Wirtschaftswissenschaftliche Fakultät der Eberhard-Karls-Universität Tübingen

Transcription

1 Wirtscaftswissenscaftlice Fakultät der Eberard-Karls-Universität Tübingen Sorting it Out: Tecnical Barriers to Trade and Industry Productivity Gabriel J. Felbermayr Benjamin Jung Tübinger Diskussionsbeitrag Nr. 315 Februar 2008 Wirtscaftswissenscaftlices Seminar Molstraße 36, D Tübingen

2 Sorting It Out: Tecnical Barriers to Trade and Industry Productivity Gabriel J. Felbermayr and Benjamin Jung Eberard Karls University, Tübingen, Germany February, 2008 Abstract Trade economists traditionally study te effect of lower variable trade costs. Wile increasingly important politically, tecnical barriers to trade (TBTs) ave received less attention. Viewing TBTs as fixed regulatory costs related to te entry into export markets, we use a model wit eterogeneous firms, trade in differentiated goods, and variable external economies of scale to sort out te ric interactions between TBT reform, input diversity, firm-level productivity, and aggregate productivity. We calibrate te model for 14 industries in order to clarify te teoretical ambiguities. Overall, our results tend to suggest beneficial effects of TBT reform but also reveal interesting sectoral variation. Keywords: Heterogeneous firms, international trade, single European market, tecnical barriers to trade, regulatory costs. JEL-Codes: F12, F13, F15 We are grateful to David Greenaway, Wilelm Koler, Julien Prat, Davide Sala, and seminar participants at te OeNB Worksop 2007 on International Trade and Domestic Growt in Vienna, te 1st FIW researc worksop in Vienna, seminars at te universities of Nottingam, Leicester and Tübingen for comments and discussion. All remaining errors are ours. gabriel.felbermayr@uni-tuebingen.de; benjamin.jung@uni-tuebingen.de. Address: Economics Department, Eberard Karls University Tübingen, Nauklerstrasse 47, Tübingen, Germany.

3 1 Introduction In te last fifty years, import duties on most relevant manufacturing goods ave fallen substantially. A rising fraction of total trade is covered by free trade agreements and is terefore exempt from tariffs. Yet, even witin te European Union only about 10% of total spending falls on products from oter EU15 countries (Delgado, 2006). Cen (2004) explains tis striking fact i.a. by te existence of tecnical barriers to trade (TBTs). TBTs impose additional export market access costs. Exporters must customize teir goods to meet te import country s tecnical norms, its ealt, safety, or environmental norms, and must undergo costly product labeling and conformity assessment procedures. Bot te European Union (EU) and te World Trade Organization (WTO) acknowledge tat TBTs may serve a multitude of legitimate goals; owever, regulation tat effectively protects incumbent domestic firms against foreign competition is deemed discriminatory and is terefore illegal. Witin te context of te Single Market Programme (SMP), te EU campions mutual recognition of tecnical standards. 1 However, Ilzkovitz, Dierx, Kovacs, and Sousa (2007) argue tat wile about 20% of industrial production and about 26% of intra EU manufacturing trade are covered by mutual recognition, practical implementation [...] is often ampered by legal uncertainty, administrative assle and lack of awareness bot from te side of te companies and of te Member States autorities (p. 61). Progress in dismantling TBTs as been slow. Te number of TBT-related complaints notified to te WTO as grown from 365 in 1995 to almost 900 in 2006 (WTO, 2007). Similarly, Conway, Janod, and Nicoletti (2005) document te persistence of discriminatory regulation in te OECD. According to te Fraser Institute, te stringency of regulatory barriers to trade as increased in many EU countries from (see Gwartney, Lawson, Sobel, and Leeson, 2007). TBT related issues are increasingly important in trade negotiations. Indeed, armonizing standards 1 Te principle of mutual recognition mandates tat a product lawfully marketed in one EU country sould be allowed to be sold in any oter EU country even wen te product does not fully comply wit te tecnical rules in te destination country. However, countries can refuse market access for public safety, ealt, and environmental reasons (Articles 28 and 30 of te EC Treaty; similar regulation appears in te WTO TBT Agreement in Article 2.) 2

4 and rules rater tan abolising tariffs and quotas are te real 21st century trade issues (Pascal Lamy). 2 Despite te importance of TBTs, te teoretical literature as usually focused on variable trade costs suc as transportation costs or tariffs. In tis paper, we model TBT liberalization as a reduction in te fixed regulatory costs of foreign market access. We study two scenarios. In te first, te reduction of regulatory burdens on importers is accompanied by domestic deregulation suc tat te implicit protection of domestic firms, T, is uncanged. We call tis situation T- neutral deregulation. In te second scenario, regulatory requirements imposed on importers are reduced wile tose on domestic firms remain uncanged. We call tis case incremental mutual recognition, since it leads towards a situation of full mutual recognition were meeting domestic regulation is enoug to access foreign markets. Bot scenarios are relevant empirically. We analyze tese two scenarios in a model of international trade in differentiated goods wit eterogeneous firms. Out setup is essentially te one of Melitz (2003), wit te difference tat we focus on aggregate productivity (not welfare) and allow for variable degrees of external scale economies in te final good production function (as Egger and Kreickemeier, 2007). We do tis, because recent literature (e.g., Corsetti, Martin, and Pesenti, 2007) as establised ow important te size of te scale effect is as a major determinant of te qualitative and quantitative implications of trade liberalization. Moreover, empirical work points towards substantial industry variance and generally rejects te implicit numerical coice of te scale effect parameter embodied in te traditional formulation of te Melitz (and, indeed, most Krugman (1980)- type trade models). As many oter autors, 3 we work wit a specific productivity distribution (Pareto) to sort out ambiguities and to parameterize te model for simulation purposes. In te proposed framework, TBT reform affects te equilibrium input diversity (i.e., te mass of imported and domestic varieties) available in an industry, wic affects te productivity of final goods producers troug an external effect. TBT reform also modifies te equilibrium productivity distribution of input producers and, ence, teir average productivity. Tese two 2 EU and Asean to pave way for trade pact talks, Financial Times, 7 September At tat time, Lamy was EU trade commissioner. He is now Director-General of te WTO. 3 Egger and Kreickemeier (2007), Baldwin and Forslid (2006), Helpman, Melitz, and Yeaple (2004), etc. 3

