Border Effects in a Free Trade Zone: Evidence from European Wine Shipments
Mona Kashiha
Department of Geography and Earth Sciences
UNC Charlotte
Charlotte, NC 28223
mkashiha@uncc.edu
Craig A. Depken, II*
Department of Economics
UNC Charlotte
Charlotte, NC 28223
cdepken@uncc.edu
Jean-Claude Thill
Department of Geography and Earth Sciences
UNC Charlotte
Charlotte, NC 28223
Jean-Claude.Thill@uncc.edu
* Corresponding author
Border Effects in a Free Trade Zone: Evidence from European Wine Shipments
Abstract
This paper examines shipments of wine from European producers to various European
ports. The research question focuses on disaggregated decisions on which port to use
controlling for port performance, distance, and whether a national border must be crossed
to reach the port. Using a mixed logit model we find heterogeneous impacts of distance,
performance, and national borders on port choice across shippers of various sizes. Using
the estimation results we calculate the distance equivalents of national borders for
shippers of various sizes. We show that border effects are non-trivial and asymmetric
even within the European free-trade zone.
JEL Classifications: F14, L66
Keywords: trade barriers, bilateral trade, distance equivalents, border effects
1. INTRODUCTION
Distance and national borders have been the focus of a number of theoretical and
empirical studies of international trade patterns. Meanwhile the economics of transportation
literature has asked what influences the decision of which port a shipper uses. This paper
contributes to both strands of research by focusing on disaggregated shipping decisions of
European wine makers that often entail crossing a national border. We first model and discuss
influences on shipment-level port decisions allowing for heterogeneity in preferences across
shippers of various sizes. We then use the estimation results to calculate distance equivalents of
national borders for shippers of various sizes.
This study diverges from the extant literature in four ways: first, using a discrete choice
framework we simultaneously account for the distances and borders between a particular shipper
and several possible destinations; second, we accommodate unobserved heterogeneity in shipper
preferences by using a mixed logit estimator; third, our data describe individual shipments rather
than aggregated trade volumes at the national or industry level, that is, we investigate individual
decisions rather than aggregations of decisions; and fourth, we measure the precise distance
between the origin of each shipment and the port chosen by the shipper rather than as between
national capitals or centers of economic activity.
Studies of aggregated trade volumes combine shipments of many different commodities,
implicitly assuming that the shipping decisions for car manufacturers are the same as for those
who produce furniture or consumer electronics. Yet, this assumption is not testable using
aggregated data. To avoid this problem, we focus on one particular commodity, wine, and we
utilize data that describe individual shipments. Wine is a relevant commodity to study because of
its regional importance (wine is the most common commodity exported from Europe to the
3
United States), it is relatively homogeneous in its shipping needs, e.g., refrigeration, and wine
shipments are divisible, i.e., it is possible to ship several cases up to several containers of wine.
We utilize data compiled from the PIERS Trade Intelligence database that describe
containerized door-to-door shipments of wine from points of origin within the European freetrade zone to one of sixteen possible ports within the same zone from July 2006 to June 2007.
To preview our empirical results, we find statistically and economically meaningful differences
in the discrete choice framework for shippers of various sizes. For instance, the impact of port
performance on the odds of choosing a particular port increases with shipper size. Further, we
find that distance from a shipper to a particular port and crossing a national border both reduce
the likelihood that the particular port will be chosen. To connect to the literature on border
effects, we use the empirical results to calculate the distance equivalents of national borders. We
find non-trivial and asymmetric distance equivalents of national borders within the Euro freetrade zone.
2. LITERATURE REVIEW
Quite naturally, geography, and particularly distance, is one of the most well-known
barriers against the movement of commodities and people. In most studies distance serves as a
proxy for transportation costs and is found to be an important determinant to trade flows. In
addition to distance, language, currencies, membership to regional or global trade agreements,
sharing a common legal culture, and whether one or both of the trading countries are landlocked,
have also been shown to influence trade patterns and flows (see, for example, Anderson and Van
Wincoop, 2003; Behar and Venables, 2011; Brun et al., 2005; and Hummels, 1999).
