WPS7156 Policy Research Working Paper 7156 Firm Heterogeneity and Costly Trade A New Estimation Strategy and Policy Experiments Ivan Cherkashin Svetlana Demidova Hiau Looi Kee Kala Krishna Development Research Group Trade and International Integration Team January 2015 Policy Research Working Paper 7156 Abstract This paper builds a tractable partial equilibrium model to and European Union. Counterfactual experiments regard- help explain the role of trade preferences given to developing ing the effects of reducing costs, both fixed and marginal, countries, as well as the efficacy of various subsidy policies. or of trade preferences offered by an importing country The model allows for firm level heterogeneity in demand are performed. The counterfactuals show that reducing and productivity and lets the mass of firms that enter be fixed costs at various levels has very different effects and endogenous. Trade preferences given by one country have suggest that such reductions are more effective in promot- positive spillovers on exports to others in this model. Pref- ing exports when applied at later stages when firms are erences given by the European Union to Bangladesh in more committed to production. A subsidy of 1.5 million an industry raise profits, resulting in entry, and some of dollars to industry entry costs raises exports by only 0.4 these firms also export to the United States. The parameters dollars for every dollar spent, but when applied to fixed of the model are estimated using cross sectional customs costs of production, it raises exports by $25 per dollar spent. data on Bangladeshi exports of apparel to the United States This paper is a product of the Trade and International Integration Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at hlkee@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Firm Heterogeneity and Costly Trade: A New Estimation Strategy and Policy Experiments. Ivan Cherkashin Svetlana Demidovay The Australian National University McMaster University Hiau Looi Keez Kala Krishnax World Bank The Pennsylvania State University and NBER Keywords: Rules of Origin, Firm Heterogeneity, Demand Shocks, Policy Experiments F12, F14, F17 Research School of Economics, College of Business and Economics, Australian National University, ACT, Australia, 0020. E-mail: ivan.cherkashin@anu.edu.au. Phone: +612 612 55637. y Department of Economics, McMaster University, Hamilton, ON, Canada L8S 4M4. E-mail: demi- dov@mcmaster.ca. Phone: (905) 525 9140, ext.26095. Fax: (905) 521 8232. z Development Research Group –Trade, The World Bank, 1818 H ST NW (MSN MC3-303), Washington, DC 20433. E-mail: hlkee@worldbank.org. Phone: (202) 473 4155. Fax: (202) 522 1159. x Department of Economics, Pennsylvania State University, University Park, PA, 16802. E-mail: kmk4@psu.edu. Phone: (814) 865 1106. Fax: (814) 863 4775. 1 Introduction When the US granted duty-free and quota-free access to Madagascar, under the African Growth and Opportunity Act 2000, exports from Madagascar exploded, from $170 million s exports to the rest of in 2000 to $500 million in 2004. Over the same period, Madagascar’ the world also increased, from $750 million to $875 million. Similarly, based on Comtrade export data, when the EU granted duty-free and quota-free access to Bangladesh under the Everything But Arms Initiative in 2001, knitwear exports from Bangladesh to the EU more than doubled, rising from $823 million to $2,351 million between 2000 to 2004. During the same period of time, knitwear exports from Bangladesh to the US increased from $316 million to $465 million.1 Exports to countries other than the US and EU rose from $60 million to $190 million.2 To the surprise of many, such generous trade preferences resulted not in trade diversion from the rest of the world to the preference granting markets, but in trade creation - tempered by the presence of quotas. The model we develop and estimate in this paper predicts exactly these changes. Trade preferences given by one country have positive spillovers on exports to others in the presence of free entry. For example, preferences given by the EU make the industry more attractive, induce entry, and some of these entering …rms export to the US so that exports to both countries rise. We use customs data from Bangladesh to estimate a heterogeneous …rm model based on the ‡agship model of Melitz (2003), but structured to be suitable for trade policy applications. Our work takes a heterogeneous …rm model literally, and confronts it with micro data and actual trade policies, to estimate all of its structural parameters, including the various levels of …xed costs. These …xed costs are at the core of the model and serve as hurdles that productive/fortunate …rms choose to jump, while those that are less so do not. Our paper then uses the estimated model to evaluate the e¤ects of the di¤erent 1 This small increase is not unexpected, given that the presence of quotas on most of these items in the US constrained export growth there. 2 If we base the numbers on import data reported by Comtrade, we get a similar pattern though the numbers are somewhat di¤erent. In the EU, knitwear imports from Bangladesh more than doubled, from $1.3 billion in 2000 to $3 billion in 2004. At the same time the US imports from Bangladesh increased by $30 million, and imports from Bangladesh to all countries other than the US and EU rose by $287 million. 3 kinds of trade polices used in practice. Finally, we compare …xed cost subsidies of various kinds in terms of their e¤ectiveness in promoting exports. In our model, there are two sources of …rm heterogeneity: …rm speci…c productivity as in Melitz (2003), and …rm and market speci…c demand shocks, which is motivated by the …ndings in Demidova, Kee and Krishna (2012). They use a …rm level data set on Bangladeshi garment producers and show that …rms roughly follow the productivity hierarchy predicted in Melitz (2003), namely, that …rms export to all markets that are easier than the toughest one they export to, and more productive …rms export to tougher markets. However, there are a number of violators. While these violators are small in terms of their numbers, they are large in terms of their output. This fact can be rationalized by introducing …rm and market speci…c demand shocks. Such shocks allow us to explain why, given its productivity, a …rm may be very successful in one market but not the other.3 We chose not to use the approach of Arkolakis (2010), who argues that …rms have a choice of penetration costs that increase with the number of consumers …rms want to access and decrease with the market size, which allows small exporters to exist - something that would be ruled out by large …xed costs of entry. However, even with his approach, but without the presence of …rm and market speci…c demand shocks, there would be a very strong positive correlation in the size s market shares across export destinations, something we do not see in our data. of the …rm’ To explain the data we need …rm and market speci…c demand shocks, as postulated here. Such demand shocks do much of the work in …tting the data, which is also consistent with the work of Roberts et al. (2012) and Eaton et al. (2011). In addition to two dimensional heterogeneity, we also incorporate, albeit simply, various real world trade policies; tari¤s, preferences, rules of origin, and quotas into our model. We focus only on the partial equilibrium interaction between Bangladeshi …rms and take the prices and actions of other …rms operating in the EU and US as …xed. 3 Eaton, Kortum and Kramarz (2011) also postulate the existence of …rm and market speci…c demand shocks. Kee and Krishna (2008) look at the patterns in the violations and what might explain them. Armenter and Koren (2014) assume heterogeneity on the …xed cost side. They show that matching the share of exporters in a standard Melitz model to the data results in having exports per …rm far larger than in the data. Fixed costs heterogeneity helps to reduce this mismatch and explain hierarchy violations. 4 A closely related paper in the literature is the work of Eaton, Kortum and Kramarz (2011) (EKK from here on). EKK use customs-level data to understand the patterns of French …rms’ exports. Their focus is on constructing the simplest model that …ts most of the facts, rather s (2010) market than on trade policy. They also add a reduced form version of Arkolakis’ access costs to explain the presence of many small …rms with a limited attachment to the market, as well as …rm and market speci…c demand shocks. We see their work as very complementary to ours. They look at the “big picture” and try to match the patterns in …rm-level exports by all French …rms, in all industries, to all countries. As a result, their model is unsuited to zooming in on a particular industry and incorporating the relevant trade policy details as our model is designed to do. Moreover, and perhaps more critically, their model, like that of Chaney (2008), assumes that the mass of potential entrants is …xed. In contrast, we treat the mass of entrants as endogenous. Since we show that this entry margin does most of the heavy lifting in the adjustments that occur in response to policy, such di¤erence in our assumptions is worth emphasizing. Our paper is also related to Bernard, Redding and Schott (2011), which also features market demand shocks in order to determine the export behavior of multi-product …rms. Our model has two policy-relevant predictions. First, it suggests that a small country can increase its exports quite considerably if granted easily accessed preferences, and through cost-reducing policies. We explicitly show how to incorporate these preferences and the costs, both …xed and variable, associated with obtaining them into a structural model that is suitable for estimation and policy analysis. Conversely, factors that raise export costs, like corruption or bad infrastructure, can really take a toll on exports. Second, the model suggests that preferences to developing countries can have a catalytic e¤ect. In our model, preferences given by one developed country can signi…cantly raise the exports to the other market rather than diverting trade away from other markets as predicted in standard competitive settings. This occurs because preferences raise the expected return to entry in the industry. Once a …rm has entered the industry, it may end up exporting to markets other than those where it was given preferences if it gets an adequate demand shock. Our low estimate of elasticity of substitution between products means that entering …rms make room for themselves in product space, so that greater entry does not quickly drive down pro…ts, which magni…es 5 the entry e¤ect of preferences. The e¤ects of such policies are, of course, blunted by the presence of quotas in other markets. In our estimation, we simulate our model and then match the generated distributions to those in the data.4 In this paper, we use only cross sectional price and quantity information and are able to generate bootstrap standard errors for our estimates. The advantage of this approach is that such cross sectional data is commonly available, which makes our procedure widely applicable in contrast to the structural dynamic approach taken in recent work, such as Das, Roberts and Tybout (2007) and Aw, Roberts and Xu (2011), which is limited to where data is available over a period of time. Finally, some caveats. Our model, which we believe captures the essential aspects in question, as usual, has a number of limitations. We make a number of modelling assumptions like constant marginal costs, restricting ourself to a static model etc., which simpli…es things considerably. These assumptions let us separate what happens in the two markets at any point of time. If, for example, marginal costs were not constant, as would be the case with capacity constraints, pricing in one market would depend on demand shocks in the other, resulting in our model being mis-speci…ed and our estimates being biased. However, if capacity constraints were binding, the correlation of sales of …rms that serve both the US and EU markets (AUS …rms) should be negative. For the 155 AUS …rms in our data, this correlation is +0.32 (and signi…cant at the 5% level), consistent with no binding capacity constraints.5 Our structural approach interprets the data through the lens of a heterogenous …rm model. Such models have been widely used in recent work. The literature has focused on whether …xed cost changes or marginal cost changes drive increases in trade into the export market.6 This is not the focus of our work, though we consider the e¤ect of both changes 4 Demidova, Kee and Krishna (2012) take advantage of a natural experiment in trade policy that provides clean predictions regarding how …rms should sort themselves across markets in this augmented Melitz model. They then show that these predictions are consistent with the data. 5 This could be because …rms in apparel can subcontract out and thereby relax capacity constraints at low cost. The absence of such evidence of capacity constraints helps to motivate our assumption of constant marginal costs. 6 The importance of marginal trade costs, such as tari¤s, in explaining the expansion of the extensive 6 in our counterfactuals. Our paper focuses on the entry margin: we …nd that the reductions in both …xed and marginal costs induce a lot of entry of Bangladeshi …rms into the export market, which occurs because ex-ante expected pro…ts shift upward and this increases the equilibrium mass of entrants. This increase is large due to the low estimated elasticity of substitution between goods made by Bangladeshi …rms. Our estimated elasticities, as we argue below, are broadly in line with those in the literature. The paper proceeds as follows. Section 2 contains a brief discussion of the empirical application and the data. Section 3 lays out the model with the details of the derivations in the Appendix. Section 4 lays out the estimation outline. The results are presented in Section 5, while policy counterfactuals are presented in Section 6. Section 7 concludes. 