5 forces determine te effect on industry productivity, wit teir relative importance given by te external scale elasticity. Incremental mutual recognition canges te extensive margin of firm beavior; i.e., it modifies te selection of input producers into exporting and domestic sales. 4 It also affects te intensive margin, as additional competitive pressure lowers sales per firm. Te two effects lead to reallocation of resources towards medium-productivity new exporters, away from te upper and lower areas of te productivity distribution. Te net reallocation effect tat drives average productivity of input producers depends on te relative importance of tese two countervailing reallocation effects. Also te effect on input variety is teoretically unclear. It depends on industry caracteristics; e.g., on te degree of productivity dispersion. It is terefore not surprising tat te total effect of TBT reform on industry productivity is a complicated function of model parameters. Te contribution of tis paper is to analytically sort out tose ambiguities. Te teoretical analysis as a couple of interesting implications. First, it may rationalize te low robustness of a positive relationsip between trade openness and economic growt (see Rodríguez and Rodrik, 2000). Bot variable and fixed cost trade liberalization lead to a iger volume of trade, tereby increasing openness. However, for similar parameter constellation, te former unambiguously improves productivity wile te latter does not. Second, our paper suggests tat te productivity effect of lower variable trade costs is importantly conditioned by te existence of fixed costs protection. Indeed, if TBTs are too ig, lower transportation costs may turn out to lower industry productivity. We offer an industry-by-industry calibration exercise in order to numerically validate weter te conditions old under wic TBT reform improves aggregate outcomes. In tis local analysis, TBT reform turns out to reduce te productivity of te average input supplier. However, te increase in input variety more tan compensates tose losses, so tat industry productivity improves. We also simulate te model and compare te status quo wit a situation were tecnical requirements are armonized across countries. Tis being a global exercise and te 4 Te selection effects rely on firm eterogeneity. Firms select temselves into exporting according to teir productivity. Tere is overwelming empirical evidence tat tis is indeed te case, see te survey by Helpman (2006). 4

6 effects of TBT reform being non-linear, it turns out tat not all industries (e.g., macinery) gain from te reform. Many do gain, but only very modestly, wile oters experience massive productivity improvements (e.g., 35% in te case of scientific equipment). Our paper is related to a number of studies, many of tem inspired by te Single Market Programme. Using a partial equilibrium framework, Smit and Venables (1988) simulate te abolisment of trade barriers in terms of tariff equivalents between European countries. Keuscnigg and Koler (1996) simulate te general equilibrium growt and welfare effects of lower variable trade costs in a multi-sector Krugman-type model, were scale economies play an important role. A similar simulation is done by Francois, Meijl, and van Tongeren (2005). Te latter autors imulate a simultaneous cut in tariffs and TBTs, and obtain a real income gain of 0.3% to 0.5% of global GDP, depending on te country coverage. Te older literature uses models wit omogeneous firms and studies te effects of lower variable trade costs. More recently, Del Gatto, Mion, and Ottaviano (2007) and Corcos, Del Gatto, Mion, and Ottaviano (2007) focus on te productivity effects of intra-eu variable trade costs reduction under quasi-linear preferences wit eterogeneous firms and provide simulation results. Our paper differs, since we use te Melitz (2003) model as a point of departure and relate TBT to fixed costs of market access. Baldwin and Forslid (2006) provide an excellent discussion of trade policy in te standard Melitz (2003) model. Tey also address lower market access costs and discuss te implication for te trade volume. Our paper differs from teirs in tat it sorts out te intricate implications of TBT reform on industry productivity. Moreover, we allow for variable degrees of external scale economies and offer a calibration exercise. Te remainder of te paper is organized as follows. Capter 2 introduces te analytical framework and solves for general equilibrium. Capter 3 teoretically derives conditions under wic TBT reform increases productivity, and Capter 4 calibrates te model in order to validate tese conditions for different industries. Finally, Capter 5 concludes. 5

7 2 Teoretical framework 2.1 Demand for inputs We study a single market (suc as te EU) wit n + 1 identical countries. Eac country is populated by a representative consumer wo as symmetric Cobb-Douglas preferences for final consumption goods produced by H industries. Final output producers in eac industry are perfectly competitive. Tey assemble teir output using a continuum of inputs q(ω) according to te same constant elasticity of substitution (CES) production function y = M η 1 σ 1 ω Ω σ 1 σ q (ω) dω σ σ 1, σ > 1, η 0. (1) Te set Ω represents te mass of available inputs in industry, and σ is te elasticity of substitution between any two varieties in tat industry. M is te measure of Ω and denotes te degree of input diversity (te number of available differentiated inputs). Expression (1) is analogous to te traditional CES production function for η = 1. 5 For η = 0, tere are no external economies of scale. In te standard treatments of te Melitz (2003) or te Krugman (1980) models, te implicit coice of η = 1 links te effect of input diversity on output directly to te elasticity of substitution σ. standard formulation (Ardelean, 2007). Recent empirical work finds tat η < 1, rejecting te Te optimal demand quantity for eac input ω is ( ) p (ω) σ R /P q (ω) = P M 1 η, (2) were R is aggregate industry spending on inputs, p(ω) is te price carged by an input producer to te final output producers, and P = M η σ 1 1 ω Ω p (ω) 1 σ dω 1 1 σ (3) 5 Te generalization is already discussed in te working paper version of te Dixit-Stiglitz (1977) paper and as been revived by Benassy (1996). Variants of it ave been adopted by Blancard and Giavazzi (2003), Egger and Kreickemeier (2007), Corsetti, Martin, and Pesenti (2007), or Felbermayr and Prat (2007). 6

