital goods and the volatility of net exports and the terms of trade. Journal of International Economics 48 (2), 347–365. Chari, V., Kehoe, P., McGrattan, E., 2002. Can sticky price models generate volatile and persistent real exchange rates? Review of Economic Studies 69 (3), 533–563. Corsetti, G., Martin, P., Pesenti, P., 2007. Productivity, terms of trade and the “home market effect”. Journal of International Economics 73 (1), 99–127. Corsetti, G., Dedola, L., Leduc, S., 2008a. International risk sharing and the transmission of productivity shock. Review of Economic Studies 75, 443–473. Corsetti, G., Dedola, L., Leduc, S., 2008b. The International Dimension of Productivity and Demand Shocks in the US Economy. CEPR Discussion Paper 7003. Corsetti, G., Dedola, L., Leduc, S., 2008c. High exchange rate volatility and low passthrough. Journal of Monetary Economics 55 (6), 1113–1128. Drozd, L., Nosal, J., 2007. Understanding International Prices: Customers as Capital. University of Minnesota working paper. Eaton, J., Kortum, S., 2001. Trade in capital goods. European Economic Review 45 (7), 1195–1235. Enders, Z., Mueller, G., Scholl, A., 2008. How do Fiscal and Technology Shocks affect Real Exchange Rates? New Evidence for the United States, Center for Financial Studies Working Paper 2008/22. Engel, C., 1999. Accounting for U.S. real exchange rate changes. Journal of Political Economics 107, 507–538. Erceg, C., Levin, A., 2006. Optimal Monetary policy with durable consumption goods. Journal of Monetary Economics 53, 1341–1359. Erceg, C., Guerrieri, L., Gust, C., 2008. Trade adjustment and the composition of trade. Journal of Economic Dynamics and Control 32 (8), 2622–2650. Erceg, C., Guerrieri, L., Kamin, S., 2009. Did Easy Money in the Dollar Bloc Fuel the Global Commodity Boom? International Finance Discussion Papers, Number 979. Feenstra, R., Levinsohn, J., 1995. Estimating markups and market conduct with multidimensional product attributes. Review of Economics Studies 62, 19–52. Gali, J., 1993. Variability of durable and nondurable consumption: evidence for six O.E.C. D. countries. The Review of Economics and Statistics 75 (3), 418–428. Ghironi, F., Melitz, M., 2007. Trade ﬂow dynamics with heterogeneous ﬁrms. American Economic Review Papers and Proceedings 97 (2), 356–361. Greenwood, J., Hercowitz, Z., Huffman, G., 1988. Investment, capital utilization and real business cycle. The American Economic Review 78, 402–417. Head, K., Ries, J., 2001. Increasing returns versus national product differentiation as an explanation for the pattern of U.S.–Canada trade. The American Economic Review 91 (4), 858–876. Heathcote, J., Perri, F., 2002. Financial autarky and international business cycles. Journal of Monetary Economics 49, 601–627.

King, R., 1990. Value and capital in the equilibrium business cycle program. In: McKenzie, L., Zamagni, S. (Eds.), Value and Capital: Fifty Years Later. MacMillan, London. Kollmann, R., 2004. Welfare effects of a monetary union: the role of trade openness. Journal of the European Economic Association 2, 289–301. Lai, H., Treﬂer, D., 2002. The Gains From Trade with Monopolistic Competition: Speciﬁcation, Estimation, and Mis-Speciﬁcation. NBER Working Paper No.9169. . Levchenko, A., Lewis, L., Tesar, L., 2009. The Collapse of International Trade During the 2008–2009 Crisis. Search of the Smoking Gun, Research Seminar in International Economics Discussion Paper No. 592. Mankiw, G., 1982. Hall's consumption hypothesis and durable goods. Journal of Monetary Economics 10 (3), 417–425. Obstfeld, M., Rogoff, K., 1996. Foundations of International Macroeconomics. MIT Press, Cambridge, MA. September 1996. Obstfeld, M., Rogoff, K., 2000. New direction for stochastic open economy models. Journal of International Economics 50, 117–153. Raffo, A., 2008. Net exports, consumption volatility and international real business cycle models. Journal of International Economics 75 (1), 14–29. Ramanarayanan, A., 2007. International Trade Dynamics with Intermediate Inputs. University of Minnesota working paper. Reinert, K., Roland-Holst, D., 1992. Armington elasticities for United States manufacturing sectors. Journal of Policy Modeling 14 (5), 631–639. Ruhl, K., 2005. Solving the Elasticity Puzzle in International Economics. University of Texas-Austin working paper. Sadka, J., Yi, K., 1996. Consumer durables, permanent terms of trade shocks, and the recent US trade deﬁcit. Journal of International Money and Finance 15 (5), 797–811. Schmitt-Grohé, S., Uribe, M., 2003. Closing small open economy models. Journal of International Economics 61, 163–185. Startz, R., 1989. The stochastic behavior of durable and non-durable consumption. Review of Economics and Statistics LXXI 2, 356–363. Wang, J., 2010. Home bias, exchange rate disconnect, and optimal exchange rate policy. Journal of International Money and Finance 29, 55–78. Warner, A., 1994. Does world investment demand determine U.S. exports? The American Economic Review 84 (5), 1409–1422. Whelan, K., 2003. A two-sector approach to modeling U.S. NIPA data. Journal of Money, Credit and Banking 35 (4), 627–656. Yi, K.-M., 2003. Can vertical specialization explain the growth of world trade? Journal of Political Economy 111 (1), 52–102. Zimmermann, C., 1999. International business cycles and exchange rates. Review of International Economics 7 (4), 682–698.

Journal of International Economics 83 (2011) 53–69

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

How do ﬁscal and technology shocks affect real exchange rates? New evidence for the United States Zeno Enders a, Gernot J. Müller a,b,⁎, Almuth Scholl c a b c

University of Bonn, Department of Economics, Adenauerallee 24-42, 53113 Bonn, Germany CEPR, United Kingdom University of Konstanz, Department of Economics, P.O. Box D 132, 78457 Konstanz, Germany

a r t i c l e

i n f o

Article history: Received 9 May 2010 Received in revised form 21 August 2010 Accepted 24 August 2010 Available online 22 September 2010 JEL classiﬁcation: F41 F42 E32

a b s t r a c t Using vector autoregressions on U.S. time series relative to an aggregate of industrialized countries, this paper provides new evidence on the dynamic effects of government spending and technology shocks on the real exchange rate and the terms of trade. To achieve identiﬁcation, we derive robust restrictions on the sign of several impulse responses from a two-country general equilibrium model. We ﬁnd that both the real exchange rate and the terms of trade—whose responses are left unrestricted—depreciate in response to expansionary government spending shocks and appreciate in response to positive technology shocks. © 2010 Elsevier B.V. All rights reserved.

Keywords: Real exchange rate Terms of trade Government spending shocks Technology shocks VAR Sign restrictions

1. Introduction How do international relative prices adjust to country-speciﬁc ﬁscal measures and productivity gains? This question is pivotal to understanding the international transmission mechanism; and yet, theoretical and empirical approaches tend to provide conﬂicting answers. Business cycle models under conventional calibrations predict that government spending raises the relative price of domestic goods, while productivity gains lower it—reﬂecting, respectively, an increase in relative demand and supply of domestic goods.1 Recent empirical studies based on estimated vector autoregressive (VAR) models suggest the opposite. Kim and Roubini (2008), Monacelli and Perotti (2006), and Ravn et al.

⁎ Corresponding author. Adenauerallee 24-42, 53113 Bonn, Germany. Tel.: + 49 228 731979; fax: + 49 228 737953. E-mail addresses: [email protected] (Z. Enders), [email protected] (G.J. Müller), [email protected] (A. Scholl). 1 See, e.g., Backus et al. (1994) and Erceg et al. (2005). Assuming debt-ﬁnance, textbook versions of the Mundell–Fleming model also predict that an exogenous increase in government spending appreciates exchange rates. In the case of tax ﬁnance, results differ as disposable income and money demand fall if money supply is unchanged, see Frenkel and Razin (1987). For similar reasons, government spending depreciates the nominal exchange rate in Obstfeld and Rogoff (1995). 0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.08.005

(2007), among others, ﬁnd that government spending depreciates the real exchange rate. Corsetti et al. (2008b), Kim and Lee (2008), and Enders and Müller (2009) document that productivity gains (or “technology shocks”) appreciate real exchange rates, measured by the terms of trade or the relative price of consumption across countries.2 In order to reassess these puzzling ﬁndings, the present paper employs a new methodological approach: while existing studies identify exogenous structural innovations through either short-run or long-run restrictions, we follow Uhlig (2005) and restrict the sign of the responses to the shocks we seek to identify.3 Importantly, and in

2 The aforementioned studies on ﬁscal shocks focus on the real exchange rate and consider data for Australia, Canada, the U.K. and the U.S. Evidence on the effect of government spending shocks on the terms of trade is somewhat mixed, see, for instance, Corsetti and Müller (2006) or Monacelli and Perotti (2008). Regarding technology shocks, evidence for an appreciation is established for U.S. data. Corsetti et al. (2008b) ﬁnd an appreciation in Japan as well, while Kim and Lee (2008) report a depreciation for the Euro area and Japan. 3 An increasing number of studies have recently employed sign restrictions. They are used, for instance, to identify government spending and technology shocks in a closed economy context by Mountford and Uhlig (2009) and Peersman and Straub (2009). In an open economy context, with a focus on identifying monetary policy shocks, sign restrictions are employed by Faust and Rogers (2003), Farrant and Peersman (2006), and Scholl and Uhlig (2008) among others.

54

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

contrast to a closely related study by Corsetti et al. (2009a), we use a quantitative general equilibrium model to formally derive the sign and the time horizon of the identiﬁcation restrictions. Our model is richly speciﬁed and nests distinct transmission mechanisms, once we consider the entire range of plausible parameterizations. Speciﬁcally, while the model delivers robust predictions for the behavior of several macroeconomic variables, it does not yield clear-cut predictions for how exchange rates respond to government spending and technology shocks. This result is key to our identiﬁcation strategy: we derive sign restrictions for several variables from the model, while remaining agnostic about exchange rate dynamics.4 We establish new evidence on how productivity gains and government spending impact U.S. exchange rates. In contrast to existing studies, which analyze the effect of either government spending or productivity gains in isolation, we assess their effects jointly in order to establish encompassing evidence on the international transmission mechanism. Speciﬁcally, we estimate our VAR model on quarterly time series for the U.S. relative to an aggregate of industrialized countries for the post-Bretton-Woods period 1975Q1– 2005Q4. The VAR includes data for consumption, output, investment, government spending, the government budget balance, inﬂation, the short-term interest rate and exchange rates. As a measure for the latter, we consider both the real effective exchange rate and the terms of trade, in order to control for the possibility that exchange rate changes merely reﬂect ﬂuctuations in the price of non-traded goods. We ﬁnd that exogenous expansions of government spending depreciate the real exchange rate as well as the terms of trade. Positive innovations to technology, instead, appreciate the real exchange rate and the terms of trade in the short-run. While the terms of trade converge back to their initial value, the real exchange rate depreciates in the medium-run after a positive technology shock. Sensitivity analysis, accounting for various complications such as possible anticipation effects of government spending, monetary policy shocks, or variations in the sample period, shows that these results are robust. Overall, our results corroborate the ﬁndings of existing studies regarding the effects of government spending and technology shocks on exchange rates, even though we employ an identiﬁcation scheme which is conceptually quite distinct. Identiﬁcation assumptions are, by their very nature, controversial, and evidence on exchange rate dynamics which is robust across identiﬁcation schemes seems particular relevant in assessing conﬂicting theoretical accounts of the international transmission mechanism. Speciﬁcally, international relative prices play an important role in allocating country-speciﬁc risk in the absence of explicit risk-sharing. Cole and Obstfeld (1991) identify conditions under which international price movements fully insure country-speciﬁc risk, thereby supporting the efﬁcient allocation. Yet, as shown in a recent theoretical contribution by Corsetti et al. (2008a), to the extent that technology shocks appreciate the real exchange rate, country-speciﬁc consumption risk is actually ampliﬁed. The reverse holds for government spending shocks. Our empirical ﬁndings are thus consistent with the notion that, in the short-run, international price movements tend to amplify rather than to mitigate country-speciﬁc consumption risk. The remainder of the paper is organized as follows. In Section 2 we describe our identiﬁcation strategy and outline a quantitative 4 The validity of our identiﬁcation assumptions thus rests on the plausibility of our theoretical framework. Yet working with a fully speciﬁed general equilibrium model imposes discipline on how to specify sign restrictions. By the same token, it allows for a quantiﬁcation of the time horizon for which restrictions may be imposed as well as for an explicit treatment of a possible anticipation of government spending shocks. We therefore consider our study complementary to Corsetti et al. (2009a) who employ sign restrictions to identify demand and technology shocks in the manufacturing sector and study their effect on the real exchange rate. Rather than using a fully speciﬁed general equilibrium model, they use sector-speciﬁc information to achieve identiﬁcation.

business cycle model from which we derive sign restrictions. In Section 3 we discuss our VAR speciﬁcation and results. Section 4 discusses sensitivity and Section 5 concludes. 2. Identifying government spending and technology shocks 2.1. Sign restrictions As discussed above, several studies use VAR models to document the effects of government spending and technology shocks on exchange rates. In these studies identiﬁcation is based on either short-run or long-run restrictions. In the following we propose to bring an alternative identiﬁcation scheme to bear on the question, because the evidence established to date conﬂicts with the predictions of business cycle models, at least if standard calibrations are considered. In the following we brieﬂy outline our approach. We start from the following reduced-form VAR model Yt = μ + Bð1Þ Yt−1 + Bð2Þ Yt−2 + … + BðmÞ Yt−m + ut ;

h i E ut u′t = Σ; ð1Þ

t = 1,..., T, for some ℓ-dimensional vector of variables Yt, coefﬁcient matrices B(i) of size ℓ × ℓ and a variance–covariance matrix for the one-step ahead prediction error Σ. Letting υt, with E[υtυ′] t = Iℓ, denote the vector of structural shocks, we need to ﬁnd a matrix A such that ut = Aυt in order to achieve identiﬁcation. Instead of restricting the matrix A a priori, we follow Uhlig (2005) and Mountford and Uhlig (2009) and identify structural shocks by imposing sign restrictions on impulse–response functions of selected P variables for a certain period k = P k ; :::; k following the shock. Intuitively, we consider various matrices A and check, for each case, whether the sign restrictions are fulﬁlled and dismiss the matrix if this is not the case. Below, we derive the sign restrictions on the basis of a quantitative business cycle model. Speciﬁcally, we assess—for a wide range of model parameterizations—whether the response of a variable to a particular shock is either robustly negative or positive for a speciﬁc time period k after the shock impacts the model economy. To ﬁx ideas, let n be the number of structural shocks that we seek to identify. Mountford and Uhlig (2009) show that identifying n shocks is equivalent to identifying an impulse matrix of rank n that is a sub-matrix of matrix A satisfying A A′ = Σ. Any impulse matrix can be written as h

ð1Þ

ðnÞ

a ; …; a

i

˜ = AQ

ð2Þ

where Ã is the lower triangular Cholesky factor of Σ and Q =[q(1), …,q(n)] is a n × ℓ matrix consisting of orthonormal rows q(s), s= 1, …,n, such that QQ′ =In. Similarly to Uhlig (2005), one can show that the impulse response to a(s) can be written as linear combination of the impulse responses obtained under a Cholesky decomposition of Σ. Let cji(k) be the impulse response of the jth variable at horizon k to the ith shock in the Cholesky decomposition of Σ and deﬁne ci(k) ∈ ℝℓ to be the vector response [c1i(k),..., cℓi(k)]. Then the impulse response r(s) a (k) to the impulse vector a(s) is given by ðsÞ

ℓ

ð sÞ

ra ðkÞ = ∑ qi i=1

ci ðkÞ:

ð3Þ

The restrictions we impose to identify an impulse vector ðsÞ ðsÞ ra ðkÞ ≥0; j ∈ J þ and ra ðkÞ ≤

characterizing shock s are that

j

j

0; j ∈ J − for some subsets of variables J þ and J − and some horizon P

k =P k ; …; k .

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

For the actual estimation we employ a Bayesian approach. Speciﬁcally, we use a ﬂat Normal-Wishart prior (see Uhlig (1994) for a detailed discussion of the properties), while the numerical implementation follows Rubio-Ramirez et al. (2005) and can be summarized as follows. We take a draw from the Normal-Wishart posterior for (B, Σ) and construct an arbitrary independent standard normal matrix M. We obtain the orthonormal matrix Q using the QRdecomposition of M such that QQ′ = I and QR = M. We construct impulse vectors a according to Eq. (2) and use Eq. (3) to compute the impulse responses. Considering orthogonal structural shocks may result in tight identifying sign restrictions in the sense that many draws from the Normal-Wishart posterior for the VAR parameters (B, Σ) are rejected because they do not permit any impulse matrices that satisfy the sign restrictions. This means that many draws receive zero prior weight, even in cases where only few of the restrictions are mildly violated. This issue gets more severe if the number of orthogonal shocks and the number of variables included in the VAR model increases. To account for this complication, we allow for small deviations ε from the sign restrictions and deﬁne

ðsÞ ωa ðkÞ

j

8 P ðsÞ > k ; :::; k and s = 1; ::; n; < maxf− ra ðkÞ ; 0g for j ∈ J þ ; k = P j = P > :maxf raðsÞ ðkÞ ; 0g for j ∈ J − ; k = P k ; :::; k and s = 1; ::; n:

55

economy consists of two symmetric countries indexed by i ∈ {1, 2}. We refer to country 1 as the domestic economy or “Home”, and to country 2 as “Foreign”. 2.2.1. Households In country i, a representative household allocates resources to consumption goods, Cit, and supplies labor, Hit, to monopolistic ﬁrms. Preferences are given by h ∞

μ

Cit ð1−Hit Þ1−μ

i1−γ

; μ b 1; 1−γ h i−1 βi0 = 1; βit + 1 = 1 + ψ Citμ ð1−Hit Þ1−μ βit;

E0 ∑ βit

ð5Þ

t =0

t ≥ 0:

Here βit is an endogenous discount factor implying higher discounting if consumption and leisure are above their steady-state values.6 The positive constants μ and γ specify the preferences of households. Labor and capital are internationally immobile; households in country i own the capital stock, Kit, and rent it to intermediate good ﬁrms on a period-by-period basis. It may be costly to adjust the level of investment, Iit. As in Christiano et al. (2005), the law of motion for capital is given by

j

Kit We keep the impulse responses if the sum of the squared deviations over all structural shocks, variables and horizons is smaller than ε: ∑ ∑ ∑ s

j

k

2 ð sÞ ωa ðkÞ b ε; j

ε ≥ 0:

ð4Þ

Inference statements are based on the posterior distribution of those draws for which Eq. (4) is satisﬁed.5

( =

5 Alternatively, Mountford and Uhlig (2009) minimize a penalty function for sign restriction violations for each draw from the posterior distribution for the VAR parameters. However, to account for several orthogonal shocks they sequentially determine the optimal impulse vectors such that the ordering of the structural shocks may be important. To avoid this, we simply allow for small deviations and draw the impulse vectors simultaneously. This also implies that, in contrast to Mountford and Uhlig (2009), we simultaneously estimate the reduced-form parameters together with the impulse matrix.

= ð1−δÞKit + ½1−ΨðIit = Iit−1 ÞIit ;

ð6Þ

where δ denotes the depreciation rate; restricting Ψ(1) = Ψ′(1) = 0, and Ψ″(1) = χ N 0 ensures that the steady-state capital stock is independent of investment adjustment costs captured by the parameter χ. Across countries there is trade in nominal noncontingent bonds, Θit, denominated in the currency of country i. The budget constraint of the representative household in country i reads as

2.2. A quantitative business cycle model We now turn to a quantitative business cycle model from which we derive sign restrictions. The model is a two-country business cycle model featuring various frictions frequently employed in earlier studies, see, e.g., Chari et al. (2002) and Kollmann (2002). Notably, we consider various degrees of price rigidity, since sticky prices potentially alter the transmission of real shocks, as forcefully argued by Galí (1999). In addition, this assumption allows us to study the behavior of international relative prices also in response to monetary policy shocks, once we turn to sensitivity analysis. Moreover, we assume that each country specializes in the production of a particular type of goods. Households in each country consume both types, but to a different extent, such that changes in their relative price govern real exchange rate dynamics. We abstract from non-traded goods, as ﬂuctuations in their relative price are of minor importance in accounting for U.S. real exchange rate changes, see Engel (1999) and Chari et al. (2002). Before we turn to model simulations in order to derive sign restrictions, we brieﬂy outline the model structure. The world

+ 1

Θ1t Θ2t

+ 1

+ 1

K ð1−τit Þ Wit Hit + Rit Kit + ϒit −Pit Cit −Pit Iit −1 −1 + D1t + 1 R1t + St Θ2t + 1 R2t − Θ1t −D1t −St Θ2t ; +

D2t + 1 R−1 2t

+ Θ1t

+ 1 ðR1t

−1

St Þ

−Θ2t −D2t −S−1 t

for i = 1; Θ1t ;

for i = 2;

ð7Þ where Wit and RitK denote the nominal wage rate and the rental rate of capital, and ϒit are nominal proﬁts earned by monopolistic ﬁrms and transferred to households. τit denotes the tax rate levied on households' income; Pit is the price of the ﬁnal good deﬁned below; Rit is the gross nominal interest rate, Dit denotes debt issued by the government in country i held by domestic residents, and St is the nominal exchange rate. In each country households maximize Eq. (5) subject to Eqs. (6), (7) and a non-Ponzi scheme condition. 2.2.2. Final good ﬁrms Consumption and investment goods are composite goods which households purchase from ﬁnal good ﬁrms. These ﬁrms operate under perfect competition and buy intermediate goods from a continuum of monopolistic competitive ﬁrms. We use j ∈ [0, 1] to index those intermediate good ﬁrms as well as their products and prices. Further, let Ait(j) and Bit(j) denote the amount of good j originally produced in country 1 and 2, respectively, and used in

6 We assume that the effect of consumption and leisure on the discount factor is not internalized by the household, see Schmitt-Grohé and Uribe (2003) for a detailed analysis. The parameter ψ determines how strongly the discount factor responds to consumption and leisure; it also pins down the value of the discount factor in steadystate.

56

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

country i to assemble ﬁnal goods Fit. These are produced under the following technology 8 2 3 σ > σ−1 > ε !σ−1 > σ 1 ε−1 > 1 ε−1 > > 6 7 σ ε dj ω ∫ A ð j Þ > >6 7 0 1t > > 6 7 > > 6 7 > > 6 7 ; for i = 1 > > 6 σ−1 7 ! > ε > 6 7 > σ 1 ε−1 > 5 4 1 ε−1 > > + ð1−ωÞσ ∫ B1t ð jÞ ε dj > > 0 < Fit =

2 3 σ > > σ−1 > ε !σ−1 > σ 1 ε−1 > 1 ε−1 > >6 7 σ ε dj ∫ B ð j Þ ω > > 6 7 0 2t > > 6 7 > > 6 7 > > 6 7 ; for i = 2 > > 6 σ−1 7 ! > ε > 6 7 > σ 1 ε−1 > 4 5 1 ε−1 > > > : + ð1−ωÞσ ∫0 A2t ð jÞ ε dj

governments which are assumed to consume only domestically produced goods, as discussed below. Production of intermediate goods is Cobb–Douglas:

ð8Þ

B

A

A

P1t ð jÞ = St P2t ð jÞ:

z

A

8 1 1−σ 1−σ 1−σ > > A B > > + ð 1−ω Þ P ; for i = 1 ω P > 1t 1t < > > 1−σ > > > : ð1−ωÞ P2tA + ω P2t

1 B 1−σ 1−σ

;

ð10Þ

; for i = 2

1 1 1 A 1 B 1−ε B 1−ε 1−ε 1−ε ∫ Pit ð jÞ dj and Pit = ∫ Pit ð jÞ dj 0

0

ð11Þ

denote the GDP deﬂators in Home and Foreign, respectively. The problem of ﬁnal good ﬁrms is to minimize expenditures in assembling intermediate goods subject to Eq. (8) and the requirement that Fit = Cit + Iit. The ﬁrst-order condition that characterizes the behavior of ﬁnal good ﬁrms in equilibrium implicitly deﬁnes the demand for a generic intermediate good, YitD(j). For future reference, taking the perspective of the home country, we deﬁne the real exchange rate as follows RXt = St P2t = P1t ;

ð14Þ

such that ρz captures the degree of autocorrelation and ρzz possible spillovers across countries. Labor and capital inputs of ﬁrm j, Hit(j) and Kit(j), are adjusted freely in each period. Price setting, however, is constrained exogenously by a discrete time version of the mechanism suggested by Calvo (1983). In a given period, each ﬁrm has the opportunity to change its price with probability 1 − ξ only. When a ﬁrm has the opportunity, it sets the new price in order to maximize the expected discounted value of proﬁts; otherwise prices are indexed to past inﬂation, where the degree of A indexation is given by ι ∈ [0, 1]. When setting the new price P1t (j) or B P2t(j), the problem of a generic intermediate good ﬁrm j in country i is given by ∞

k

max ∑ ξ Et

8 > < ρit;t > : ρit;t

h

ι A PitA ð jÞ PitA + k−1 = Pit−1 −MCit h ι B B B + k Yit + k ð jÞ Pit ð jÞ Pit + k−1 = Pit−1 −MCit

+ k Yit + k ð jÞ

i + k + k

i

= Pit

+ k;

for i = 1

= Pit

+ k;

for i = 2

ð15Þ subject to the production function (13) and the optimal choice of factor inputs which minimizes marginal costs, MCit.7 As households own ﬁrms, proﬁts are discounted with ρit, t + k, which equals the household's marginal rate of substitution between consumption in period t and t + k.

where

Pit =

ð13Þ

ð9Þ

The price for ﬁnal goods is given by

Pit =

;

Zit = ρz Zit−1 + ρzz Z3−i;t−1 + εit ;

k=0

B

1−θ

where Zit denotes the level of technology common to all ﬁrms. It evolves exogenously according to

where σ denotes the elasticity of substitution between foreign and domestic goods (“trade price elasticity”, for short) and ε measures the elasticity of substitution between goods produced within the same country. The parameter ω measures the home bias in the composition of ﬁnal goods. Let PitA(j) be the price in country i of an intermediate good produced in country 1 and PitB(j) the price in country i of a good produced in country 2. Assuming that the law of one price holds, we have P1t ð jÞ = St P2t ð jÞ;

θ

Z

Yit ð jÞ = e it Kit ð jÞ Hit ð jÞ

ð12Þ

such that an increase corresponds to a depreciation. The terms of trade are deﬁned as the price of imports relative to the price of B A / P1t . exports: P1t 2.2.3. Intermediate good ﬁrms At the intermediate good level, ﬁrms specialize in the production of differentiated goods. A generic ﬁrm j ∈ [0, 1] in country i engages in monopolistic competition facing imperfectly-elastic demand from domestic and foreign ﬁnal good producers, as well as domestic

2.2.4. Fiscal and monetary policy Government policies are characterized by feedback rules. Turning to ﬁscal policy ﬁrst, we assume that government spending, Git, consists of a bundle of intermediate goods. Speciﬁcally, we assume an aggregation technology isomorphic to Eq. (8), except that only domestically produced goods enter the consumption basket of the government.8 Government consumption evolves according to the following feedback rule: g Git = 1−ρg Gi + ρg Git−1 + φy ðYit −Yi Þ−φd ðDit −Di Þ + εit;t−n ; ð16Þ ε−1

where Yit =

∫10 Yit + k ð jÞ ε

! dj

ε ε−1

is an index for aggregate domes-

tic production (“output”, for short); letters without time subscript refer to steady-state values; ρg captures persistence, while φy and φd measure to what extent government spending responds to the deviation of output and debt from their steady-state values.9 εit,g t − n is an i.i.d. innovation to current government spending, which may have been correctly anticipated n periods in advance because, say, of institutional features of the legislative process.