The significant negative effect of distance on trade seems obvious, but counter to
intuition estimated distance effects have been increasing rather than decreasing over time. This
4
contradicts the perception of the “death of distance” brought about by the current wave of
globalization and this distance puzzle has attracted attention in many international trade studies.
Several researchers, including Brun et al. (2005) and Coe et al. (2007), find that the impact of
distance is falling over time whereas studies reviewed by Leamer and Levinsohn (1995) and
Disdier and Head (2008) fail to find a decrease in the impact of distance over time. Disdier and
Head conclude that their systematic analysis of 1,467 distance coefficient estimates from 103
separate studies “represent a challenge for those who believe that technological change has
revolutionized the world economy causing the impact of spatial separation to decline or
disappear.”
After long use, the standard gravity model, which models bilateral trade volume as a
primarily a function of the national product of the two countries and the distance between them,
was revolutionized by Anderson and Van Wincoop (2003). Their non-linear gravity model
implies that bilateral trade is homogenous of degree zero in trade costs, which they point out,
may be a primary reason why gravity model estimations have not found trade becoming less
sensitive to distance over time.
Borders
Another spatial dimension that has caught the attention of economists concerns national
borders (for example, Combes et al., 2008). Empirical evidence suggests that crossing a national
border, while itself a rather trivial distance, can add considerable effective distance to
international trade patterns. Given the worldwide trends in reducing tariff and nontariff barriers
to trade due to expanding economic freedom, multilateral and bilateral free trade agreements,
and a reduction in logistics and transportation costs, the remaining significant border effect is
5
puzzling and a large empirical literature provides various estimates for the magnitude of national
borders on international trade.
McCallum (1995) found that the U.S.-Canada border caused Canadian provinces to trade
22 times more with each other than with U.S. states. Further research confirmed a substantial
home bias in trade between integrated and culturally homogenous countries, where a free trade
agreement is supposed to otherwise enhance international trade. For instance, Helliwell (1996
and 1998) confirms McCallum’s original findings, by analyzing trade between the US and
Canada from the post-NAFTA period and shows that international trade fell by a factor of
twelve.
European Union national borders have been also studied extensively. Chen (2004)
estimates that the average EU country trades about six times more with itself than with other EU
countries. Nitsch (2000) finds that a border reduces trade by a factor of 6.8 between European
countries. Turrini and Ypersele (2010) shows that the absence of borders raises trade by a factor
by approximately 9.5 among OECD countries.
The frictional effect of national borders on trade has been routinely documented in
empirical studies yet there is no consensus on its underlying causes. Turrini and Ypersele (2010)
point out that several factors such as exchange rate volatility, non-tariff barriers and regularity
differences across countries, informational barriers and the role of commercial networks, weak
institutions, and widespread corruption have all been suggested as partial causes of border
effects. Using data describing OECD countries, they show that, after controlling for countryspecific factors, distance, the presence of a common border, and sharing a common language, a
national border matters most when there are significant differences in legal systems between the
two trading partners. They found that trade between two countries with “identical legal
6
procedures to refund an unpaid check” is seventy percent higher than with a “fully differentiated
[legal] procedure”.
Some studies show a relationship between the magnitude of a border effect and the
degree of substitution between domestic and imported goods. For instance, Chen (2004) shows
that in the EU bulk commodities like concrete, stone, or mortars experience the largest border
effect. Moreover, Wolf (1997) and Hillberry and Hummels (2005) argue that a border effect can
arise endogenously as a result of spatial clustering of firms seeking to reduce transportation
costs.