2 The Empirical Application We use data on the woven apparel sector in Bangladesh, in particular, the subcategory Mens/Boys Cotton Trousers (HS 620342). Firms that produce garments from woven fabrics s typically use imported or domestically produced fabrics to make apparel items, such as men’ cotton shirts or ladies dresses, and export the …nished products to the EU or US. Given that fabrics make up more than 50 percent of the costs, the origin of the fabric has implications for the tari¤s faced by these products in each market, due to the existence of trade preferences in some markets. We describe the setting in some detail below as it is the basis for how we margin of trade can be found in Yi (2003), who focuses on multi-stage production, and Kehoe and Ruhl (2013) and Debaere and Mostashari (2010), who focus on the new goods margin. Giovanni and Levchenko (2013) argue that reductions in marginal costs would have a far larger impact than reductions in …xed costs in a world with signi…cant fractions of very large …rms, i.e., where the size distribution of …rms has fat tails as in the U.S. data. In such settings, small …rms contribute little to trade and reductions in …xed costs, which a¤ect the entry of small …rms, do little to increase trade. Lincoln and McCallum (2011) look at the U.S. census data and …nd that in the US, both the number of …rms and their average size grew from 1987 to 2006, consistent with a fall in marginal costs driving entry and larger size. However, they attribute this to rising incomes in export markets rather than to a fall in …xed or marginal cost. On the other hand, by focusing on cross sectional data of a wide range of developed and developing countries, Helpman, Melitz and Rubinstein (2008) show that …xed costs of trade could determine the extensive margin of trade through the selection of …rms into the export market. 7 incorporate the trade policy environment into our model.7 We focus on this subcategory for two reasons. First, we want to have a relatively homo- geneous product industry: woven apparel as a whole might be seen as too aggregated. Next, we want a fair number of …rms to be present. Both requirements are met in this sector.8 2.1 The Trade Policy Environment There are three main components of the trade environment: the trade policy of the US and EU, the trade preferences (if any) they grant to Bangladesh, and the Rules of Origin, or ROOs, upon which preferences are conditioned.9 They can take a variety of forms. The important thing to note is that, whatever the form, if ROOs are binding, then the choice of inputs used in production di¤ers from their unconstrained levels. Thus, from an analytical viewpoint binding ROOs must raise the marginal costs of production. In addition, ROOs can also raise the …xed cost of production as compliance with ROOs must be documented. A large part of these documentation costs involves learning the ropes and can be treated as …xed. We explicitly allow for such costs of meeting ROOs in our model. The US Environment In 2004, the US had tari¤s of about 20% applied on a Most Favoured Nations (MFN) basis, as well as Multi…bre Arrangement (MFA) quota restrictions in place for the imports of wovens from most developing countries, including Bangladesh. Quotas under the MFA were country speci…c, so exporting was contingent on obtaining origin: unless the good was shown to originate from Bangladesh, it could not enter un- 7 The apparel sector of Bangladesh also produces non-woven products, but the production technique and the trade policy environment faced by the non-woven …rms are vastly di¤erent from those of the woven producers. Hence, we exclude them from this paper to simplify the modelling and estimation. See Kee and Krishna (2008) and Demidova, Kee and Krishna (2012) for details. Both papers use the di¤erences in policies applied to woven and non-woven garment exporters to identify the sorting behavior of …rms with unobserved heterogeneity in productivity and market demand shocks. 8 This subcategory accounts for 31.3% of exports to the EU and 15.3% to the US. We have a total of about 800 …rms that export to the US or EU in this category in 2004, and of the …rms that export to the EU, 72.5% meet ROOs and invoke preferences. 9 ROOs specify conditions on production that must be met in order to obtain origin and thereby qualify for country speci…c quotas or trade preferences. For a relatively comprehensive survey see Krishna (2006). 8 der its quota. US ROOs regarding apparel products are governed by Section 334 of the Uruguay Round Agreements Act. For the purpose of tari¤s and quotas, an apparel product is considered as originating from a country if it is wholly assembled there.10 No local fabric requirement is necessary. Thus, the products of a Bangladeshi …rm are not penalized if the …rm chooses to use imported fabrics. Bangladesh did not have any trade preferences in the US and had to compete with producers from other countries such as India and China. However, since there were quotas on other exporters as well, competition among supplying countries was limited.11 In wovens, 65-75% of Bangladeshi exports to the US in value terms are under quota. For the subgroup we focus on this number is close to 100%. These quotas are bilateral and product speci…c, so …rms have no choice but to meet origin.12 Note that quotas in the US have a license price associated with them. As these quotas are binding, it is positive. A …rm survey in Bangladesh conducted by the World Bank in 2004 gave the quota license price to be about 7%.13 The EU Environment In 2004, the EU had a MFN tari¤ rate of 12%-15% on woven apparel. However, Bangladeshi …rms could avoid paying this tari¤: under the “Everything- But-Arms”(EBA) initiative in 2001, Bangladesh together with 48 other LDCs was allowed to export duty and quota free to the EU, provided that ROOs were satis…ed. The EBA initiative e¤ectively removed any inklings of a quota and granted a 100% preference margin for garment exports of Bangladesh to the EU. It signi…cantly improved the market environment in which Bangladeshi garment exporters operated. EU ROOs on apparel products were considerably more restrictive than the US ones. As such, an item exported to the US may be considered as a product of Bangladesh and imported under its quota allocation. However, the same item may fail to meet EU ROOs 10 For details, please see the online Appendix available at the authors’web-sites. 11 Note that less competitive countries are at less of a disadvantage in the US than they would be in the absence of the quota. The quota in e¤ect guarantees them a niche as long as they are not too ine¢ cient. Their ine¢ ciency reduces the price of their quota licenses, while the quota licenses of a very competitive country would be highly priced. 12 The …ll rate of the quota is close to 80% suggesting the quotas are binding. 13 In the survey administered by H.L. Kee to a sample of Bangladeshi woven …rms the average cost increase from having to buy a license was 7%, which is in line with estimates in Mlachila and Yang (2004) for 2003. 9 and would not qualify for the tari¤ preference under EBA. According to Annex II of the GSP (Generalized System of Preferences) guidebook, which details ROOs of all products, for an apparel product to be considered as having originated from a country, it must start its local manufacturing process from yarn,14 i.e., the use of imported fabrics would result in the item failing to meet ROOs for the purpose of tari¤ and quota preferences under GSP or EBA for the case of LDCs. It would, thus, be subject to MFN tari¤s of about 12% to 15% on price. Table 1a: Trade policy environment US EU ROOs ROOs (local assembly) Meeting ROOs (yarn forward rule) is optional are required Option 1: no ROOs met Option 2: ROOs are met Quotas Yes No No Tari¤s t 20% (MFN tari¤) t 12% (MFN tari¤) 0% (Preferences are given) Firms making garments from woven material (woven …rms) mostly assemble cut fabrics into garments. THus, domestic woven cloth was essential to meeting the ROOs. Given its limited supply,15 such cloth commanded a corresponding price premium. The 20% higher price for domestic cloth translates into a signi…cantly higher cost of production, as cloth is s share of the input cost. The cost of cloth to FOB (free on board) price is roughly the lion’ 70-75% for shirts, dresses, and trousers,16 resulting in about a 15% cost disadvantage from meeting ROOs.17 Thus, the costs and bene…ts of invoking preferences under the EBA were similar. For this reason, not all woven …rms chose to invoke these preferences when exporting to the EU. This feature allows us to estimate the …xed documentation costs of invoking 14 For the details on the EBA user guide and Annex II on GSP, please see the online Appendix available at the authors’web-sites. 15 Of 1320 million meters of total demand in 2001, only 190 was supplied locally in wovens according to a study by Development Initiative in 2005. 16 See Table 33 in Development Initiative (2005). 17 In contrast, India has the ability to meet its woven cloth needs domestically at competitive prices so that its …rms can avail themselves of GSP preferences in the EU. As a result, Bangladeshi …rms …nd themselves at a disadvantage in woven garments. 10 preferences and meeting ROOs.18 Table 1a summarizes the trade environment in the US and EU. 2.2 The Data We use customs data at the transaction level for the …scal year 2004 as our main source. This has information on sales, quantity, product (at the HS 8 level), weight, destination, currency of transaction as well as a …rm identi…er. Since the customs data has no information on whether preferences were invoked or not, we obtained a list of …rms that export woven apparel to the EU and obtain preferences for their exports. For these …rms, we obtained information on their exports of wovens to the EU, their quantity and price. However, the …rm IDs in the two data sets are not the same. To match the …rms, we did the following. In our customs data, we aggregated the sales per …rm to the EU of woven apparel, the quantity exported to the EU, and its unit value over the …scal year. We then matched the …rms according to sales and quantity, which resulted in 799 …rms.19 Table 1b provides some summary statistics. Firms that do not meet ROOs are smaller than those that do, especially if they sell to both the US and EU. This is what one would expect, if more productive …rms choose to meet ROOs and enter the US market, which is the tougher one to be in. Table 1b: Summary Statistics Markets …rms export to Only EU Both EU and US Only US Total Do …rms meet ROOs in EU? Yes No Yes No Number of …rms 373 134 107 48 137 799 000) Average sales in EU ($’ 568 294 1,993 373 729 Average price in EU ($ per unit) 3.7 4.4 4.7 5.1 4.1 000) Average sales in US ($’ 922 597 776 800 Average price in US ($ per unit) 4.7 4.4 4.2 4.4 18 We could not estimate documentation costs separately from other …xed costs of exporting if all …rms choose to meet ROOs as in non-wovens. This is the main reason why we focus on the woven sector here. 19 There is little need to worry about mismatches that might have occurred in this procedure as we only target the share of …rms that meet ROOs in our estimation procedure. 11 3 The Model We develop a simple partial equilibrium setting based on the setup in Melitz (2003), to which we add another dimension of …rm heterogeneity: …rm and market speci…c demand shocks. Demidova, Kee and Krishna (2012) use the same setup to see how …rms with di¤erent productivities, facing these demand shocks, are predicted to sort themselves and behave as a result of di¤erences in tari¤s, quotas, and ROOs of the EU and US. The way in which they do so is shown to be consistent with the model. For example, they …nd that, as predicted by the model, the probability a …rm only exports to the EU falls with increases in productivity, favorable demand shocks in the US, and adverse demand shocks in the EU. Conversely, the probability a …rm exports to both the EU and US rises with increases in productivity and favorable demand shocks in the US and EU. They also found evidence suggesting that …rms who only export to the US (whose presence is impossible without demand shocks) are mainly driven by favorable demand shocks in the US together with adverse demand shocks in the EU, but not by productivity. The fact that the predictions of the model are seen in the data allows us to use it in a structural estimation procedure with a fair degree of con…dence that it is not being arbitrarily imposed on the data. While we explicitly model a small open economy partial equilibrium, Melitz (2003) has a general equilibrium model. We do not have the data to con…dently estimate a general equilibrium model and for this reason we stick to partial equilibrium and make the equivalent of a “small country” assumption as explained in more detail later. We also focus only on the US and EU as export markets and do not model the domestic Bangladeshi market at all. The US and EU are most of the Bangladeshi export market, since …rms do not produce much (about 3%) for the domestic market. This is not surprising since the domestic market demands di¤erent products from those exported. We …rst set up the demand side, where we describe preferences and how we incorporate demand shocks into the model. Then we explain the timing of decisions and model how …rms behave in the presence of ROOs. Following this, we outline the equilibrium conditions in our partial equilibrium model. Next, we explain how we estimate our model and provide our estimation results. Finally, we explain the counterfactuals we ran and what they mean. 12 3.1 Utility Utility in country j; j 2 fU S; EU g ; is given by Uj = (Nj )1 (Cj ) ; 0< < 1; (1) where Nj is a competitively produced numeraire good, which is freely traded and takes a unit of e¤ective labor to produce. Cj can be thought of as the services produced by consuming the exports of apparel from all trading partners. Thus, 0 1 j =( j 1) X Cj = @ [Xij ]( j 1)= j A ; (2) i2 j where j is the set of trading partners for country j: Xij denotes the services produced by the exports of a trading partner i to country j that produces and sells a continuum of varieties indexed by ! . q (! ) is the quantity