TRADE FACILITATION AND THE EXTENSIVE MARGIN OF EXPORTS

Transcription

1 bs_bs_banner Vol. 65, No. 2, June 2014 Te Journal of te Japanese Economic Association TRADE FACILITATION AND THE EXTENSIVE MARGIN OF EXPORTS By ROBERT C. FEENSTRA and HONG MA doi: /jere University of California Davis and National Bureau of Economic Researc Tsingua University Tis paper examines te link between trade facilitation and export variety for a broad cross-section of countries. We measure trade facilitation using port efficiency. We also include te bilateral import tariff and OECD membersip and regional trade agreements. We find tat port efficiency contributes significantly to te extensive margin of exports, and tat te bilateral import tariff negatively impacts te variety of exports. Te positive effect is confirmed wen examining trade between countries witout common land borders, or between OECD member countries and non-oecd countries. Results are not as strong wen we look at witin-oecd trade, or focus on bilateral trade in te intensive margin. JEL Classification Numbers: F13, F Introduction Recent literature in international trade as empasized te extensive margin, by wic we mean te variety tat a country exports and imports. Wen a country imports more varieties, its consumers gain, as as been sown empirically for te USA by Broda and Weinstein (2006). However, as a country exports more product varieties, its producers also gain, as sown by Feenstra and Kee (2008). Tis gain arises due to improved productivity in a model wit eterogeneous firms (see Melitz, 2003). Wen a country faces improved market opportunities abroad, te ig-productivity firms will begin to export. Te entry of tose firms will drive up factor prices in te sector and force out lower-productivity domestic firms. Hence, improved market opportunities abroad are associated wit greater product variety of exports and iger overall productivity in a sector. Tis linkage between export variety and productivity raises te potential for welfareenancing government policies via trade facilitation. Witin tis broad category of policies we include actions tat allow for enanced exports, troug, for example, infrastructure development, foreign marketing opportunities and institutions. 1 Te logic of te Melitz model is tat suc actions tat facilitate trade will raise export variety and average productivity. Tere is a rapidly-growing body of literature discussing various actions tat promote or facilitate trade flow across countries. Recent researc by Limao and Venables (2000) sows tat a 10% decrease in transport cost raises trade volume by more tan 20%. Te reduction in transport costs could be due to improvement in infrastructure: for example, improvement in ocean port facilities. Clark et al. (2004) analyse te effect of port efficiency on bilateral trade flows and sow tat improving port efficiency does reduce sipping costs considerably. Finally, a series of World Bank working papers empirically explores te trade-facilitating impact of standards (Cen et al., 2008), road network quality 1 In te WTO, trade facilitation is defined as te simplification and armonization of international trade procedures covering te activities, practices and formalities involved in collecting, presenting, communicating and processing data required for te movement of goods in international trade. 158

2 R. C. Feenstra and H. Ma: Trade Facilitation and Exports (Seperd and Wilson, 2006), and oter factors suc as port efficiency, custom regimes, regulatory policy and tecnology (Wilson et al., 2005; Soloaga et al., 2006). Most current works focus on te effect of trade facilitation on trade flows, instead of on te extensive margin or export variety. One exception is Keoe and Rul (2004), wo sow tat trade liberalization suc as te Nort American Free Trade Agreement drives growt in te extensive margin, wic is an important source of new trade. A closely-related study considers te impact of a currency union. For example, Baldwin and Di Nino (2006) examine te effect of te euro on te extensive margin of trade among European countries. Bergin and Lin (2009) use te NBER-UN world trade data set to study te different effects of excange rate regimes on te extensive and intensive margins. Furtermore, researc empasizing te export diversification of developing countries naturally relates itself to studies of te export variety and its determinants. For example, Dennis and Seperd (2007) and Amurgo-Paceco and Pierola (2008) bot focus on patterns of trade diversification in developing countries. In sort, tere is increasing attention being paid to te cross-border trade in varieties, instead of to trade volume. In tis paper, we examine te link between trade facilitation and export variety for a broad cross-section of countries over 13 years ( ). We measure trade facilitation using te data on te efficiency of ports, from Blonigen and Wilson (2008). Te extensive margin of exports is constructed as in Hummels and Klenow (2005), but allowing bot cross-sectional and time-series variations. We also include oter variables tat can impact trade in our specification; notably, te trade restrictiveness index due to Kee et al. (2009), as well as institutional variables suc as regional trade agreements (RTA). Tis paper first adopts te Feenstra (1994) metod to develop a panel of bilateral export variety measures, as discussed in Section 2. Tis metod as been developed by Hummels and Klenow (2005) for a cross-section sample, were export variety is named te extensive margin of exports. Following Feenstra and Kee (2008), we define te export variety (or extensive margin of exports) of country to country j as te worldwide average export over all years to country j in tose categories were country actually exports to j, relative to te worldwide average export to j over all years in all categories. Using tis metod, our measure of export variety is consistent bot across countries and over time, as discussed in Section 3. We ten adopt a conventional empirical metod, te gravity regression, to estimate ow te export variety between two countries is influenced by, for example, ocean port efficiency, bilateral trade tariffs, international institutions, suc as te OECD, and regional trade agreements. It is important to single out export variety for empirical investigation and to understand te role and te determinants of te extensive margins of trade. Using a conventional gravity equation wit total trade flows as te dependent variable, altoug armed wit te micro-foundation provided in Anderson and van Wincoop (2003), migt be misleading because te extensive margin and te intensive margin migt correspond differently to trade costs and trade facilitation factors. As revealed in Bernard et al. (2007), te extensive margins of trade are central to understand te effect of trade costs on trade flows. In fact, following a metod proposed in Eaton et al. (2004), Bernard et al. (2007) decompose export flow into tree components: te number of firms exporting to a destination, te number of products exported to tat destination and average exports per product per firm. Separately regressing eac component on te usual gravity variables suc as distance and income, tey find tat it is te first two items (te extensive margin) tat explain te dampening effect of distance, wile te average export value (te intensive margin) is increasing in distance. Tis is in sarp contrast to te conventional belief tat 159