8 is te price index dual to (1). Clearly, demand for variety ω is larger te smaller te price p (ω) relative to te average price of competing varieties P, and te larger real spending R /P. Higer input diversity M affects demand troug two cannels: indirectly, troug its effect on te price level, and, if η 1, directly, troug te reduction of relevant real spending (R /P ) /M 1 η on eac variety ω. Markups over marginal costs are constant in tis framework; neverteless we find it useful to call M a competition effect. 2.2 Production of inputs Differentiated inputs are produced by a continuum of monopolistically competitive firms. Eac industry draws on a single industry-specific factor L, wic is inelastically supplied in equal quantities to all industries in all countries. Industry specificity of factors and te Cobb-Douglas utility function make sure tat trade reforms generate only witin rater tan between-industry resource reallocation. Input producers differ wit respect to teir productivity index ϕ; in te following we use tis index instead of ω to identify firms. 6 Tey sare te same domestic and foreign market entry costs, f d and f x, and te same iceberg variable trade costs τ 1. All fixed costs ave to be incurred in terms of te industry-specific factor. T f x /f d measures te competitive disadvantage of imported relative to domestically produced inputs. To ensure te existence of te selection effect (and in line wit empirical evidence) we assume τ σ 1 T > 1. Following Melitz (2003), we assume tat firms are ex ante identical but face uncertainty regarding teir productivity ϕ. Tey learn about ϕ only after sinking te entry cost f e. Not all of tose entrants turn out to be productive enoug to bear te domestic fixed costs f d. Hence, tey remain inactive. Firms wit intermediate productivity sell on te domestic market, but cannot recover te additional fixed costs associated to foreign sales, f x. Te most productive firms are active on all markets. Under te assumption τ σ 1 T > 1, tere exist tresold productivity levels 0 < ϕ < (ϕx ), wic partition te distribution of input producers into 6 Tis is possible because, in equilibrium, eac input is produced by one firm only and te distribution of ϕ is assumed to ave no mass points. 7

9 inactive firms, purely domestic ones, and exporters. We caracterize te ex ante productivity distribution by te Pareto. 7 Te c.d.f. is G (ϕ) = 1 ϕ γ wit support on [1, ), were te sape parameter γ > σ 1 controls te dispersion of te distribution. 8. Larger values of γ caracterize industries in wic te productivity distribution is skewed towards inefficient input producers. We can ten write te probability tat a given entrant (tat as just paid te entry fee f e ) starts production by pin = 1 G (ϕ ) = (ϕ ) γ. Analogously, p x = 1 G[(ϕx ) ] = [ϕ 1 G(ϕ ) / (ϕx ) ] γ is te ex-ante (and ex-post) probability tat one of tese successful entrants will export. Input producers ave linear production functions q(ϕ) = ϕl (ϕ), were l (ϕ) denotes te employment of te industry- specific factor in firm ϕ. Profit maximization of input producers results in te standard rule for determining te ex-factory (f.o.b.) price, i.e. p (ϕ) = w / (ρ ϕ), were ρ = 1 1/σ. Since te description of tecnology (1) is identical over all countries, we may pick te factor price specific to some industry, w, as te numeraire. In te following, we focus on tat industry. Optimal demand (2) and te pricing rule of input producers imply tat revenues earned on te domestic market are given by r d (ϕ) = R (P ρ ϕ) σ 1 /M 1 η. (4) By symmetry, producers wo find it optimal to sell to a foreign market generate revenues of r (ϕ) = r d (ϕ) ( 1 + nτ 1 σ ). In turn, profits from selling domestically and exporting to one foreign market are respectively given by π d (ϕ) = rd (ϕ) /σ f d, (5) π x 1 σ (ϕ) = τ r d (ϕ) /σ f d T. (6) 7 Tis assumption is not necessary for many properties of te model; see Melitz (2003). However, it allows to understand te importance of industry dispersion to sort out te potentially ambiguous effects of various forms of trade liberalization on industry productivity. Te Pareto as been used, i.a., by Melitz, Helpman, and Yeaple (2004), Baldwin and Forslid (2006) or Egger and Kreickemeier (2007). It fits well empirically, see Axtell (2001) or Corcos, Del Gatto, Mion, and Ottaviano (2007). 8 Te assumption γ > σ 1 makes sure tat te equilibrium sales distribution converges. 8

