A Theory on the Role of Wholesalers inInternational Trade Based on Economies of Scope∗

Anders Akerman†

January 5, 2014 (first version January, 2010)

Abstract

This paper offers an explanation for the existence of wholesalers and other inter-mediaries in international trade, and analyses their effect in an economy withheterogeneous manufacturing firms and fixed costs of exporting. Wholesalerspossess a technology such that they can buy manufacturing goods domesticallyand sell in foreign markets, and they can, unlike manufacturers, handle morethan one good. A wholesaler therefore faces an additional fixed cost which isincreasing in the number of goods it handles. The entry of wholesale firms leadsto productivity sorting. The most productive firms export on their own bypaying a fixed cost, but a range of firms with intermediate productivity levelsexport through international wholesalers. A higher fixed cost of exportingmeans that wholesalers handle (i) a higher share of total export volumes and(ii) a higher share of the exported product scope (i.e. the number of exportedproducts). A higher fixed cost of exporting gives wholesalers a larger role,since these can spread the fixed cost across more than one good. The wholesaletechnology therefore exhibits economies of scope. An empirical analysis usingSwedish firm-level data supports the main assumption and predictions of themodel.

Keywords: heterogeneous firms, international trade, intermediationJEL codes: D21, F12, F15

∗I am grateful for helpful comments and suggestions from Andrew Bernard, Gregory Corcos,Matthieu Crozet, Jonathan Eaton, Gabriel Felbermayr, Harry Flam, Rikard Forslid, Benjamin Jung,Sebastian Krautheim, James Markusen, Lindsay Oldenski, Horst Raff, Peter Schott, Heiwai Tangand Stephen Yeaple as well as seminar participants at the Midwest International Trade Meetingsin Chicago, NOITS in Helsinki, RIEF in Kiel, Uppsala International Trade Workshop, the EEAMeetings in Glasgow, the ETSG in Lausanne and the EITI in Tokyo. Financial support from JanWallander’s and Tom Hedelius’ Research Foundation is gratefully acknowledged.†Department of Economics, Stockholm University, e-mail: anders.akerman@ne.su.se

1 Introduction

It is by now a well-known empirical regularity that wholesalers and other intermedi-aries account for a substantial share of exports from most countries and that theyassist less productive firms in overcoming barriers to trade to foreign markets. Recentresearch has shown that intermediary firms handle about 22% of aggregate Chineseexport sales (Ahn et al., 2011), 10% of US and Italian exports (Bernard et al. 2011a;2010a), 20% of French exports (Crozet et al., 2013) and about 35% of imports intoChile (Blum et al., 2010). Feenstra and Hanson (2004) examine the role of HongKong in re-exporting goods from China during the period 1988-1998 and find that53% of Chinese exports were shipped through Hong Kong as re-exports. The authorsargue that intermediation was one of the more important reasons. These ratios areeven larger for some goods typically associated with China’s rapid export growthin the 1990s, such as 77% for footwear and 83% for toys. Basker and Van (2011)document the role of Wal-Mart as a catalyst for US imports from China and notethat this large US retailer accounts for 15% of US imports from China.

Moreover, it has also become clear that firms in most countries and sectors arehighly heterogeneous in several dimensions and that this is strongly correlated withexport performance (see for example Mayer and Ottaviano, 2008). It is thus highlylikely that the effect of intermediaries on producing firms is asymmetric. Bernardet al. (2010a) find that exporting firms in the US exhibit substantial heterogeneityas regards to export mode, i.e. whether firms manage their own exporting activitiesor export through intermediaries. McCann (forthcoming) and Abel (2013) useEast European and Turkish datasets, respectively, to find that “indirect” exporting(exporting through intermediaries) is correlated with important firm characteristics.Most importantly, the largest and most productive firms export directly while firmsin the middle of the productivity distribution choose to export indirectly through anintermediary instead.

This paper develops a model with an explicit characterization of wholesalersin a setting with heterogenous manufacturing firms (as in Melitz, 2003) based onan assumption on why wholesalers exist and how they operate.1 The model ischaracterized by free entry in all sectors and therefore allows for a welfare analysisof the presence and structure of a wholesale industry in a general equilibriumframework. As in Melitz (2003), exporting is subject to an initial fixed cost. Therole of wholesalers in international trade is that they are able to pool the fixed costof exporting across more than one good. They do not produce goods themselves but

1The term wholesalers will be used in the paper since the empirical analysis uses data forwholesale firms, but the theoretical model could be applied to intermediaries in general.

1

instead buy goods in their local market and export these goods to foreign markets.They only have to incur the fixed cost of establishing a distribution network in aforeign market once (regardless of how many goods they export). However, theyface a fixed cost of their distribution network, which is convex and increasing inproduct scope (the number of goods that a wholesaler exports) since a more complexproduct portfolio is more costly to handle. For a producer, exporting through awholesaler is therefore an alternative to setting up its own distribution channel.Wholesalers are assumed to be homogeneous since the main focus of the paper ison how aggregate trade flows and heterogeneous producers of goods respond to thepossibility of exporting through wholesalers, as an alternative to managing their owndistribution networks.

The core mechanism that operates in the model is thus the following: Wholesalersspread the fixed cost of exporting across more than one good. But to cover thefixed cost they need to charge a markup between the procurement price of the goodand what it charges the final consumer in the foreign country. This markup thatwholesalers charge causes productivity sorting in the choice of export mode: themost productive firms choose to incur their own fixed cost of exporting since theiroperating profit is large enough. However, some goods, which cannot be profitablyexported by the producer itself, can be exported at a lower fixed cost (per good) bywholesalers, who export several goods but only have to make one investment in thefixed cost.2 This means that the wholesale technology exhibits economies of scope.When fixed costs increase, wholesalers become more important, since fewer firms canexport on their own. Moreover, wholesalers have to expand and handle more goodsto be able to cover the higher fixed cost. Wholesalers therefore help producers inovercoming destination-specific barriers to entry. In an appendix section, I develop amulti-country gravity model where the “multilateral resistance variable” of a countryis found to include the number of wholesalers of trading partners, and their presencelowers a country’s price index as well as mitigates the positive effect that fixed tradecosts have on the price level. Since the price level is indirectly a measure of welfare,wholesalers therefore contribute positively to welfare.

This generates a number of predictions about how exporting is conducted inthe presence of intermediation. First, producers sort according to productivity indetermining their mode of exporting. The most productive firms continue to managetheir own exporting activities and incur the fixed cost associated with this as in thestandard model. However, some firms that were almost productive enough to export

2The mechanisms which cause sorting are analogous to those in Helpman et al. (2004). Theymodel horizontal FDI as a mode of foreign market entry which, compared to exporting, is associatedwith a higher fixed cost but a lower variable cost.

2

on their own in the standard model now choose to do so, but through wholesalersinstead. The least productive firms do not export through any of the two modes.Second, a higher fixed cost is associated with a larger importance of wholesalersboth in terms of export volumes and export scope. Finally, a higher fixed cost isassociated with each wholesaler handling more goods (having a larger scope).

Swedish data is used to test the main predictions of the model. In this context,Sweden represents a small and highly open economy for which merchandise tradeamounted to 63% of GDP in 2005 according to the World Bank, which can becompared to 21% for the United States in the same year. Specifically, I use Swedishfirm-level data from 2005, which matches data from the Swedish customs office withproduction data for the universe of Swedish firms. The dataset contains a sectorclassification of firms according to main business activity, and also contains theirtrade flows for each eight-digit Combined Nomenclature (CN8) product code anddestination. It can therefore be observed what goods are exported by firms listed aseither “wholesalers” or “manufacturers” where the latter are treated as the producersof goods.

The empirical analysis supports the main predictions of the model. Cruciallyfor the basic assumption about economies of scope, Swedish wholesalers export alarger scope of products (or CN8 product categories) than producers. A wholesalerexports about 54% more products per firm when destination-specific effects and firmsize are accounted for, an empirical fact which supports the notion that there is atechnological difference between wholesalers and producers. Moreover, as predictedby the model, wholesalers export, on average, substantially lower volumes (between 35and 57% less) per firm within a product category than producers. Finally, wholesalersplay a more important role in aggregate exports (both in terms of total volumesand in the number of CN8 product categories exported) from Sweden to countriescharacterized by higher fixed costs. The result for aggregate export volumes is inline with previous studies by Ahn et al. (2011) for China, Bernard et al. (2011a) forItaly and Crozet et al. (2013) for France. Admittedly, no perfect measure of fixedcosts exists so these are proxied by institutional variables from the World Bank thatrelate to the difficulty of importing as done by, for example, Helpman et al. (2008).

This paper is most directly related to a growing literature that studies theeffects of intermediary firms on international trade in a setting with heterogeneousproducing firms. Ahn et al. (2011) introduce the choice of indirect exporting throughan intermediary into a model of firm heterogeneity along the lines of Melitz (2003).This new export mode is assumed to be associated with lower fixed costs but highervariable costs and therefore causes productivity sorting among producers as regards

3

to export mode. These assumptions also mean that intermediaries become moreimportant for destinations characterized by higher fixed costs of entry. Felbermayrand Jung (2011) develop a similar model but where the difference in the ratio betweenfixed costs versus de facto variable costs instead come from contracting frictionsbetween producers and intermediaries. The predictions of these papers are in linewith the predictions I generate and I view my paper as complementary to these papersbut different for mainly three reasons. My paper adds the dimension of productscope and find that wholesalers handle a larger share of the number of exportedproducts to destinations characterized by high fixed costs. This is also confirmedto hold empirically in Sweden but has to my knowledge not been shown before.Moreover, I introduce a specific microeconomic mechanism for why intermediariesexist and how they operate in international markets, namely that they specialize ingenerating economies of scope, and find that this mechanism generates the patternsthat we have seen hold in most empirical firm-level studies of intermediation orwholesalers in international trade. Finally, this paper adds an explicit theoreticalframework for the intermediation sector where wholesalers are profit maximizingfirms and subject to free entry in a general equilibrium setting. This facilitates ananalysis of how aggregate welfare is affected by the existence of an intermediarysector and by the structure of this sector. Blum et al. (2011) also analyze therole of intermediaries in a setting with heterogeneous firms but focus instead onhow intermediaries reduce the costs of matching between producers and foreignconsumers and find that intermediaries are especially important in matching smallproducers with small foreign consumers. Finally, Tang and Zhang (2011) and Crozetet al. (2013) analyze how quality differentiation among producers interact with thepossibility of choosing intermediation as an alternative export mode.

Other theoretical work has also explored the issue of intermediation in inter-national trade but has instead focused on the role of intermediaries in reducingcontracting or matching frictions between buyers and sellers, as in Antràs andCostinot (2010, 2011), Biglaiser (1993), Olsen (2013) or Rubinstein and Wolinsky(1987); or on their positions in networks in international trade, as in Rauch (1999,2001), Rauch and Watson (2004) or Petropoulou (2008).

The paper is structured as follows. Section 2 develops the model by introducinga wholesale industry into the model in Melitz (2003) and derives the main results.Section 3 describes the Swedish data and provides the empirical analysis. Section 4concludes.

4

2 The model

2.1 Basic setup without wholesalers

The model builds on the structure in Melitz (2003) but with an additional “agricul-tural” sector characterized by constant returns to scale. It depicts two economies(Home and Foreign, the latter denoted by superscript F ) with a primary productionfactor labor, L, which is used in all sectors. Ignoring the wholesale sector for themoment, there are two main sectors in the economy. First, the agricultural sector is aWalrasian, homogeneous-good sector with costless trade. Second, the manufacturingsector is characterized by increasing returns, Dixit-Stiglitz monopolistic competitionand iceberg trade costs. Manufacturers face constant marginal production costsand three types of fixed costs. The fixed cost, FE, is the standard Dixit-Stiglitzcost of developing a new variety. The other two fixed costs involved reflect theone-time expense of introducing a new variety into a market: FD(θ, wL) if it isthe domestic market and FX(θF , wFLF ) for the foreign market, where θ denotesa vector of country-specific characteristics that determine the difficulty of entry(such as, for example, the quality of institutions) and w denotes the wage level in acountry. Consequently, wL denotes the gross domestic product of a country. Whenanalyzing the comparative statics of the model, I will use these relationships to seehow country characteristics affect outcomes through the fixed cost of entry but forease of exposition I denote F j ≡ F j(θj, wjLj) until then.3

There is heterogeneity with respect to firms’ productivity levels, ϕ. Each Dixit-Stiglitz firm/variety is therefore associated with a particular labor output coefficientdenoted by ϕi for firm i. After sinking FE units of labor in the product innovationprocess, the firm is randomly assigned ϕi from the cumulative distribution functionG(ϕ). The analysis focuses on steady-state equilibria and intertemporal consider-ations are ignored; the present value of firms is kept finite by assuming that firmsface a constant Poisson hazard rate δ of forced exit.

Consumers in each country have two-tier utility functions with the upper tier(Cobb-Douglas) determining the consumer’s division of expenditure among the sectorsand the second tier (CES) dictating the consumer’s preferences over the variousdifferentiated varieties within the manufacturing sector. All individuals in Home

3I will not make any assumption on exactly how the fixed cost is affected by country size sincethis is difficult to know. For example, Arkolakis (2010) points out that due to returns to scale inthe technology of entry (i.e. advertising technology) firms tend to pay a smaller fixed cost percustomer in large economies but a higher total fixed cost. Moreover, larger destinations tend toattract more exporters, see for example Mayer and Ottaviano (2008), which (controlling for thedegree of competitive pressure) possibly makes it easier for a potential new exporter to do so aswell since it can observe the methods of entry of other exporters.

5

have the utility function:

U = CµMC

1−µA

where µ ∈ (0, 1), and CA is the consumption of the homogeneous good. Manufacturesenter the utility function through the index CM , defined by

CM =(ˆ N

0cσ−1σ

i di

) σσ−1

where N is the mass of varieties consumed, ci the amount of variety i consumed andσ > 1 the elasticity of substitution between varieties.

Each consumer spends a share µ of his income on manufactures, and demand fora variety i is therefore

xi = Ap−σi

where

A ≡ µL

P 1−σ , (1)

pi is the consumer price of variety i, L is the population size and P ≡(´

i∈Ω p1−σi di

) 11−σ

the price index of manufacturing goods available in the Home country. The set ofavailable varieties is denoted by Ω.

The unit factor requirement of the homogeneous good is one unit of labor. Thisgood is freely traded and since it is chosen as the numeraire

pA = w = 1

where w is the nominal wage of workers.In an economy without wholesalers, shipping the manufactured good involves a

frictional trade cost of the iceberg form: for one unit of a good from Home to arrivein the Foreign country, τ > 1 units must be shipped. It is assumed that trade costsare equal in both directions. Profit maximization by a manufacturing firm i locatedin Home exporting to the Foreign country leads to the following consumer price inForeign of firm i’s good:

pFi = σ

σ − 1τ

ϕ.

6

Entrepreneurs entering the manufacturing sector draw their marginal productivity,ϕi, from the probability distribution G(ϕ) after having sunk FE units of labor todevelop a new variety. Having learned their productivity, firms decide on entry in thedomestic and foreign market, respectively. Doing so is associated with fixed marketentry costs; firms pay FD to enter the domestic market and FX to enter the foreignmarket. Firms will therefore enter a market as long as the operating profit in thismarket is sufficiently large to cover the market entry cost associated with the market.Because of the constant mark-up pricing, it is easily shown that operating profitsequal revenues divided by σ. The critical cut-off levels of productivity needed inorder to enter the domestic and foreign markets (for the operating profit to be aslarge as the discounted fixed cost of entry) are given by:

ϕσ−1D A = FD (2)

ϕ′σ−1X τ 1−σAF = FX (3)

where FD ≡ δ (σ − 1)1−σ σσFD and FX ≡ δ (σ − 1)1−σ σσFX . A and AF indicate “per-firm demand” of the Home market and the Foreign market, respectively. Accountingalso for free entry, which means that E(π) = FE, yields that A = AF . The reason forusing the notation ϕ′X is that the export cut-off will be different when wholesalersare introduced. Equations (2) and (3) yield the well-known result from Melitz (2003)which can be summarized as follows (provided that 1

τ1−σFXFD

> 1):

Proposition 1. In a world without wholesalers, only firms with a marginal pro-ductivity above ϕ′X choose to export, firms with a productivity between ϕD and ϕ′Xserve the domestic market only and firms with a marginal productivity below ϕD exitimmediately.

2.2 Introducing wholesalers

The third sector, which is the novel feature of the model, is the wholesale sector(variables relating to this sector are denoted by the superscript “W ”). Wholesale firmsare indexed by j and are homogeneous. The wholesale technology gives a wholesalerfirm j the ability to source a range of goods and ship these to the Foreign country.The mass of goods that wholesaler j ships (equal to the mass of manufacturing firmsthe wholesaler is buying goods from) is denoted by mW

j . The sector is characterizedby free entry. A wholesaler faces the same cost as manufacturing firms to establish aretail channel in the foreign country, FX , but has the technology to export severalgoods. Operations are assumed to become more costly the more goods a firm handles,so it also faces a per-period fixed cost that is monotonically and convexly increasing

7

in the range of goods it handles. Its total fixed cost of foreign market entry istherefore:

FWXj = FX +

(mWj

)γγ

(4)

where γ > 1 and mWj is the mass of domestic manufacturing goods the wholesale

firm j is handling. Since manufacturing firms are atomistic and therefore has a massequal to zero, if a wholesale firm were to only export one single good, its scopemeasure mW

j would be zero and limmWj →0FWXj = FX . Therefore, a wholesale firm

which only exports one good has the same fixed cost of exporting as a manufacturer.Moreover, the functional form in (4) is chosen both for technical and intuitive reasons.The technical reason is that some convexity needs to be included in this functionto put an upper bound on the scope of wholesalers. If it was not present (i.e. ifγ = 0), the economies of scope would be infinite and only one wholesaler wouldexist and would export all goods. The intuitive reason is that it is more costly tomaintain an international distribution system the more different goods are in nature.If mW

j is low, the wholesaler is more specialized in a more narrow range of products,for example sport cars of different kinds. However, as mW

j increases, the productsnecessarily become more different; in the specific example the wholesaler firm alsostarts to export other types of cars or motor vehicles etc., which means that the levelof specialization decreases and the cost per product increases. The wider the scope,the more costly each product is to export.

I assume that a wholesaler gets the exclusive right to sell the manufacturinggood in the foreign market (thereby excluding the possibility that more than onewholesaler sell the same manufacturing good). Since the wholesaler faces a CESdemand abroad, its demand function towards manufacturing firms when procuringtheir products is also characterized by a CES structure. This causes manufacturingfirms to charge the same price to wholesalers as they do to consumers. And sincethe manufacturer imposes a CES markup over its marginal cost and the wholesalerdoes the same, the final consumer price in the foreign economy for a good sold bya wholesaler is characterized by a “double marginalization”; the CES markup ofσ/ (σ − 1) is imposed twice:

pWij =(

σ

σ − 1

)2 τ

ϕi(5)

where pWij is the price charged by wholesaler j selling manufacturing firm i’s good in

8

the foreign market.4

Proposition 2. A wholesaler imposes a double marginalization over the initialmarginal cost of the producer.

Proof. The marginal cost of wholesaler j consists of two parts. First, it pays aniceberg trade cost, τ , and, second, it pays the procurement price of the domesticmanufacturing good (from manufacturer i). Since a monopolistically competitivemanufacturer does not necessarily charge the same price to final consumers andwholesaler firms, the price manufacturer i charges wholesaler j is for now denoted bypij,P . The wholesaler’s marginal cost, MCW

ij of procuring and shipping manufactureri’s good is then equal to

MCWij = τpij,P .

The wholesaler faces the demand AF (pWij )−σ in the foreign economy where pWij isthe price that wholesale firm j charges for manufacturing firm i’s good in Foreign.Faced with a CES type of demand, it will charge a constant markup over its marginalcost:

pWij = σ

σ − 1MCWij

= σ

σ − 1τpij,P .

The demand for good i sold by wholesaler j in Foreign will be equal to xij =AF

(σσ−1τpij,P

)−σand wholesaler j’s cost function is therefore:

CWij (pij,P , xi) = τpij,Pxij.

4It could be argued that imposing a double marginalization is a strong assumption. There aretwo aspects to this that makes me believe that this is not the case. First, it is an endogenousoutcome of a CES setting which is standard in the literature: wholesalers are free to set their priceand choose a CES markup when doing so since this is profit maximizing. Since the demand oftheir consumers is characterized by CES preferences, this causes their input demand (as derived byShepard’s lemma from the cost function of wholesalers) to be characterized by CES preferencesas well, which motivates the markup choice by manufacturing firms. Second, even if wholesalersand manufacturers are allowed to negotiate, there are many reasons to believe that this contractin reality would not be completely non-distortionary and that there would indeed be some extramargin component which is paid to the wholesaler. But even so, in order for wholesalers to generatenon-negative profits, they must charge the manufacturer (whose product it is buying) some strictlypositive price for their services in order to cover their fixed costs. This additional price will make itmore profitable for the manufacturer to export on its own if it can. Finally, the empirical factsthat wholesalers and manufacturers co-exist in most export markets and that the most productivemanufacturers choose to export on their own, as in McCann (forthcoming) and Abel (forthcoming),suggest that there is some reason for manufacturers to avoid using intermediation services if theycan.

9

Applying Shephard’s lemma to find wholesaler j’s demand function for good i(i.e. the demand function that manufacturer i faces from wholesaler j), yields:

DWij (pij,P ) =

∂CWij (pij,P )∂pij,P

= τxij = τ 1−σAF(

σ

σ − 1pij,P)−σ

.

This result has two important implications. First, the producer firm faces theexact same demand elasticity from wholesale firms as that from domestic consumers,and will therefore charge the same price to wholesalers as it does to domesticconsumers (a constant markup over its marginal cost), pij,P = pi. Second, it canbe seen that the wholesaler will charge the following price in the foreign economy(foreign consumers have CES demand and the wholesaler will charge a standard CESmarkup over its marginal cost):

pWij =(

σ

σ − 1

)2 τ

ϕ.

The term(

σσ−1

)2decreases in the elasticity of substitution which means that

the degree of double marginalization is smaller in sectors where varieties are moresubstitutable.

Proposition 3. The double marginalization is higher in less competitive sectors, i.e.in sectors characterized by low elasticities of substitution.

Proof. See proof to Proposition 2.

Since manufacturers can choose their export mode (i.e. by establishing their owndistribution system or exporting through a wholesaler), their choices are determinedby what mode yields the highest profits from exporting. The expected profits(discounted by the forced exit rate δ) of a manufacturing firm i that exports throughwholesaler j will be

1δ

(τpij,Pxij

(pWij)− τ

ϕixij

(pWij))

= 1δτ 1−σϕσ−1

i AF(σ − 1σ

)2σ 1σ − 1 (6)

where xij(pWij)indicates foreign sales of good i at price pWij , i.e. the price set by the

wholesaler. The expected discounted profits of a manufacturing firm exporting onits own would be

1δ

(τpixi (τpi)−

τ

ϕixi (τpi)

)− FX = 1

δτ 1−σϕσ−1

i AF(σ − 1σ

)σ 1σ − 1 − FX .

(7)

10

Comparing the profits for a manufacturing firm choosing between the two exportmodes yields the following condition, using (6) and (7), for the firm to choose toexport on its own:

ϕσ−1i >

FXτ 1−σAF

1(1−

(σ−1σ

)σ) . (8)

This means that more productive firms will want to export on their own ratherthan through a wholesaler. This is due to the fact that they are productive enoughto take the fixed cost of exporting themselves and avoid the markup incurred onthem by the wholesaler. Note also that (8) defines the new export cut-off

ϕσ−1X = FX

τ 1−σAF1(

1−(σ−1σ

)σ) > FXτ 1−σAF

= ϕ′σ−1X .

The inequality sign demonstrates that, with wholesalers present, some producersthat previously exported on their own now decide to use wholesalers instead. There-fore the export cut-off is higher with wholesalers in the model than without. Thisphenomenon is illustrated in figure 1. The lines πX and πW show the operating profitsof manufacturing firm exporting on its own and exporting through a wholesaler,respectively, as functions of their productivity levels. The line πW starts from zerosince a manufacturing firm does not have to pay any fixed costs when exportingthrough a wholesaler. However, exporting directly requires paying a fixed cost FX .The slope of πW is lower than that of πX due to the additional markup charged bythe wholesaler. A manufacturer with productivity higher than ϕX will always havehigher profits from exporting on its own (πX > πW if and only if ϕ > ϕX).

Supposing that wholesalers only find it profitable to buy goods with a productivityhigher than ϕW means that firms with productivity levels between ϕW and ϕX willprefer to export through a wholesaler rather than on their own. Also, the exportcut-off in an economy without wholesalers, ϕ′X , will always lie to the left of ϕX ,which can be seen in the graph. Note also that since wholesaler prices are higher,they also export smaller volumes per good. The reason that their prices are higher isdue to (i) their additional markup and (ii) the fact that they export goods producedby less productive firms.

Finally, equation (8) shows that the degree of competition (the elasticity ofsubstitution across varieties) in a sector affects the productivity cutoff for directexporting since it determines the impact of the double marginalization imposed bywholesalers. Less competition (i.e. a smaller σ) means that the cutoff for directexporting is lower since the wholesaler’s additional margin is higher which makes it

11

relatively more expensive to use.

Proposition 4. The model generates productivity sorting as regards choice of exportmode. The most productive firms, ϕ > ϕX , export their products on their own, firmswith intermediate productivity levels, ϕ ∈ [ϕW , ϕX), export through wholesalers andthe least productive firms, ϕ ∈ [ϕD, ϕW ), do not export. That ϕD < ϕW has, however,to be assumed.

Proposition 5. Export sales per good are lower for wholesalers than for producersexporting on their own.

Proposition 6. Sectors characterized by less competition (low elasticity of substitu-tion) have a lower export cutoff due to the stronger effect of the double marginalizationimposed by wholesalers.

Proof. See appendix A.1.

Wholesale firms are homogeneous and I make the simplifying assumption thatthe atomistic manufacturing firms that use wholesalers for the distribution of theirgoods are randomly matched with wholesaler firms (see figure 2). This ensures thatwholesaler firms in equilibrium will, on average, have identical baskets of goods thatthey export. They will therefore have the same number of products and also thesame productivity distribution among the goods in their baskets.5 The scope ofgoods that wholesaler firms handle will be equal to the mass of manufacturing firmsthat use wholesalers for exporting (i.e. those with a level of productivity betweenϕW and ϕX). The mass of manufacturers in this range is MM G(ϕX)−G(ϕW )

1−G(ϕD) whereMM is the mass of manufacturing firms in total. The scope per wholesaler is thenequal to this expression divided by the number of wholesale firms, nW :

mWj = MM

nWG (ϕX)−G (ϕW )

1−G (ϕD) . (9)

5Wholesalers control more than one good and are therefore no longer atomistic like the manufac-turing firms. This could potentially have implications for how they affect prices, but it is assumedhere that parameters are such that these baskets of atomistic products are still small enough not tohave an effect on the aggregate price level. In a sense, a wholesaler could be labelled “moleculistic”,i.e. larger than an “atomistic” firm but still small enough not to have an effect on the aggregateprice level. In this context, it can also be noted that in the seminal contributions by, for example,Dixit and Stiglitz (1977) and Krugman (1979, 1980), the distribution of differentiated varieties arediscrete and not continuous such as in the more recent trade literature. With a discrete and finiteset of varieties, all manufacturing firms are theoretically able to affect the aggregate price index soit has to be assumed that they do not.

12

The assumption of random matching between manufacturing firms and wholesalersmay appear to be strong, especially given the convexity in the fixed cost functionof wholesalers which indicates that wholesalers prefer to export products which arecloser to each other in nature. However, the results would hold even if manufacturingfirms were matched only within, for example, broader sector categories as long aswholesalers are homogenous in terms of costs. Moreover, the key mechanism inthe model, that wholesalers face a lower fixed cost per product, would hold evenif manufacturers were matched less randomly. One example would be that somewholesalers are matched with more productive manufacturers than others. Thesewholesalers with a portfolio of more productively produced products would, however,still demand a markup from the producers, σ/ (σ − 1) under CES preferences, andtherefore deter the most productive exporters but encourage manufacturers from theintermediate productivity range. Finally, even if wholesalers were assumed to beheterogeneous in terms of their cost function, and even if this would be correlatedwith the markup they choose to set, this key difference between the two exportmodes would prevail due to the fact that all wholesalers, regardless of costs, mustcharge some markup for each good in order for them to finance the extra cost itmeans for them to add an additional product to their existing scope.

The total fixed cost of a wholesaler, as specified in equation (4) can be written:

FWXj = FX +

(mWj

)γ

γ

= FX + 1γ

(MM

nWG(ϕX)−G(ϕW )

1−G(ϕD)

)γ

where ϕW is the marginal productivity of the least productive manufacturing firmthat exports through wholesalers.

A wholesaler takes as given the number of other wholesale firms and the mass ofdomestic manufacturing firms and its pricing mechanism is, as described in equation(5), a constant markup over the marginal cost. Therefore the number of wholesalefirms and the range of goods they consume can be determined by two conditions.First, the free entry condition for wholesalers states that the profits of wholesalefirms should be zero. Second, an optimal scope condition by wholesalers, i.e. thatthe marginal increase in operating profits for a wholesaler firm to expand its setof goods distributed (its scope) must equal the resulting marginal increase in fixed

13

costs, specifies the scope of each wholesale firm.

FX +

(mWj

)γγ

= mWj π

Wj (Zero profit condition)

∂

∂mWj

FX +

(mWj

)γγ

= ∂

∂mWj

(mWj π

Wj

)(Optimal scope)

These two conditions jointly determine the mass of manufacturing firms exportingthrough each wholesaler, mW

j , and the weighted average of profits per good handled,πWj ,

mWj = F

1γ

X

(γ

γ − 1

) 1γ

(10)

πWj = Fγ−1γ

X

(γ

γ − 1

) γ−1γ

. (11)

The fixed cost of exporting, FX , is the key variable in understanding how the sizeof wholesaler firms is determined. A higher fixed cost of exporting forces wholesalefirms to expand their scope so that the fixed cost is spread across more goods. Italso makes the operating profit per good handled to be larger in equilibrium. Theparameter determining how difficult it is for wholesalers to handle more goods, γ,also plays an important role. The elasticity of the optimal scope with respect to thefixed cost in equation (10) is 1

γwhich decreases in γ; the more difficult it is to handle

many goods, the less responsive is the scope of wholesalers to fixed costs.

Proposition 7. The optimal scope of wholesalers increases in the size of the fixedcost of exporting.

Proposition 8. The elasticity of the optimal scope of wholesalers with respect tothe fixed cost of exporting is lower when it is more difficult for wholesalers to expandtheir scope (when γ is high).

To close the equilibrium, I note that the operating profit of wholesaler j sellinggood i is:

πWij = pWij xij − pij,P τxij = AF τ 1−σϕσ−1i

(σ

σ − 1

)1−2σ 1

σ − 1 .

14

Therefore, the total operating profit of wholesale firm j is

mWj π

Wj (ϕW , ϕX) =

mWj τ

1−σAF

G (ϕX)−G (ϕW )

(σ

σ − 1

)1−2σ 1

σ − 1

ˆ ϕX

ϕW

ϕσ−1dG (ϕ)

(12)

where πWj (ϕW , ϕX) is the average operating profit per good handled given the rangeof productivity in the basket.

Combining (11) and (12) gives:

τ 1−σAF

G (ϕX)−G (ϕW )

(σ

σ − 1

)1−2σ 1

σ − 1

ˆ ϕX

ϕW

ϕσ−1dG (ϕ) = Fγ−1γ

X

(γ

γ − 1

) γ−1γ

(13)

where ϕW is the equilibrium level of the lowest productivity needed for a manufactur-ing firm to use a wholesaler firm to export.6 The export cut-off, ϕX , is determinedaccording to (3) by τ , FX and AF . The variable and fixed trade costs are exogenousbut AF is endogenous. Since the left side of (13) is monotonically increasing in ϕW ,equation (13) yields an implicit solution for ϕW as a function of AF .

Using the equilibrium value for ϕW , it is also possible to find a solution for thenumber of wholesale firms by combining (9) and (10):

MM

nW

(G (ϕX)−G (ϕW )

1−G(ϕD)

)= F

1γ

X

(γ

γ − 1

) 1γ

. (14)

Finally, the free entry condition for manufacturing firms says that, in expectation,the expected total profit of entrepreneur must equal the fixed entry cost:

ˆ ∞ϕD

(ϕσ−1A− FD

)dG (ϕ) +

ˆ ϕX

ϕW

ϕσ−1τ 1−σAF(σ − 1σ

)σdG (ϕ) (15)

+ˆ ∞ϕX

(ϕσ−1τ 1−σAF − FX

)dG (ϕ) = FE.

where FE ≡ δ (σ − 1)1−σ σσFE.The set of equations (1), (2), (8), (13), (14) and (15) yield implicit solutions for the

productivity cutoffs ϕD, ϕW , ϕX and the mass of wholesale and manufacturing firms,6There is nothing in the model that ensures that ϕW > ϕD (that manufacturers exporting

through wholesalers need to be more productive than manufacturers producing for the domesticmarket). This has to be assumed but this restriction is similar to the assumption in the literaturethat ϕX > ϕD (that exporters are more productive than non-exporters). This does, however, notmatter for the final results.

15

nW , MM . The “per-firm demand” A in Home and AF in Foreign are determined bythe mass of firms, number of wholesalers together with the productivity cutoff levelsthrough the expression for the price levels. This set of equations therefore defines ageneral equilibrium in the sense that there is free entry in all sectors, and that pricelevels and market demand are endogenous.

2.3 Imposing the Pareto distribution

To find exact expressions for the importance of wholesalers in the economy, theexact distribution of productivity, G (ϕ) has to be specified. I therefore impose thescale-free Pareto distribution. which has been found to correspond reasonably wellwith observed distributions of firm productivity, see Axtell (2001) or Luttmer (2007).Now

G (ϕ) = 1− ϕ−k

where ϕ ∈ [1,∞). For solutions to exist it is also required that β ≡ kσ−1 > 1.

To calculate the relative export volumes that occur by firms exporting on theirown versus through wholesalers, it can be noted that the export volume of a goodthrough the two export modes is:

Vi,X (ϕi) = pixi =(

σ

σ − 1

)1−στ 1−σϕσ−1

i AF (direct exporting)

Vj,W (ϕi) = pWij xWij =

(σ

σ − 1

)2(1−σ)τ 1−σϕσ−1

i AF (wholesale).

The ratio of total export volumes will therefore be

VWVX

=´ ϕXϕW

VW (ϕ) dG (ϕ)´∞ϕXVX (ϕ) dG (ϕ) =

(σ − 1σ

)σ−1(ϕX

ϕW

)k−(σ−1)

− 1 , (16)

which is an explicit function of the relative productivity cut-off levels ϕX and ϕW .The relative mass of firms exporting on their own versus through wholesalers versuscan be written

´ ϕXϕW

dG (ϕ)´∞ϕXdG (ϕ) =

(ϕXϕW

)k− 1 (17)

which is also an explicit function of the relative productivity cutoffs.To see what affects the relative importance of wholesalers in total export volumes,

16

equation (16), and scope, equation (17), it is necessary to understand what affectsthe relative productivity cut-off, ϕX

ϕW. An explicit solution for ϕW cannot be found

but by evaluating equation (13), the following nonlinear relationship can be found:

(ϕXϕW

)k− 1(

ϕXϕW

)k−(σ−1)− 1

= λ1FX(θ, wFLF

) 1γ (18)

where λ1 is a constant of parameters.7 The expression shows that the relativeproductivity cut-off increases in the fixed cost of exporting where I now, again,include country-specific characteristics as mentioned in the beginning of this section.

First, a higher fixed cost causes the relative productivity cut-off to increase. This,therefore, causes: (i) more firms to export through wholesalers (which is equivalent tomore varieties or products being exported through wholesalers) and (ii) the relativeexport volume that is managed by wholesalers to increase. This result originatesin the central mechanism provided by the model: the wholesale industry pools theexport fixed costs across goods and therefore reduces the fixed cost per good, afeature which is more important when fixed costs are large. We saw previously that ahigher fixed cost causes wholesale firms to expand the set of goods that they handle.By doing so, the fixed cost per good decreases. Second, neither the variable tradecost, τ , or per firm demand, AF , play any direct role for the choice of export mode.This is due to the fact that for the operating profit, these variables affect wholesalersand direct exporters in identical ways (the only way in which they can possiblyaffect the relative productivity cut-off is through their effect on the fixed cost).8 Thewholesale technology therefore exhibits an increasing returns to scale property withregard to product scope. An increase in the number of products that a wholesalerexports lowers the fixed cost per good, making wholesalers more important as fixedcosts increase.

The net effect of the elasticity of substitution, however, is unclear. As we sawin Proposition 6, more competition (higher σ) causes firms at the margin betweenchoosing direct or wholesale exporting to opt for exporting through wholesalers.This should make σ increase the role played by wholesalers. However, a higher σ

7 λ1 ≡ kk−(σ−1)

(γ−1γ

) γ−1γ 1

1−(σ−1σ )σ

1σ

(σ−1σ

)2(σ−1).8Distance appears to significantly affect the importance of intermediaries in activities relating to

foreign trade in many empirical studies. Indeed, the empirical analysis in this paper shows thatwholesalers are more important in distant markets. While contrary to the results of this paper, arange of explanations can be argued to give this pattern. An obvious one is that the fixed cost ofentering a foreign market increases with distance, since cultural barriers, supervision costs etc. arelarger in markets far away than in those close to the home country of a firm. This would be in linewith the theoretical predictions presented here.

17

also makes price differences more important for revenues and profitability. Whenthe elasticity of substitution is high, the additional markup imposed by wholesalersbecomes more important for how much is actually sold and this has a negative effecton the aggregate role played by wholesalers in exporting.

The conclusions above can be summarized in the following two propositions.

Proposition 9. A higher fixed cost is associated with (i) a higher share of totalexports being shipped by wholesalers and (ii) a larger number of firms exportingthrough wholesalers relative to exporting on their own. This is due to the fact thatwholesalers spread the fixed cost of exporting across more goods.

Proof. The following is a proof that

∂

∂x

(xk − 1

xk−(σ−1) − 1

)> 0 if x > 1 and k > σ − 1 > 0

which corresponds to equation (18) where x would be ϕXϕW

.First, the condition for the derivative to be positive can be simplified to

xk(σ − 1− kxσv−1−k) + k − (σ − 1) > 0.

Now, consider the factors of the first term. xk is minimized when x = 1 andσ − 1− kxσ−1−k is minimized when x = 1 (recall that x > 1 and k > σ − 1). Whenx= 1, the expression on the left hand side is equal to 0. However, as x increases,both xk and σ − 1 − kxσ−1−k increase, meaning that the whole expression on theleft hand side will increase. Therefore, the condition holds since x has to be strictlygreater than 1.

Proposition 10. The net effect of the elasticity of substitution on the aggregateimportance of wholesalers in exporting is uncertain.

Proof. See appendix A.2.

Finally, an analysis of a multi-country setting of the model is conducted inAppendix section A.3. This adds two main insights in addition to what has beendescribed so far. First, the ratio of the productivity cutoffs, ϕJX

ϕJW, to each export

destination J depends on country J ’s fixed cost of entry. Since such costs differacross countries, many manufacturing firms would export directly to some countriesbut through a wholesaler to other countries. It also depends on each country’s“multilateral resistance” (as proposed by proposed by Anderson and van Wincoop,2003) which takes into account the industry structure in the destination but also

18

the level of exports from other countries to that specific market. In this case, theequilibrium would contain firms that use both export modes at the same time but not,however, both export modes to the same country. Second, I find that the presence ofwholesalers contribute positively to welfare by assisting firms at intermediate levelsof productivity to overcome fixed costs of exporting. Wholesalers therefore increasethe range of products available for consumption in each country and they thereforealso reduce price indices and increase real wages.

3 Empirical evidence

3.1 Data

I use a dataset from Statistics Sweden which contains information on the economicactivities of the universe of Swedish firms in 2005 (a firm is defined as the legal unit).The data is collected by the Swedish Tax Authority (Skatteverket) and containsinformation on, for example, total annual revenues, number of employees (reportedonce every year at a certain point in time), total fixed assets. A firm in this dataset isclassified according to its main business activity, and my analysis utilizes firms thatare listed as “wholesalers” and firms active in any of the manufacturing sectors.9 Theonly restriction I make is to exclude firms with no employees. Because misreportingis prosecuted and these data are subjected to quality controls by statisticians atStatistics Sweden, measurement errors are most likely rare.

Through a common firm identifier, I match this data with a trade dataset collectedby Swedish Customs (Tullverket) which records all trade flows per firm, productcode (according to the Combined Nomenclature, CN, up to 8 digits) and destinationcountry. The CN is a classification system used by the European Commission in itsexternal statistics and it is based on the Harmonized System of product classification.I use only exporting firms (about 35% of both manufacturers and wholesalers export)and in 2005 there were 7,248 wholesaler firms exporting and 8,768 manufacturingfirms. In total 468,710 transactions are reported over 8,272 CN8 categories and 194destination countries. I only conduct a cross-sectional analysis in this study since apanel study with several years would only use variation in the definition of a firm’smain business activity (which would not be very informative) or in the destinationcountry characteristics and this variation is very small.

Due to the extreme detail of the CN8 classification, I use the number of CN8categories exported by a firm as a proxy for the number of products exported by a firm.As stated, I observe more than 8,000 product categories in the data. An example

9Confidentiality requirements prevent me from giving specific examples of how firms are classified.

19

of its level of detail is the subcategory CN code 6601 containing umbrellas. Theseare divided into “Garden or similar umbrellas” and “Other umbrellas”. The lattercategory is divided into umbrellas with a “telescopic shaft” and those without. Theseare then divided into umbrellas with a “cover of woven textiles” and those without.It therefore seems reasonable to view a CN8 category as a manufacturing variety (ora product, the two are equivalent in the model) in the analysis. Moreover, I interpretthe value of exports observed in my data as the export volume of firms. Of course,differences in pricing behavior may then affect my results but it is impossible todifferentiate between price differences due to differences in productivity, quality andmarkups (which are all likely to be correlated with each other and also likely to affectprices). The average number of CN8 product categories that a firm exported was 11for wholesalers and 9 for manufacturers. However, manufacturers were much largerand accounted for 86% of aggregate export volumes (measured in SEK). As a referencepoint, 1 US dollar was worth between 6.9 and 8 SEK in 2005. As for market size(GDP) and the institutional variables used, all data comes from the World Bank’sWorld Development Indicators (WDI) and Doing Business databases. Distancemeasures are from Centre d’Etudes Prospectives et d’Informations Internationales(CEPII). A full list and description of each variable can be found in the AppendixTable A1.

Table 1 reports descriptive statistics for exporting wholesalers and manufacturers.The main conclusion from this table is that manufacturing firms are much larger thanwholesale firms on average, both in terms of turnover, number of workers and capitalemployed. Moreover, despite the smaller size of wholesalers, they export a largernumber of products than manufacturers. This difference between manufacturingfirms and wholesale firms in the sense that manufacturing firms tend to be largerin terms of export volumes while wholesalers, instead, export a larger scope canalso be seen in Figure 3 which shows how wholesalers’ share of export volumes andexport scope, respectively, are distributed across destinations. It can be seen thatthe distribution of the share of the scope controlled by wholesalers clearly lies to theright of the distribution of the share of export volumes controlled by wholesalers.

3.2 Main assumption and predictions

This section will assess empirically the main predictions of the model:

1. Wholesalers export more products than manufacturers.

2. Export sales per good are lower for wholesalers than for producers exportingon their own.

20

3. A larger share of aggregate export volumes is handled by wholesalers to countrieswith high fixed costs of entry.

4. A larger share of the number of exported products is handled by wholesalersto countries with high fixed costs of entry.

5. Producers sort according to productivity when choosing export mode.

The underlying assumption of the model is that the wholesale technology ischaracterized by economies of scope, while that of manufacturers is not. This is asomewhat stylized assumption used in order to focus on the mechanisms linked to acase in which wholesalers have more economies of scope compared to manufacturers(that wholesalers have a comparative advantage in generating economies of scopein exporting). We already know from a vast literature that manufactures often sellmore than one product and that firms doing so are often important in terms of sizeand export volumes.10 However, one way to see whether the assumption of this paperis reasonable is to look at whether wholesalers on average sell more products thanmanufacturers, both generally and when controlling for different characteristics ofthese firms such as sector, export destinations and size. This is the first analysisperformed in the empirical section.

The second prediction originates from the fact that goods sold abroad throughwholesalers are produced in a less productive way than other export goods and thefact that wholesalers charge an additional markup. The following two predictions,however, are both related to the core mechanism highlighted in the theoretical section:when fixed costs increase, the ability of wholesalers to generate economies of scope byspreading the fixed cost of entry across more goods becomes more valuable. Variationin both variables listed is therefore driven by variation in fixed costs. Measures offixed costs of entry are always imperfect but a reasonable proxy would be measuresof institutional quality which relate to trade so these are used in the analysis.

Finally, one important prediction remains which I cannot test with the datacurrently at hand: that manufacturing firms sort according to productivity as regardsexport mode (Result 4). Due to the fact that intermediated exporting, as opposedto direct exporting, entails a lower fixed cost but a higher variable cost, it followsthat, in a framework as in Melitz (2003), the most productive firms will choose toexport on their own while firms with intermediate productivity will choose to exportthrough a wholesaler. Since I cannot observe sales by manufacturers to wholesalersinside Sweden, I do not test the prediction of productivity sorting and export mode.

10See, for example, Bernard et al. (2010b).

21

However, evidence by, for example, McCann (forthcoming) and Abel (forthcoming),build on firm survey data from Eastern Europe and Turkey, respectively. Interestingly,this data includes mode of export and both authors find robust evidence that thefirms which choose direct exporting are more productive than those which chooseindirect exporting.

3.3 Economies of scope

As stated, the underlying assumption of the paper is that the wholesale technology ischaracterized by economies of scope: wholesale firms are able to distribute more thanone product internationally. We know, however, that many manufacturers in realitydistribute more than one good but the theoretical section is set up such that theeffect of a difference between wholesalers and manufacturers is seen in its simplestform. One way to test this assumption in the data, however, would be to see whetherwholesalers, on average, export more products controlling for firm size, destinationand sector effects. I do this by running the following OLS regression:

scopeisc = α + βWi + γsizei + fs + fc + εijl

where scopeisc denotes the number of CN8 product codes exported by firm i tocountry c in sector s. By sector I refer to a wider range of products and in thiscase CN4 categories. Wi takes the value 1 if the product is sold by a wholesale firmand 0 if it is sold by a manufacturer. The variable sizei denotes the logarithm ofrevenues of firm i and is included since it is well known that larger firms tend toproduce and export a wider scope of products. To account for characteristics ofspecific product categories and specific markets, fixed effects are included for sectors,fs, and destination countries, fc. Otherwise, if wholesalers export within differentcategories than manufacturers, or serve different markets, the coefficient might bebiased. Finally, it is quite possible that the errors are correlated within countriesand I therefore cluster standard errors at the country level.

The coefficient β measures how much larger the product scope per country isfor wholesalers compared to manufacturers. The assumption used in the paperwould correspond to β being positive. Table 2 reports the results. It can be seenthat estimated difference in scope is positive and significant in all columns, whichsupports the main assumption used in the theoretical section. When the size of firmsis not included, the difference is not too large and around 10%. But when I takeinto account that wholesalers are much smaller in size, on average, by including thelogarithm of firms’ revenues, the estimated difference is fairly large, around 0.443

22

(translating into a 56% difference). This is a fairly large number given that theaverage product scope of firms is around 10. I therefore conclude that there is animportant difference in terms of scope between wholesalers and manufacturing firms.

3.4 Export sales per product

This part examines whether sales per good are lower for wholesalers than manufac-turers. The theoretical motivation for this is that wholesalers handle goods thatare produced with lower productivity than manufacturers and are therefore moreexpensive. Moreover, wholesalers need to charge an extra markup which exportingmanufacturers do not. As noted, manufacturers account for 86% of aggregate exportswhich gives an indication that this prediction holds. However, to account for effectsthat are specific for products and destination countries, a regression analysis usingfixed effects for these variables is carried out. The regression equation is:

xipc = α + βWi + γsizei + fp + fc + εipc

where xipc is the logarithm of the value of exports of product p (at the CN8 level)by firm i to country c. The coefficient β therefore measures the difference in exportlevels per good between wholesalers and manufacturers. It is possible that the errorterms are correlated within countries and the standard errors reported here aretherefore clustered at the country level. It is, of course, possible, that the errorsare correlated at the product level too but since the product categories are of suchdetail (there are 8,272 CN8 categories) this is most surely of less worry than anycorrelation at the country level. Indeed, the standard errors are substantially smallerwhen they are clustered at the product level instead of the country level. Table 3lists the result of these regressions.

The regressions indicate that wholesalers, on average, sell between 33% and 54%less of a given product to a given country. The negative effect is significant at the1% level for all four combinations of fixed effects. The results therefore suggest thatthe prediction holds: wholesalers sell less per good than producers. The changein magnitude of the estimated difference, when I include fixed effects, is of somemagnitude (which means that there is some selection of firms, especially as regardsproduct categories). This effect, however, is not large and the results are qualitativelysimilar. The effect of including the size of firms does not influence the outcomeeither. This last result might seem surprising but is not if one remembers thatthis estimation is at the product level. For example, Bernard et al. (2011b), showhow larger firms export a larger portfolio and how the export volume per product

23

therefore might be uncorrelated with the firm’s size or productivity (although afirm’s total export volume across products is clearly correlated with a firm’s size andproductivity).

Finally, the results in Tables 2 and 3 are in line with the graphs in Figure 3. Thedistribution of wholesalers’ share of export volumes is much more skewed towardsthe left than the distribution of wholesalers’ share of export scope. This correspondswith the findings in the regressions that while wholesalers export less per product,they tend to export more products than manufacturing firms.

3.5 Importance of wholesalers and fixed costs in destination countries

Regarding predictions 2 to 3, the explanatory variable is the fixed cost of entry inboth cases. Therefore, these predictions will be examined in the same context in thissection. The theory states that a higher fixed cost increases the importance of thewholesalers’ ability of generating economies of scope. Wholesalers are able to spreadthe fixed cost across several products and are therefore better equipped to exportto markets where the fixed cost is high. They also control a larger share of exportvolumes and export scope to markets with higher fixed costs of entry.

Since observable measures of variation in fixed entry costs across countries do notexist, I use instead proxy variables. These variables have to fulfill two criteria: (i)they have to be relevant proxies for fixed costs in international trade, and (ii) theyhave to be available for a large enough number of countries. This leads me to usemeasures of institutions relating to international trade as proxy variables for fixedcosts of foreign market entry, and more specifically the following measures from theWorld Bank: (i) the time required to start a business, (ii) the number of proceduresneeded to open a warehouse, (iii) the time required to open a warehouse, and (iv) thecost, time and number of documents needed to import.11 These variables have beenused in other studies for this purpose and are often argued to be reasonable proxiesfor the type of fixed costs present in the model. This argument is based on theassumption that the costs of the bureaucracy, and possible corruption, involved wheninstitutions are poor are much more fixed than variable in nature, such as the costsof obtaining the initial permits to enter the market or constructing the buildingsand infrastructure needed for a viable distribution network. One may expect thesevariables to be strongly correlated with each other which would make it unnecessaryto use all of them in the analysis. However, as can be seen in Appendix Table A2,

11The first three variables (reported in Table 4) are collected from the World Bank’s WorldDevelopment Indicators and the last three (Table 5) are from the World Bank’s Doing BusinessSurvey. The last three variables were not available for 2005 so instead data from 2006 is used. Amore detailed description of these variables is given in Appendix Table A1.

24

these variables are far from perfectly correlated with each other which means thatthere is a certain degree of independent variation across countries in all of the sixvariables.

To see the effect that fixed costs have on the importance of wholesalers, thefollowing regressions are used:

(VW

VW + VX

)sc

= β0 + β1Qc + β2 log Yc + β3 log distc + fs + εi (19)(NW

NW +NX

)sc

= β0 + β1Qc + β2 log Yc + β3 log distc + fs + εi (20)

where(

VWVW+VX

)sc

denotes wholesalers’ share of aggregate export volumes in sector s(i.e. CN4 category) to destination country c,

(NW

NW+NX

)sc

the share of CN8 productshandled by wholesalers in sector s which are exported to country c, Qc the institutionalmeasure for country c (logarithmic values are used), Yc the level of GDP of countryc, distc the distance from Sweden to country c and fs a sector fixed effect. BothGDP and distance are in logarithms as well. Fixed effects are used here to controlfor technological differences in export mode across sectors. In this and the followingsection, only product and destination pairs where wholesalers are active are includedin the analysis, since this is the situation on which the model focuses. As for theerror terms, these are likely to be correlated both within countries as well as withinsectors. I therefore apply the multiway clustering method as described in Cameronet al. (2011) and cluster on both destination and sector.

Figures 4 and 5 give a graphical representation of equations (19) and (20). Foreach quintile of the six different proxy variables, I plot the mean across CN4 sectorsof the relative shares (i.e. the share for each CN4 category and destination minusthe mean share for each CN4 category across all countries:

(VW

VW+VX

)sc−(

VWVW+VX

)s)

of export volumes and export scope controlled by wholesalers. The graphs show aconsistent and positive relationship between the importance of wholesalers and thedifficulty of entering a foreign market. This relationship holds for all six measuresand for both export volumes and export scope.

Tables 4 and 5 report the regression results for equations (19) and (20). Consistentwith the graphical evidence, all the measures of institutional qualities have positivesigns in the regression and almost all are significant.12 If these variables are goodproxies of the fixed cost of entering a specific market, then the regression resultsmean that wholesalers are more important, both in quantity and scope, for entering

12If the standard errors are not clustered, all coefficients are significant at the one percentagesignificance level, indicating that the errors are to some extent correlated within countries andsectors.

25

markets characterized by large fixed costs. Distance appears to significantly affectthe importance of wholesalers and with a positive sign. Regarding the effect of GDP,I find that wholesalers account for a smaller share to larger countries. Analyzing theeffect of distance and GDP through the lens of my model, where they only affect theoutcome variable through the fixed cost, would indeed mean that a larger distance isassociated with a higher fixed cost of entry and that larger markets are associatedwith smaller fixed costs. As for distance, it is not unlikely that the fixed cost ofentering a foreign market does indeed increase with distance since cultural barriers,supervision costs etc. are presumably larger in markets far away than in those closeto the home country of a firm. As regards GDP, a number of studies (see for exampleEaton et al., 2004) have found that larger economies attract more exporters whichsuggest a smaller fixed cost of entry. Arkolakis (2010), for example, argues thatthis depends on decreasing returns to scale in the technology of market entry andthat exporters with a low productivity can enter a foreign market by just targetinga limited share of the consumers in that market. It might also be the case thatthe fact that many firms export to a larger market in some way causes “knowledgespillovers” among firms about how to enter that market which makes it relativelyeasier to enter a larger market. It is, for example, presumably easier for a U.S. firmto obtain information about how to enter Germany than the Netherlands.

4 Conclusion

The paper presents a model that gives a rationale for the presence of wholesalers andintermediaries in international trade. Wholesalers possess a technology that allowsthem to use their international distribution network to handle more than one good(although the fixed cost increases in the number of goods they handle), while theproducers of these goods (manufacturing firms) can only export their own good. Inorder to cover the fixed cost of exporting, wholesalers charge a markup between theprice at which they procure the good and the final price that they charge in theforeign country. This markup causes manufacturing firms to export on their ownas long as they are productive enough to cover the fixed cost of doing so. However,if they are not productive enough, they will instead try to sell their good to theforeign market through a wholesaler’s distribution network. This process resultsin productivity sorting in the choice of export mode: the most productive firmsexport on their own while less productive firms export through wholesalers. Theleast productive firms do not export at all.

When aggregated, this generates new results for what affects the importance

26

played by wholesalers in international trade, especially for the share of exports aswell as the share of products that are shipped through wholesalers’ networks. Thefixed cost of exporting plays an important role: the higher the fixed cost of exporting,the more important are wholesalers. This is because wholesalers can spread the fixedcost over several goods, as opposed to manufacturing firms which need to incur onefixed cost for their single good. As fixed costs become more important, firms shift toexporting through wholesalers (which expand the number of goods they handle) tobenefit from the lower fixed costs per good exported. Wholesalers therefore assistthose firms which are not productive enough to export on their own to do so.

This microeconomic foundation is also used to derive equilibrium price indices ina multi-country framework. It is found that the presence of wholesalers lowers priceindices in countries with which they trade. Therefore, wholesalers increase welfareand competition in these countries. Finally, the presence of wholesalers mitigatesthe effect of fixed costs on price levels.

The empirical section supports the main assumption and predictions of themodel. First, Swedish wholesalers export a broader range (56%) of products thanmanufacturers. Second, wholesalers export smaller amounts of each product, evenwhen the effect of which particular product it concerns or which country the goodis exported to is controlled for. Finally, wholesalers are more important, both inquantity and scope, for markets characterized by high fixed costs. The empiricalanalysis therefore supports the notion that wholesalers play a larger role in exportingto countries characterized by higher fixed costs of entry.

27

References

Abel, J. (2013). Who Uses Intermediaries in International Trade? Evidence fromFirm-level Survey Data. The World Economy, 36(8):1041–1064.

Ahn, J., Khandelwal, A. K., and Wei, S. (2011). The Role of Intermediaries inFacilitating Trade. Journal of International Economics, 84(1):73–85.

Anderson, J. E. and van Wincoop, E. (2003). Gravity with Gravitas: A Solution tothe Border Puzzle. American Economic Review, 93(1):170–192.

Antràs, P. and Costinot, A. (2010). Intermediation and economic integration.American Economic Review, Papers & Proceedings, 100(2):424–428.

Antràs, P. and Costinot, A. (2011). Intermediated trade. The Quarterly Journal ofEconomics, 126(3):1319–1374.

Arkolakis, C. (2010). Market Penetration Costs and the New Consumers Margin inInternational Trade. Journal of Political Economy, 118(6):1151 – 1199.

Axtell, R. L. (2001). Zipf Distribution of U.S. Firm Sizes. Science, 293:1818–1820.

Basker, E. and Van, P. H. (2011). Putting a Smiley Face on the Dragon: Wal-Martas Catalyst to U.S.-China Trade. Working paper.

Bernard, A. B., Grazzi, M., and Tomasi, C. (2011a). Intermediaries in internationaltrade: Direct versus indirect modes of export. NBER Working Paper 17711.

Bernard, A. B., Jensen, J. B., Redding, S. J., and Schott, P. K. (2010a). Wholesalersand Retailers in U.S. Trade (Long Version). NBER Working Paper 15660.

Bernard, A. B., Redding, S. J., and Schott, P. K. (2010b). Multiple-Product Firmsand Product Switching. American Economic Review, 100(1):70–97.

Bernard, A. B., Redding, S. J., and Schott, P. K. (2011b). Multi-product firms andtrade liberalization. Quarterly Journal of Economics, 126(3):1271–1318.

Biglaiser, G. (1993). Middlemen as Experts. RAND Journal of Economics, 24(2):212–223.

Blum, B. S., Claro, S., and Horstmann, I. J. (2010). Facts and Figures on Intermedi-ated Trade. American Economic Review, Papers & Proceedings, 100(2):419–423.

Blum, B. S., Claro, S., and Horstmann, I. J. (2011). Intermediation and the Natureof Trade Costs: Theory and Evidence. University of Toronto, mimeograph.

28

Cameron, A. C., Gelbach, J. B., and Miller, D. L. (2011). Robust Inference withMulti-Way Clustering. Journal of Business and Economic Statistics, 29(2):238–249.

Chaney, T. (2008). Distorted Gravity: The Intensive and Extensive Margins ofInternational Trade. American Economic Review, 98(4):1707–1721.

Crozet, M., Lalanne, G., and Poncet, S. (2013). Wholesalers in international trade.European Economic Review, 58:1–17.

Dixit, A. K. and Stiglitz, J. E. (1977). Monopolistic Competition and OptimumProduct Diversity. American Economic Review, 67(3):297–308.

Eaton, J., Kortum, S., and Kramarz, F. (2004). Dissecting Trade: Firms, Industries,and Export Destinations. American Economic Review, Papers & Proceedings,93:150–154.

Feenstra, R. and Hanson, G. (2004). Intermediaries in entrepôt trade: Hong Kongre-exports of Chinese goods. Journal of Economics & Management Strategy,13(1):3–35.

Felbermayr, G. and Jung, B. (2011). Trade Intermediation and the Organization ofExporters. Review of International Economics, 19(4):634–648.

Helpman, E., Melitz, M. J., and Rubinstein, Y. (2008). Estimating Trade Flows: Trad-ing Partners and Trading Volumes. Quarterly Journal of Economics, 123(2):441–487.

Helpman, E., Melitz, M. J., and Yeaple, S. R. (2004). Export versus FDI withHeterogeneous Firms. American Economic Review, 94(1):300–316.

Krugman, P. R. (1979). Increasing Returns, Monopolistic Competition, and Interna-tional Trade. Journal of International Economics, 9:469–479.

Krugman, P. R. (1980). Scale Economies, Product Differentiation, and the Patternof Trade. American Economic Review, 70(5):950–959.

Luttmer, E. G. J. (2007). Selection, Growth, and the Size Distribution of Firms.Quarterly Journal of Economics, 122(3):1103–1144.

Mayer, T. and Ottaviano, G. (2008). The Happy Few: the Internationalisation ofEuropean Firms. Intereconomics: Review of European Economic Policy, 43(3):135–148.

29

McCann, F. (forthcoming). Indirect exporters. Journal of Industry, Competition andTrade.

Melitz, M. J. (2003). The Impact of Trade on Intra-Industry Reallocations andAggregate Industry Productivity. Econometrica, 71(6):1695–1725.

Olsen, M. (2013). How Firms Overcome Weak International Contract Enforcement:Repeated Interaction, Collective Punishment, and Trade Finance. Working Paper.

Petropoulou, D. (2008). Information Costs, Networks and Intermediation in Interna-tional Trade. CEP Discussion Papers 0848.

Rauch, J. E. (1999). Networks versus Markets in International Trade. Journal ofInternational Economics, 48(1):7–35.

Rauch, J. E. (2001). Business and Social Networks in International Trade. Journalof Economic Literature, 39(December):1177–1203.

Rauch, J. E. and Watson, J. (2004). Network Intermediaries in International Trade.Journal of Economics and Management Strategy, 13(1):69–93.

Rubinstein, A. and Wolinsky, A. (1987). Middlemen. The Quarterly Journal ofEconomics, 102(3):581–593.

Tang, H. and Zhang, Y. (2011). Quality Differentiation and Trade Intermediation.mimeo.

30

5 Tables

Table 1. Descriptive statistics (only exporting firms included).

Wholesalers ManufacturersNumber of firms 7,248 8,768

Total exports (1,000 SEK)Mean 15,034 76,31910th perc. 7 9Median 238 62690th perc. 16,503 65,648

Product scope (CN8)Mean 11 910th perc. 1 1Median 3 390th perc. 22 19

Export destinationsMean 4 710th perc. 1 1Median 2 290th perc. 11 22

Capital intensity (%)Mean 179 34710th perc. 6 14Median 44 15790th perc. 424 736

Labor intensity(

employeesTSEK output

)Mean 1,302 1,99010th perc. 503 696Median 1,020 1,39690th perc. 2,198 3,341

Revenues (1,000 SEK)Mean 26,650 160,98610th perc. 995 2,279Median 5,779 16,68990th perc. 41,748 185,526

31

Table 2. Product scope.

Dep. variable: Product scope per firm and destination (in logarithms)(1) (2) (3)

Wholesale dummy 0.144*** 0.0972** 0.443***(0.0494) (0.0415) (0.0463)

Log(revenues) 0.216***(0.0099)

Fixed effects:Destination

√ √

Observations 96,130 96,130 96,130Note: The dependent variable is the logarithmic value of the number of CN8 products exportedby each firm. The explanatory variables are (i) a dummy for whether the firm is classified as awholesaler according to Statistics Sweden and (ii) the logarithmic value of firms’ revenues. Standarderrors are clustered at the country level and reported in parentheses. *** p<0.01, ** p<0.05, *p<0.1.

Table 3. Export volumes.

Dep. variable: Log(export volume)(1) (2) (3) (4) (5)

Wholesale dummy -0.685*** -0.398*** -0.768*** -0.518*** -0.571***(0.0560) (0.0401) (0.0560) (0.0413) (0.0700)

Log(revenues) -0.0254(0.0206)

Fixed effects:Product

√ √ √

Destination√ √ √

Observations 468,710 468,710 468,710 468,710 468,710Note: The dependent variable is the logarithmic form of the volume of exports of firms for eachproduct and destination (measured in 1,000 SEK). The explanatory variables are (i) a dummyfor whether the firm is classified as a wholesaler according to Statistics Sweden and (ii) thelogarithmic value of firms’ revenues. Standard errors are clustered at the country level and reportedin parenthesis. All regressions include fixed effects as specified. *** p<0.01, ** p<0.05, * p<0.1.

32

Table 4. Importance of wholesalers and destination country characteristics (a).

Dep. var.: Share of total exports handled Share of scope handledby wholesalers by wholesalers

(1) (2) (3) (4) (5) (6)

Time to start 0.0201** 0.0190***business (0.00966) (0.00665)Procedures to 0.0396** 0.0273***open a warehouse (0.0190) (0.0102)Time to open a 0.0213 0.0158warehouse (0.0142) (0.0100)

GDP -0.0558*** -0.0575*** -0.0573*** -0.0363*** -0.0381*** -0.0380***(0.00456) (0.00471) (0.00456) (0.00335) (0.00351) (0.00344)

Distance 0.0480*** 0.0493*** 0.0500*** 0.0274*** 0.0290*** 0.0295***(0.0101) (0.00977) (0.0102) (0.00636) (0.00653) (0.00709)

Fixed effectsSector

√ √ √ √ √ √

Observations 21,760 21,760 21,760 21,760 21,760 21,760R2 0.50 0.50 0.50 0.50 0.50 0.50

Note: The dependent variable is the relative importance of wholesalers relative to manufacturingfirms in exporting, both in terms of total export volumes as well as product scope. The explanatoryvariables are the logarithmic values of institutional measures from the World Bank as well as GDPand distance. All regressions are run at the sector (CN4) and destination level. Standard errors inparentheses. These are clustered on both destination and sector (CN4) according to the multiwayclustering method described in Cameron et al. (2011). All regressions include fixed effects at theCN4 sector level. *** p<0.01, ** p<0.05, * p<0.1.

33

Table 5. Importance of wholesalers and destination country characteristics (b).

Dep. var.: Share of total exports handled Share of scope handledby wholesalers by wholesalers

(1) (2) (3) (4) (5) (6)

Documents to 0.0791*** 0.0507***import (0.0196) (0.0133)Cost to import 0.0263 0.0323**

(0.0226) (0.0157)Time to import 0.0672*** 0.0464***

(0.0125) (0.00772)

GDP -0.0539*** -0.0565*** -0.0490*** -0.0353*** -0.0363*** -0.0317***(0.00453) (0.00490) (0.00411) (0.00333) (0.00351) (0.00310)

Distance 0.0345*** 0.0475*** 0.0306*** 0.0191*** 0.0258*** 0.0156**(0.00957) (0.0105) (0.00958) (0.00665) (0.00716) (0.00622)

Fixed effectsSector

√ √ √ √ √ √

Observations 21,688 21,688 21,688 21,688 21,688 21,688R2 0.50 0.50 0.51 0.51 0.50 0.51

Note: See Table 4.

34

6 Figures

Figure 1. Relative profits for different export modes. πX indicates the operatingprofit of a producer which exports on its own and πW indicates the operating profitof a producer exporting through a wholesaler.

35

Figure 2. Productivity sorting and export mode.

02

46

810

Density

0 .2 .4 .6 .8 1Share of total exports per market

01

23

4D

ensity

0 .2 .4 .6 .8Share of total scope per market

Note: Both graphs show the distribution across destinations of the share of export volumes/scopethat is controlled by wholesalers.

Figure 3. The role of wholesalers for aggregate export volumes and export scope.

36

−.0

50

.05

.1.1

5.2

Whole

sale

r share

of volu

mes (

res.)

1 2 3 4 5Time to start business

−.0

50

.05

.1.1

5.2

Whole

sale

r share

of volu

mes (

res.)

1 2 3 4 5Procedures to build warehouse

−.0

50

.05

.1.1

5.2

Whole

sale

r share

of volu

mes (

res.)

1 2 3 4 5Time to build warehouse

−.1

0.1

.2.3

Whole

sale

r share

of volu

mes (

res.)

1 2 3 4 5Documents to import

−.1

0.1

.2.3

Whole

sale

r share

of volu

mes (

res.)

1 2 3 4 5Cost to import

−.1

0.1

.2.3

Whole

sale

r share

of volu

mes (

res.)

1 2 3 4 5Time to import

Note: The horizontal axis shows quintiles of the six different proxy variables of the fixed cost ofentry. The horizontal axis shows the mean residual share (across CN4 categories and destinations)that wholesalers control of export volumes within each quintile. Residual shares are generated bysubtracting the actual share that wholesalers control for each CN4 category and destination by themean share for each CN4 category across all destinations (therefore similar to fixed effects as usedin the regressions).

Figure 4. Importance of wholesalers and destination country characteristics (exportvolumes).

37

−.0

50

.05

.1.1

5W

hole

sale

r share

of scope (

res.)

1 2 3 4 5Time to start business

−.0

50

.05

.1.1

5W

hole

sale

r share

of scope (

res.)

1 2 3 4 5Procedures to build warehouse

−.0

50

.05

.1.1

5W

hole

sale

r share

of scope (

res.)

1 2 3 4 5Time to build warehouse

−.0

50

.05

.1.1

5.2

Whole

sale

r share

of scope (

res.)

1 2 3 4 5Documents to import

−.0

50

.05

.1.1

5.2

Whole

sale

r share

of scope (

res.)

1 2 3 4 5Cost to import

−.1

0.1

.2W

hole

sale

r share

of scope (

res.)

1 2 3 4 5Time to import

Note: The horizontal axis shows quintiles of the six different proxy variables of the fixed cost ofentry. The horizontal axis shows the mean residual share (across CN4 categories and destinations)that wholesalers control of the export scope within each quintile. Residual shares are generated bysubtracting the actual share that wholesalers control by the mean share with each CN4 category(therefore similar to fixed effects as used in the regressions).

Figure 5. Importance of wholesalers and destination country characteristics (exportscope).

38

A Appendix (not for publication)

A.1 Proof to Proposition 6.

I define

v ≡ log(σ − 1σ

)σ= σ log

(σ − 1σ

).

This means that:

v′ = log σ − 1σ

+ 1σ − 1

v′′ = − 1σ (σ − 1)2 .

Therefore

lim σ→1+v = −∞

lim σ→∞v = lim σ→∞v′ = 0

where l’Hôpital’s rule is used in the second equality. To find the sign of v′, I notethat

v′′ = − 1σ (σ − 1)2 < 0.

The limit values of v′ are

lim σ→1+v′ = ∞

lim σ→∞v′ = 0.

Since we know that v´ decreases in σ at all possible values of σ (because v′′ < 0)and approaches zero when σ →∞, it must therefore always be strictly positive inthe space σ ∈ (1,∞). Therefore, we know that v always increases in σ. Furthermore,we know that

(σ−1σ

)σ∈ (0, 1) since limσ→1+ ev(σ) = 0, limσ→∞ e

v(σ) = 1 and v′ >

0 ∀σ ∈ (1,∞).

A.2 Proof to Proposition 10.

Equation (18) can be rewritten as

u

(ϕXϕW

, σ

)= θ(σ)v (σ)

where

u

(ϕXϕW

, σ

)≡

(ϕXϕW

)k− 1(

ϕXϕW

)k−(σ−1)− 1

v (σ) ≡ 1σ

(σ − 1σ

)2(σ−1)

θ (σ) ≡ 11−

(σ−1σ

)σ k

k − (σ − 1)F1γ

X .

Then, the total derivative of equation (18) can be formulated as

u ϕXϕW

dϕXϕW

+ uσdσ = (θσv + θvσ)dσ

d ϕXϕW

dσ= θσv + θvσ − uσ

u ϕXϕW

. (21)

We know that u ϕXϕW

is positive from the proof to Proposition 9. Then,

uσ =

(ϕXϕW

)k− 1((

ϕXϕW

)k−(σ−1)− 1

)2

(ϕXϕW

)k−(σ−1)

log(ϕXϕW

)> 0

since ϕXϕW

> 1 and k > (σ − 1).Turning to vσ, evaluating the sign of the derivative of log v (σ) instead of v(σ) is

equivalent since the logarithm is a monotonic transformation.

log v (σ) = − log σ + 2 (σ − 1) log(σ − 1σ

)∂

∂σlog v (σ) = 1

σ+ 2 log

(σ − 1σ

)∂2

∂σ2 log v (σ) = σ + 1σ2 (σ − 1) > 0

since σ > 1. Since the second derivative is always positive, the first derivative willhave its maximum when σ approaches infinity. The first derivative must thereforealways be negative since limσ→1+

∂∂σ

log v (σ) = −∞ and limσ→∞∂∂σ

log v (σ) = 0.Therefore, ∂

∂σlog v (σ) < 0 and also ∂

∂σv (σ) < 0 ∀σ ∈ (1,∞).

However, we also know that θσ > 0 since

∂

∂σ

1(1−

(σ−1σ

)σ) > 0

from the proof to Proposition 6. This means that the net effect in the numerator inequation (21) is unclear since θσ > 0 but vσ and −uσ are negative.

A.3 Gravity and welfare

The model outlined in section 2 gives predictions for when wholesale firms areimportant for aggregate trade flows (both in terms of export volumes and in termsof product scope). It also gives predictions for what firms select into which of thetwo export modes available. The advantage with the setup used so far is that itassumes free entry in both the manufacturing and wholesale industries and thereforegenerates a general equilibrium model. However, in this appendix section, I make animportant modification of the assumptions for two reasons. First, I wish to attainanalytically solvable expressions for welfare, which in this model is measured bythe price index. Second, I want to compare my model to a version of the model byMelitz (2003) that has become relatively standard in the literature, namely that ofChaney (2008). He estimates the effect of firm heterogeneity on the gravity modeland modifies the “multilateral resistance” variable proposed by Anderson and vanWincoop (2003).

This section therefore imposes the same assumptions for the manufacturing sectoras Chaney (2008). Most importantly, free entry in the manufacturing sector, asin equation (15), is restricted.13 Instead, there is a constant number of “potentialentrants” in country i, MM

i , which is proportional to the size of that economy:

MMi = wiLi.

This means that the expected profit of firms is no longer zero. Instead, profitsare collected into a global fund in which workers own shares according to theirincome. For comparability, I also assume that the productivity in the homogenous(agricultural) sector differs across countries which generates wage differences acrosscountries. I denote the wage level in country i by wi. The dividend per share in theglobal fund is equal to π and is the same across countries. Therefore, the outputlevel in country i is equal to

Yi = (1 + π)wiLi.13The reason for not choosing this restriction of entry already from the start is that the model

generates all the results until this point while allowing for free entry which is the most preferablesetup for a general equilibrium model.

The dividend, π, can therefore be written

π =

∑iwiLi

(´∞ϕiDπiD (ϕ) dG (ϕ) +∑

j 6=i

(´∞ϕijXπijX (ϕ) dG (ϕ) +

´ ϕijXϕijW

πijW (ϕ) dG (ϕ)))

∑iwiLi

where πijW (ϕ) indicates the profits of a manufacturing firm with productivity ϕ fromexporting through a wholesaler from country i to country j and πijX (ϕ) indicates thesame but for direct exporting instead. The remaining assumptions are as before andwholesalers are assumed to be present in all countries. This provides the equilibriumconditions for all endogenous variables: ϕiD, ϕ

ijX , ϕ

ijW , Y i, nWi , Pi ∀i, j and π. The

variable nWi denotes the number of wholesale firms in country i.Since nominal wages (measured in terms of the numeraire) are determined

exogenously by each country’s level of productivity in the homogenous sector, therelevant measure of welfare is the price index in the manufacturing sector weightedby the consumption share of manufacturing goods in the utility function: P−µi . Itturns out that the model generates an expression for the price level of manufacturinggoods which is very similar to that found by Chaney (2008):

Pi = λ2Y1k− 1σ−1

i θi

where

θ−ki ≡∑jYjYWw−kj τ−kji (ριijFji) 1−β

1− σγµ(γ−1)

∑j 6=i

nWjYiFji

(22)

and where ιij takes the value 1 if i 6= j and 0 otherwise.14

Two components of this result are especially important. First, the presence ofwholesalers in other countries, nWj > 0, lowers the price level in the Home country(country i above) which increases competition there and also raises welfare. Second,an increase in the fixed cost in a model without wholesalers increases the price level.However, when wholesalers are present, the positive effect of an increase in the fixedcost of entry from country j into country i is lower due to the presence of nWj Fji inthe denominator. Wholesalers therefore mitigate the effect of fixed costs on pricelevels.

Finally, the variable described in equation (22) is this model’s version of theso-called “multilateral resistance” term highlighted by Anderson and van Wincoop(2003), an aggregate index of the remoteness of country i. Chaney (2008) calculates

14ρ ≡ 11−(σ−1

σ )σ and λ2 ≡(

σσ−1

)1β µ

1k−

kσ−1

(β−1β

) 1k

Y− 1k

W (1 + π) 1k .

the multilateral resistance variable with his assumptions and finds:

θ−ki =∑j

YjYW

w−kj τ−kji F1−βji . (23)

A comparison of equations (22) and (23) shows that the expressions are virtuallyidentical except for the entry of ρ in the numerator and the effect of wholesale firmsin the denominator. Since nWj > 0, it is therefore the case that the presence ofwholesale firms lowers the price index in all countries and increases both welfare andcompetition as compared to the standard model in Chaney (2008). This effect isespecially strong in bilateral relationships characterized by high fixed costs of entry.Moreover, it can be seen that the presence of wholesale firms (higher nWj ) mitigatesthe effect of a higher fixed cost on the price level.

Proposition 11. The presence of wholesalers in the economies of trading partnerslowers a country’s price index and increases welfare and competition in this country.Moreover, the presence of wholesalers mitigates the effect of an increase in the fixedcost on the price index.

A.4 Data appendix

Table A1. Variable description.

Variable DescriptionFirm accounts Source: Statistics Sweden (originally the Swedish Tax Authority).

Revenues Total sales by a firm.Capital Value of total stock of fixed assets of a firm.Number ofemployees

Number of employees at each firm.

Sector 5-digit code classifying a firm’s main activity in year t according to theSNI system.

Trade variables Source: Statistics Sweden (originally the Swedish Customs).

Export volume The value of exported goods within each CN8 category and to eachdestination.

Destination The country of destination.Product code The CN8 (Combined Nomenclature 8 digits) product category.Fixed costs ofentry

Source: World Bank’s World Development Indicators and DoingBusiness databases. Definitions below are as listed in the World Bankdatabase.

Time to startbusiness

The number of calendar days needed to complete the procedures tolegally operate a business.

Procedures toopen a warehouse

The number of procedures to build a warehouse is the number ofinteractions of a firm’s employees or managers with external parties,including government agency staff, public inspectors, notaries, landregistry and cadastre staff, and technical experts apart from architectsand engineers.

Time to open awarehouse

The number of calendar days needed to complete the requiredprocedures for building a warehouse.

Documents toimport

The number of documents required for clearance by governmentministries, customs authorities, port and container terminal authorities,health and technical control agencies and banks in order to import.Since payment is by letter of credit, all documents required by banks forthe issuance or securing of a letter of credit are also taken into account.

Cost to import Cost measures the fees levied on a 20-foot container in U.S. dollars. Allthe fees associated with completing the procedures to export or importthe goods are included (but not tariffs or trade taxes).

Time to import The number of days required to import a good. The time calculation fora procedure starts from the moment it is initiated and runs until it iscompleted.

GDP Gross Domestic Product measured in US dollars.Distance variable Source: CEPII.

Distance The distance between two countries calculated with the great circleformula based on the geographic coordinates of the largestcities/agglomerations (in terms of population) of countries.

Tab

leA

2.Correlatio

ntableforvaria

bles

relatin

gto

fixed

costsof

entry.

Tim

eto

start

Procedures

toTim

eto

build

Docum

ents

Costto

Tim

eto

busin

ess

build

wareho

use

wareho

use

toim

port

impo

rtim

port

Tim

eto

startbu

siness

1.00

Procedures

tobu

ildwa

reho

use

0.10

1.00

Tim

eto

build

wareho

use

0.36

0.35

1.00

Docum

ents

toim

port

0.28

0.30

0.35

1.00

Costto

impo

rt0.20

0.07

0.25

0.36

1.00

Tim

eto

impo

rt0.46

0.25

0.44

0.73

0.62

1.00

Note:

The

logarit

hmic

valueof

allv

ariables

isused.