7 In this formulation we impose the constraint that demand is met by actual G production at all times: Yit(j) = YD it (j) + Yit (j), where the last term denotes the demand stemming from government consumption. 8 Put differently, we assume that government goods are assembled in the same way as in Eq. (8), with ω = 1. The evidence discussed in Corsetti and Müller (2006) suggests that the import content in government spending is generally less than half the import content in private spending. As a ﬁrst approximation it is thus reasonable to assume zero import content in government spending. 9 Rules of this type have been estimated by Galí and Perotti (2003), among others.

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Regarding the tax rate we assume that τit = τi + φτ(Dit − Di) / Yi, with φτ ≥ 0 measuring how strongly the tax rate adjusts to the level of debt.10 The budget constraint of the government in country i is given by

57

We approximate the equilibrium conditions of the model around a deterministic, symmetric, zero-debt steady-state and compute the model solution numerically. In order to determine parameter values we focus on steady-state relationships which link particular parameters to ﬁrst moments of the data or, in case parameters have no bearing on the steady-state, turn to empirical studies which report appropriate estimates. We account for uncertainty of measurement by specifying a particular interval of plausible values for each parameter. As our VAR model is estimated on time series data for the U.S. relative to an aggregate of industrialized countries, we mostly rely on evidence for the U.S., but also account for non-U.S. observations. In order to generate sign restrictions which are robust across the entire range of plausible model parameterizations, we adopt the following procedure. Given the speciﬁed intervals and assuming a uniform and independent distribution, we draw a total of 100,000 realizations of the parameter vector. For each realization we compute impulse responses to a government spending and a technology shock. Finally, we compute conﬁdence bounds containing 99% of the responses and analyze which variables respond unambiguously either positively or negatively to a particular shock for a speciﬁc number of periods after the shock.11 In addition, we compute impulse responses to monetary policy shocks and anticipated government spending shocks, because in our empirical analysis we also account for these shocks once we assess the robustness of the results obtained under the baseline speciﬁcation. Table 1 summarizes the range of parameter values used in the model simulations. A period in the model is one quarter. As the discount factor relates, via the Euler equation, to observed after-tax returns, we chose the interval for the steady-state value of the discount factor to be consistent with annual after-tax returns between 4.2 and 7.5%, see Gomme and Rupert (2007) and references therein. The elasticity of substitution between varieties determines the markup, for which Rotemberg and Woodford (1993) ﬁnd values between 20 and 40%. Correcting for a potential bias due to intermediate inputs reduces the lower bound to 7%.12 For parameters

governing the capital share and the depreciation rate, we allow for values consistent with a range of observations for the labor share and annual depreciation rates in various sectors of the economy, see Rotemberg and Woodford (1999) and Gomme and Rupert (2007), respectively. Government spending on domestic goods in steadystate, g, is assumed to vary between 14 and 23% of output. Regarding the degree of home bias, we consider an interval which accounts for an export share between 7 and 12.5%.13 The parameters μ and γ jointly determine the Frisch elasticity and the intertemporal elasticity of substitution (IES). Our parameter intervals account for the uncertainty regarding appropriate values for these elasticities, see Basu and Kimball (2002) and Domeij and Flodén (2006). We also allow for higher values, in line with the early RBC literature. The parameter governing the average price duration is somewhat controversial in the literature. Here we rely on international evidence reported by Dhyne et al. (2006). As a lower bound we employ the value for the U.S., yielding an average price duration of 6.7 months. The upper bound of 13 months is set in line with observations for the Euro Area. Regarding χ, the parameter capturing investment costs, we consider an interval which is centered around the point estimate of Christiano et al. (2005) and accounts for two standard errors. Concerning price indexation, we allow for the whole range from zero to full indexation. In order to specify monetary policy rules we consider an interval given by estimates for the U.S. and the Euro Area, reported by Clarida et al. (2000) and Enders et al. (2009), respectively. We specify an admissible range of parameter values for the coefﬁcients in the government spending feedback rule drawing on estimates reported by Galí and Perotti (2003) for a sample of OECD countries. The tax rule coefﬁcient is chosen such that deﬁcits display considerable persistence as in the data, see Corsetti and Müller (2008). Finally, for the persistence of technology shocks we sample from a range of two standard deviations around point estimates reported by Backus et al. (1992) and Heathcote and Perri (2002). Possible cross-country spillovers are admitted to take any value which does not lead to explosive behavior of the model. Regarding the trade price elasticity σ, we consider two distinct intervals of empirically plausible values. One interval allows for low values, in line with evidence reported in a number of recent macroeconometric studies, see Enders and Müller (2009) for further discussion. A second interval allows for higher values of up to 2.5, the highest value which Backus et al. (1994) consider on the basis of surveying model-independent evidence. In our simulations we consider both intervals (each for 50,000 draws), but omit the middle range, as this will allow us to highlight the importance of the trade price elasticity for the international transmission mechanism and, in particular, for the sign of the exchange rate response.14 Turning to our results, we display in Figs. 1 and 2 the impulse responses to an unanticipated innovation in government spending and technology, respectively. We consider both intervals for the trade price elasticity: the shaded area covers 99% of the responses in case values are drawn from the low interval (pointwise), the dashed lines cover 99% of the responses if values are drawn from the high interval. On the horizontal axis we measure periods after the shock (in quarters), on the vertical axis we measure responses in percentage deviation from steady-state values. We display the response of

10 In the simulation of the model we only allow for values of {φτ, φd} such that government debt is stationary. It is interesting to observe that in the case of φτ = 0 all ﬁnancing of government spending occurs through reduced future spending. As a result, the standard wealth effect of government spending is absent. 11 Computing the impulse responses for a large number of realizations of the parameter vector ensures the robustness of our sign restrictions. Assuming a uniform distribution over the speciﬁed interval, we consider the entire range of parameter values, while dismissing all values outside the interval as implausible on a priori grounds. In order to dismiss very unlikely implications of realizations of the parameter vector, we consider 99% coverage bands. 12 Here and in the following we draw parameters sequentially, and treat the earlier realizations as given when calculating the implied target values.

13 The values for the export and government shares correspond to observations for the U.S. over the sample period used in our estimation (data sources are discussed below). 14 To be precise, the trade price elasticity interacts with all model parameters, but most importantly with the degree of home bias and the persistence of shocks in determining the sign of the exchange rate response. As a result, there is no single threshold value for σ, which is independent of the values assumed for the other model parameters. Hence, we exclude a sizeable range between the two intervals. Note that we do not judge parameter values in this range as empirically implausible, but simply omit them in the simulation to illustrate that the trade price elasticity governs the sign of the exchange rate response. Note also that we do not restrict the exchange rate response in order to identify shocks in our estimated VAR model.

G K Dit + Pit Git = τit Wit Hit + Rit Kit + ϒit + Dit

−1 + 1 Rit ;

ð17Þ

where PGit is the price index of government consumption. Monetary policy is characterized by an interest feedback rule, whereby the policy rate is adjusted in response to domestic (i.e., producer-price) inﬂation, Πit, and a measure of the output gap, yit = (Yit − Yi) / Yi as, for instance, in Galí and Monacelli (2005): r Rit = ρr Rit−1 + ð1−ρr Þ R + ϕπ ðΠit −ΠÞ + 0:25ϕy yit + εit :

ð18Þ

Here ρr ≥ 0 captures interest rate smoothing, while ϕπ and ϕy measure the long-run inﬂation and output gap response of the policy rate; εitr is an i.i.d. exogenous monetary policy shock. 2.3. Generating sign restrictions

58

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Table 1 Parameter values used in simulation.

β ε θ δ g ω μ γ ξ χ ι ϕπ ϕy ρr ρg φy φd φτ ρz ρzz σ

Parameter description

Range

Target/source

Discount factor (steady-state) Elasticity of substitution Capital share Depreciation rate (× 100) Government spending (steady-state) Home bias in ﬁnal goods Consumption weight in utility Risk aversion Fraction of prices kept unchanged Investment adjustment costs Indexation of prices Inﬂation response of interest rate Output response of interest rate Interest rate smoothing Government spending persistence Output gap response of G-spending Debt response of G-spending Debt response of tax rate Technology persistence Technology spillover Trade price elasticity

[0.982, 0.99] [3.50, 15.0] [0.15, 0.39] [0.38, 6.02] [0.14, 0.23] [0.84, 0.92] [0.02, 0.89] [1.00, 20.0] [0.55, 0.77] [2.05, 2.91] [0.00, 1.00] [1.68, 2.15] [0.78, 0.93] [0.62, 0.79] [0.70, 0.85] [− 0.2, 0.20] [1 − β, 0.04] [0.00, 0.04] [0.83, 0.98] [0, 0.99 − ρz] [0.10, 0.33] or [1.00, 2.50]

After-tax return Markup Labor share Various sectors Government share Export share Frisch elasticity IES Price duration (months) Christiano et al. '05 All admissible values Clarida et al. '98/Enders et al. '09 " " Galí and Perotti '03 " " Corsetti and Müller '08 Backus et al. '92/Heathcote and Perri '02 Remaining admissible values Enders and Müller '09/Backus et al. '94

[0.042, 0.075] [1.07, 1.40] [0.57, 0.78] [0.015, 0.22] [0.14, 0.23] [0.07, 0.125] [0.18, 4.00] [0.48, 1.00] [6.70, 13.0]

Notes: Parameter values used in simulation of the model. “Range” speciﬁes interval from which values of the parameter vector are drawn for each simulation of the model.

relative variables to a domestic shock, i.e., the difference in the response of a domestic variable and its foreign counterpart, because we are concerned with the behavior of the real exchange rate, which is determined by these relative variables.15 For the real exchange rate and net exports we report the response of the domestic variable. We do not show responses for the terms of trade, because they move proportionally to the real exchange rate, as we assume home bias throughout. Fig. 1 shows how the economy adjusts to unanticipated government spending shocks. Relative government spending increases for at least four quarters and possibly falls below its steady-state level thereafter, as it is systematically cut in response to higher public debt.16 Output increases for at least two quarters, but a decline below the steady-state cannot be ruled out for later periods. The government budget deteriorates for at least four quarters. Importantly, the sign of the response of private consumption is ambiguous, once we consider the entire range of plausible model parameterizations.17 Government spending crowds out investment for at least six quarters and triggers a rise in inﬂation. The nominal interest rate rises for at least four quarters. The responses of net exports and the real exchange rate are ambiguous at all horizons. The sign of the response of net exports is determined by the trade price elasticity: if we sample from the low (high) interval, the response is positive (negative).18 The real exchange rate appreciates, i.e., falls, in case we assume a high trade price elasticity—at least as long as government spending expands. If,

15 Given the symmetry of the model, results are unchanged if we consider relative innovations, e.g., an exogenous increase in domestic government spending relative to foreign government spending. 16 For ease of exposition, we do not always explicitly refer to the fact that we are dealing with variables in relative terms in the following discussion. 17 Galí et al. (2007) analyze the transmission of government spending shocks in a new Keynesian closed economy model. They show that government spending raises private consumption only in the presence of labor market frictions and if a considerable fraction of households consumes disposable rather than permanent income. Our simulation results show that, although our model does not feature these frictions, we nevertheless cannot rule out a positive response of consumption to government spending. This is because we allow for a wide range of preference speciﬁcations while assuming infrequent price adjustment, see Bilbiie (2009) for a detailed analysis. 18 See Müller (2008) for a detailed analysis of how the trade price elasticity (in relation to the IES) determines whether net exports rise or fall in response to government spending shocks.

instead, we assume a low trade price elasticity, the sign of the exchange rate response is ambiguous. Corsetti et al. (2008a) show that, absent explicit risk-sharing across countries, the trade price elasticity determines the sign of the exchange rate response to technology shocks, thereby possibly amplifying consumption risk. Our simulation results suggest that this is the case for government spending shocks, too. Fig. 2 shows how the economy adjusts to a positive technology innovation in the home country. The response of government spending is ambiguous as we allow for a positive and negative adjustment of government spending to output, which increases unambiguously for more than eight quarters. The budget does not fall on impact, because tax revenues are procyclical. The response of consumption is distinctly positive from quarter two to eight, after an initial period when a drop in consumption cannot be ruled out. The response of investment is positive during the ﬁrst four quarters after the shock, inﬂation falls for at least two quarters and the interest rate drops for at least six quarters. As with government spending shocks, the responses of net exports and the real exchange rate are governed by the trade price elasticity. If we consider only low values, net exports tend to fall, while they increase if a high trade price elasticity is assumed. Similarly, if a high elasticity is assumed the real exchange rate depreciates robustly, while its response is not clear-cut if a low trade price elasticity is assumed, in line with the results of Corsetti et al. (2008a). In sum, the model delivers sign restrictions for a number of variables, as their responses are qualitatively robust with respect to the entire range of plausible model parameterizations. Yet, at the same time, the model does not deliver unambiguous predictions as to how net exports and the real exchange rate respond to government spending and technology shocks. Table 2 summarizes the sign restrictions implied by the model simulations. Speciﬁcally, the length and sign of the restrictions are given by the maximum number of quarters for which the simulations provide robust predictions for the sign of the response of a particular variable. Table entries indicate whether a variable is restricted to respond nonnegatively (+), non-positively (−) to a speciﬁc shock, or whether it is left unrestricted (ø). Numbers indicate the quarters for which the response is restricted, with “0” indicating the impact period of the shock. In addition to unanticipated government spending shocks (column 1) and technology shocks (column 3), Table 2 reports sign restrictions for monetary policy shocks (column 4) and for possibly anticipated

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Government spending 1.5

−1.5

Output

2

4

6

8

10

−0.3

0.3

0

2

Consumption

6

8

10

0

2

4

6

8

10

−0.9

0

2

4

6

8

10

4

6

8

−0.2

0

Net exports

2

10

−0.1

4

6

8

10

2

4

6

8

10

Real exchange rate

0.1

2

0

0.2

Interest rate

0

−0.3

Inflation

0.9

0.2

−0.2

4

Investment

0.2

−0.2

Budget balance

0.3

0

59

0.3

0

2

4

6

8

10

−0.3

0

2

4

6

8

10

Fig. 1. Model responses to exogenous and unanticipated increase in domestic government spending under various parameterizations. Notes: Responses are measured in relative terms, i.e., Home less Foreign, except for net exports and the real exchange rate (Home); shaded area covers 99% of responses (pointwise) assuming a low trade price elasticity; dashed lines display the same statistic assuming a high trade price elasticity; number of draws: 100,000. Horizontal axes: quarters; vertical axis: percentage deviation from steady-state.

government spending shocks (column 2). In this case, we compute for each realization of the parameter vector impulse responses to both unanticipated as well as to anticipated shocks considering an anticipation horizon of up to two quarters, i.e., we allow for n ∈ {0,1,2}. If a government spending shock is anticipated, the date at which it becomes known deﬁnes the impact period.19 Our set of sign restrictions ensures that productivity and ﬁscal shocks are distinguishable along several dimensions. The same holds for monetary policy shocks and potentially anticipated spending shocks. Nevertheless, without explicitly analyzing the responses to other shocks we cannot rule out that these shocks satisfy a particular set of sign restrictions, too. In this respect, however, including a relatively large number of variables provides some assurance. For instance, considering preference shocks which trigger an exogenous increase in private consumption demand, we ﬁnd that responses are fairly similar to those of government spending shocks—except, that is, for the response of the government budget balance, which improves in response to preference shocks. Before turning to our empirical speciﬁcation, we note that several studies employ sign restrictions to identify ﬁscal and technology shocks. 19 In principle, recovering structural shocks from estimated VAR models is complicated by “ﬁscal foresight”, see Leeper et al. (2009). For our setup, however, we ﬁnd on the basis of the test suggested by Fernández-Villaverde et al. (2007) that the mapping from VAR shocks to structural shocks is typically invertible.

In particular, Corsetti et al. (2009a) investigate the effects of productivity and demand shocks on the real exchange rate using a six variable VAR estimated on U.S. time series relative to an aggregate of industrialized countries. They impose sign restrictions on labor productivity, manufacturing production (in country differences) and manufacturing production relative to GDP, for a horizon of 20 quarters. In order to discriminate between demand and productivity shocks they restrict the price of traded goods relative to the price of non-traded goods. In other words, their identiﬁcation assumptions relate to a number of variables not included in our analysis. Interestingly, regarding the restrictions on output, we ﬁnd that our model simulations provide no justiﬁcation for a priori restricting the response over such a long horizon. Yet, as we will show below, the output restriction imposed by Corsetti et al. (2009a) is in fact satisﬁed by our estimated impulse response function. This observation may explain why we ﬁnd similar effects of technology shocks on real exchange rates.20

20 Mountford and Uhlig (2009) take a closed economy perspective and restrict output, consumption, and investment to increase in response to a business cycle shock. Government spending shocks are assumed to be orthogonal to business cycle shocks and to raise government spending. Fratzscher and Straub (2009) analyze the effects of asset price shocks on the current account and also discriminate shocks to technology and government spending through their differential impact on inﬂation. Both studies restrict responses for one year, rather than considering a variable-speciﬁc horizon.

60

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Government spending 5.2

−5.2

Output

0

2

4

6

8

10

−1.6

0.7

0

2

Consumption

6

8

10

0

2

4

6

8

10

−12.3

0

2

4

6

8

10

4

6

8

−1

0

Net exports

2

10

−0.3

4

6

8

10

2

4

6

8

10

Real exchange rate

0.3

2

0

1

Interest rate

0

−0.7

Inflation

12.3

0.7

−0.7

4

Investment

3.2

−3.2

Budget balance

1.6

5.2

0

2

4

6

8

10

−5.2

0

2

4

6

8

10

Fig. 2. Model responses to positive technology innovation under various parameterizations. Notes: see Fig. 1.

3. New evidence on the behavior of U.S. real exchange rates 3.1. Data and baseline speciﬁcation We estimate the VAR model (1) on time series data for the U.S. relative to an aggregate of industrialized countries consisting of the Euro Area, Japan, Canada and the U.K. (“rest of the world”, for short). We include a constant and 4 lags of endogenous variables in the VAR model. The vector of endogenous variables consists of, in logs and real terms, private consumption, GDP, private investment, government spending as well as the primary budget balance scaled by GDP, inﬂation (measured using the GDP deﬂator), the nominal short-term interest rate and the log of the real exchange rate. We also report results for a speciﬁcation which includes the log of the terms of trade instead of the real exchange rate. We focus on a post-Bretton-Woods sample, with data ranging from 1975Q1 to 2005Q4, as we omit the ﬁrst two turbulent years after the collapse of the Bretton-Woods system. For all variables we consider time series data for the U.S., which we express, except for the real exchange rate and the terms of trade, relative to the rest of the world. Data for real output, private consumption, government spending, the GDP deﬂator, and private ﬁxed investment (excluding stockbuilding) are taken from the OECD Economic Outlook database. Government spending includes government spending on goods and services (government consumption),

but neither investment nor transfers.21 In addition, except for the Euro Area, we obtain from the same source data for the short-term interest rate, the primary government balance (measured in percent of GDP), exports of goods and services (value, local currency), imports of goods and services (value, local currency), and GDP (market prices). Net exports, as a fraction of GDP, are computed on the basis of these series. We use the series for the export price of goods and services (local currency) and the import price of goods and services (local currency) to measure the terms of trade. For the Euro area we obtain several series from the ECB's AWM database, see Fagan et al. (2001).22 We obtain the CPI-based real effective exchange rate for the U.S. from the Main Economic Indicators of the OECD. In constructing the rest-ofthe-world aggregate, we eliminate national basis effects by aggregating quarterly growth rates, weighted by each currency area's GDP.23

21 We do not consider government investment, because of an accounting problem in the U.K. in 2005Q2. We neither consider transfers to ensure consistency with our business cycle model. 22 Speciﬁcally, we use the short-term interest rate (STN), the deﬂator of exports of goods and services (XTD), the deﬂator of imports of goods and services (MTD), and the government primary surplus (GPN_YEN). 23 Euro area growth rates include West-Germany until 1990Q4, and uniﬁed Germany from 1991Q1 onwards (in case OECD data is used, similar adjustments have been applied in constructing the AWM database). Weights are based on PPP-adjusted values for the year 2000, as reported in the World Economic Outlook database (2007) of the IMF.

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69 Table 2 Sign restrictions implied by model simulations. Response of

Expansionary shock to Government spending No anticipation + Anticipation

Private consumption Output Investment Government spending Government budget Net exports Nominal interest rate Inﬂation Real exchange rate/terms of trade

ø

ø

+ 0−2 − 0−6 + 0−4 − 0−4

+ 2 − 0−6 + 2−4 − 2−4

ø

Technology Monetary policy + 2−8 + 0−8 + 0−4

+ 0 + 0 + 0

ø

ø

+ 0

+ 0

ø

ø

ø

+ 0−4 + 0

+ 2−4

− 0−6 − 0−2

ø

ø

− 0 + 0 + 0

ø

ø

Notes: responses of variables (in relative terms) are restricted to be non-negative (+), non-positive (−) or unrestricted (ø). Numbers indicate the time periods after the shock for which the responses are restricted. The column “+ Anticipation” refers to potential anticipation of up to two quarters.

Under our baseline speciﬁcation, we jointly identify unanticipated government spending shocks and technology shocks on the basis of the sign restrictions summarized in Table 2. Since the search for impulse responses that fulﬁll all sign restrictions exactly is very cumbersome, we allow for small deviations using criterion Eq. (4) introduced above, setting ε = 0.005. Inference is based on 1000 draws satisfying the identiﬁcation restrictions. The results reported below are robust with respect to assuming lower values of ε. In these cases, however, many draws from the Normal-Wishart posterior for the VAR parameters (B, Σ) receive zero prior weight, even if only few restrictions are mildly violated. 3.2. Effects of government spending and technology shocks Given the estimated VAR model and the identiﬁed shocks to government spending and technology, we compute and display the corresponding impulse responses in Figs. 3 and 4. In all panels we plot the median as well as the 16 and 84% quantiles of the posterior distribution of impulse responses. In our discussion of the results we will use the term “signiﬁcance” whenever both quantiles are either above or below zero at a particular point in time. Shaded areas indicate that the sign of a response has been restricted over the corresponding horizon. Fig. 3 shows the effects of an exogenous innovation in relative government spending.24 Government spending, displayed in the upper left panel, increases persistently. In line with the evidence reported by Perotti (2005) for a post-1980 sample as well as by Mountford and Uhlig (2009), we ﬁnd a very short-lived increase in output in response to government spending shocks. The increase is limited to the period for which we restrict output to respond nonnegatively. In fact, we ﬁnd that signiﬁcance bands cross the zero line while the response is still restricted to be non-negative, as a result of admitting small violations of our sign restrictions, see Eq. (4). The budget deteriorates persistently. Consumption, the response of which

24 Given that identiﬁcation is based on sign restrictions derived from a symmetric business cycle model, we are agnostic as to whether relative government spending rises because domestic government spending rises in absolute terms or merely relative to foreign government spending. The same applies to technology shocks.

61

is left unrestricted, does not display a signiﬁcant response, in line with evidence reported by Mountford and Uhlig (2009).25 Investment shows a protracted decline, while inﬂation increases considerably. Interest rates, in turn, increase initially as long as they are restricted to respond non-negatively, but fall thereafter. The middle and right panel of the last row show the response of the terms of trade and the real exchange rate. As discussed in the introduction, business cycle models under standard calibrations predict that government spending appreciates exchange rates, as do textbook versions of the Mundell–Fleming model. Yet, as shown above, our quantitative business cycle model fails to deliver robust predictions for how government spending impacts the real exchange rate (and the terms of trade), if one considers the entire range of plausible model parameterizations. Consequently, we do not restrict their responses and the obtained estimates constitute fresh evidence: government spending depreciates (raises) both the real exchange rate and the terms of trade. This ﬁnding largely conﬁrms evidence obtained under alternative identiﬁcation schemes. Following Blanchard and Perotti (2002), several authors assume that government spending is predetermined in order to achieve identiﬁcation. Kim and Roubini (2008) analyze U.S. time series for the period 1973–2002 and ﬁnd that government spending shocks depreciate the real exchange rate; Monacelli and Perotti (2006) report similar results for Australia, the U.S. and the U.K., but not for Canada. Ravn et al. (2007) pool the data of all four countries, reporting a depreciation, too. Results for the terms of trade on the basis of this identiﬁcation scheme are less clear-cut, see Corsetti and Müller (2006), Müller (2008), and Monacelli and Perotti (2008).26 Fig. 4 shows the effects of a positive innovation to relative productivity. There is no effect on government spending, but output rises for an extended period, beyond the horizon that is restricted to be characterized by a non-negative response. There is also a beneﬁcial and persistent effect on the government budget. The response of consumption and investment is also strongly positive, at horizons before and after restrictions are imposed. Inﬂation shows a persistent decline, as do interest rates. In the later case, the fall is limited to the period for which we impose restrictions. The middle and right panels of the last row of Fig. 4 show the response of the terms of trade and the real exchange rate. As discussed in the introduction, business cycle models under standard calibrations predict that gains in relative productivity depreciate international relative prices. A notable exception is known as the Balassa– Samuelson effect, according to which productivity gains in the production of traded goods may appreciate the real exchange rate via their effect on the price of non-traded goods. However, even within our model, which does not allow for non-traded goods, we fail to detect a robust depreciation of the real exchange rate in response to positive technology shocks, since we consider a wide range of plausible model parameterizations. Consequently, we do not restrict the response of the real exchange rate and the terms of trade. We ﬁnd that exchange rates appreciate (fall) signiﬁcantly during the ﬁrst few quarters after a technology shock. To the extent that technology shocks not only appreciate the real exchange rate, but also the terms of trade (computed on the basis of import and export price

25 The response of consumption to government spending shocks has been the subject of a considerable debate with different results emerging from different identiﬁcation schemes based on short-run restrictions and narrative identiﬁcation schemes, see Perotti (2007) and Ramey (2009), respectively. 26 Beetsma et al. (2008) consider a panel of European countries and ﬁnd that government spending appreciates the real exchange rate. The narrative approach to the identiﬁcation of government spending shocks, suggested by Ramey and Shapiro (1998), is a widely considered alternative to the Blanchard–Perotti approach. It is employed by Monacelli and Perotti (2006) who ﬁnd that government spending falls in response to the Carter–Reagan military build-up, while the real exchange rate depreciates.

