1
Internet technology and the extensive margin of trade:
Evidence from eBay in emerging economies
Andreas Lendle
1
Pierre-Louis Vézina
2
July 2014
Abstract
Online platforms such as eBay offer technologies that make it easier for firms to export. This paper
dissects a new firm-level dataset that covers sales made through eBay by sellers based in 21 emerging
economies to provide a new lens through which to look at the effect of trade costs on the extensive
margin of trade. Comparing eBay sellers to “offline” firm-level data from the World Bank’s Exporter
Dynamics Database allows us to test whether the observed trade patterns on eBay fit with the trade-
liberalization predictions of heterogeneous-firm models. We find that eBay firms export to more
destinations, suggesting low destination-specific fixed costs on eBay. We then show that the
distribution of export destinations across eBay sellers is well approximated by a balls-and-bins model
of frictionless trade, suggesting eBay indeed lowers fixed export costs. Finally, we compare the
gravity of eBay to that of offline trade and find geographic distance, languages, and trade agreements
to matter less for online trade.
JEL CODES: F14, F17, L81
Keywords: Firm-level data, online trade, eBay, gravity equation
1
Office of the Chief Trade Adviser for Forum Island Countries, Vanuatu. Email: alendle@octapic.org
2
Dept. of Economics, University of Oxford. Email: pierre-louis.vezina@economics.ox.ac.uk
* We thank participants at DEGIT 2013 in Lima for their comments as well as eBay and the World Bank’s
“Exporter Dynamics Database” team for providing us with data.
2
1 Introduction
The last decade has seen a wave of papers studying the heterogeneity of firms in international
trade. These have been mostly motivated by Melitz (2003), who combined firm heterogeneity in
productivity with fixed export costs in trade models to show that extensive-margin effects, i.e., the
growth in the number of exporters, destinations, and varieties, were important following trade
liberalization.
Three accompanying empirical regularities (see Bernard et al. 2007), often attributed to high fixed
export costs, have been 1) the scarcity of exporters among firms (only the most productive export), 2)
the few number of markets reached by those firms that do export, and 3) the concentration of exports
among a small number of very large exporters – export “superstars” (Freund & Pierola 2012) or “the
happy few” (Mayer & Ottaviano 2007). To put it differently, these empirical papers have suggested
that exporting is rare and that this is most likely due to high export costs.
Yet recent empirical studies do not agree on the magnitude of the extensive-margin effects of trade
liberalization. For example, Kehoe and Ruhl (2013) suggest that extensive-margin growth is an
important part of export growth following trade liberalization. They find that over the period 1995–
2005, a 10% increase in trade between two countries was associated with a 36% increase in the
extensive margin, defined as newly traded goods. They add that the extensive margin accounted for
9.9% of the growth in trade between the US, Mexico and Canada following the North American Free
Trade Agreement in 1994. Similarly, Dutt, Mihov & Van Zandt (2013) show that the impact of the
World Trade Organization on trade is concentrated almost exclusively on the extensive margin. Their
preferred specification suggests that WTO membership increases the number of exported products by
25%. On the other hand, Baier, Bergstrand & Feng (2014) have estimated the effect of economic-
integration agreements on the margins of trade during 1962-2000 and found that intensive-margin
effects were more important.
Needless to say, understanding how a reduction in trade costs affects firms and trade is important
for the design of trade policy. This paper provides another lens though which to look at the effect of
3
trade costs on the extensive margin of trade. We use a new dataset covering firms based in 21
emerging economies that are selling on the eBay platform – one of the world’s largest online
marketplaces – to dig into the predictions of heterogeneous-firm models.
More precisely, we start by suggesting that online technology can reduce trade costs and therefore
can be thought of as trade liberalization. Specifically, we suggest that eBay reduces abruptly the
destination-market-specific costs of exporting. As argued by Hortaçsu et al. (2009) and Goldmanis et
al. (2010), the main benefit of the internet as a trade facilitator is to reduce search costs, and it is
reasonable to think of online marketplaces as being almost “frictionless'' in this regard – especially
when compared with traditional sales channels. A prospective eBay seller only needs to register with
eBay and list items to sell. Prospective buyers in any country then search eBay for products they want
to buy, facing almost no international frictions. Unlike for traditional “offline” trade, exporters no
longer need to hire consulting firms to deal with local procedures and attend trade fairs and
networking events to make foreign contacts. The export costs are thus mostly reduced to the variable
part including shipping and contract enforcement.