3 distance or trade cost reduces aggregate and average trade flow (assuming tat all firms export), and tis also motivates our investigation in te present paper into te different effects of port efficiency on te extensive and intensive margins. Finally, a few papers attempt to explain te teoretical reason wy te extensive margin and te intensive margin adjust differently wit respect to trade costs. One example is Bergin and Lin (2009). Tey construct a stocastic general equilibrium model, assuming sticky prices and fixed costs of entry, to explain te different effect of excange rate uncertainty on te extensive margin (firm number) and te intensive margin (average value). Section 4 summarizes our data and Section 5 presents te estimation. In our bencmark regression we find tat bilateral port efficiency as a significant and positive impact on export variety (i.e. te extensive margin). Te impact on export volume (i.e. te intensive margin) is positive, but not significant. A 10% improvement in te bilateral port efficiency increases export variety by 1.5 to 3.4%, wile it only increases te intensive margin of exports by 0.2 to 1.0%. Port efficiency appears to matter muc more for te extensive margin tan for te intensive margin. Similarly, for trade barriers, we look at te simple average of bilateral import tariffs. Not surprisingly, tariffs appear to discourage expansion in export variety, but tey ave an insignificant impact on te intensive margin. Furtermore, saring te same language seems to promote trade at bot te extensive and intensive margins. RTA and saring common land borders bot promote export volume, wile te former discourages export of variety. Interestingly, being an OECD member for importing countries is sown to promote te range of export variety tremendously, wile exporting countries OECD membersip does not ave a significant effect on export variety. To furter explore te impact of trade facilitation on export variety, we divide te sample into five subgroups; namely, trade among OECD countries, trade between member country and non-member country, and trade among non-oecd countries, and, finally, trade between countries witout common land borders. Te results are not as strong wen we use only te OECD countries, or focus on te intensive margin rater tan variety, but continue to old wen at least one trading nation is not an OECD member, or wen we exclude countries wit contiguous borders wit eac oter. Section 6 concludes. 2. Measuring new varieties in international trade Feenstra (1994) sows ow to construct an exact import price index tat accounts for bot newly-created import varieties over time and taste or quality canges in existing varieties. Tis exact price index can be derived from a non-symmetric constant elasticity of substitution (CES) utility function. Consider ome country importing from many countries, eac of wic exports many types of commodities. For simplicity, we aggregate tese goods into a single sector, but te extension to multiple sectors will be immediate. 2 For eac period t, let te set of goods consumed in country be denoted by I t {, 123,,.}. Ten te quantity vector of eac type of good consumed in country in period t is denoted by q t > 0. Te representative consumer s preferences are caracterized by a nonsymmetric CES utility function, wic is: 2 See, for example, Broda and Weinstein (2006), for an aggregate exact price index derived from a composite CES function incorporating many sectors and many countries. 160

4 R. C. Feenstra and H. Ma: Trade Facilitation and Exports ( σ 1)/ σ U = U( q, I ) = a ( q ) σ σ, σ > 1, (1) t t t it it i I t /( 1) were σ denotes te elasticity of substitution among all varieties, wic is assumed to exceed unity; a it > 0 denotes a parameter measuring taste (or quality) for good i, wic is allowed to vary over time; and I t denotes te set of goods available in period t, at prices p it > 0. By duality, te minimum unit-cost function is also a CES form: c( p, I ) = 1 σ b ( p ) σ, σ > 1, b a. (2) t t it it i It 1/( 1 σ ) Te CES unit-cost function specified above canges wit evolving variety set I t ; terefore, it cannot be evaluated witout knowledge of te taste (or quality) parameter b it. However, a result from index number teory is tat te ratio of cost functions can be evaluated using data on prices and qualities in te two periods or two countries. Our interest is in te ratio c ( p t, I t ), wic can be measured as follows: c( p, I ) t 1 t Proposition (Feenstra, 1994) Assume tat b it 1 = b it for i I It 1 It, and tat te quantities are costminimizing. Ten, for σ > 1: c pt I I P p pt qt qt I t (, ) λ ( ) = (, 1,, 1, ) c( p, I ) λ ( I ) t SV t 1 ω ( I ) t 1 it it 1/( σ 1) p were PSV pt pt qt qt I it i (, 1,, 1, ) = i I pit is te price index due to Sato (1976) 1 and Vartia (1976), constructed as a geometric mean of te price ratios wit te weigts ω i (I ), wic are constructed from te expenditure sares as in Equations (4) and (5),, (3) s it ( I )= pq pq it it it it i I (4) sit ( I ) sit 1( I ) ω i ( I ) sit I sit 1 I ln ( ) ln ( ) i I s sit 1( I ). (5) ln s ( I ) ln s ( I ) it ( I ) it it 1 Te values λ t (I ) and λ t 1 (I ) are constructed as: p q = = 1 i τ i i I τ λ τ ( I ) pi τqi i I τ τ p q i I τ τ τ, i I i i pi q i I τ iτ τ, τ = t 1, t. (6) Tis teorem states tat te exact price index wit variety cange is equal to te Sato Vartia P SV, multiplied by a term of [λ t (I )/λ t 1 (I )] 1/(σ 1), wic captures te creation and destruction of varieties over time. 161