62

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Government spending

Output

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0

0

−0.1

0.2 0.1 0 −0.1

−0.2

−0.1

−0.3

−0.2 −0.3

Budget balance

−0.2

−0.4 0

10

20

30

0

Consumption

10

20

30

0

Investment

10

20

30

20

30

Inflation

0.5

0.2

0.2

0.1

0 0.1

0 −0.5

−0.1

0

−0.2

−1 −0.1

−0.3 −1.5 −0.4

−0.2 0

10

20

30

0

Interest rate

10

20

30

0

Terms of trade

0.4

10

Real exchange rate

0.8 1

0.6

0.2

0.4 0.5

0

0.2 0

−0.2

0

−0.2 −0.4 0

10

20

30

0

10

20

30

0

10

20

30

Fig. 3. Responses to identiﬁed government spending shock (baseline speciﬁcation). Notes: solid lines display the median response and the 16 and 84% quantiles. Shaded areas indicate sign restrictions. Horizontal axes: quarters; vertical axis: percent. All variables, except for the real exchange rate and the terms of trade, are expressed in relative terms (U.S. vs. ROW).

indices), the exchange rate response does not merely reﬂect a Balassa–Samuelson effect, but a general increase in the relative price of domestically produced goods. Our ﬁndings are in line with results reported by Corsetti et al. (2008b), Kim and Lee (2008), and Enders and Müller (2009), who, drawing on Galí (1999), use longrestrictions to identify technology shocks.27 Moreover, Corsetti et al. (2009a), identifying demand and productivity shocks on the basis of a set of sign restrictions which differs from ours, also ﬁnd that productivity shocks appreciate U.S. exchange rates.28 A new ﬁnding relative to earlier studies, however, concerns the medium-term dynamics of the real exchange rate. We ﬁnd an

27 Corsetti et al. (2008b) and Kim and Lee (2008) specify their VAR model in relative terms and identify technology shocks by assuming that only these shocks affect relative labor productivity in the long-run. Corsetti et al. (2008b) report that relative technology shocks appreciate the real exchange rate and the terms of trade in the U.S. and Japan. Kim and Lee (2008) ﬁnd an exchange rate appreciation for the U.S., but not for Japan and the Euro area. Enders and Müller (2009) assume instead that only technology shocks affect the level of U.S. labor productivity in the long-run. They also ﬁnd an appreciation of the U.S. terms of trade and the exchange rate. 28 They ﬁnd that demand shocks appreciate exchange rates, too. This result does not conﬂict with our result on government spending shocks, which, in equilibrium, simultaneously affect the supply and demand of domestic goods.

appreciation only for the ﬁrst couple of quarters. Afterwards, the exchange rate start to rise above its pre-shock level and shows a signiﬁcant depreciation for an extended period. This strikes us an interesting ﬁnding, notably because there is no evidence for a reversal of the sign of the terms of trade response.29

3.3. Exchange rate dynamics: further analysis In the following we take up a number of issues to shed further light on our results. First, to give a more systematic account of the uncertainty surrounding the median responses, we follow Scholl and Uhlig (2008) and report in Fig. 5 the posterior joint distribution of the timing and the size of the peak responses of the real exchange rate and the terms of trade. The distribution of peak responses to government spending and technology shocks is displayed in the left and right column, respectively, against the size of the response and the quarter when the peak response occurs. Overall, the distribution of peaks is fairly well behaved, with almost the entire mass of the distribution leaning towards the median response.

29 Corsetti et al. (2009a) also detect some signs of a long-run depreciation in response to productivity shocks, but only at about 35 quarters after impact.

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Government spending

63

Output

0.3

Budget balance

0.5

0.2

0.3

0.4 0.2

0.3

0.1

0.2 0

0.1

0.1

−0.1

0

−0.2

−0.1

0 −0.1

−0.2

−0.3 0

10

20

0

30

Consumption

10

20

0

30

1.5

0.2

1

0.1

0.5

0

0

0

−0.1

−0.2

−0.5

−0.2

0.2

10

20

0

30

Interest rate

10

20

20

30

20

30

Inflation

Investment

0.4

0

10

0

30

Terms of trade

10

Real exchange rate

0.4

0.3

1

0.2

0.2

0 0.1

0.5

−0.2

0

0

−0.4

−0.1

−0.6

−0.2

−0.5

−0.8

−0.3

−1

−1 0

10

20

30

0

10

20

30

0

10

20

30

Fig. 4. Responses to identiﬁed technology shock (baseline speciﬁcation). Notes: see Fig. 3.

Note, however, that the posterior distribution reﬂects both sampling and model uncertainty. In order to gauge the extent of model uncertainty, we rule out sampling uncertainty by holding the coefﬁcients ﬁxed at the OLS point estimates when computing the posterior distribution of impulse response functions. The solid lines in Fig. 6 display the median as well as the 16 and 84% quantiles of the posterior distribution of the responses of the real exchange rate and the terms of trade. While considerable model uncertainty is apparent, the posterior distribution of the responses is tighter relative to the results reported in Figs. 3 and 4. Finally, we note that focusing on the median of the posterior distribution of impulse responses might be problematic, particularly if several structural shocks are identiﬁed. Fry and Pagan (2007) point out that the posterior distribution of impulse responses is a distribution across different identiﬁed models such that the median impulse response functions to two shocks are generated by two different impulse matrices. Consequently, the identiﬁed shocks associated with the median responses are not necessarily orthogonal. To explore this issue further, we generate impulse responses on the basis of a single model, by computing an impulse matrix which implies responses as close to the median responses as possible, as suggested by Fry and Pagan (2007). The responses, displayed by the dashed line in Fig. 6, show that the results obtained under our baseline speciﬁcation are not very sensitive to this adjustment. In particular, we still ﬁnd a short-run appreciation and the medium-run depreciation of the real exchange rate in response to technology shocks.

3.4. Accounting for ﬂuctuations We now turn to a brief analysis of the actual incidence of government spending and technology shocks as suggested by our identiﬁcation scheme. In Fig. 7 we plot four-quarter moving averages of the estimated innovations. In the left panels, the solid lines display the median estimates, while the dashed lines display 16 and 84% quantiles. However, since the median shock series result from different identiﬁed models, they are potentially correlated. We therefore apply once more the procedure suggested by Fry and Pagan (2007) and compute innovations from a single model. The results are displayed in the middle column of Fig. 7. The solid line refers to the median innovations holding the VAR parameters ﬁxed at the OLS point estimates, while the dashed lines display the innovations obtained under the single model. Despite considering four-quarter moving averages, the picture of the entire history of identiﬁed shocks is fairly complex. Yet a few episodes stand out and may be related to familiar narratives concerning important macroeconomic episodes during the last three decades. Focusing on the results obtained under the single model (middle panels, dashed line) and turning ﬁrst to government spending shocks, we observe spikes during the Carter–Reagan military build-up in the early 1980s as well as after 9/11. Regarding technology shocks, strong positive innovations during the late 1990s can be detected, in line with the notion of a distinct productivity driven upturn in the U.S. at that time.

64

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Technology shock Percent of Responses

Percent of Responses

Real exchange rate

Government spending shock

3 2 1 0

0

3 2 1 0 −2 −1

Si 1 ze

Si 2

ze

10 68 2 4 arters u

10 68 s 24 er t r a u

0

Q

Percent of Responses

Percent of Responses

Terms of trade

Q

3 2 1 0 0

4

2

0 −2 −1

Si

Si 1 ze 2

10 68 s 24 rter

Qua

ze

0 1

24

68

10

rs

rte Qua

Fig. 5. The posterior distribution of the peak responses. Notes: posterior joint distribution of quarter and size of the maximal absolute value of the response to a government spending shock (left) and technology shock (right) within the ﬁrst 10 quarters.

The right panels of Fig. 7 plot a historical decomposition of the real exchange rate, i.e., a comparison of the time series predicted by the VAR model assuming that all shocks occurred (solid line) and that either only technology or government spending shocks occurred (dashed line). In both scenarios we assume zero as the starting value of the vector of endogenous variables in order to abstract from initial

conditions. A casual inspection suggests that technology shocks— more than government spending innovations—account for a considerable fraction of actual exchange rate dynamics. A similar picture emerges once we compute a business cycle variance decomposition as in Altig et al. (2005). Speciﬁcally, we compute, on the basis of counterfactual simulations using the single

Fig. 6. Impulse responses of real exchange rate and terms of trade. Notes: VAR coefﬁcients are ﬁxed at their OLS estimates. Solid lines: median response and 16 and 84% quantiles; dashed line: impulse responses implied by a single model as proposed by Fry and Pagan (2007). Horizontal axes: quarters; vertical axes: percent.

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Spending shocks

Innovations

Innovations, single model

Exchange rate decomposotion 15

1

1

0.5

10

0.5

0

5 0

0

−0.5

−0.5

−1

−1

−5 −10

1980 1985 1990 1995 2000 2005 1.5

Technology shocks

65

1

−15 1980 1985 1990 1995 2000 2005

1.5

15

1

10

0.5

5

0.5

0

0 0

−5

−0.5 −0.5

−10

−1

−15

−1 −1.5

1980 1985 1990 1995 2000 2005

1980 1985 1990 1995 2000 2005

1980 1985 1990 1995 2000 2005

1980 1985 1990 1995 2000 2005

Fig. 7. Estimated innovations and historical decomposition of real exchange rate ﬂuctuations. Notes: left panels show four-quarter moving average of estimated innovations (median, 16 and 84% quantiles); middle panels show innovations for VAR parameters ﬁxed at OLS point estimates (solid line) and innovations implied by single model (dashed line); right panels show historical decomposition of real exchange rate predicted by VAR assuming zero as starting value for the vector of endogenous variables: all shocks (solid line) vs. spending or technology shocks only (dashed line).

model, the fraction of the variance of each time series that is accounted for by either government spending or technology shocks. In Table 3, we report the fraction of the variance of the corresponding counterfactual time series relative to the variance of the actual data, after applying the HP-ﬁlter with a smoothing parameter of 1600 to each series (in brackets, we report the corresponding statistics based on unﬁltered data). According to this measure, technology shocks account for 21, 11, and 10% of the short-run ﬂuctuations of consumption, output, and investment, while government spending shocks account for 7, 12, and 12%, respectively. Both shocks have a large impact on the cyclical volatility of inﬂation: 29 and 21% of the variation is due to technology and spending shocks, respectively. Turning to the real exchange rate, we ﬁnd that 18% of its business cycle variance is due to shocks to technology, while government spending shocks account for 9% of ﬂuctuations. Yet the role of both shocks in accounting for ﬂuctuations appears to be much smaller once we consider unﬁltered times series, suggesting that both shocks are of minor importance at median and long-run frequencies. 4. Sensitivity analysis

spending not to respond for four quarters and ﬁnd, in comparison with unanticipated shocks, more persistent and stronger effects of government spending, notably on output and consumption; the latter increasing signiﬁcantly only in response to anticipated shocks. In order to allow for the possibility that spending shocks are anticipated, we impose the sign restrictions reported in the second column of Table 2. In other words, we rely on our model simulations which explicitly allow for implementation lags in government spending innovations of up to two quarters. Recall that the sign restrictions reported in the second column are satisﬁed by anticipated and unanticipated spending shocks, such that we are agnostic as to whether shocks have been anticipated or not. In our view, working with a fully speciﬁed general equilibrium model to derive sign restrictions ensures capturing non-trivial feedback effects of anticipated innovations. For instance, we ﬁnd that government spending, if anticipated to rise exogenously, may nevertheless adjust instantaneously—to the extent that it responds endogenously to the state of the economy. As in our baseline speciﬁcation we identify (possibly anticipated) government spending and technology shocks jointly. In fact, the only difference relative to the baseline speciﬁcation is the set of sign restrictions: we use, in addition to column 3, column 2 of Table 2, rather

4.1. Anticipation of government spending shocks As a result of the institutional features of the budget process, innovations in government spending may become known before they are actually implemented. Blanchard and Perotti (2002) discuss this issue and show how accounting for anticipation requires stronger identiﬁcation assumptions within their framework. They explicitly investigate the possibility that shocks are known one quarter in advance and ﬁnd somewhat stronger output effects.30 Mountford and Uhlig (2009) argue that the sign restriction approach to identiﬁcation is particularly well-suited to address the issue of announcement effects. In order to identify anticipated spending shocks, they restrict government 30

See Mertens and Ravn (2010) for a general treatment of how to account for anticipation effects while employing the Blanchard–Perotti identiﬁcation scheme.

Table 3 Business cycle variance decomposition. Variable

Private consumption Output Investment Government spending Government budget Nominal interest rate Inﬂation Real exchange rate

Business cycle variation due to shocks to Government spending

Technology

0.07 0.12 0.12 0.08 0.06 0.17 0.21 0.09

0.21 [0.08] 0.11 [0.08] 0.10 [0.04] 0.06 [0.04] 0.10 [0.04] 0.17 [0.08] 0.29 [0.26] 0.18 [0.06]

[0.02] [0.05] [0.05] [0.04] [0.03] [0.08] [0.17] [0.03]

Notes: Variance of counterfactual relative to actual time series (after applying HP-ﬁlter with smoothing parameter of 1600 [without ﬁltering in brackets]); counterfactual time series are computed on the basis of VAR model and identiﬁed shocks (single model).

66

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Technology shock

Real exchange rate

Government spending shock 1

1

0.5 0.5

0 −0.5

0

−1 0

10

20

30

0

10

20

30

0

10

20

30

Terms of trade

0.4 0.8

0.2

0.6

0

0.4

−0.2

0.2

−0.4

0

−0.6

−0.2

−0.8 −1 0

10

20

30

Fig. 8. Responses to technology and possibly anticipated government spending shocks; Notes: see Fig. 3.

Real exchange rate

Government spending shock

Monetary policy shock

1

1

1 0.5 0.5

0.5

0 0 −0.5

0

−0.5 −1 0

10

20

30

0.8

Terms of trade

Technology shock

0

10

20

30

0.4

0.8

0.2

0.6

0.6

0

0.4

−0.2

0.2

−0.4

0

−0.6

0

−0.8

−0.2

−0.2 10

20

30

10

20

30

0

10

20

30

0.4 0.2

−1 0

0

−0.4 0

10

20

30

Fig. 9. Responses under three-shock identiﬁcation scheme. Notes: see Fig. 3.

than column 1. Fig. 8 shows the resulting impulse responses of the terms of trade and the real exchange rate. We detect only minor differences in the effects of government spending and technology shocks.31 4.2. Monetary policy shocks Monetary policy shocks may contribute considerably to real exchange rate ﬂuctuations, see, e.g., Clarida and Galí (1994) and 31 We also ﬁnd responses very similar to those obtained under the baseline speciﬁcation, once we impose sign restrictions which are generated by model simulations while allowing for an anticipation horizon of up to four quarters. Results are available on request.

Eichenbaum and Evans (1995). We investigate whether results obtained under our baseline speciﬁcation are sensitive to a modiﬁcation where we explicitly identify monetary policy shocks, in addition to government spending and technology shocks. To do so, we rely on sign restrictions implied by our model simulations and reported in the right column of Table 2.32 We restrict all three structural shocks to be orthogonal to each other. Fig. 9 displays the impulse responses of the real exchange rate and the terms of trade obtained under this three-shock identiﬁcation

32 Note that we restrict the impact response of exchange rates to monetary policy shocks, in line with the predictions of our model.

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

Real exchange rate

67

Terms of trade

Net exports

Spending shock

0.8 1

0.2

0.6 0.4

0.1

0.5 0.2

0 0

0

−0.1

−0.2 −0.5

0

10

20

30

0

10

20

30

Technology shock

10

20

30

0

10

20

30

0.2

0.4 1

0

0.2 0.1

0

0.5

−0.2 0

0

−0.4 −0.6

−0.5

−0.1

−0.8 −0.2

−1

−1 0

10

20

30

0

10

20

30

Fig. 10. Responses of VAR model which includes net exports. Notes: see Fig. 3.

Real exchange rate

Government spending shock

Technology shock 1

1 0.5 0.5

0

0

−0.5 −1

−0.5

0

10

20

30

0

10

20

30

0

10

20

30

0.8

Terms of trade

0.2 0.6

0

0.4

−0.2

0.2

−0.4 −0.6

0

−0.8 −0.2

−1 0

10

20

30

Fig. 11. Responses for different samples. Notes: see Fig. 3; solid line: baseline sample; dashed line: sample starts in 1973Q1; dashed–dotted line: sample starts in 1980Q1.

scheme. We ﬁnd the responses to government spending and technology shocks hardly altered relative to the baseline speciﬁcation. 4.3. Further sensitivity analysis In the following we take up additional complications to assess the robustness of the results obtained under our baseline speciﬁcation. First, we investigate whether our results are sensitive to the inclusion of net exports as an additional variable in the VAR model. Our quantitative business cycle model does not deliver clear-cut prediction for how net exports respond to any of the shocks we seek to identify. We thus leave their response unrestricted and impose the same restrictions as in the baseline speciﬁcation.

Fig. 10 shows that our results are not much affected by the inclusion of net exports. The initial exchange rate appreciation after technology shocks, however, is no longer signiﬁcant, most likely reﬂecting the fact that the number of variables in the VAR model increased, while the number of restrictions did not. The response of the trade balance displays signiﬁcant dynamics only if the terms of trade are included in the VAR.33 Government spending shocks tend to improve U.S. net exports, albeit by a limited amount, see Kim and Roubini (2008) and Corsetti and Müller (2006). For technology shocks, we ﬁnd a hump-shaped decline of net exports, after an initial 33 These are shown in Fig. 10. Results for the real exchange rate speciﬁcation are very similar, but insigniﬁcant.

68

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69

increase; Enders and Müller (2009) report similar adjustment dynamics. Finally, we consider two alternative starting dates for our sample. First, we use data from the end of the Bretton-Woods system onwards, i.e., we set 1973Q1 as a starting date. Second, we use 1980Q1 as a starting date, as U.S. ﬁscal policy transmission arguably changed afterwards, see Perotti (2005). Fig. 11 displays the median responses of the exchange rate and the terms of trade to government spending and technology shocks, contrasting results for the baseline sample (solid line) with those for the earlier (dashed line) and the later (dashed–dotted line) starting period. We ﬁnd that results are fairly robust across sample periods. 5. Conclusion In this paper we establish evidence on the response of international relative prices to government spending and technology shocks. We start from three observations. First, the behavior of the real exchange rate and the terms of trade carries important information regarding the international transmission mechanism. Second, the existing evidence on the behavior of international relative prices in response to technology and government spending shocks conﬂicts with the predictions of international business cycle models under standard calibrations. Third, this evidence is mostly based on estimated VAR models where identiﬁcation is achieved either through short-run or long-run restrictions. We provide new evidence by employing an alternative identiﬁcation scheme based on sign restrictions. In order to generate robust sign restrictions, we simulate a quantitative business cycle for a wide range of parameter values. Moreover, we document that the model does not deliver clear-cut predictions for the sign of the exchange rate response to government spending and technology shocks. To identify these shocks, we thus restrict the responses of several variables, but leave the response of exchange rates unrestricted. Estimating a VAR model on time series data for the U.S. relative to an aggregate of industrialized countries, we ﬁnd that expansionary government spending shocks depreciate the real exchange rate and the terms of trade. Positive technology shocks, in contrast, appreciate the real exchange rate and the terms of trade in the short-run. These results are robust with respect to several variations of our baseline speciﬁcation, such as accounting for monetary policy shocks or anticipation of innovations in government spending. In principle, our empirical results can be rationalized within standard business cycle models. After all, our model simulations illustrate that the exchange rate response to both shocks can go either way, depending on the parameterization. As conventionally calibrated models predict that government spending (technology) shocks appreciate (depreciate) exchange rates, one might interpret our ﬁndings as evidence in support of an alternative calibration, characterized, in particular, by a low trade price elasticity. For this case, Corsetti et al. (2008a) show that strong wealth effects in response to technology shocks raise the demand for domestic goods above supply and thus appreciate the exchange rate. Our simulation results suggest that the reverse holds for government spending shocks. However, a more fundamental reassessment of the international transmission holds considerable promise, too. Results by Ravn et al. (2007) and Corsetti et al. (2009b), for instance, suggest that “deep habits” or “spending reversals” may account for an exchange rate depreciation following an exogenous increase in government spending. Further research into the international transmission mechanism along these lines seems warranted. Acknowledgements We thank Almira Buzaushina, Giancarlo Corsetti, Charles Engel (the Editor), two anonymous referees, and seminar participants at HEI

Geneva, University of Amsterdam, University of Macedonia, Goethe University Frankfurt, University of Bonn, University of Zurich, University of Mannheim, the German Economic Association Meeting 2007, the EEA meeting 2007 and the CEF 2007 conference for very useful comments and suggestions. Part of this paper was written while Müller was visiting EPFL Lausanne. Its hospitality is gratefully acknowledged. The usual disclaimer applies.

References Altig, D., Christiano, L.J., Eichenbaum, M., Lindé, J., 2005. Firm speciﬁc capital, nominal rigidities and the business cycle. NBER Working Paper 11034. Backus, D.K., Kehoe, P.J., Kydland, F.E., 1992. International real business cycles. The Journal of Political Economy 100 (4), 745–775. Backus, D.K., Kehoe, P.J., Kydland, F.E., 1994. Dynamics of the trade balance and the terms of trade: the J-curve? The American Economic Review 84 (1), 84–103. Basu, S. and Kimball, M. 2002, Long-run labor supply and the elasticity of intertemporal substitution of consumption. University of Chicago. Mimeo. Beetsma, R., Giuliodori, M., Klaasen, F., 2008. The effects of public spending shocks on trade balances and budget deﬁcits in the European Union. Journal of the European Economic Association 6 (2–3), 414–423. Bilbiie, F.O., 2009. Non-separable preferences and Frisch labor supply: one solution to a ﬁscal policy puzzle. CERP Discussion Paper 7484. Blanchard, O.J., Perotti, R., 2002. An empirical characterization of the dynamic effects of changes in government spending and taxes on output. Quarterly Journal of Economics 117 (4), 1329–1368. Calvo, G., 1983. Staggered prices in a utility maximizing framework. Journal of Monetary Economics 12, 383–398. Chari, V.V., Kehoe, P.J., McGrattan, E.R., 2002. Can sticky price models generate volatile and persistent real exchange rates? Review of Economic Studies 69, 533–563. Christiano, L.J., Eichenbaum, M., Evans, C., 2005. Nominal rigidities and the dynamic effects of a shock to monetary policy. Journal of Political Economy 113 (1), 1–45. Clarida, R., Galí, J., 1994. Sources of real exchange-rate ﬂuctuations: how important are nominal shocks? Carnegie–Rochester Conference Series on Public Policy 41, 1–56. Clarida, R., Galí, J., Gertler, M., 2000. Monetary policy rules and macroeconomic stability: evidence and some theory. Quarterly Journal of Economics 115, 147–180. Cole, H., Obstfeld, M., 1991. Commodity trade and international risk sharing: how much do ﬁnancial markets matter. Journal of Monetary Economics 28, 3–24. Corsetti, G., Müller, G.J., 2006. Twin deﬁcits: squaring theory, evidence and common sense. Economic Policy 48, 598–638. Corsetti, G., Müller, G.J., 2008. Twin deﬁcits, openness and the business cycle. Journal of the European Economic Association 6 (2–3), 404–413. Corsetti, G., Dedola, L., Leduc, S., 2008a. International risk sharing the transmission of productivity shocks. Review of Economic Studies 75, 443–473. Corsetti, G., Dedola, L., Leduc, S., 2008b. Productivity, external balance and exchange rates: evidence on the transmission mechanism among G7 countries. In: Reichlin, L., West, K.D. (Eds.), NBER International Seminar on Macroeconomics 2006. Corsetti, G., Dedola, L. and Leduc, S. 2009a. The international dimension of productivity and demand shocks in the US economy. European University Institute. Mimeo. Corsetti, G., Meier, A., Müller, G.J., 2009b. Fiscal stimulus with spending reversals. IMF Working paper 09/106. Dhyne, E., Álvarez, L.J., Bihan, H.L., Veronese, G., Dias, D., Hoffmann, J., Jonker, N., Lünnemann, P., Rumler, F., Vilmunen, J., 2006. Price changes in the euro area and the United States: some facts from individual consumer price data. The Journal of Economic Perspectives 20 (2), 171–192. Domeij, D., Flodén, M., 2006. The labor-supply elasticity and borrowing constraints: why estimates are biased. Review of Economic Dynamics 9 (2), 242–262. Eichenbaum, M., Evans, C.L., 1995. Some empirical evidence on the effects of shocks to monetary policy on exchange rates. The Quarterly Journal of Economics 110 (4), 975–1009. Enders, Z., Müller, G.J., 2009. On the international transmission of technology shocks. Journal of International Economics 78, 45–59. Enders, Z., Jung, P., Müller, G.J., 2009. Has the Euro changed the business cycle? Bonn Econ Discussion Papers 06/2009. Engel, C., 1999. Accouting for U.S. real exchange rate changes. Journal of Political Economy 107 (3), 507–538. Erceg, C.J., Guerrieri, L., Gust, C., 2005. Expansionary ﬁscal shocks and the U.S. trade deﬁcit. International Finance 8, 363–397. Fagan, G., Henry, J., Mestre, R., 2001. An area-wide model AWM for the Euro area. ECB Working Paper 42. Farrant, K., Peersman, G., 2006. Is the exchange rate a shock absorber or a source of shocks? New empirical evidence. Journal of Money Credit and Banking 38 (4), 939–962. Faust, J., Rogers, J.H., 2003. Monetary policy's role in exchange rate behavior. Journal of Monetary Economics 50 (7), 1403–1622. Fernández-Villaverde, J., Rubio-Ramírez, J., Sargent, T.J., Watson, M.W., 2007. A, B, C's (and D's) for understanding VARs. The American Economic Review 97 (3), 1021–1026. Fratzscher, M., Straub, R., 2009. Asset prices and current account ﬂuctuations in G7 economies. IMF Staff Papers 56 (3), 633–654. Frenkel, J.A., Razin, A., 1987. The Mundell–Fleming model: a quarter century later. IMF Staff Papers 34, 567–620.

Z. Enders et al. / Journal of International Economics 83 (2011) 53–69 Fry, R., Pagan, A., 2007. Some issues in using sign restrictions for identifying structural VARs. NCER Working Paper 14. Galí, J., 1999. Technology, employment, and the business cycle: do technology shocks explain aggregate ﬂuctuations? The American Economic Review 89, 249–271. Galí, J., Monacelli, T., 2005. Monetary policy and exchange rate volatility in small open economy. Review of Economic Studies 72, 707–734. Galí, J., Perotti, R., 2003. Fiscal policy and monetary integration in Europe. Economic Policy 37, 534–572. Galí, J., López-Salido, J.D., Vallés, J., 2007. Understanding the effects of government spending on consumption. Journal of the European Economic Association 5 (1), 227–270. Gomme, P., Rupert, P., 2007. Theory, measurement and calibration of macroeconomic models. Journal of Monetary Economics 54, 460–497. Heathcote, J., Perri, F., 2002. Financial autarky and international business cycles. Journal of Monetary Economics 49, 601–627. Kim, S., Lee, J., 2008. International macroeconomic ﬂuctuations: a new open economy macroeconomics interpretation. Hong Kong Institute of Monetary Research Working Paper. Kim, S., Roubini, N., 2008. Twin deﬁcit or twin divergence? Fiscal policy, current account, and real exchange rate in the U.S. Journal of International Economics 74 (2), 362–383. Kollmann, R., 2002. Monetary policy rules in the open economy: effects on welfare and business cycles. Journal of Monetary Economics 49, 989–1015. Leeper, E.M., Walker, T.B., Yang, S.-C.S., 2009. Fiscal foresight and information ﬂows. NBER Working Paper 14630. Mertens, K., Ravn, M.O., 2010. Measuring the impact of ﬁscal policy in the face of anticipation: a structural VAR approach. The Economic Journal 120 (544), 393–413. Monacelli, T. and Perotti, R. 2006. Fiscal policy, the trade balance and the real exchange rate: Implications for international risk sharing. Mimeo. Monacelli, T., Perotti, R., 2008. Openness and the sectoral effects of ﬁscal policy. Journal of the European Economic Association 6 (2–3), 395–403. Mountford, A., Uhlig, H., 2009. What are the effects of ﬁscal policy shocks? Journal of Applied Econometrics 24 (6), 960–992.