consumed and v (! ) is the demand shock for variety !: A lower value of v (! ) corresponds to a worse demand shock. Let the sub-utility function be Z ! j =( j 1) 1= j ( 1)= Xij = vij (! ) qij (! ) j j d! ; (3) !2 ij where ij is the set of varieties from country i available to consumers in country j; and j = 1=(1 j) > 1 is the elasticity of substitution between these varieties in country j . De…ne the price index for exports from country i to country j as "Z #1=(1 j) 1 Pij = vij (! ) pij (! ) j d! ; (4) !2 ij so that the demand for variety ! produced by country i for export to country j is j pij (! ) qij (! ) = vij (! ) Xij : (5) Pij Thus, our demand function looks just like the standard one à la Melitz, except it has a mul- tiplicative demand shock. It is worth emphasizing that the above speci…cation implies that 13 the expenditure on all the di¤erentiated goods taken together will be constant. Thus, any increase in Bangladeshi exports must come at the cost of producers from other countries.20 3.2 Pricing and Equilibrium Firms are heterogeneous in their productivity as well as their demand shocks. The production structure is summarized in Figure 1. Stage 1 Stage 2 Stage 3 Pay f and export Low without Productivity Exiters meeting Entrants ROOs Survivors `s are learnt with high Potential by firms demand Entrants shock High Productivity Market Pay f and d, Entrants Entrants and invoke ROOs Pay entry costs Pay fixed costs to access a desirable `s are learnt and market or markets (EU and/or US) by firms for Exiters randomly draw and learn a market and firm specific all markets productivity demand shock tried Figure 1: Production Structure for Bangladeshi Exporters. Bangladeshi …rms …rst pay fe in order to get their productivity draw ' from the pro- ductivity distribution G (') : After observing '; they decide whether to enter the US and/or US EU EU markets and pay a …xed cost of fm and fm ; respectively.21 Once they have entered, 20 Had we allowed greater substitutability between C and N , we would have generated larger responses to policies that enhanced Bangladeshi competitiveness with a less adverse impact on other suppliers. 21 We assume these …xed costs are not per country but for the EU as a whole. Moxnes (2010) suggests 14 they see the market speci…c demand shocks, vU S and vEU; drawn from distributions Hj (v ) ; j 2 fEU; U S g ; where the draws for each …rm are independent across markets. This as- sumption is convenient; it allows us to separate the decisions on entry made by a …rm in each market.22 It is also consistent with the facts: the correlation between the estimates of demand shocks in Demidova, Kee and Krishna (2012) is close to 0: If …rms decide to sell in market j , they incur a …xed cost of production, f; as well as transportation costs of the iceberg form BD;j > 1; so that marginal costs are increased by this factor. Given wage w, the marginal cost of a …rm with productivity ' is w BD;j =': If …rms further choose to meet ROOs, they pay in addition dj ; the documentation cost of meeting ROOs. s decision on whether to sell in a market or not depends on its value of ' and v A …rm’ in the market. As all varieties are symmetric, while productivities and demand shocks di¤er across …rms, we can drop ! from our notation, keeping only ' and v: A Bangladeshi …rm with productivity ' and market demand shock vBD;j in market j will charge price pBD;j (') and sell qBD;j (') units, so its revenue will be 1 rBD;j ('; vBD;j ) = (1 tBD;j ) qBD;j (') pBD;j (') = (1 j tBD;j ) vBD;j PBD;j pBD;j (')1 j RBD;j ; (6) where RBD;j = PBD;j XBD;j is the total sales from Bangladesh and tBD;j is the tari¤ on Bangladeshi exports by country j: The ad-valorem tari¤ tBD;j is levied on the price so the …rm receives (1 tBD;j )pBD;j per unit sold at the price pBD;j . As the demand shock is s price depends only multiplicative, it does not a¤ect the price set by a …rm, so that a …rm’ on its productivity. The pro…ts earned by a …rm are w BD;j BD;j ('; vBD;j ) = (1 tBD;j ) pBD;j (') qBD;j (') qBD;j (') f (7) ' w BD;j = (1 tBD;j ) pBD;j (') qBD;j (') f: (8) ' (1 tBD;j ) w BD;j It is easy to see that …rms set consumer prices as if their marginal costs were '(1 tBD;j ) , while receiving only (1 tBD;j ) of their variable pro…ts. As usual, due to the CES framework, that costs may be per country. 22 If demand shocks were correlated, a …rm may enter just to get information on the state of demand. 15 w the price paid by consumers is pBD;j (') = BD;j :23 We set labor units to be such that (1 tBD;j ) j ' wage (w) is equal to a dollar in our partial equilibrium model. In e¤ect, we assume that w is …xed and this can be rationalized by the existence of labor surplus in Bangladesh. Thus, j all …xed costs f; fe ; fm ; and dj are in terms of labor units and are expressed in dollars. To sell in a market, a …rm has to pay a …xed production cost f and, if it chooses to meet ROOs and avoid paying tari¤ tBD;j , documentation costs dj as well. However, meeting ROOs could raise direct marginal costs and this possibility is allowed for by having direct 1 marginal costs be ' when ROOs are met. Of course, 1 as ROOs are costly to meet. The bene…t of meeting the ROOs is that tari¤s are not paid. Finally, to model binding quotas in the US, we have marginal cost of an exporter to the US with productivity ' given by ( BD;U S + ) ='; where denotes the price of a quota license in ad-valorem form (i.e., = 0:07).24 As marginal costs remain constant despite these complications, we can look at the decision-making in each market separately. 3.2.1 Stage 3 As usual, the model is solved backwards. In Stage 3 we can de…ne the demand shock v ('; PBD;j ) ; which allows a Bangladeshi …rm with productivity ' to earn zero pro…ts in market j . As pro…ts are increasing with v in each market, all …rms with productivity ' and v v ('; PBD;j ) sell in market j: In addition, for the EU market we de…ne the demand shock v ROO ('; PBD;EU ) such that additional pro…ts from invoking EU ROOs just cover the documentation costs of meeting them.25 From the zero pro…t conditions (see the Appendix for more detail), the relationship between v ('; PBD;EU ) and v ROO ('; PBD;EU ) is 23 Demidova, Kee and Krishna (2012) estimate TFP for each …rm and show that, as predicted by the model, the correlation between TFP and price is negative and the shapes of 2 distributions are very similar. 24 We model license prices per unit in ad-valorem rather than speci…c terms to ease the analysis and avoid severe computational di¢ culties. Such speci…cation tends to reduce the tari¤ paid by more productive …rms, as they charge lower prices and so pay a lower dollar tari¤. Irarrazabal et al. (2012) argue that if trade barriers are of the per unit form, gains from trade liberalization may be signi…cantly higher than if they are of the ad-valorem form. 25 Note that there is no such shock for …rms in the US market as all Bangladeshi exporters have to meet US ROOs since the US has country speci…c quotas. 16 v ROO ('; PBD;EU ) = C ROO v ('; PBD;EU ) ; (9) dEU where C ROO = h EU i: If only some …rms meet ROOs, C ROO > 1: f (1 EU ) (1 tBD;EU ) 1 If C ROO 1; then all …rms that enter the market meet ROOs. As expected, C ROO rises and the fraction of …rms that meets the ROOs falls, as preferences become less attractive: i.e., as tari¤s are lowered, or the documentation costs or marginal costs of meeting preferences increase. Equation (9) points out that once we know cuto¤ v ('; PBD;EU ) ; we also know the corresponding one for meeting ROOs. 3.2.2 Stage 2 In Stage 2, we de…ne productivity 'BD;j of the marginal Bangladeshi …rm in market j: For any '; the expected pro…t from selling in market j is the integral of pro…ts over v v ('; PBD;j ). Firms enter the market Firms choose not to enter and do not pay fmUS Firms draw demand shock, but stay out Figure 2: Demand Shock-Productivity Trade-o¤ for the US market. The …rm with 'BD;j is, by de…nition, indi¤erent between trying to access market j and not j 26 doing so, i.e., its expected pro…ts from accessing market j are equal to the …xed cost fm . 26 As shown in equation (40) in the Appendix, the expected pro…ts for the EU market consist of 2 parts: the expected pro…ts from exporting without invoking EBA preferences plus the expected pro…t conditional 17 As expected, pro…ts rise with ': only …rms with ' > 'BD;j earn non-negative pro…ts on average once their demand shocks are realized, and therefore, only such …rms try their luck s parameters. in market j . This gives the cuto¤ productivity 'BD;j in terms of the model’ (See equations (45) and (46) in the Appendix.) Knowing 'BD;j and v ('; PBD;j ) allows us to depict the trade-o¤ between the demand shocks and productivities of …rms in each market, as done in Figures 2 and 3: the downward sloping locus re‡ects the fact that the demand shock needs to be really low to force a very e¢ cient …rm to exit the market.27 3.2.3 Stage 1 In Stage 1 we use the free entry condition to derive the mass of entrants in the equilibrium. Our solutions for 'BD;j ; j 2 fEU; U S g ; depend on the aggregate price indices in the markets, which fall with increases in the mass of entrants. This reduces pro…ts at any given ' and v; which shifts the cuto¤ locus upward and raises the cuto¤ productivity in each market, thereby reducing ex-ante expected pro…ts from entry. The equilibrium entry level is such that the expected pro…ts from entering the industry, obtaining a productivity draw, and choosing optimally from there onwards equal the cost of doing so, fe : (See equation (47) in the Appendix.) We use the model and the available data on Bangladeshi …rms to estimate s parameters. The solution of the model is described in the Appendix. the model’ 4 Estimation Outline Identi…cation of the parameters is conditional on a number of basic assumptions stated and brie‡y discussed below. First, the model is structured so that decisions across markets are made separately. This simpli…es the derivations signi…cantly. We assume in particular that marginal costs are constant, but they could fall or the …rm could be subject to capacity on invoking EBA preferences multiplied by the probability of getting a demand shock high enough to warrant meeting the ROOs needed to obtain EBA preferences. 27 This is analogous to the setup in Lileeva and Tre‡er (2010), where there are also two sources of …rm heterogeneity, namely, initial productivity and productivity gains from investing. In their analysis, …rms with initial productivity and productivity gains from investing that are above a downward sloping locus select into making an additional investment. 18 Firms enter the market and pay to meet ROOs Firms enter the market, Firms choose but do not pay not to enter to meet ROOs and do not pay Firms draw demand shock, but stay out Figure 3: Demand Shock-Productivity Trade-o¤ for the EU market. constraints so that its costs would rise steeply at some point.28 In addition, incurring some …xed market entry costs may reduce or raise others, or demand shocks may be correlated across markets so that entering one market may provide information valuable in another. We abstract from all such issues and assume that all costs are speci…c to the market with no spillovers across markets. Second, we assume the US and EU make up the entire world market for Bangladesh. This is a reasonable assumption as in 2004 about 93% of total Bangladeshi exports in apparel went to 16 countries in the EU or to the US. Relaxing this assumption would a¤ect the ex-ante entry condition and tend to raise the estimate of fe : Third, we make assumptions about the parametric form taken by the distributions we recover. We assume all entrants draw their productivities (as well as demand shocks) from a Weibull distribution with density function: x 1 x h(x) = e ( ) ; (10) where and are the shape and scale parameters. Such a distribution has a very ‡exible form: it can approximate the exponential or log normal distributions and, when truncated as 28 See, for example, Blum, Claro and Horstmann (2013). In their model, capital investments are made prior to the realization of the demand shock, resulting in increasing marginal cost curves faced by …rms. 19 required by the model, closely …ts the observed distributions. The distribution functions for productivity and demand shocks are given by G(') and Hj (v ); j 2 fU S; EU g, respectively. 4.1 Estimation Strategy We distinguish between what we take as given, the data, and the parameters to be estimated. For the analysis below we de…ne three kinds of …rms: …rms that sell to the US only (OUS …rms), to the EU only (OEU …rms), and to both the EU and US (AUS …rms). 4.1.1 Trade Policy Data We take the values for (the per-unit cost of meeting ROOs), t (tari¤s), and (transport costs) to be set at levels roughly in line with the speci…cs of the market. As ROOs involve using domestic cloth, which is about 20% more expensive than imported cloth and as roughly 75% of the cost is the cloth, we assume a 15% cost increase from meeting ROOs and so set = :85 in wovens. As there are quotas in the US, ROOs must be met by all …rms so that for the US market we cannot separately estimate documentation and …xed costs. As ROOs are easy to document in the US since only assembly is required, we set dU S = 0: We denote a license price associated with the US quotas (estimated at 7%) by below. Table 2: Trade Policy Parameters t tROO + EU 0.85 0.12 0 1.14 US 1 0.2 0.2 1.14 + 0.07 In the EU, as only some …rms meet ROOs, we can estimate d and f separately. Transport cost estimates for the apparel industry range from a low of about 8% to a high of roughly 14%.29 Bangladesh has poor infrastructure in transportation and for this reason we use the upper end of the estimates in the literature and set transport costs to be 14% in our 29 See World Bank (2005), p.110, and Gajewski and Riley (2006), p. 6. 20 estimation.30 Tari¤s are 12% and 20% in the EU and US, respectively, and t is set accordingly. This is all summarized in Table 2. 