5 q 2 D B C A q 1 FIGURE 1. Gains from consumer variety Notice tat eac of te terms λ τ (I ) 1 can be interpreted as te period τ expenditure on te varieties in te overlapping set I, relative to te period τ total expenditure. Alternatively, tis can be interpreted as one minus te period τ expenditure on new varieties (not in te set I ), relative to te period τ total expenditure. Wen tere is a greater number of new varieties in period t, or, more precisely, wen te new varieties take greater sare of expenditure tan disappearing varieties, te value of λ t (I ) will be lower tan γ t 1 (I ). Ten te exact price index will be lower relative to te conventional price index, wic does not take into account te cange in varieties. Tus, λ t (I ) provides an inverse measure of new varieties in period t. Te term [λ t (I )/λ t 1 (I )] 1/(σ 1) measures te decrease in unit cost (or price index) due to expansion of te range of varieties. Figure 1 illustrates te gains from adding a new variety. We consider a two-good case: wen only one product (q 1 ) is available (at A), te minimum cost of acieving te utility level U (represented by te indifference curve ACD) is te budget line AB. Wit te introduction of a second product (q 2 ), te minimum expenditure of getting te same utility is te budget line tangent to te indifferent curve troug C. Tis downward sift of te budget line sows consumers gain from product variety. Wile Figure 1 focuses on te consumer impact from new varieties, tere is an analogous argument on te benefits from output variety for producers, wic we turn to next. 3. Measurement of output or export variety Now we turn to te case of measuring output variety. Consider a world economy wit many = 1,..., H countries, eac of wic produce many types of goods. For simplicity in tis section we aggregate tese goods into a single sector, but te extension to multiple sectors will be immediate. For eac period t, let te set of goods produced in country be denoted by I t {, 123,,.}. Ten te quantity vector of eac type of good produced in country c in period t is denoted by q t > 0. Te aggregate output of eac country, Qt = f( qt, It ), is in te same form as in Equation (1). However, te elasticity of substitution between outputs is σ < 0. Note tat Q t could be interpreted as a scalar measure of resources needed to produce tose outputs, q t > 0. We assume tat total output obtained from te economy is constrained by te transformation curve: 162

6 R. C. Feenstra and H. Ma: Trade Facilitation and Exports q 2 D B C A q 1 FIGURE 2. Gains from output variety F[ f( qt, It ), Vt ] = 0, (7) were Vt v = ( 1t, v2t,, vmt ) > 0 is te endowment vector for country in year t. Because for outputs, we suppose tat σ < 0 in Equation (1), wic means tat te set of feasible output varieties, q it, in any country will lie along a strictly concave transformation curve defined by Equation (7). Tis is sown in Figure 2, were we draw te transformation frontier between two product varieties, q 1t and q 2t, witin a country. For a given transformation curve, and given prices, an increase in te number of output varieties will raise revenue. For example, if only output variety 1 is available, ten te economy would be producing at te corner A, wit output revenue sown by te line AB. Ten if variety 2 becomes available, te new equilibrium will be at point C, wit an increase in revenue: it is clear to see tat te revenue line sifts out. Tis illustrates te benefits of output variety. Figure 2 considers maximizing te value of output obtained in eac industry. Under te assumption of perfect competition, te value of output obtained in eac country will be PQ t t, were P t is a CES index of te prices of all output varieties produced in te country: P c p I =( b p ) 1 σ b = σ (, ) ( ), a > 0, = 1,, H, (8) t t t i it i It 1/( 1 σ ) were p t > 0 is te domestic price vector for eac country. Te rigt-and side of Equation (8) is a CES cost function, so, again, te exact price index teorem developed in Feenstra (1994) applies ere. In particular, te ratio of te output price indices over two countries, a and b, equals te product of te Sato Vartia price index of goods tat are common, It ( It a It b ), multiplied by terms reflecting te revenue sare of unique goods: i i P P t a t b = pit a i It ω ( ) t a 1/( It σ 1) λ ( ) pit b t b i I It λ ( ), ab, = 1,, H, (9) were te weigts ω i (I t ) are constructed from te revenue sares in te two countries: 163

7 s ( I ) s ( I ) ln ( ) ln ( ) it a t it b t ω i ( I t ) sit a It sit b It sit a ( It ) sit ( ) b It ln ( ) ln ( ) (10) sit a It sit b i I t It sit ( It) pq it it pq it it, for = ab, (11) I λ t ( I ) t pq pq it it it it i I i It, i I = = 1 pq it it p i I it t i q I it t, for = a, b. (12) Notice tat te output sares in Equation (11), for eac country, are measured relative to te common set of goods I t. Ten, te weigts in Equation (10) are te logaritmic mean of te sares sit a ( It ) and sit b ( It ), and sum to unity over te set of goods i I t. 3 To interpret Equation (12), notice tat λ t ( I t ) 1 due to te differing summations in te numerator and denominator. Tis term will be strictly less tan one if tere are goods in te set I t tat are not found in te common set I t. In oter words, if country a is selling a some goods in period t tat are not sold by country b, tis will make λ t ( I t )<1. Ten we could use λt a ( It) λt b ( It ) as an inverse measure of country a s export variety, relative to country b. Having more export variety in a (i.e. aving lower λt a ( It) λt b ( It )) leads to a iger price index for a (because σ < 0), reflecting an increase in a s revenue. For cross-section comparison of λt a ( It) λt b ( It ), we could coose te worldwide exports to all destinations as a consistent comparison country. Let It w = It be te complete set of varieties exported by te world in year t, and let pit w qit w be te total value of imports for good i. Ten, comparing country to te world in year t, it is obvious tat te common set is te goods exported by ; i.e. It It It w = It ; = a, b. Terefore, λ t ( It ) = 1 ; = a, b, and a direct measure of country s export variety (recall σ < 0) is: λ t w w w p t i I it q it t ( I ) =, = a, b. p q i It w it w it w (13) Te system of Equations (9) (13) above is exactly a cross-country analogue to te time-series import price index in Feenstra (1994). New varieties lead to a fall in prices (from reservation level on demand) for consumers or importers, but a rise in prices (from reservation level on supply) for producers or exporters. Te time-series version of import variety is used by Broda and Weinstein (2006), wereas te cross-section version provides us te teoretical base to te derivation of te export extensive margin, as defined in Hummels and Klenow (2005). From Equation (13), a country s export variety is measured as te worldwide export in goods exported by te country, relative to te worldwide export in all goods. Tis is exactly wat is used in Hummels and Klenow using teir worldwide data! 3 More precisely, te numerator of Equation (10) is te logaritmic mean of te output sares of te two countries, and lies in-between tese sares. Te denominator of (10) is introduced so tat te weigts ω i(i t) sum to unity. 164