69

Müller, G.J., 2008. Understanding the dynamic effects of government spending on foreign trade. Journal of International Money and Finance 27 (3), 345–371. Obstfeld, M., Rogoff, K., 1995. Exchange rate dynamics redux. Journal of Political Economy 103, 624–660. Peersman, G., Straub, R., 2009. Technology shocks and robust sign restrictions in a Euro Area SVAR. International Economic Review 50 (3), 727–750. Perotti, R., 2005. Estimating the effects of ﬁscal policy in OECD countries. CEPR Discussion Paper 4842. Perotti, R., 2007. In search of the transmission mechanism of ﬁscal policy. NBER Working paper 13143. Ramey, V.A., 2009. Identifying government spending shocks: it's all in the timing. NBER Working paper 15464. Ramey, V.A., Shapiro, M.D., 1998. Costly capital reallocation and the effects of government spending. Carnegie Rochester Conference on Public Policy 48, 154–194. Ravn, M.O., Schmitt-Grohé, S., Uribe, M., 2007. Explaining the effects of government spending on consumption and the real exchange rate. NBER Working Paper 13328. Rotemberg, J., Woodford, M., 1993. Dynamic general equilibrium models with imperfecly comptetitive product markets. NBER Working Paper 4502. Rotemberg, J., Woodford, M., 1999. The cyclical behavior of prices and goods. In: Taylor, J., Woodford, M. (Eds.), Handbook of Macroeconomics. North Holland, Amsterdam. Rubio-Ramirez, J.F., Waggoner, D.F., Zha, T., 2005. Markov-switching structural vector autoregressions: theory and application. Federal Reserve Bank of Atlanta, Working Paper. Schmitt-Grohé, S., Uribe, M., 2003. Closing small open economy models. Journal of International Economics 61, 163–185. Scholl, A., Uhlig, H., 2008. New evidence on the puzzles: results from agnostic identiﬁcation on monetary policy and exchange rates. Journal of International Economics 76 (1), 1–13. Uhlig, H., 1994. What macroeconomists should know about unit roots: a Bayesian perspective. Econometric Theory 10, 645–671. Uhlig, H., 2005. What are the effects of monetary policy on output? Results from an agnostic identiﬁcation procedure. Journal of Monetary Economics 52 (2), 381–419.

Journal of International Economics 83 (2011) 70–82

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

Sovereign default, private sector creditors, and the IFIs Emine Boz International Monetary Fund, United States

a r t i c l e

i n f o

Article history: Received 21 March 2008 Received in revised form 30 September 2010 Accepted 2 October 2010 Available online 15 October 2010 JEL classiﬁcation: F3 F34 G15

a b s t r a c t The data reveal that emerging market sovereign borrowing from International Financial Institutions (IFIs) is small, intermittent and countercyclical compared to that from private sector creditors. The IFI loan contracts offered to sovereigns differ from the private ones in that they are more enforceable and have conditionality arrangements attached to them. Taking these contractual differences as given, this paper builds a quantitative model of a sovereign borrower and argues that better enforceability of IFI loan contracts is the main institutional feature that explains the size and cyclicality while conditionality accounts for the intermittency of borrowing from the IFIs. © 2010 Elsevier B.V. All rights reserved.

Keywords: Emerging markets Sovereign debt and default IFIs

1. Introduction Why do sovereigns borrow less from International Financial Institutions (IFIs) compared to private sector creditors even though the interest rates charged by the IFIs are signiﬁcantly lower? This paper argues that better enforceability is the main reason why sovereigns borrow little from the IFIs. In other words, when IFI debt contracts are enforceable while those of the private sectors creditors are defaultable, on average, the sovereign chooses to borrow less from the IFI.1 More enforceable debt contracts are less attractive implying that the beneﬁts of introducing default in terms of completing markets is large. In addition to implying lower levels of IFI debt, enforceability also leads to countercyclicality of IFI debt ﬂows. This ﬁnding is related to the debt ﬂow predictions of a canonical consumption savings model or a small open economy real business cycle model (SOE-RBC) under the full commitment assumption. Both of these models with transitory ﬂuctuations around a deterministic trend would predict more borrowing in bad times and less in good times, hence countercyclical debt ﬂows. Similarly, the non-defaultable IFI debt modelled in our setup behaves in a countercyclical fashion. Conditionality, typically attached to IFI debt contracts, appears to be the main contributor to the intermittency of IFI debt. As it is modelled in this paper, conditionality generates a ﬁxed cost-like effect. Only when output is low enough, the sovereign ﬁnds it optimal

to borrow from the IFI. Therefore, IFI debt ﬂows are lumpy, intermittent and very small positive levels of IFI debt are never optimal. In addition to mapping the cyclical properties to the contractual features of IFI debt, this paper contributes to the literature on sovereign debt and default by empirically documenting the cyclical properties of this kind of debt. Taking the International Monetary Fund (IMF) as an example, the data reveal that the cyclical properties of lending by IFIs are in stark contrast to those of private sector creditor lending. The average correlation of IMF debt ﬂows with output for a group of emerging market economies is −0.19 while the same correlation in the case of commercial debt ﬂows is 0.37.2 In addition, the variability of commercial debt ﬂows is about four times as large as that of IMF debt.3 Finally, borrowing from the IMF is intermittent; we ﬁnd the unconditional probability of the use of IMF credit to be around 50%. This also contrasts with commercial debt as all countries in our sample were indebted to private sector creditors at all times. Our framework features incomplete markets and explicitly models a sovereign, private sector creditors, and an IFI. The sovereign cannot commit to repay its debt to private sector creditors and thereby strategically defaults depending on the level of its commercial debt, ofﬁcial debt, and output.4 Punishment for default is exclusion from the

2

See Section 2 for more details on the data and calculations. Throughout the paper, we use the term “commercial debt” to refer to debt to the private sector creditor and “ofﬁcial debt” to refer to debt to the IFI. 4 Reinhart et al. (2003) document that sovereigns have borrowed and defaulted since the 19th century with the maximum number of defaults incurred by Venezuela (9) and Mexico (8). 3

E-mail address: [email protected] 1

In this setting enforceability of IFI contracts refers only to the repayment of the loan and not to the enforcement of conditionality. 0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.10.001

E. Boz / Journal of International Economics 83 (2011) 70–82

71

commercial credit markets and direct output losses. Assuming perfectly competitive, risk-neutral private sector creditors, the interest rate charged by private creditors is determined by endogenous default probabilities. The IFI offers a different contract than the private sector creditors. These differences are based on the institutional framework of IMF loans, more speciﬁcally Stand-By Arrangements (SBAs). First, contracts with the IFI are enforceable while those with commercial creditors are not. This is loosely implied by the IFI having a preferred creditor status and also the fact that the IMF has almost always been repaid, particularly by the emerging market economies that we focus on in this paper. Second, the interest rate associated with IFI lending is assumed to be the sum of the risk-free rate and a charge that increases with the amount borrowed from the IFI. This speciﬁcation for the IFI interest rate captures the surcharges that may apply in the case of SBAs, depending on the amount borrowed. Note that this is signiﬁcantly different from commercial interest rates that depend on the endogenous default probability determined by the “riskiness” of the sovereign. Finally, conditionality associated with IFI debt is accounted for by a higher discount factor in periods when the sovereign is indebted to the IFI. In this setting, a higher discount factor tilts the consumption proﬁle by shifting consumption from the present to the future, thereby lowering debt levels and default probabilities. This can be interpreted as similar to implementing tighter ﬁscal policies that have traditionally been part of IMF conditionality.5 The model featuring the aforementioned types of creditors, when calibrated to Argentina, a representative emerging market economy, performs well in matching several features of the data. Most importantly, the average IFI debt is smaller than commercial debt. Even when conditionality is eliminated, the sovereign on average borrows more commercially. The ratio of ofﬁcial debt to commercial debt goes up from 24% in the baseline calibration to 47% when there is no conditionality attached to ofﬁcial debt. Even without conditionality, the sovereign chooses to borrow from the IFI only about half of what she borrows from the private sector creditors despite the interest rates associated with these loans being lower. This suggests that what matters is the difference in enforceability as opposed to conditionality. The model generates procyclical commercial debt as well as countercyclical IFI debt. This is driven by two intertwined features of the model. First, the fact that IFI debt contracts are enforceable leads to a canonical small open economy real business cycle model (SOERBC) type prediction for the cyclical properties of the sovereign's borrowing from the IFI. With commitment, SOE-RBC models predict more borrowing in bad times and less in good times, hence countercyclical debt ﬂows. The non-defaultable IFI debt modelled in our setup behaves in the countercyclical fashion that canonical SOERBC would predict. On the contrary, defaultable commercial debt is procyclical highlighting the role of lack of commitment to account for the cyclical characteristics of this kind of debt ﬂow. Second, the inelasticity of the IFI interest rates with respect to country characteristics leads the sovereign to ﬁnd it optimal to borrow more commercially at relatively lower rates in good times with lower default probability and without incurring the costs associated with borrowing from the IFI. However, in bad times, in order to avoid the high risk premia charged by private sector creditors, the sovereign reallocates its portfolio by giving more weight to borrowing from the IFI.

The sovereign's default probability increases with the level of ofﬁcial debt as it does with commercial debt. The relationship between the default probability and commercial debt has already been established in the literature. With regards to ofﬁcial debt, there are two effects working in opposite directions. Ceteris paribus, higher IFI debt in the current period implies higher total debt service in the following period. With high debt service in the following period, the sovereign is more likely to default on commercial debt to avoid low levels of consumption. This channel suggests that the higher the IFI debt, the higher the default probability on commercial debt. Second, positive IFI debt forces the sovereign to discount the future less and act more prudently, lowering default probabilities. In this setting, we ﬁnd that the ﬁrst effect is quantitatively larger generating higher country spreads when the sovereign is indebted to the IFI as opposed to when it is not.6 The elasticity of interest rates to ofﬁcial debt is lower than commercial debt. Higher commercial debt increases default probability signiﬁcantly as the beneﬁt of default is the commercial debt that is forgiven. Meanwhile, as mentioned above, higher ofﬁcial debt in the current period increases total debt service in the following period, making the sovereign more likely to default to avoid low consumption levels. This channel is less direct. The sovereign internalizes the effect of its borrowing on interest rates and acknowledges these different elasticities, and therefore, reallocates its portfolio in response to output shocks. Finally, our welfare analysis reveals that in our setup with lack of commitment and incomplete markets, IFI lending reduces welfare by 0.34%. However, a state by state comparison suggests that conditional on the sovereign being highly indebted to the private sector creditors and output being low, the scenario with the IFI yields higher welfare than the one in which IFI lending is non-existent. Note that we assume that the output process remains unchanged in the case of borrowing from the IFI and therefore our welfare calculations yield a pessimistic estimate of the impact of the presence of the IFI. Since there is no investment in our model, the foregone consumption in the current period can reduce future debt but does not improve dynamics of future output. From the empirical side, the evidence regarding the impact of IMF programs on output is mixed.7 However, one-third of a percent improvement in output can be considered a rather small magnitude likely to be achieved with an IMF program. Our paper is closely related to the literature on sovereign debt, particularly emerging market debt that builds on the seminal work of Eaton and Gersovitz (1981). Recently, Arellano (2008) studied the quantitative implications of Eaton and Gersovitz's setup after introducing stochastic output and exclusion from capital markets for a period with stochastic length and direct output losses as punishment for default. This framework has also been extended to include trend growth shocks to output and debt renegotiation by Aguiar and Gopinath (2006) and Yue (2010), respectively. However, this literature overlooks the existence of IFIs in credit markets and the role they might play in the sovereign's debt and default decisions. To our knowledge, the only analysis in this area is conducted by Aguiar and Gopinath (2006), who analyze the implications of a third party bailout limited to a ratio of the defaulted debt. In their extension, this third party is not explicitly modelled, and the bailout takes the form of a grant that is never repaid. In contrast, we explicitly model the third party as an IFI that lends with terms similar to the SBAs of the IMF. Differently from previous literature, our framework incorporates an

5 In good times, countries may not have access to borrowing from an IFI. The IMF, for example, agrees to an SBA only when a country has balance of payments difﬁculties. To account for this, one can assume that IFI lending is available only when output falls below a certain level and the interest rates charged by the private sector creditors exceed a threshold. However, in equilibrium, these constraints would not be binding because the sovereign would never choose to borrow from the IFI if these conditions are not met.

6 Note that we only establish correlation but not causality between spreads on commercial debt and IFI lending. In our setting, higher commercial spreads coupled with higher IFI lending are generated by negative endowment shocks which are the underlying “cause” of aggregate ﬂuctuations. 7 This is also the reason why we do not model in our baseline scenario any potential output improvement due to a program with the IFI.

72

E. Boz / Journal of International Economics 83 (2011) 70–82

enforceable and a non-enforceable contract into a uniﬁed framework and synthesizes the canonical SOE-RBC implications with those of sovereign debt models with limited commitment. An incomplete list of other studies in the sovereign debt literature include Bai and Zhang (2005), Guimaraes (2006), Alfaro and Kanczuk (2007), Arellano and Ramanarayanan (2008), and Mendoza and Yue (2008). Alfaro and Kanczuk (2007) introduce international reserves into the standard strategic default model and allow the sovereign to smooth its consumption during autarky by using these reserves. This extension is similar to ours in the sense that it also brings in an additional instrument that is available during autarky periods. However, reserve holding is different in spirit because it has a lower bound of zero. Applying this lower bound, the authors conclude that it is optimal for the sovereign not to hold any reserves. With the discount rate much higher than the interest rate on reserves (as is usually assumed in the sovereign debt literature), they conclude that zero reserve accumulation is optimal. On the contrary, in our setup, the sovereign is assumed to be a net debtor vis-a-vis the IFI, making it possible to obtain a well-deﬁned stationary distribution for this type of debt. Our modelling of conditionality is reminiscent of the strategies used by Cole et al. (1995), Alfaro and Kanczuk (2005), Hatchondo et al. (forthcoming) and D'Erasmo (2008) to study “patient vs. impatient” or “aligned vs. misaligned” governments using different discount factors. Conditionality implies that the sovereign has to act as a patient/aligned type so long as it is indebted to the IFI. An important difference between our model and those utilized in the aforementioned studies is that in those models, different types of governments alternate in power stochastically while in ours, the sovereign chooses its own type endogenously because the sovereign acknowledges that it has to act as a patient/aligned type as long as it is indebted to the IFI. In addition, being indebted to the IFI is a choice made by the sovereign. In our study, we abstract from any informational role that an IMF program might play and, as a result, from any catalytic effect that such a program might have. There is extensive literature on this topic where the debate is centered on the question of whether IMF lending helps countries avoid or mitigate crises by reducing domestic political costs of adjustment or whether it exacerbates moral hazard. Cottarelli and Giannini (2002) and Ghosh et al. (2002) provide empirical evidence suggesting that IMF assistance leads to little increase in private sector ﬂows, suggesting a minor catalytic role, if any. Mody and Saravia (2006) ﬁnd that whether IMF programs provide a positive signal depends on the likelihood that these programs lead to policy adjustments. In the theoretical literature, Morris and Shin (2006) and Corsetti et al. (2006) analyze the conditions under which the catalytic role exists using a global games framework. Finally, our paper is related to the literature analyzing the impact of IMF programs on countries' macroeconomic conditions. Eichengreen et al. (2006) provide a comprehensive survey of this literature and investigate potential links between sudden stops (severeness and probability) and IMF programs. The authors ﬁnd that IMF programs somewhat reduce the incidence of sudden stops. An incomprehensive list of studies looking at the impact of IMF programs on growth includes Barro and Lee (2005), and Ghosh et al. (2005). Ghosh et al. (2005) ﬁnd that during IMF programs, output growth follows a V-shaped trajectory, i.e., initially it falls but then recovers relatively quickly.8 For the rest of the paper we take the IMF as the representative IFI and focus on the properties of its lending. The paper is organized as follows: Section 2 documents the institutional framework of IMF lending and the cyclical properties of IMF and private sector creditor lending for a set of emerging market economies. Section 3 describes the benchmark two-creditor model. Section 4 elaborates the calibra8 Haque and Khan (2002) provide a survey both from a methodological and results perspective.

tion, reports model statistics comparing them with the data, conducts sensitivity analysis and further investigates the role of IFI lending. Finally, Section 5 concludes. 2. IMF lending 2.1. Institutional framework The IMF lends to member countries with balance of payments difﬁculties to provide them with temporary ﬁnancing.9 Lending of the IMF involves “conditionality,” that is, the borrower needs to follow appropriate policies to resolve the balance of payments problems. Conditionality is aimed at enabling the borrower to repay the IMF on time. The amount borrowed from the IMF determines whether any conditions need to be met as well as the level of the interest rate. Borrowing up to 25% of quota involves essentially no conditionality.10 Any amount beyond the 25% threshold requires the borrower to propose a policy program described in a letter of intent, also called an IMF arrangement or program. IMF lending to provide funding in the face of balance of payments difﬁculties takes place under the General Resources Account which include the SBAs usually lasting 4–6 quarters with repayment in 3–5 years. The interest rate associated with SBAs is the sum of the Special Drawing Rights (SDR) interest rate, margin, burden sharing adjustment, service fee, commitment fee and surcharges. • SDR interest rate is a weighted average of the 3-month U.S. T-bill rate, 3-month UK T-bill rate, Japanese Government 13-week ﬁnancing bills, and 3-month Eurepo rate. • Margin is intended to cover the intermediation costs of the IMF and to help build reserves against credit risk. It is set at the beginning of every ﬁscal year by the management of the IMF. The average from May 1993–September 2008 is 62 bps and it is currently set at 100 bps.11 • A charge for burden sharing is added to cover losses resulting from overdue obligations to the IMF. These charges are refunded as overdue obligations are repaid. The average of the burden sharing adjustment from May 1986–September 2008 is 19 bps, and it is currently set at 2 bps. • A service fee of 50 bps applies to cover part of the intermediation costs and this fee is paid at the time the loan is disbursed. • A commitment fee is collected but refunded as the loans are withdrawn.12 This fee is 25 bps for loans up to the country's quota, and 10 bps for those in excess of the quota. • In the case of SBAs, a surcharge of 100 bps apply to the portion of those loans that are greater than 200% of the country's quota, and 200 bps for the portion greater than 300%. The sum of the SDR interest rate, margin, and burden sharing adjustment is called the adjusted rate of charge which is plotted in Fig. 1 with historical averages reported in Table 1. Both in the table and the ﬁgure, the IMF's adjusted rate of charge is compared with U.S. government bonds with 2, 3, and 5-year maturities since under SBAs, the loans are repaid within 3–5 years. The averages for the IMF's adjusted rate of charge are in line with the yield on 3-year U.S. government bonds. This result is somewhat expected given that the SDR interest rate constitutes a signiﬁcant portion of the adjusted rate of charge and the SDR rate itself is a weighted average of developed 9 There is also lending to low-income countries for poverty reduction. In this paper, we focus on emerging markets that borrow for balance of payments reasons and the types of loans that are available to them. 10 Each member is assigned a quota when it enters the IMF. Quotas are largely based on the size of the country and determine the voting power and borrowing limits. 11 The margin was negative for 1981–May 1993 since the rate of charge was not based on the SDR rate, and the imputed margin for that period is negative. 12 Therefore, commitment pertains to borrowing from the IMF rather than to repayment.

E. Boz / Journal of International Economics 83 (2011) 70–82

16 IMF adjusted rate of charge Yield on 2−yr US govt bonds Yield on 3−yr US govt bonds Yield on 5−yr US govt bonds

percent per year

12

8

4

0 1986

1990

1994

1998

2002

2006

Fig. 1. IMF interest rates vs. U.S. government bond yields.

country T-bill rates with shorter maturities. Similarly, Jeanne and Zettelmeyer (2001) conclude that the IMF's non-concessional lending rates, which include SBAs, are comparable to the risk-free rate. Note that the adjusted rate of charge and the above mentioned fees do not vary across countries. The only loan or country speciﬁc component of interest rates associated with SBAs is surcharges. These are determined by the size of the loan compared to the country's quota but not by the riskiness of a sovereign. That said, there is an indirect channel through which the riskiness of a sovereign might impact the interest rate it faces upon borrowing from the IMF. The riskier the sovereign, or the more negative the output shock, the more it might borrow from the IMF thereby facing higher surcharges and interest rates. Another important feature of IMF lending is conditionality. IMF program-related conditions include macroeconomic and structural measures that fall under the core areas of the IMF's expertise. According to IMF (2002a), these areas are “…macroeconomic stabilization, monetary, ﬁscal and exchange rate policies, including the underlying institutional arrangements and closely related structural measures, and ﬁnancial system issues related to the functioning of both domestic and international ﬁnancial markets.” IMF (2002b) reports that about 34% of the structural measures were related to the ﬁscal sector from 1994 to 1999 followed by the ﬁnancial sector and privatization-related measures that constituted 16 and 14% of total measures, respectively. IMF conditionality is usually implemented using “performance criteria” that include performance targets on net international reserves, monetary policy, ﬁscal policy, and structural variables. The ﬁrst one involves setting a ﬂoor for net international reserves. The second one typically includes a ceiling for net domestic assets of the Central Bank. The ﬁscal policy criteria consist of ceilings on overall balance and domestic ﬁnancing of the deﬁcit. It may also include limits to the size and level of external debt when indebtedness is considered to be an issue. Finally, structural performance criteria span a wide range of measures including ﬁnancial sector reforms, privatization, etc.

Table 1 Average interest rates (in percent).

1986–2008 1986–1996 1997–2008

IMF 1.2

U.S. government bonds

Adjusted rate of charge

2-year

3-year

5-year

5.5 6.8 4.2

5.3 6.6 4.1

5.5 6.8 4.3

5.8 7.1 4.6

Note: Historical averages for the quarterly IMF adjusted rate of charge, market yield on U.S. Treasury securities at 2-year, 3-year and 5-year constant maturity in percentages. Source: IMF Finance Department and The Federal Reserve Board.

73

Inarguably, most of the aforementioned policies are aimed at containing aggregate demand. In our model this translates into a reduction in consumption. In addition, the ﬁscal contraction along with external debt performance criteria can be mapped into a reduction in the commercial debt stock of the sovereign. To understand whether our modelling of conditionality is consistent with these two, we calculate mean consumption and debt to the private sector in periods when the sovereign is indebted to the IFI but is not in default. In this setting it is important to exclude default periods since during default output falls (by assumption) and therefore consumption would also fall. Similarly, since all the debt is forgiven after default, commercial debt stock would also decline dramatically. In order to abstract from these dynamics that are simply due to default per se, we focus on those periods with IFI borrowing but without default. Our model using the calibration elaborated in Section 3 implies that, consistent with the policies prescribed under conditionality, mean consumption during IFI borrowing periods is 6.5% lower than without IFI borrowing. Moreover, the mean commercial debt to GDP ratio falls by 13 percentage points of quarterly GDP. Using the available data on conditionality to conduct a comprehensive quantitative exercise or to quantitatively compare model implications with data is problematic because even when one focuses on the conditionalities related to ﬁscal policy, there is a wide variety of performance criteria and these criteria are often applied to different levels of government. For example, these criteria are applied to the central government budget in some countries, and to the general government in others. The aggregate that is used in most cases is the overall balance but there is a variation across countries in this regard as well with the use of extrabudgetary funds. This kind of heterogeneity makes it difﬁcult to present meaningful cross country evidence on the quantity of adjustment that IMF programs involve. 2.2. Cyclical properties Although cyclical properties of commercial debt ﬂows have been extensively studied in the literature, those of IMF lending have remained largely unexplored. In this section, we examine these properties for a set of emerging market economies and document our ﬁndings in Tables 3, 4, and 5. We choose the countries in our sample based on quarterly GDP data availability and on whether the country had an IMF program in the last two decades. Table 2 reports the main business cycle statistics associated with commercial debt for comparison. Table 2 establishes that commercial debt ﬂows are procyclical, and highly variable. In addition, on average most sovereigns in our sample

Table 2 Data moments: private sector creditor lending. Annual

Argentina Brazil Ecuador Indonesia Mexico Peru Philippines Thailand Turkey Mean

Period

ρ(Δd/y, y)

σ(Δd/y)

E[d/y] (%)

1971–2005 1971–2005 1971–2005 1971–2005 1971–2005 1971–2005 1971–2005 1971–2005 1971–2005

0.32 0.22 0.33 0.43 0.25 0.14 0.40 0.62 0.62 0.37

2.87 1.58 4.26 2.58 2.67 2.78 3.66 10.10 2.94 3.71

20.63 13.27 27.63 9.81 19.17 14.36 14.30 10.47 12.35 15.78

Note: Correlation of commercial debt ﬂows with real GDP, its standard deviation, and mean stock of commercial debt. Source: Commercial debt data is from World Bank's Global Development Finance Database and it includes bonds, commercial banks, and other debt to private creditors. Real GDP data are GDP volume index from IFS.

74

E. Boz / Journal of International Economics 83 (2011) 70–82

Table 3 Data moments: IMF lending.

Table 4 Spreads and use of IMF credit.

Annual

Argentina Brazil Ecuador Indonesia Mexico Peru Philippines Thailand Turkey Mean

Period

ρ(Δd⁎/y, y)

σ(Δd⁎/y)

E[d⁎/y] (%)

1971–2006 1971–2006 1971–2006 1971–2006 1971–2006 1971–2006 1971–2006 1971–2006 1971–2006

− 0.30 − 0.29 − 0.16 − 0.20 − 0.38 − 0.01 − 0.10 0.26 − 0.54 − 0.19

1.15 0.78 0.55 1.21 0.81 0.44 0.49 0.53 1.45 0.82

2.40 0.78 1.07 1.45 1.12 1.67 2.17 0.95 1.89 1.50

Note: Correlation of IMF debt ﬂows with real GDP, its standard deviation, and mean stock of IMF debt. Source: IFS. IMF debt data is the use of Fund credit: GRA and real GDP data are GDP volume index.

appear to be highly indebted to private sector creditors.13 Their standard deviations and correlations with the corresponding country's real GDP are reported in the third and fourth columns of the Table, and the ﬁfth column documents the mean commercial debt stock as a share of annual GDP. On average, the correlation of net lending by private sector creditors with real output is 0.37. This correlation is positive for all of the countries in the sample, lying between 0.14 (Peru) and 0.62 (Thailand and Turkey). The average standard deviation of this type of lending is 3.71 with Thailand being an outlier with a standard deviation of 10.10%. Finally, mean commercial debt stock is around 16% of annual GDP. Table 3 reveals that, IMF lending is countercyclical and has lower mean and variability than commercial lending. The countercyclicality of this type of lending with a correlation of − 0.19 stands in contrast to lending by private sector creditors, which has a correlation of 0.37 with output.14 Countercyclicality holds for all countries except Thailand (0.26). The standard deviation of IMF debt ﬂows reported in the third column of the same table suggests that IMF debt ﬂows are remarkably less variable than commercial debt ﬂows (0.82 vs. 3.71). Comparing the debt ratios, the mean debt ratio to the IMF is lower than that to the private sector creditors (15.8 vs. 1.5%). Therefore, on average, IMF debt is around one tenth of private sector creditor debt. These results are in line with those of Ratha (2001) who ﬁnds that during 1996– 1998, private ﬂows to developing countries were 10–12 times multilateral ﬂows. Table 4 reports statistics on the relationship between the use of IMF resources and country spreads based on J.P. Morgan's EMBI Global spread data from Bloomberg. Since the ﬁrst observation of spread data varies across countries, we report in the second column of the table, the periods considered in the calculations. The following two columns compare the average EMBI spreads for periods when there was positive use of IMF resources and those when the country was not using Fund resources. The number of observations is reported in parenthesis. Note that of the available sample of EMBI data, only Mexico and Thailand had a signiﬁcant number of periods with and without use of Fund resources. For both of these countries spreads in periods with use of Fund credit are about three times the spreads in periods without Fund credit. All other countries were indebted to the IMF during a signiﬁcant portion of the sample period. We ﬁnd average spreads to be 617 (199) bps when Mexico (Thailand) was using Fund

13 Net debt ﬂows by private sector creditors data reported in Table 2 are annual and are from the World Bank's Global Development Finance Database (includes bonds, commercial banks, other debt to private creditors). Real GDP series are also from IFS. They are logged and HP ﬁltered with smoothing parameter 100. 14 The ﬁnding, that in bust periods a country is more likely to have an IMF program, is not to say that one causes the other. Correlation does not imply causality.