10 2.3 Industry aggregation Te productivity of final output producers (industry productivity) depends on input diversity (te number of available inputs), and on te average productivity level of input producers. Input diversity as a domestic and an imported component: M = M d + nm x, were n is te number of identical import (and, by symmetry: export) markets. Since M x = px M d, one can express M as M = M d (1 + npx ). Te average productivity level of domestic input producers, ϕ d, is defined as te mean over sales-weigted productivities of all active producers. 9 Using te Pareto assumption, ( ϕ ) d σ 1 ϕ ϕ σ 1 dg = (ϕ) 1 G ( ) ϕ = γ γ (σ 1) (ϕ )σ 1. (7) Equation (7) sows tat te endogenously determined entry cutoff productivity level ϕ sapes te average productivity of domestically produced inputs. Te average over exporters, ϕ x, is constructed analogously, and crucially depends on te export cutoff productivity level (ϕ x ). Given perfect symmetry across countries, te average productivity of inputs used in production of te final good, ϕ, is ( ϕ ) σ 1 = np x ( ϕ d ) σ 1 + np x 1 + np x ( ϕ x τ ) σ 1, (8) were productivities of foreign firms are adjusted for iceberg transportation costs τ and average productivities of domestic and imported varieties are weigted by teir respective sares in total input diversity. Te weigting in (8) implies tat q ( ϕ ) = R M η +σ 1 σ 1 /P. In η = 0 (i.e., in te absence of industry externalities), te output of te average firm is equal to average output R / (P M ). Similarly, applying (8) to te industry price index (3), one as P = M η σ 1 p ( ϕ ). Hence, if η = 0, te price index is equal to te price cosen by te average firm. Using optimal pricing of inputs in P and recognizing tat aggregate productivity A = 1/P, we are now ready to write te level of aggregate industry productivity as a function of average 9 See Melitz (2003), p

11 productivity and input diversity η σ A = ρ ϕ M 1. (9) Industry productivity 10 increases as ρ goes up so tat markups and te amount of resources used for fixed costs are lower. Industry productivity is directly proportional to average productivity of input producers ϕ. It depends positively on input diversity M as long as η is strictly positive. Te term η σ 1 is te elasticity of industry productivity wit respect to input diversity.11 Te aim of te subsequent analysis is to understand ow A canges wit different types of TBT reform. To do tis, we need to endogenize ϕ and M. Typically, TBT liberalization moves tese two components of industry productivity in opposite directions. Hence, te elasticity play a crucial role. η σ 1 will 2.4 General equilibrium In tis section, we solve for te equilibrium values of M and ϕ. Te discussion is deliberately brief, since it is close to Melitz (2003) and to Baldwin and Forslid (2006); te only difference comes troug η 1. Equilibrium is determined by four conditions. Zero cutoff profit (ZCP) conditions. Te domestic ZCP condition identifies te firm ϕ tat is indifferent between selling domestically and remaining inactive; te foreign ZCP condition locates te firm (ϕ x ) tat is indifferent between selling domestically and also selling on te n symmetric foreign markets. Formally, te ZCPs are π d (ϕ ) = 0, πx [(ϕx ) ] = 0. (10) Using te profit functions derived in (5) and (6), te zero cutoff profit conditions imply tat r d (ϕ ) = σ f d and rd [(ϕx ) ] = σ f dτ σ 1 T. Ten (4) links te export cutoff (ϕ x ) and te 10 Tis is te ideal measure of industry productivity. Measuring productivity empirically is not trivial (see Levinson and Petrin, 2003). Gibson (2006) points out tat productivity effects induced by Melitz (2003)-type reallocation of market sares witin industries are not reflected by data-based measures of productivities, (e.g., value added per worker). 11 If η = 1, (9) is formally equivalent to te expression describing total welfare in Melitz (2003). 10

12 domestic entry entry cutoff ϕ 12 1 σ 1 (ϕ x ) = ϕ τ T. (11) Moreover, using te definition of ϕ d (7) and te condition linking te two cutoff productivities (11), one can link te average productivity of domestic firms wit tose of exporters 1 σ 1 ϕ x = τ T ϕ d. (12) It follows from (11) tat te probability of exporting conditional on successful entry is given by p x = τ γ T γ σ 1. (13) Our results so far allow to express average productivity ϕ, as defined in (8), by using (12) ϕ = ϕ d ( 1 + np x T 1 + np x ) 1 σ 1. (14) Bot te domestic and te foreign market ZCPs can be combined and graped in (ϕ, π ) space by using te definition of average profits (defined over active firms, ex post perspective) π = π d ) d ( ϕ + np x x π x ( ϕx ) and noting tat ϕd and ϕx are bot functions of ϕ. It is well known tat, given te Pareto assumption, average profits π do not depend on ϕ.13 Free entry. Te free entry condition ensures tat expected profits (from te ex ante perspective) cover entry costs f e : were δ is te exogenous Poisson exit rate of producers and p in p in π δ = f e, (15) = (ϕ ) γ is te likeliood tat a random productivity draw allows a producer to at least break even on te domestic market. Clearly, tis free entry condition defines an upward-sloping relationsip between π and ϕ. Equating tat condition wit te combined ZCP condition discussed above, one can determine te entry cutoff productivity level ϕ as a function of exogenous variables only14 [ ϕ = σ 1 f d γ (σ 1) δ f e 12 Derivations of analytical results are detailed in te Appendix. 13 See, e.g., Baldwin and Forslid (2006). 14 Noting tat (13) relates p x to exogenous variables. ] 1 (1 + np x T γ ). (16) 11

13 Substituting ϕ into te definition of te domestic average productivity (7) one can determine ϕ d. Finally, using (13) and (14) allows to compute te average productivity defined over all input producers, ϕ. Note tat te above analysis as not used any factor market clearing condition; ϕ is terefore independent from L. Moreover, wen solving for ϕ, input diversity is irrelevant. Input diversity M can be found recursively, i.e., given ϕ. Stationarity condition. Te fourt equilibrium condition allows to pin down input diversity. In a stationary equilibrium, in any country, te mass of successful entrants p in M e te mass of producers it by te exit sock δ M d. Hence, must equal p in M e = δ M d. (17) As sown in Melitz (2003, p. 1704), under stationarity, aggregate revenue R is fixed by te size of L (and te normalization of te factor price). Tis determines downs equilibrium input diversity by M = R /r d ( ϕ ). Using te zero cutoff profit conditions (10), we obtain equilibrium industry diversity M = L σ f d ( ϕ ϕ ) σ 1. (18) 3 Industry productivity effects of TBT reform Totally differentiating industry productivity (9) yields  = η σ 1 ˆM + ϕ. (19) We use te conventional at notation to denote an infinitesimally small deviation of a variable from its initial level (ˆx = dx/x). Any type of trade liberalization as potential implications for te cutoff productivity levels ϕ and (ϕx ), and ence for productivity averages of domestic and international firms, ϕ d and ϕx, respectively. Te productivity level of te average firm ϕ is a weigted average over domestic and international firms, wit te relative weigts potentially being affected by TBT reform, too. Different trade liberalization scenarios may ave similar effects on cutoff productivities (te extensive margin); yet, tey may lead to drastically different 12

14 patterns of inter-industry resource reallocation along te intensive margin, and, ence, different results for industry productivity. Te literature as not fully recognized tis point yet. Input diversity adjusts to canges in te entry cutoff ϕ and average productivity ϕ suc tat factor markets clear (see equation (18)). For te assessment of industry productivity, bot effects need to be combined, wit te elasticity η / (σ 1) playing a crucial role. Hence, te overall effect of TBT reform on industry productivity works troug a number of different mecanisms and is likely to be ambiguous teoretically. In te extreme case were η = 0 (Blancard and Giavazzi, 2003), (19) simplifies substantially as variation in input diversity as no bearing on industry productivity. Also te case were η = 1, typically studied in te literature, turns out offer more clear-cut results. In tis special case ((19)) is formally isomorpic to te description of welfare in te Melitz (2003) model. Te contribution of te present paper is to discuss te empirically relevant situation were η (0, 1) and to focus on TBT reform rater tan on te more widely studied case of variable trade cost liberalization. We assume tat fixed market costs f d and f x ave two components: fixed distribution costs, f d and f x, and fixed regulatory costs, f d and f x, tat relate to approval and conformity assessment costs. Te latter is set by national autorities, but differs from a tax since it does not generate revenue. We define as a TBT reform any policy measure tat reduces regulatory costs for foreign firms f x. Hence, armonization of standards, i.e., f d = f x, need not be a TBT reform. Full-fledged mutual recognition of standards, in contrast, would make licensing procedures for imported varieties redundant, ence f x = 0. Only in tis case do TBTs disappear entirely. We consider two scenarios of TBT reform. In te first, policy makers lower te burden on foreign firms f x, but also adjust regulatory costs for domestic firms f d suc tat te competitive disadvantage of foreign firms, T, remains uncanged. We term tis case T-neutral deregulation. In te second scenario, f x is reduced, wile f d remains fixed. Any marginal reduction in f x brings te economy closer to te ideal situation of full mutual recognition. Hence, we call our second scenario incremental mutual recognition. Trougout, we assume tat distributionrelated fixed costs are suc tat te partitioning of firms into exporters and purely domestic firms is maintained (i.e. f x /f d > τ 1 σ ) Deregulation of entry costs (f e ) is beyond te scope of tis paper. Lower entry costs induce additional 13

15 3.1 T-neutral deregulation In tis scenario, f x and f d bot fall, but T remains constant. Terefore, te export probability p x (13), wic depends on fixed market access costs only troug T, is fixed. It is also clear, tat te domestic ϕ and te export cutoff productivity levels (ϕx ) move proportionally (see (11)). To understand te effect of T-neutral deregulation, note tat ϕ is determined in (ϕ, π ) space by te intersection of te ZCP condition and te free entry condition. In te present context, te first is a orizontal line, wile te latter is upward-sloping. Domestic deregulation does not affect te free entry locus. However, te ZPC condition sifts downwards, so tat ϕ falls. Te reasoning is as follows. Te ZCP locus summarizes combinations of π and ϕ marginal firm ϕ just breaks even. Wen fixed costs f d for wic te fall, te firm starts to make profits. To restore zero profits, te firm s revenue as to fall. Tis is acieved by tigter competition: eiter relative prices ave to increase or residual demand as to drop. Tis is owever not limited to firm ϕ ; profits fall for all firms; ence π goes down. Te effect on te cutoff productivities at and, one can now use Figure 1 to gain some intuition on te reallocation of market sares tat domestic deregulation entails. Te figure sows sales r (ϕ) per firm as a function of productivity. Tis locus is upward-sloping as more efficient firms ave iger sales (given σ > 1). Since total sales R are pinned down by L, r can be read as a measure of market sare. Te sales function canges wit T-neutral deregulation. 16 Due to te increase in te number of traded varieties, competition goes up, wic means tat incumbent exporters and domestic-only firms lose market sare (intensive margin). Since domestic and foreign market entry costs fall in proportion, te probability of exporting, given successful entry, does not cange (see equation (13)). Moreover, te entry cutoff levels sift proportionally. Hence, te reallocation of market sares towards less productive firms directly translates into a decrease in average productivity. We sall discuss te effect on average entry, wic increases competition and reduces realized profits, resulting in a (proportional) sift of te cutoff productivity levels to te rigt and an increase in average productivity (see Bernard, Redding, and Scott, 2005, and Felbermayr and Prat, 2007, in sligtly different settings). 16 Figure 2 in Melitz (2003) wic studies te reallocation of market sares as an economy moves from autarky to trade. Our Figure 1 is similar, but studies incremental trade liberalization. 14

16 Figure 1: Witin-industry reallocation of market sares as response to T-neutral deregulation. productivity, input diversity, and industry productivity in more detail below. Average productivity of input producers. Te cange in average productivity is completely driven by te cange in te entry cutoff productivity level, i.e. ϕ / ˆf d = ˆϕ / ˆf d (see (14) and (7)). Totally differentiating (16) yields 17 ϕ ˆf d = 1 γ > 0. (20) Tus, average productivity declines in response to T-neutral deregulation. Te parameter γ is inversely related to te degree of productivity dispersion (eterogeneity) in te industry. In te extreme case were γ, all firms are identical and tere cannot be any selection or reallocation effect (as long as all firms remain exporters or purely domestic). Te room of reallocation is bigger as γ is smaller and industry eterogeneity is larger. It is terefore natural tat te effect of T-neutral deregulation on average productivity is larger te smaller γ. 17 Recall tat canges in te regulatory component directly translate into canges in total market access costs, d i.e. ˆf = b f d 15

17 Input diversity. As argued above, lowering fixed market entry costs attracts new input producers to start production and makes it profitable for additional firms to export. Te cange in input diversity is given by ˆM = ˆf ) d + (σ 1) (ˆϕ ϕ < 0. In te present scenario, te entry cutoff productivity level ˆϕ and average productivity ϕ move proportionally. Hence, te elasticity of input diversity wit respect to f d is ˆM ˆf d = 1. (21) Industry productivity. Te industry productivity effect combines te input diversity effect and te effect on input producers productivity. Tis leads to te following proposition. Proposition 1 (T-neutral deregulation) Industry productivity only increases in response to T-neutral deregulation, if te degree of external economies of scale is larger tan te inverse dispersion measure of te Pareto η σ 1 > 1. (22) γ Proof. Follows from using (20) and (21) in (19). Hence, te elasticity of aggregate productivity wit respect to input diversity as to be sufficiently large in order to overcompensate te loss in average productivity. Note tat, in te case of η 1, te above inequality always olds (by te regularity condition γ > σ 1). Hence, domestic deregulation always makes te final goods producer more productive. However, tis result is not general: in te empirically relevant case, were η < 1, te industry productivity effect is ambiguous. 3.2 Incremental mutual recognition Tis scenario implies a reduction of T wit f d eld constant. Consider again te determination of te domestic cutoff productivity ϕ in (ϕ, π ) space. Te free entry condition does not cange as T falls. However, te ZPC condition now sifts up, so tat ϕ goes up. Te marginal domestic producer is not an exporter; ence tere is no direct effect of te reduction in f x. However, te entry of foreign importers makes competition touger, revenue per firm goes 16

18 Figure 2: recognition. Witin-industry reallocation of market sares as response to incremental mutual down, and te ϕ firm starts to make losses. To restore zero profits, tere must be an upwards adjustment of π. Te number of competitors or teir average productivity (or bot) ave to go down. Hence, ϕ increases wile (ϕx ) goes down. Figure 1 provides some intuition on te reallocation of market sares: te emergence of new exporters causes a loss of market sare to incumbent exporters and domestic firms. Since new exporters are firms wit medium levels of productivity, te net effect on average productivity is ambiguous. Average productivity of input producers. ϕ only increases in response to a cut in T if te sape parameter γ is large enoug. Te intuition is straigtforward: Te larger te sape parameter γ, te more mass is given to low productive firms, tus giving a ig potential for reallocation from fairly unproductive, exiting firms to new exporters. If te initial level of competitive disadvantage of importers is already smaller tan 1, tere is almost no export selection effect, and ϕ never increases in response to a TBT reform regardless 17

19 of te sape parameter γ. We may summarize te result in Lemma 1. Lemma 1 (Average productivity) Fix f d and reduce T. Average productivity ϕ increases in response to incremental mutual recognition if and only if te te dispersion measure of te Pareto distribution is large enoug, i.e. ϕ < 0 1 < ˆT γ γ σ np x Proof. Follows immediately from totally differentiating (14). T 1 T (23) At te extensive margin, te least productive firms are forced to exit (selection effect), wile new exporters enter (adverse export selection effect). Only if te selection effect is large enoug as compared to te adverse selection effect, ϕ rises as stated in condition (23). Input diversity. If te productivity distribution is not extremely skewed towards te least productive firms (i.e. if te sape parameter γ is sufficiently small), te number of input varieties lost troug exposure to trade is overcompensated by additionally imported inputs, resulting in an increase in input diversity. Lemma 2 (Input diversity) Fix f d and reduce T. Input diversity M increases in response to incremental mutual recognition if and only if te dispersion measure of te Pareto γ is sufficiently small, i.e. ˆM < 0 1 > 1 γ 1 ˆT γ σ np x T 1 T. (24) Proof. Follows immediately from totally differentiating (18). Lemma 2 presents a necessary condition (24). T < 1. Note tat a simple sufficient condition is Industry productivity. Using (19) and Lemmata 1 and 2, average productivity and input diversity increase in response to a incremental mutual recognition, if te value γ is not too extreme, i.e., if γ > γ > γ. (25) 18

20 Ten, industry productivity improves unambiguously regardless te degree of external economies of scale. However, even if condition (25) is violated, industry productivity can actually increase, depending on te degree of external economies of scale. If ϕ is falling and M rising, te degree of external economies of scale as to be sufficiently large for industry productivity to increase, and vice versa. Tere exists te following trade-off: If te sape parameter γ is sufficiently small unproductive firms ave little relative mass. Hence, tere is little potential for reallocation from te exiting, low-productivity firms to new exporters. Ten average productivity declines. In contrast, input diversity increases, since more imported varieties are attracted tan domestic ones are forced to exit. If, on te oter and, te sape parameter is γ is sufficiently large, te logic reverses, and average productivity increases wereas input diversity declines. Consider tat input diversity decreases in response to incremental mutual recognition, wic means a violation of condition (24) in Lemma 2. Ten, by condition (23) average productivity unambiguously rises, and te degree of external economies of scale as to be sufficiently small. Te negative diversity effect is always offset for te empirically relevant cases η 1. Turn now to te case were average productivity declines in response to incremental mutual recognition, i.e. a violation of condition (23) in Lemma 1. Ten, by (24) industry diversity always increases, and η / (σ 1) as to be sufficiently large to generate an increase in industry productivity, wic is always true for te special Melitz case (η = 1). Tese results are summarized in te following Proposition. Proposition 2 (Incremental mutual recognition) Let υ be te tresold degree of external economies of scale ( υ 1 γ γ γ σ 1 γ ). γ (i) Violation of condition (24). A decrease in input diversity in response to incremental mutual recognition is overcompensated by an increase in average productivity, if and only if te degree 19

