The Annals of Regional Science

, Volume 51, Issue 2, pp 495–513

The agglomeration of exporters by destination


    • School of Economic SciencesWashington State University
  • Katherine N. Schmeiser
    • Department of EconomicsMt. Holyoke College
Original Paper

DOI: 10.1007/s00168-012-0538-9

Cite this article as:
Cassey, A.J. & Schmeiser, K.N. Ann Reg Sci (2013) 51: 495. doi:10.1007/s00168-012-0538-9


Precise characterization of informational trade barriers is neither well documented nor understood. Using Russian customs data, we document that regional destination-specific export spillovers exist for developing countries, extending a result that was only known for developed countries. This result suggests behavior responding to a destination barrier. To account for this fact, we build on a monopolistic competition model of trade by postulating an externality in the international transaction of goods. We test the model’s prediction on region-level exports using Russian data and find improvement over gravity-type models without agglomeration. This finding has important development implications in that export policy that considers current trade partners may be more effective than policy that focuses only on the exporting country’s industries. Furthermore, our findings can be considered in the burgeoning literature refining transaction costs beyond the traditional iceberg cost.

JEL Classification


1 Introduction

One of the unsolved problems in international trade and economic development is the lack of knowledge about the nature of direct and indirect barriers to trade. Great theoretical and empirical strides have been made by country-level and firm-level studies, yet the precise characterization of barriers to trade continues to elude. In this paper, we show evidence consistent with a barrier to trade that firms respond to by exporting to the same markets as their neighbors.

Previous empirical studies find export spillovers that are destination specific. Examples are Lovely et al. (2005) using data from the United States, Koenig (2009) for France, and Choquette and Meinen (2011) for Denmark. We take the next step in this literature by documenting that these destination-specific export spillovers also exist in a developing country, Russia. This agglomeration is beyond the clustering of exporters around gross domestic product (GDP), regions that are physically close to trade partners, or domestic ports (Glejser et al. 1980; Nuadé and Matthee 2007). We document this destination-specific agglomeration using customs data from Russia at the regional level, confirming the previous work using a different method, and extend it to developing countries.

We then go beyond the current literature to develop a theory accounting for this destination-specific agglomeration. We assume there are economies of scale in the transaction costs of international trade at the firm level and test the sufficiency of this assumption. This differs from Krautheim (2012) who explores the theoretical implications of an exporter spillover in the fixed costs to trade. We acknowledge that other theories may be consistent with the regional destination-specific agglomeration of which we find evidence. See Cassey (2012) for an accounting of our data assuming regional comparative advantage in transportation technology to specific countries.

In principle, destination-specific export spillovers could work through the fixed or variable costs of trade. Econometric results by Choquette and Meinen (2011) provide evidence for export spillovers working through variable costs as well as fixed costs. Furthermore, Koenig et al. (2010) find evidence of a variable cost externality when the total number of exporters is small. In our data, the number of exporters selling to the same country from the same region in Russia is often small and thus is more likely to have the destination-specific export spillovers working through variable costs. We also consider an alternative measure of agglomeration beyond simple firm counts and that of the aggregate region–country weight of shipments.

The Russian customs data show that, by far, most shipments are small in weight. As long as shipping prices to exporters include a fixed cost per container (buying the container) or per destination (permit to unload) and are based on weight (Micco and Pérez 2002), then there are shipping cost savings in grouping small exports together on a boat or plane heading to the same destination. Alternatively, aggregate export weight may be thought of as the result of an informational spillover about foreign market access. The exact nature of the externality is not important for our model or results.

We find that using the region-level export data from Russia, our model improves upon the benchmark gravity equation. Therefore, we have uncovered empirical evidence that is consistent with there being a barrier to trade large enough for firms to respond to it by agglomerating, and unlike previous literature, we provide theoretic justification for it.

The transaction-level data that motivates our theory are from Russia. We choose to study Russian exporters because Russia is a developing country that is geographically large and diverse. It borders fourteen countries (with China and Ukraine being major trading partners) and has four coasts (Arctic Ocean, Pacific Ocean, Baltic Sea, and Black Sea) and diverse economic subregions. Finally, the laws in Russia limit the movement of firms regionally. This allows us to model a firm’s export destination decision without also modeling its location decision within Russia. Though Russia has a unique history that may affect development (relationships with former Soviet countries, corruption, and remnants of central planning to name a few), Cassey and Schmeiser (2013) show that the Russian data qualitatively and quantitatively match firm-level data sets from Colombia and France. Therefore, the development lessons from this work apply more generally than Russia alone.

Section 2 begins by describing the Russian export data in detail. Then we document that there is agglomeration by destination. This is the empirical motivation for the model developed in Sect. 3. We propose a theory that generates agglomeration by destination from an externality in the variable transaction costs of trade. We then test the consistency of our model in Sect. 4 and find that our theory is consistent with the data.