Argentina Brazil Ecuador Indonesia Mexico Peru Philippines Thailand Turkey Mean

Period

E[s|d⁎ N 0]

E[s|d⁎ = 0]

ρ(Id⁎ N 0, s)

1994Q1–2006Q4 1994Q2–2006Q4 1998Q1–2006Q4 2004Q2–2006Q4 1994Q1–2006Q4 1997Q4–2006Q4 1994Q1–2006Q4 1998Q1–2006Q1 2000Q1–2006Q4

2143 (48) 826 (46) 1280 (34) 275 (10) 618 (26) 460 (37) 405 (51) 199 (22) 516 (28) 860

343 (4) 257 (5) 3544 (2) 183 (1) 242 (26) – (0) 191 (1) 61 (11) – (0) 342

0.22 0.47 – – 0.67 – – 0.57 – 0.48

Note: Mean bond spreads in periods with and without use of IMF credit and the correlation of IMF debt dummy with spreads. Source: Spreads are based on J.P. Morgan's EMBI Global spread data from Bloomberg.

resources, while the average spreads were 242 (61) bps when it was not. Suspecting that the spread data might have outliers during crises, we also calculate median spreads and ﬁnd these to be 488 (163) bps in periods when Mexico (Thailand) was indebted to the IMF vs. 214 (58) bps when it was not. Finally, we compute the correlation between spreads and a dummy variable that is set equal to 1 when the country is indebted to the IMF. We ﬁnd this correlation is positive for all countries in our sample with an average of 0.48.15 These results on the relationship between spreads and IMF borrowing are roughly in line with those of Mody and Saravia (2006). Mody and Saravia (2006) ﬁnd that spreads on bonds issued during an IMF program are higher. The authors collect data on launch spreads for more than 3000 emerging market and developing country bonds and separate them based on whether they were issued during an IMF program or not. Their calculations reveal that the average spreads during IMF programs were 406 while they were 223 in the absence of programs. IMF borrowing is intermittent in that periods with borrowing are nested within periods without any borrowing from the IMF. This kind of pattern manifests itself as a higher standard deviation of net IMF ﬂows than it would have been if borrowing from the IMF was more continuous. In addition to the standard deviations already reported in Table 3, we also calculate the probability of positive use of IMF resources from its GRA account (Pr[d N 0]) to get a sense of the frequency of IMF borrowing. Table 5 documents that the unconditional probability of having a program with the IMF is 46% for 1945– 2007 and 55% for 1970–2007 when we expand our sample to 17 emerging market economies. The rationale for expanding our sample for this exercise is to get a more unbiased picture because one of our selection criteria for the initial sample was to include countries with signiﬁcant IMF borrowing in the last several decades. So by construction, the probability of the use of IMF credit would be biased upwards. The additional countries included in the calculation of the mean reported in the last row of Table 5 are Chile, Colombia, Egypt, India, Korea, Malaysia, Singapore, and Venezuela. Note that these probabilities are higher for our original sample with 9 countries.16 Table 5 reveals that countries are more likely to be indebted to the IMF when output is below trend. The second and third columns of Table 5 report the unconditional probabilities and probabilities

15 Averages reported in the last row for E[s|d* N 0] and E[s|d* = 0] are weighted by the number of observations. The average for ρ(Id * N 0, s) includes only those countries with at least four observations without use of Fund credit (Argentina, Brazil, Mexico, and Thailand). 16 Barro and Lee (2005) calculate the fraction of time that a country had an IMF program (SBA or EEF) to be 0.185 using data for 130 countries from 1975 to 1999. They also calculate IMF loan approval frequency to be 0.364. Approval frequency is calculated by assigning ‘1’ if the country had a loan approved within a 5-year interval. Our numbers differ from theirs mainly because we focus only on large emerging market economies and also our dummy is based on quarterly data rather than 5-year intervals.

E. Boz / Journal of International Economics 83 (2011) 70–82 Table 5 Probability of use of IMF credit.

Argentina Brazil Ecuador Indonesia Mexico Peru Philippines Thailand Turkey Mean Mean w/8 other EMEs

1945–2007

1970–2007

Pr(d⁎ N 0)

Pr(d⁎ N 0)

Pr(d⁎ N 0|y)

0.62 0.60 0.53 0.54 0.33 0.59 0.73 0.74 0.65 0.49 0.46

0.79 0.62 0.70 0.60 0.57 0.93 0.99 0.57 0.81 0.73 0.55

0.32 0.32 0.38 0.49 0.32 0.50 0.53 0.35 0.41 0.40 0.30

Note: The additional countries included in the last row are Chile, Colombia, Egypt, India, Korea, Malaysia, Singapore, and Venezuela. Source: IFS.

conditional on output being below trend, respectively.17 Subtracting the third column from the second yields probabilities conditional on output being above trend. On average, the probability of having an IMF program when output is low is greater (30%) than when output is high (25%). This holds for all countries in our sample except Argentina. Note that the probability of borrowing from the IMF is signiﬁcant even when output is above trend. One potential reason for this could be that most emerging markets recover relatively quickly after a crises. If a country gets an IMF loan during crisis and output bounces back quickly before the repayments to the IMF are complete then the country can be considered a user of IMF resources while its output would have already recovered. Another difference between IFI debt and commercial debt pertains to repayment history. According to Reinhart et al. (2003), all countries in our sample have defaulted at least once on their external commercial debt with the exception of Thailand. In contrast, essentially all IMF loans to major emerging market economies have always been repaid. Zettelmeyer (2005) writes: “In the past, the IMF has virtually always been repaid, but this leaves the possibility that currently open lending relationships may eventually result in arrears or debt forgiveness.” Jeanne and Zettelmeyer (2001) ﬁnd that “most open lending relationships with emerging market countries are statistically similar to past lending cycles that eventually ended in repayment while the very long open lending cycles of many poor countries statistically ‘look’ like they might continue forever.” Similarly, Eichengreen (2003) and Saravia (2004) argue that the IMF loans are typically repaid. Only during the debt crises of the 1980s was there an increase in non-repayment cases. According to Zettelmeyer (2005), there were thirteen countries in protracted arrears to the IMF.18 However, most of these countries have repaid, and in particular the only emerging market among these countries, Peru, has repaid. To summarize, this section establishes that: • The IMF lending rate under SBAs can be approximated by the sum of the risk-free rate and a charge that increases with the amount borrowed. • Lending by private sector creditors is procyclical whereas IMF lending is countercyclical. • Debt to the private sector creditors is about four times more variable than that to the IMF.

17 We restrict our analysis to a shorter sample in these two columns because real GDP series become available for all countries starting in 1967. 18 Cambodia, Guyana, Haiti, Honduras, Liberia, Panama, Peru, Sierra Leone, Somalia, Sudan, Vietnam, Zaire, and Zambia.

75

• The average debt stock to the IMF is about one tenth of the debt to private sector creditors. • Country spreads are higher in periods when countries are indebted to the IMF. • The unconditional probability of the use of IMF resources is about 50% for emerging markets. • The IMF has almost always been repaid, particularly by emerging market economies. 3. Model The model features competitive, risk-neutral private sector creditors, an IFI, and a sovereign that maximizes the utility of its households. The model mainly builds on the work of Eaton and Gersovitz (1981) on sovereign debt with potential repudiation. The distinguishing and appealing feature of this framework is the lack of commitment of the sovereign to repay its debt to the private sector creditors. Due to this commitment problem, the sovereign may optimally default on its debt to this type of creditors. Therefore, this model analyzes a “willingness to pay” scenario rather than “ability to pay.” Every period the sovereign is in one of two states: default or nondefault vis-a-vis the private sector creditors, D, N. Starting in the nondefault state, the sovereign has access to borrowing commercially. In these non-default states, the sovereign observes its output level, makes a decision on whether to default or not, and also whether to borrow from the IFI. If it chooses to default, it does not repay its existing debt, but as a punishment, it loses access to commercial borrowing for a period with exogenously and stochastically determined length. If it does not default, however, it repays its debt obligations in full, chooses how much to borrow commercially and from the IFI, and maintains the option to default in the following period. During exclusion from commercial borrowing, the sovereign only chooses whether to borrow from the IFI based on its remaining output after incurring losses due to default. The IFI lends to the sovereign at the risk-free rate plus a debt elastic component intended to capture the surcharges of the IMF.19 The sovereign does not face a commitment problem in repaying its debt to the IFI because of the preferred creditor status of the IMF and also the observation that major emerging market economies have almost always repaid their debt to the IMF. There is only one IFI with a nonproﬁt role as an international ﬁnancial institution in this context. In our setup, the existence of the IFI provides the sovereign with an additional instrument to smooth consumption that is available even during exclusion from commercial credit markets. Borrowing from the IFI is preferable in times when the default probability on private sector debt is high (and as a result the interest rate is high) or when the sovereign is in default and does not have access to commercial credit markets at all. The sovereign makes its default decision in the following fashion: n N o D V d; d⁎ ; y = max V d⁎ ; y ; V d; d⁎ ; y D

ð1Þ

where D is the binary variable that represents the decision to default (D = 1) or not (D = 0), y is output as explained below, and d and d* denote the debt stock to private sector creditors and the IFI, respectively. Note that the value of default, V D, depends on output and IFI debt because the level of commercial debt becomes irrelevant when it is not repaid. However, the value of non-default, V N, and the decision on whether to default or not, D, depend on the level of debt to IFI and commercial debt as well as output. Default, D = 1, is chosen if

19 As discussed in Section 2, the rate applied by the IMF includes a surcharge determined by the amount borrowed.

76

E. Boz / Journal of International Economics 83 (2011) 70–82

and only if V D N V N. For state (d⁎, y) the value of default can be expressed as: V

D ⁎ d ;y

( ) h i D = max uðcÞ + βðd⁎ ′Þ ∑ θV ð0; d⁎ ′; y′Þ + ð1−θÞV ðd⁎ ′; y′Þ πðy′jyÞ d⁎ ′

y′

ð2Þ 1−σ

c , c is consumption, and where u is CRRA per-period utility, uðcÞ = 1−σ β(d⁎′) is the discount factor elaborated further below.20 An implicit assumption in this formulation is that once the country regains access to commercial borrowing, all of its commercial debt is forgiven and it starts with d = 0. θ is the exogenously determined probability of entry to the credit market after default; therefore, it gives the average duration of exclusion once the sovereign defaults. The budget constraint during exclusion is

y

def

= c + d⁎ −q⁎ d⁎′

ð3Þ

where q⁎ is the price of IFI debt which can be written as: q⁎ ðd⁎′Þ =

1 : 1 + r + ϕðd⁎ ′Þ

ð4Þ

Note that this price depends only on d*′ and not on d′ or y. Output is exogenous and is characterized by an AR(1) process: z

ð5Þ

y=e

where zt = ρzt − 1 + t and ∼ N(0, σ2z ). We approximate this normal process with a Markov chain with S states and transition probability matrix π. π(yj|yi) denotes the probability of transiting from state i to j. The output process is truncated during default as in Arellano (2008) in order to bring the default probability implied by the model in line with those in the data. We assume that the direct output cost of default increases with output. In other words, if a country defaults when its output is high, the output loss due to the default decision would be higher as opposed to a sovereign that defaults when its output is low. Mathematically: ( y

def

=

y;

if y b yˆ

ˆ if y N y: ˆ y;

In non-default periods, the sovereign decides whether or not to default, and how to allocate its ﬁnancing needs among the private sector creditors and the IFI. The value of non-default in this setup can be written as: V

N

( ) d; d⁎ ; y = max uðcÞ + β d⁎′ ∑ V d′; d⁎′; y πðy′ jyÞ d′ ;d⁎′

y′

ð7Þ

As mentioned previously, the risk-neutral and perfectly competitive private sector creditors lend at a rate that is mainly determined by the default probability of the sovereign. The commercial bond price schedule is given by: 1−E λ d′; d⁎ ′; y q d′; d⁎ ′; y = 1+r

ð8Þ

where λ(d′, d⁎′, y) = Pr[V D(d⁎, y) N V N(d, d⁎, y)] captures the default probability and the spread is deﬁned as s = 1/q − 1 − r. 20

Throughout the paper, primes refer to the following period.

σ r θ ρ σz yˆ k βL βH

2 1% 10% 0.91 1.92% 0.947 0.175 0.97000 0.97008

Literature Literature Literature Estimated from data Estimated from data Calibration Calibration Calibration Calibration

Note: Target moments that determine yˆ, k, βL, and βH are the number of defaults, mean public debt to commercial debt ratio (E[d⁎]A/E[d]A), the standard deviation of IFI ﬂows (σ(Δd⁎/y)), and mean IFI spread (E[s⁎]).

In order to capture the conditionality of IMF loans, and also to account for the regularity that sovereigns do not borrow from the IMF every period as documented in Table 5, we assume that the sovereign has to switch to a higher discount factor (that is, a lower discount rate) as long as it is indebted to the IFI. Models of sovereign default have traditionally calibrated discount rates so that they are signiﬁcantly greater than the risk-free interest rate. In this fashion, more realistic debt ratios and default probabilities are achieved because, with a low discount factor, the sovereign has a strong desire to consume today and is willing to hold larger amounts of debt in the long run. By assuming that borrowing from the IFI is associated with a higher discount factor, we are building “more prudent” policies that come with conditionality. This can also be interpreted as a reduction in consumption (public consumption in particular) that typically is part of conditionality. More formally: βðd⁎′Þ =

βH ; βL ;

if d⁎′ N 0 if d⁎′ = 0:

The recursive equilibrium of this economy is characterized by price schedules for commercial debt, q(d′,d ′,y) and ofﬁcial debt, q⁎(d⁎′), debt and consumption allocations, d′(d,d⁎,y), d⁎′(d,d⁎,y), and c(d,d⁎,y) such that: 1. The sovereign maximizes its utility subject to the budget constraint taking the price schedules as given. 2. Given the default probabilities implied by the debt position and output of the sovereign, private sector creditors lend at price q(d′,d⁎′,y) and break even in expected value while the IFI lends at q⁎(d⁎′) to the sovereign. 3. The goods market clears. 4. Quantitative analysis 4.1. Calibration and data

ð6Þ

subject to the budget constraint: c = y−d + qd′−d⁎ + q⁎ d⁎′:

Table 6 Parameters.

We have three sets of parameters, those borrowed from the sovereign debt literature, those estimated from the data and those calibrated to match certain model statistics to the data. All parameters are reported in Table 6. The risk aversion coefﬁcient and risk-free rate are standard both in the sovereign debt and small open economy business cycle literatures which are set to 2 and 1%, respectively. Following Arellano (2008) and Aguiar and Gopinath (2006), we set the probability of redemption to 10%, implying an average exclusion of 10 periods (2.5 years with quarterly calibration). We estimate the parameters governing the output process using quarterly Argentine GDP data for the two decades before its default (1980Q1–2000Q4) obtained from Neumeyer and Perri (2005).21 The 21 Similarly, we report model statistics after excluding default episodes to make them comparable with the data. However, for those statistics regarding borrowing from the IFI, we include the default periods both in the data and in the model. This is to capture IFI borrowing dynamics during default as well.

E. Boz / Journal of International Economics 83 (2011) 70–82 Table 7 Business cycle statistics.

σ(y) σ(c)/σ(y) σ(TB/y) σ(Δd/y)A σ(Δd⁎/y) σ(s)A ρ(Δd/y, y)A ρ(Δd⁎/y, y) ρ(s, y) ρ(c, y) ρ(TB/y, y) ρ(Id⁎ N 0, s) E[d]A E[d⁎] E[d⁎]A/E[d]A E[s] E[s⁎] No. of defaults Pr(d⁎ N 0)

77

1 Data

Model

4.32 1.17 1.44 3.14 3.26 3.17 0.29 − 0.15 − 0.59 0.98 − 0.89 0.22 25.99 17.08 − 7.08 0.46 67.2 0.62

4.39 1.21 1.83 1.84 3.41 4.50 0.28 − 0.08 − 0.34 0.93 − 0.29 0.25 4.64 3.09 0.24 2.75 0.49 64.6 0.34

Note: Series marked with superscript ‘A’ are annual.

GDP series is deseasonalized, logged and HP ﬁltered with smoothing parameter 1600. With our notation of y = ez where zt = ρzt − 1 + t and ∼ N(0, σ2z ), we estimate an AR(1) process and ﬁnd ρ = 0.91 and σz = 0.019. Using these AR(1) parameters, we estimate a Markov process with nine states using Tauchen's algorithm. We assume ϕ(d⁎)=kd⁎ for simplicity. As a result, we have four ˆ k, βL, and βH. These parameters remaining that need to be calibrated: y, parameters are calibrated to match four model moments: number of defaults, mean public debt to commercial debt ratio (E[d⁎]A/E[d]A), the standard deviation of IFI ﬂows (σ(Δd⁎/y)), and mean IFI spread (E[s⁎]). The values obtained as a result of this calibration exercise are reported in Table 6. Argentine business cycle statistics are reported in the ﬁrst column of Table 7. We use the same series on private sector creditor debt and IMF lending as in Section 2 except that we utilize quarterly IMF lending data available in International Financial Statistics as our model calibration is quarterly. Since private sector creditor debt data is annual, simulated data related to borrowing from private sector creditors are also annualized to facilitate comparison. The number of defaults is calculated based on Reinhart et al. (2003) who report that Argentina defaulted four times in the period 1824–1999. Adding the most recent default episode in 2001, we have ﬁve defaults in 186 years which translates into 67.2 defaults per 10,000 quarters.

4.2. Findings Existing sovereign debt models with only private sector creditors establish that a high level of debt strengthens the incentive to default because the gain from default is the amount of debt that is forgiven.22 Conversely, a high output realization strengthens repayment incentives as the sovereign can afford to repay without facing low levels of consumption. As a consequence, these models imply countercyclical interest rates in line with the emerging markets data. This prediction of countercyclical interest rates is in fact one of the distinguishing features of the last generation of sovereign debt models with incomplete markets from those under complete markets that imply procyclical interest rates.23 These features are present in our twocreditor setup as well as those that arise due to the introduction of the IFI discussed below.

22 See Arellano (2008), and Aguiar and Gopinath (2006) for a discussion of the equilibrium, solution and implications of models with one type of creditor. 23 See Kehoe and Levine (1993) and Alvarez and Jermann (2000).

High y

0.8

Low d*’ High d*’ Low d*’ High d*’

0.6

0.4 Low y

0.2

0 0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

d’ 1 q*(d*’)

0.98

0.96

0.94

0.92

0.9

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

d*’ Fig. 2. Debt price schedules, q(d′, d⁎′, y) and q⁎(d⁎′).

The default probability on commercial debt increases with the level of IFI debt although not as fast as it does with commercial debt. This is evident in the commercial debt price schedule plotted in the top panel of Fig. 2. This pricing schedule is steep suggesting that small increases in d′ increase interest rates signiﬁcantly; however, the interest rate difference between high and low levels of d⁎′ is not large. The steepness is intuitive given that commercial debt is directly linked to the beneﬁt of default. For the case of IFI debt, there are two effects working in opposite directions. First, since IFI debt is not defaultable, a high level of it implies a higher IFI debt service in the following period leaving the sovereign with less resources to repay commercial debt. This channel suggests that the higher the IFI debt, the higher the default probability on commercial debt. Second, positive IFI debt forces the sovereign to discount the future less and act more prudently lowering default probabilities. As suggested by Fig. 2, ceteris paribus the ﬁrst effect is quantitatively larger leading to commercial interest rates that increase with IFI debt in equilibrium. An implication of Fig. 2 is that a shift in sovereign's portfolio towards IFI debt in response to a negative output shock does not sufﬁce to reduce spreads.24 As mentioned above, any type of debt, ofﬁcial or commercial, leads to higher interest rates. However, the interest rates are much more elastic with respect to commercial debt than IFI debt. Given a certain level of total debt, a portfolio with only commercial debt leads to higher spreads than one that includes any IFI debt. 24 This is not to say that the existence of the IFI leads to higher spreads. We conjecture that spreads would have been even higher in absence of IFI borrowing and portfolio reallocation of the sovereign.

78

E. Boz / Journal of International Economics 83 (2011) 70–82

The model performs well in matching the non-target moments. Those moments related to IFI borrowing that are not used in the calibration exercise are the correlation of IFI debt ﬂows with output, ρ(Δd⁎/y, y), the correlation of IFI debt dummy with spreads, ρ(Id⁎ N 0, s), and the frequency of borrowing from the IFI, Pr(d⁎ N 0). In addition to the model doing a good job in matching these statistics (perhaps with the exception of the frequency of IFI borrowing that appear to be lower in the model than compared to the data), it matches well those statistics related to commercial debt ﬂows. The weak countercyclicality of IFI debt is due to the decoupling of the interest rate charged by the IFI from default probability, which is an implication of enforceability of IFI debt contracts. Countercyclical debt ﬂows are a prediction of canonical SOE-RBC models that assume enforceable debt contracts. In those models, as a result of the consumption smoothing motive, agents choose to save in good times and borrow in bad times, hence countercyclical ﬂows. Our framework in some sense synthesizes these strands of models by introducing enforceable and non-enforceable contracts into a uniﬁed framework. Our framework that incorporates both enforceable and non-enforceable contracts nicely preserves the cyclical properties of each type of debt ﬂow established in the literature. The sovereign makes debt allocation decisions in non-default periods by comparing the cost of borrowing commercially and the cost of IFI debt. The cost of borrowing commercially increases when the economy is hit by a negative output shock. However, the cost associated with IFI debt depends on the realizations of these shocks only indirectly through their impact on d⁎′. As a result, the sovereign chooses to hold less commercial and more ofﬁcial debt in its portfolio in bad times, (ρ(Δd/y, y) = 0.28, ρ(Δd⁎/y, y) = − 0.08). The model performs particularly well in matching the correlation between the Fund credit use dummy and spreads (0.22 in the data vs. 0.25 in the model). In addition, in the model, spreads in periods with positive IFI debt are 2.46 times those with no IFI debt. This is in line with the data documented in Section 2. For Mexico and Thailand, this ratio is 2.55 and 3.25, respectively. For Argentina, E[s|d⁎ N 0] is as high as 2100 bps, which is driven by a few outliers in the data. For this country, the ratio of median spreads conditional on being and not being indebted to the IMF is 2.32, remarkably close to the result generated by the model. The intermittency of borrowing from the IFI is evident in Pr(d⁎ N 0) in Table 7. Introduction of two different discount factors generates a ﬁxed cost-like effect making it possible for the model to account for intermittent IFI borrowing. When output is high, the sovereign prefers d⁎ =0 keeping the discount factor low. However, when output is low, the sovereign chooses relatively higher values of d⁎. Therefore, small but positive values for d⁎ are almost never optimal. These dynamics allow us to match the lumpiness and probability of use of Fund credit in the data. To ensure that these ﬁxed cost-like effects are in fact due to the introduction of two different discount factors, we strip out the other ingredients of the model and analyze a simple small open economy consumption–savings setting explained in the Appendix. The aforementioned effect is present in that simple model, conﬁrming our intuition. The sovereign in our model borrows from the IFI with a 34% probability. This is fairly close the average borrowing probability of 46% reported in Table 4. In terms of spreads, we used the mean IFI spread, E[s⁎], as a target moment. Mean commercial debt spread, E[s], implied by our model is 2.75, substantially higher than the mean IFI spread. In the data, mean commercial debt spreads are as high as 7.08%. Unfortunately, the sovereign default models have difﬁculty in generating the high level of spreads that we see in the data, mainly because the spreads are determined in equilibrium by risk-neutral pricing using the default probabilities. (See Arellano (2008) for a discussion.) Our model does not provide an improvement in this regard. Similarly, the introduction of an IFI does not help reduce the discrepancy between the level of commercial debt in the model and in the data, another shortcoming established in the literature.

Table 8 IFI debt during high and low spreads. E[d⁎] s

Data

Model

b 25 pct N 75 pct D=1

7.83 8.23 31.87

2.39 2.76 26.74

To further compare model implications with the data, we analyze the behavior of IFI debt in high and low spread periods. Table 8 reports average IFI debt when spreads are above the 75th percentile, below the 25th percentile, and when the sovereign is in default. This exercise allows us to assess model performance in yet another dimension not captured by the business cycle statistics. The model generates a mean IFI debt stock of 2.76 (2.39) percent when spreads are high (low). These are lower than those implied by the data, 7.83 (8.23).25 However, note that the difference in mean IFI debt between high and low spreads is close to the data. Average IFI debt during default is 26.74% in the model, remarkably close to that in the data (31.87).26 Finally, we conduct a similar exercise looking at the mean IFI debt during booms and busts deﬁned as periods when endowment falls at least one standard deviation and periods when it exceeds its mean by at least one standard deviation. As reported in Table 9, the model matches the qualitative features of the data by generating a mean IFI debt during busts (5.50) that is higher than that during booms (0.06). In the data, these mean IFI debts are higher than those implied by the model with 9.23% during busts and 7.43% during booms. 4.3. Sensitivity analysis Our sensitivity analysis focuses on the parameters that govern the cost of IFI borrowing, which we could not estimate from the data. In Table 10, we present ﬁve sets of results with different levels of conditionality and values for k, the parameter governing the elasticity of the IFI interest rate with respect to the level of IFI debt. In this section, we compare the ﬁrst four scenarios to the baseline and ﬁnd that our main results (weak countercyclicality of IFI debt ﬂows and higher spreads when indebted to the IFI) are by and large robust. The ﬁfth scenario is discussed in the next section. The ﬁrst column of Table 10 reproduces the results with baseline parametrization to facilitate comparison. The second column reports those with no conditionality, that is βH = βL = 0.97. Without conditionality, the average ofﬁcial debt holdings increase from 3.09 to 6.02 with frequency of d⁎ N 0 increasing from 34% to 98%. Mean IFI debt spreads go up proportionally to E[d⁎]. The other statistics do not change signiﬁcantly except ρ(Id⁎ N 0, s) and σ(Δd⁎/y). The result that IFI debt is chosen to avoid high levels of interest rates somewhat strengthens and the countercyclicality increases marginally. Without the ﬁxed cost of IFI borrowing, the sovereign can more effectively utilize this instrument in bad times to avoid high interest rates and low consumption levels. When IFI lending has no conditionality, we ﬁnd that the sovereign still borrows more commercially than from the IFI. A priori one might expect the opposite, i.e., the sovereign would borrow more from the IFI considering that the mean spreads are lower for the IFI than the commercial debt. Despite higher spreads associated with commercial debt, it has an important advantage over IFI debt in that it is defaultable. The E[d⁎]A/E[d]A ratio increases from 0.24 in the baseline to 0.47 in the no conditionality scenario revealing that the sovereign chooses to hold a signiﬁcantly higher IFI debt in proportion to 25 This discrepancy is largely due to our calibration strategy of matching E[d⁎]/E[d] to data as opposed to E[d⁎] which leads to lower levels of IFI debt generated by the model. 26 Calculating similar statistics for commercial debt was not possible because of short time series data on spreads at annual frequency and commercial debt data is annual.