4.2 The Estimation Routine j Thirteen parameters we need to estimate are j; fm =f; d=f; fe =f; j; j; TFP ; TFP ; and f; j = U S; EU: In such procedures, the strategy is to guess the values of the parameters and generate data from the model given the guesses. The parameters are then chosen to best …t certain moments of the data. In other words, we use the method of moments in estimation. First, we guess values of all the above parameters. Given these, we can simulate …rms’ productivity and demand shocks, and solve numerically for the cuto¤ values of demand shocks for any given productivity, as well as the cuto¤ productivity level in each market. (See the Appendix for details.) Once we do this, we know which of the simulated …rms will actually produce in each market, the price indices in each market, the shares of OUS/OEU/AUS …rms, and the fraction of …rms meeting ROOs. Then we are able to generate the distributions of prices, demand shocks, and quantities from the model that are the counterparts of those we choose to match from the data. We then choose parameters to make generated data as close to the actual data as possible. What remains to be speci…ed is the objective function being minimized in the above procedure. Let us denote vector of parameters by , the k -th percentile of the distribution ::: of prices and quantities for …rm type b 2 fOEU; AU S; OU S g31 in the simulation by pb k :::b and qk ; correspondingly, and the data set we have by X . The moment conditions for the distributions of prices and quantities for …rm z; z = 1; :::; Z; are: :::b :::b :::: :::b mp zbk (X; ) = I pzb 2 pk ( ); pk+" ( ) "; and mq b zbk (X; ) = I qzb 2 qk ( ); qk+" ( ) "; where " is the bin size chosen, and I denotes an indicator function. In other words, given 30 Hummels (2007) shows that the e¤ect of distance has diminished over time so that an exporter 14,000 km away paid only 30% more than the one 2,000 km away. As our data is a few years after his sample period we decided to use the same value of transport costs for both the US and EU markets. 31 For AUS …rms we distinguish between the distributions in each market and give them an equal weight. 21 the parameters, ; the model would have a fraction " of the data in each bin, and mp zbk (X; ) and mq zbk (X; ) are the di¤erences between the fraction of the data in each bin and ": If the parameters are the true ones, these should have a mean of zero. Unlike prices and quantities, which we can observe directly in our data, we have to back s revenue is out demand shocks, which we do in the following way. Recall that a …rm’ 1 j pij (! ) qij (! )pij (! ) = vij (! ) Pij Xij : Pij We know the left hand side of the above equation as it corresponds to sales in the data. s sales from the data, as well as the We know the price charged as the unit value of the …rm’ price index (Pij ), which is obtained from the simulation, as well as the total value of exports to a given market (Pij Xij ) from the data. This leaves the demand shock vij (! ) as the only ::: unknown to be solved for. Let us denote percentile points of this distribution by 1 bk ( ; X ): At the same time, given a vector of parameters ; we can also recover the model implied ::: distribution of demand shocks for those …rms that survive and its percentile points 2 bk ( ; X ). We minimize the di¤erence between the data/model and model implied distributions of demand shocks in the same way as we did for the price and quantity distributions, and de…ne moment conditions as: :::1 :::1 :::2 :::2 mzbk (X; ) = I zb 2 bk ( ; X ); bk+" ( ; X) I zb 2 bk ( ; X ); bk+" ( ; X) : We use seven bins total for each distribution we match. The …rst four bins have 20% of the data in each bin, and the next three bins have 10%, 7.5%, and 2.5% of the data, respectively.32 Finally, we denote the shares of …rms (in percentages) that are of the OEU, AUS, and e e OUS types in the data by Sb : Similarly, let SROO denote the share of …rms (in percentages) that meet EU ROOs in the data. The moment condition for the share of AUS …rms is: mShare z;AU S (X; ) = I [F irm z is AU S; ] e SAU S: 32 In essence we are approximating continuous distribution functions by step functions. Our choice of bins puts more weight on matching the model and the data for the largest …rms. In this we follow EKK, who similarly match 4 bins - from 0% to 50%, from 50% to 75%, from 75% to 95%, and from 95% to 100%. 22 I [F irm z is AU S ] equals unity if …rm z is a AUS …rm in the simulation. This occurs if its productivity is above the cut-o¤ level for the tougher market and demand shock draws are above the corresponding cut-o¤s for both markets. The expectation of mShare z;AU S (X; ) over the distributions of productivity and demand shocks for the true parameters should be zero. The moment conditions for OEU, OUS, and ROOs …rms are de…ned analogously. Stacking moment conditions for price, quantity, demand shock distributions, and shares of di¤erent types of …rms yields the vector of moment conditions mz (X; ) for …rm z . The choice of which moments to …t comes from the need to identify all our parameters. It is worth providing some intuition on how all the parameters are being identi…ed. Matching the shares of each …rm type and the distributions of demand shocks for each …rm type helps identify the parameters of the distributions of demand shocks. Matching the share of …rms meeting ROOs identi…es documentation costs, while matching the distributions of prices for each …rm type identi…es the parameters of the TFP distributions. Matching the position of the quantity distributions helps pin down …xed costs of production. The value of the elasticity of substitution a¤ects price, and through it, quantity, so that matching quantity distributions for the di¤erent …rm types also helps pin down these parameters. The shape of the demand shock distribution in a market helps pin down the …xed market entry costs as is evident from equations (49) and (50) in the Appendix. The objective function that we minimize with respect to is: " #0 " # 1X 1X Z Z mz (X; ) W mz (X; ) ; Z z=1 Z z=1 where W is the weighting matrix. Following the literature we start with the unitary weighting matrix and obtain the set of estimates ^u of parameters :33 Up to this moment our method is an example of the classical generalized method of moments. With the …rst step estimates in hand, we have to calculate the optimal weighting ma- trix. Following Wooldridge (2010), we use the weighting matrix Qmm , where Qmm = 33 We also tried to use a weighting matrix with higher weights on the moment conditions on shares. The estmation results are similar. 23 E [m(X; )m(X; )0 ] ; which is usually approximated by XZ bmm = 1 Q mz (X; bu )mz (X; bu )0 : Z z=1 bmm , we employ a simulation based approach. Based on the …rst step estimates To evaluate Q we simulate a large number of arti…cial …rms (NA ). Knowing prices, quantities, and demand shocks for these …rms, and combining these data with the real data, we calculate moment conditions ms (X; ) …rm by …rm. Then the optimal weighting matrix (for a given draw)34 is "NA # 1 X ^s = 1 W ms (X; b)ms b0 k (X; ) : NA k=1 k To minimize the e¤ect of random sampling error, we take a large number of draws S and ^ = 1 PW S ^ as: W approximate W ^ s: S s=1 Standard errors for the estimates are obtained using bootstrap techniques. At each of the 100 bootstrap iterations we repeat the estimation routine outlined above.35 Follow- ing Horowitz (1996, 2001), we re-center the moment conditions to achieve the asymptotic re…nement. 5 Results of the Estimation The results of the estimation are presented below. Tables 3 and 4 show the estimated parameters for the TFP and demand shock distributions in the two countries, respectively.36 Table 3 reports the means and standard deviations of the TFP distribution. The means and standard deviations of the distributions of demand shocks are reported in Table 4. The mean demand shocks in the US are larger than those in the EU and have a greater coe¢ cient of variation. This is consistent with the di¤erences in the distribution systems in the two 34 Following EKK, we use the generalized inverse to calculate the optimal weighting matrix. 35 According to Efron and Tibshirani (1994), having 100 of bootstrap iterations is enough for estimating a standard error. 36 The …gures with the distributions of quantities, prices, and demand shocks in the data and from the estimated model can be found in the working version of the paper available online. 24 countries. Large retailers, like Walmart, play a much bigger role in the US than in the EU. Firms that are lucky enough to land an order from such a large buyer will look like they had a higher positive demand shock. The higher mean demand shock in the US is also consistent with the fact that the US is the older market for Bangladesh, which ties in with the work of Foster, Haltiwanger and Syverson (2008, 2012). They allow for both TFP and demand shock di¤erences among …rms and argue that younger …rms tend to be at least as productive as old ones, but are smaller, i.e., have smaller demand shocks. They interpret this result in terms of …rms having “market capital” (due to advertising or consumer experience with their goods), which grows slowly over time.37 Table 3. TFP distribution Estimate Std. Err. Shape ( TFP ) 0.81 0.21 Scale ( TFP ) 0.42 0.15 Implied moments of the distribution Implied mean shock38 0.47 Coe¢ cient of Variation 1.24 Table 4. Distribution of demand shocks EU US Estimate Std. Err. Estimate Std. Err. Shape ( ) 0.32 0.008 0.17 0.003 Scale ( ) 1.39 0.087 0.57 0.020 Implied moments Implied mean shock 10.4 421.8 Coe¢ cient of variation 4.9 30.7 37 Among other papers that emphasize the dominant role of changes in demand over productivity shocks are Roberts, Xu, Fan and Zhang (2012), who show that demand shocks are the driving force behind …rms’ turnover, and Eaton, Kortum, Neiman and Romalis (2011), who demonstrate that changes in demand explain more than 80% of the drop in trade/GDP during the Great Recession of 2008-2009. 38 The mean equals (1 + 1 ); where (:) denotes the standard gamma function, and the variance equals 2 2 (1 + 2 ) (1 + 1 ) . 25 Given our estimated distributions, we can check whether the US is a tougher market. To do this, we compare the probability of being active in the US market versus the EU market by integrating over the relevant demand and productivity shocks. In our estimates, the probability of trying the EU (US) market is 17% (13%), while the probability of surviving, conditional on having tried the market, is 60% (23%) for the EU (US). These numbers are consistent with the US being a tougher market and having a larger extent of failed entry.39 Table 5. Elasticities of substitution EU US 1.34 1.45 Std. Error 0.033 0.027 Table 5 gives the demand elasticities in each market. These are more than unity and close to the estimates obtained in Demidova, Kee and Krishna (2012). They are a little higher in the US and are similar in magnitude to those found in other structural models like Foster, Haltiwanger and Syverson (2012).40 The estimates of various …xed costs are given in Table 6. The costs of entering the industry are about $77,000, which are about a third of the cost of entering the EU market and similar in magnitude to the cost of entering the US market. Fixed costs of production as well as the documentation costs are low. Note that these numbers, when added up, give a …gure close to the sunk cost estimates in Das, Roberts and Tybout (2007) for knit wear (of about .5 million), which is a part of the non-woven apparel industry. Our estimates are, unfortunately, not directly comparable to theirs for two reasons. First, our numbers are for a particular subset of wovens, while theirs are for knit wear. Second, our numbers should be interpreted as annualized values since our model is static. 39 In previous work, see Demidova, Kee and Krishna (2012), we showed that when the EU o¤ered EBA preferences to Bangladesh, ROOs were less restrictive in non-wovens than in wovens. As expected, we saw the number of …rms in wovens rising by 33% and in non-wovens by 80% from 1999 to 2003. This con…rms the importance of entry as a channel of adjustment. 40 The estimated elasticity in Foster, Haltiwanger and Syverson (2012), who focused on homogenous goods producers in the US, is 1.8. Given that Mens/Boys Cotton Trousers can be viewed as more di¤erentiated goods, our estimates of elasticities being less than theirs are not very surprising. 26 Table 6. Fixed costs in $ Estimate Std. Error f 6,404 476 EU fm 251,250 19,054 dEU 4,240 317 US fm 67,869 5,237 fe 77,348 5,372 While interpreting our cost estimates it is important to understand that they include all …xed costs, sunk and not sunk, monetary, opportunity, or psychic that the producer takes into account in making decisions. For example, if getting around corrupt o¢ cials to enter an industry creates costs in terms of bribes, time spent, or headaches, these would show up in industry entry costs.41 It is worth noting that market entry costs to the EU are quite high. As marginal costs of exporting to the US are higher than those of exporting to the EU (tari¤s are higher and there are quotas), …xed costs of entering the US market are estimated to be low to help match the relatively large share of …rms (37%) that export to the US. This makes sense in terms of the institutions as the presence of large US retailers from the US looking for suppliers abroad reduces the …xed cost of entering the US market. Conversely, for the EU: despite preferences and the absence of quotas, the share of …rms exporting to the EU is only 60%. As a result, the EU market entry costs come out to be large. Documentation costs are low, which is consistent with a large share, 73%, of the …rms that export to the EU choosing to meet ROOs. Fixed cost of production are also estimated to be low to account for the presence of small exporters. Finally, industry entry costs come from the free entry condition. It is also worth noting that such a rich structure of …xed costs is rarely estimated. Es- timates for documentation costs, for example, are almost impossible to …nd. Our approach allows us to provide such estimates. 41 According to the World Bank’s Worldwide Governance Indicators, Bangladesh scores poorly on control of corruption. Doing-Business Indicators also consistently place Bangladesh among the countries with the highest costs of doing business. Both facts are consistent with high …xed entry costs. 