Border effects have been detected using the dispersion of prices of similar goods between
city pairs. Engel and Rogers (1996) show that crossing the US–Canada border is equivalent to
traversing a distance of 75,000 miles. Depken and Sonora (2002) found the effect of the Mexico–
U.S. border is equivalent to crossing between 3,642 and 44,765 more miles of distance. Parsley
and Wei (2001) estimate a remarkably wide border between US and Japan, finding it is
equivalent to adding 43,000 trillion miles or 7,314.79 light years. Distance, unit-shipping costs,
and exchange rate variability are possible explanations of the observed border effect.
Port Performance
The data utilized in this study describe individual shipments from wine makers to various
ports of departure in Europe. As ports are nodal points of the worldwide maritime network that
connect inland distribution networks, port efficiency, infrastructure, accessibility, and
connectivity have been recognized as important factors in port choice. Despite the large literature
focusing on port characteristics, there are no universal port-level efficiency indicators nor an
aggregated index of port performance available to researchers. While the Global
Competitiveness Report (GCR) reports port efficiency at the national level, some economists
7
have estimated individual port efficiency through data envelopment analysis, stochastic frontier
analysis, multivariate analysis, and regression models (Cullinane and Song 2006; Wanke et al.
2011; Tang et al., 2008; Clark et al., 2004; and Blonigen and Wilson, 2007). We cannot utilize
their estimates of port efficiency because their samples are not comprehensive and do not include
the ports we use in this study.
Acknowledging the difficulty in measuring port efficiency and performance, Malchow
and Kanafani (2004) used average capacity of vessels (in TEUs) as a proxy for port
attractiveness, Veldman et al. (2003) included liner service frequency, and Tiwari et al. (2003)
used vessel size and port throughput in their analysis of port choice. Tongzon (2009) regresses
port’s throughput (a port performance indicator) on port efficiency, cargo-handling capacity and
reliability, frequency of ship visits, number of containers lifted per crane, number of container
berths, and delay time. The high value of the coefficient of determination suggests that port
throughput might be a reliable aggregate of a port’s attributes. Therefore, we use port throughput
measured as total number of containers loaded as our measure of port performance.
3. EMPIRICAL METHODOLOGY AND DATA
Methodology
It has been shown that the gravity model can be related to logit-type models (Anderson
2010). However, the standard logit model, which is popular in the port-choice literature, cannot
capture heterogeneous preferences and treats repeated choices by the same decision-makers as
independent (cross-sectional) observations. Another well-known limitation of the logit model is
the Independence of Irrelevant Alternatives (IIA), which imposes restrictions on the relationship
between choices. To avoid these limitations and to exploit the richness of our data, we employ a
8
mixed logit framework that matches the economic reality in the shipping industry of
unobservable or difficult-to-measure variables.
Computational development and ease of simulation allows estimation of more realistic
behavioral models using more complicated methods, which were previously unapproachable due
to computational complexity (Train 2003). One established solution to the limitations of logit is
to allow for variation in preferences across individuals. In other words, to decompose the error
term to two components; the first component can be correlated over alternatives and captures
unobservable taste heterogeneity by allowing for random preferences, and the remainder is an
identically and independently distributed error.
Let the utility for person n choosing alternative i be Uni = nxni + ni, where
is a vector
of preferences for observed attributes that vary across individuals with density ( ), and
is
an iid extreme value error term. The probability that individual n chooses alternative i is a
weighted average of the logit formula evaluated at different values of β, with the weights given
by the density f(β); in most applications the density function is specified as normal or lognormal.
The mixed logit probability is obtained by the integrals of standard logit probabilities over all
possible
:
The mixed logit accommodates a wide range of behaviors, avoids the aforementioned
limitations from the logit (Behrens and Pels, 2012), and has been extensively used in the
9
industrial organization and transportation literature (for example, Hastings et al., 2005; Hensher
and Greene, 2002; and Hess and Polak, 2005).1
Data
The data used in this study describe 32,097 separate European wine door-to-door
shipments to sixteen European ports made by 5,732 distinct shippers from July 2006 through
July 2007. Many shippers make repeated choices which we accommodate in our estimation
strategy. The data are comprised of information submitted through both the U.S. Customs and
Border Protection Automated Manifest System (AMS) and manifests submitted at the various
ports. These original documents have been corrected, cross-referenced, and improved by data
supplier Port Import Export Reporting Service (PIERS) and distributed to users as the PIERS
Trade Intelligence data product (PIERS, 2007). This data include attributes of each shipment
including bill of lading, the shipping company’s name and address, a description of the
commodity being shipped based on the Harmonized Commodity Description and Coding
2
System , shipment quantity in twenty-foot-equivalent-units (TEU), the shipment's estimated
value and weight, the carrier, forwarding port, pre-carrier location, U.S. port of entry, and U.S.
consignee and its address (if the shipment is not in transit to a third country).
Figure 1 depicts the spatial distribution of European ports and wine shippers in the
sample, differentiated by total TEUs. Figure 1a shows that there are three very large ports, four
mid-size ports, and several smaller ports used by European wine makers shipping to the United
States. Figure 1b depicts the spatial distribution of European wine shippers throughout Europe.
1
The mixed logit specification above can be generalized to accommodate repeated choice situations, by calculating
the integral of the product of logit probabilities, one for each time period, instead of only one logit probability. We
employ this extension
10
The majority of wine shippers are, as would be expected, in Spain, France, Germany, and Italy
(the four countries in the region that produce the most wine). Table 1 reports the countries of
origin and their aggregate level of shipments (in TEUs) included in the sample.
As can be seen in Figure 1b the size of shippers varies dramatically. Figure 2 shows that
the distribution of shipper size, based on total TEUs reported in the data, is skewed toward
smaller shippers.3 While the five largest shippers in our data account for 58% of wine shipping in
Italy, 40% in France, and 46% in Spain, there are a large number of very small shippers who also
export wine to the U.S. Of the 5,732 unique shippers, 3,072 shipped less than one TEU over the
year of the sample, which collectively accounts for 0.6% of the total Italian exports to the U.S,
1.3% for France, and 2.5% for Spain. This reflects the fragmented structure of the European
wine industry discussed by several researchers (Bardají and Mili, 2009; Chandes et al., 2003; and
Garcia et al., 2012).
Table 2 reports the number of wine shipments during the sample year and the percentage
of all wine shipments that go through each port in the sample. As can be seen, the port of
Leghorn was selected most often at 22%, followed by La Spezia (18%), Le Havre (16%),
Barcelona (8%), Antwerp (8%), Genoa (7%), Fos (7%), and Rotterdam (5%). No other port was
selected more than five percent of total choices.
Our empirical analysis aims to explain port choices through a number of possible
variables. Unlike a standard logit model in which there is only one observation per choice, in a
mixed logit framework there are J observations for each choice, where J is the number of choices
available to the decision-maker. Thus, for each decision there are sixteen observations (one for
each port), fifteen of which have a dependent variable that takes a value of zero and one that
takes a value of one.
3
A single TEU can contain 800 12-bottle cases of 750 ml bottles.
11
To capture the importance of proximity and transportation costs, the explanatory
variables include the distance from the shipper’s location to the various ports (distance); an
indicator that takes a value of one if the port is the closest port and zero otherwise (home port);
an indicator that takes a value of one if the border is crossed and zero otherwise (bordercross);
throughput as a measure of port performance (performance); the oceanic distance to the U.S. port
of destination (oceanic distance).4
We control for shipper heterogeneity in various ways. First, the model is estimated
separately for groups of shippers based on size (smaller than 5 TEUs of wine per year, between 5
and 50 TEUs, between 50 and 300 TEUs, between 300 and 600 TEUs).5 We include the
interactions between distance and estimated shipment value, between having to cross a national
border with estimated shipment value,6 and between distance and crossing a national border.
To test if a shipper’s current choice is affected by its past choice a lagged dependent
variable is added to the model (lagchoice). Since the mixed logit model assumes that the error
terms are independent over time, the lagged dependent variable, Yt-1, is uncorrelated with the
error term at time t. In this regard, the lagged dependent variable does not cause inconsistency in
the estimation (Train, 2003).
One of the advantages of the mixed logit model is that it allows for heterogeneous
preferences across individual economic agents. We initially allow the preferences for distance,
border, and port performance to be random, and drawn from a Normal distribution.7 This seems
4
The oceanic distance is the great circle distance between two ports and does not reflect the vessels’ actual routes,
perhaps consisting of a sequence of scheduled ports.
5
Including the handful of shippers larger than 600 TEUs per year cause the mixed logit model to fail to converge.
6
We specified three different models with interactions of distance and border with shipment value, TEUs and
weight, respectively. The results turn out very similar. We report the interaction with shipment value here.
7
In the literature, the distribution of coefficient that are known to have the same sign for every decision maker, such
as price (Revelt and Train, 1998) and distance (Hastings et al., 2005) are assumed to follow a lognormal distribution.
In this study we do not expect that distance is always negatively perceived, for instance, if the process of port choice
12
advantageous as there are many aspects of shippers and particular shipments that we cannot
observe and restricting the impact of distance to be homogeneous across all shippers might
introduce specification bias. If the estimated standard deviation of the random coefficients is
sufficiently small it would suggest that heterogeneity is not a substantial problem in the data.
Table 3 reports the descriptive statistics for the sample of 32,079 shipping decisions. On
average wine ships approximately 225 miles to reach the port of departure and approximately
twenty-two percent of all decisions entail crossing a national border. The average throughput of a
port is approximately 8,777 TEUs per month, the average distance from the port of departure to
the port of destination in the United States is approximately 4,400 miles, approximately fortyfour percent of all port decisions correspond to the “home port,” and approximately fifty-two
percent of all decisions are a repeat choice.
4. Empirical Results
Table 4 reports the estimates of the mixed logit models for the four subsamples of
shippers. The coefficient estimates differ significantly across the four groups and within each
group the random coefficients have significant variances.
Distance
The average shipper in each group places a negative value on distance, but the effect is
different for each group. Each additional mile of distance reduces the odds of choosing a
particular port by approximately 0.50% for the smallest shippers and approximately 0.9% for
large shippers. The impact of distance is less important and generally less variant in the first
group as reflected in the relatively low value of the standard deviation in estimated
heterogeneous preferences. Distance becomes more important for larger shippers. The variation
is done under consideration of contractual relationships between shippers and ports. Therefore, we assume that
preferences are normally distributed.
13
of the impact of distance increases over groups; it is high relative to the mean in the two groups
of largest shippers.
The closest port is not significant in the choice of port by the smallest shippers and is
negatively related to the choice of port by the next smallest shippers. Only for the larger shippers
is homeport positive and statistically significantly related to port choice. These differences might
reflect the non-linear impact of distance for shippers of various size.
Overall, distance is an important factor in port choice for the average shipper and there is
significant heterogeneity across shippers in their valuation of distance, some shippers value
distance negatively and five times larger than the average shipper, while some shippers place a
positive value on distance. Using the cumulative standard normal distribution, given the mean
and standard deviation of the distance coefficients, in the first two groups almost all shippers
place a negative value on distance but that nineteen percent of Group 3 and thirty-six percent of
Group 4 place a positive value on distance, most likely reflecting infrastructure or idiosyncratic
connections between shippers and ports that are unobserved in the data.
All groups of shippers place negative weight on oceanic distance, but always smaller than
the weight on land distance. This makes sense given lower overseas transportation costs
(Hummels, 2001) and that large temperature fluctuations, which can damage wine, are more
likely to occur on the land than the water (Marquez, et al., 2012)
Performance
As discussed by Hess and Polak (2005), the linear specification of the underlying utility
function of the mixed logit may not be appropriate for attributes with decreasing marginal
returns. Here it is reasonable to assume that improvement at a lower level of performance
14
increases utility more than improvement at a higher level of performance. Hence we use the
natural logarithm transformation to introduce a non-linear impact of performance to port choice.