E. Boz / Journal of International Economics 83 (2011) 70–82 Table 9 IFI debt during booms and busts. E[d⁎] y

Data

Model

b − σy N + σy D=1

9.23 7.43 31.87

5.50 0.06 26.74

commercial debt. However, this ratio does not reach one for the aforementioned reason. The higher conditionality scenario reported in the third column largely works in the opposite direction compared to the no conditionality scenario. The average IFI debt levels and frequency of IFI borrowing are lower. This scenario implies lower number of defaults (43 vs. 64 in the baseline) and consistently lower mean commercial debt spreads. The standard deviation of IFI debt ﬂows fall along with its mean mainly because 96% of the time, the sovereign does not borrow at all from the IFI lowering both mean and variability of this type of debt. The ability of the sovereign to use IFI debt to avoid high commercial interest rates disappears in this case evidenced by a weak negative correlation between the IFI debt dummy and spreads. Reducing k to 0.165 from 0.175 in the baseline causes average IFI debt and frequency of IFI borrowing to be higher than the baseline. This is expected considering that k determines the spreads of IFI debt over the risk-free rate for a given level of d⁎. All other results remain more or less similar to the baseline with somewhat higher variability of IFI debt and commercial debt spreads that appear to go hand in hand in our simulations. Note that mean IFI debt spreads are higher in this case despite k being lower because higher mean IFI debt more than offsets a potential decline in the spreads due to lower k. The ﬁfth column of Table 10 reports the statistics of the scenario with a higher k. Contrary to the lower k scenario, this scenario reduces average IFI debt and its frequency. The high k scenario leads to lower IFI debt holdings as well as a smoother pattern for its ﬂows. Mean IFI debt spreads fall in this case because of the signiﬁcant decline in mean IFI debt. The weak countercyclicality result survives in all of the cases we examine in our sensitivity analysis suggesting that this result is mainly related to IFI debt being non-defaultable which leads to the inelasticity of the IFI interest rate to output shocks and to the level of commercial debt. Finally, not reported in Table 10, is an experiment we conducted assuming no IFI conditionality and no surcharges, i.e., βH = βL = 0.97 and k = 0. The sovereign therefore has to allocate its portfolio Table 10 Sensitivity. β = 0.99

Conditionality

σ(y) σ(c)/σ(y) σ(TB/y) σ(Δd/y)A σ(Δd⁎/y) σ(s)A ρ(Δd/y, y)A ρ(Δd⁎/y, y) ρ(s, y) ρ(c, y) ρ(TB/y, y) ρ(Id⁎ N 0, s) E[d]A E[d⁎] E[d⁎]A/E[d]A E[s] E[s⁎] No. of defaults Pr(d⁎ N 0)

Baseline

None

High

k =.165

k =.185

k = 100

4.39 1.21 1.83 1.84 3.41 4.50 0.28 − 0.08 − 0.34 0.93 − 0.29 0.25 4.64 3.09 0.24 2.75 0.49 64.6 0.34

4.44 1.19 1.79 2.11 4.11 4.54 0.26 − 0.10 − 0.36 0.93 − 0.28 0.30 5.21 6.02 0.47 3.01 1.07 62.0 0.98

4.44 1.26 2.38 1.44 1.67 3.95 0.36 − 0.07 − 0.31 0.90 − 0.25 − 0.05 4.30 1.25 0.02 2.00 0.22 43.6 0.04

4.37 1.20 1.82 1.76 3.76 5.31 0.29 − 0.09 − 0.07 0.93 − 0.26 0.11 4.26 3.63 0.37 4.30 0.61 64.8 0.37

4.69 2.46 1.91 1.63 3.24 4.29 0.31 − 0.06 − 0.34 0.93 − 0.29 0.13 4.69 2.46 0.15 2.49 0.45 56.4 0.16

4.51 1.21 2.25 1.28 0.17 0.81 0.34 0.01 − 0.07 0.93 − 0.24 − 0.00 8.32 0.00 0.00 0.25 0.14 5.80 0.00

79

between a defaultable bond where the interest rate is determined by the default risk and a non-defaultable bond with interest rate the same as the risk-free rate. We ﬁnd low mean commercial debt, but nonzero, high frequency of default coupled with high mean commercial debt spreads. These ﬁndings imply that the sovereign rarely borrows commercially and whenever it does, it ends up defaulting on it. The weak countercyclicality of IFI debt is also robust in this case conﬁrming that the key feature of the debt driving this result is its being non-defaultable. 4.4. Role of IFI lending Having set up a model that captures key characteristics of IFI lending, in this section, we further explore the role played by IFI lending. To do so, ﬁrst, we look at the model moments when loans from the IFI are unavailable. We achieve this by setting both k and βH to high values such that effectively borrowing from the IFI is not an option. Second, we conduct welfare analysis. Model moments without IFI lending are reported in the last column of Table 10. Comparison of these results with those from our baseline model with IFI lending delivers mixed results as to whether the sovereign is better off when borrowing from the IFI is possible. Consumption variability to output variability ratio and consumption's correlation with output are unchanged across these two scenarios. When borrowing from the IFI is not possible, the mean total (both private and IFI) debt stock of the sovereign is higher and at the same time, default is less frequent and hence the spreads on private sector debt lower. The sovereign has more incentives to repay commercial debt in this case since default implies autarky leaving the sovereign without any instruments through which consumption can be smoothed. From a narrow perspective, the presence of IFI debt does not help complete markets since it is not a contingent contract paying only on certain exogenous states of nature, i.e. endowment shocks. However, the price function associated with it is contingent on only part of the state space (including endogenous and exogenous state variables) unlike that of private sector creditor debt which is contingent on the entire state space. In other words, q⁎(d⁎′) while q(d′, d⁎′, y). It has different terms from private sector creditor debt and therefore does provide the sovereign with a non-redundant instrument. A priori it is difﬁcult to conjecture whether IFI lending would improve or reduce sovereign's welfare. At a ﬁrst glance one may be tempted to think that IFI lending is another instrument provided to the sovereign that can potentially facilitate better transfer of wealth over time, and it cannot make it worse off since the sovereign can always choose not to borrow from the IFI. This logic would hold in an environment without frictions where competitive equilibrium allocations are necessarily Pareto efﬁcient. However, our setup incorporates limited commitment and incomplete markets that imply a deviation from the assumptions of the ﬁrst fundamental theorem of welfare economics. The closest study from the literature that examines efﬁciency in a limited commitment framework is Zame (1993). He shows that opening new markets for securities does not necessarily improve efﬁciency if these securities are non-defaultable. He argues, however, that making default an option may improve efﬁciency by allowing the debtor not to pursue repayment in states when debt exceeds income. Based on this argument, the presence of IFI debt, which in some sense is opening a new securities market, may not improve efﬁciency as it is non-defaultable. Given the mixed results based on model moments and the limited help from the literature on this issue, we proceed numerically to evaluate the impact of IFI lending on the welfare of the sovereign. Our measure of welfare is compensating variations in consumption that equate expected lifetime utility in the baseline model with that of the setup without IFI lending. More formally, we solve for g in the

80

E. Boz / Journal of International Economics 83 (2011) 70–82

equation g 1 − σ V IFI (d, d ⁎ , y) = V no − IFI (d, d ⁎ , y) and evaluate its expected value where V IFI and V no − IFI denote the value functions of the scenarios with and without the IFI, respectively. We ﬁnd that, as it is modelled, the presence of IFI lending, on average, leads to a welfare decline of 0.34% in units of consumption. A state by state comparison suggests that conditional on the sovereign being highly indebted to the private sector creditors and output being low, the baseline scenario with the IFI yields higher welfare than the one in which IFI lending is non-existent. In fact, for the lowest realization of output and highest level of commercial debt, the improvement in welfare in the baseline scenario reaches about 1% as plotted in Fig. 3. With this realization of output, the improvement in welfare declines monotonically to −0.75%, a reduction in welfare, as commercial debt falls. In the case of the highest output realization, the sovereign's welfare declines in the baseline compared to when there is no IFI lending. Note that in our setting we did not incorporate any output improvement that IFI lending and its conditionality could provide to the sovereign. There is no investment or capital accumulation, and as a result, the foregone consumption in the current period can reduce future debt but does not improve dynamics of future output which are determined by an exogenous process. In this paper we do not empirically estimate the impact of IFI lending on output but argue that a 0.34% improvement in output is a relatively small number. It is plausible that most of the beneﬁts of IMF programs can be reaped in the long-term well after the program is over, especially if the program involves structural measures as argued by Ghosh et al. (2005) and Haque and Khan (2002).

would without any IFI debt. The model successfully accounts for the weak countercyclicality of IFI debt ﬂows as well as the fact that IFI debt stock tends to be higher in times of high spreads. These ﬁndings are shown to be robust to the key parameters. Moreover, we conclude that conditionality plays a key role in intermittency of borrowing from the IFI while enforceability of IFI debt contracts explains its relatively small size and countercyclicality observed in the data. The recent increase in demand for IMF loans by emerging market economies underscores the importance of understanding the dynamics of IMF lending, or more generally, IFI lending and its impact on a sovereigns' decisions. In addition to providing a setting that accounts for key regularities in the data, the model laid out in this paper can be utilized to answer policy questions regarding the design of IMF lending facilities. The current setting mimics SBAs, however, one could model other facilities such as the new ﬂexible credit line and compare its implications on debt and default.

5. Conclusion

Appendix: Solution

This paper proposed a framework for analyzing dynamics of lending by IFIs in international credit markets. This framework sheds light on the decisions of a sovereign borrower regarding when to default on its commercial debt in the existence of an IFI and how to allocate its borrowing needs among these two types of creditors. Differences in the contracts offered by private sector creditors and IFIs take us a long way in accounting for the stark differences in cyclical properties of these two types of debt. Characteristics of IFI lending are modelled based on the institutional framework of the IMF's Stand-By Agreements. The interest rate charged does not depend on the sovereign's riskiness but only on the size of the loan. In addition, as long as the sovereign is indebted to the IFI, it has to follow more prudent policies, modelled as the sovereign acting as if it had a higher discount factor, valuing future consumption more than it

In this setting with two types of creditors, the sovereign not only makes default decisions but also faces a portfolio allocation problem in non-default periods. Depending on the realizations of output and its current level of debt to each type of creditor, the sovereign allocates its borrowing needs. Given that these creditors lend with different terms, we can solve the sovereign's portfolio choice problem using standard techniques explained below. Initially, we solve for the competitive equilibrium of the model with state spaces of d and d⁎ spanning a wide interval. We ﬁnd that commercial debt has a well-deﬁned stationary distribution in the interval [0.02, 0.34] and d = 0 is imposed in default periods. This reveals that any commercial debt level in the (0, 0.02) range is not optimal. We use this observation to “economize” on the nodes. We place (NB − 1) equally spaced nodes in the [0.02, 0.34] interval, {d1, d2, d3, …, dNB − 1} ∈ [0.02, 0.34] and concatenate them with {dNB} = {0}. The ergodic distribution of IFI debt also displays such discontinuity, however; it is not as strong as the one for commercial debt. Therefore, we set the state space for d⁎ to span the interval [0, 0.36].27 In order to ﬁnd the equilibrium allocations and price schedule for private sector debt, we implement the following algorithm:

1.5 Low y Neutral y High y

Welfare gain (percent)

1

Acknowledgments I am grateful to Sheila Bassett, Enrica Detragiache, Bora Durdu, Chris Jarvis, Anton Korinek, Enrique Mendoza, Ydahlia Metzgen, Alex Mourmouras, Marco Rossi, two anonymous referees and the editor as well as the participants of the World Congress of the International Economic Association in Istanbul, European Meetings of the Econometric Society in Milan, and the IMF Institute Lunch Seminar for helpful comments and suggestions. Correspondence: The views expressed in this paper are those of the author and should not be attributed to the International Monetary Fund, its Executive Board, or its management.

0.5

1. Discretize the state space as explained above. 2. Conjecture a price schedule for private sector debt, qold ðd′; d⁎ ′; yÞ = 1 1 + r. 3. Solve the sovereign's problem recursively using value function iteration by taking the conjectured price schedule for private sector debt and IFI debt price schedule as given and obtain default decisions D(d, d⁎, y), policy functions d′ (d, d⁎, y), d⁎′ (d, d , y), and c(d, d⁎, y). 4. Given these policy functions, calculate default probability λ(d′, d⁎′, y).

0

−0.5

−1

−1.5

0

0.1

0.2

0.3

0.4

d Fig. 3. Welfare gains from the availability of IFI loans (d⁎ = 0).

0.5

27 Note that by limiting the lower limit of IFI debt to zero, we assume that the sovereign cannot lend to the IFI. In practice, some countries in particular advanced economies can be creditors to the IMF. With our calibration, even if the sovereign is allowed to be a creditor, it would never choose to do so in the model. We consider this a plausible scenario given our focus on emerging market economies.

E. Boz / Journal of International Economics 83 (2011) 70–82

5. Using the default probability λ(d′, d⁎′, y), calculate a new price 1−E½λðd′ ;d⁎′ ;yÞ . schedule, qnew d′ ; d⁎′ ; y = 1 + r 6. Update the conjectured price schedule qold(d′, d⁎′, y) with a weighted average of qold(d′, d⁎′, y) and qnew(d′, d⁎′, y). Repeat these steps until convergence is obtained, that is, max |qold(d′, d⁎′, y) −qold(d′, d⁎′, y)| b ξ where ξ is a very small number.28

81

baseline two different β s

Note that Hatchondo et al. (forthcoming) need to keep track of the government type as a state variable and evaluate different continuation values for different types. In our setting, sovereign essentially chooses its own type and the information with regards to its type can be inferred from its debt to the IFI making it unnecessary to include it as a separate state variable. Appendix: Simple small open economy model

−0.5

In this section we analyze a simple small open economy model to elaborate further the implications of switching to a higher discount factor when in debt. We solve the following model with a constant interest rate assuming that commitment technology exists, so there is no default. In order to simplify the analysis further and increase the accuracy of the results, we assume there is only one lender that lends at the risk-free rate. However, whenever the country borrows from this creditor, it has to switch to a higher discount factor. Similar to our benchmark model, we assume an endowment economy but with an i. i.d. exogenous stochastic process. More formally, the small open economy would solve: ∞ max E0 ∑ β bt ct

t =0

t + 1

uðct Þ

subject to ct + bt

+ 1

= yt + bt ð1 + rÞ

ð9Þ

b denotes bond holdings, y is endowment, and interest rate r is assumed to be constant. We assume a CRRA utility function and set β(1 + r) b 1 to ensure stationarity for bonds. This is a standard consumption–saving problem in the case of a small open economy except that the lender requires the economy to switch to a different discount factor as a condition for lending: βðbÞ =

βH ; L β;

if bb0 if b = 0:

This speciﬁcation for the discount factor is the same as that in our benchmark two-creditor model with default. Fig. 4 plots the stationary distributions for two parameterizations of this model, a baseline parameterization with βH = βL = 0.98 and one with different discount factors βH = 0.98, βL = 0.979965. All other parameters are kept the same with the standard deviation of output set to 4.5%. We set βL such that b′ b 0 is chosen around 60% of the time, again similar to our benchmark model. We ﬁnd that one needs to set βH − βL to a very low number in order to obtain the 6y probability of a negative bond position. Larger differences between the two discount factors quickly lead to the country choosing autarky as opposed to borrowing and switching to a higher discount factor. This is likely to be a result of the fact that in these models the welfare cost of living in autarky is small. Having to switch to a higher discount factor leads the country not to choose negative but small (in absolute value) bond positions. Instead of borrowing a small amount, it becomes optimal not to borrow at all and stay in the low discount factor. This is evident in 28 Note that the calculation of q⁎ does not require solving for a ﬁxed point. For given ϕ(d⁎′), we can evaluate q⁎(d⁎′) and plug in the budget constraint of the sovereign.

−0.4

−0.3

−0.2

−0.1

0

bond position Fig. 4. Stationary bond distributions for the simple SOE model.

Fig. 4 which reveals that bond positions in the interval [− 0.04, 0] are almost never chosen. References Aguiar, Mark, Gopinath, Gita, 2006. Defaultable debt, interest rates and the current account. Journal of International Economics 69, 64–83. Alfaro, Laura, Kanczuk, Fabio, 2005. Sovereign debt as a contingent claim: a quantitative approach. Journal of International Economics 65, 297–314. Alfaro, Laura, Kanczuk, Fabio, 2007. Optimal reserve management and sovereign debt. Harvard Business School working paper, No. 07-010. Alvarez, Fernando, Urban J. Jermann, 2000. Efﬁciency, equilibrium, and asset pricing with risk of default. Econometrica 68, 775–797. Arellano, Cristina, 2008. Default risk, the real exchange rate and income ﬂuctuations in emerging economies. The American Economic Review 98, 690–712. Arellano, Cristina, Ramanarayanan, Ananth, 2008. Default and the Term Structure in Sovereign Bonds. University of Minnesota mimeo. Bai, Yan, Zhang, Jing, 2005. Financial Integration and International Risk Sharing. University of Minnesota mimeo. Barro, Robert J., Lee, Jong-Wha, 2005. IMF programs: who is chosen and what are the effects? Journal of Monetary Economics 52, 1245–1269. Cole, Harold, Dow, James, English, William B., 1995. Default, settlement and signalling: lending resumption in a reputational model of sovereign debt. International Economic Review 36, 365–385. Corsetti, Giancarlo, Guimaraes, Bernardo, Roubini, Nouriel, 2006. International lending of last resort and moral hazard: a model of IMF's catalytic ﬁnance. Journal of Monetary Economics 53, 441–471. Cottarelli, Carlo, Giannini, Curzio, 2002. Bedfellows, hostages, or perfect strangers? Global capital markets and the catalytic effect of IMF crisis lending. IMF Working Paper No. 02/193. D'Erasmo, Pablo, 2008. Government Reputation and Debt Repayment in Emerging Economies. University of Maryland mimeo. Eaton, Jonathan, Gersovitz, Mark, 1981. Debt with potential repudiation. Review of Economic Studies 48, 289–309. Eichengreen, Barry, 2003. Restructuring sovereign debt. The Journal of Economic Perspectives 17, 75–98. Eichengreen, Barry, Gupta, Poonam, Mody, Ashoka, 2006. Sudden stops and IMFsupported programs. NBER Working Paper No. 12235. Ghosh, A., Lane, T., Shulze-Ghattas, M., Bulir, A., Hamman, J., Mourmouras, A., 2002. IMF supported programs and capital account crises. IMF Occasional Paper No. 210. Ghosh, A., Christoﬁdes, C., Kim, J., Papi, L., Ramakrishnan, U., Thomas, A., Zalduendo, J., 2005. The design of IMF-supported programs. IMF Occasional Paper No. 241. Guimaraes, Bernardo, 2006. Optimal External Debt and Default. London School of Economics mimeo. Haque, Nadeem, Khan, Mohsin S., 2002. Do IMF-supported programs work? A survey of the cross-country empirical evidence. Macroeconomc Management: Programs and Policies, pp. 38–57. Hatchondo, Juan Carlos, Leonardo Martinez and Horacio Sapriza, 2009. Heterogeneous Borrowers in Quantitative Models of Sovereign Default, International Economic Review 50, 1129–1151. International Monetary Fund, 2002a. Guidelines on Conditionalityavailable at http:// www.imf.org/External/np/pdr/cond/2002/eng/guid/092302.htm2002. International Monetary Fund, 2002b. The Modalities of Conditionality — Further Considerationsavailable at: http://www.imf.org/External/np/pdr/cond/2002/eng/ modal/010802.htm2002. Jeanne, Olivier, Zettelmeyer, Jeromin, 2001. International bailouts, moral hazard and conditionality. Economic Policy 409–432. Kehoe, Tim, Levine, David, 1993. Debt constrained asset markets. Review of Economic Studies 105, 211–248.

82

E. Boz / Journal of International Economics 83 (2011) 70–82

Mendoza, Enrique, Yue, Vivian, 2008. A solution to the default risk-business cycle disconnect. NBER Working Paper No. 13861. Mody, Ashoka, Saravia, Diego, 2006. Catalyzing capital ﬂows: do IMF supported programs work as commitment devices? The Economic Journal 116, 843–867. Morris, Stephen, Shin, Hyun Song, 2006. Catalytic ﬁnance: when does it work? Journal of International Economics 70 (1), 161–177. Neumeyer, Pablo, Perri, Fabrizio, 2005. Business cycles in emerging economies: the role of interest rates. Journal of Monetary Economics 52 (2), 345–380. Ratha, Dilip, 2001. Complementarity between multilateral lending and private ﬂows to developing countries. World Bank Policy Research Working Paper 2746.

Reinhart, Carmen, Rogoff, Kenneth, Savastano, Manuel, 2003. Debt intolerance. Brookings Papers on Economic Activity, No. 1. Saravia, Diego, 2004, “On the Role and Effects of IMF Seniority”, University of Maryland Ph. D. Thesis (www.lib.umd.edu/drum/bitstream/1903/1579/1/umi-umd-1650.pdf). Yue, Vivian, 2010. Sovereign Default and Debt Renegotiation, Journal of International Economics. 80, 176–187. Zame, William R., 1993. Efﬁciency and the role of default when security markets are incomplete. The American Economic Review 83/5, 1142–1164. Zettelmeyer, Jeromin, 2005. Implicit transfers in IMF lending, 1973–2003. IMF Working Paper 05/8.

Journal of International Economics 83 (2011) 83–91

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

Unemployment and relative labor market institutions between trading partners Hervé Boulhol ⁎ University Paris I Panthéon-Sorbonne, 13 rue de Chabrol, 75010 Paris, France

a r t i c l e

i n f o

Article history: Received 13 February 2009 Received in revised form 27 September 2010 Accepted 3 October 2010 Available online 16 October 2010 JEL classiﬁcation: F16 J50 F10 F41

a b s t r a c t This paper contributes to the literature that highlights the role of trading partners' institutions for a country's unemployment rate. The objective is to study whether the results established in the minimum wage–setting of Davis (1998) hold when unemployment is driven by search frictions. This paper ﬁnds that relative labor market institutions matter for equilibrium unemployment as they generate comparative advantages, but there are two main differences with Davis. With North–North trade, unemployment decreases in the lowregulation country. When South is brought into the picture, low-regulation North is not insulated, and unemployment increases in both developed countries as a result of specialization. © 2010 Elsevier B.V. All rights reserved.

Keywords: Unemployment Labor Market Institutions Trade

“A more subtle – but by this more important – reason for considering a global approach is that the consequences even of purely local institutions and shocks often depend crucially on the links to the global market” Donald Davis 1. Introduction The impact of labor market institutions on unemployment is generally assessed without taking into account increasing international economic linkages between countries. Despite globalization, most researchers focus on domestic labor market regulation to explain differences in unemployment rates both across countries and through time. This choice is motivated by theories of unemployment based on job search frictions, according to which more stringent domestic labor market regulation raises unemployment, an explanation herein referred to as the “regulation view“. In doing so, the effect of foreign labor market institutions on domestic unemployment is thus ignored. In their seminal paper, Blanchard and Wolfers (2000) highlight that, even though labor market institutions could explain much of the differences in unemployment across countries either in the 1980s or the 1990s, changes in institutions were too small to account for the changes in unemployment rates. Blanchard and Wolfers ﬁnd evidence

⁎ Tel.: +33 1 45 24 84 58. E-mail address: [email protected]. 0022-1996/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jinteco.2010.10.002

that the mostly common shocks that affected developed countries, such as changes in real interest rate and productivity growth, had differentiated impacts on unemployment rates based on differences in labor market institutions. The rationale is, ﬁrst, that rigidities can delay the adjustment of wages in the advent of negative shocks, which might generate unemployment, and, second, that differences in rigidities are related to differences in institutions. In contrast, SaintPaul (2004) argues that changes in institutions have been signiﬁcant in the last decades and can explain by themselves the magnitude of the trends in unemployment rates.1 In turn, Blanchard (2006) considers that these explanations are only partly satisfactory and encourages researchers to consider other shocks and other interactions. This domestic focus is surprising given the prominent attention placed on the employment consequences of globalization in the public and political debate. Theoretically, Brecher (1974) shows how labor market rigidities generated by a binding minimum wage are magniﬁed by international trade, and Davis (1998), building on Brecher's idea, draws attention to the key interactions between labor

1 The most comprehensive effort to match the changes in unemployment rates with those in institutions is probably that of Nickell et al. (2005) who ﬁnd support for the regulation view and assess that the shock interactions à la Blanchard–Wolfers are not robust once added to their thorough speciﬁcation. On the other hand, Baker et al. (2005) and Baccaro and Rei (2005) present a sceptical analysis of the evidence produced in support of the regulation view. Following a rigorous empirical strategy, Bassanini and Duval (2006) reach more moderate conclusions as for the role of institutions, either directly or through the interactions with shocks.