2 Facts on the location and agglomeration of Russian exporters

In this section, we show that (1) exporters are clustered regionally given output and the location of ports and (2) exporters shipping to the same destination are further clustered. The purpose of this section is to establish these two agglomeration facts for a developing country. We do so by using a different method than the literature. Koenig (2009) and Choquette and Meinen (2011) document the same agglomeration facts for developed countries using a logit estimator to see whether the odds that a firm exports to a particular country depends on the ratio of other firms in the same area also exporting to that country to total firms. Our method also differs from Lovely et al. (2005) who put an industry concentration variable on the left-hand side of a linear model with the difficulty in reaching an export destination on the right-hand side.

2.1 The Russian data

To document the existence of destination-specific agglomeration of exporters, we use 2003 data from the Russian External Economic Activities (REEA) set reported by Russian customs. The REEA data are shipment-level customs data of uniquely identified firms, providing value of goods shipped (in 2003 USD), weight of shipment, exporter location within one of 89 regions in Russia, and destination country. See Cassey and Schmeiser (2013) for complete description of these data, as well as descriptive statistics and stylized facts. We only consider manufacturing exports because firms that exported natural resources were necessarily located in the region endowed with those resources (Bradshaw 2008).1

To check for consistencies in the data, we refer to Cassey and Schmeiser (2013) who compare the REEA data with product data from the United Nations and find close agreement. They also compare the REEA data to statistics obtained with firm-level data from Colombia (Eaton et al. 2008) and France (Eaton et al. 2011). The export patterns are qualitatively and quantitatively similar. We are therefore confident that our data are not greatly influencing or biasing our interpretation or generality of results, in particular toward any geographic region.

We combine the REEA data with information about the gross domestic products (GDP) of each region using Russia: All Regions Trade and Investment Guide (CTEC Publishing 2004, 2006). (We do not have data on employment or the number or size of domestic firms.) We also use the World Economic Outlook Database (IMF 2006) for GDP information for the 175 countries in our sample. Finally, as is standard in the gravity literature, we calculate the great circle distance from the capital of each Russian region to the capital of each country in the world.2

The top destinations for Russian exports in 2003 were, in order, the following: China, United States, United Kingdom, Japan, Ukraine, Kazakhstan, Turkey, the Netherlands, Germany, and Iran. Of these, only Ukraine and Kazakstan were part of the Soviet Union. Additionally, Russia does not currently have a Warsaw Pact nation as a top destination. Because we do not see current evidence of favored trade relations with former Soviet or Eastern bloc countries, we do not treat them any differently than Western countries. (In a robust check reported later, we remove Warsaw Pact and Former Soviet Union countries from our analysis.)

Our data have three drawbacks that limit the forthcoming empirical analysis. First, though we have transaction-level data on shipments that we can match to unique firms, we are not able to pinpoint exporter location by exact physical location or longitude and latitude. The best we can do is assign each exporter to one of Russia’s 89 federal regions.3 Second, we only have the customs data, and thus, we are not able to match Russian firms with their characteristics. We neither know the employment, age, or total sales of the firms in our data nor do we know anything about firms that are not exporting. Therefore, we cannot compare the clustering of Russian manufacturing exporters to that of the Russian manufacturing industry or subsectors therein, nor systematically exclude other sources of firm clustering such as natural, labor, or infrastructural advantages. Given this limitation, we instead show that Russian firms exporting to the same country are nonrandomly located within Russian regions, controlling for variables such as distance, importer GDP, and preferences of the importer for particular goods. Third, we only have one year of the Russian customs data. Therefore, we cannot study dynamics from entry and exit of firms into various export markets or entry and exit of existence.

2.2 The geography of Russian exporters

In this section, we document exporter clustering by destination within Russian regions that will then be motivation for a model in Sect. 3. We document this agglomeration-by-destination fact for a developing country using an additional method beyond the logit estimator used in the literature.

To show that exporters are clustered, we first use the customs data to calculate the number of firms in each region that export anywhere in the world. Next, we weight the exporter count observations by regional GDP. We compare each region’s exporter count to the overall mean. Figure 1 highlights regions with a concentration of exporters far from the mean as well as the location of major ports. Darker colors indicate regions whose exporter count, weighted by GDP, is much larger than the mean. As seen in the figure, there exist regions in Russia that have a statistically significant larger number of exporters than expected based on regional GDP. Furthermore, many of these export-concentrated regions do not have a major port. Figure 1 also shows there are regions (weighted by GDP) with less exporting firms than the mean. These regions, indicated with the lightest shades, are not nearly as far below the mean as the regions much larger than the mean.
Fig. 1

Deviations from the average number of exporting firms, by region, weighted by GDP. Moran’s I \((89) =0.16^*\) (\(z\text{-}score=6.37\))

We use this information to calculate the Moran’s I statistic for spatial autocorrelation. The expectation of the statistic is \(-1/(89-1)\) under a GDP-weighted random distribution of exporters to Russia’s 89 regions. To calculate the Moran’s I, we convert all regions to the geographic centroid (thus avoiding discontinuity issues with Kaliningrad) and use an inverse distance weighing scheme. We also assume normality. With our data, the Moran’s I statistic is \(0.16^*\) (\(z\text{-}score=6.37\)), indicating that we reject spatial (GDP-weighted) randomness with 99 % confidence. Additionally, the clustering is not based on regions that have major ports. We conduct two other tests (available on the author’s Web site) using location quotients and the balls and bins method of Armenter and Koren (2010) to confirm this result.