27 6 Policy Experiments Before turning to the policy experiments, we need to outline how the partial equilibrium assumption is implemented in the simulations. From the estimation procedure, we get price US 1 EU 1 indices of Bangladeshi …rms exporting to the EU and US: (PBD;U S ) and (PBD;EU ) . s share of demand times total Just as demand for a variety is the product of the variety’ demand, the revenue of Bangladeshi …rms in country j is: (PBD;j )1 j RBD;j = P Rj ; (11) (PBD;j )1 j + i2 ( BD );j [Pi;j ]1 j where ( BD);j is the set of countries exporting to j; excluding Bangladesh. RBD;U S and RU S are approximated by the total Bangladeshi sales and total exports of woven apparel to P the US, respectively. We can invert equation (11) to obtain i2 ( BD) ;j [Pi;U S ]1 U S denoted by P BD;U S : We can obtain P BD;EU in an analogous manner.42 In our simulations, we keep P BD;EU and P BD;U S …xed in accordance with our partial equilibrium assumptions, as we assume that Bangladesh is a small country. However, it is worth exploring what is being missed by making this assumption. We argue below that this would result in under estimating the e¤ects of policy on exports, while over estimating the e¤ect of consumer surplus. Consider a policy that reduces PBD;EU : This will raise Bangladeshi exports, while reducing the pro…ts of non-Bangladeshi …rms and causing their exit, which, in turn, would raise P BD;EU ; making Bangladeshi exports even more competitive, suggesting that the Bangladeshi export increases due to the policy (that reduced PBD;EU to begin with) that are predicted by our simulations would tend to be under estimates. However, because P BD;EU would rise, while the simulations assume it is …xed, the consumer surplus gain in the EU would be over estimated in our simulations. In addition, as Bangladesh has a small share of the world market, there is an asymmetry that is worth noting. Any change in PBD;EU will evoke a small response in P BD;EU : However, even a small change in P BD;EU will have 42 Note that even if we assume that instead of the term P i;j we have the price indices for other countries P 1 i;j calculated with elasticities di¤erent from j (i.e., i2 ( BD);j [Pi;j ] ), our results will not be a¤ected P 1 i;j at all, since i2 ( BD);j [Pi;j ] will be solved for exactly the same way as the P i;j : 28 large e¤ects on Bangladeshi demand. It is also worth emphasizing that the increases in exports of Bangladesh would tend to come at the expense of other exporters. Thus, EU preferences to less developed countries, including Bangladesh, may end up hurting possibly even poorer countries in Africa, as Bangladesh is likely to be able to better take advantage of such preferences than some other developing countries.43 In our experiments below, we distinguish between “endogenous license price”and “exoge- nous license price”scenarios. We transform the license price for a quota into its ad-valorem equivalent in the model. In the presence of quotas, policy changes will a¤ect the license price and its ad-valorem equivalent. In the “exogenous”scenario, we keep the license price …xed. In the “endogenous”one, we allow the equilibrium value of the license price to change. For each experiment we …rst check whether the US quota remains binding or not. If it is no longer binding, we set the new license price to be zero. If it remains binding, we …nd the license price that equates the demand for quota licenses at the new license price to the …xed supply of licenses. This keeps the quantity sold by Bangladeshi …rms to the US constant. We are now ready to look at some policy questions. Our …rst experiment deals with a question of considerable policy importance, namely, the costs of preferences. Developed countries typically give preferences to developing ones, but require that exporters meet origin requirements as done by the EU in the EBA. Thus, obtaining preferences can be quite costly. Consequently, such preferences can be much less generous than they seem. We use our model to quantify the impact of making such preferences easier or harder to obtain.44 We show, for example, that removing the home yarn requirement results in a surge of entry and exports. The second set of experiments looks at the policy e¤ectiveness of subsidies to …xed costs. Which subsidies are the most e¤ective in terms of promoting exports? This question is relevant for developing countries for a number of reasons. Foreign exchange may be valuable 43 In our tables, we calculate welfare changes directly using indirect utility normalized so that the marginal utility of a dollar of income is unity. We use estimates of $12,465 and $11,855 billion dollars for the EU and US GDP in 2004 taken from the IMF World Economic Outlook Database (April 2012 Edition), tari¤ revenues of $105 and $308 million dollars collected by the EU and US, respectively, and their expenditure shares of 0.00024 and 0.0005 on woven apparel in our calculations. 44 Mattoo et al. (2003) look at the Africal Growth and Opportunity Act and (based on back of the envelope calculations in a simple competitive model) argue that preferences are undone by restrictive ROOs. 29 in itself due to the existence of a “foreign exchange gap.”Also, exports may provide needed tax revenues, or more generally, may be a source of externalities. To examine this question, we look at the e¤ectiveness of a given dollar value of a subsidy to di¤erent kinds of …xed costs as in Das, Roberts and Tybout (2007). We consider both the short run e¤ects (when the mass of entrants is …xed at the level before the policy change) and long run e¤ects (when everything, including the mass of entrants, adjusts) and …nd that they can go in opposite directions. We also look at welfare and revenue e¤ects. Our work suggests that in the absence of any response from other countries, as might be expected for a small country (with the given price index of competing products), a fall in the …xed costs …rms face can greatly increase their exports. We discuss the reason for this and provide a decomposition of the relevant margins. It is worth emphasizing that, without quotas, the cross market spillovers of EU liber- alization discussed earlier seem to be large, being of the same order of magnitude as the own market e¤ects. Also, while the e¤ects of changes in trade policies in most standard models give welfare changes in the millions, the welfare changes our model generates are in the hundreds of millions of dollars. Dixit (1988) argues that the magnitude of welfare changes is in the billions only when there are large pre-existing distortions in other markets like those created by labor unions setting arti…cially high wages. However, in our model free entry magni…es the e¤ects of trade policies, generating changes in the hundreds of millions rather than in the millions, even though the sector we are using accounts for a small part of expenditure. Of course, these e¤ects are muted in the endogenous license price scenario. We report the e¤ects on Bangladesh focusing on the change in its exports. How might Bangladeshi welfare be a¤ected by increases in exports? The average wage in wovens in 2003 is about $530 per year, while GDP per capita is $334. The sales to employment ratio is $4,876 per worker, so an additional $10,000 of exports creates 2 jobs, with a rent of about $196 per job, resulting in a welfare gain of $392 per $10,000 increase in Bangladeshi exports.45 45 If the wage premium of $196 per worker is a compensating di¤erential associated with employment in the exporting sector, our estimates for welfare increases in Bangladesh would be an upper bound for welfare gains there. However, there is evidence, see Bryan, Chowdhury and Mobarak (2014), suggesting that credit constraints keep agents from migrating despite gains from doing so. 30 6.1 Documentation Costs, Preferences, and ROOs The preferences given to Bangladeshi exporters by the EU in the woven industry are costly for two reasons. First, there is the requirement of using more expensive domestic fabric. Second, there are documentation costs involved (see Table 6). 6.1.1 Long Run E¤ects We begin by considering the e¤ects of series of policies in the long run, i.e., when entry has time to occur. Table 7 looks at four policy changes and their e¤ects in the long run. Most of the table deals with the endogenous quota price scenario, though the last few rows describe the key outcomes for the exogenous quota price scenarios. Column 1 describes the outcome under the status quo. Column 2 looks at the e¤ect of eliminating preferences. This involves making the tari¤ in the EU 12% for all Bangladeshi …rms.46 Column 3 shows the e¤ects of doubling documentation costs. Column 4 captures the e¤ect of policies, like regional cumulation,47 which make ROOs less costly to meet. To approximate this, we make ROOs costless to meet in terms of marginal production costs in wovens. Finally, Column 5 gives the e¤ect of removing the documentation costs and the marginal cost of meeting ROOs. All reductions in costs make Bangladeshi …rms more optimistic about their expected pro…ts ex-ante. Thus, the mass of entrants rises. In addition, more relaxed EU ROOs allow a greater share of Bangladeshi exporters to meet them. This e¤ect also expands the market share of Bangladeshi exporters in the EU market. There are also cross-market e¤ects that are particularly pronounced when the quota price is …xed. This emphasizes the role of existing quotas in limiting the e¢ cacy of trade liberalization elsewhere. A more liberal policy in the EU results in a greater mass of entrants into the industry, which raises Bangladeshi exporters’share in the US market and reduces the price index there. Quotas in the US blunt such e¤ects reducing the impact of unilateral liberalization on the part of the EU. 46 In calculating welfare changes, we add the net increase in tari¤ revenues from Bangladeshi and from non-Bangladeshi …rms as predicted by the model. Details of the calculations are available upon request. 47 For example, if cheap Indian cloth could be used in production without compromising Bangladeshi origin, costs of meeting ROOs would fall. 31 Table 7. Long-run equilibrium implications of policy changes (values in dollars) Baseline No preferences Higher doc. costs No home yarn req. Costless pref. Tari¤ in EU (tBD;EU ) 12% 12% 12% 12% 0% Tari¤ in EU, ROO (tROO BD;EU ) 0% 12% 0% 0% 0% Tari¤ in US (tBD;U S ) 20% 20% 20% 20% 20% Cost disadvantage ( ) 0.85 1.00 0.85 1.00 1.00 Documentation costs (dEU =f ) 0.66 0.00 1.32 0.66 0.00 Endogenous quota price case Change in quota price in US, % Change in quota price in US 0.070 -100% -5.7% +43.4% +49.3% Change in Bangladeshi exports, % EU imports from Bangladesh 482.3m -31.7% -1.5% +17.1% +19.0% US imports from Bangladesh 233.6m -11.9% -0.1% +1.1% +1.2% Change in mass of entrants, % Implied mass of entrants 4,712 -22.3% -0.7% +5.8% +6.6% Change in productivity cuto¤ s, % Productivity cuto¤ in EU 0.8508 +15.9% +0.91% -6.06% -7.42% Productivity cuto¤ in US 1.0355 -6.8% -0.34% +2.61% +2.97% Change in demand shock cuto¤ s, % Demand shock cuto¤ in EU 0.1866 -10.24% -0.59% +5.55% -10.24% Demand shock cuto¤ in US 6.9570 0.00 0.00 0.00 0.00 Share of …rms invoking ROO, % Share of ROO …rms (model) 70.2% 0% 57.0% 77.7% 100% Change in EU and US price indices, % Price index in EU 100% +19.1% +0.87% -9.38% -10.41% Price index in US 100% +1.1% +0.01% -0.10% -0.11% Change in tari¤ revenues, % Tari¤ revenue in EU 447k +8,742% +125.9% -34.2% -100% Tari¤ revenue in US 46,728k -11.9% -0.1% +1.1% +1.2% Change in welfare, $ Change in welfare in EU ($) — -480,935k -25,208k 293,418k 327,162k Change in welfare in US ($) — -68,538k -709k 6,191k 6,964k Exogenous quota price case Change in Bangladeshi exports, % EU imports from Bangladesh 482.3m -45.5% -2.24% +22.7% +25.5% US imports from Bangladesh 233.6m -41.6% -1.94% +14.3% +16.1% Change in EU and US price indices, % Price index in EU 100% +28.15% +1.28% -12.39% -13.8% Price index in US 100% +3.84% +0.18% -1.30% -1.5% Change in mass of entrants, % Implied mass of entrants 4,712 -45.5% -2.3% +16.9% +19.0% Change in welfare, $ Change in welfare in EU ($) — -707,594k -37,343k 393,918k 441,279k Change in welfare in US ($) — -238,328k -11,193k 82,661k 92,933k 32 An important thing to note in Table 7 is that despite ROOs being costly to meet, the industry relies greatly on the presence of the EU preferences. Our model suggests that in the absence of these preferences, as shown in Column 2, entry would fall considerably. If the quota price is …xed, entry falls, EU imports from Bangladesh fall from $482 million to $262 million as do US imports from $233.6 million to $136 million, highlighting the cross-market e¤ects of the EU policies. When quota prices are endogenous, the fall in entry reduces the US quota price so that the fall in entry is lower than in the exogenous quota price scenario. As a result, EU imports fall by less, to only $330 million. US imports fall as well (to $206 million), and the quota becomes non-binding. Doubling documentation costs (Column 3 of Table 7) also reduces EU imports (to $475 million, -1.5%) and US imports (to $233.4 million, -0.1%) in the endogenous quota price case. Note this is a far smaller e¤ect than the removal of preferences. With exogenous quota prices, the exit induced is larger so that the fall in EU and US exports is more pronounced than when the quota price adjusts. The reason why raising documentation costs by a factor of two has a relatively small e¤ect is that these have a limited e¤ect on entry and operate through their e¤ect on marginal …rms choosing whether or not to use ROOs. Policies that a¤ect marginal costs (like removal of preferences, as in Column 2, or reducing the cost of meeting preferences, as in Column 4) a¤ect entry to a greater extent. They do so because they a¤ect …rms of all types, and so tend to have more bang than ones that a¤ect only marginal …rms. When the home yarn requirement is removed, as in Column 4, so that preferences are not costly to obtain, both EU and US imports rise. The model suggests this would result in exports to the EU of $565 million (+17.1%) and to the US of $236.1 million (+1.1%) when the quota price adjusts, and by 22.7% and 14.3% when the quota price is …xed.48 Finally, when both the marginal and …xed costs of meeting ROOs are removed (Column 5), the e¤ects are slightly more pronounced: with endogenous (exogenous) quota prices, exports rise by 19% (25.5%) and 1.2% (16.1%) to the EU and US, respectively. When exporting becomes less promising, the direct e¤ect on pro…ts in the EU is negative, which 48 In 2011, the EU changed its ROOs to require only one stage of processing to occur in the exporting country to obtain origin. 33 raises the productivity cuto¤ there. However, there is a fall in entry of Bangladeshi …rms that raises the price index in both the US and EU (as evident in Table 7), making pro…ts swing upwards, which, in turn, acts to reduce the cuto¤ productivity. The former e¤ect dominates in the EU, while in the US, only the latter operates so that the cuto¤ falls. Giving preferences results in substantial e¤ects (both in the US and EU), even when there are restrictive ROOs, compared to back of the envelope calculations that ignore the role of entry like Mattoo et al. (2003). Our numbers for e¤ects on trade and welfare below are larger because entry does most of the heavy lifting in such experiments. It is also worth emphasizing that when the quota price is endogenous, the e¤ects are qualitatively similar, but muted, suggesting that quotas maintained by the US may have signi…cantly hindered the ability of the EU to help Bangladeshi exports. At the same time, since US quotas are product and country speci…c, they insulate Bangladesh from competition by others. 6.1.2 Long Run Welfare Consequences What about welfare e¤ects of these policies? There are two main channels through which policy regarding Bangladesh a¤ects welfare of the EU households: via consumer surplus and tari¤ revenue. Changing policies impacts the value of tari¤ revenues, T REU , earned by the EU both through the number of Bangladeshi exporters who pay a tari¤ and through the volume of their sales. In addition, policy changes a¤ect the EU price index. In particular, 2 31 1 X EU PEU = 4(PBD;EU )1 EU + [Pi;EU ]1 EU 5 ; (12) i2 ( BD );EU P where i2 ( BD );EU [Pi;EU ]1 EU = P BD;EU . Recall that P BD;EU was calculated earlier and is held …xed at this level in our counterfactual experiments. However, (PBD;EU )1 EU changes as we change the EU policies. The e¤ect on tari¤ revenue and the percentage change in the price index are reported in Table 7. When the quota price adjusts (does not adjust) removing preferences given to Bangladesh by the EU decreases EU welfare by roughly $481 ($708) million dollars ao that giving pref- s own narrow self interest. The removal of preferences reduces ex-ante erences is in the EU’ 34 pro…ts and entry. Such a reduction in the mass of entrants results in a large fall in exports to both the EU and US, with a consequent fall in consumer surplus and tari¤ revenue. Re- moving EU preferences reduces US welfare by about $69 million when quota prices adjust, and by $238 million when they do not. There are positive spillovers to the US of the EU liberalization, and negative spillovers of the US quota on the EU. When documentation costs are raised, ex-ante pro…ts fall, as does the mass of entry, raising prices, which acts to reduce surplus, but as fewer …rms invoke ROOs, tari¤ revenues increase, raising welfare. The former e¤ect dominates when the quota price adjusts so that welfare in the EU falls by about 25 million dollars, while welfare in the US decreases by about 0.7 million dollars. Finally, removing the home yarn requirement raises ex-ante pro…ts and entry, with consequent increases in surplus and tari¤ revenue. Note that welfare rises in both the US and EU by 293 million and 6 million dollars, respectively. Removing the documentation costs as well (see Column 5) raises welfare even more. As expected, all e¤ects are larger when the quota price is exogenous as can be seen in Table 7. 6.1.3 Short Run Results Table 8 looks at the same policy changes, but limits the analysis to the short run. In calculating these impact e¤ect estimates, we turn o¤ the entry channel and look at the e¤ect on …rms that have already decided to be in an industry and market. We keep the mass of …rms that enter the industry and the productivity cuto¤ of …rms that enter a particular market …xed at their initial estimated levels and allow the experiment only to a¤ect the position of the productivity-demand shock trade-o¤s, and via this, all other variables. Preferences and Documentation Costs Before we begin, note that as de…ned, the price charged, pij ('); is the price consumers pay. Firms give part of what consumers pay to the ij w government. In the short run, without preferences pij (') = (1 tij )' ; while with preferences it is ij w= ': Since tari¤s are 12%, ((1 tij ) = :88), while the cost disadvantage is 15% ( = :85), the price paid by consumers falls when preferences are removed! This is what lies behind the fall in the price index when preferences are removed. It also raises the sales of 35 Bangladeshi …rms and their share in EU imports, but reduces the revenues after tari¤s,49 which results in exit in the long run.50 Tari¤ revenues rise quite considerably in the short run, but this will only be temporary. Note that the welfare e¤ects in the short run are the opposite of those in the long run as we have turned o¤ the main channel, namely, entry/exit. Documentation costs do not a¤ect marginal costs, but make some …rms choose not to meet ROOs, reducing their cost and price, and in turn, the price index. As a result, Bangladeshi sales rise, though their revenues (net of tari¤s) fall. With no home yarn requirement more …rms meet ROOs. As there is no marginal cost disadvantage associated with doing so, their prices fall, reducing the price index, while their sales and revenues rise. With costless preferences, there is an additional e¤ect as all …rms meet ROOs. In all the experiments, welfare in the EU rises. In Columns 2 and 3 of Table 8, the driving forces are a rise in consumer surplus and tari¤ revenues, while in Columns 4 and 5 welfare rises despite a fall in tari¤ revenues. Table 8. Short-run equilibrium implications of policy changes. Baseline No preferences Higher doc. costs No home yarn req. Costless pref. Tari¤ in EU (tBD;EU ) 12% 12% 12% 12% 0% Tari¤ in EU, ROO (tROO BD;EU ) 0% 12% 0% 0% 0% Cost disadvantage ( ) 0.85 1.00 0.85 1.00 1.00 Relative doc. costs (dEU =f ) 0.66 0.00 1.32 0.66 0.00 Change in mass of …rms, % Mass of successful exporters to EU 485 0.00% 0.00% -0.21% -0.21% Change in demand shock cuto¤ , % Demand shock cuto¤ in EU 0.1866 +0.37% 0.00 +0.37% +0.37% Change in price index in EU, % Aggregate price index in EU 100% -1.63% -1.09% -3.67% -3.68% Change in tari¤ revenues collected, % Tari¤ revenues in EU 447k +12,964% +130% -43.0% -100% Change in Bangladeshi exports Bangladeshi exports, RBD;EU 482.3m +0.97% +0.01% +4.64% +4.65% Change in revenues of Bangladeshi …rms Revenue of Bangladeshi …rms 481.8m -11.06% -0.11% +4.68% +4.75% Change in welfare Change in welfare in EU ($) — 107,433k 33,712k 111,610k 111,627k It is worth emphasizing the di¤erence in the long run export e¤ects (both in the US and 49 The after-tari¤ revenues, rij (') ; are given by (6), and the gross sales of Bangladeshi …rms can be calculated as rij (') = (1 tij ) : 50 Of course, as entry is …xed, there are no e¤ects on the US market so we ignore it for the time being. 36 EU) of preferences, even when there are restrictive ROOs, and the short run e¤ects in Table 8. Back of the envelope calculations that ignore the role of entry like Mattoo et al. (2003) as well as more sophisticated calculations based on models with the …xed mass of entrants,51 could easily underestimate these long run e¤ects, or even get the e¤ect on welfare reversed. 6.2 Subsidizing Fixed Costs Which …xed costs should be subsidized? Is there a di¤erence? Table 9 looks at these questions in terms of promoting exports. It compares the e¤ectiveness of a given dollar value ($1,500,000) of a subsidy to di¤erent kinds of …xed costs. In this, it follows Das, Roberts and Tybout (2007). The results suggest the export e¤ects vary considerably depending on where the subsidy is applied. A policy maker wanting to raise exports would get up to a $25 ($81) increase in export revenue for every dollar spent reducing …xed costs when the quota price is endogenous (exogenous). In general, applying the subsidy at a later stage so that it is not wasted on …rms that end up exiting produces greater results. Thus, subsidies to market entry are the least e¢ cient (with $.4 increase in exports per dollar spent), while compensating …xed costs of production raises exports the most (by $24.8 per dollar spent). Also, subsidizing market entry costs for markets, where the market share is lower, gives more leverage as …rms can “steal” business from a greater fraction of competitors. Thus, subsidizing US entry with the Bangladeshi market share around 4% gives $11.4 per dollar spent, while doing the same for the EU with Bangladeshi market share around 16% gives only $5.5. We also …nd that cross market e¤ects are large: if the EU market entry is subsidized, the US exports (and tari¤ revenue) rise but only when the quota price is exogenous. As expected, these e¤ects are much more muted when the quota price is endogenous. Thus, policies have large cross market e¤ects, though quotas dilute these e¤ects considerably. 6.3 The Responsiveness of Trade Flows to Trade Barriers It is worth explaining how we get such a large e¤ect on exports given there is free entry. 51 Recall this assumption is made in Eaton, Kortum and Kramarz (2011) and Chaney (2008). 37 Table 9. Fixed costs compensation: Government spends $1.5 million ( 2%). Baseline Industry EU market US market Documentation Fixed case entry costs entry costs entry costs costs Costs $ compensation per …rm / entrant — 318 1,826 2,328 3,192 2,117 Endogenous quota price case Change in quota price in US, % Quota price in the US +1.43% +3.67% +61.3% +2.85% +98.6% Change in Bangladeshi exports,% EU imports from Bangladesh 482.3m +0.11% +1.68% +1.30% +1.37% +6.54% US imports from Bangladesh 233.6m +0.04% +0.08% +5.78% +0.06% +3.19% Change in mass of …rms,% Implied mass of entrants into industry 4712 +0.22% +0.47% +2.62% +0.39% +12.34% Change in productivity cuto¤ s, % Productivity cuto¤ for EU 0.8508 +0.06% -1.16% +0.74% -0.95% +3.25% Productivity cuto¤ for US 1.0355 +0.09% +0.22% -3.52% +0.17% +4.37% Change in demand shock cuto¤ s, % Demand shock cuto¤ in EU 0.1856 0.00% +0.72% 0.00% +0.58% -32.9% Demand shock cuto¤ in US 6.9570 0.00% 0.00% +3.48% 0.00% -32.6% Share of …rms invoking ROOs, % Share of ROOs …rms (model) 70.2% 70.2% 70.2% 70.2% 95.4% 58.6% Change in tari¤ revenues, % Tari¤ revenue in EU 447k +0.12% +2.14% +1.49% -93.14% +86.26% Tari¤ revenue in US 46,728k +0.04% +0.08% +5.78% +0.06% +3.19% Change in EU and US price indices, % Price index in EU 100% -0.06% -0.95% -0.74% -0.78% -3.67% Price index in US 100% -0.004% -0.01% -0.53% -0.01% -0.29% Change in welfare, $ Change in welfare in EU ($) — 1,887k 28,533k 22,077k 22,782k 111,733k Change in welfare in US ($) — 241.3k 456.6k 33,409k 367.6k 18,403k Policy e¢ ciency (dollars of extra net exports per dollar of subsidy) Policy e¢ ciency — 0.4 5.5 11.4 4.8 24.8 Exogenous quota price case $ compensation per …rm / entrant 317 1,820 2,001 3,185 1,912 Change in Bangladeshi exports, % EU imports from Bangladesh 482.3m +0.28% +2.07% +8.59% +1.76% +14.69% US imports from Bangladesh 233.6m +0.46% +1.04% +23.6% +0.95% +27.75% Change in EU and US price indices, % Price index in EU 100% -0.16% -1.17% -4.80% -1.00% -8.12% Price index in US 100% -0.04% -0.10% -2.14% -0.09% -2.50% Change in welfare, $ Change in welfare in EU ($) — 4,808k 35,173k 146,585k 29,402k 252,637k Change in welfare in US ($) — 2,643k 6,019k 136,785k 5,497k 159,761k Policy e¢ ciency (dollars of extra net exports per dollar of subsidy) Policy e¢ ciency — 1.5 8.3 57.1 7.1 81.2 38 First, subsidies raise the mass of entrants considerably. Second, spillover e¤ects of policies across markets magnify the export increase due to any given increase in entry. When subsi- dies attract …rms into the industry, these entrants export not just to the EU, but wherever they have a good demand shock. This is in contrast to what happens in simple competitive settings, where the EU preferences given to Bangladesh would raise exports to the EU but reduce them to the US. Third, due to demand shocks, the marginal and average …rms are large. Without shocks, variable pro…ts of the marginal …rm just cover its …xed costs of pro- duction. With demand shocks, this is true only at the cuto¤ demand shock for each …rm in the economy. Hence, at all demand shocks above the cuto¤ one, the …rm with the cuto¤ productivity has higher pro…ts and sales than it needs to produce. Thus, the marginal and average …rms tend to be larger in the presence of demand shocks. This also helps explain why exports rise greatly with rising mass of …rms. Finally, we look at small interventions. The e¢ cacy of the intervention declines with its size, as marginal returns fall. 6.3.1 The Relevant Margins We want to decompose export changes due to policy in our counterfactuals into their com- ponent parts. The basic idea is quite simple. We ask how much of the exports change is due to changes in the exports of existing …rms (the intensive margin), how much is due to changes in cuto¤s (the extensive margin via cuto¤s), and how much is due to the entry of …rms (the extensive margin via entry). Let total exports be X . Let x denote the exports of an individual …rm. Total exports di¤er in the two periods, 0 and 1; as the mass, productivity, and demand shock cuto¤s change, resulting in the changes of exports per …rm. The change in total exports can (by adding and subtracting the relevant terms) be decomposed as: e e X (M1 ; ' 1 ; v 1 ; x1 ) X (M0 ; ' 0 ; v 0 ; x0 ) e e = [X (M0 ; ' 0 ; v0 ; x 1 ) X (M0 ; '0 ; v0 ; x0 )] (Intensive Margin) (13) e e + [X (M0 ; ' 0 ; v 1 ; x1 ) X (M0 ; '0 ; v0 ; x1 ) (Extensive Margin (14) e e + X (M0 ; ' 1 ; v1 ; x 1 ) X (M0 ; '0 ; v1 ; x1 )] via