84

H. Boulhol / Journal of International Economics 83 (2011) 83–91

market institutions designed at the country level and global goods markets.2 In a stylized trade model between ﬂexible-wage “America” and minimum wage “Europe”, Davis shows that trade ties up factor prices between countries and leads to an increase in long-term unemployment in “Europe”. Davis' main intuition lies in the fact that “even when factor markets are strictly national, with idiosyncratic institutional features, they cannot be considered in isolation when goods markets are global”. There is a major difference between the Blanchard–Wolfers hypothesis and the Brecher–Davis mechanism. The former seems implicitly optimistic in that the effects of labor market institutions, albeit persistent, are not a long-term phenomenon since “bad” institutions merely slow the necessary adjustments. In contrast, the Brecher–Davis interactions between trade and labor market institutions affect the unemployment rate in the long run. The effects of a country's labor market regulations on its trading partners are the subject of a growing attention. Davidson and Matusz (2005), Moore and Ranjan (2005) and Cuñat and Melitz (2007) provide evidence that labor market institutions affect comparative advantages. Boulhol (2009) and Helpman and Itskhoki (2010) highlight how trade liberalisation affects unemployment through specialization effects when countries differ in either union power or hiring costs, respectively. For example in Boulhol (2009), the threat of relocations, which trade integration makes even more credible, encourages labor market deregulation in the highly-regulated country to avoid capital outﬂows, thus leading to a decrease in unemployment. In an earlier contribution, Davidson et al. (1999) show that trade liberalisation between two countries, one of which is a capital abundant large country with a more efﬁcient labor market, leads to higher unemployment in that country. In these models, labor market regulation typically affects sectors asymmetrically, which generates comparative advantages when either labor market institutions or factor endowments differ across countries. Dutt et al. (2009) focus on trade between countries having different development levels with search-generated unemployment. They show that unemployment increases with trade in the labor scarce country if comparative advantages arise from differences in factor endowments, but decreases if they are based on relative technological differences, and ﬁnd empirical support for the latter. The current paper is part of the still nascent literature trying to incorporate unemployment in trade models. Its main contribution consists in investigating the extent to which Davis' main idea can be generalized to a broader type of labor market institutions than the simple minimum wage context. Using the two-factor matching framework of Pissarides (2000), the model developed herein highlights that foreign labor market institutions affect a country's unemployment rate through the trade channel. The main mechanism through which institutions of trading partners inﬂuence unemployment is straightforward. To the extent that labor market institutions matter for unemployment, they affect the cost of labor and, therefore, relative factor and good prices. It follows that labor market regulation contributes to comparative advantages in an open economy. More speciﬁcally, the paper brings three main results. First, in the case of trade between two developed countries, as in Davis, the highregulation (“rigid”) country that has relative high labor costs tends to specialize in the capital-intensive good and unemployment increases as a result of the induced fall in labor demand. Second, and this is the main difference with Davis in the North–North context, the “ﬂexible” economy beneﬁts from trade with the “rigid” country in terms of total employment, through the induced increase in demand for the laborintensive good. Through trade and induced changes in factor prices,

2 In a different context, Krugman (1995) emphasizes that the impact of trade with developing countries on wages and employment depends on the functioning of the labor market: trade effects are likely to be mostly reﬂected by changes in wages in ﬂexible economies and in employment in rigid ones.

comparative advantages in labor market institutions enable the “ﬂexible” economy to transfer some of the search-friction costs to the “rigid” economy. Third, when South is introduced to this North– North equilibrium, overall specialization results from the combination of endowment- and institution-driven comparative advantages. In the case where the endowment effect dominates, labor abundant South specializes in the labor-intensive good. Consequently, unemployment decreases in the developing country and increases in both developed countries from their North–North equilibrium levels: in contrast with Davis, the “rigid” developed country does not absorb the whole permanent shock related to trade with South, and “ﬂexible” North is also negatively affected in terms of employment. Combining North– North with North–South trade, the unemployment rate rises unambiguously from the autarky level in high-regulation North, whereas the total effect is ambiguous for low-regulation North. The rest of the paper is organized as follows. Section 2 extends the matching model to a two-sector economy. Section 3 embeds this framework into a standard trade model focusing on North–North trade, while Section 4 introduces a labor abundant country into the picture. Finally, Section 5 concludes. 2. A two-sector extension of Pissarides' matching framework This section embeds the large-ﬁrm version of the matching model of Pissarides (2000, chapter 3), which captures the main features of the regulation view, into a two-sector model. Each sector produces a homogeneous good under perfect competition. There are two factors of production, capital K and labor L. The two factors could alternatively be thought of as being skilled and low skilled labor, as in Davis, with rigidities affecting mainly low skilled labor. This framework lays the ground for an extension of the Heckscher–Ohlin trade model to search frictions, which is developed in Section 3. Firms are identical within each sector, and sector 1 produces the capital-intensive good of price p in terms of good 2, which is chosen as the numeraire.3 Let Kij and Lij be the capital and employment of ﬁrm i in sector j, and let Fj(Kij, Lij) be the constant returns-to-scale production function of all ﬁrms in that sector. Each ﬁrm is large enough so that there is no uncertainty about its ﬂow of labor, and unemployed workers are assumed to be perfectly mobile across sectors. Wages are bargained at the individual level and ﬁrms choose the number of jobs by taking the bargained wages as given. Labor market characteristics are assumed to be identical in both sectors.4 While employed workers might be attached to the sector in which they work due to search frictions, it is assumed that unemployed workers have no attachment to a speciﬁc sector: they randomly take the ﬁrst job possible. Hence, during a small interval dt, a vacant job is matched to an unemployed worker with the same probability m(θ)dt in both sectors, where m(·) is the matching function which decreases with labor market tightness, and θ, deﬁned as the ratio of total vacancy to unemployment rates. Usual properties of the matching function, discussed at greater length in Pissarides (2000), are supposed to hold: m′ðθÞb0; jθm′ðθÞj b mðθÞ

3

ð1Þ

As goods are homogenous, all ﬁrms within the same sector set the same price. There are two reasons for this assumption. The ﬁrst one is analytical simplicity. The second is that the impact of differences in labor market features across sectors on trade specialization has been studied elsewhere, as pointed out in the introduction. For example, Boulhol (2009) and Helpman and Itskhoki (2010) consider differences in bargaining power and hiring costs, respectively. In these papers, the analytical complexity is resolved by assuming that there is no imperfection in one sector (i.e. it is always possible to ﬁnd a job in that sector). Differences between countries in labor market institutions (of the other sector) inﬂuence specialization. In contrast here, specialization is determined by the interactions between sectoral factor intensities and labor market institutions, which differ between countries but not across sectors. 4

H. Boulhol / Journal of International Economics 83 (2011) 83–91

85

such that θm(θ) increases with θ. It is assumed that labor market tightness is exogenous to the ﬁrm's control and that each ﬁrm loses workers at the exogenous separation rate λ. Each vacancy costs the ﬁrm h in recruitment costs and returns a worker at the rate m(θ). Therefore, denoting Vij the vacancies at ﬁrm i in sector j, the law of motion of job is:

In equilibrium, proﬁt opportunities drive rents from vacant jobs to zero, i.e. Jv, j = 0, implying that proﬁts from occupied jobs are the same in both sectors:

L˙ ij = −λLij + mðθÞVij :

Let We, j and Wu denote the present-discounted values of the expected income stream of an employed worker in sector j and of an unemployed worker, respectively. It follows that:

ð2Þ

Aggregating across all ﬁrms in each sector shows that in the steady-state: Vj = Lj = λ = mðθÞ:

ð3Þ

Summing across sectors gives the steady-state unemployment rate: u=

λ : λ + θmðθÞ

ð4Þ

Unambiguously, given Eq. (4), the unemployment rate is negatively related to labor market tightness.5 I representing the investment of price pK, δ the depreciation rate of capital stock, w the gross wage, h the cost of a vacant position and r the discount rate, ﬁrm i in sector j maximizes the present-discounted value of expected proﬁts: ∞ −rt

Max Πij =∫ e Vij ;Iji

0

h i pj Fj Kij ; Lij −wj Lij −hVij −pK Iij dt

Jo; j =

pj FjL −wj h : = λ+r mðθÞ

ð8Þ

V V rWu = z + θmðθÞ 1 We;1 + 2 We;2 −Wu V V

ð9aÞ

rWe; j = wj = ρ−λ We; j −Wu

ð9bÞ

where z and ρ denote unemployment beneﬁts and the tax wedge between gross and net of labor tax wages, respectively.6 Eq. (3) implies that the sector shares in total vacant positions are equal to their respective sector employment shares, se, j. Using the expression of We, j in Eq. (9b), Eq. (9a) becomes: w1 = ρ λ w2 = ρ λ rWu = z + θmðθÞ se;1 + Wu + se;2 + Wu −Wu λ+r λ+r λ+r λ+r " = z + θmðθÞ

ð5Þ

w= ρ r − W λ+r λ+r u

#

where w is the average wage, and therefore:

with p1 = p and p2 =1 ˙ =−λLij + mðθÞVij : s:c: K˙ ij =Iij −δKij ; Lij Let kij ≡ Kij/Lij be the capital per unit of labor and fj(kij) ≡ Fj(Kij,Lij)/Lij. The ﬁrst-order condition with respect to the investment decision implies that the capital–labor ratio is negatively related to the real user cost of capital (and is the same for all ﬁrms in a given sector; the i sub-script is hence dropped below): marginal product of capital = pj FjK Kij ; Lij = pj f′j kj = pK ðr + δÞ≡cK :

ð6aÞ The adjustment cost of labor, represented by h, creates a wedge between the marginal product of labor and the gross wage. Indeed, proﬁt maximization entails that the marginal product of labor is equal to the sum of the gross wages, w, and the expected capitalized value of the ﬁrm's hiring costs: h i hðλ + r Þ : marginal product of labor = pj FjL Kij ; Lij = pj fj kj −kj f′j kj = wj + mðθÞ

ð6bÞ Jo, j and Jv, j being the present-discounted values of expected proﬁt from an occupied and vacant job in sector j, respectively, Bellman equations lead to: rJv; j = −h + mðθÞ Jo; j −Jv; j rJo; j = pj fj kj −pK ðr + δÞkj −wj −λ Jo; j −Jv; j

rWu =

rWe; j

ðλ + r Þz + θmðθÞw = ρ λ + r + θmðθÞ

ð10Þ λz + ðr + θmðθÞÞwj = ρ + λθmðθÞ w−wj ρ = ðλ + rÞ : = λ + r + θmðθÞ

=

The negotiated wage in each sector is the outcome of the Nash bargaining which boils down to maximizing the weighted product of the worker's and the net ﬁrm's surpluses from the match, the weight γ representing workers' bargaining power: wj = arg max (We,j − Wu)γ ((Jo,j − Jv,j) / ρ)1 − γ. The ﬁrst-order maximization condition satisﬁes: We,j − Wu = γ ((We,j − Wu) + (Jo,j − Jv,j) / ρ) which, using Eqs. (8) and (10) and after some manipulations, leads to:

wj +

θmðθÞ γ hðλ + r + θmðθÞÞ P wj − w = ρz + : λ+r 1−γ mðθÞ

ð11Þ

This proves that wages are the same across sectors: w1 = w2 ≡ w. In short, this result stems from the perfect mobility of unemployed workers and the assumption that both sectors have the same labor market characteristics. It enables to preserve all features of the one sector model, as shown now. It is straightforward to derive the expressions of the wage–setting and labor demand schedules from the earlier mentioned analysis. Using the equality of wages across sectors, Eq. (11) simpliﬁes into:

ð7Þ wage setting: w = ρz +

γ λ+r h θ+ : 1−γ mðθÞ

ð12Þ

This positive relation between tightness and bargained wages is the wage–setting curve. 5 For example, referring to the previous footnote, if one allows for different separation rates across sectors, Eq. (4) remains valid, but with λ = λ1s1 + λ2(1 − s1) where s1 is share of employment in sector 1, which is affected by specialization.

6 This type of framework typically ignores how unemployment beneﬁts are ﬁnanced and taxes spent.

86

H. Boulhol / Journal of International Economics 83 (2011) 83–91

Combining Eqs. (6a) and (6b) for the numeraire good deﬁnes a negative relationship between the marginal product of labor and the user cost of capital, cK ≡ pK(r + δ):

w+

hðλ + r Þ = g2 ðcK Þ; g′2 ðcK Þ = −k2 b0 mðθÞ

ð13Þ

where g2(cK) ≡ f2(f2′− 1(cK)) − cKf2′− 1(cK) depends on the characteristics of the numeraire good 2. For a given user cost of capital, this expression implies a downward-sloping relationship between wages and labor market tightness, which is similar to labor demand7: labor demand: hðλ + r Þ = mðθÞ = g2 ðcK Þ−w; g′2 b0:

ð14Þ

The wage–setting and labor–demand curves deﬁne the unique equilibrium determining labor market tightness as a function of the cost of capital and of the parameters characterizing labor market institutions: labor market equilibrium: γθ +

λ+r 1−γ ½g2 ðcK Þ−ρz = mðθÞ h

ð15Þ

which ensures that the following sensitivities are satisﬁed: ∂θ ∂θ ∂θ ∂θ ∂θ ∂θ b 0; b 0; b 0; b 0; b 0; b0 ∂γ ∂z ∂ρ ∂r ∂h ∂λ

ð16Þ

or equivalently given Eq. (4), ∂u ∂u ∂u ∂u ∂u ∂u N 0; N 0; N 0; N 0; N 0; N 0: ∂γ ∂z ∂ρ ∂r ∂h ∂λ

ð17Þ

This extension of the Pissarides' matching model to a two-sector framework still captures the main features of the regulation view. Workers' bargaining power, and therefore, union density, is positively related to the unemployment rate, because a greater union power tends to push up wages, which reduces labor demand. An increase in the unemployment beneﬁts leads to an increase in the unemployment rate, as it improves the outside option of workers in the bargaining process and therefore boosts wages. For the same reason, unemployment is positively related to the tax wedge, but only to the extent that an increase in the tax wedge, which drives down net wages, is not offset by lower unemployment beneﬁts, i.e. to the extent that the wedge between net wages and beneﬁts is reduced.8 3. Relative labor market institutions with respect to trading partners The purpose of this section is to introduce trade in this set-up, thus providing a theoretical framework in which the labor market institutions of trade partners affect a country's unemployment rate. Davis' (1998) rationale is followed closely by combining the two-sector matching model, developed in Section 2, with the standard trade framework under homothetic preferences, keeping in mind that good 1 is assumed to be relatively capital-intensive at any factor prices. Subsection 3.1 deals with autarky and sub-section 3.2 with the open economy. 7 Section 3 shows that the focus on sector 2 is not restrictive: equilibrium conditions ensure that the reciprocal relation derived from sector 1 is actually identical (see next). 8 Employment protection can be seen both as increasing the vacancy costs and as decreasing the separation rate. The former has a positive impact on the unemployment rate, while the latter has a negative one. Therefore, employment protection has an ambiguous effect overall in this framework.

3.1. Autarkic equilibrium In order to understand how factor endowments determine equilibrium unemployment based on domestic labor market regulation, it is convenient to explore the relationships between the user cost of capital and the unemployment rate in detail, ﬁrst by focusing on the labor market, then on the product market. What is the effect of changes in the user cost of capital on the unemployment rate in this two-factor framework, as a result e.g. of a change in relative endowments? According to the expression of the labor market equilibrium (Eq. (15)), an increase in the cost of capital triggers an increase in unemployment. Fig. 1 represents the labor market equilibrium based on the labor demand and wage–setting schedules. The equilibrium is initially at (E0). An increase in the cost of capital is associated with a shift in labor demand, which results in lower wages and higher unemployment. The new equilibrium is at (E1), and the mechanism is the following. A greater user cost of capital means a lower capital–labor ratio in both sectors such that the marginal product of capital increases to match the rise in the user cost. This adjustment in the capital–labor ratio is associated with a decrease in the marginal product of labor, w + h(λ + r) / m(θ), which results in a decrease in both wages and labor market tightness along the wage–setting schedule, hence a rise in unemployment. This mechanism by which an increase in the user cost of capital is associated with a higher unemployment level (and reciprocally for a decrease in the user cost) plays a key role in the open economy through changes in factor prices that are induced by trade with countries having either different labor market institutions or different factor endowments. In sum, the labor market equilibrium deﬁnes a positive relationship between the unemployment rate and both the cost of capital cK and the level of (unfavorable) labor market regulation (LMR), which represents unemployment beneﬁts, the tax wedge, the cost of search, etc.: RR

labor market equilibrium ðRRÞ: u =αðcK ; LMRÞ; ∂α=∂cK N 0 ∂α=∂LMR N0:

ð18Þ The function α captures the main features of the regulation view. It depends only on labor market institutions and the production functions (see Eq. (15), and footnote 10 below), i.e. not directly on the country's factor endowments which effects on unemployment are channeled through the product market equilibrium only and the resulting capital cost level. The positive relation between the user cost of capital and the unemployment rate is represented by the positively-sloped RR curve in Fig. 2, where R stands for regulation. The discussion in the previous section implies that an increase in (bad) regulation shifts the RR curve to the upper-left. This RR curve holds for both autarky and trade equilibriums as it depend only on labor market regulation and technological parameters. In autarky, the product market equilibrium through demand for both goods determines the cost of capital as a function of employed factor endowments. Indeed, according to the Heckscher– Ohlin theorem, there is a negative relation between the price of the capital-intensive good and the effective capital–labor endowment: p=ζ

κ K =ζ ; ζ′ b0 Nð1−uÞ 1−u

ð19Þ

where K is the country's capital stock, N the labor force, and κ ≡ K/N denotes relative endowments. As long as the country produces both goods, the Stolper– Samuelson type-relation, adapted here to take into account the adjustment costs to labor, still implies a positive relation between cK

H. Boulhol / Journal of International Economics 83 (2011) 83–91

87

Wage High Regulation

BD locus (product market equilibrium)

unemployment rate

RR (labour market equilibrium)

(WS)

E0

A

Low Regulation

W0

RR* A*

W1 E1 user cost of capital

(LD)

Fig. 2. Regulation view in a two-sector model.

L1

L0

Employment

(LD): labor demand schedule (WS): wage-setting / labor supply schedule Fig. 1. Labor market equilibrium and the impact of an increase in the user cost of capital.

and p.9 Indeed, as shown in the previous section, wages and therefore the marginal products of labor are equalized across sectors at equilibrium, which implies:

implies a negative relationship between the unemployment rate and both the cost of capital and the price of the capital-intensive good: an increase in the unemployment rate, by reducing the effective labor available to the economy, makes the labor-intensive good relatively more expensive, hence a decrease in both cK and p12: A BD −1 product market equilibrium BD : uA = 1−κ = ζ ð pÞ; ς′b 0 BD

⇔uA = φðcK Þ; φ′ N 0: g2 ðcK Þ = f2 ðk2 Þ−cK k2 = pf1 ðk1 Þ−cK k1 = pg1 ð cK = pÞ

where g1 is the counterpart of g2 for good 1.10 Given that good 1 is capital-intensive (i.e. k1 N k2 at any factor prices), Eq. (20a) implicitly deﬁnes a positive relation between the relative price p of the capitalintensive good and the user cost of capital as differentiation leads to11: dp αK ðk −k2 Þ dck : = K 2 1 p α2 k1 + 1−αK2 k2 cK

Importantly, both decreasing functions only depend on the technical parameters of the production functions, on the relative factor endowment κ ≡ K/N and on preferences. This means that labor market regulation does not affect this relation, which is labelled as the BD locus in Fig. 2, where BD stands for Brecher–Davis, while A stands for autarky in Eq. (21). Therefore, the autarkic equilibrium is deﬁned at the intersection of the BD and RR schedules: −1 uA = α cK;A ; LMR ; uA = 1−κ = ζ ðpA Þ; cK;A = ψðpA Þ:

This also captures the well-known result that absolute changes in the user cost amplify absolute changes in the relative price of the capitalintensive good. This positive relation between the cost of capital and the price of the capital-intensive good (Eq. (20a)) is denoted: cK = ψð pÞ; ψ′ N cK = p N 0:

ð21Þ

ð20aÞ

ð20bÞ

ð22Þ

For example, a more stringent regulation moves the (RR) curve on the upper-left, increasing the unemployment rate while reducing the cost of capital (and the price of the capital-intensive good) along the (BD) locus. Reciprocally, a decrease in capital endowment moves the (BD) curve towards the upper-right, increasing the cost of capital (and p) and the unemployment rate along the (RR) locus. 3.2. The magniﬁcation effect of trade

The Heckscher–Ohlin and Stolper–Samuelson theorems (Eqs. (19) and (20b)) characterize the product market equilibrium, which 9 Using Samuelson's (1962) terminology, the equilibrium factor price cK is obtained at the intersection of the factor price frontier of both goods at price p. 10 Eq. (20a) can also be written as p2 g2(cK/p2) = p1 g1(cK/p1), which makes it clear that the focus on good 2 to characterize the labor market equilibrium (Eq. (15)) is innocuous as product market equilibrium links the user cost of capital to good prices based on the properties of both production functions. For example, Eq. (15) could be written as well as γ θ + (λ + r)/m(θ) = (1 − γ)/h [p g1(cK/p) − ρ z]. 11 In this model, the usual properties of the Heckscher–Ohlin model, e.g. those linking changes in factor prices to changes in good prices are preserved, provided that wages are adjusted to take into account adjustment costs. That is, the appropriate price of labor should be taken as ω ≡ w + h(λ + r)/m(θ). Differentiating Eq. (12) leads to dω = dw − h(λ + r)m′/m2dθ = [1 − (1 − γ)/γ(λ + r)m′/(m2 − (λ + r)m′)] dw, which shows that wages and the adjusted labor price ω move in the same direction (given that m′ b 0). One can check that when the match is perfect (m(θ) = 1∀θ), dω = dw or when h → 0, then θ → + ∞ (i.e. u → 0) and given Eq. (1) m′ → 0 and dω = dw also.

The mechanism highlighted by Davis, who treats the case of trade between a minimum wage and a ﬂexible-wage economy, is now extended to the labor market framework presented earlier. Suppose that two countries having the same relative factor endowments and preferences open up to trade: κ≡ NK = K * ≡κ* = κW ≡ K + K * where N* N + N* the * and W superscripts refer to the foreign country and the world, respectively. In that case, both countries share the same BD locus in

12 It should be noted that the framework is a static one. Taking the dynamics of capital accumulation into account would amplify these mechanisms because of the substitution of capital to labor.

88

H. Boulhol / Journal of International Economics 83 (2011) 83–91

the closed economy, which is also the BD relation in the integrated equilibrium linking the price p to the world employed factors: BD uA

−1

= 1−κ = ζ

BD* ð pÞ; uA

= 1−κ = ζ

−1

BD;W ðp* Þ; uT

= 1−κ = ζ

−1

A BD

BD locus in the closed economies and in the integrated equilibrium

ðpÞ:

High Regulation RR

unemployment rate

ð23Þ In autarky, the low-regulation equilibrium is represented at point A⁎ in Fig. 2 for the foreign country. In the domestic country, assumed without loss of generality to have a more stringent regulation, the autarkic equilibrium settles at a point like A along the common BD locus. Stricter regulation implies a larger unemployment rate, and both a lower user cost of capital and price p: uA N uA⁎ and pA b pA⁎. Consequently because regulation affects relative prices it creates comparative advantages even if factor endowments are identical. Let s ≡ N/(N + N⁎) be the share of the domestic country in world labor force. The world RR curve is given straightforwardly as follows, since the world unemployment rate uW is equal to su + (1 − s)u⁎: W RR;W RR RR RRT : uT = su + ð1−sÞu * = sαðcK ; LMRÞ+ ð1−sÞαðcK ; LMR* Þ:

T

A

BD*

uW A*

B

W

RR

BD

W

ð25Þ u = µ (u *, pT )

ð pÞ:

ð27Þ

This deﬁnes the (BD) locus in the trade equilibrium for the domestic country, in which the domestic unemployment rate is inversely related to the foreign unemployment rate as follows: 1−s 1−κ = ς−1 ðpÞ ∂μ ∂μ T BD u* + ≡μ ðu*; pÞ; b 0; b 0: BD : uT = − s s ∂u* ∂p ð28aÞ Reciprocally, for the foreign country: s 1−κ = ς−1 ðpÞ ∂μ* ∂μ* T BD* u+ ≡μ* ðu; pÞ; b 0; b 0: BD * : uT = − 1−s 1−s ∂u ∂p ð28bÞ

13 I am especially grateful to one anonymous referee for having suggested this formalization.

BD pT

u*

p , p*

T* A*

W RR* BD*

ð26Þ

−1

T A

The trade equilibrium is represented in Fig. 3A. As a result, from autarky to trade, changes in the output mix induce the (BD) locus to shift rightwards for the domestic high-regulated country and leftwards for the foreign low-regulated country. Indeed, for a given level of unemployment in the foreign country, u⁎, Eq. (23) implies13: suT + ð1−sÞu* = 1−κ = ζ

p*

u u = µ (u *, p A )

It follows that trade ampliﬁes the regulation-induced differences in unemployment rates between countries. Indeed, pA ≤ pT ≤ pA⁎ W BD,W (pT) ≥ implies, according to Eq. (23), that uA = uBD A (pA) ≥ uT = uT BD⁎ ⁎ ⁎ uA =uA (pA ): in the trade equilibrium, the world unemployment rate lies in between the autarkic rates. Also, based on the RR curves, uT = uRR(pT) ≥ uRR(pA) = uA, and reciprocally uT⁎ ≤ uA⁎. Hence, uT ≥uA ≥uT ≥u⁎A ≥u⁎T :

pT

p, relative price of the capital intensive good (or user cost of capital)

Hence, the trade equilibrium is deﬁned by:

sα½ψðpT Þ; LMR + ð1−sÞα½ψðpT Þ; LMR* = 1−κ = ζ−1 ðpT Þ:

RR* Low Regulation

T* p

ð24Þ

uT = α cK;T ; LMR ; uT = α cK;T ; LMR* ; cK;T = ψðpT Þ

RR (World )

u*

Fig. 3. Trade between two countries having identical relative factor endowment.

the BD schedules being linked by: BD

BD*

suT ðpÞ + ð1−sÞuT

−1

ðpÞ = 1−κ = ζ

BD BD* ð pÞ = uA ð pÞ = uA ðpÞ:

ð28cÞ

With trade, each country's BD curve depends on the other country's effective endowments, i.e. the other country's unemployment rate. Therefore, as an alternative to Eq. (25), the trade equilibrium can be characterized by: uT = α cK;T ; LMR ; u⁎T = α cK;T ; LMR* ; cK;T = ψðpT Þ; uT = μ u⁎T ; pT : ð29Þ As one can easily verify that the autarky (BD) loci are given by uA = μ(uA, pA) and uA⁎ = μ⁎(uA⁎, pA⁎) = μ(uA⁎, pA⁎), Eqs. (28a)–(28c) imply the aforementioned shifts from autarky to trade. The underlying channel is specialization through trade along the following line. Due to differences in labor market regulation, this world behaves as if Home is relatively capital abundant. At the domestic autarky price, foreign demand for the capital-intensive good increases, the more so the lower the regulation in the foreign country (as it implies a higher pA⁎). The resulting boost in the Home price of the capitalintensive good generates an increase in the user cost of capital, which rises unemployment. Due to labor market rigidities, supplying more of the capital-intensive good can only happen if the domestic economy

H. Boulhol / Journal of International Economics 83 (2011) 83–91

becomes effectively more capital abundant by shifting resources to the capital-intensive sector, thus generating an increase in unemployment. Fig. 3B illustrates how the BD curves shift with trade. Equilibrium in the domestic and foreign countries is shown in the North–East and South–East quadrants, respectively, with autarky at points A and A⁎. The assumption that both countries have the same relative factor endowments is reﬂected by both countries having the same (symmetric in the chart) (BD) locus in autarky. Because the domestic country has a more stringent labor market regulation, the autarkic unemployment rate is higher than in the foreign country. The integrated equilibrium W settles at a price pT, which lies somewhere between pA and pA⁎ based on the relative size of the two countries. The North–West quadrant represents the negative relationship between the unemployment rates in both countries at price pA and pT according to u = μ(u⁎, pA) and u = μ(u⁎, pT), respectively. Finally, the South– West quadrant uses the 45° line to transform the foreign unemployment rate from the North–West to the South–East quadrant. Starting from p = pA, the domestic RR curve implies u = uA and, according to u = μ(u⁎, pA), u⁎ = uA, which in turn based on RR⁎ would mandate a price p⁎ N p = pA clearly inconsistent with the trade equilibrium. Any point on the South–West of A along the domestic RR schedule will produce a similar outcome. Price convergence occurs therefore at a price such as pT whereby the equilibrium in each country moves along their respective RR locus at points T and T⁎: because the domestic country specializes in the capital-intensive good, the BD locus shifts rightwards for the domestic country and leftwards for the foreign country according to Eq. (28a–28c).14 In the domestic country, trade induces a joint increase in the user cost of capital and the unemployment rate, while the converse applies to the foreign country, leading to the aggregated unemployment rate uW. In this model the unemployment rate of a given country is therefore negatively related to the regulation level of its trading partner.15 Hence, the qualitative results obtained by Davis are extended to a more general context, with one noticeable difference. In Davis, the low-regulation country has ﬂexible-wages and therefore no unemployment in autarky, as well as in the trade equilibrium. With labor market frictions, the lowregulation country beneﬁts from trade with a high-regulation country in terms of employment, through a decrease in the user cost of capital, which boosts labor demand. Davis' special (and extreme) case, where regulation is limited to minimum wages in one country, is obtained as follows, with s = 1/2. The labor market in the ﬂexible economy implies uT⁎ = 0, ∀p, while the binding minimum wage in the rigid one induces pT = pA. Hence, uT = μ(uT⁎, pT) = 2[1 − κ/ζ− 1(pA)]= 2uA: the unemployment rate doubles in the domestic country. Davis' case is quantitatively extreme as Eq. (26) implies that with s = 1/2 the doubling of the unemployment rate in the rigid country is an upper bound.