To see that this clustering cannot be accounted for entirely by regional comparative advantage, consider a specific example: the location of Russian firms that export to the United States and the location of Russian firms that export to Canada. We choose these two countries because they have similar economies (though the United States is much larger), distance from Russia, and consume similar goods (at the 2-digit level, six of the top ten imported industries from Russia are the same). The Moran’s I for the location of Russian exporters shipping to the United States is \(0.04^{***}\) (\(z=1.93\)), and it is \(0.03^{***}\) (\(z=1.68\)) for Canada. Thus, there is clustering for exporters to the United States and Canada with 90 % confidence. To see that it is not the same regions (and therefore the same firms within a region) that specialize in exporting to both countries, see Fig. 2.
Fig. 2

Deviations from the average number of exporting firms, by region, weighted by GDP. For the United States, Moran’s I (89) \(=0.04^{***}\) (\(z=1.93\)) and for Canada, Moran’s I (89) \(=0.03^{***}\) (\(z=1.68\))

Figure 2 shows the deviation from the mean number of firms that export to the United States and the deviation from the mean number of firms that export to Canada. Notice that there are more regions in Russia that have exporters shipping goods to the United States than to Canada, but that is not surprising given the difference in GDP and that the United States is the second largest receiving country of Russian exports. The second fact is that the regions where exporters shipping to the United States are most concentrated (shaded black in Fig. 2 and indicating more than 2.5 standard deviations from the mean) are not the same regions as those in which exporters to Canada are concentrated. Therefore, Russian exporters to the United States are clustered in a different pattern than the Russian exporters to Canada. The fact that the Russian regions that specialize in exporting to the United States are largely different from the Russian regions that specialize in exporting to Canada despite the similarities in export composition and distance strongly suggests that more must be going on than comparative advantage at the region level.

To support this claim of exporter agglomeration by destination, we perform a bilateral Moran’s I statistic for these two countries. The statistic differs in the number of exporters in each region \(i\) shipping to country \(c\) from the average number of exporters also shipping to \(c\) and compares it to the difference between the number of exporters in another region \(j\) shipping to country \(d\) from the average also shipping to \(d\). Because of its construction with respect to the other country \(d\), the bivariate Moran’s I is not symmetric. We want to test whether the clustering pattern of Russian exports to Canada is the same as the clustering pattern of Russian exports to the United States. If the statistic is different from \(\mathbb{E }(I)\), then the clustering patterns are the same whereas a statistic not different from \(\mathbb{E }(I)\) suggests that the clustering patterns are different. The bivariate Moran’s I for Canada–United States is 0.0194, and for United States–Canada it is 0.0189. Neither of these statistics is different from \(\mathbb{E }(I)\), indicating that the geographic location of Russian exporters shipping to Canada is statistically different from those shipping to the United States, even though the composition of exports and the distance to the countries is roughly the same.

This pattern of exporter agglomeration holds when we expand the sample. Consider Table 1, taken from Cassey (2012). Observations on the diagonal are starred if there is statistically significant clustering, indicating that Russian exporters are clustered around the destinations of Canada, Germany, Great Britain, and the United States.
Table 1

Moran’s I for location of firms exporting to select countries





Great Britain



United States





























Great Britain



























United States











\( -0.0426^{***}\)







Source: Cassey (2012)

Observations on the diagonal are the Moran’s I for the location of exporters shipping to that country. Off-diagonals are the bivariate Moran’s I representing the degree of spatial correlation between the exports to the “row” countries and exports to the “column” countries. Observations have been normalized with respect to regional GDP. Bold observations are the bivariate statistic where both countries have a statistically significant univariate Moran’s I

\(^{***}\), \(^{**}\), and \(^{*}\)p values less than .10, .05, and .01

The lack of stars on the off-diagonal indicates the clustering pattern differs to these two countries if there are stars on the diagonal. (Stars on the off-diagonal indicate the spatial autocorrelation of Russian exporters is the same for the two countries.) Therefore, the pattern of clustering is not the same for seven of twelve possibilities (in bold in Table 1).4 Our evidence contrasts somewhat with that in Lovely et al. (2005) in that we find clustering around relatively open countries rather than relatively closed countries. Perhaps, this is a difference between a developing country such as Russia and a developed country such as the United States, or perhaps this is due to Russia’s relationship with countries otherwise considered relatively closed.

Using customs data from Russia, we have shown that Russian exporters are clustered but that these clusters differ depending on the country receiving the exports. The example with Canada and the United States provides strong evidence that regional comparative advantage cannot fully account for the pattern of exporter agglomeration. Next, we develop one possible theory consistent with this fact and show that destination-specific spillovers occur across industries.

3 A theory of exporter agglomeration based on shipment destination

Motivated by the evidence documented in Sect. 2.2, we develop a static theory of trade where exporters agglomerate with other exporters shipping to the same location due to international transaction costs. Ours is a monopolistic competition trade model in the spirit of Melitz (2003) and Chaney (2008) but modified to incorporate agglomeration as in Ottaviano et al. (2002) and Fujita and Thisse (1996).