3
We then compare the extensive margin of exports on eBay with comparable “offline” firm-level
data from the World Bank’s Exporter Dynamics Database to check for differences between eBay
sellers and traditional firms (henceforth called “offline firms”) with regard to the extensive margin of
exports. This allows us to test whether the observed patterns on eBay fit with the trade-liberalization
predictions of heterogeneous-firm models.
Our results support our hypothesis that eBay is akin to a drastic reduction in fixed export
costs. It is well-known from the literature that most “offline” firms don’t export, and those that do
export reach customers in only a very small number of markets. What we observe on eBay in
emerging economies is remarkably different. Practically all sellers export (98% of them), and most
exporters sell to a wide range of foreign markets. We find that eBay firms export to 8.4 markets on
average, compared to 2.8 for offline firms. These results are all the more striking given the fact that
most eBay sellers are very small in terms of annual sales compared to offline exporters.
3
These variable costs for small-scale sales through the postal network can be quite high, especially shipping
costs. This may explain why the overall level of online sales is still small.
4
Yet Armenter & Koren (2012) argue that the prevalence of “zeroes” in trade data, i.e. the low
number of destination reached per exporter, is not necessarily due to high trade costs. They argue
instead that this could simply be a consequence of the “sparsity” of trade data. In their view, there are
simply not enough trade transactions to “fill” all the zeroes, especially when looking at highly
disaggregated data.
4
They suggest a balls-and-bins model of trade that takes into account data sparsity
in explaining the extensive margin. In their model, individual trade transactions, i.e. balls, are
independent and randomly allocated among destination countries, i.e. bins, with no export costs.
We show that the extensive margin of eBay exports, in terms of seller-destination
combinations, is indeed very well approximated by this Armenter & Koren’s balls-and-bins model of
frictionless trade. This suggests that the zeroes on eBay are well explained by data sparsity only, and
that eBay trade is almost frictionless, at least with respect to destination-specific fixed costs. Due to
data availability, we are only able to compare balls-and-bins results for online and offline firms for
Peru. We find that the balls-and-bins model fits the online distribution much better, suggesting eBay
trade is closer to frictionless trade and that zeroes in offline firms’ exports are explained not only by
data sparsity but most-likely also by fixed exports costs.
As a final step we investigate whether eBay indeed reduces trade costs by estimating gravity
equations on eBay and offline trade. Our aim here is to see whether measured trade costs, such as
geographic distance, languages, or free-trade agreements, matter as much on eBay as offline.
We find a distance elasticity on eBay trade of about -0.3, which is 50% smaller than for
offline exports, consistent with lower distance-related export costs on eBay. We also find that free-
trade agreements matter 60% less for eBay exporters, suggesting again that the trade costs associated
with lack of economic integration are much lower on eBay. These results confirm that measured trade
costs are less important on eBay and thus that we can think of eBay as a trade-liberalization
technology. When we decompose eBay exports into extensive and intensive margins, we find that
distance affects equally the number of exporters and the average value of exports, consistent with
distance capturing both variable and fixed costs associated with shipping and information respectively,
4
For example, most exporters would only sell to a few countries among all the 200+ possible destinations that
can be found in trade data. But many of these destinations are so small that it would be impossible for all firms
from a given exporting country to export there.
5
while free-trade agreements affect only the number of exporters. The latter result suggests that fixed
export costs related to a lack of economic integration do hinder the extensive margin of online trade
and hence that online trade is not completely frictionless.
Our study thus stands apart by focusing on a unique form of trade liberalization, i.e., online
technology, and using firm-level rather than product-level data to define the extensive margin of
exports. Our results support the idea that online technologies bring down trade costs and highlight the
potential of internet technologies to help countries realize gains from trade.
The rest of the paper is structured as follows: The next section describes our data and
methodology. A third section presents our results and a fourth section concludes.
2 Data and methodology
Our first step is to compare the extensive margin of exports, i.e., the average number of
different export destinations per exporter, on eBay with offline data from the World Bank. We use a
new firm-level dataset that covers all international sales conducted by eBay sellers based in 21
emerging economies.
5
We refer to sellers from a country as sellers being based in that country,
independent of which eBay site is used for their transactions.
6
The data was provided by eBay and is
at the level of individual transactions made. It is therefore much more detailed than most “offline”
trade data that is usually available to researchers. It contains data for all sellers for the period 2006 to
2011. For each seller, the dataset includes the number and value of transactions by destination country,
by product category, and by eBay site used. The data also contains the number of shipped items (some
transactions include more than one item), and the shipping fee. The data is anonymous - all eBay
sellers are only identified by a number.
5
The 21 countries are Albania, Bulgaria, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador,
Egypt, Guatemala, Jordan, Kenya, Lebanon, Morocco, Mexico, Macedonia, Pakistan, Peru, El Salvador, Turkey,
and South Africa. In Lendle et al. (2013) we focused on eBay sellers based in the US.
6
For example, an eBay seller based in Pakistan could sell items on the US eBay site (eBay.com) or any other
eBay site. Also, eBay sites are linked: If the item is posted on eBay.com, then it can also be found by eBay
sellers looking for items on other eBay sites.
6
We then use “offline” firm-level data for comparison, which is taken from the World Bank’s
Exporter Dynamics Database (World Bank 2012). The World Bank provided us with more detailed
data for Peru that closely resembles transaction-level data, which we analyze in more detail.
7
Yet this simple comparison of descriptive statistics may not be enough to draw conclusions about
the trade-liberalization effect of internet technologies on the extensive margin of exports. Indeed,
Armenter & Koren (2014) argue that the analysis of the extensive margin of exports should take into
account the sparsity of the data. Their main idea is that the number of transactions that a seller
undertakes, while not purely random, is certainly limited and small compared to the number of
possible export markets. Many sellers are small and have very few transactions, which means they will
“naturally” only ship to a few markets. Armenter & Koren thus suggest a simple statistical model that
predicts the number of export destinations per exporter by taking the sparsity of the data into account
and implying that there are no fixed costs to export. Their model provides a statistical benchmark
against which one can assess the extensive margin of exports. Finding that eBay trade can be predicted
by the model would thus be strong evidence of low export costs.
In the balls-and-bins model, trade transactions can be thought of as balls that are randomly thrown
into bins, and the width of bins being proportional to the size of the market. For eBay, one should
intuitively think of buyers picking sellers randomly – the buyers are “throwing balls” at them. This is
shown schematically in Error! Reference source not found.1. It shows a number of buyers and
sellers, each from four different countries. Each “ball” is a purchase, and the number of balls per buyer
country is assumed to be exogenous. If all buyers make random purchases, then the seller (and thus
country) from which a transaction is made will only depend on the width of a seller’s bin, which is
proportional to the number of items that each seller lists on eBay.
7
Unfortunately, we do not have access to such detailed “offline” data for other countries.
7
Figure 1. A balls-and-bins model of eBay exports
Australia
Buyer 1 Buyer 2 Buyer 3 Buyer 4 Buyer 5 Buyer 6 Buyer 7 Buyer 8 Buyer 9 Buyer 10
●●●●●● ● ●●● ●●●● ● ●●● ●●●●●● ●● ●●●●● ●
Seller 3 Seller 5 Seller 6
Kenya Peru
UK
France
Seller 1
Seller 2
Brazil
Seller 4
Turkey
US
If purchases are made randomly, then the expected number of different export destinations
(n_dest) out of C countries for a seller with t transactions can be calculated as follows:
In an extreme scenario of a perfectly “flat” world without fixed costs to export, the probability for
a specific transaction to be made with a country would only depend on demand from that country. We
would expect that sellers export more items to a large country as compared to a small country. The
probability that a specific country c receives at least one export is being given by s
c
, for which we use
the country’s share in total exports. For t going towards infinity, the number of different destinations
for an individual seller goes towards C, the total number of countries. In other words, everybody
would export everywhere:
8
With the number of sales going towards infinity, a seller would export to all possible
destination countries, and the distribution of sales across foreign countries would resemble world
demand shares of these countries. In practice, the number of transactions by each eBay seller is small,
which explains why they do not sell to all countries. Because the distribution of s
c
is very unequal,
with many destinations having extremely low probabilities to receive an individual shipment, even a
8
Note that by definition, both the expected and the actual number of destinations cannot be higher than the
number of transactions.
8
seller who ships 1,000 items in a given year is very unlikely to ship to all countries in the world – even
in the complete absence of any country-specific fixed costs.
9
This model does not in itself provide an
explanation for trade patterns but suggests that the geography of trade flows may be close to random.
We will thus use the balls-and-bins model to compare the extensive margins of trade online
and offline while taking into account data sparsity.
Last but not least, we also use the gravity model (Anderson and van Wincoop 2003) to look at
how the different nature of trade costs online should also affect the gravity of eBay trade. Our aim here
is to see whether measured trade costs, such as geographic distance, languages, free-trade agreements,
etc. matter as much on eBay as offline and thus to determine whether eBay can indeed be thought of as
a trade-cost reducing technology.
The gravity model suggests that trade between two countries is proportional to their economic
mass and the multilateral resistance indices, i.e. the weighted averages of price indices in the
importer's and exporter's trading partners, and inversely proportional to trade costs between the two
countries, captured by the geographic distance between them, i.e.:
where
are imports of country i from country j,
is total income in importing country i,
total
income in exporting country j,
is total world income,
are trade costs between country i and
country j, is the trade cost elasticity of imports, and
are the multilateral resistance terms in
the importing and exporting country. Bilateral trade costs
can be modeled as a function of
geographic distance and other trade cost variables:
where all α are parameters,
is the geographic distance between countries,
is a dummy variable
taking the value 1 when countries share a border,
is a dummy variable taking the value 1 when
9
Because the distribution of s
c
is very unequal, with many destinations having extremely low probabilities to
receive an individual shipment, even a much higher number of shipments would usually not cover all
destinations.
9
countries share a language,
is a dummy variable taking the value 1 when countries share a colonial
link,
is a dummy variable taking the value 1 when countries share a common legal system,
is a dummy variable taking the value 1 when countries are part of the same free trade
agreement, and
is a dummy variable taking the value 1 when countries share a currency.
We then substitute (4) into (3) and take logs on both sides to obtain:
where all β are parameters to be estimated and
=
, where k is the subscript indicating the trade
cost variable. We control for incomes and multilateral resistance including importer and exporter fixed
effects and estimate (5) separately for eBay and offline flows. This allows us to confirm that eBay is
akin to a reduction in trade costs and furthermore to understand why the extensive margin of eBay is
much more important than offline.
10
3 Results
We first look at descriptive statistics on the extensive margins online and offline. Bernard et al.
(2007) show that out of those US firms that export, 64% export to a single country. Only 14% of
exporting firms sell to five or more countries, and the average number of destination markets is 3.5.
Similar evidence has been shown for French firms. Mayer & Ottaviano (2007) find that 43% of French
firms sell to a single country and only 15% sell to ten or more countries. In the 21 countries for which
we have both eBay and offline data, the average number of destinations per offline firm is 2.8 – and it
is around 3 in most other markets for which we have offline data (but no eBay data).
How does this compare with eBay exporters? A first striking difference between online and offline
data is the fact that online sellers export to a very high number of different countries. This is seen in
Figure 2,Error! Reference source not found. which plots the average number of export destinations
10
In previous research (Lendle et al. 2012), we have shown – based on a dataset of bilateral eBay trade between
61 countries – that distance still matters for eBay, but about 65% less when compared to traditional offline trade
for the same countries and similar products. Here we replicate this analysis with the new seller-level data
covering exporters in 21 emerging economies and 204 destinations.
10
per firm, on eBay and offline. On eBay, across the 21 countries, exporters reach on average 8.4
different destinations. The average number of destinations per exporter ranges from 4 in Kenya to 16
in Egypt. Offline, only Turkish firms reach more than 4 destinations on average. This first finding is
consistent with our hypothesis that eBay reduces fixed destination-specific export costs.
This is all the more surprising considering the smaller size of eBay firms. Indeed, the literature
covering data for US and France shows that multi-country exporters, while few in number, are much
larger. In the 21 developing and emerging markets for which we have detailed data, offline exporters
are on average much larger than exporters on eBay, with average annual exports of around $2 Million
for offline exporters rather than around $5,000 on eBay. All in all, eBay exporters reach many more
destination markets than offline exporters, despite the fact that their aggregate exports are considerably
smaller. In fact, the differences are even more striking if one only considers “commercial” sellers on
eBay. Many of the sellers for which we have data only ship a small number of transactions per year,
and while they may nevertheless reach many markets, the number of markets they could reach is
limited. Using a threshold of $10,000 in annual sales to define “commercial” sellers, we find that such
sellers reach even more markets (32 on average) – despite still being small compared to “offline”
firms.
11
Figure 2. Average number of destinations per exporter
Despite these significant differences, most eBay sellers still do not export to most markets. While
the average seller reaches about eight different markets, this also means they do not reach the other
190 markets.
11
However, as explained in the previous section, a low number of destinations does not
necessarily indicate fixed market-specific costs because one also has to take into account the sparsity
of the data. A simple example illustrates this. Assume one had data for US-based eBay sellers’
domestic transactions by destination defined as ZIP code area (of which there are more than 40,000 in
the US). Now, most sellers would only sell to a tiny fraction of those, but obviously there are no fixed
costs to “enter” a ZIP code.
12
11
While eBay data includes slightly different definitions of destination markets than the World Bank data (e.g.,
some tiny islands may be included in one dataset, but not in the other), both types of data contain about 200
possible destination markets.
12
Taking into account the product dimension, consider Amazon US’ domestic shipments of books by ZIP code
area and ISBN code of the book – an extreme example of “sparse data”. While Amazon probably ships some
books to any ZIP code, many books will only be shipped to a few zip codes. This does not indicate that there are
ZIP-code specific “entry costs” for a book. See Hillberry & Hummels (2008) for a paper using US domestic
shipments by ZIP code, where they make a similar argument.
ALB
BGR
CHL
COL
CRI
DOM
ECU
EGY
GTM
JOR
KEN
LBN
MAR
MEX
MKD
PAK
PER
SLV
TUR
ZAF
45° line
0 4 8
12 16
eBay
0 4 8 12 16
World Bank
12
We thus calculate the expected number of destinations for each eBay exporter given its number of
transactions and as predicted by the balls-and-bins model. In Figure 3, we compare the actual and
simulated average number of export destinations for given numbers of transactions.
13
The number of
markets reached by eBay exporters is almost as high as the balls-and-bins model predicts. This
suggests eBay trade is indeed close to frictionless.
Figure 3. Expected vs. actual number of destinations per eBay exporter
Figure 4 shows the results for such a simulation made with detailed offline data for Peru, the
only country for which we have sufficient data – and provide a direct comparison with Peru-based
eBay exporters. We can see that the model predictions are close to eBay data but far from the offline
data. For example, offline exporters in Peru with around 100 export transactions “should” export to 29
different destinations, but in reality only sell to 6 whereas eBay exporters reach about 20. The balls-
13
Such a simulation cannot easily be done for offline firms because firm-level data usually contains no
information on the number of transactions. This also applies to the publicly available data from the World
Bank’s database, except for data for Peru.
1 2 3 5
10
25
50
100 200
Number of export destinations
1 10 100 1000 10000 100000
Number of export transactions
eBay Model predictions
13
and-bins model thus does not describe well the behavior of the extensive margin of offline exports in
Peru. As one can see, some fairly large exporters in Peru (in terms of the number of export
transactions) only sell to a single market.
Figure 4. Expected vs. actual number of destinations per Peruvian exporter: eBay vs. offline
What do we learn from this? Offline exporters sell to fewer markets than one should expect
from the balls-and-bins model, which indicates that their exports are not “random” and independent of
each other. This can be interpreted as evidence for fixed costs to enter specific markets (or to enter into
a buyer-seller relationship).
What can we conclude? It appears that the extensive margin of eBay exporters – the number of
markets they reach, can be fairly well replicated by the balls-and-bins model. There are several
plausible explanations for the small deviation from the predicted outcome of the model that do not
assume any market-specific entry costs.
14
The comparison with offline data shows that the balls-and-
14
For example, sellers might often make repeated sales to the same customer, which implies that transactions are
not fully independent from each other – unlike what is assumed in the balls-and-bins model.
1 2 3 5
10
25
50
100 200
Number of export destinations
1 10 100 1000 10000
Number of export transactions
eBay
World Bank Model predictions
14
bin model does not perform very well in predicting offline patterns, as one would expect if one
assumes that entry costs are much more important offline than online. This is strong evidence that
such costs are indeed lower online.