89

differences. sN being the share of North in the world labor force, the world RR schedule is still given by: W RR;W N N N S = s α cK ; LMR + 1−s α cK ; LMR : RRT : uT

Differences in relative endowments imply that the two countries have different BD schedules: BD;N

uA

N

= 1−κ = ζ

−1

N BD;S S −1 S p ; uA = 1−κ = ζ p

W BD;W W −1 N BD;N N BD;S ðpÞ = 1−κ = ζ ðpÞ = s uA ðpÞ + 1−s uA ðpÞ: BDT : uT ð32Þ Hence, it is a general feature that the integrated RR and BD schedules are the labor-force-weighted average of the countries' (autarky) schedules. As a result, the trade equilibrium price pT is given by Eqs. (30) and (32), with cK,T = ψ(pT). Given labor market regulation in both countries, North will have a comparative advantage in the capital-intensive sector if it is effectively 1−uN

capital abundant: pNA bpSA ⇔κN N κS 1−uAS . With trade, the North BD A locus shifts as follows, given Eq. (32): BD;N

uT

=−

1−sN S 1−κW = ς−1 ðpÞ N S ∂μ N ∂μ N u + ≡μ u ; p ; S b 0; b0 N N ∂p s s ∂u ð33Þ

while the trade equilibrium is now given by: N N S S N N S uT = α cK;T ; LMR ; uT = α cK;T ; LMR ; uT = μ uT ; pT ; cK;T = ψðpT Þ:

ð34Þ Assume ﬁrst that both countries have the same regulation, with North being capital rich/labor scarce relative to South. Differences in relative endowments imply that the BD schedule differs between the two countries in autarky, and autarky equilibrium is at a point like N in North and S in South in Fig. 4: the labor abundant country has both a higher cost of capital and unemployment rate. As a result of specialization, the integrated equilibrium is at a point like W on the common RR schedule, ensuring convergence of the unemployment rate across countries according to Eq. (18) as a result of price convergence. In

BD locus in the integrated economy

BD in the closed South economy

4.1. One North, one South

14 Factors are supposed to be immobile internationally in the Heckscher–Ohlin context. This can be problematic if K is explicitly thought of as physical capital. However, in such a case, low return to capital in the “rigid” economy would lead to capital outﬂows that would shift the BD schedule rightwards. This would produce similar results to the case of trade without capital mobility, because factor mobility is a substitute for trade in this model. 15 If there remain some impediments to trade, full convergence to the world price is not complete. As a result, the equilibrium would be somewhere between A and T for the domestic country and between A⁎ and T⁎ for the foreign country, depending on the level of trade costs.

ð31Þ

The world trade BD curve, uBD,W (p) = 1 − κW/ζ− 1(p), is thus a T weighted average of the countries' autarkic BD curves as κW = sNκN + (1 − sN)κS:

4. North–South trade

RR Same level of regulation

unemployment rate

When countries have different relative factor endowments (κS b κN), labor market institutions contribute to the establishment of comparative advantages on top of those driven by endowment

ð30Þ

S

uW

W

N

BD in the closed North country

pN

pT

pS

Fig. 4. Trade between two countries having identical labor market regulation, but different relative factor endowments.

90

H. Boulhol / Journal of International Economics 83 (2011) 83–91

this model with search frictions, trade leads to an increase in unemployment in the labor scarce country and to a decrease in the labor abundant country, a result that bears some resemblance with the analysis of Davidson et al. (1999), and is consistent with the Hechscher– Ohlin framework of Dutt et al. (2009) even though the settings are different. When differences in labor market regulation are brought into the picture, comparative advantages are driven by both differences in endowments and in labor market institutions, with the total effect combining those in Fig. 3 (differences in institutions only) and Fig. 4 (differences in endowments only): high-regulation in the developed country relative to the developing country amplify the endowmentdriven comparative advantages, whereas the latter would be attenuated, and potentially reversed, if the developing country highly regulates. 4.2. Two North, one South The ﬁnal case investigates the impact of trade with a laborintensive country (South) on a pair of developed countries (North, domestic and foreign) with low- and high-regulated labor market. W (p) = 1 − The BD locus in the trade equilibrium is still given by uBD, T N S N BD, N * N (p) + (1 − s)s u (p) + (1 − s )u BD, (p), κ W /ζ − 1 (p) = ss N u BD, A A A where s is the share of the domestic country in North labor force, which implies: 1−s 1−sN S 1−κW = ς−1 ðpÞ N T;N BD S u + ≡μ u*; u ; p : u*− : uT = − BD N N s ss ss ð35Þ Provided that each of the three countries produces both goods, the full equilibrium is given by: uT = α cK;T ; LMR ; uT = α cK;T ; LMR* ; uST = α cK;T ; LMRS uT = μ u⁎T ; uST ; pT ; cK;T = ψðpT Þ:

ð36Þ

As in Davis, it is helpful to start from the equilibrium corresponding to free-trade between the developed countries, as represented in Fig. 3, then to introduce of a previously isolated labor-intensive country, South, and apply the North–South analysis developed in the previous subsection. If, given labor market regulations, South maintains a comparN ative advantage in the labor-intensive good, i.e. if pSA N pN N, where pN is the price in the North–North equilibrium, the world cost of capital will increase from its North–North level triggering an increase in unemployment in both Northern countries along their respective RR schedule. Contrary to Davis, the high-regulated developed country does not absorb the whole shock. In contrast, the user cost will fall from its autarky level in South, inducing a decrease in unemployment. Compared with autarky, unemployment raises unambiguously in the high-regulated developed country. In the low-regulated developed country, however, trade with North and with South produces opposite effects, such that the overall change in unemployment is undetermined. 5. Conclusion There is ample evidence that a country's labor market institutions are important determinants of unemployment, either directly or through the propagation of shocks. This paper contributes to the growing awareness that, with globalization, the institutions of trading partners matter also for a country's equilibrium unemployment rate. Because labor market institutions affect relative prices, they contribute to comparative advantages and boost or weaken demand for labor-intensive goods depending on differences in labor market regulations between countries. When labor markets work perfectly, wages adjust to ensure full employment. In contrast, in the presence

of rigidities, shifts in sectoral demands affect unemployment. Consequently, trade might magnify the consequences of the institutional setting on total employment.16 Davis (1998) highlights such a mechanism in a Heckscher–Ohlin trade model between two developed countries with identical factor endowments, in the special case where one country has a ﬂexiblewage and the other a minimum wage labor market. This paper asks whether Davis' results extend to the workhorse labor market search framework of Pissarides (2000), which is embedded into a two-sector context. In Davis, trade between these two countries induces a doubling of unemployment in the minimum wage economy while full employment is maintained in the ﬂexible country. With searchgenerated unemployment, it is shown herein that the unemployment rate increases in the high-regulation country (although it less than doubles), but decreases in the low-regulation one. When trade with a developing country is brought into this setting, comparative advantages result from both relative factor endowments and relative labor market institutions. Assuming that South specializes in the laborintensive sector, unemployment increases in both developed countries relative to their respective North–North integrated trade equilibrium levels. Hence, the speciﬁc Davis' case in which the minimum wage of its Northern trading partner insulates the ﬂexiblewage country from the (negative) impact of trade with South on total employment does not hold any more. In fact, unemployment might even rise more in the low-regulation country, albeit from a lower level, depending on the speciﬁc parameters of both labor markets. Conﬁrmation of these theoretical results could have important policy implications. Economic integration might foster labor market deregulation because of the dissuasive costs of maintaining a high level of regulation in the open economy. In addition, incentives to non-cooperatively deregulate the labor market might be reinforced with integration and, by anticipation, the desire to preserve the socalled “social models” might create resistance to opening up to lowregulation economies. Also, deregulating the labor market might generate a negative externality for trading partners, raising the possibility that cooperation in setting labor market policies enables to reach a better equilibrium. Acknowledgments I am grateful to Matthieu Crozet, Romain Duval, Cyril Nouveau, Joaquim Oliveira-Martins and the participants of the Fourgeaud seminar for their helpful comments, as well as to the editor and two anonymous referees for having provided guidance to upgrade the paper substantially. References Baccaro, L., Rei, D., 2005. Institutional determinants of unemployment in OECD countries: a time series cross-section analysis (1960–1998). ILO Discussion Paper, No 160. Baker, D., Glyn, A., Howell, D.R., Schmitt, J., 2005. Labor market institutions and unemployment: a critical assessment of the cross-country evidence. In: Howell, D. R. (Ed.), Fighting Unemployment: The Limits of Free Market Orthodoxy. Oxford University Press. Bassanini, A., Duval, R., 2006. Employment patterns in OECD countries: reassessing the role of policies and institutions. OECD Economic Studies 42 (1), 7–86. Blanchard, O., 2006. European unemployment: the evolution of facts and ideas. Economic Policy 21 (45), 5–59. Blanchard, O., Wolfers, J., 2000. The role of shocks and institutions in the rise of European unemployment: the aggregate evidence. Economic Journal 110, C1–C33. Boulhol, H., 2009. Do capital market and trade liberalization trigger labor market deregulation? Journal of International Economics 77 (2), 223–233. Boulhol, H., Dobbelaere S., Maioli S., forthcoming. Imports as product and labor market discipline. British Journal of Industrial Relations. doi:10.1111/j.1467-8543.2009.00760. x/abstract. 16 Eventually trade might “discipline” labor market institutions; Boulhol et al. (forthcoming) provide some evidence that rising imports from developed countries exerted downward pressure on workers' bargaining power in the United Kingdom.

H. Boulhol / Journal of International Economics 83 (2011) 83–91 Brecher, R., 1974. Minimum wage rates and the pure theory of international trade. Quarterly Journal of Economics 88, 98–116. Cuñat, A., Melitz, M., 2007. Volatility, labor market ﬂexibility, and the pattern of comparative advantage. NBER Working Paper, No. 13062. Davidson, C., Matusz, S., 2005. Trade and turnover: theory and evidence. Review of International Economics 13 (5), 861–880. Davidson, C., Martin, L., Matusz, S., 1999. Trade and search generated unemployment. Journal of International Economics 48, 271–299. Davis, D.R., 1998. Does European unemployment prop up American wages? National labor markets and global trade. American Economic Review 88 (3), 478–494. Dutt, P., Mitra, D., Ranjan, P., 2009. International trade and unemployment: theory and cross-national evidence. Journal of International Economics 78, 32–44. Helpman, E., Itskhoki, O., 2010. Labor market rigidities, trade and unemployment. Review of Economic Studies 77 (3), 1100–1137.

91

Krugman, P.R., 1995. Growing world trade: causes and consequences. Brookings Papers on Economic Activity 1, 327–362. Moore, M.P., Ranjan, P., 2005. Globalisation vs skill-biased technological change: implications for unemployment and wage inequality. Economic Journal 115, 391–422. Nickell, S., Nunziata, L., Ochel, W., 2005. Unemployment in the OECD since the 1960s. What do we know? Economic Journal 155, 1–28. Pissarides, C., 2000. Equilibrium Unemployment Theory. MIT Press, Cambridge, MA. Saint-Paul, G., 2004. Why are European countries diverging in their unemployment experience? Journal of Economic Perspectives 18 (4), 49–68. Samuelson, P.A., 1962. Parable and realism in capital theory: the surrogate production function. Review of Economic Studies 29, 193–206.

Journal of International Economics 83 (2011) 92–94

Contents lists available at ScienceDirect

Journal of International Economics j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j i e

Book reviews China's Growing Role in World Trade, Robert C. Feenstra, Shang-Jin Wei (Eds.), The University of Chicago Press, 2010

China's transformation and rapid integration into the global economy has been extraordinary. In 1990 China had 21% of the world's population, but only about 1.5% of the total world trade. After two decades of mercurial economic growth, China is now the world's largest exporter with trade representing nearly 10% of the world total. How has this growth occurred and how has it affected markets for goods, labor, and capital both inside and outside of China? This is a critical set of questions, of great interest to academics, managers and policymakers alike. It is also an extremely broad set of questions, and this ambitious NBER volume attempts to wrap its arms around all of them. The book is long (nearly 600 pages), wide ranging (14 chapters on distinct topics, plus an extensive introduction by the editors), and draws on a contributor cast of thousands (okay, 43) well known for their expertise in international trade and ﬁnance. Helpfully, most of the book's chapters appeared in the NBER Working Paper series, so the reader has the option of sampling chapters there. The research here is data-intensive, in many cases exploiting rich new datasets and providing a wealth of descriptive tables that are worth skimming even if the time required for a more careful read proves daunting. Especially illuminating is the use of statistics that separate “processing” trade, in which foreign inputs are imported, assembled and exported, from “ordinary” trade, as well as the ability to separate trade on the basis of ownership and identify FIE's or “foreign invested enterprises”. More on this below. While few of the chapters offer signiﬁcant theoretical or methodological advances, most of them provide nice applications of important recent empirical methods. Indeed, graduate students wanting a primer in applied technique in international economics would be well served to read the book through. Apart from capturing a wealth of insights about China's trade, it provides an instructive collection of distinct data and empirical methods brought to bear on related questions. Turning to particulars, what do we learn about China? To begin, I'd suggest reading Feenstra and Wei's introductory distillation of the chapters to follow. Apart from setting the scene nicely, they provide both a useful summary of the main ﬁndings and some context in which to place these ﬁndings. When I am reading slightly outside my area of expertise, I always ﬁnd it useful to know in advance what I should or should not ﬁnd surprising. Given the broad range of topics here, their quick backgrounder will be helpful to almost all readers. In fact, if you have a free hour put this review down right now and grab their introduction from the NBER working paper series (Feenstra and Wei, 2009). You'll get a quick summary of 20 years of Chinese trade, a great sense of the volume, a useful compendium of statistics, and a table of contents which you can then use to ﬁnd the rest of the volume in the working paper series. Because the editors have provided an excellent and easily accessible summary, this book review is not going to run through the table of contents, remarking on each chapter in turn. Instead, it will proceed thematically through some questions and what we've learned, but then also discuss questions still on the table.

China's exports grew 10-fold in the last two decades. How did they grow? The opening chapters describe a variety of ways to decompose this growth on the basis of broad industrial aggregates, intensive v. extensive margins, factor intensity, and unit values. A number of previous authors have noted seemingly unique features of Chinese exports relative to countries at a similar level of development. These features include greater skill intensity, higher quality (as measured by unit values), and the importance of Chinese exports in providing “new goods” for importers. The contributions in this volume — especially Chapter 2 (Amiti and Freund) and Chapter 3 (Wang and Wei) — discount all these, showing that what is seemingly anomalous about Chinese trade can be attributed entirely to processing exports. That is, China's ability to shift its export bundle from simple labor-intensive manufactures into complex goods may seem extraordinary but it is really a matter of combining labor with sophisticated imported inputs. Further, that processing is primarily being done by foreign invested enterprises (FIE). To underscore this point, machinery and electronics are by a wide margin the largest category of processing exports with 63% of the total, and have increased 37-fold since 1992. It was not always so. Textiles and footwear dominated early on, representing 43% of processing exports in 1992 but only 10% in 2006. Meanwhile, the share of textiles and footwear in “ordinary” exports has been about one-third of the total throughout. This looks like rapid industrial evolution and growing sophistication, but it is only occurring in processing. Completing the story, processing exports themselves have undergone a revolution of ownership, with FIEs now comprising 84% of the total. Putting the boot in, Chapter 12 (Blonigen and Ma) points out the dearth of evidence that the FIEs have generated spillovers to Chinese-owned ﬁrms, and by some measures the Chinese-owned ﬁrms appear to be falling further behind their foreign-owned counterparts. On the one hand, that seems to resolve the anomalies of Chinese export sophistication: it is not, in a strict sense, Chinese sophistication. It is imported sophistication. Fine, but that leaves open an enormous question. What is it about China that has drawn in the FIE's, and encouraged them to ramp up the sophistication of their activities? Does the sheer size of the Chinese economy provide some special inducement? Perhaps ﬁrms want to establish beachheads in China to access the domestic consumption market. But it seems that these goods are primarily produced for export not domestic use. Or maybe the vast number of workers is an attraction, offering a highly elastic labor supply to the manufacturing sector, or the capability to produce an unusually diverse “industrial base” of inputs. But, there are other populous economies in the world, and calculations in Chapter 5 (Feenstra and Hong) suggests that employment growth due to exporting activity was not that great, adding only 2.5 million workers per year in an economy of 1.3 billion. Further, the whole idea of processing exports based on sophisticated foreign inputs suggests that FIE's aren't relying on supplier networks in China. If not size, then what? There are some hints at the role of policy. Chapter 3 (Wang and Wei) use differences in policy applied to an alphabet-soup of special zones within China and ﬁnd some systematic differences in the types of goods emanating from particular zones. More conventionally, Chapter 9 (Brambilla, Khandelwal, and Schott)

Book reviews

shows that a relaxation of Multi-Fibre Accord quotas after WTO accession led to rapid export growth in the affected categories. These points aside, my sense is that policy alone is not enough to explain the special case of China. Or put another way, do we think that another country facing a similar reduction of tariffs and selectively relaxing restrictive policies against entry would have attracted the same level of investment and trade growth? Perhaps China is sui generis. With a handle on how China's trade grew, if lacking a similarly strong grasp on why, we can turn to the question of who was affected. Presumably, China's export success reﬂects the ability to produce goods that are better or cheaper than competitors, which should mean dropping prices for importers and reduced market shares for competitor nations. Chapter 6 (Broda and Weinstein) argues that import prices have not dropped much for Japan, acquitting China of responsibility for deﬂation there. Chapter 4 (Hanson and Robertson) calculate the degree to which China's exports have depressed exports for other developing countries, and conclude that the effect has been quite modest. A similar leaning against received wisdom can be found throughout the volume. Exports play a smaller role than previously thought in affecting Chinese domestic employment (Chapter 5, Feenstra and Hong) and declining environmental quality (Chapter 11, Dean and Lovely). Despite the importance of the US as a destination for Chinese exports, Chapter 13 (Branstetter and Foley) shows that US ﬁrms are not particularly important as a source of FDI in China either in the aggregate or in the exporting sector, and that investment in China is not displacing other investment by those ﬁrms. Despite concerns that WTO accession would trigger a wave of anti-Chinese protectionism, anti-dumping cases against China have not been unusually high, nor has China been a common litigant before the WTO (Chapter 8, Bown). Finally, concerns that a reduction in Chinese agricultural protection after WTO accession would impoverish Chinese farmers also appears overblown (Chapter 10, Huang, Liu, Martin and Rozelle.) Has conventional wisdom ever been so wrong on so many counts? Maybe this China thing isn't such a big deal after all. Or perhaps China's export growth is a big deal, just not as extraordinary or catastrophic as hyperbolic public accounts would have had us believe. A sober and data-driven accounting of actual trends, their causes and consequences, has been overdue and is ably delivered by this volume. Reference Feenstra, R.C., Wei, S.-J., 2009. Introduction to “China's Growing Role in World Trade”. NBER Working Paper 14716.

David Hummels Purdue University, 100 S. Grant St, West Lafayette, IN 47907-2056, United States Tel.: +1 765 494 4495. E-mail address: [email protected].

doi:10.1016/j.jinteco.2010.10.004

Challenges in central banking: The current institutional environment and forces affecting monetary policy, Pierre L. Siklos, Martin T. Bohl, Mark E. Wohar, Cambridge University Press, 2010

Challenges in Central Banking is a bricolage of a book, covering a wide range of issues relating to monetary and ﬁnancial policy. The volume's eleven chapters grew out of papers that were originally presented at a conference held in Budapest in 2007, a time of general consensus on best practices in central banking. Four propositions in particular had emerged from this consensus: ﬁrst, that granting

93

independence to central banks reduced inﬂation; second, that ﬂexible inﬂation targeting delivered optimal macroeconomic performance; third, that central banks should be as transparent as possible in communicating their intentions to the public; and fourth, that delegating ﬁnancial supervisory authority to an independent agency would allow the central bank to pursue its monetary policy objectives more effectively. The worthy goal implicit in many of the papers collected in the volume is to critically reexamine this conventional wisdom. These four consensus desiderata were overwhelmed by the 2007– 2009 ﬁnancial crisis, which presented central bankers with an entirely new (or at least forgotten) set of challenges. These problems included overcoming the zero lower bound on nominal interest rates, determining the efﬁcacy and implementation of unconventional policies, and the appropriate response of monetary policy to ﬁnancial conditions. This volume's inconvenient timing means it was unable to address directly any of these problems. Its focus is instead on issues of institutional design: incentive contracts, committee structure, independence, transparency, operating procedures, and the location of the supervisory function. Recent events have pushed many of these issues to the background, but they remain important questions in the study of central banking. The chapters collected in the volume, most of which are surveys, provide an excellent compendium of current thinking on these topics. The book largely avoids issues relating to optimal design and performance of monetary policy rules, a technical topic that has come to dominate the recent literature on monetary policy. The sole exception is the opening chapter by Frank Smets, “Is the Time Ripe for Price Level Path Stability in the New Keynesian Model?”, which takes up the important question of whether a price level target is preferable to the implicit or explicit inﬂation targets employed by most central banks. The argument for a price level target is familiar from Svensson (1999) and others. The logic is compelling: with forward-looking expectations, credibly targeting the price level generates countercyclical inﬂation expectations, which in turn create stabilizing ﬂuctuations in the real interest rate. Smets deftly recapitulates the rationale, using simulations of a New Keynesian model to illustrate and quantify the policy's beneﬁts. He then considers and then largely dismisses the objections that uncertainty and lack of credibility render the policy unworkable, and comes down squarely in favor of targeting the price level. The volume contains two contributions that turned out to be quite prescient in light of recent events. The ﬁrst of these is “Analysis of Financial Stability,” by Charles A.E. Goodhart and Dimitri P. Tsomocos, which consists of two distinct and only tenuously related sections. The ﬁrst is a brief but rich historical perspective on central banks' role in promoting ﬁnancial stability, duties that encompass both lender of last resort and prudential regulation functions. The second half summarizes a theoretical model of illiquidity and default developed by the same two authors in a series of papers written in collaboration with Pojanart Sunirand. The model, a descendant of Diamond and Dybvig (1983), purports to offer central bankers a framework for assessing the probability of default during crises, although no exercise along those lines is presented in the chapter. “National Banks in a Multinational System,” by David Mayes and Geoffrey Wood, complements the Goodhart–Tsomocos paper with its discussion of the challenges faced by national central banks in coordinating international responses to ﬁnancial crises. Their conclusion, which has been borne out by recent events, is that while liquidity problems can be handled in the traditional manner, the resolution of insolvent banks is far harder to handle without the appropriate transnational institutions. The chapter by Donato Masciandaro, Marc Qintyn and Michael W. Taylor, “Independence and Accountability in Supervision Comparing Central Banks and Financial Authorities,” looks at the governance of the ﬁnancial regulatory authority, a function that in recent years has been spun off from many central banks and delegated to independent

94

Book reviews

agencies. The authors observe that ﬁnancial supervisors choosing between forbearance and failure, face a time consistency problem—a dilemma illustrated all too vividly during the recent ﬁnancial crisis. The standard prescriptions of independence and accountability would presumably mitigate this problem, just as in the case of monetary policy. The chapter's main undertaking is to quantify these attributes for a group of 55 countries, and to assess empirically the impact of various economic, institutional and political factors on independence and accountability. The reduced-form empirical analysis produces some intriguing results, including the puzzling ﬁnding that removing ﬁnancial supervision from the central bank does not enhance supervisory independence, although accountability tends to increase. Two chapters revisit the well-trod ground of central bank transparency and independence. Both provide excellent surveys of the relevant literature. “The Economic Impact of Central Bank Independence: A Survey,” by Carin van der Cruijsen and Sylvester C.W. Eijfﬁnger is breathtakingly comprehensive—both in its recapitulation of the theoretical arguments for and against transparency, and in its summary of twenty years' worth of research the topic. Particularly impressive is the paper's tabular summary of 53 theoretical and 47 empirical papers written on the topic in recent years. Adopting a refreshingly skeptical point of view, “The Complex Relationship Between Central Bank Independence and Inﬂation,” by Bernd Hayo and Carsten Hefeker, critically surveys the vast literature on central bank independence. The prevailing interpretation of Alesina and Summers (1993), and its often-overlooked antecedent Bade and Parkin (1988), is that granting independence leads to lower inﬂation. Hayo and Hefeker question this presumption, arguing that central bank independence is to a large extent endogenous. Independence and low inﬂation may jointly depend on the “inﬂation culture” and historical conditions; or, as argued by Posen (1993, 1995), on the strength of the ﬁnancial sector's opposition to inﬂation. Two chapters are noteworthy for the wealth of institutional information they contain. The ﬁrst of these is Corinne Ho's contribution, “Implementing Monetary Policy in the 2000s: Operating Procedures in Asia and Beyond.” The chapter's overarching message is that central banks should no longer be thought of as simply manipulating the quantity of reserves in order to achieve the desired overnight interest rate. Most central banks have discontinued the use of quantitative operational targets, and many (though not the U.S. Federal Reserve) have de-emphasized discretionary operations in favor of regular standing facilities. The chapter provides a valuable tabular summary of the main features of central banks' operating procedures, although its usefulness is somewhat limited by its focus on East Asian economies. In a similar spirit, Philipp Maier's chapter, “How Central Banks Take Decisions: An Analysis of Monetary Policy Meetings,” contains an encyclopedic tabulation of the attributes of the policy committees of 43 central banks. Included in the tabulation are such features as the number of committee members, their backgrounds, how the meetings are organized, and who makes the interest rate proposal. While Maier does not attempt to grade or rank the banks' decision-making processes, he does identify several features that can affect the likely prevalence of problems free-riding, groupthink, and information cascades. On these criteria, Maier credits the Bank of England with the closest adherence to best practice. The remaining pair of papers is a study in contrasts. Both deal with the broad question of how to design a central bank to achieve low inﬂation, but one paper takes a very narrow theoretical approach,

while the other is a broad, reduced-form econometric analysis. The theoretical perspective is supplied by Georgios E. Chortareas and Stephen M. Miller in their chapter, “The Principal–Agent Approach to Monetary Policy Delegation,” which reviews the familiar Barro and Gordon (1983) time inconsistency problem as it applies to monetary policy, and describes how incentive contracts can be used to solve the problem. While attractive theoretically, no central bank has yet adopted a contract-based solution. One explanation is the institutional environment within which central banks operate is far richer than the single selﬁsh decision-maker model used in this approach; another is that the time inconsistency problem is simply not relevant for most central banks. Pierre L. Siklos' concluding chapter, “Institutional Rules and the Conduct of Monetary Policy,” abandons the highly stylized principal–agent model, hypothesizing instead that inﬂation is determined by many aspects of good governance—not just central bank autonomy—that collectively contribute to a sense of trust. He supports his view with an informal appeal to signaling theory, and with a reduced-form regression of inﬂation performance on various proxies for governance quality. The general conclusion is that many of these indicators do seem to matter, albeit in ways that are not immediately apparent in a reduced-form regression framework. While thin in terms of new theory and original empirical work, this eclectic collection makes a valuable contribution with its surveys of central bank transparency and independence, its discussion of ﬁnancial stability, and its compilation of data on central banks' practices. It is a worthwhile addition to the bookshelf of any connoisseur of central banking. References Alesina, A., Summers, L.H., 1993. Central Bank independence and macroeconomic performance: some comparative evidence. Journal of Money, Credit and Banking 25, 151–162. Bade, R. and M. Parkin, 1988, Central Bank Laws and Monetary Policy, unpublished manuscript, University of Western Ontario. Barro, R.J., Gordon, D.B., 1983. A positive theory of monetary policy in a natural-rate model. Journal of Political Economy 91, 589–610. Diamond, D.W., Dybvig, P.H., 1983. Bank runs, deposit insurance and liquidity. Journal of Political Economy 91, 401–419. Posen, A.S., 1993. Why Central Bank independence does not cause low inﬂation: there is no institutional ﬁx for politics. In: O'Brien, R. (Ed.), Finance and the International Economy 7. Oxford University Press, New York, NY. Posen, A.S., 1995. Declarations are not enough: ﬁnancial sector sources of Central Bank independence. In: Bernanke, B.S., Rotemberg, J.J. (Eds.), NBER Macroeconomics Annual. The MIT Press, Cambridge, MA, pp. 253–274. Svensson, L.E.O., 1999. Price-level targeting versus inﬂation targeting: a free lunch? Journal of Money, Credit and Banking 31, 277–295.

Kenneth N. Kuttner Williams College, Department of Economics, 326 Schapiro Hall, 24 Hopkins Hall Drive, Williamstown, MA 01267, United States NBER, United States Tel.: +1 413 597 2300; fax: +1 413 597 4045. E-mail address: [email protected].

doi:10.1016/j.jinteco.2010.10.003

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents