Barriers to Entry, Trade Costs, and Export Diversification in Developing Countries Allen Dennis and Ben Shepherd*, † The World Bank 1818 H St. NW Washington D.C. 20433 Draft April 24, 2007 Abstract: We show that lower entry costs for firms, and lower internal and external trade costs, are strongly and robustly associated with export diversification in developing countries. Specifically, a 10% reduction in internal trade costs increases the number of products exported by 2.5%, while a similar cut in entry costs increases diversity by 1%. These results withstand numerous robustness checks including changes in country sample, diversification measure, and econometric methodology. Our analysis, using new Doing Business data, covers the direct, official costs of market entry, as well as both internal trade costs (document preparation, inland transport, customs, and port charges) and external trade costs (distance and tariffs). In policy terms, our results suggest that one effective way for developing countries to promote export diversification is to focus regulatory reform efforts on making entry procedures simpler and less expensive, as well as on trade facilitation measures. Keywords: International trade; Trade policy; Product variety; Diversification; Economic development. JEL Codes: F12; F13; 024. * The authors are respectively Economist (Doing Business) and Consultant (Development Research Group—Trade). This work is part of a broader project on trade facilitation and development supported through a Trust Fund of the U.K. Department for International Development. We are grateful to Simeon Djankov, Matthias Helble, Bernard Hoekman, Will Martin, and Beata Smarzynska Javorcik for helpful comments on a previous draft. Correspondence to: [email protected] or [email protected]. † The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. 1 Introduction Export diversification has long been a stated policy goal for many developing countries. In this paper, we leave to one side the debate as to how far that goal should be pushed, or which sectors and products should be given priority over others. We approach the question from a different angle: taking the goal of export diversification as a given, what policies are available to developing countries in order to pursue it at minimum economic cost? We show that one set of policy measures that has not received much attention in this area, namely lower barriers to firm entry and lower international trade costs, constitutes an important way in which developing countries can help diversify their export baskets. In concrete terms, we find that 10% reductions in trade costs and entry costs increase export product diversity by 2.5% and 1% respectively. To establish these results, we draw on new data from the World Bank’s Doing Business database, which provides a unique cross-country resource on firm entry restrictions and trade costs. We also construct an original database on export diversity, using highly disaggregated (8digit) import data for the European Union. This dataset allows us to distinguish amongst more than 10,000 different traded products, across 100+ developing countries. Using count data econometric methods, we find that the association between lower barriers to entry and trade costs, and greater export diversity, is extremely robust to model specification, country sample, estimation technique, and choice of diversification metric. Why has export diversification historically been important to developing countries? As is well known, some of them depend on a relatively small number of products, normally agricultural commodities, for a large proportion of their overall export earnings. Intuitively, it seems plausible that there may be scope for welfare enhancing diversification as a means of reducing 1 price and terms of trade risk in the aggregate (Brainard and Cooper, 1968; De Rosa, 1991). Optimal policy in such a framework is therefore a tradeoff between the traditional trade gains from specialization, and a “hedging” gain from diversification. Alongside this approach to diversification, there has also long been a view in development circles that in order to grow rich, poor countries need to alter the composition of their exports so that it looks more like that of rich countries. (Again, we leave to one side the validity of that argument, and the causal chain it suggests.) Traditionally, that point of view has found expression in a preference for exporting manufactures rather than raw materials (e.g., Prebisch, 1959), or more recently in the argument that one way in which countries develop is by latching onto high productivity goods (Hausmann et al., 2006). The view that we take of diversification in this paper is more closely related to the first of these concerns (i.e., hedging) than to the second, since we do not privilege, a priori, one sector or product group over another. Against this background, a number of previous studies have examined the extent to which export diversification is associated with particular economic outcomes. Funke and Ruhwedel (2001a) find a positive link between export diversification and per capita GDP, as well as TFP growth, amongst OECD countries.1 Feenstra and Kee (2006) find that greater export variety is associated with a 4.5% productivity gain for exporters over the 1980-2000 period. There is also mounting evidence of the importance of product variety as a determinant of aggregate trade values (Hummels and Klenow, 2005; Funke and Ruhwedel, 2001b), and as a major source of importers’ gains from trade (Broda and Weinstein, 2006). Coming at the question from the opposite 1 For Chile, the evidence presented by Amin Gutiérrez de Piñeres and Ferrantino (1997) seems to point in the opposite direction: lagged export specialization is found to be positively associated with growth in exports and GDP. However, the authors’ interpretation is that their results “are consistent with the possibility that in the long run, export diversification enhanced Chilean growth performance” (p. 390). 2 direction, Jansen (2004) finds evidence that export concentration is associated with greater volatility both in the terms of trade and national income. Surprisingly, there is much less empirical work on the determinants of export diversification, which will be the focus of this paper.2 Chandra et al. (2007) follow Imbs and Wacziarg (2003) in focusing attention on GDP per capita, as well as additional aggregate factors such as technology, infrastructure, and openness. However, their results do not lend themselves to a direct policy interpretation due to the rather broad nature of the policy variables they use. Feenstra and Kee (2006) address the question of policy determinants using more detailed indicators, though in an indirect way. They use an instrumental variables strategy to isolate exogenous shifts in export variety, and thereby show that it is in part determined by tariffs, distance and resource endowments. The signs on tariffs and distance are generally negative and statistically significant, suggesting that variable trade costs play a role in determining export diversity. Finally, Klinger and Lederman (2004, 2006) examine the factors underlying the rate of export “discoveries”, i.e. instances when a country starts exporting a product that it has not previously exported. They consider per capita income, historical exports, a proxy for the return to exporting relative to domestic production, and Doing Business data on barriers to domestic market entry. They conclude that the first three factors have significant impacts on discovery, but that the direct effect of entry barriers is not significant or robust. 2 In related work, Imbs and Wacziarg (2003) show that production (not export) concentration and per capita income appear to follow an inverse U relationship. However, their analysis using nonparametric techniques is primarily descriptive and cannot be said to establish causality in a strict sense (although the authors discuss a number of theoretical issues which could be developed to account for the relation they observe). In any case, their conclusions cannot simply be extrapolated to the case of export diversification: production diversity is likely to be a poor proxy for export diversity, since countries generally export only a small subset of the total range of goods they produce. 3 Our paper seeks to provide greater detail on the policy determinants of export diversity, and thereby to contribute to the debate as to what might be an appropriate policy set for developing countries seeking to promote export diversification. To do that, we extend the literature in three main ways. Firstly, we embed our analysis in the well-known heterogeneous firms framework of Melitz (2003). In doing so, we follow Feenstra and Kee (2006), as well as a different strand of the literature that looks at the determinants of the extensive margin of trade (e.g., Baldwin and Di Nino, 2006; Amurgo Pacheco, 2006), although it does not make an explicit link to export diversification in a development context. Our version of the Melitz (2003) model suggests that export diversification should be negatively associated with fixed and variable trade costs— including components that are both internal and external to the exporting country—and with domestic barriers to entry. We indeed find strong and robust evidence in favor of that hypothesis. Secondly, our empirical approach treats export diversity as integer count data, namely the number of product lines with positive imports into the EU-15 in 2005. Such a treatment immediately suggests the use of econometric models specific to count data. This departs from the gravity model framework more commonly used to look at the impact of various factors on the extensive margin of trade via zero entries in the bilateral trade matrix (e.g., Helpman et al., 2007). Doing so has a number of potential advantages. On the one hand, the estimation problem—even with over 10,000 products and 100+ countries—is highly tractable. Moreover, the Poisson quasi-maximum likelihood estimator that we use does not suffer from bias or lack of consistency in the presence of fixed effects—a potential problem for some commonly used estimators in the gravity literature, including Tobit and Heckit (see Greene, 2004, for a review of this issue; cf. Santos Silva and Tenreyro, 2006, who apply a Poisson QML estimator to the gravity model.) 4 The closest relative to our approach is to be found in the two papers by Klinger and Lederman (2004, 2006) which we have previously referred to. However, our approach is fundamentally different from theirs in that our dependent variable is not “new” products, but the full range of exported products at a given time. Moreover, we expand on their approach by including a wider range of policy variables motivated by the Melitz (2003) framework, in particular the fixed and variable costs of international trade. The third way in which our paper adds value is in terms of the datasets used. To construct our measure of export diversification, we use highly detailed 8-digit import data for the European Union. These data are largely unexploited in the literature, and offer nearly twice the level of product detail that can be found in their more commonly used 6-digit counterparts. To measure fixed and variable trade costs internal to the exporting country, we use new data from the World Bank’s Doing Business database. It provides by far the most detailed and easily comparable cross-country data on entry costs and internal trade costs. It allows us to distinguish amongst different varieties of internal trade costs, such as transport, customs, port and terminal charges, and document preparation. As far as we know, this is the first occasion on which these trade cost data are used in empirical work. The paper is set out as follows. The next Section presents a brief theoretical motivation for our empirical work, in terms of the Melitz (2003) heterogeneous firms framework. Section 3 presents our dataset, and discusses in detail the way in which we have measured export diversification, trade costs, and barriers to entry. Our empirical results are in Section 4, and we conclude with some policy implications and suggestions for further research in Section 5. 5 2 Theoretical Motivation Since our contribution in this paper is primarily empirical, we do not present a full-blown theoretical model. Rather, this Section provides a simple theoretical motivation for linking export diversification with trade costs and barriers to entry. To do so, we use a simplified version of the Melitz (2003) heterogeneous firms model, due to Baldwin (2005). Our basic contention is simple, and follows directly from the model’s comparative statics. Lower costs mean that domestic firms face lower hurdles when exporting. This enables less efficient producers to start exporting, in addition to those firms already exporting. When each producer is associated with a distinct product variety (as is the case in most monopolistic competition trade models), this means that the country’s export structure becomes more diversified as costs fall, in the sense of including a greater number of different products. To fix ideas, we present a simple formalization following Baldwin (2005). Each of two identical countries (Home and Foreign) is endowed with a single numéraire factor of production (labor) and produces a continuum of goods in a single, differentiated sector with constant elasticity of substitution σ. Trade costs are symmetric at all times. A representative consumer has utility 1 1 1 1 1 U qv dv , where v represents each variety and V is the full set of varieties. Each vV firm produces a single variety of the differentiated product, under increasing returns to scale. On the production side, the dynamics of entry is broken down into three stages. First, each firm pays a fixed innovation cost Fi, which can be likened to the research costs incurred in developing a new variety. Once Fi is paid, each firm is assigned its marginal production cost a randomly, by repeated draws from the distribution G[a], which is usually assumed to be Pareto with shape 6 parameter k and support 0 a a0 (e.g., Chaney, 2006; Eaton and Kortum, 2002). That is, k a Ga . It is the stochastic nature of this process which results in firms having different a0 productivity levels. The second and third stages of the process relate to specific market entry decisions. Selling to the domestic market requires the firm to pay a fixed cost Fd , while exporting entails Fx . These can be conceptualized as, for example, market-specific product adaptation costs (e.g., Baldwin, 2000), or the costs of setting up market-specific distribution and servicing networks (e.g., Helpman et al., 2004). We assume that Fx Fd . One justification for that assumption is that information on the overseas market (such as consumer tastes) is more costly to obtain than for the domestic market. As will become clear below, it also provides a rationale for the wealth of empirical evidence suggesting that only a small percentage of active firms in a country actually export (see Bernard et al., Forthcoming, for a review). The presence of these two fixed costs leads firms to sort into three groups. The present value of firms with high marginal cost draws will be less than Fd , and those firms will therefore maximize profits by not producing at all. For intermediate marginal cost draws, firms will be able to cover Fd but not Fx , and will therefore produce for the domestic market only. Only those firms with low marginal cost draws will be able to support Fx , and they will sell in both markets. In productivity terms, the least efficient firms will exit and the most efficient will produce for both the domestic and export markets, while an intermediate group will sell to the domestic market only. 7 We can use this analysis to define marginal cost cutoffs for domestic production (ad) and export (ax), by association with the corresponding levels of fixed costs. By symmetry, we assume a d a x . For a given cost distribution G[a], the mass of firms that produces in equilibrium will be larger for higher ad. Similarly, the equilibrium mass of exporting firms is larger for higher ax. Free entry by firms implies that the ex ante expected value of developing a new variety that can be profitably produced must be equal to the ex ante expected fixed cost of doing so. Baldwin (2005) shows that this condition makes it possible to derive explicit expressions for the equilibrium domestic and export marginal cost cutoffs ad and ax, as well as the total number of producers n, in terms of fixed costs (Fi, Fd and Fx), “iceberg” trade costs τ, the elasticity of substitution σ, and the parameters of the marginal cost distribution (a0 and k). Defining openness 1 Fx1 Fd 1 such that it ranges between zero (infinite fixed or variable trade costs) and unity (free trade with Fx Fd ), and using k to simplify notation, allows us to express 1 those solutions as follows: 1 F 1 k a x a0 i Fx 1 (1) 1 F 1 k a d a0 i Fd 1 (2) L 1 Fd 1 (3) n The marginal cost distribution G[a] drives the process that leads firms to self-select into production for the domestic market only, or for the domestic and export markets jointly. The 8 equilibrium proportion of exporters in the total mass of firms will therefore be simply Ga x a x Ga d a d k , where the equality follows from our assumption of a Pareto distribution for marginal costs. This allows us to derive an explicit expression for the number of exporting firms in equilibrium: nx Ga x L 1 n Ga d Fx 1 (4) Since each firm produces a distinct variety of the differentiated product, nx is also a measure of export diversification. It is related to openness Ω, which suggests that export diversity will be affected by changes in the fixed and variable costs of trading, as well as domestic market entry costs Fd. To see this, we conduct some simple comparative statics (cf. Results 2-3 and 11 in Baldwin, 2005). As a preliminary, we note that openness is decreasing in both fixed and variable trade costs, but increasing in the fixed costs of domestic market entry: d 1 0 dFx Fx (5) d 1 0 d (6) d 1 1 0 1 dFd F F x d (7) While there is nothing surprising about (5) and (6), the intuition behind (7) is not immediately clear. In fact, the reason for the sign of the derivative in (7) is largely mechanical. The model is of course highly stylized in its presentation of Fd and Fx, the only constraint being that the fixed cost of exporting must be greater than the fixed cost of domestic market entry. It is therefore 9 legitimate to assume that Fx in fact includes the cost set contained in Fd, with some additional amount to represent the added burden of foreign market entry. We therefore set dFx = dFd when deriving (7). As a result, a given absolute change in Fd causes an identical absolute change in Fx which—assuming as always Fx Fd —changes the ratio of export to domestic entry costs. And it is this ratio which matters for openness as we have defined it. Using the above results, we now differentiate (4) sequentially with respect to Fx , τ and Fd :3 dn x n x 0 1 Fx dFx (8) dn x n x 1 0 1 d (9) 1 1 dn x 1 1 nx 0 1 Fx Fd Fx dFd (10) Equations (8) through (10) directly support our contention that lower fixed and variable trade costs, and lower barriers to entry, are associated with export diversification. The intuition behind these results is simple. Within the framework of this model, exporters must pay both the costs of domestic market entry and a supplementary cost related specifically to exporting. Hence, lower barriers to domestic market entry translate directly into lower costs both for Home-only producers, and for exporters. Similarly, lower fixed and variable trade costs also lower costs for 3 The sign of the derivative in (10) is ambiguous. It is negative provided Fx 1 Fd , which we interpret as a condition that export costs not be “too high” with respect to domestic entry costs. In what follows, we assume that that condition holds. 10 exporters. For given country technology (i.e., marginal cost distribution), lower costs for exporters make it easier for firms to shift from producing for the domestic market only, to producing for both the domestic and export markets. The mass of exporting firms therefore increases, which mechanically leads to greater product variety in exports since the new exporters make different varieties from existing ones—the country’s export bundle becomes more diversified. Conceptualizing the process of diversification in this way is important. It emphasizes that when “new” products appear in export data, they do not necessarily represent pure innovation in the exporting country. They can also arise from existing domestic producers who decide that it is now profitable to begin exporting (cf. Klinger and Lederman, 2004 and 2006). Intuitively, we would argue that this is an appealing feature of the model which reflects an easily recognizable stylized fact. 3 Data We have used a simplified Melitz (2003) framework to show that lower trade costs and barriers to entry should promote export diversity. In this and the following Section, we take that hypothesis to the data. Our variables and sources are fully set out in Table 1, and the list of countries we consider is in Table 2. A number of our data sources are standard, and do not require any particular discussion. However, our dataset also contains two novel aspects that do require a more detailed presentation. Firstly, we measure export diversity using largely unexploited 8-digit trade data for the European Union. Secondly, we use data from the World Bank’s Doing Business report to measure entry restrictions and trade costs. We now address each of those particularities in turn. 11 3.1 Measuring Export Diversification The literature discloses a number of different ways of measuring export diversification. One set of papers uses variations on the Herfindahl-Hirschman index (e.g., Amin Gutiérrez de Piñeres and Ferrantino, 1997; Imbs and Wacziarg, 2003), while another follows the relative variety approach of Feenstra (1994). Still another strand of the literature analyzes diversification indirectly, by looking at zero entries in disaggregated trade data (Helpman et al., 2007; Baldwin and De Nino, 2006). We choose to take a more direct approach, by defining diversification in terms of a count of the number of products exported, either by a country as a whole, or in a particular sector. Simple counts have the advantage of operationalizing a transparent and natural definition of diversification, which corresponds directly to the way in which diversity is conceptualized in our theoretical framework presented in the previous Section. Given sufficiently detailed trade data, it should be a simple matter to calculate country-bycountry counts of the number of product lines in which exports take place. But therein lies the problem: the most detailed level of trade data available on a worldwide basis is at the HS 6-digit level, which differentiates among 5,000 or so “products”. However, each of these 5000 codes in fact aggregates together a number of individual varieties, and counts based upon them therefore tend to understate the true level of diversity in exports. To resolve this problem, we use detailed import data for the European Union. These data are freely available through the Eurostat website (http://fd.comext.eurostat.cec.eu.int/xtweb/). They distinguish amongst 10,000 or so individual “products” at the CN 8-digit level. While there is still some degree of aggregation involved, it is considerably less than for the HS 6-digit classification. We therefore expect count data constructed using this dataset to paint a more 12 accurate picture than could be obtained with 6-digit data. These data are comparable in detail to the US 10-digit data used by, for instance, Feenstra and Kee (2006). To our knowledge, they have not previously been used in empirical work on product variety in trade.4 The main objection that can be made to our use of these data is that they in fact capture diversity of exports to the EU, rather than towards all trading partners. That is indeed the case. However, the EU represents an important export outlet for many developing countries, particularly in Africa, which suggests that we can be confident that we are capturing a significant part of their overall export pattern. In any case, the only ready alternative—use of HS 6-digit export data— has two major disadvantages which lead us to favor the approach we have just set out. Firstly, 6digit data is arguably too aggregated to serve as a reliable guide to export diversity. Secondly, export data from developing countries are generally considered less reliable than “mirror” import data from developed countries. We therefore believe that our dataset represents a defensible compromise between accuracy and completeness. In any case, replication of our results using alternative data is a potential avenue for future research. Taking 2005 as our base year, we use these data to construct two measures of export diversification for the trade partners of the EU-15 (aggregated to a single importer). We start with a dataset of 470,035 observations across 246 countries and customs areas (including the EU), and 10,753 distinct products. In this paper, we focus only on the developing country component of that dataset, namely countries that are neither members of the EU-25 nor the OECD. (We return to this definition in the context of robustness checks below.) Our first measure of export diversification, lines, is a simple count of the number of CN 8-digit product 4 Indeed, these data are an underused resource more generally. The only published applications appear to be Fontagné et al. (1998), Henry de Frahan and Vancauteren (2006), and Manchin (2006). 13 lines in which a given country exported to the EU-15 in 2005. It has one observation per country. To provide greater detail, we also construct lines_cn2 following the same pattern as for lines, but with counts by 2-digit sector rather than aggregated to the country level. Lines_cn2 therefore has 97 observations per country (the number of 2-digit Chapters in the CN classification). Given that the CN 8-digit classification scheme is inconsistent in the level of detail (i.e., individual “products”) it accords to each sector, we will need to take care to correct for this when using lines_cn2 as an indicator of export diversification. We return to this issue below. In Table 3, we provide a ranking of countries in terms of their level of export diversification as measured by lines (i.e., at the aggregate, not sectoral, level). On average, the countries in our sample—excluding the EU-25 and OECD members—exported 1,138 8-digit product lines to the EU in 2005. However, the range is extremely wide: from 9 lines (Palau) to 8053 (China), out of a possible maximum of 10,753. In broad terms, the country rankings accord with the sensible prior that larger, more developed countries tend to have more diversified export bundles. Thus, we find China, India, and Brazil at the top of the table, while Palau, Micronesia, and the Comoros are at the opposite end. In order to gauge the robustness of our data against other diversification measures that have been used in the literature, we also calculate Feenstra’s lambda measure of relative variety,5 and the Herfindahl-Hirschman index of export concentration.6 Rankings across 5 Feenstra and Kee (2006) show that country e’s export variety relative to the world as a whole can be measured as p p w w i i lambdae q iJ e iJ w w w i i q , where the numerator is the total value of world exports in product lines exported by e, and the denominator is the total value of world exports across all products. 14 all three measures are reasonably similar, and are quite close for lines and lambda. The sample correlations with lines are 0.95 and -0.52 for lambda and hh_index respectively. In other words, we can be confident that our measure of diversification squares well with alternative approaches—but this is, in any case, a robustness issue that we return to below. 3.2 Trade Costs Our theoretical presentation in Section 2 identified four distinct costs facing potential exporters: the fixed costs of innovation, domestic entry and export market entry, and variable “iceberg” trade costs. While these costs are well articulated in the theoretical model, their empirical content is quite broad. One the one hand, they cover “internal” trade costs, i.e. those incurred by an exporter between the factory gate and the exporting country dock. But they also include “external” trade costs incurred between the exporting country dock and the importer. The breadth of factors involved makes data difficult to come by on a cross-country basis. For instance, the annual Global Competitiveness Reports published by the World Economic Forum have previously been used to measure trade costs related to infrastructure and logistics performance (e.g., Wilson et al., 2005). However, the indices included in the GCR are perception based and are therefore subject to potentially significant measurement error. Secondly, data are not collected for some developing countries, meaning that the sample size is likely to be considerably reduced in a study like this one. 6 The index for exporter e is simply the sum of the squared market shares of each export product i (with prices p and 2 J pi qi quantities q): hh _ indexe J . i 1 p j q j j 1 15 The World Bank Doing Business survey now provides a complement to existing perception measures in its “Trading Across Borders” section. For its 2006 and 2007 reports, Doing Business has included information on the number of procedures required for importing and exporting, as well as the time taken to comply with them. (See Djankov et al., 2006, for an empirical application.) For 2007, “Trading Across Borders” was expanded to include a trade costs indicator as well. The Doing Business trade costs indicator captures the total costs for importing and exporting a standardized cargo of goods (“Export Cost”), excluding ocean transit and trade policy measures such as tariffs. The four main components of the costs that are captured are: costs related to the preparation of documents required for trading, such as a letter of credit, bill of lading, etc. (“Document Costs”); costs related to the transportation of goods to the relevant sea port (“Inland Transport Costs”); administrative costs related to customs clearance, technical controls, and inspections (“Customs Costs”); and ports and terminal handling charges (“Ports Costs”). These trade costs are provided for both imports and exports on the basis of a standard set of assumptions, including: the traded cargo travels in a 20ft full container load; the cargo is valued at $20,000; and the goods do not require any special phytosanitary, environmental, or safety standards beyond what is required internationally. Data are collected from local freight forwarders, shipping lines, customs brokers, and port officials. For the purposes of this study, we use the export component of the Doing Business trade cost dataset, which covers 175 countries. To our knowledge, these data have not previously been used in empirical work. They disclose a considerable range of country experiences: in some countries these export operations cost as little as $300-$400 (Tonga, China, Israel, Singapore, and UAE), whereas they run at nearly ten times that level in others (Gabon, and Tajikistan). On average, the cost is around $1278 per container (excluding OECD and EU countries). 16 The broad definition of export costs used by Doing Business is useful in the present context since it includes most additional cost components faced by domestic firms that seek to export. As the above discussion suggests, some of these costs have both fixed and variable components: for instance, export documentation needs to be agreed and drafted prior to any export activity taking place (fixed cost), but then needs to be copied and slightly adapted for each shipment (variable cost). However, Doing Business data do not cover two important aspects of trade costs, namely tariffs in importing countries (the EU in this case) and the cost of shipping from the exporting country to the importing countries. In what follows, we use WITS-TRAINS data on tariffs,7 while we proxy international transport costs using distance, as in the gravity model literature. We take Germany to be the “center” of Europe, and therefore measure the distance between each exporting country and Germany. 3.3 Barriers to Entry Doing Business also provides information on the costs of domestic market entry in 175 countries, through its “Starting a Business” section. (For a detailed discussion of methodology, see Djankov et al., 2002.) It includes indicators on the costs, time, and number of procedures required for an entrepreneur to start-up and formally operate a local limited liability company with general industrial or commercial activities. This includes pre-registration, registration, and post-registration activities legally required in the country (“Entry Cost”). Information on the paid-in minimum capital required before starting a business is also recorded (“Minimum Capital”). Only official costs are considered. The company law, commercial code, and specific 7 In future versions of this paper, we will replace TRAINS tariff data with applied tariffs from the MAcMap v2 database (Laborde et al., forthcoming). The advantage of MAcMap over WITS is that it fully accounts for preferential agreements, and is fully disaggregated to the HS 6-digit level even when no trade is observed for a given country pair-product combination. 17 regulation and fee schedules are used as sources for calculating costs. The information is provided by incorporation lawyers and government officials. As far as we are aware, this is the most comprehensive source of information on business start-up costs, and hence it is the dataset best suited to being used as a measure of domestic market entry costs (i.e., Fd) in our study. It has previously been used in related contexts by Klinger and Lederman (2004, 2006), and Helpman et al. (2007). 4 Empirical Model and Results Summarizing from Section 2, the Melitz (2003) framework leads us to expect that lower entry barriers and trade costs (both fixed and variable) will be positively associated with export diversification. We start the empirical testing of that hypothesis by conducting some exploratory analysis using nonparametric regression techniques (cf. Imbs and Wacziarg, 2003). Figures 1 and 2 show output from locally weighted regression (Lowess smoothing) of lines on entry cost (in % GNI per capita) and export cost (in US dollars).8 Both Figures show a clear, negative association between export diversification, as measured by an aggregate country-level count of products exported, and entry or trade costs. The relationship between the variables is not, however, a strictly linear one. Visually, it would appear to be more consistent with a logarithmic relation. The plots suggest that a small absolute reduction in very high trade or entry costs is associated with only a relatively small absolute increase in export diversification, while a similar reduction starting 8 Each figure is obtained by running a separate regression of lines on the relevant trade costs variable at each data point, using 80% of the total effective sample. For each regression, observations are downweighted according to their distance from the central data point around which the regression takes place. All calculations are performed in Stata 9.2SE. In this case, we use the mlowess module developed by Cox (2006). 18 from moderate cost levels can be expected to have a considerably greater impact. Although we can only speculate at this point as to the reasons why the cost-diversification relationship takes this form, it seems plausible that very high levels of costs are close to prohibitive, and in any case are well in excess of the levels that would allow moderately productive firms to overcome them and start producing and, perhaps, exporting. In terms of the Melitz (2003) framework we have used above, very high absolute costs of entry and export place the domestic and export market productivity cutoffs a long way to the right in terms of the productivity distribution of domestic firms. We have assumed that that distribution is Pareto, which means that it is strongly skewed towards the left. Small cost shifts that keep the productivity cutoffs in the right tail of the distribution would therefore not be expected to have a large impact on the equilibrium number of firms with sufficiently high productivity to meet them. In any case, a deeper investigation of such questions could be a fruitful avenue for future research. Suggestive though these results are, they do not yet account for the many other influences which could simultaneously be acting on our outcome variable, namely export diversification. To try and separate out the role that entry and trade costs play, we now move to a more traditional parametric regression framework. As discussed in the previous Section, our measure of diversification by sector (lines_cn2) is observed repeatedly across a number of countries. This suggests that we can exploit both cross-country and cross-sectoral variation in estimating the relationships of interest. Adopting a panel data framework also has the advantage of allowing us to control for one dimension of variation using econometric methods rather than additional data. Since the variables that are of main interest to us vary in the country dimension, we choose to model cross-sectoral variation in terms of unobserved heterogeneity to be captured by a set of fixed effects. This method will allow us to take proper account of factors which are specific to 19 sectors but common to all exporters, such as non-discriminatory trade barriers (e.g., MFN tariffs and non-tariff barriers) and regulatory measures (e.g., product standards) imposed by the European Union on exports from third countries. It also controls for differences in the level of recorded 8-digit product detail within individual 2-digit sectors. Our dependent variable, lines_cn2, displays an important characteristic that will need to be taken into account in estimation: it is discrete (not continuous), and is composed exclusively of nonnegative integer values. Estimators built to deal with count data will be more appropriate than a general estimator—such as OLS—that deals well with continuous, unbounded data. We therefore turn to the workhorse model for count data, namely the Poisson estimator. (On Poisson and other count data methods, see generally: Cameron and Trivedi, 2001; Wooldridge, 1997; and Winkelmann, 2000). Concretely, we postulate that lines_cn2, can be described by a Poisson process with mean and variance equal to μes (indexing over e for exporters and s for sectors). Its density conditional on the set of independent variables Xes is given by: f lines _ cn2es X es exp es es lines _ cn2es ! lines _ cn 2es (11) The conditional mean is itself assumed to be a function of the set of independent variables Xes. We follow standard practice, and parameterize the conditional mean as an exponential function. We include exporter barriers to entry (entry), internal and external trade costs (internal_costs and external_costs), a full set of sectoral fixed effects (fs), and J additional exporter dimension control variables (control). The first three elements come directly from our theoretical presentation above, and are designed to capture entry costs, and fixed and variable trade costs in terms of the Melitz (2003) model (i.e., Fd, Fx, and τ). We include additional exporter control 20 variables to account for other country-specific factors that might also influence export diversification, such as the economy’s size, structure, and level of development, as well as macroeconomic conditions.9 Finally, the fixed effects account for unobserved cross-sectoral heterogeneity, as discussed above. J es f s expa logentrye b loginternal_costse c logexternal_costse d i logcontrolei (12) i 1 Conditional maximum likelihood techniques can be used to provide parameter estimates for the model defined by (11) and (12), namely a, b, c, and the J control coefficients d (Hausman et al., 1984). Given the functional form of (12), all coefficients can be interpreted as elasticities.10 It may seem overly restrictive to apply the Poisson model—which assumes equality of mean and variance in lines_cn2—and indeed the general literature referred to above discloses a number of alternative count data estimators. The most common is the negative binomial, which differs from Poisson in that the assumed distribution for the dependent variable exhibits over-dispersion: i.e., its variance is larger than its mean (whereas the two are identical under Poisson). This might appear to be a strong factor in favor of the negative binomial over Poisson in many empirical settings, since real world data are often over-dispersed. However, Poisson has an important property which, we would argue, makes it more appropriate as a workhorse model in this context: it can be shown using quasi-maximum likelihood reasoning that Poisson estimates are consistent under quite weak assumptions (Gourieroux et al., 1984). Indeed, the data do not have 9 We note in passing that larger countries should clearly be expected to export a more diversified product basket. By inspection of (4), it is obvious that dn x 0. dL 10 Prior to transforming the independent variables into logarithms, we add unity to those series containing zeros. Where small negative values are observed—for instance, in the GDP deflator series—we replace them with zeros prior to transformation, so as to conserve sample size. 21 to follow a Poisson process at all, and may be over- or under-dispersed. Essentially all that is required for consistency is that the conditional mean function (12) be properly specified. The negative binomial model, by contrast, is only consistent if the conditional distribution of the dependent variable is in fact a negative binomial. As a result, the potential gain from preferring the negative binomial over Poisson—increased efficiency if the data are over-dispersed—needs to be balanced against the more restrictive conditions that have to be met to ensure consistency. It is for this reason that we use the Poisson estimator to produce our baseline results, and use the negative binomial as a robustness check only. (For a more detailed exposition of this reasoning, see Wooldridge, 1997.) Table 6 presents regression results using the fixed effects Poisson estimator.11 We estimate six models in all, using different combinations of dependent variables. In terms of (12), we use the following sets of variables (see Table 1 for a full description): entrye = {Entry Capital, Entry Cost} (13a) internal_costse = {Export Cost} (13b) external_costse = {Distance, Tariffs} (13c) controle = {GDP, GDP per Capita, Manufacturing % GDP, Agriculture % GDP, GDP Deflator, Exchange Rate, Interest Rate} (13d) Our preferred specification in Table 6 is Model 1. It includes all variables listed above except Tariffs and Exchange Rate. The reason for excluding them is simply data availability: Model 1 represents in our view the best compromise between variable set and number of observations 11 We report robust quasi-maximum likelihood standard errors (Wooldridge, 1999), as computed by the xtpqml module in Stata (Simcoe, 2007). 22 (10,088 across 104 countries if the two variables are excluded, versus 2,997 across 31 countries on average if they are included). But in any event, little turns on our preference for Model 1 in terms of the qualitative interpretation of our results: the main coefficients are remarkably stable in terms of sign, significance, and even magnitude, across Models 1-6. The first thing to note about Model 1 is that it represents an extremely good fit to the data: R2 is 0.97, and a comparison of the kernel densities for the dependent variable and its predicted values shows that the two are virtually indistinguishable on the basis of visual evidence (see Figure 3).12 Addressing the results for Model 1 in detail, we see that the coefficient on entry costs is negative (with elasticity -0.1) and statistically significant at the 1% level; the minimum capital requirement has an unexpected positive sign, but is statistically insignificant at the 10% level. Internal trade costs are also negative (elasticity = -0.25) and statistically significant (1%), as is distance (elasticity = -0.46). From the estimated coefficients, we conclude that export diversification appears to be particularly sensitive to international transport costs—as proxied by distance—but is also significantly affected by internal trade costs and entry costs. These results are clearly consistent with the theoretical framework developed in this paper. Most of the control variables in Model 1 also carry the expected signs, and are statistically significant at the 10% level. This is the case for aggregate and per capita GDP and the size of the manufacturing sector (all of which enter positively), as well as for the GDP deflator (which enters negatively). In other words, bigger and richer countries tend to produce more varieties, as do those with more moderate rates of inflation, which could be associated with a stable 12 We are conscious that the standard R2 measure that we present (1-ESS/TSS) does not, strictly speaking, apply to the Poisson model, since there is no residual as such. However, we follow Wooldridge (1997) and present this R2 nonetheless, since it provides a convenient summary indicator of model fit. Its interpretation in that sense, if somewhat approximate, is more straightforward than the pseudo-R2 measures more commonly used for count data models. 23 investment climate. The estimated coefficient on the size of the agricultural sector is positive, but not statistically significant. Only the real interest rate coefficient is surprising, since it is positive and statistically significant at the 5% level. One possible reason behind both of these less encouraging results could be correlation amongst the dependent variables: the size of the agricultural sector is strongly related to GDP per capita (ρ = -0.5), while the interest rate is correlated with the GDP deflator (ρ = -0.8). All in all, we conclude that the Poisson estimates of Model 1 provide strong evidence in favor of our primary contention, namely that lower costs of entry and exporting are associated with greater export diversification in developing countries. As already noted, these results are quite insensitive to model specification: entry and trade cost variables are consistently negative and significant at the 1% level in Table 6. The only exceptions are tariffs in Model 3, and entry costs in Model 6; however, these variables are negative and statistically significant in all other formulations. Results are more mixed for entry capital, which has the wrong sign in two models and is not statistically significant in our preferred specification. We interpret that as indicating that it is the direct costs of entry which represent the more serious constraint on potential exporters. Comparing results from Models 1 and 3-6 with those from Model 2—which includes only our three core variables for entry and trade costs—shows that addition of control variables lowers (in absolute value) our estimate of the sensitivity of export diversification to these costs, but does not in any sense eliminate the qualitative result. Concretely, we find that adding controls reduces the entry cost elasticity from -0.26 to -0.10, and the export cost elasticity from -0.90 to -0.25 (all significant at the 1% level). 24 Thus far, we have analyzed the impact of trade and entry costs in a global sense. However, the Doing Business dataset allows us to look in more detail at the individual components of our internal trade costs measure, in order to gauge whether particular cost types have a stronger or weaker impact on diversification. We therefore disaggregate internal trade costs into four categories: document preparation, customs, inland transport, and terminal handling (ports). Table 7 shows results for Model 1 with export costs replaced by these four variables (which together sum to the original export costs variable). Again, the qualitative conclusion to be drawn is clear: entry costs and trade costs impact negatively on export diversification. In terms of the particular cost categories we have identified, we find that customs and document preparation costs have the strongest negative impacts (elasticities of -0.05 and -0.01 respectively). The coefficient on inland transport costs is also negative, but is very small (elasticity = -0.001) and is statistically insignificant. The only surprising result is that the estimated coefficient on port costs is positive and statistically significant. Again, this might be related to correlation with other explanatory variables, such as GDP (ρ = -0.3). While this question deserves further research in the future, we would argue that it does not significantly detract from our results in an overall sense. 4.1 Robustness Checks There are a number of potential criticisms of our results that we can deal with via simple robustness checks. We divide these into three groups: estimation method, choice of country sample, and construction of our variety measure lines_cn2. We consider first the issue of estimation technique. Our preferred method is Poisson conditional fixed effects. As discussed above, Poisson provides consistent estimates under relatively weak assumptions. However, the negative binomial can provide efficiency gains under appropriate circumstances. We therefore re-estimate Model 1 using the fixed effects negative binomial 25 estimator (Hausman et al., 1984). For comparative purposes only, we also present results for simple OLS fixed effects and Tobit estimation. Finally, to take account of the fact that our main variables of interest vary only at the country level, and not the sector level (cf. Moulton, 1990), we re-estimate Model 1 using the lines variable which is calculated for each country based on exports of all products (i.e., one observation per country.) Table 8 shows clearly that none of the alternative estimation methods produces substantially different results from those obtained via Poisson. Both the negative binomial and OLS estimates suggest that all trade and entry cost variables—including the minimum capital requirement— enter negatively, and are 1% significant. These coefficients are also negative using Tobit, although entry capital is only significant at the 5% level; all other entry and trade costs variables are 1% significant. Even the somewhat extreme step of aggregating the data to the country level—and reducing our sample size from over 10,000 to just 104 observations—does not substantially alter our results: all three entry and trade costs variables have the expected negative signs and coefficients very close to those reported in Table 6. Export costs remain significant at the 1% level, but entry costs are now only significant at the 15% level (prob. = 0.12). Nonetheless, we would argue that this is a relatively minor loss in significance given the extreme drop in sample size that this exercise imposes. It should not be seen as altering our conclusions. A second set of potential criticisms relates to our country sample: all non-OECD and non-EU countries represents a very broad definition of the developing country group, which could mask heterogeneity according to income level. To address this criticism, Table 9 presents Poisson estimation results for Model 1 using progressively narrower country samples. The first column excludes, in addition to those countries already excluded from Model 1, all countries in the World Bank’s high income group. The second column excludes both high income and upper 26 middle income countries, while the third column includes only low income countries. Our conclusions emerge unaltered from this process: export costs and entry costs retain their negative coefficients and 1% significance. Indeed, our conclusions are arguably strengthened, since there is weak evidence that the elasticities on trade costs and entry barriers increase in absolute value as group income falls. In other words, relatively poor developing countries would seem to have the most to gain in terms of diversification from reductions in entry barriers and trade costs. Finally, there are a number of grounds on which our metric of export diversification could be called into question. On the one hand, it could be argued that our theoretical framework is more appropriate to manufactured goods than to agricultural products, yet lines_cn2 includes both groups. Alternatively, the same variable might be said to give too much weight to very small export flows that do not represent diversification that is meaningful in a policy sense. Poisson results in Table 10 show that neither criticism holds water. Our results are just as strong when we construct a new dependent variable, lines_cn2_man, which only includes products in HS chapters 25-97 (i.e., manufactured goods). The same applies when we construct lines_cn2_100k and lines_cn2_1m which only count as exports product lines with a value to the exporting country of at least €100,000 or €1 million in 2005. Our use of simple product line counts in order to capture diversification could also be questioned on the basis that the previous literature devotes greater attention to two alternative, and potentially quite different, measures. On the one hand, the diversification literature privileges measures of concentration such as the Herfindahl-Hirschman (HH) index, which take account of both the number and value of products exported. The product variety and extensive margin literature, on the other hand, prefers the measure of relative variety developed by Feenstra (1994) based on a CES aggregator. (See above and Table 1 for the formulae relating to each measure.) 27 To ensure that our results are robust to the choice of diversification measure, we therefore reestimate Model 1 using alternately hh_index_cn2 and lambda_cn2. In light of our theoretical framework, we expect trade and entry cost coefficients to be positive in the first formulation (since diversification is decreasing in the HH index), and negative in the second. As can be seen from Table 10, the change in diversification measure does not have any significant impact on the qualitative interpretation of our results: entry costs and trade costs carry the expected signs, have sensible magnitudes, and are statistically significant at the 10% level or better in all five columns of the Table. Moreover, Table 10 makes clear that lines_cn2 has an important potential advantage over the two other measures when applied at the sectoral level. From the way in which both the HH Index and Feenstra’s lambda are constructed, they break down when a country does not export any goods in a particular sector. This is why the last two columns of Table 10 contain many missing observations—approximately one third of the sample used in Table 6 (3,393 observations). Our count variable, on the other hand, codes these observations quite naturally as zeros. As a result, our Poisson estimates take account of all available information, including the fact that some countries doe not export any product lines at all in particular sectors. In sum, we conclude that our model survives very well in the face of extensive robustness checks. Our results are, we would argue, unlikely to be an artifact either of the way in which we measure diversification, the countries we analyze, or the econometric methods we apply. 5 Conclusions, Policy Implications, and Questions for Future Research We have used highly disaggregated EU trade data along with new information from the Doing Business database to show that export diversification is inversely related to trade costs and entry 28 restrictions. This association is found to hold using both parametric and nonparametric techniques, and is very robust to changes in methodology and control variables. Moreover, we would argue that our results should be given a causal interpretation: they sit well with the theoretical motivation presented in Section 2, and our policy variables can reasonably be regarded as exogenous to diversification. This suggests that developing countries can promote export diversification by acting to reduce entry and trade costs. Trade costs here include both those which are internal to developing countries themselves—customs and port charges, and document preparation—and those which are external to them—distance and tariffs. This in turn suggests that countries can act to reduce trade costs both unilaterally, by focusing on their internal constraints, and regionally or multilaterally in terms of their external constraints. Whilst our study does not constitute a full welfare analysis of the various policy options available for promoting diversification, its conclusions are nonetheless suggestive of a simple, but important, result: encouraging diversification does not have to equate with protecting “infant” industries or pursuing a selectively active industrial policy. Rather, the same ends can be achieved using different means, such as making it easier for firms to enter the market, and making it cheaper for them to export once they are producing. In a static sense, we expect that the net balance of costs and benefits from diversification through trade and entry cost reduction is likely to be significantly more positive than for diversification through infant industry protection. After all, one set of measures reduces distortions in the economy, while the other increases them. Moreover, such policies are likely to be considerably more efficient in a dynamic sense, since they reduce rents and sharpen competitive forces. Infants raised on such a diet are, we expect, more likely to “grow up” than those which are over-protected. However, our 29 conclusions on this point are only suggestive, and we leave it to future research to elaborate further on the costs and benefits of these various policy measures. In concrete policy terms, developing countries have a number of means at their disposal by which to pursue export diversification through lower entry barriers and trade costs. In terms of trade costs, a starting point might be regulations governing customs procedures and documentation: Table 7 suggests that export diversification is relatively sensitive to changes in these costs. In broad terms, this can be seen as an additional incentive to make progress on trade facilitation, i.e. the set of policies designed to reduce such costs. Similarly, a number of potential policy measures are available for lowering domestic barriers to entry: reducing the number of procedures required for company registration, applying a fixed registration fee rather than a percentage of capital, and removing the need for pre-tax payments, are examples. We can see many avenues for future research in this area. On the one hand, it will be important to replicate these results using other data, and perhaps taking account of other potential influences on diversification. In light of the cost-based approach we have adopted here, we consider it likely that financial sector development and FDI might turn out to be important factors in promoting diversification (Harding and Javorcik, 2007). Second, our reasoning throughout this paper has been in partial equilibrium. Historically, however, the general equilibrium aspect of diversification has also been important on a policy level. The Melitz (2003) framework has been extended in a general equilibrium sense by, for example, Bernard et al. (2007). However, it remains for future research to take the model’s predictions to the data in the diversification context, as we have done here. Finally, an important caveat to our research is that due to data limitations, we have not been able to analyze the cost-diversification link in the time dimension. We are aware that diversifying a 30 country’s export base is unlikely to be an instantaneous process, and would not want to be understood as implying such through our use of a cross-country regression in this paper. As more data on trade costs and barriers to entry become available—Doing Business is updated annually—we hope that researchers will return to the questions we have raised, and investigate in more detail the dynamics of the processes underlying them. The flipside of this is that our results will need to be supplemented with detailed work, including cost-benefit analysis, on particular mechanisms that governments use to promote diversification over time. Given the recent revival of interest in export diversification, we anticipate that future research in this area has the potential to be particularly fruitful in policy terms. 31 References Amin Gutiérrez de Piñeres, Sheila, and Michael Ferrantino, 1997, “Export Diversification and Structural Dynamics in the Growth Process: The Case of Chile”, Journal of Development Economics, 52(2), 375-391. Amurgo Pacheco, Alberto, 2006, “Preferential Trade Liberalization and the Range of Exported Products: The Case of the Euro-Mediterranean FTA”, Working Paper No. 18/2006, Graduate Institute of International Studies (Geneva). Baldwin, Richard E., 2000, “Regulatory Protectionism, Developing Nations, and a Two-Tier World Trading System” in Susan Collins and Dani Rodrik (eds.) Brookings Trade Forum, Washington, D.C.: The Brookings Institution. Baldwin, Richard E., 2005, “Heterogeneous Firms and Trade: Testable and Untestable Properties of the Melitz Model”, Working Paper No. 11471, NBER. Baldwin, Richard E., and Virginia Di Nino, 2006, “Euros and Zeros: The Common Currency Effect on Trade in New Goods”, Working Paper No. 12673, NBER. Baldwin, Richard E., and Rikard Forslid, 2006, “Trade Liberalization with Heterogeneous Firms”, Working Paper No. 12912, NBER. Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott, Forthcoming, “Firms in International Trade”, Journal of Economic Perspectives. Bernard, Andrew B., Stephen J. Redding, and Peter K. Schott, 2007, “Comparative Advantage and Heterogeneous Firms”, Review of Economic Studies, 74, 31-66. 32 Brainard, William C., and Richard N. Cooper, 1968, “Uncertainty and Diversification in International Trade” in Studies in Agricultural Economics, Trade, and Development, Stanford: Stanford University. Broda, Christian, and David E. Weinstein, 2006, “Globalization and the Gains from Variety”, The Quarterly Journal of Economics, 121(2), 541-585. Cameron, A. Colin, and Pravin K. Trivedi, 2001, “Essentials of Count Data Regression” in Badi H. Baltagi (ed.), A Companion to Theoretical Econometrics, Malden, Ma.: Blackwell. Chandra, V., J. Boccardo, and I. Osorio, 2007, “Export Diversification and Competitiveness in Developing Countries”, Working Paper. Chaney, Thomas, 2006, “Distorted Gravity: Heterogeneous Firms, Market Structure and the Geography of International Trade”, Working Paper, University of Chicago. Cox, Nicholas J., 2006, “MLOWESS: Stata Module for Lowess Smoothing with Multiple Predictors”, http://fmwww.bc.edu/repec/bocode/m/mlowess.ado. De Rosa, Dean A., 1992, “Increasing Export Diversification in Commodity Exporting Countries: A Theoretical Analysis”, International Monetary Fund Staff Papers, 39(3), 572-595. Dixit, Avinash K., and Joseph E. Stiglitz, 1977, “Monopolistic Competition and Optimum Product Diversity”, The American Economic Review, 67(3), 297-308. Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes and Andrei Shleifer, 2002, “The Regulation of Entry”, The Quarterly Journal of Economics, 117(1), 1-37. Eaton, Jonathan, and Samuel Kortum, 2002, “Technology, Geography and Trade”, Econometrica, 70(5), 1741-1779. 33 Feenstra, Robert C., 1994, “New Product Varieties and the Measurement of International Prices”, The American Economic Review, 84(1), 157-177. Feenstra, Robert C., and Hiau Looi Kee, 2006, “Export Variety and Country Productivity: Estimating the Monopolistic Competition Model with Endogenous Productivity”, Mimeo. Fontagné, Lionel, Michael Freudenberg, and Nicolas Péridy, 1998, “Intra-Industry Trade and the Single Market: Quality Matters”, Discussion Paper No. 1959, CEPR. Funke, Michael, and Ralf Ruhwedel, 2001a, “Product Variety and Economic Growth: Empirical Evidence for the OECD Countries”, IMF Staff Papers, 48(2), 225-242. Funke, Michael, and Ralf Ruhwedel, 2001b, “Export Variety and Export Performance”, Journal of Asian Economics, 12(4), 493-505. Gourieroux, C., A. Monfort, and A. Trognon, 1984, “Pseudo Maximum Likelihood Methods: Applications to Poisson Models”, Econometrica, 52(3), 701-720. Greene, William, 2004, “The Behavior of the Maximum Likelihood Estimator of Limited Dependent Variable Models in the Presence of Fixed Effects”, Journal of Econometrics, 7, 98119. Harding, Torfinn, and Beata Smarzynska Javorcik, 2007, “Note on the Effect of FDI on Export Diversification in Central and Eastern Europe”, Working Paper. Hausman, Jerry, Bronwyn H. Hall and Zvi Griliches, 1984, “Econometric Models for Count Data with an Application to the Patents-R&D Relationship”, Econometrica, 52(4), 909-938. Hausmann, Ricardo, Jason Hwang, and Dani Rodrik, 2006, “What You Export Matters”, Working Paper, http://ksghome.harvard.edu/~drodrik/hhr.pdf. 34 Helpman, Elhanan, Marc Melitz, and Yona Rubinstein, 2007, “Trading Partners and Trading Volumes”, Working Paper, http://post.economics.harvard.edu/faculty/helpman/papers/TradingPartners.pdf. Helpman, Elhanan, Marc J. Melitz, and Stephen R. Yeaple, 2004, “Export Versus FDI With Heterogeneous Firms”, The American Economic Review, 94(1), 300-316. Henry de Frahan, Bruno, and Mark Vancauteren, 2006, “Harmonization of Food Regulations and Trade in the Single Market: Evidence from Disaggregated Data”, European Review of Agricultural Economics, 33(3), 337-360. Hummels, David, and Peter J. Klenow, 2005, “The Variety and Quality of a Nation’s Exports”, The American Economic Review, 95(3), 704-723. Imbs, Jean and Romain Wacziarg, 2003, “Stages of Diversification”, The American Economic Review, 93(1), 63-86. Jansen, Marion, 2004, “Income Volatility in Small and Developing Economies: Export Concentration Matters”, Discussion Paper No. 3, World Trade Organization. Klinger, Bailey, and Daniel Lederman, 2004, “Discovery and Development: An Empirical Exploration of New Products”, Policy Research Working Paper No. 3450, The World Bank Klinger, Bailey, and Daniel Lederman, 2006, “Diversification, Innovation and Imitation Inside the Global Technological Frontier”, Policy Research Working Paper No. 3872, The World Bank Laborde, David, Houssein Boumelassa, and Maria C. Mitaritonna, Forthcoming, “A Consistent Picture of the Protection Across the World in 2004: MAcMap HS6 V2”, Working Paper, CEPII. Manchin, Miriam, 2006, “Preference Utilization and Tariff Reduction in EU Imports from ACP Countries”, The World Economy, 29(9), 1243-1266. 35 Mayer, Thierry, and Soledad Zignago, 2006, “Notes on CEPII’s Distance Measures”, Working Paper, CEPII. Melitz, Marc J., 2003, “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity”, Econometrica, 71(6), 1695-1725. Moulton, Brent R., 1990, “An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Units”, Review of Economics and Statistics, 72(2), 334-338. Prebisch, Raúl, 1959, “Commercial Policy in the Underdeveloped Countries”, The American Economic Review, 49(2), 251-273. Santos Silva, J.M.C. and Silvana Tenreyro, 2006, “The Log of Gravity”, Review of Economics and Statistics, 88(4), 641-658. Simcoe, Timothy, 2007, “XTPQML: Stata Module to Estimate Fixed-Effects Poisson Quasi-ML Regression with Robust Standard Errors”, http://fmwww.bc.edu/repec/bocode/x/xtpqml.ado. Wilson, John S., Catherine L. Mann, and Tsunehiro Otsuki, 2005, “Assessing the Benefits of Trade Facilitation: A Global Perspective”, The World Economy, 28(6), 841-871. Winkelmann, Rainer, 2000, Econometric Analysis of Count Data, 3rd edition, Berlin: Springer. Wooldridge, Jeffrey M., 1997, “Quasi-Likelihood Methods for Count Data” in M.H. Pesaran and P. Schmidt (eds.), Handbook of Applied Econometrics, Vol. 2, Oxford: Blackwell. Wooldridge, Jeffrey M., 1999, “Distribution-Free Estimation of Some Nonlinear Panel Data Models”, Journal of Econometrics, 90, 77-97. 36 Tables Table 1: Data description and sources. Variable Description Agriculture % GDP Customs Costs Distance Document Costs Entry Capital Entry Cost Units Size of the agricultural sector in the exporting country. Proxied by agriculture value added as a percent of GDP. Official customs clearance and technical control fees levied on a 20 foot container leaving the exporting country. Distance of the exporting country from Europe. Proxied by the average of the great circle distances between the main cities of the exporting country and Germany, weighted by population shares. Official document preparation costs in respect of a 20 foot container leaving the exporting country. Paid-in minimum capital requirement in the exporting country, i.e. amount an entrepreneur must deposit with a bank prior to company registration. Converted from percent GNI per capita to USD. Official cost of starting up and formally operating an industrial or commercial business in the exporting country. Converted from percent GNI per capita to USD. Year 2005 or most recent Source World Development Indicators USD 2006 Doing Business Km NA Mayer and Zignago (2006) USD 2006 Doing Business USD 2006 Doing Business USD 2006 Doing Business % Exchange Rate USD exchange rate for the exporting country. Proxied by the real effective exchange rate index. Index (2000 = 100) 2005 or most recent World Development Indicators Export Cost Official fees levied on a 20 foot container leaving the exporting country. Includes document preparation costs, administrative fees for customs clearance and technical control, terminal handling charges, and inland transit. Sum of customs costs, document costs, inland transport costs, and port costs (below). USD 2006 Doing Business GDP Gross domestic product of the exporting country. Constant 2000 USD GDP Deflator Inflation in the exporting country. Proxied by the GDP deflator. % GDP/capita Per capita gross domestic product of the exporting country. Constant 2000 USD 2005 or most recent 2005 or most recent 2005 or most recent World Development Indicators World Development Indicators World Development Indicators 37 Variable Description Year Source NA 2005 Eurostat NA 2005 Eurostat HH_Index pi qi Herfindahl-Hirschman index for exporter e = J i 1 p j q j j 1 HH_Index_CN2 Js pi qi Herfindahl-Hirschman index for exporter e in sector s = J s i 1 p jq j j 1 Inland Transport Costs Official inland transit costs for a 20 foot container leaving the exporting country. USD 2006 Doing Business Interest Rate Interest rate in the exporting country. Proxied by the real interest rate. % 2005 or most recent World Development Indicators NA 2005 Eurostat NA 2005 Eurostat NA 2005 Eurostat NA 2005 Eurostat NA 2005 Eurostat J Units 2 p Lambda Feenstra’s relative diversification measures for exporter e = w i 2 q iw iJ e p iw qiw , where Jw is the iJ w set of products exported by the world and Je is the set of products exported by country e. p Feenstra’s relative diversification measures for exporter e in sector s = Lambda_CN2 iJ s e w i q iw p iw q iw , where iJ s w Lines Lines_CN2 Lines_CN2_100k Jw is the set of products exported by the world and Je is the set of products exported by country e (both in sector s). Count of the total number of product lines in which a country has strictly positive exports to the EU. Count of the number of product lines in a 2-digit sector for which a country has strictly positive exports to the EU. Count of the number of product lines in a 2-digit sector for which a country has exports to the EU of a value greater than €100,000. 38 Variable Lines_CN2_1m Lines_CN2_Man Manufacturing % GDP Description Count of the number of product lines in a 2-digit sector for which a country has exports to the EU of a value greater than €1,000,000. Count of the number of product lines in a 2-digit sector for which a country has strictly positive exports to the EU. Limited to HS sectors 25-97 only (manufactured goods). Size of the manufacturing sector in the exporting country. Proxied by manufacturing value added as a percent of GDP. Units Year Source NA 2005 Eurostat NA 2005 Eurostat 2005 or most recent 2006 World Development Indicators Doing Business 2005 WITS-TRAINS % Official terminal handling charges for a 20 foot container leaving the exporting country. USD Simple average applied tariff by HS2 sector, including ad valorem equivalents of specific % ad Tariffs tariffs. valorem a) Log transformations use ln(1+…) for series containing zero values. b) Negative observations on the GDP deflator and the interest rate are coded as zero prior to taking the log transform. Port Costs Table 2: Country group definitions. Country Members Group Australia, Austria, Belgium, Canada, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Japan, Korea, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovakia, Spain, Sweden, Switzerland, Turkey, United Kingdom, United OECD States Austria, Belgium, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, EU-25 Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, United Kingdom Antigua and Barbuda, Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong, China, Iceland, Ireland, High Income Israel, Italy, Japan, Korea, Kuwait, Netherlands, New Zealand, Norway, Portugal, Saudi Arabia, Singapore, Slovenia, Spain, Sweden, Switzerland, United Arab Emirates, United Kingdom, United States Argentina, Belize, Botswana, Chile, Costa Rica, Croatia, Czech Republic, Dominica, Equatorial Guinea, Estonia, Gabon, Grenada, Hungary, Upper Latvia, Lebanon, Lithuania, Malaysia, Mauritius, Mexico, Oman, Palau, Panama, Poland, Romania, Russia, Seychelles, Slovakia, South Africa, Middle St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Trinidad and Tobago, Turkey, Uruguay, Venezuela Income Albania, Algeria, Angola, Armenia, Azerbaijan, Belarus, Bolivia, Bosnia and Herzegovina, Brazil, Bulgaria, Cameroon, Cape Verde, China, Colombia, Congo, Rep., Djibouti, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Georgia, Guatemala, Guyana, Honduras, Indonesia, Lower Iran, Iraq, Jamaica, Jordan, Kazakhstan, Kiribati, Lesotho, Macedonia, FYR, Maldives, Marshall Islands, Micronesia, Moldova, Morocco, Middle Namibia, Nicaragua, Paraguay, Peru, Philippines, Samoa, Serbia, Sri Lanka, Suriname, Swaziland, Syria, Thailand, Tonga, Tunisia, Ukraine, Income Vanuatu Afghanistan, Bangladesh, Benin, Bhutan, Burkina Faso, Burundi, Cambodia, Central African Republic, Chad, Comoros, Côte d'Ivoire, Eritrea, Ethiopia, Gambia, Ghana, Guinea, Guinea-Bissau, Haiti, India, Kenya, Kyrgyz Republic, Lao PDR, Madagascar, Malawi, Mali, Mauritania, Low Income Mongolia, Mozambique, Nepal, Niger, Nigeria, Pakistan, Papua New Guinea, Rwanda, Senegal, Sierra Leone, Solomon Islands, Sudan, São Tomé and Principe, Tajikistan, Tanzania, Togo, Uganda, Uzbekistan, Vietnam, Yemen, Zambia, Zimbabwe 39 Table 3: Export diversification as measured by counts of the number of 8-digit products exported to the EU in 2005 (lines). Rank Country Name Lines Rank Country Name Lines China 8053 69 Jamaica 423 1 India 6449 70 Georgia 379 2 Brazil 5477 Suriname 379 3 71 Romania 5148 72 Uganda 377 4 Thailand 5073 73 Gabon 369 5 Hong Kong, China 5062 74 Azerbaijan 364 6 Israel 4830 75 Mongolia 350 7 South Africa 4806 76 Paraguay 345 8 Russia 4451 77 Uzbekistan 316 9 Bulgaria 4297 78 Cape Verde 315 10 Malaysia 4071 Congo, Rep. 314 11 79 Indonesia 4059 80 Yemen 308 12 Singapore 3970 81 Lao PDR 296 13 Croatia 3807 82 Mali 281 14 United Arab Emirates 3454 83 Armenia 279 15 Morocco 3396 84 Mauritania 258 16 Tunisia 3297 85 Burkina Faso 257 17 Argentina 3158 86 Togo 249 18 Ukraine 3077 Mozambique 248 19 87 Egypt 3000 88 Guinea 246 20 Philippines 2858 89 Sudan 223 21 Vietnam 2806 90 Benin 217 22 Pakistan 2612 91 Zambia 212 23 Chile 2032 92 Botswana 210 24 Saudi Arabia 1993 93 Sierra Leone 206 25 Iran 1935 94 Afghanistan 200 26 Sri Lanka 1876 Papua New Guinea 200 27 95 Bosnia and Herzegovina 1815 96 Fiji 194 28 Lebanon 1682 97 Seychelles 192 29 Colombia 1680 98 Nicaragua 188 30 Peru 1666 99 Antigua and Barbuda 184 31 Macedonia, FYR 1528 100 Swaziland 164 32 Syria 1491 101 Kyrgyz Republic 160 33 Belarus 1439 102 Niger 158 34 Mauritius 1262 Iraq 154 35 103 Bangladesh 1225 104 Equatorial Guinea 153 36 Nigeria 1155 105 Malawi 152 37 Kenya 1118 106 Guyana 136 38 Albania 1087 107 Gambia 132 39 Ecuador 1052 108 Maldives 131 40 Jordan 1050 109 Belize 118 41 Algeria 1028 110 Rwanda 118 42 Venezuela 1018 Haiti 115 43 111 Uruguay 970 St. Lucia 112 44 112 40 Rank Country Name Lines Rank Country Name Lines Moldova 937 Chad 98 45 113 Oman 905 Dominica 96 46 114 Dominican Republic 895 Tajikistan 93 47 115 Kuwait 894 Eritrea 88 48 116 Ghana 876 Djibouti 86 49 117 Madagascar 858 Marshall Islands 83 50 118 Costa Rica 843 St. Kitts and Nevis 81 51 119 Nepal 813 Central African Republic 71 52 120 Côte d'Ivoire 764 St. Vincent and the Grenadines 71 53 121 Senegal 746 São Tomé and Principe 66 54 122 Panama 685 Burundi 54 55 123 Kazakhstan 678 Lesotho 51 56 124 Cameroon 642 Vanuatu 49 57 125 Guatemala 611 Grenada 45 58 126 Tanzania 552 Guinea-Bissau 44 59 127 Bolivia 505 Kiribati 43 60 128 Zimbabwe 475 Bhutan 40 61 129 Angola 472 Samoa 38 62 130 Honduras 456 Tonga 36 63 131 Trinidad and Tobago 446 Solomon Islands 35 64 132 El Salvador 440 Comoros 27 66 133 Namibia 440 Micronesia 12 65 134 Cambodia 439 Palau 9 67 135 Ethiopia 432 68 c) Sample is limited to non-OECD and non-EU25 countries. d) Calculation is based on 10,753 distinct products identified in the CN8 classification. e) Countries ranked by lines in decreasing order of export diversification (higher lines indicates greater diversification). 41 Table 4: Export diversification as measured by Feenstra’s lambda. Rank Country Name Lambda Rank Country Name China 0.84 Iraq 1 69 Brazil 0.82 Uzbekistan 2 70 Russia 0.82 Honduras 3 71 India 0.81 El Salvador 4 72 South Africa 0.72 Namibia 5 73 Romania 0.71 Uganda 6 74 Israel 0.68 Nepal 7 75 Thailand 0.67 Armenia 8 76 United Arab Emirates 0.66 Equatorial Guinea 9 77 Ukraine 0.66 Bolivia 10 78 Singapore 0.66 Zimbabwe 11 79 Hong Kong, China 0.64 Cape Verde 12 80 Malaysia 0.64 Zambia 13 81 Indonesia 0.64 Mozambique 14 82 Bulgaria 0.63 Mali 15 83 Croatia 0.62 Paraguay 16 84 Tunisia 0.60 Benin 17 85 Argentina 0.59 Sierra Leone 18 86 Egypt 0.58 Antigua and Barbuda 19 87 Saudi Arabia 0.57 Chad 20 88 Morocco 0.54 Togo 21 89 Iran 0.53 Suriname 22 90 Philippines 0.46 Seychelles 23 91 Algeria 0.44 Afghanistan 24 92 Bosnia and Herzegovina 0.43 Cambodia 25 93 Vietnam 0.43 Botswana 26 94 Chile 0.43 Mongolia 27 95 Pakistan 0.42 Mauritania 28 96 Kuwait 0.41 Sudan 29 97 Venezuela 0.41 Lao PDR 30 98 Colombia 0.40 Guinea 31 99 Belarus 0.39 Burkina Faso 32 100 Lebanon 0.39 Maldives 33 101 Macedonia, FYR 0.36 Dominica 34 102 Nigeria 0.36 Nicaragua 35 103 Albania 0.36 Malawi 36 104 Syria 0.36 Papua New Guinea 37 105 Jordan 0.34 Rwanda 38 106 Côte d'Ivoire 0.34 Guyana 39 107 Oman 0.33 Kyrgyz Republic 40 108 Peru 0.32 Fiji 41 109 Sri Lanka 0.32 Tajikistan 42 110 Mauritius 0.31 St. Vincent and the Grenadines 43 111 Kazakhstan 0.31 Belize 44 112 Kenya 0.30 Haiti 45 113 42 Lambda 0.16 0.15 0.15 0.15 0.15 0.15 0.14 0.14 0.14 0.14 0.14 0.13 0.13 0.13 0.12 0.12 0.12 0.12 0.11 0.11 0.11 0.11 0.11 0.11 0.11 0.10 0.10 0.10 0.10 0.10 0.10 0.09 0.08 0.08 0.08 0.08 0.08 0.07 0.07 0.07 0.07 0.07 0.07 0.06 0.06 Rank Country Name Lambda Rank Country Name Lambda Cameroon 0.29 Gambia 0.06 46 114 Panama 0.29 St. Lucia 0.05 47 115 Moldova 0.28 Djibouti 0.05 48 116 Uruguay 0.27 Swaziland 0.05 49 117 Gabon 0.27 Central African Republic 0.05 50 118 Costa Rica 0.26 Eritrea 0.05 51 119 Guatemala 0.26 Burundi 0.04 52 120 Bangladesh 0.26 Marshall Islands 0.04 53 121 Ecuador 0.26 Vanuatu 0.04 54 122 Angola 0.25 Grenada 0.04 55 123 Senegal 0.25 São Tomé and Principe 0.04 56 124 Ghana 0.24 St. Kitts and Nevis 0.03 57 125 Dominican Republic 0.24 Lesotho 0.03 58 126 Azerbaijan 0.23 Kiribati 0.03 59 127 Trinidad and Tobago 0.23 Comoros 0.03 60 128 Madagascar 0.21 Bhutan 0.02 61 129 Jamaica 0.19 Samoa 0.02 62 130 Tanzania 0.19 Tonga 0.02 63 131 Congo, Rep. 0.19 Guinea-Bissau 0.01 64 132 Georgia 0.19 Solomon Islands 0.01 65 133 Yemen 0.18 Micronesia 0.01 66 134 Niger 0.17 Palau 0.01 67 135 Ethiopia 0.16 68 a) Sample is limited to non-OECD and non-EU25 countries. b) Lambda ranges between 0 and 1. c) Countries ranked by lambda in decreasing order of export diversification (higher lambda indicates greater diversification). 43 Table 5: Export diversification as measured by the Herfindahl-Hirschman index. Rank Name HH_Index Rank Name China 0.01 Uganda 1 69 Croatia 0.01 Yemen 2 70 Romania 0.01 Panama 3 71 India 0.01 Senegal 4 72 Hong Kong, China 0.01 Algeria 5 73 Indonesia 0.01 Namibia 6 74 Bulgaria 0.01 Honduras 7 75 Pakistan 0.01 Costa Rica 8 76 Thailand 0.01 Mongolia 9 77 Vietnam 0.02 Belarus 10 78 Tunisia 0.02 Micronesia 11 79 Sri Lanka 0.02 Cameroon 12 80 Lebanon 0.02 Ghana 13 81 Bosnia and Herzegovina 0.02 El Salvador 14 82 Ukraine 0.02 Dominica 15 83 Macedonia, FYR 0.03 Congo, Rep. 16 84 Israel 0.03 Afghanistan 17 85 Malaysia 0.03 Samoa 18 86 Brazil 0.03 Armenia 19 87 South Africa 0.04 Russia 20 88 Singapore 0.04 Palau 21 89 Albania 0.04 Guyana 22 90 Uruguay 0.04 Central African Republic 23 91 Philippines 0.04 Suriname 24 92 Moldova 0.05 São Tomé and Principe 25 93 Egypt 0.05 Sudan 26 94 Lao PDR 0.05 Comoros 27 95 Peru 0.06 Maldives 28 96 Madagascar 0.06 Belize 29 97 Kyrgyz Republic 0.06 Oman 30 98 Morocco 0.07 Grenada 31 99 Bangladesh 0.07 Nicaragua 32 100 Zimbabwe 0.07 Mali 33 101 Kenya 0.08 Ethiopia 34 102 Tanzania 0.09 Jamaica 35 103 Burkina Faso 0.09 Togo 36 104 Benin 0.10 Chad 37 105 Tonga 0.10 Marshall Islands 38 106 Mauritius 0.10 Guinea 39 107 Dominican Republic 0.10 Swaziland 40 108 Argentina 0.11 Mauritania 41 109 Georgia 0.11 Solomon Islands 42 110 Eritrea 0.11 Kuwait 43 111 Zambia 0.12 St. Lucia 44 112 Djibouti 0.12 Kiribati 45 113 44 HH_Index 0.21 0.21 0.21 0.21 0.23 0.23 0.23 0.23 0.23 0.24 0.25 0.25 0.25 0.26 0.26 0.26 0.27 0.27 0.28 0.28 0.28 0.28 0.28 0.29 0.30 0.30 0.30 0.30 0.31 0.31 0.33 0.34 0.34 0.35 0.37 0.40 0.44 0.45 0.47 0.48 0.48 0.49 0.49 0.51 0.51 Gambia 0.12 St. Kitts and Nevis 0.52 46 114 Malawi 0.13 Paraguay 0.54 47 115 Trinidad and Tobago 0.13 Saudi Arabia 0.55 48 116 Uzbekistan 0.13 Vanuatu 0.56 49 117 Colombia 0.13 Antigua and Barbuda 0.61 50 118 Bolivia 0.13 Nigeria 0.61 51 119 Gabon 0.13 Equatorial Guinea 0.62 52 120 Chile 0.14 Rwanda 0.67 53 121 Cambodia 0.14 Mozambique 0.68 54 122 Jordan 0.14 Kazakhstan 0.68 55 123 Guinea-Bissau 0.14 Angola 0.68 56 124 Côte d'Ivoire 0.14 Syria 0.70 57 125 Papua New Guinea 0.15 Sierra Leone 0.74 58 126 Tajikistan 0.15 Iran 0.75 59 127 Guatemala 0.15 St. Vincent and the Grenadines 0.75 60 128 Cape Verde 0.15 Azerbaijan 0.76 61 129 Bhutan 0.16 Niger 0.77 62 130 Haiti 0.17 Fiji 0.79 63 131 Ecuador 0.19 Burundi 0.93 64 132 Nepal 0.19 Botswana 0.94 65 133 Venezuela 0.20 Lesotho 0.94 66 134 United Arab Emirates 0.20 Iraq 0.99 67 135 Seychelles 0.21 68 a) Sample is limited to non-OECD and non-EU25 countries. b) HH_Index ranges between 0 and 1. c) Countries ranked by hh_index in decreasing order of export diversification (lower hh_index indicates greater diversification). 45 Table 6: Estimation results. Model 1 0.002 Entry Capital 0.003 -0.097*** Entry Cost 0.010 -0.250*** Export Cost 0.017 -0.463*** Distance 0.031 Model 2 0.019*** 0.002 -0.259*** 0.016 -0.895*** 0.020 Tariffs GDP GDP/cap. Mftg. % GDP Ag. % GDP GDP Defl. Model 3 -0.029*** 0.002 -0.168*** 0.010 -0.921*** 0.035 -0.413*** 0.039 0.321*** 0.027 0.515*** 0.016 0.032* 0.018 0.330*** 0.025 0.007 0.015 -0.256*** 0.017 Ex. Rate Int. Rate 0.019** 0.008 10088 97 Model 5 -0.023*** 0.003 -0.078*** 0.011 -0.356*** 0.022 -0.472*** 0.034 -0.086*** 0.019 0.453*** 0.013 0.025 0.018 0.331*** 0.024 -0.054*** 0.014 Model 6 -0.024*** 0.005 0.121*** 0.019 -0.146*** 0.026 -0.545*** 0.036 -0.041* 0.024 0.494*** 0.015 -0.257*** 0.033 0.566*** 0.066 -0.061*** 0.016 -0.519*** 0.029 0.731*** 0.071 -0.084*** 0.021 2997 96 31.2 (Unbal. ave.) 4721.38*** 8286 8210 7849 96 96 96 86.3 85.5 81.8 No. Countries 104 (Balanced) 132 (Balanced) (Unbal. ave.) (Unbal. ave.) (Unbal. ave.) 3437.19*** 2459.90*** 1280.66*** 2531.63*** 3471.62*** Wald Chi2 a) Dependent variable is lines_cn2. Independent variables are in logarithms. b) Estimation is by Poisson QML with conditional fixed effects by 2-digit sector. Robust standard errors are in italics under the parameter estimates. c) Sample is limited to non-OECD and non-EU25 countries. Obs. No. Sectors 12804 97 Model 4 -0.029*** 0.002 -0.089*** 0.009 -0.285*** 0.016 -0.455*** 0.028 -0.062*** 0.019 0.453*** 0.013 0.067*** 0.021 46 Table 7: Estimation results using broken down export costs. Model 7 -0.001 Entry Capital 0.003 -0.072*** Entry Cost 0.009 -0.050*** Customs Costs 0.005 -0.010** Document Costs 0.005 -0.001 Inland Transport Costs 0.006 0.039*** Port Costs 0.004 -0.451*** Distance 0.031 0.531*** GDP 0.015 0.042** GDP Per Capita 0.018 0.369*** Manufacturing % GDP 0.025 -0.008 Agriculture % GDP 0.015 -0.238*** GDP Deflator 0.019 0.025*** Interest Rate 0.008 10088 Observations 97 No. of Sectors 104 (Balanced) No. of Countries 4704.43*** Wald Chi2 a) Dependent variable is lines_cn2. Independent variables are in logarithms. b) Estimation is by Poisson QML with conditional fixed effects by 2-digit sector. Robust standard errors are in italics under the parameter estimates. c) Sample is limited to non-OECD and non-EU25 countries. 47 Table 8: Robustness checks—estimation methodology. OLS Tobit Neg. Bin. Poisson (Aggregate) -0.009*** -0.008** -0.003*** -0.002 Entry Capital 0.003 0.004 0.003 0.015 -0.069*** -0.098*** -0.098*** -0.086 Entry Cost 0.010 0.014 0.011 0.055 -0.136*** -0.192*** -0.183*** -0.262*** Export Cost 0.014 0.022 0.020 0.091 -0.314*** -0.445*** -0.409*** -0.469*** Distance 0.016 0.019 0.016 0.076 0.388*** 0.520*** 0.507*** 0.515*** GDP 0.005 0.007 0.005 0.023 0.048*** 0.067*** 0.044*** 0.032 GDP Per Capita 0.012 0.017 0.014 0.060 0.161*** 0.248*** 0.291*** 0.305*** Manufacturing % GDP 0.013 0.019 0.016 0.094 0.041*** 0.051*** 0.046*** 0.010 Agriculture % GDP 0.012 0.016 0.012 0.068 -0.223*** -0.188*** -0.272*** -0.197*** GDP Deflator 0.012 0.018 0.015 0.061 -0.019** 0.033*** 0.021** 0.004 Interest Rate 0.010 0.013 0.010 0.039 -3.638 -5.454*** -6.748*** 0.506 Constant 0.239 0.337 0.276 1.154 10088 10088 10088 104 Observations 97 97 97 NA No. of Sectors 104 (Balanced) 104 (Balanced) 104 (Balanced) 104 No. of Countries 1457.81*** 21601.40*** 13598.69*** 1140.68*** Wald Chi2/F a) Dependent variable is log(1+lines_cn2) in columns 1 and 2, lines_cn2 in column 3, and lines in column 4. b) Independent variables are in logarithms in all models. c) Estimation (column 1) is by OLS with fixed effects by 2-digit sector. Robust standard errors are in italics under the parameter estimates. d) Estimation (column 2) is by Tobit with fixed effects by 2-digit sector. Robust standard errors are in italics under the parameter estimates. e) Estimation (column 3) is by negative binomial ML with conditional fixed effects by 2-digit sector. Non-robust standard errors are in italics under the parameter estimates. f) Estimation (column 3) is by Poisson QML using aggregate country-level data. Robust standard errors are in italics under the parameter estimates. g) All samples are limited to non-OECD and non-EU25 countries. 48 Table 9: Robustness checks—country sample. Low + Low + Low Income Middle Income Lower Middle Income 0.002 0.011*** -0.005 Entry Capital 0.002 0.004 0.011 -0.148*** -0.171*** -0.293*** Entry Cost 0.012 0.012 0.042 -0.240*** -0.234*** -0.302*** Export Cost 0.015 0.026 0.052 -0.442*** -0.383*** -0.379** Distance 0.030 0.038 0.164 0.500*** 0.505*** 0.506*** GDP 0.016 0.017 0.023 -0.016 -0.002 0.240*** GDP Per Capita 0.015 0.017 0.087 0.450*** 0.384*** 0.413*** Manufacturing % GDP 0.048 0.051 0.090 -0.077** -0.073 0.141** Agriculture % GDP 0.032 0.048 0.063 -0.284*** -0.228*** -0.074** GDP Deflator 0.016 0.018 0.034 0.017* -0.016 -0.080** Interest Rate 0.010 0.013 0.034 9603 7372 3007 Observations 97 97 97 No. of Sectors 99 (Balanced) 76 (Balanced) 31 (Balanced) No. of Countries 2836.71*** 2666.85*** 1077.42*** Wald Chi2 a) Dependent variable is lines_cn2. Independent variables are in logarithms. b) Estimation is by Poisson QML with conditional fixed effects by 2-digit sector. Robust standard errors are in italics under the parameter estimates. c) The sample in column 1 is limited to low and middle income countries (non-OECD and non-EU25). In column 2, it is limited to low and lower middle income countries. In column 3, low income countries only are included. 49 Table 10: Robustness checks—diversification measure. >$100k >$1m -0.007* -0.013* Entry Capital 0.004 0.007 -0.141*** -0.160*** Entry Cost 0.019 0.034 -0.397*** -0.422*** Export Cost 0.063 0.102 -0.621*** -0.636*** Distance 0.060 0.083 0.697*** 0.750*** GDP 0.031 0.035 -0.034 -0.075 GDP Per Capita 0.045 0.054 0.492*** 0.545*** Manufacturing % GDP 0.029 0.041 0.001 -0.009 Agriculture % GDP 0.023 0.030 -0.391*** -0.396*** GDP Deflator 0.039 0.082 0.023 0.003 Interest Rate 0.022 0.038 Manufactures 0.004 0.003 -0.091*** 0.011 -0.263*** 0.019 -0.501*** 0.032 0.526*** 0.019 0.016 0.020 0.338*** 0.029 -0.022** 0.010 -0.248*** 0.018 0.024*** 0.009 Constant Observations No. of Sectors 10088 97 10088 97 7592 73 No. of Countries 104 (Balanced) 104 (Balanced) 104 (Balanced) HH Index 0.005** 0.002 0.032*** 0.008 0.047*** 0.013 0.153*** 0.012 -0.185*** 0.004 0.018* 0.011 -0.098*** 0.012 0.020* 0.011 0.110*** 0.011 -0.011 0.008 1.567*** 0.206 6675 97 68.8 (Unbal. ave.) 340.60*** Lambda -0.001* 0.001 -0.012*** 0.002 -0.006* 0.004 -0.046*** 0.003 0.063*** 0.001 0.004 0.003 0.030*** 0.003 -0.002 0.003 -0.034*** 0.003 0.003 0.002 -0.744*** 0.056 6695 97 69 (Unbal. ave.) 546.86*** 1925.73*** 1113.07*** 3087.39*** Wald Chi2 a) Independent variables are in logarithms. b) Dependent variable in columns 1-3 is lines_cn2, adjusted to include only exports greater than €100,000 (column 1), exports greater than €1 million (column 2), or exports in HS chapters 25-97 only (column 3). c) Dependent variables in columns 4-5 are the Herfindahl-Hirschman index, and Feenstra’s lambda respectively. d) Estimation for columns 1-3 is by Poisson QML with conditional fixed effects by 2-digit sector. Columns 4-5 are estimated by OLS with fixed effects by 2-digit sector. In all cases, robust standard errors are in italics under the parameter estimates. e) Sample is limited to non-OECD and non-EU25 countries. 50 Figures Figure 1: Nonparametric regression results for entry costs. 0 2000 lines 4000 6000 8000 Lowess smoother 0 1 2 3 4 5 ent_cost bandwidth = .8 a) Dependent variable is lines. Independent variable is entry cost. b) Sample limited to non-OECD and non-EU25. c) Observations (3) with ent_cost > 500% of GNI per capita excluded as outliers. Figure 2: Nonparametric regression results for export costs. 0 2000 lines 4000 6000 8000 Lowess smoother 0 1000 2000 exp_cost 3000 4000 bandwidth = .8 a) Dependent variable is lines. Independent variable is export cost. b) Sample limited to non-OECD and non-EU25. c) Observations (3) with ent_cost > 500% of GNI per capita excluded as outliers. 51 6000 4000 lines 2000 0 -2000 -2000 0 2000 lines 4000 6000 8000 Figure 3: Nonparametric regression results for entry costs and export costs together. 0 1 2 3 ent_cost 4 5 0 1000 2000 3000 exp_cos t 4000 a) Sample limited to non-OECD and non-EU25. b) Three observations with ent_cost > 500% of GNI per capita excluded as outliers. c) Calculations performed in Stata using the MLOWESS module (Cox, 2006). 0 .05 .1 .15 Figure 4: Kernel densities of lines_cn2 and predicted values from Model 1 (Table 3). 0 200 400 x kdensity lines_cn2 600 800 kdensity lines_cn2_hat 52
© Copyright 2026 Paperzz