8 Our main interest is nations export varieties (or te extensive margin of exports). Given te panel property of our data wic cover 1988 to 2005 and a large sample of countries, we will adopt te union metod developed in Feenstra and Kee (2008). Tis union metod combines cross-section and time-series and provides consistent measures of export variety, and is briefly summarized as follows. Suppose tat te set of exports from country and w differ, but ave some varieties in common. Denote tis common set by It ( It It w ).Aninverse measure of export variety from is λ ( I ) λ ( I ), were λ t a R. C. Feenstra and H. Ma: Trade Facilitation and Exports t t t w t ( I ) t pq pq it a it a it a it a i I i It, i I = = 1 pq it a it a a a p i I it t i q I it t, for a=, w. (14) We will use te worldwide exports in all year as a comparison: denoting tis comparison w country by w, so tat te set I = It is te total set of traded varieties over all years, t, and pi w qi w is te average value of exports for variety i. Tat is, we take te union of all products sold in any year, and also average te export sales of eac product over years. Ten, comparing country to country w, it is immediate tat te common set of goods exported or imported is I It I w = It ; terefore, we ave tat λ t ( I t )=1, so a direct measure for bilateral export variety is given by: λ t w p q t i It ( I ) =. p q i Iw w i w i i w i w In words, according to Equation (15), te bilateral export variety (or te extensive margin of exports) from to j is defined as te world s average exports to j in categories tat are exported by to j, relative to te world s average exports to j in all categories. By coosing te world s average exports over all sample years as comparison, our measure of te extensive margin is consistent across nations and over time periods. Ten to summarize te bilateral export variety into a multilateral export variety index for eac country, we adopt te Sato Vartia index number metod: a geometric mean of bilateral export varieties from te same country to different destinations, wit weigts defined as te logaritmic mean of te sares of j in te overall exports of and te world w, wic is also normalized so tat te weigts sum up to unity. (15) 4. Trade facilitation In tis section we turn to our empirical estimation on factors tat facilitate trade. Our main interest is to look at te influence of different factors on trade, and, in particular, on export variety. We first describe our data sources. 4.1 Export data We draw our trade flow data from te Commodity and Trade Database (COMTRADE) database of te United Nations Statistics Division. Te data are reported in te Harmonized System (HS) classification code at 6-digit level, wic means altogeter 5017 HS-6 165

9 products, and include sipment values and quantities. Te database combines bilateral import data collected by te national statistical agencies of importing countries over all teir exporting partners. Tis is te same data set used in Hummels and Klenow (2005), but our data cover a muc longer time period ( ) instead of teir single year sample (only 1995). Note tat te HS classification was not widely adopted until te mid-1990s. 4 For example, in our data set tere are only 11 countries tat reported teir imports using te HS classification in In addition, countries taking an important role in international trade participate in te system only in later years: te USA since 1991, Cina since 1992, UK and Russia since 1993, and France and Italy since Altoug tis lack of reporting countries in te early years would not urt our calculation of bilateral extensive margins and related regressions, it would bias our estimation on a compreensive multilateral extensive margin for eac country. Tis is because to calculate a compreensive multilateral extensive margin, we need to know te importance of eac of te country s exporting destinations in world trade and use te relative importance as weigts. Hence, we start from 1994 to construct te multilateral export varieties, wile using te full set of data to construct a separate measure of bilateral export varieties for our regressions. 4.2 GDP and population Te standard formulation of te gravity model includes variables on country endowments and economic capacity, wic play important roles in determining countries bilateral trade flow. Our empirical specification will be built on suc a gravity equation, were we include te total population of te country and te real GDP per capita (in 2000 US$). Tose data are taken from te World Development Indicators (World Bank, various years). Te data set covers a broad number of countries 5 (228 countries) over a long time period ( ). 4.3 Gravity factors Besides income level and population, we also need institutional and geograpical elements as additional controls. To control cultural and geopolitical factors, we introduce dummies for, for example, common land border, RTA and usage of common language, into our gravity model. Tose data are taken from Rose (2004), wo constructs a ric data set covering all tose variables. Transportation costs are an important factor determining trade volume and trade components; one key factor affecting transportation costs is te distance between trade partners. Rose (2004) also provides data on distance, wic measures te Great Circle distance between capital cities of eac country. Because tis data set ends in 1999, we will update information wenever possible. It is argued in te literature tat OECD membersip could possibly promote bilateral trade, so we will also use information on countries OECD membersip. 4 Te year (1995) used in Hummels and Klenow (2005) is good enoug because te 59 importers in data represent te vast majority of world imports. 5 However, Taiwan is not included in te World Development Indicators. In tis case, we use data from te Penn World Table instead, wic is up to