Cuto¤s; (15) e e + [X (M1 ; ' 1 ; v 1 ; x1 ) X (M0 ; '1 ; v1 ; x1 )] via Entry), (16) 39 e e where X (M1 ; '1 ; v1 ; x1 ) denotes total exports when M1 mass of …rms enter, the productivity and demand shock cuto¤s are those in period 1; and the output per …rm corresponds to that e e in period 1: Similarly, X (M0 ; '0 ; v1 ; x1 ) denotes total exports when M0 mass of …rms enter, the productivity cuto¤s for entry and for each market are that in period 0; while the demand shock cuto¤s in each market for each productivity correspond to those in period 1: What would we expect to happen through these margins? Any policy will have an impact on exports via the exports of existing …rms, i.e., the intensive margin. In addition to the direct e¤ect on exports of the policy, changes in the price index in response to the policy will also a¤ect the exports of existing …rms. If, for example, the price index of Bangladeshi apparel falls, each Bangladeshi …rm faces more competition from other Bangladeshi …rms, s exports which lowers the aggregate price index. This force works to reduce an existing …rm’ at any given price, which is captured in the intensive margin in our decomposition. In our simulations, independent of what they are, the exports of existing …rms do not change very much in response to policy, so that the intensive margin counts for little. In addition, the changes in entry a¤ect the price index via the extensive margin in terms of the demand shock and productivity cuto¤s. Again, these e¤ects tend to be small. For example, when quotas are exogenous, new entrants drive over 78% of the increase in exports. The question still remains how such a small subsidy could result in such a large increase in entry. The answer is that the relationship between pro…ts ex-ante and the mass of …rms is very ‡at in the estimated model. As the mass of …rms that enter rises, pro…ts fall o¤ very slowly. A subsidy, for example, shifts these ex-ante pro…ts upwards. As the above mentioned curve is ‡at, even a small shift up results in a large change in the intersection of the curve with the x axis (which is the zero pro…t condition pinning down entry). Such curve is likely to be ‡ s exports (so that there are a lot at when Bangladeshi …rms are a small part of the world’ of other exporters to steal consumers away from) and when is low so that …rms make room for themselves in the market since they produce unique products. Simulations revealed that this is indeed the case: as the share of Bangladesh in exports rises, the increase in exports in this kind of a simulation falls very fast, suggesting that developing countries, especially small ones, whose exports are not large enough to disrupt markets, might be able to raise exports a lot by focusing on policies that reduce entry costs of various kinds. These polices 40 need not even be subsidies. Nor do they need to be very costly to implement. For example, promoting export fairs that allow buyers and sellers to meet more easily could reduce …xed costs of exporting, as could workshops on how to institute the quality requirements needed by foreign buyers. Putting the needed documentation for obtaining preferences on the web to reduce documentation costs is another example of a potentially low cost, high return policy. 6.3.2 The Role of What is the relation between the estimated low values of and the predominance of the entry channel we …nd? Krugman (1980) predicts that in a homogeneous …rm setup, low makes demand inelastic, reducing the impact of trade barriers on trade ‡ows. In other words, the e¤ect of trade barriers via the intensive margin is weak when substitution is limited. Chaney (2008) argues that a low elasticity of substitution between goods magni…es the e¤ect of trade barriers on trade ‡ows when …rm heterogeneity is added to the model. Note this is exactly the opposite of what Krugman predicts. In the presence of …rm heterogeneity, there are additional e¤ects via productivity cuto¤s. Trade barriers raise prices and, in turn, the price index. The increased price index allows less productive …rms to survive. When is low, such …rms are not at a severe disadvantage, as their products di¤er considerably from those of other …rms. Hence, these …rms can sell a good deal so that with low trade ‡ows are very responsive to trade barriers via the extensive margin. In other words, when is low, the extensive margin e¤ects on trade ‡ows are strong and dominate in the comparison. However, Chaney (2008) (and for that matter EKK) assumes the mass of entrants is …xed and so ignores the entry margin completely. When is low, the ex-ante pro…t condition is quite ‡ s products do not compete directly with those of existing …rms: at as new entrant’ their goods make room for themselves in product space. Thus, trade barriers which shift ex- ante pro…ts will have a large e¤ect on entry and trade ‡ows. Both cuto¤ and entry margins are more powerful when is low. But most of the action on trade ‡ows, at least empirically, comes from the entry margin, not the cuto¤ or intensive one. Thus, while it is fair to say that the low value of makes trade ‡ows more responsive to trade barriers, which, in turn, translates into large leverage for policy in our counterfactual experiments, the channel by which it does so empirically is not the margin emphasized in Chaney (2008). 41 7 Conclusion We provide a simple way of estimating the structural parameters of a heterogeneous …rm model. One of the advantages of our approach is that it uses only cross sectional data to recover all the structural parameters of the model, including …xed costs at di¤erent levels. These include entry costs at the industry and market levels as well as …xed costs of production and documentation costs needed to obtain preferences. Moreover, all our estimates seem reasonable and roughly in line with previous work. The policy implications inherent in our counterfactual simulations are as follows. We think of these as making a case for “trade as aid.”Recently, there have been serious doubts cast on the e¢ cacy of direct aid. It may be diverted to the pockets of those in power or used ine¤ectively. Giving aid and having it be e¤ective in terms of growth or a reduction in poverty are two very di¤erent things. For example, governments may cut back their own support for the poor as aid grows. In contrast, “trade aid”works through market forces. For example, in our application, preferences given by the EU are responsible for a huge increase in export ‡ows from Bangladesh to the EU and to the US, rather than diverting trade away from the US market to the EU market if there had been no US quotas (exogenous quota price). In this manner, trade preferences or other forms of trade facilitation by one country can have a powerful e¤ect on exports to all markets and on output, exports, and employment in the recipient developing country but this e¤ect is sharply attenuated by the presence of quotas in other importing countries. Our counterfactuals also suggest where subsidies might be the most e¤ective in increasing exports. The rule seems to be to subsidize late in the process so that subsidies are not wasted on failed …rms and to subsidize where existing market share is low so that there is more room to poach from foreign competitors. It is worth emphasizing that trade preferences are a form of aid that can create a scenario where most participants win, which is much easier to sell to all parties concerned. The developed country giving preferences wins as its consumers face lower prices. Other developed countries also stand to gain as entry reduces the price of the goods they import. In addition, the developing country gets to increase its exports, earning foreign exchange and employing its labor force. The losers would be developing countries, who do not get these preferences 42 or are unable to make e¤ective use of them and whose exports end up falling. Under the EBA, the former are the better o¤ developing countries (India, China, etc.), while the latter are the worst o¤ of the developing countries, whose other features make them inhospitable production sites. Our results have some lessons for developing and transition countries. Corruption and bureaucracy raise …xed and marginal costs faced by …rms. Our work suggests that even small increases in such costs can result in huge reductions in entry, production, and exports of a country. Conversely, reining in such costs can do much good. Our work can also be seen as highlighting the importance of other initiatives that reduce search costs or inherent uncertainties in the market that raise costs. Thus, export fairs, tribunals for dealing with complaints about product quality, and other policies that reduce the costs of doing business in developing countries may have unexpectedly large e¤ects. Acknowledgements We thank the editor and two anonymous referees for their helpful and perspective comments. We are very grateful for comments from Susumu Imai, David Levine, Ralph Ossa, Joris Pinske, Andrés Rodríguez-Clare, Ina Simonovska, James Tybout, and Anthony Venables. We are especially grateful to Jonathan Eaton for detailed comments on an early draft. We also thank the IGC for support that granted us access to the customs dataset used. Krishna is grateful to the Human Capital Foundation for research support and to the International Economics Section at Princeton University, where this work was started, for a Peter Kenen Fellowship. Demidova thanks the Social Science and Humanities Research Council of Canada and the Arts Research Board of McMaster University for …nancial support. References [1] Arkolakis, C., 2010. Market Penetration Costs and the New Consumer Margin in Inter- national Trade. Journal of Political Economy 118(6), 1151-1199. 43 [2] Armenter, R., Koren, M., 2014. Economies of Scale and the Size of Exporters. Forth- coming in the Journal of the European Economic Association. [3] Aw, B. Y., Roberts, M. J., and Xu, D. Y. , 2011. R&D Investment, Exporting, and Productivity Dynamics. American Economic Review 101(4), 1312-44. [4] Bernard, A., Redding, S., and Schott, P., 2011. Multi-Product Firms and Trade Liber- alization. Quarterly Journal of Economics 126(3), 1271-1318. [5] Blum, B.S., Claro, S., and Horstmann, I. J., 2013. Occasional and perennial exporters. Journal of International Economics 90 (1), 65-74. [6] Bryan, G., Chowdhury, S., and Mobarak, A. M., 2014. Under-Investment in a Pro…table Technology: The Case of Seasonal Migration in Bangladesh. Econometrica 82(5), 1671- 1748. [7] Chaney, T., 2008. Distorted Gravity: the Intensive and Extensive Margins of Interna- tional Trade. American Economic Review 98(4), 1707–1721. [8] Das, S., Roberts, M., and Tybout, J., 2007. Market Entry Costs, Producer Heterogene- ity, and Export Dynamics. Econometrica 75(3), 837-873. [9] Debaere, P., Mostashari, S., 2010. Do Tari¤s Matter for the Extensive Margin of Trade? An Empirical Analysis. Journal of International Economics 81(2), 163-169. [10] Demidova, S., Kee, H. L. , and Krishna, K., 2012. Do Trade Policy Di¤erences In- duce Sorting? Theory and Evidence From Bangladeshi Apparel Exporters. Journal of International Economics 87(2), 247–261. [11] Development Initiative (2005): GSP, SAARC Cumulation and the RMG Textile Sector of Bangladesh, mimeo. [12] Dixit, A., 1988. Optimal Trade and Industrial Policies for the U.S. Automobile Industry, in: Empirical Methods in International Trade, ed. Feenstra, Cambridge, MA: MIT Press, 141-165. 44 [13] Eaton, J., Kortum, S., and Kramarz, F., 2011. An Anatomy of International Trade: Evidence from French Firms. Econometrica 79(5), 1453-1498. [14] Eaton, J., Kortum, S., Neiman, B., and Romalis, J., 2011. Trade and the Global Reces- sion. NBER Working Paper No. 16666. [15] Efron, B., Tibshirani, R. J., 1994. An Introduction to the Bootstrap, Monographs of Statistics and Applied Probability, 57, New York: Chapman & Hall/CRC. [16] Foster, L., Haltiwanger, J., and Syverson, C., 2008. Reallocation, Firm Turnover, and Ef- …ciency: Selection on Productivity or Pro…tability? American Economic Review 98(1), 394-425. [17] Foster, L., Haltiwanger, J., and Syverson, C., (2012): The Slow Growth of New Plants: Learning About Demand?, NBER Working Paper No. 17853. [18] Gajewski, G., Riley, A., 2006. Bangladesh Export Trade Practices and their E¤ect on the Competitiveness of the Garment Industry. USAID case study. s World. [19] di Giovanni, J., Levchenko, A., 2013. Firm Entry, Trade, and Welfare in Zipf’ Journal of International Economics 89(2), 283-296. [20] Helpman, E., Melitz, M. J., and Rubinstein, Y., 2004. Trading Partners and Trading Volumes. The Quarterly Journal of Economics 123 (2), 441-487. Bootstrapping Time Series Models.’Econometric [21] Horowitz, J. L., 1996. Comments on ’ Reviews 15, 165-168. [22] Horowitz, J. L., 2001. The Bootstrap and Hypothesis Tests in Econometrics. Journal of Econometrics 100, 37-40. [23] Hummels, D., 2007. Transportation Costs and International Trade in the Second Era of Globalization. Journal of Economic Perspectives 21(3): 131-154. [24] Irarrazabal, A., Moxnes, A., and Opromolla, L.D., 2012. The Tip of the Iceberg: A Quantitative Framework for Estimating Trade Costs, forthcoming in of Economics and Statistics. 45 [25] Kee, H. L., Krishna, K., 2008. Firm Level Heterogeneous Productivity and Demand Shocks: Evidence from Bangladesh. American Economic Review 98(2), 122-128. [26] Kehoe, T., Ruhl, K., 2013. How Important is the New Goods Margin in International Trade? Journal of Political Economy 121(2), 358-392. [27] Krishna, K., 2006. Understanding Rules of Origin, Estevadeordal et al. (eds.), Rules of Origins, Oxford University Press. [28] Krugman, P. R., 1980. Scale Economies, Product Di¤erentiation, and the Pattern of Trade. American Economic Review 70, 950-959. [29] Lileeva, A., Tre‡er, D., 2010. Improved Access to Foreign Markets Raises Plant-Level Productivity. . . for Some Plants. Quarterly Journal of Economics 125(3), 1051-1099. [30] Lincoln, W. F., McCallum, A. H., 2011. Entry Costs & Increasing Trade, William Davidson Institute working paper 1024. [31] Mattoo, A., Roy, D., and Subramanian, A., 2003. The Africa Growth and Opportunity Act and its Rules of Origin: Generosity Undermined? The World Economy 26(6), 829- 851. [32] Melitz, M. J., 2003. The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity. Econometrica 71(6), 1695-1725. [33] Mlachila, M., Yang, Y., 2004. The End of Textiles Quotas: A Case Study of the Impact on Bangladesh, IMF Working Paper No. 04/108. [34] Moxnes, A., 2010. Are sunk costs in exporting country speci…c? Canadian Journal of Economics 43, 467–493. [35] Roberts, M. J., Xu, D. Y., Fan, X., and Zhang, S., 2012. A Structural Model of Demand, Cost, and Export Market Selection for Chinese Footwear Producers, NBER Working Paper No. 17725. [36] Wooldridge, J. M., 2010. Econometric Analysis of Cross Section and Panel Data, Cam- bridge, MA: MIT Press, 525-530. 