The different nature of trade costs online should also affect the gravity of eBay trade, i.e., the
impact of bilateral trade costs on the geography of trade. In previous research (Lendle et al. 2012), we
have shown – based on a dataset of bilateral eBay trade between 61 countries – that geographic
distance still matters in affecting eBay trade, but about 65% less when compared to traditional offline
trade for the same countries and similar products. In this paper, we replicate this analysis with the new
seller-level data covering exporters in 21 emerging economies and 204 destinations. While this data
only covers exports from 21 countries, it includes all destination countries, whereas our previous
dataset only included trade among 61 countries.
The eBay dataset also allows us to decompose exports into the extensive margin (number of
exporters) and the intensive margin (exports per seller). This allows us to verify whether the distance
effect on eBay is driven by the extensive margin (few sellers export to distant markets), or by the
intensive margin (sellers export less to distant markets).
Table 1 shows the results of a gravity equation for eBay exports and “offline” exports. We find
that distance matters almost 50% less on eBay – with a distance elasticity of - 0.31 vs. - 0.59. When
we decompose eBay exports into margins, we find that distance affects both margins pretty much to
the same extent. This is consistent with distance capturing both variable and fixed costs associated
with shipping and information respectively. While we cannot replicate this for offline exports, other
authors have done so using detailed “offline” firm-level datasets, including most recently some authors
from the World Bank (Cebeci et al., 2012), with a dataset covering 45 countries.
15
They find that for
offline trade, it is mostly the extensive margin that is affected by geographic distance. This suggests
that the trade costs measured by geographic distance are less of a barrier for the eBay firms than
offline firms. Bernard et al. (2007) also found that the extensive margin of US exports decreases most
with distance.
15
The dataset used by the authors is very similar to the data we received from the World Bank, but the World
Bank authors have access to more disaggregated raw data that allows them to decompose exports accordingly.
15
We also find that a common language increases eBay exports slightly less than traditional
exports, suggesting eBay may reduce information frictions captured by language differences. Yet this
difference is not statistically significant. Moreover, the language effect on eBay seems to be mostly on
the intensive margin, suggesting languages are also associated with variable cost, for example related
to contract enforcement. Sharing a border has a similar effect as sharing a language, both online and
offline. Common legal systems and common currencies do not seem to affect any type of trade among
our group of exporters. We also find that free-trade agreements matter much less for eBay exporters,
though they still matter significantly for the extensive margin of exports.
Table 1. The gravity of eBay exports from 21 emerging economies
(1)
(2)
(3)
(4)
Offline
eBay export value
Nb of exporters
Avg value per exporter
Distance
-0.586***
-0.313***
-0.153***
-0.190***
(0.0499)
(0.0478)
(0.0184)
(0.0420)
Shared Border
0.426
0.438**
0.0408
0.431**
(0.357)
(0.208)
(0.0935)
(0.194)
Common language
0.543***
0.416***
0.162***
0.300***
(0.0993)
(0.103)
(0.0417)
(0.0880)
Colonial link
0.350
0.344
0.250*
0.0467
(0.342)
(0.216)
(0.129)
(0.193)
FTA
0.761***
0.313***
0.345***
-0.0711
(0.130)
(0.0977)
(0.0495)
(0.0858)
Common legal system
-0.0430
0.0709
0.0330
0.0332
(0.0807)
(0.0808)
(0.0352)
(0.0691)
Common currency
0.499
-0.213
0.0821
-0.274
(0.444)
(0.346)
(0.112)
(0.319)
Observations
4,284
4,284
4,284
4,284
R-squared
0.907
0.850
0.868
0.746
OLS regressions with importer and exporter fixed effects. Trade flows are transformed using the inverse hyperbolic sine
rather than the log transformation, as the former is defined at 0 and allows for coefficients to be interpreted as elasticities,
as when taking logs. Country-pair clustered standard errors in parenthesis. ***significant at 1%, ** 5%, * 1%. Data for
offline exports is taken from Comtrade and gravity controls are from CEPII.
To conclude, this section has confirmed the previous findings of Lendle et al. (2012) that eBay
trade is less affected by geographic distance for a dataset covering exports from 21 emerging markets
to all importing markets. This suggests that the trade costs measured by geographic distance are less of
a barrier for the eBay firms than offline firms and thus that online technologies can be thought of as
trade-liberalizing.
16
4 Conclusion
Understanding how a reduction in trade costs affects firms and trade is important for the design of
trade policy yet recent empirical studies do not agree on the magnitude of the extensive-margin effects
of trade liberalization. Our paper, by looking at trade on an online platform, provides a new lens
though which to look at the effect of trade costs on the extensive margin of trade.
We have shown that eBay firms export to 8.4 markets on average, compared to 2.8 for offline
firms, and that the extensive margin of eBay exports, in terms of seller-destination combinations, is
well approximated by this Armenter & Koren’s balls-and-bins model of frictionless trade. This
suggests that eBay trade is almost frictionless, at least with respect to destination-specific fixed costs.
Our comparison with offline data shows that this is not the case for offline firms. Our paper thus also
shows that the balls-and-bins model can be a useful tool to analyze sparse trade data as well as the
relevance of fixed costs. What’s more, the eBay platform can be seen as an example to overcome
traditional trade barriers in particular any destination-specific fixed costs to enter a market.
Yet these results should not be interpreted as evidence for the total absence of trade barriers for
online trade. Exporters may not face high fixed entry costs into markets, yet exports are still
constrained significantly by gravity barriers such as distance, language, and trade policies. The gravity
results on the intensive margin of eBay exports also suggest that variable costs are much present, most
likely including those related to complicated customs procedures and expensive or unreliable postal
services. Identifying more carefully the barriers faced by international eBay trade would be a key area
for further research.
5 References
Anderson, James and Eric van Wincoop (2003). Gravity with gravitas: a solution to the border puzzle.
American Economic Review 93, 170-92.
Armenter, Roc, and Miklós Koren. 2014. A Balls-and-Bins Model of Trade. American Economic
Review, 104(7): 2127-51.
Baier, S. L., Bergstrand, J. H., & Feng, M. (2014). Economic integration agreements and the margins
of international trade. Journal of International Economics.
17
Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding & Peter K. Schott (2007). Firms in
International Trade. Journal of Economic Perspectives 21(3):105-130.
Cebeci, T., Fernandes, A., Freund, C. & M. Pierola (2012). Exporter Dynamics Database. World Bank
Policy Research Working Paper 6229, World Bank.
Dutt, Pushan & Mihov, Ilian & Van Zandt, Timothy, 2013. The effect of WTO on the extensive and the
intensive margins of trade," Journal of International Economics. Elsevier, vol. 91(2), pages 204-219.
Freund, C. & M. Pierola (2012). Export Superstars. World Bank Policy Research Working Paper
6222, World Bank.
Goldmanis Maris & Ali Hortaçsu & Chad Syverson & Onsel Emre, (2010). E-Commerce and the
Market Structure of Retail Industries. Economic Journal, Royal Economic Society, vol. 120(545),
pages 651-682, 06.
Hillberry, Russell & Hummels, David, (2008). Trade responses to geographic frictions: A
decomposition using micro-data. European Economic Review, Elsevier, vol. 52(3), pages 527-550,
April.
Hortaçsu, A, A. Martinez Jerez & J. Douglas (2009). The geography of trade in online transactions:
Evidence from eBay and MercadoLibre. American Economic Journal: Microeconomics 1(1), p. 53-74.
Kehoe, Timothy J. & Kim J. Ruhl, 2013. How Important Is the New Goods Margin in International
Trade? Journal of Political Economy, University of Chicago Press, vol. 121(2), pages 358 - 392.
Lendle, A., M. Olarreaga, S. Schropp & P.-L. Vézina (2012). There Goes Gravity: How eBay Reduces
Trade Costs. CEPR Discussion Paper No. 9094.
Lendle, A. (2012). An anatomy of online trade: evidence from eBay exporters. Mimeo.
Lendle, A., M. Olarreaga, S. Schropp & P.-L.Vézina (2013). eBay’s anatomy. Economics Letters.
Mayer, Thierry & Gianmarco I. P. Ottaviano (2007). The Happy Few: The internationalisation of
European Firms. Bruegel, Brussels.
Melitz, Marc J. (2003). The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry
Productivity. Econometrica, Econometric Society, vol. 71(6), pages 1695-1725, November.
World Bank (2012). Exporter Dynamics Database. Washington D.C.: World Bank.