10 R. C. Feenstra and H. Ma: Trade Facilitation and Exports 4.4 Bilateral import tariff We draw te tariff data from UNCTAD s Trade Analysis and Information System (TRAINS). Te tariff lines between countries are available by 6-digit HS categories. We take simple averages to generate a measure of bilateral import tariffs Port efficiency Ocean ports are a central and necessary component in facilitating international trade. Blonigen and Wilson (2008) develop a straigtforward measure of port efficiency. In teir metodology, port inefficiency adds additional cost to te total import costs, including port administration and financing costs. Tey run an ordinary least squares regression, regressing import carges on tose observable cost terms, suc as distance and freigt costs, and ten port inefficiency is uncovered from a ports fixed-effect indicator. Because tis fixed-effect dummy measures te ports contribution to te import costs, it is inversely correlated wit a measure of port efficiency: te lower is tis estimate of fixed effects, te more efficient is te port. Due to limitation of data, Blonigen and Wilson could, at best, provide te estimate on te top 100 foreign ports, over 1991 to Following Blonigen and Wilson, we use a weigted-average port efficiency index were te weigts are eac port s sare in te imports of te USA. Using one minus te original fixed effect estimate, we obtain a positive measure of port efficiency. By taking te natural log, we are able to measure te elasticity of export variety in response to improvement in port efficiency. One caveat is tat teir estimation only uses US import data, so it actually only provides te foreign ports efficiency in teir trade wit one or more US local ports. In our estimation using countries exports to te wole world, ideally we want to collect a compreensive efficiency index uncovering eac port s performance to all its destinations. In adopting Blonigen and Wilson s measure, we ave to sacrifice some accuracy in several respects: first, tat a port s efficiency does not vary wit its destination; and, second, tat a weigted average of efficiency indices of tose ports tat are utilized in transporting products to or from te USA is a close approximation of te efficiency summary of all ports in te same country. 5. Estimation results Our bencmark estimation is based on te following gravity model of bilateral international trade: ln( EVijt ) = α ln( gdppcit ) + α ln( gdppc jt ) + β ln( popit ) + β ln( pop ) + γ port + κtariffijt + φ1ln( distij ) + φ2comlangij + φ3borderij + φ4regionalijt + φ oecd + φ oecd + τ + θ + λ + ε jt ijt 5 it 6 jt t i j ijt, (16) were i refers to te exporter, j denotes te importer and t denotes year. Te left-and side dependent variable EV ijt indicates te bilateral extensive margin of exports between te 6 Conceptually, it is te applied tariffs including preferential tariffs, wic is importer-exporter-pair specific. Besides te simple average tariff, we could also use a weigted average of bilateral import tariffs, wic leads to very similar results. 167

11 TABLE 1 Summary statistics for key variables Variable Observations Mean Standard deviation Minimum Maximum Port efficiency 17, L.1 period port efficiency 15, L.3 period port efficiency 12, L.5 period port efficiency 9, Bilateral import tariff 16, Log imp gdp p/c (2000 US$) 17, Log exp gdp p/c (2000 US$) 17, Log imp country population 17, Log exp country population 17, Log distance 17, Common language 17, Common border 17, Regional trade agreement 17, Exporter OECD member 17, Importer OECD member 17, trade partners i and j, atyeart; i.e. te bilateral export variety from country i to country j. As a comparison, we will also use te intensive margin, te bilateral trade flow from i to j relative to te average world export to j in te same categories. As derived in Section 3 (Equation (15) in particular), te bilateral export variety from to j is defined as te world s average export to j in categories tat exported by to j, at time t, relative to world s average export to j in all categories over all time periods. Accordingly, te intensive margin of exports from to j is defined as value of exports from to j, relative to te world s average export to j in all categories tat actually exports. Among te rigt-and side variables of te equation, te two conventional explanatory variables are gdppc and pop, wic are te per capita GDP and population for trade partners in eac year, respectively; port represents te log of bilateral port efficiency; Tariff represents te simple average of bilateral import tariffs; and dist ij represents te distance between nations i and j. We also add a set of binary indicators depending on weter bot countries use te same common language (comlang), weter tey sare te same border (border), weter tey are witin a RTA (regional) and weter tey are OECD members. Finally, we add τ t as year fixed effect, θ i as exporter fixed effect and λ i as importer fixed effect, and ε ijt, te ortogonal error term. Using separate importing and exporting country fixed effects, we are able to capture te multilateral resistance terms in Anderson and van Wincoop (2003). As described in Section 4, after combining data from various sources, our final sample for te bencmark regression covers 41 countries, 819 country pairs, and spans from 1991 to Table 1 gives te summary statistics for all major variables. Hummels and Klenow (2005), wo estimate cross-exporter extensive margins of exports using a sample of single-year observations, sow tat large countries export more, not only in greater volume but also a wider range. Tis is also confirmed by our bencmark regression reported in Table 2. Larger/ricer countries also tend to import more, bot in volume and in variety range. Columns (1) and (4) give te basic specification sowing te positive and significant effect of port efficiency on trade in te extensive margin, and positive but insignificant effect on trade in te intensive margin. To control for te possible underestimation of standard errors, we cluster te standard errors by importer exporter pairs in te bencmark specification and all te following regressions. 168