We explore whether the agglomeration by destination is achievable through a reduction in the variable costs of exporting. We study variable costs of exporting because Koenig et al. (2010) find nonrobust evidence of agglomeration occurring through the variable cost when the number of firms in the area exporting to the same country is small, as it often is in Russia. We test the consistency of this assumption. To see the implications of a spillover working through the fixed cost to trade, see Krautheim (2012). See Cassey (2012) to see the consistency of an assumption of regional comparative advantage in transportation technology to specific destinations.

3.1 Destination-specific information spillovers and economics of scale in transportation

Our theory may take a few forms in the real world. The first version is a traditional treatment of an externality as described by Marshall (1920) and more recently by Head et al. (1995) among many others, where agglomeration occurs within a region because of knowledge spillovers that are limited in geography. In our case, the knowledge that is spilled is about exporting to a specific country because of good contacts, translation, or effective bribes paid as a percent of shipment value, for example.

A second story is that firms benefit from economies of scale in transportation and shipments. Because of containerization in rail, ship, and air transportation, there is a fixed size to the shipment.5 As documented in Besedeš and Prusa (2006), most shipments are small. This is true in our Russian data as well. Fifty percent of shipments weighed less than 185 kg (407 lbs) and 25 % of less than 14 kg (31 lbs). Thus, most shipments are small, with a few huge shipments.

Because most shipments are small, exporters cannot fill a standardized container or a ship on their own. Hence, if firm A is exporting to China, firm B can add its goods to the shipment. Either the transportation company would be willing to sell its remaining room to firm B below average cost, or firm A would be willing to sell its extra container room. Thus, both firm A and B pay a lower transportation cost than if they shipped alone. Micco and Pérez (2002) find evidence of this by estimating that a doubling of trade volume weight from a particular port reduces transportation costs on that route by 3–4 %. They write: “In general, even though most of these economies of scale are at the vessel level, in practice they are related to the total volume of trade between two regions.”

Regardless of which of these motivating tales is the most reasonable, modeling of them is the same. The particular stories above should not be taken too seriously. For example, we know that some countries require that imports in containers only come from one firm. But we think these two stories exemplify the spirit of why we posit an externality in transaction costs.

3.2 The model and equilibrium

Although this strand of literature on exporter agglomeration relies on the gravity framework, there is little theoretical background for why gravity is appropriate for analyzing agglomeration patterns. We show that by incorporating the externalities described, we can derive a gravity-type equation that explicitly contains an agglomeration variable. The resulting gravity with agglomeration equation has greater explanatory power than the traditional gravity model.

Our model is a variation of Melitz (2003) as expanded by Chaney (2008). There are \(N\) nonsymmetric destination countries spread spatially across the world. We do not model the characteristics of countries other than that each is a passive consumer of Russian goods and each differs in size and location. The model is static.

The representative consumer in each country \(j\) has standard taste-for-variety utility on the consumption of an aggregate good composed of the varieties of Russia goods available in that county. Countries differ exogenously in their income endowment \(Y_j\) and their spatial location. They differ endogenously in the availability of Russian goods, their aggregate price index, and their aggregate consumption.

There is a continuum of Russian manufacturing exporters distributed spatially throughout Russia. These exporters differ in productivity in addition to their location in space, which is exogenously fixed. Firm productivity, \(\varphi \), is drawn from a Pareto distribution and cannot be changed. Each firm, identified by their productivity and the region where they are located, produces a unique good. Because our model is static, firms do not choose either to enter or to exit nor their location within Russia. Instead, given their fixed location and the location of others, firms choose where to export to (including nowhere).6

There is a per-unit cost of production \(w_i\) that is common to all firms in a region, but differs across regions. (We believe this is plausible given the relative lack of labor mobility within Russia. It also allows for unilateral regional comparative advantages due to natural resource inputs or other regional specific unilateral cost advantages.) Finally, there is a fixed cost, \(f_{ij}\), associated with exporting from each region \(i\) in Russia to each country \(j\) in the world. Given their productivity, the cost of production, and the fixed cost to export, a firm in region \(i\) maximizes profits in each market \(j\) it sells to by choosing the price, \(p_{ij}(\varphi )\), and quantity, \(x_{ij}(\varphi )\), in that market: \(\pi _{ij}(\varphi )=p_{ij}(\varphi )x_{ij}(\varphi )-{w_i}\left(\frac{x_{ij}(\varphi )}{\varphi }\tau _{ij}(A_{ij}, \cdot )+{f}_{ij}\right).\)

Notice in the profit there is a region-destination variable transportation cost, \(\tau _{ij}(A_{ij}, \cdot )\). This cost is not a constant, but a function of the agglomeration term\(A_{ij}\), among other variables such as physical distance that will be specified later and are now represented with a dot. Turning \(\tau _{ij}\) into a function of other exporter activities based on the activities of the entire region is the primary theoretical contribution of this work. For technical purposes, the form of \(\tau _{ij}(A_{ij})\) can be very general, though we think it economically reasonable to suppose \(\tau _{ij}(A_{ij})>0\), \(\tau ^{\prime }_{ij}(A_{ij})<0\), and \(\tau ^{\prime \prime }_{ij}(A_{ij})>0\).

Unlike the constant markup common in Melitz-style models, in our equilibrium the per-unit price charged by each firm in a destination also depends on our agglomeration term \(A_{ij}\) such that the greater the agglomeration, the lower the transaction cost, and thus the lower the price: \(p_{ij}(\varphi )=\frac{\sigma }{\sigma -1}\frac{w_i}{\varphi }\times \tau _{ij}(A_{ij}, \cdot )\) where \(\sigma >1\) is the elasticity of substitution in the utility function. Equilibrium aggregate exports to country \(j\) from region \(i\) are:
$$\begin{aligned} X_{ij}=Y_iY_j\left(\frac{w_i\tau _{ij}(A_{ij}, \cdot )}{\theta _j}\right)^{-\gamma }(w_if_{ij}) ^{1-\frac{\gamma }{(\sigma -1)}}\lambda \end{aligned}$$
where \(\theta _j\) may be interpreted as a weighted trade barrier as in Chaney (2008) or multilateral resistance as in Anderson and van Wincoop (2003), \(\gamma >\sigma -1\) is the parameter on the Pareto distribution of firm productivities, and \(\lambda \) is a constant. Our agglomeration effect \(A_{ij}\) shows up in Eq. (1) in two places. It shows up directly in \(\tau _{ij}(A_{ij}, \cdot )\) but also indirectly in \(\theta _j=\left[\sum _i Y_i(w_if_{ij})^{1-\gamma /(\sigma -1)}[w_i\tau _{ij}(A_{ij}, \cdot )]^{-\gamma }\right]^{-1/\gamma }\).

4 Estimation and empirical results

Firms will export to a country if the profits of doing so are positive. Our theory in Sect. 3.2 results in the equilibrium firm profit in a particular country depending on GDP, agglomeration, distance, and other unobserved characteristics of the firm, region, and country. The probability a firm exports to a given country is then a function of these variables. From our theory, we estimate the following logistic regression at the firm level to analyze the effect of the externality on the extensive margin of exports. Note that our theory supports the regressions of Koenig (2009) and Choquette and Meinen (2011) when agglomeration depends on the number of exporters.
$$\begin{aligned} E_{ij}&= \underset{(0.069)^{*}}{4.576}+\underset{(0.004)^{*}}{0.078}\log M_{ij}-\underset{(0.007)^{*}}{1.456}\log D_{ij}-\underset{(0.004)^{*}}{0.108}\log Y_i+\underset{(0.003)^{*}}{0.252}\log Y_j\qquad \nonumber \\&N=3{,}979{,}007\ \end{aligned}$$
where \(E_{ij}=1\) if a firm in region \(i\) exports to country \(j\) and \(M_{ij}\) is the number of firms in region \(i\) exporting to country \(j\) and \(^*\) indicates a p value of less than .01. These estimates support our results using the Moran’s I that agglomeration exists and is destination specific. Though we do not have firm-level controls, our estimate on \(M_{ij}\) of 0.078 is above the range of 0.002–0.011 reported by Choquette and Meinen (2011) but within the range of 0.068–0.140 reported by Koenig (2009).

4.1 The standard gravity equation and our model

Next, we consider the effect of agglomeration on the intensive margin. Though the gravity equation takes several forms, we use the version derived by Anderson and van Wincoop (2003). In this formulation, the exponent on exporter GDP, \(Y_i\), and importer GDP, \(Y_j\), must be one, so we divide both sides by exporter and importer GDP. Aggregate region–country exports are \(X_{ij}\) and great circle distance are \(D_{ij}\). As is standard in the trade literature, we estimate using ordinary least squares (OLS).7

We derive our theoretical reduced form counterpart to gravity from Eq. (1) by taking logs and rewriting to get
$$\begin{aligned} \log \frac{X_{ij}}{Y_iY_j} = \alpha +\beta _1\log \tau _{ij}(A_{ij}, \cdot )+\beta _2\log w_i+\beta _3\log \theta _j+\varepsilon _{ij} \end{aligned}$$
where \(\alpha =\log \lambda \), \(\beta _1=-\gamma \), \(\beta _2=1-\gamma -\frac{\gamma }{\sigma -1}\), \(\beta _3=\gamma \), and \(\varepsilon _{ij}=(1-\frac{\gamma }{\sigma -1})\log f_{ij}\).
In order to estimate (3), we assume a functional form for how the agglomeration term, \(A_{ij}\), relates to bilateral trade costs, \(\tau _{ij}\). We assume trade costs are a function of physical distance, \(D_{ij}\), and the agglomeration term, but nothing else. Often the gravity literature uses variables such as common language, colonial history, location of overseas trade offices, and exchange rates as barriers to trade. But these variables are not nearly as important in our data because there is not much variation across regions within Russia.8 Therefore, we assume:
$$\begin{aligned} \tau _{ij}(A_{ij}, \cdot )=D_{ij}\times A_{ij}^{-\eta }. \end{aligned}$$
This modeling of technology nests the gravity model by setting \(\eta =0\). Later, we test if \(\eta =0\).

We have data on \(X_{ij}\), \(Y_i\), \(Y_j\), \(D_{ij}\), as well as the number of exporters \(M_{ij}\) and weight of shipments \(W_{ij}\). We do not have reliable data on \(w_i\) and \(f_{ij}\) and therefore cannot calculate \(\theta _j\). We use a set of region-specific binary variables \(S_i\) to account for \(w_i\) and any other unobserved unilateral region features such as differences in cultural and ethnic backgrounds, infrastructure, within-region transportation network, resources, and regional cost advantages. We use a set of country-specific binary variables \(T_j\) to account for \(\theta _j\) and any other unobserved unilateral country features such as exchange rate and idiosyncratic demand for certain goods. We let the bilateral fixed cost be the error term.

We take (3) to our Russian data. We perform our experiments at the region level to document the importance the agglomeration term provides for empirical trade flows. By doing so, we do not directly distinguish whether the agglomeration works through the variable or fixed cost at the firm level, since both would be realized as an increase in aggregate exports. Also, we cannot directly distinguish what the correct agglomeration term is. We consider two possibilities. The first is the number of exporters, similar to the agglomeration measure used by Koenig (2009) and Choquette and Meinen (2011). The second agglomeration measure we consider is aggregate weight of exports from each region to each country. We are the first to consider this variable as an agglomeration term.

Our first results are for the traditional gravity model (setting \(\eta =0\) and thus turning off agglomeration). The results are in Table 2 under column (A). The regression results suggest that the overall explanatory power of the gravity equation to account for Russian regional exports is low compared to the \(R^2\ge 0.7\) common in country-level data sets that often do not use binary variables to increase the goodness of fit.9

Our second results introduce agglomeration as measured by the number of exporters. This is the measure used by Koenig et al. (2010) and Choquette and Meinen (2011). The results, reported in Table 2 column (B), indicate that the number of firms in region \(i\) exporting to country \(j\) is not statistically different from zero in accounting for aggregate exports. This result is different than what Koenig et al. (2010) andChoquette and Meinen (2011) find. We think this is a combination of two factors: Russia is a developing country whereas France and Denmark are developed, and many regions in Russia tend to have fewer exporters than you would find in developed countries. Schmeiser (2012) shows that the developed/developing country distinction can be crucial in understanding trade barriers such as these. That agglomeration as measured by the number of exporters is insignificant seems counter the results in Sect. 2. But exporter count is just one measure of agglomeration, and perhaps, it is not always the best agglomeration measure to understand export knowledge spillovers in a gravity context.

As an alternative to measuring agglomeration by the number of exporters, we consider aggregate weight. At the industry or product level, it must be the case that an increase in shipment weight will cause an increase in shipment value. However, this mechanical relationship between weight and exports is less strict across industries. That is, the value of a shipment of computers and steel is not as strongly related to the combined weight as either one or the other of these products. Let \(W_{ij}\) be the aggregate weight of exports shipped from region \(i\) to country \(j\).10

Compare the results of column (C) to the benchmark gravity equation in column (A). In addition to our agglomeration term being significant at the 99 % level, the adjusted \(R^2\) improves from 0.58 to 0.64. We perform Wald tests using a model including all variables (\(D_{ij}, M_{ij}, W_{ij}\)) to see which model provides the best fit of the data. We first look at columns (A)–(C) and find that restrictions to the aggregate weight term \(W_{ij}\) seriously harm the fit of the nesting model whereas restricting exporter count \(M_{ij}\) does not. Finally, we compare Akaike information criterion (AIC) and confirm that the specification in (C) is preferred.

Because of our choice of how weight enters the transportation cost function (4), our model predicts that the coefficient of physical distance \(D_{ij}\) is equal to \(-\gamma \) and the coefficient on aggregate export weight \(W_{ij}\) is equal to \(\gamma \eta \). From these two predictions, we calculate \(\eta =0.310\), which is economically significant. We conduct a nonlinear Wald test of whether \(\eta =0\). We reject that \(\eta =0\) with 99 % confidence \((F(1, 2{,}\!755)=46.35)\), so \(\eta \) is statistically significant as well. This suggests our choice of technology is reasonable. Note that when we use the number of exporters as our agglomeration term, \(\eta =0.024\) and we cannot reject that \(\eta =0\).

Another endorsement of our model comes from our estimate of \(\gamma \). We cannot reject that \(\gamma =1\) with 99 % confidence, which agrees with Axtell (2001) who reports that \(\gamma \) is near one for many countries. Furthermore, notice that the estimate for \(\gamma \) in (A) without the agglomeration term is larger than that reported by Axtell.

Using our estimate on \(\gamma \) and Chaney’s estimate for \(\frac{\gamma }{\sigma -1}\), we calculate \(\sigma =1.40\). This is consistent with the requirement that \(\gamma >\sigma -1\) and that \(\sigma >1\). At first, \(\sigma \) seems too low for an estimate of the elasticity of substitution, since it implies very large markups in price over marginal cost. Additionally, our \(\sigma \) is low compared to the 3–8 range estimated by Hummels (2001), but Hummel’s estimates are within a product category whereas ours is for all manufactured tradeables. Crozet and Koenig (2010) estimate an average across trade-weighted industries of 2.25. Furthermore, as argued by Rauch (1999), if there are informational costs to trade in the data that do not appear as physical distance as in Eq. (4), then the estimate for \(\sigma \) will be low. When we use the estimates from Table 2 column (B), \(\sigma =0.34\) indicating that goods are complements. We find this further evidence that measuring agglomeration by aggregate weight is appropriate.

One issue with our approach is whether multiplant firms with a headquarter in a different region affect our estimates. Our customs data are unique to the firm level, not the establishment. We thus rely on the result in Koenig et al. (2010), who find their results are robust to using the headquarter locations and its neighbors. Another issue is possible simultaneity. Bernard and Jensen (2004) introduce a lag on the right-hand side to sever any mechanical relationships as past aggregate weight cannot affect current export value in any way other than a reduction in the transaction costs of trade. They find their results remain statistically significant. While we cannot perform a similar exercise with our Russian data, we appeal to this previous finding. Also, we perform a similar exercise with US state-level data (available at the author’s Web site) and find the results hold.

4.2 Product-level regressions

Though we derived our reduced form aggregate equation from firm-level theory, there is a potential problem with the estimates in Table 2 column (C) due to the mechanical relationship between aggregate shipment weight and export value. Though the relationship between export weight and value across industries is far less strong than within industries, there still exists the possibility of an endogeneity problem. To address this endogeneity concern, we perform product-level regressions.

We add product controls to our regression, \(P_k\). We also subtract out product weight from aggregate weight to get:
$$\begin{aligned} \log \frac{X_{ijk}}{Y_i Y_j}&= \alpha +\beta _1\log D_{ij}+\beta _2\left(\log \left(W_{ij}-W_{ijk}\right)\right)\\&+\sum _{i=2}^{89}\kappa _iS_i+\sum _{j=2} ^{175}\delta _jT_j+\sum _{k=2}^{56}\xi _kP_k+\varepsilon _{ijk}. \end{aligned}$$
The results for the benchmark gravity and with the addition of the agglomeration term are in Table 2 columns (D) and (E). They reveal that the agglomeration effect is strengthened. A Wald test indicates the restriction of (E) to generate (D) is strongly rejected by the data, and it also degrades model fit substantially as indicated by the large increase in the AIC measure. But the results also show unusual estimates for distance, which have bizarre implications for \(\eta \). Nonetheless, the fact that the coefficient on aggregate weight less product weight is statistically significant indicates that our results are not being driven by bias in the results of column (C) and that spillovers do occur across industries.

4.3 Preferential relationships

Though we do not find evidence that Russia’s history causes trade relations with former Warsaw Pact and Soviet countries to be systematically different, we repeat our regression (C) in Table 2 excluding these countries and find no change in statistical or economic significance of weight (with coefficient of \(0.245^*\) (0.016)). The evidence suggests that possible preferential treatment is not driving the results of the regression in (C). Importantly, that Russia’s unique history is not driving results indicates that the lessons we learn about export agglomeration by destination can be generalized to other developing countries.

4.4 Firm size

Now that we have shown that agglomeration as measured by aggregate weight is a statistically and economically significant variable in the gravity equation of trade, we do exploratory analysis to determine whether our agglomeration mechanism (whether through economies of scale in shipping or informational spillovers) is prevalent for firms of all sizes. Our model cannot distinguish between economies of scale in shipping and agglomeration occurring because of an informational spillover on how to export to a specific destination. However, we expect that savings from consolidating shipments to the same country would be more prevalent for small firms than for big firms, who presumably have their own distribution networks and can fill a container by themselves.

We re-estimate for separate subsamples based on firm exports to the world. These estimates, reported in Table 3, show that the effect of agglomeration is statistically and economically significant for all size firms. The fact that the key variable is statistically significant for each firm size group is suggestive of the importance of an informational spillover, while not refuting that small- and medium-size firms may benefit from consolidation of shipments.
Table 3

Firm size estimates

































The estimating equation: \(\log \frac{X_{ij}}{Y_i Y_j} = \alpha +\beta _1\log D_{ij}+\beta _2\log W_{ij}+\sum _{i=2}^{89}\kappa _iS_i+\sum _{j=2}^{175}\delta _jT_j+\varepsilon _{ij}\). Exporter and importer binary variables and a constant are estimated but not reported. Standard errors (in parentheses) are robust

\(^*\)p values less than .01

5 Conclusion

The findings from the theoretical and empirical trade literature indicate there are barriers to exporting whose precise characterization continues to elude researchers. Koenig (2009), Koenig et al. (2010), and Choquette and Meinen (2011) show export spillovers are important for firm export decisions in developed countries if those firms are exporting to the same destination. We use the physical location of exporting establishments in Russia to see whether there is exporter agglomeration by destination in a developing country. Using the Moran’s I, we find there is agglomeration of exporters, not only in general and around ports, but also in terms of the destination to which exports are shipped.

While acknowledging that other explanations are possible, we posit that this agglomeration of exporters by destination might be an externality from how-to-export information spillovers or economies of scale in shipping costs. We then study the aggregate ramifications of this line of thinking. We measure this agglomeration in two ways: by the number of firms exporting and by the aggregate weight of exported shipments. We find that the agglomeration term as measured by aggregate weight improves the fit of the gravity model of trade.

Using aggregate weight, our model’s prediction is verified with the Russian data and outside sources, and our model accounts for more of the variation in export share than the benchmark Anderson and van Wincoop (2003) or Chaney (2008) gravity-type model. We perform a number of robustness checks including product controls and eliminating Warsaw Pact and Soviet countries from the sample and find the agglomeration term remains economically and statistically significant. We also find that the agglomeration term is important for firms of all sizes, and even the smallest exporters benefit from informational and/or shipping cost spillovers.

The documentation of agglomeration by destination and the confirmation of our model takes a step toward understanding the nature of the international barriers to trade. Our combination of localization techniques and externality modeling from industrial organization with the tools from firm-level international trade models yields new empirical and theoretical insights into the nature and size of barriers to international trade, and the behavior of firms to overcome these barriers. The implications for trade-focused development are that policy perhaps should consider where exporters are operating and who they are serving when promoting exports rather than just focusing on particular industries or countries.


Cassey (2009) shows that for the United States, manufacturing data are the only data that can be reliably used for the state of production of the export instead of the state where the export began its journey abroad, whereas agriculture and mining data are not reliable.


Le Gallo and Dall’erba (2008) find no significant difference in results if they use great circle distances or travel time by road from the most populous town.


Russia’s federal regions are somewhat similar in geographic scope to US states. The number of regions has decreased since 2003 because of mergers. It is common to study Russia at this level of geographic disaggregation. See Broadman and Recanatini (2001) for example.


These results are robust to using export volume instead of the number of exporting firms.


Korinek and Sourdin (2010) show that shipping companies quote transportation rates in cost per container.


Martin et al. (2008) find that agglomeration positively affects firm productivity. To address this issue, Koenig et al. (2010) introduce a total factor productivity (TFP) variable to prevent overestimation of spillovers. In our data, we are unable to construct such a variable; however, Koenig et al. show that including productivity, while significant, does not affect the coefficient on the spillover.


Koenig et al. (2010) write, “Results are coherent with a linear specification since the effect on starting to export of having one neighbor exporting the same product to the same destination compared to zero (0.072) is very similar to the effect of having two neighbors instead of one, and of having three neighbors instead of two.”


We do, however, acknowledge that there other are bilateral variables such as immigration. But we believe that these are of secondary importance because the patterns of regional Russian immigrants (or the 185 identified ethnic groups in Russia) to the 175 countries in the world are unlikely to be empirically relevant compared to the product of GDPs.

The estimates for a “naive” gravity equation are
$$\begin{aligned} \log X_{ij}&= -\underset{(0.580)}{0.84}+\underset{(0.043)^{*}}{0.56}\log Y_i+\underset{(0.020)^{*}}{0.29}\log Y_j-\underset{(0.065)^{*}}{1.17}\log D_{ij}\\&N=2,\!985, \quad \hat{R^2}=0.15, \; \text{ RMSE}=2.45 \end{aligned}$$
Table 2

OLS estimates on Russian data






















































Product controls






Wald \(\chi ^2\)






p value












The benchmark gravity equation in (A): \(\log \frac{X_{ij}}{Y_i Y_j} = \alpha +\beta _1\log D_{ij}+\sum _{i=2}^{89}\kappa _iS_i+\sum _{j=2}^{175}\delta _jT_j+\varepsilon _{ij}\). Exporter and importer binary variables and a constant are estimated but not reported. Standard errors (in parentheses) are robust. Specification (B) adds the number of exporters in region \(i\) shipping to country \(j\), whereas specification (C) adds the aggregate shipping weight. A Wald test is performed compared to an encompassing model with all right-hand variables included. Specifications (D) and (E) repeat the analysis at the product level and include product binary variables. Own industry weight, \(k\), is subtracted from aggregate weight in (E)

\(^*\)p values less than .01

A logistic regression similar to (2) except replacing the number of exporting firms with aggregate weight yields:
$$\begin{aligned} E_{ij}&= \underset{(0.074)^{}}{0.567}+\underset{(0.002)^{*}}{0.230}\log W_{ij}-\underset{(0.008)^{*}}{0.764}\log D_{ij}-\underset{(0.004)^{*}}{0.329}\log Y_i+\underset{(0.003)^{*}}{0.076}\log Y_j\\&N=3{,}979{,}007\nonumber \end{aligned}$$


The authors thank Thomas Holmes, Yelena Tuzova, Anton Cheremukhin, and seminar participants at Amherst College, Vasser College, Washington State University, Kansas State University, University of Scranton, University of Richmond, and the Midwest International Trade Conference and New York Economics Association annual meetings. Cassey thanks Qianqian Wang and Pavan Dhanireddy for research assistance, and Jeremy Sage for help with ArcGIS. Cassey also thanks the Western Regional Science Association and the editors of the Annals of Regional Science. Portions of this research are supported by the Agricultural Research Center Project #0540 at Washington State University. This manuscript received the Springer Award for best paper by an early career scholar at the 51st Annual WRSA Meeting, Kaui HI, February 2012.

Copyright information

© Springer-Verlag 2012