21 of external economies of scale is below te tresold value υ, i.e. were υ > 1/ (σ 1). η σ 1 < υ, (26) (ii) Violation of condition (23). A decrease in average productivity in response to incremental mutual recognition is overcompensated by an increase in input diversity, if and only if te degree of external economies of scale is above te tresold value υ, i.e. were 0 < υ < 1/ (σ 1). η σ 1 > υ, (27) Proof. Te conditions follow from equation (19). 3.3 Comparing lower variable trade costs and TBT reform As wit incremental mutual recognition, lower variable trade costs induce an upward-sift in te ZCP. Te reason for tis effect is te same as before. Hence, tariff liberalization (or any reduction in variable trade costs) as similar effects on te cutoff productivity levels as lower T wit f d fixed. However, lower trade costs on net benefit incumbent exporters, as additional competitive pressure is over-compensated by lower trade costs. 18 It follows, tat te direction of market sare reallocation is unequivocally towards more productive firms. Note, owever, tat te sales function depicted in Figure 3 does not suffice to determine te effect on average productivity, wic depends on te masses of firms engaged in exporting relative to purely domestic ones. It turns out tat te productivity effect is a priori ambiguous and depends on T, wic governs te size of te selection and export selection effect. If T > 1, imported inputs are on average more productive (tey ave to cover iger fixed market entry costs). Tis implies lower prices and, in turn, given CES preferences, results in iger expenditure. Tus, more tan one domestically produced input as to be displaced in 18 Tis result olds for all productivity distributions. 20

22 Figure 3: Witin-industry reallocation of market sares as response to variable trade cost liberalization. order to import one additional input variety, and input diversity drops. 19 Reallocation of market sares towards more productive firms and te reduced availability of te least productive inputs, result in iger average productivity. If T < 1, we end up wit iger input diversity. It turns out tat in tis case average productivity actually declines. 20 For te empirical relevant case T > 1, input diversity drops at te lower end of te productivity distribution, resulting in an increase in average productivity. As mentioned above, te condition under wic average productivity increases is less strict: Lemma 3 (Average productivity) Average productivity increases in response to variable trade cost liberalization if and only if te dispersion measure of te Pareto distribution is small 19 A similar explanation as been put forward by Baldwin and Forslid (2006). 20 T > (<) 1 is a necessary condition for input diversity to decrease (rise), wereas for average productivity to increase (drop) it is a sufficient condition. Te necessary condition would be less strict and depend on te skewness of te productivity distribution. 21

23 enoug, i.e. ϕ < 0 1 > 1 ˆτ γ σ np x T 1 T. (28) Proof. Follows from totally differentiating (14). Condition (28) clearly olds if M decreases, i.e. T > 1. However, average productivity also rises if te selection effect is sufficiently large, sifting input production to more productive firms. Proposition 3 (Lower variable trade costs) Let ψ economies of scale ψ 1 σ T γ T 1 (1 + npx ). be te tresold degree of external (i) Assume T > 1, so tat input diversity decreases and average productivity increases. Ten industry productivity goes up if and only if te degree of external economies of scale is below te tresold value ψ, i.e. η σ 1 < ψ. (29) (ii) Assume T < 1 and a violation of condition (28), so tat input diversity increases and average productivity decreases. Ten industry productivity increases if and only if te degree of external economies of scale is above te tresold value ψ, i.e. Proof. Te conditions follow from totally differentiating (18). η σ 1 > ψ. (30) Conditions (29) and (30) always old if respectively η 1, and η 1. Hence, in te special Melitz case (η = 1), industry productivity always increases in response to variable trade cost liberalization. In contrast, incremental mututal recognition reduces te market sares of existing exporters, tereby inducing reallocation of market sares towards less productive firms, and T > 1 is not sufficient to guarantee an increase in average productivity. Tere are tree interesting corollaries tat follow from te comparison between TBT reform and variable trade cost reductions. First, in te empirically relevant case T > 1 and η 1, 22

24 lower variable trade costs unambiguously improves industry productivity, wile te effect of TBT reform is still ambiguous. However, in bot situations, total export sales increase. 21 Hence, tere is no clear link between increased trade openness and industry (or even economy-wide) productivity measures. Tis teoretical result rationalizes te low degree of robustness tat empirical cross-country analyses of te openness-productivity (or more often: GDP per capita) link suffer from; see, e.g., Rodríguez and Rodrik (2000). Second, te effect of lower trade costs is conditioned by te importance of competitive disadvantage of foreign firms as measured by T. We ave seen above, tat if T > 1 lower variable trade costs may lead to a fall in industry productivity. In oter words: industries can be urt by reductions in tariffs or transportation costs if te degree of fixed-cost protection is too ig. Tis allows two policy conclusions: first, before engaging in variable trade cost reforms, countries sould lower TBTs. Only countries wit sufficiently low TBTs benefit from te (exogenous) downward trend in transportation costs. Hence, productivity gains from tecnical progress in transportation can be tapped only if TBTS are low enoug. Tird, tere seems to be substantial resistance against TBT reforms. Gwartney, Lawson, Sobel, and Leason (2007) argue tat te EU25 countries ave failed on average to decrease regulatory costs to importers. Our paper allows two interpretations of tis result. First, based on efficiency considerations, TBT reform is not desirable per se, at least not under arbitrary parameter constellations. Second, TBT reform even if it leads to industry productivity gains inflicts losses to te vast majority of firms due to te implied reallocation of resources towards new exporters by nature a relatively small fraction out of all domestic firms. Hence, it may not be overly surprising tat total resistance against TBT reform is strong, and, in particular, stronger tan against lower variable trade cost reductions, wic tend to be beneficial for incumbent exporters. 21 Total sales abroad are given by X cif finds tat X cif = L np x T / (1 + np x T ), and X cif / T < 0. = nm x r x ( ϕ x ). Recall tat M x = p x M d. Using (4), (17) and (18) one 23