12 R. C. Feenstra and H. Ma: Trade Facilitation and Exports TABLE 2 Export variety and trade facilitating factors Extensive margin Intensive margin (1) (2) (3) (4) (5) (6) Port efficiency 0.218** 0.336** 0.154** (4.14) (5.59) (3.50) (1.06) (1.55) (0.43) Bilateral import tariff 1.234** 2.511** 1.771** ** (2.87) (7.45) (3.38) (1.69) (3.36) (0.79) Log imp gdp p/c (2000 US$) ** 1.222** 0.289* 1.886** (1.34) (1.20) (6.21) (7.56) (2.34) (10.66) Log exp gdp p/c (2000 US$) ** ** 1.125** (1.92) (2.85) (0.88) (3.17) (9.57) (0.88) Log imp country population 3.170** 3.454** * (9.30) (11.77) (0.37) (2.15) (0.87) (0.36) Log exp country population 2.819** 3.613** * (8.44) (12.17) (0.72) (2.13) (0.28) (1.64) Log of distance 0.606** 0.606** 0.416** 0.412** (15.09) (15.14) (11.59) (11.48) Common language 0.455** 0.454** 0.262** 0.263** (7.27) (7.25) (4.42) (4.45) Land border dummy * 0.335* (0.50) (0.49) (2.30) (2.31) Regional agreement 0.937** 0.944** 0.540** 0.554** (3.95) (3.99) (5.23) (5.42) Exporter OECD member * (0.36) (0.25) (0.33) (2.00) (1.70) (1.05) Importer OECD member 0.173* 0.146* 0.134* ** (2.39) (2.42) (2.26) (0.54) (6.82) (0.60) Year fixed effects Yes Yes No Yes Yes No Country fixed effects Yes No Yes Yes No Yes Country pair fixed effects No Yes No No Yes No Year trend No No Yes No No Yes Trend * country fixed effects No No Yes No No Yes Observations 14,686 14,686 14,686 14,686 14,686 14,686 R Notes: Robust t-statistics are in parenteses. *Significant at 5%; **Significant at 1%. Te estimated coefficients are in line wit our expectations. First, larger countries (in GDP per capita and in population) tend to export more varieties and import more varieties. Te only exception is for te exporting country s population, wic is negative. Second, aving more efficient ocean ports will substantially facilitate trade; tis is more pronounced for trade in variety, as in column (1). For te intensive margin of exports, te coefficient is positive but not significant, as in column (4). Furtermore, a bilateral import tariff significantly reduces exports in te extensive margin. In summary, improving ports efficiency by 10% increases export variety by 2.18%, and increases te intensive margin of exports by 0.65%. A drop of 10% in te average bilateral import tariff leads to an increase in export variety by 12.3%, wile te intensive margin of exports decreases by 6.8%. Notice tat te intensive margin of exports from i to j is measured as te bilateral exports relative to te world s average exports to te same importing country in te product categories exported from i to j; terefore, a positive coefficient of port efficiency on te intensive margin does not necessarily mean trade costs suc as tariffs promote trade volume in absolute values. 169

13 Not surprisingly, being more distant from eac oter significantly reduces trade, in bot extensive and intensive margins, wile speaking te same language appears to promote trade in bot margins. Moreover, if two countries are geograpically contiguous, or ave regional RTA wit eac oter, tey trade more in te intensive margin, rater tan in te extensive margin. Interestingly, it seems tat te importing country s OECD membersip matters more tan te exporting country s, especially for te extensive margin of exports. Tis indicates tat, controlling for oter factors, OECD countries (i.e. developed countries) demand muc more variety tan non-oecd countries (i.e. developing countries). In columns (2) and (5), we add country pair fixed effects, aiming to control for all time-invariant factors (bot observed and unobserved) pertaining to te country pair. Tis treatment makes te identification of te bilateral port efficiency coefficient only depend on time-series variation and takes out all time-invariant features (bot observed and unobserved), suc as border dummy and distance. Bencmark results are confirmed. Efficient ocean ports elp in facilitating export in variety, but ave little to do wit promoting relative export volume. TABLE 3 Export variety and trade facilitating factors: Extension Extensive margin Intensive margin (1) (2) (3) (4) (5) (6) Exporter port efficiency ** (0.14) (4.39) (1.00) (0.81) (0.09) (0.84) Importer port efficiency 0.096** 0.075* 0.044** 0.073** (4.43) (2.55) (2.73) (2.95) (1.63) (0.24) Bilateral import tariff 1.229** 2.499** 1.781** ** (2.86) (7.42) (3.40) (1.68) (3.40) (0.79) Log imp gdp p/c (2000 US$) ** 1.220** 0.284* 1.889** (1.37) (1.11) (6.45) (7.57) (2.32) (10.66) Log exp gdp p/c (2000 US$) ** ** 1.125** 0.16 (1.89) (2.91) (1.08) (3.17) (9.58) (0.90) Imp country population 3.169** 3.445** * (9.29) (11.70) (0.29) (2.16) (0.91) (0.36) Exp country population 2.784** 3.603** * (8.30) (12.10) (0.61) (2.18) (0.24) (1.64) Log of distance 0.605** 0.606** 0.416** 0.412** (15.08) (15.13) (11.59) (11.48) Common language 0.455** 0.454** 0.262** 0.263** (7.28) (7.25) (4.42) (4.45) Land border dummy * 0.335* (0.50) (0.49) (2.30) (2.31) Regional agreement 0.936** 0.944** 0.540** 0.554** (3.95) (3.99) (5.23) (5.42) Exporter OECD member (0.52) (0.12) (0.26) (1.95) (1.73) (1.05) Importer OECD member 0.171* 0.136* 0.142* ** (2.36) (2.25) (2.42) (0.60) (6.82) (0.60) Year fixed effects Yes Yes No Yes Yes No Country fixed effects Yes No Yes Yes No Yes Country pair fixed effects No Yes No No Yes No Year trend No No Yes No No Yes Trend * country fixed effects No No Yes No No Yes Observations 14,686 14,686 14,686 14,686 14,686 14,686 R Notes: Robust t-statistics are in parenteses. *Significant at 5%; **Significant at 1%. 170

14 R. C. Feenstra and H. Ma: Trade Facilitation and Exports TABLE 4 Export variety and trade facilitating factors: Witin OECD countries Extensive margin Intensive margin (1) (2) (3) (4) (5) (6) Port efficiency * (0.34) (1.84) (2.60) (1.23) (0.66) (0.80) Bilateral import tariff 1.493* (2.20) (1.33) (1.24) (0.53) (0.44) (0.99) Log imp gdp p/c (2000 US$) 0.424* 0.572** 0.484* 2.237** 2.254** 1.974** (2.29) (2.89) (2.29) (6.17) (6.32) (5.32) Log exp gdp p/c (2000 US$) * 0.720* (1.95) (1.48) (2.60) (2.15) (1.68) (0.91) Log imp country population 1.393** 1.672** 1.681** (2.85) (2.85) (3.24) (0.98) (1.82) (1.42) Log exp country population 2.439** 2.772** 3.619** (4.45) (4.49) (4.71) (0.42) (0.95) (1.08) Log of distance 0.260** 0.252** 0.246** 0.842** 0.831** 0.825** (5.64) (5.57) (5.56) (8.43) (7.95) (7.57) Common language 0.245** 0.254** 0.245** (3.95) (3.97) (4.22) (1.54) (1.45) (1.55) Land border dummy 0.266** 0.257** 0.268** (2.96) (2.81) (3.00) (1.02) (0.48) (0.55) Regional agreement (1.77) (1.79) (1.61) (0.58) (0.69) (0.61) Lagged port efficiency No 3 period 5 period No 3 period 5 period Year fixed effects Yes Yes Yes Yes Yes Yes Country fixed effects Yes Yes Yes Yes Yes Yes Observations 2,220 1,741 1,482 2,220 1,741 1,482 R Notes: Robust t-statistics are in parenteses. *Significant at 5%; **Significant at 1%. One concern wit using country fixed effects or country pair fixed effects is tat it does not adequately control for time varying multilateral resistance effects, as pointed out in Anderson and van Wincoop (2003) and Baldwin and Taglioni (2006). To deal wit tis, in columns (3) and (6) we use a time trend, country fixed effects, as well as te interaction between time trend and country dummies. 7 Te results confirm te previous estimations tat port efficiency substantially promotes trade in te extensive margin, altoug to a lower magnitude. A 10% improvement in port efficiency leads to a 1.5% increase in te extensive margin of exports. Furtermore, lowering bilateral import tariffs also significantly promotes te extensive margin of exports. In Table 3, we furter separate te importer port efficiency and te exporter port efficiency. 8 Te findings are mostly consistent wit tose of Table 2. Furtermore, importer port efficiency plays a substantial role in increasing te extensive margin of exports from te source country. Te exporter port efficiency also matters but is not precisely estimated, except for te specification wit country pair fixed effects. In most cases, bot importer and exporter port efficiency do not matter for te intensive margin of exports. 7 Ideally we sould use country-year fixed effects. However, given te limited variation of our sample, using country-year fixed effects generates serious multicollinearity problem. 8 We tank an anonymous referee for suggesting tis estimation. 171

15 TABLE 5 Export variety and trade facilitating factors: Importing OECD countries Extensive margin Intensive margin (1) (2) (3) (4) (5) (6) Port efficiency 0.117* 0.404** 0.550** (1.96) (3.16) (3.10) (1.67) (0.54) (1.33) Bilateral import tariff 5.352** 5.771** 5.989** (5.22) (5.46) (5.51) (0.94) (0.82) (0.55) Log imp gdp p/c (2000 US$) * 0.933* 1.269** 1.233** 0.984* (1.27) (2.37) (2.36) (3.30) (2.96) (2.21) Log exp gdp p/c (2000 US$) * 0.373* 0.617** 0.799** 0.869** (1.44) (2.44) (2.22) (3.42) (3.64) (3.71) Log imp country population 3.147** 2.976** 2.770** ** 4.784** (4.82) (4.11) (2.74) (1.65) (4.16) (4.01) Log exp country population 1.289** 1.964** 2.508** * (3.20) (4.04) (3.68) (1.68) (1.96) (1.98) Log of distance 0.512** 0.467** 0.456** 0.294** 0.267** 0.255** (9.13) (8.24) (7.97) (4.07) (3.46) (3.24) Common language 0.280** 0.257** 0.259** 0.277** 0.288** 0.237* (4.65) (3.89) (3.84) (3.16) (2.96) (2.44) Land border dummy (0.82) (0.83) (0.85) (0.76) (0.39) (0.23) Regional agreement 0.413** 0.471** 0.478** 0.640** 0.597** 0.568** (3.52) (3.48) (3.28) (4.43) (4.15) (3.81) Lagged port efficiency No 3 period 5 period No 3 period 5 period Year fixed effects Yes Yes Yes Yes Yes Yes Country fixed effects Yes Yes Yes Yes Yes Yes Observations 4,195 3,250 2,657 4,195 3,250 2,657 R Notes: Robust t-statistics are in parenteses. *Significant at 5%; **Significant at 1%. Troug Tables 4 to 8, we investigate alternative specifications to our bencmark model. We first investigate port efficiency and trade among OECD member countries in Table 4. Ten, in Table 5, we investigate a subsample were only te importing country is an OECD member, wile te exporting country is not. Table 6 examines te case were only te exporting country is an OECD member wile te importing country is not. Finally, in Table 7, we consider te case were none of te trading partners belongs to te OECD. We run tree regressions for eac subsample. Column (1) redoes te regression as specified in Equation (16) but only a subsample of data is used. Columns (2) and (3) run te same regression, lagging port efficiency by 3 and 5 years, respectively. 9 Tis is to at least partially control for te potential endogeneity of port efficiency: countries tat ave large trade transactions wit eac oter are more likely to invest to improve port efficiency. Ten columns (4) to (6) repeat te same regression from columns (1) to (3), wit te intensive margin of exports as te dependent variable. It is expected tat more trade in varieties will be observed among industrial countries, rater tan between Nort Sout country pairs or among Sout countries. Tis is because intra-industry trade, wic is mainly in differentiated products, is dominant between te industrial nations, wo ave similar income levels and consumer preference. In columns 9 Lag by 1 year gives similar results as using current year port efficiency. 172