46 [37] World Bank, 2005. Bangladesh Growth and Export Competitiveness. Poverty Reduction and Economic Management Sector Unit, South Asia Region, Report No. 31394-BD. [38] Yi, K. M., 2003. Can Vertical Specialization Explain the Growth of World Trade? Journal of Political Economy 111, 52–102. 8 Appendix This Appendix contains the detailed derivations of the equilibrium conditions for the model described in the paper as well as the way of solving the model numerically. As usual, the model is solved backwards. We begin with Stage 3: 8.1 Model Derivations 8.1.1 Stage 3: Production Decisions Exporting to the US Consider a Bangladeshi exporter with productivity ' and demand shock v . The US does not give tari¤ preferences to Bangladeshi garments, and the presence of country-speci…c quotas in most categories makes meeting ROOs mandatory for exports. This means that Bangladeshi …rms exporting to the US have no choice but to meet ROOs. (Recall that we assume that documentation costs are zero in the US and that as only assembly is required for origin, meeting ROOs is costless so = 1:) They have to pay the tari¤ of 20%. As a result, the …rm sells quantity US pBD;U S RBD;U S 1 BD;U S+ qBD;U S (pBD;U S ; v ) = v ; pBD;U S = ; (17) PBD;U S PBD;U S (1 tBD;U S ) US' and earns the following revenues and pro…ts: 1 US US 1 1 BD;U S + rBD;U S ('; v ) = (1 tBD;U S ) vRBD;U S (PBD;U S ) ; (1 tBD;U S ) US' (18) BD;U S ('; v ) = rBD;U S ('; v ) = US f: (19) The price set by a …rm does not depend on its market speci…c shock v . However, v a¤ects 47 s pro…ts: for any '; there exists a minimal demand shock v ('; PBD;U S ) ; such that the …rm’ BD;U S ('; v ('; PBD;U S )) = 0; so that from (18); (20) US 1 USf 1 BD;U S + v ('; PBD;U S ) = 1 : (21) (1 tBD;U S ) RBD;U S (PBD;U S ) US 1 tBD;U S US' The mass of Bangladeshi exporters selling in the US is, thus: Z +1 Z +1 BD MBD;U S = ME hU S ( )g (')d d'; (22) 'BD;U S (';PBD;U S ) BD where ME denotes the mass of entrants in Bangladesh and 'BD;U S is de…ned below. Exporting to the EU When Bangladeshi …rms export to the EU, they have an additional choice: invoke ROOs and pay zero tari¤s, or ignore the preferences and pay the tari¤ tBD;EU without meeting ROOs. If …rms meet ROOs, they incur an additional documentation cost of dEU as well as an increase in marginal costs due to not using the least cost input mix. As a result, only …rms with very favorable demand shocks will choose to meet EU ROOs. ROO The …rm maximizes max 0; BD;EU ('; v ) ; BD;EU ('; v ) ; where BD;EU BD;EU ('; v ) = max (1 tBD;EU ) pqBD;EU (p; v ) qBD;EU (p; v ) f ; p ' ROO BD;EU BD;EU ('; v ) = max pqBD;EU (p; v ) qBD;EU (p; v ) f + dEU : (23) p ' The pricing rules for exporters not to invoking preferences and invoke them are given by 1 BD;EU BD;EU pBD;EU = and pROO BD;EU = ; 1 tBD;EU EU ' EU ' respectively. Exporters that do not invoke preferences earn revenues and pro…ts of: 1 EU EU 1 1 BD;EU rBD;EU ('; v ) = (1 tBD;EU ) vRBD;EU (PBD;EU ) ; (24) 1 tBD;EU EU ' rBD;EU ('; v ) BD;EU ('; v ) = f; (25) EU 48 while exporters that invoke preferences have 1 EU ROO EU 1 BD;EU rBD;EU ('; v ) = vRBD;EU (PBD;EU ) ; EU ' ROO ROO rBD;EU ('; v ) BD;EU ('; v ) = f + dEU : (26) EU For any '; we can de…ne 2 demand shock cuto¤s, v ('; PBD;EU ) and v ROO ('; PBD;EU ) : Firms with v 2 v ('; PBD;EU ) ; v ROO ('; PBD;EU ) choose not to invoke preferences as their demand shocks are too small to cover the …xed documentation costs of dEU . Firms with v > v ROO ('; PBD;EU ) choose to meet EU ROOs. The cuto¤ shocks are implicitly de…ned by BD;EU ('; v ('; PBD;EU )) = 0; (27) ROO BD;EU '; v ROO ('; PBD;EU ) BD;EU '; v ROO ('; PBD;EU ) = 0; (28) where (28) comes from setting the additional pro…ts from invoking EU ROOs to zero. Thus, EU 1 EU f 1 BD;EU v ('; PBD;EU ) = 1 ; (29) (1 tBD;EU ) RBD;EU (PBD;EU ) EU 1 tBD;EU EU ' h i EU 1 EU BD;EU EU d EU ' v ROO ('; PBD;EU ) = 1; or (30) ( EU 1 (1 tBD;EU ) EU ) RBD;EU (PBD;EU ) EU v ROO ('; PBD;EU ) = C ROO v ('; PBD;EU ) ; where (31) ROO dEU C = > 1: f EU 1 (1 tBD;EU ) EU 1 Note that from (30) and (31), v ROO ('; PBD;EU ) and v ('; PBD;EU ) are decreasing in ': As shown below, only …rms with productivity ' > 'BD;j will try to access market j , where cuto¤s 'BD;j are de…ned below in the stage 2 problem. Thus, the masses of exporters that sell in the EU, but do not or do meet the EU ROOs are, respectively: Z +1 Z ROO (';P BD;EU ) N ROO BD MBD;EU = ME dHEU ( )dG('); (32) 'BD;EU (';PBD;EU ) 49 Z +1 Z +1 ROO BD MBD;EU = ME dHEU ( )dG('): (33) 'BD;EU ROO (';P BD;EU ) 8.1.2 Stage 2: Market Entry Decision Consider a …rm who has drawn a productivity level and has to decide whether to enter a market. Firms who expect non-negative pro…ts from trying to enter market j will do so. The US Market For any ' and v , the pro…t of selling in the US is rBD;U S ('; v ) v BD;U S ('; v ) = f =f 1 : US v ('; PBD;U S ) Thus, the expected pro…t of entering the US market is Z +1 US US Ev [ BD;U S ('; v )] fm = BD;U S ('; v ) dHU S (v ) fm (34) v (';PBD;U S ) Z +1 v US = f 1 dHU S (v ) fm : v (';PBD;U S ) v ('; PBD;U S ) The expected pro…ts of accessing the US market are increasing in '; since v ('; PBD;U S ) is decreasing in ': Denote the productivity of a marginal Bangladeshi …rm, which is indi¤erent between accessing the US market or not, by 'BD;U S : Then all …rms with ' > 'BD;U S will try to access the US market. 'BD;U S is de…ned by US Ev BD;U S 'BD;U S ; v = fm ; () (35) Z " # +1 US v fm 1 dHU S (v ) = : v ('BD;U S ;PBD;U S ) v 'BD;U S ; PBD;U S f Equation (35) is important for several reasons. First, by solving it, we obtain the minimal demand shock for the marginal …rm from Bangladesh, v 'BD;U S ; PBD;U S ; which is a key step in the estimation procedure: Second, it shows that this demand shock does not depend US fm on the per-unit costs of selling there, only on f and the demand shock distribution HU S (v ) : Finally, knowing v 'BD;U S ; PBD;U S ; we can express the expected pro…ts at Stage 2 for any …rm as a function of its own productivity ' and 'BD;U S : To see this, note that from the 50 de…nition of v ('; PBD;U S ) in equation (21): 1 v ('; PBD;U S ) 'BD;U S US = ; (36) v 'BD;U S ; PBD;U S ' Z +1 v Ev [ BD;U S ('; v )] = f 1 dHU S (v ) (37) v (';PBD;U S ) v ('; PBD;U S ) 2 3 Z +1 6 v 7 =f ' US 1 4 US 1 15 dHU S (v ) : BD;U S 'BD;U S ' v ('BD;U S ;PBD;U S ) v 'BD;U S ; PBD;U S ' Thus, the expected pro…ts of a …rm depend on its own productivity ', the cuto¤ productivity level, 'BD;U S ; the demand shock for that level, v 'BD;U S ; PBD;U S ; and, of course, the distribution of demand shocks. Now let us look at the exporters to the EU. The EU Market As in the US case, for a …rm with productivity '; v BD;EU ('; v ) = f 1 ; (38) v ('; PBD;EU ) ROO v BD;EU ('; v ) BD;EU ('; v ) = dEU 1 : (39) v ROO ('; PBD;EU ) Thus, the expected pro…t of entering the EU market is ROO EU Ev max BD;EU ('; v ) ; BD;EU ('; v ) fm (40) Z vROO (';PBD;EU ) Z +1 ROO EU = BD;EU ('; v ) dHEU (v ) + BD;EU ('; v ) dHEU (v ) fm v (';PBD;EU ) v ROO (';PBD;EU ) Z +1 = BD;EU ('; v ) dHEU (v ) v (';PBD;EU ) Z +1 ROO EU + BD;EU ('; v ) BD;EU ('; v ) dHEU (v ) fm v ROO (';PBD;EU ) Z +1 v = f 1 dHEU (v ) v (';PBD;EU ) v ('; PBD;EU ) Z +1 EU v EU +d ROO 1 dHEU (v ) fm : v ROO (';PBD;EU ) v ('; PBD;EU ) 51 The careful reader may wonder what ensures that the above maximum of two functions can be written in this simple way. This follows from the fact that for any '; v ROO ('; PBD;EU ) = C ROO v ('; PBD;EU ) ; so that, as long as C ROO > 1; at all values of '; the demand shock cuto¤ line in Figure 3 lies below the demand shock cuto¤ line to invoke ROOs. From the expression above, the expected pro…ts of accessing the EU market are increasing in '; so if we denote the productivity of a marginal …rm exporting to the EU by 'BD;EU ; then all …rms with ' > 'BD;EU will try to access the EU market. 'BD;EU is de…ned by ROO EU Ev max BD;EU 'BD;EU ; v ; BD;EU 'BD;EU ; v = fm ; or Z +1 " # v 1 dHEU (v ) (41) v ('BD;EU ;PBD;EU ) v 'BD;EU ; PBD;EU Z " # dEU +1 v EU fm + 1 dH EU (v ) = : f vROO ('BD;EU ;PBD;EU ) v ROO 'BD;EU ; PBD;EU f Again, solving the equation above for v 'BD;EU ; PBD;EU ; we can express the expected pro…ts at Stage 2 for any …rm as a function of its own productivity ' and 'BD;EU : 'BD;EU EU 1 v ('; PBD;EU ) = v 'BD;EU ; PBD;EU ; (42) ' Ev [ BD;EU ('; v )] 2 3 Z +1 6 v 7 = f ' EU 1 4 ' EU 1 15 dHEU (v ) : BD;EU ' v ('BD;EU ;PBD;EU ) BD;EU v 'BD;EU ; PBD;EU ' Moreover, the expected additional pro…ts coming from the possibility of getting a favor- able enough demand shock to invoke ROOs can be expressed as ROO Ev BD;EU ('; v ) BD;EU ('; v ) (43) 2 3 Z +1 EU 6 v 7 = d ' EU 1 4 EU 1 15 dHEU (v ) ; BD;EU 'BD;EU ' v ROO ('BD;EU ;PBD;EU ) v ROO 'BD;EU ; PBD;EU ' 52 where, from the analysis above, v ROO 'BD;EU ; PBD;EU = C ROO v 'BD;EU ; PBD;EU : (44) For our estimation exercise, we use the analysis of Stage 2 with the data available to calculate 'BD;U S and 'BD;EU . Using (20) together with (18) and (19), " #1 1 US 1 US (1 tBD;U S ) RBD;U S (PBD;U S ) BD;U S+ 'BD;U S = v 'BD;U S ; PBD;U S : f US (1 tBD;U S ) US (45) Similarly, the productivity of a marginal EU exporter is given by " #1 1 EU 1 EU (1 tBD;EU ) RBD;EU (PBD;EU ) BD;EU 'BD;EU = v 'BD;EU ; PBD;EU : EU f (1 tBD;EU ) EU (46) 8.1.3 Stage 1: Entering the Industry Entry occurs until the expected pro…ts that could be earned by a Bangladeshi …rms in all their potential markets equal entry costs: US E' max Ev [ BD;U S ('; v )] fm ;0 (47) ROO EU +E' max Ev max BD;EU ('; v ) ; BD;EU ('; v ) fm ;0 = fe : Using the analysis of Stage 2, we can rewrite this expression as 2 2 3 Z +1 Z +1 6 6 v 7 4 ' US 1 4 US 1 15 dHU S (v ) BD;U S 'BD;U S 'BD;U S ' v ('BD;U S ;PBD;U S ) v 'BD;U S ; PBD;U S ' Z (Z US +1 +1 fm v dG (') + 1 dHEU (v ) f 'BD;EU v (';PBD;EU ) v ('; PBD;EU ) Z ) +1 dEU v EU fm fe + ROO 1 dHEU (v ) dG (') = : (48) f v ROO (';PBD;EU ) v ('; PBD;EU ) f f 53 8.2 Solving the Model Numerically Given 13 guessed parameters; how can we solve the model? Take the expression for ex-ante pro…ts from entering EU market, given by equation (41): Z " # +1 v 1 dHEU (v ) v ('BD;EU ;PBD;EU ) v 'BD;EU ; PBD;EU Z " # dEU +1 v EU fm + 1 dH EU (v ) = : (49) f C ROO v('BD;EU ;PBD;EU ) C ROO v 'BD;EU ; PBD;EU f Note that the LHS depends only on the cuto¤ demand shock for the marginal …rm, v 'BD;EU ; PBD;EU , (think of this as a number), while the RHS equals one of the parameters we have set. Simi- larly, if ROOs must be met for the US market, we have (see equation (35)): Z " # +1 US v fm 1 dHU S (v ) = : (50) v ('BD;U S ;PBD;U S ) v 'BD;U S ; PBD;U S f Thus, for …xed values of the eleven parameters, equations (49) and (50) are nonlinear equations in only one unknown, v ('BD;EU ; PBD;EU ) and v ('BD;U S ; PBD;U S ), respectively: Each has at most one solution as the LHS is a decreasing function in v ('BD;j ; PBD;j ). As 'BD;j j 1 shown in (42) and (36), v 'BD;j ; PBD;j = ' v 'BD;j ; PBD;j so that v 'BD;j ; PBD;j can now be written as a function of only 'BD;j and ': This is the key in what follows. Next, we will derive the cuto¤ productivities 'BD;EU and 'BD;U S : To solve for the pro- ductivity cuto¤s, we de…ne a system of two equations with two unknowns. The …rst relation between productivity cuto¤s: The price index of exporters from Bangladesh to the EU (where some …rms meet ROOs and others do not) is given by Z +1 Z ROO (';P BD;EU ) 1 BD (PBD;EU ) EU = ME pBD;EU (')1 EU hEU ( )g (')d d' 'BD;EU (';PBD;EU ) Z +1 Z +1 BD +ME pROO BD;EU (') 1 EU hEU ( )g (')d d' 'BD;EU ROO (';PBD;EU ) BD ME D('BD;EU ); (51) BD where ME is the mass of entrants in Bangladesh. Similarly, the price index of exporters 54 from Bangladesh to the US (where everyone has to meet ROOs to obtain origin) is given by Z +1 Z +1 1 BD (PBD;U S ) US = ME pBD;U S (')1 US hU S ( )g (')d d' 'BD;U S (';PBD;U S ) BD ME D('BD;U S ): (52) Recall that D('BD;j ) does not depend on PBD;j as we solved for ('; PBD;j ) in terms of 'BD;j =' and v 'BD;j ; PBD;j : Thus, from (52) and (51), the ratio of the price indices is just: (PBD;EU )1 EU D('BD;EU ) = : (53) (PBD;U S )1 US D('BD;U S ) Since ('; PBD;j ) is pinned down once we know 'BD;j ; this gives one relation between the price index ratios and the productivity cuto¤s. Also, D(' ) rises as ' falls. From the model, we also know that the ratio of the price indices is de…ned by two zero pro…t conditions (see equations (21) and (29)) so that: BD;EU 1 EU 1 ! (PBD;EU ) 1 EU (1 tBD;EU ) v 'BD;EU ; PBD;EU US RBD;U S (1 tBD;EU ) EU 'BD;EU EU = : (PBD;U S )1 US (1 tBD;U S ) v 'BD;U S ; PBD;U S RBD;EU EU BD;U S + 1 US 'BD;U S US 1 (1 tBD;U S ) US (54) Again, v ('BD;j ; PBD;j ) is solved for. Then the RHS is a function of the cuto¤s alone. Equating the RHS of (53) and (54) gives one equation in 2 unknowns, 'BD;EU and 'BD;U S : 1 EU BD;EU 1 EU v 'BD;EU ; PBD;EU (1 tBD;EU ) U S RBD;EU (1 tBD;EU ) EU 'BD;EU D('BD;EU ) 1 = 1 : v 'BD;U S ; PBD;U S (1 tBD;U S ) EU RBD;U S BD;U S + US 'BD;U S US D('BD;U S ) (1 tBD;U S ) US RBD;U S and RBD;EU are approximated by total Bangladeshi exports of wovens to the US 1 j and EU. Also, since 'BD;j D('BD;j ) falls with 'BD;j ; this equation gives a negative relation between 'BD;EU and 'BD;U S : The second relation between productivity cuto¤s: The next equation we use is the free entry condition (48). It gives a negative relation between 'BD;EU and 'BD;U S : if one cuto¤ rises, the expected pro…ts from that market decline. To keep ex-ante pro…ts constant, the expected pro…ts from the other market must rise, i.e., its productivity cuto¤ must fall. 55