25 4 Numerical exercise at te industry level In tis capter, we use estimates of te key parameters of our model from te literature or calibrate tem according to te model. Since tere is substantial cross-industry variation of parameters, we do a separate analysis for 14 industries. Te numerical exercise serves several purposes. First, it allows to calibrate te degree of external economies of scale, η / (σ 1), and te level of competitive disadvantage of importers, T. Second, it enables us to ceck te inequalities derived in te teoretical section of tis paper and to empirically sort out te ambiguous effects of different trade liberalization scenarios, industry by industry. Finally, te exercise allows to compute te productivity gains (or losses) relative to status quo acieved by setting T = 1, i.e., to a situation, were tecnical requirements are armonized. 4.1 Calibration Several studies quantify te elasticity of substitution and productivity dispersion on industry level for US and European data. However, we do not ave estimates from a structural econometric approac, in wic σ and γ are separately identified under te relevant regularity conditions tat ave to old in te present teoretical framework. 22 Corcos, Del Gatto, Mion, and Ottaviano (2007) estimate industry-level dispersion measures γ using European data. 23 Teir estimates of γ are on average close to 2. Caney (2007) sows tat elasticities of substitution obtained from standard gravity models are distorted under te presence of eterogeneous firms. Terefore, we draw on estimates of te sape parameter of te sales distribution ς = γ (σ 1) obtained from Helpman, Melitz, and Yeaple (2004) to back out te values of σ given te estimates of γ. 24 Our sources for ς and γ bot are consistent wit eterogeneous firm models and use te same European firm-level data (Amadeus). Since ς is close to 1 for all industries, te values of σ cluster around Table 1 reports our findings 22 For example, γ > σ Tey estimate firm-level productivities using te Levinson and Petrin (2003) estimator, and fit a Pareto distribution for eac industry. For all industries te regression fit (te adjusted R squared) is close to In oter words: we take γ and ς as data and calibrate σ. 25 Our values of σ seem low; owever, te are consistent wit oter estimates, e.g., tose by Acemoglu and 24

26 Table 1: Parameter description (preferred specification) Data Calibration Industry τ p x η γ σ T η σ 1 η min σ 1 η max σ 1 Cemicals Rubber and plastics Leater and footwear Lumber and wood Paper products Textile Apparel Non-ferrous metals Macinery except electrical Electrical macinery Road veicles Transport equipment Scientific/measuring equip Optical/potograpic equip Notes. τ from Hanson and Xiang (2004); η from Ardelean (2007). T calibrated to meet export participation rate p x from Eaton, Kortum, Kramarz (2004). n calibrated to meet openness of 40%. σ imputed from sape parameters estimated by Corcos, Del Gatto, Mion, and Ottaviano (2007) and sales dispersion measures from Helpman, Melitz, and Yeaple (2004) 25

27 for 14 sectors. 26 Ardelean (2007) provides te first industry-level estimates of te parameter tat governs te external scale effect. Se identifies η by decomposing te price index into a traditional part and te extensive margin, following Feenstra (1994), and exploiting cross-importer variation. For all industries, se rejects te standard assumption of η = 1. On average, er estimate of η is Given te importance of tis parameter, and te fact tat te available estimates are for te US (wile our calibration targets Europe), we run tree scenarios. (A) uses te estimates found by Ardelean and allows for industry variation. (B) disallows for industry variation and sets η for all industries to te lowest available estimate found in Ardelean η min. (C) is similar to (B) but sets η = η max. Note tat even in (B) and (C) te elasticity of A wit respect to input variety (η / (σ 1)) still exibits industry-level variance. We take data on industry transport costs from Hanson and Xiang (2004). Using data on freigt rates for U.S. imports from Feenstra (1996), tey identify te implicit U.S. industry freigt rate (insurance and freigt carges/import value), and regress it on log distance to te origin country. Transport cost for an industry are reported as te projected industry freigt rate from tese coefficient estimates evaluated at median distance in teir sample of importers and exporters. 27 Finally, we calibrate te competitive disadvantage of importers T suc tat te model replicates te export participation rates p x by industry reported by Eaton, Kortum, and Kramarz (2004) for European firms. 28 Equation (13) sows tat our coice of τ, γ and σ and te observed values of p x directly imply T. 29 Our calibration yields values of T varying between 1 and 3, in all industries strictly above unity (see Table 1). Tis finding is well in line wit te oter calibration exercises in te literature, e.g., Gironi and Melitz (2005). Te calibration Ventura (2002). We conduct some robustness analysis wit respect to σ below. 26 Table 4 in te Appendix reports ow data organized in different industry classifications as been mapped into our sectoral structure (wic is essentially tat of Corcos, Del Gatto, Mion, and Ottaviano, 2007). 27 We are grateful to Gordon Hanson for providing tose estimates. 28 More specifically, teir data is from France. 29 We also use information on openness (40%), te average transportation costs, and te average T to calibrate te number of trading partners n. 26