16 R. C. Feenstra and H. Ma: Trade Facilitation and Exports TABLE 6 Export variety and trade facilitating factors: Exporting OECD countries Extensive margin Intensive margin (1) (2) (3) (4) (5) (6) Port efficiency ** 0.380* 0.220* (0.69) (2.97) (2.05) (2.52) (0.21) (1.58) Bilateral import tariff 0.533* * 1.127** (1.98) (1.21) (2.20) (3.07) (1.57) (0.29) Log imp gdp p/c (2000 US$) 0.290* 0.619** 0.622** 1.734** 2.467** 2.591** (2.30) (5.05) (5.33) (8.39) (12.71) (12.56) Log exp gdp p/c (2000 US$) (0.36) (0.10) (0.15) (0.93) (0.33) (0.47) Log imp country population 2.389** 2.625** 2.745** 3.255** 3.393** (4.30) (4.17) (5.08) (4.84) (3.75) (0.54) Log exp country population 6.129** 5.274** 2.833** 2.320* (7.75) (5.19) (3.95) (2.14) (0.40) (0.71) Log of distance 0.529** 0.465** 0.433** 0.764** 0.762** 0.780** (8.92) (8.42) (8.38) (13.97) (13.63) (13.78) Common language 0.369** 0.378** 0.345** 0.489** 0.498** 0.466** (6.35) (6.27) (5.93) (5.72) (5.32) (5.03) Land border dummy * 0.325* (1.62) (2.57) (2.32) (0.86) (0.64) (0.94) Regional agreement * 0.565** (0.58) (0.65) (0.59) (1.55) (1.99) (2.59) Lagged port efficiency No 3 period 5 period No 3 period 5 period Year fixed effects Yes Yes Yes Yes Yes Yes Country fixed effects Yes Yes Yes Yes Yes Yes Observations 3,185 2,507 2,081 3,185 2,507 2,081 R Notes: Robust t-statistics are in parenteses. *Significant at 5%; **Significant at 1%. (1) and (4) of Table 4, we redo te regression as specified in Equation (16) but only look at trade among OECD countries. Countries wo are OECD members are, on average, larger and ricer tan non-oecd countries, and trade between two OECD members is expected to be more pronounced in differentiated products and more on te extensive margin. As expected, te importing country s GDP per capita as a more pronounced positive effect tan tat of te full sample. In contrast, te coefficient of port inefficiency becomes insignificant, probably reflecting te fact tat a ig level of port efficiency as already been obtained witin te OECD. Wen we are using lagged port efficiency, it regains significance as in column (3). Table 5 examines te case in wic only te importing country is an OECD member. In tis case, port efficiency significantly promotes exports in te extensive margin, wile bilateral import tariff significantly discourages exports in te extensive margin. Bot variables ave no significant effects on trade in te intensive margin. In tis case, GDP per capita and population of bot importing country (OECD member) and exporting country (non-oecd member) become important in determining export variety. Table 6 is te oter side of te coin for Table 5: in tis case, te exporting country is an OECD member wile te importing country is not. Estimates of port efficiency sow te same pattern as in te bencmark regression and Table 5, wen we use lagged measures of port efficiency. However, te coefficients on tariffs are controversially positive for bot 173

17 TABLE 7 Export variety and trade facilitating factors: Non-OECD countries Extensive margin Intensive margin (1) (2) (3) (4) (5) (6) Port efficiency * * (1.66) (1.97) (0.39) (1.83) (2.41) (0.81) Bilateral import tariff 1.283* * (2.34) (1.87) (2.49) (0.31) (0.74) (0.03) Log imp gdp p/c (2000 US$) ** 1.009** 1.011** 1.163** 1.652** (1.11) (3.04) (3.03) (3.58) (3.40) (4.47) Log exp gdp p/c (2000 US$) * 0.698* (1.27) (1.55) (1.51) (1.76) (2.35) (2.10) Log imp country population 4.593** 4.115** 4.271** 2.220* (4.94) (3.84) (2.78) (2.20) (1.10) (0.29) Log exp country population 2.939** 3.181** (3.60) (3.22) (1.90) (1.25) (0.73) (0.03) Log of distance 0.796** 0.731** 0.710** 0.360** 0.382** 0.395** (13.25) (12.29) (11.82) (5.51) (5.43) (5.42) Common language 0.338** 0.339** 0.339** * 0.279* (2.92) (2.92) (2.83) (1.93) (2.01) (2.02) Land border dummy (0.48) (0.10) (0.04) (1.48) (1.38) (1.41) Regional agreement 2.163** 2.167** 2.109** 1.246** 1.245** 1.325** (6.82) (6.08) (5.74) (5.78) (5.55) (5.56) Lagged port efficiency No 3 period 5 period No 3 period 5 period Year fixed effects Yes Yes Yes Yes Yes Yes Country fixed effects Yes Yes Yes Yes Yes Yes Observations 5,086 4,001 3,198 5,086 4,001 3,198 R Notes: Robust t-statistics are in parenteses. *Significant at 5%; **Significant at 1%. extensive and intensive margins. Tis reflects tat for less developed importing countries, iger tariffs probably make it even arder to import from oter less developed countries and, terefore, relatively increases imports from OECD countries in bot margins. Table 7 examines trade between non-oecd country pairs. Tis time, port efficiency is positive and marginally significant (at te 10% level) in promoting export variety in column (1), wile a bilateral import tariff as a negative and significant effect on export variety. In regressions sown in Tables 2 to 7, trade flows are not distinguised by its transport mode. Tis potential raises doubt regarding te validity of using te ocean port efficiency measure adopted from Blonigen and Wilson (2008). Due to a lack of detailed data on trade wit different transport modes, we instead investigate a special case were no country pairs sare any common border, sown in Table 8. Most of te regression estimates are quite similar in magnitude and in economic meaning to te story tat Table 2 delivers, and are not discussed in detail. It is interesting to note tat te coefficient estimates for RTA sow diverging patterns for estimations wit different groups of countries. Namely, for our bencmark regressions (Tables 2 and 3) wit all countries, or regressions wit OECD importing countries, or regressions wit countries tat do not sare common borders, RTA seem to discourage te export at te extensive margin wile increase te export at te intensive margin (te effect 174