A Normal distribution is assumed for the coefficient on performance. The results indicate
that the preference for port performance is rather low for smaller shippers but increases with the
shipper size.8 The heterogeneity in preference for performance is substantially large, where the
standard deviation is estimated to be three times larger than the mean in Group 2. We assume
that this variation in preference for port performance arises due to other port characteristics that
are not observed in this study. In the next section we control for port country fixed-effects that
control for all the fixed (during the sample period) port characteristics in terms of performance,
costs, and wine handling facilities.
Border
A large body of literature investigates trade-barrier effects caused by borders and the
puzzle that border effects are found even in integrated and free trade regimes, such as the
European Union and NAFTA. The results in Table 4 indicate that an average European wine
shipper perceives a border crossing as a negative regardless of shipper size, but that the impact of
a border crossing decreases with shipper size. For a shipment with average value, a border is a
strong barrier for the smallest group but the borders impact decreases with shipment value.
Although the border is a smaller barrier to larger shippers, its negative impact increases with the
value of the shipment. There is considerable heterogeneity across shippers in their sensitivity to
crossing a border.
8
We also interacted performance and shipment value to test whether a higher value of a shipment increases the
preference for more efficient ports. However, this variable turns out to be insignificant across the four groups, and
was removed from the reported results.
15
Variation of Border Effects across Countries
To explain the negative and heterogeneous effect of a border crossing in the free trade
zone of Europe we include interactions of bordercross and country fixed effects to control for
the local business culture, local regulations and norms, and local business climate.
These
interactions identify movement across one of the four country borders from within the country.
For example, From_Italy_border equals one if an Italian shipper chooses a port outside of Italy
and zero otherwise. We also create indicator variables for the countries hosting the most popular
ports (Italy, Spain, France, Belgium, and the Netherlands) and interact them with bordercross.
These new interactions allow for asymmetric border effects depending on whether a domestic
shipper is leaving a particular country or a non-domestic shipper is coming into a particular
country. For example, To_Italy_Border takes a value of one if a non-Italian shipper were to
choose an Italian port. The main bordercross variable is replaced by four from-within
interactions and five to-outside interactions. In both cases, countries such as Portugal,
Netherlands, and Greece comprise the “omitted category.” For the landlocked countries a border
crossing is required and therefore the bordercross interactions have zero variation and are
therefore not included. The results are reported in Table 5.
In Table 5 only distance has a randomly distributed coefficient. The large variation of the
preference for border and performance observed in Table 4 is captured by the interactions of
bordercross and country fixed effects. The results show that the border effects from within vary
across countries: Spanish shippers are least reluctant to crossing their own border, as reflected in
the lowest distance equivalents of the border, whereas the Italians are the most reluctant, as
reflected in the highest distance equivalents of the border. Furthermore, the border effect from
outside a country varies: the borders of the Netherlands, Spain, and Belgium are perceived to be
16
the least costly followed by France, and Italy. As can be seen, the impact of a port’s national
border is almost always positive, reflected in the negative distance equivalents of the border
crossing, suggesting that the country attracts more shipments relative to the omitted countries.
Only for the largest shippers do the borders of France and Italy dissuade shipments from other
countries.
Next, we specifically test if shippers differentiate between equidistant ports but one port
requires a border crossing. To do so, we interacted distance and border effects in Table 5.
Groups 1, 2, and 4 evaluate distance more negatively when there is a border to cross. In other
words, a port that is located across a border is treated as farther than its actual physical distance.
Trade-off between Proximity, Port Performance and Border
While the results from the mixed logit models provide some insight about how borders
influence decision making, it is of interest to see how decision makers weigh different port
characteristics on the margin. When using the mixed logit, the trade-off between distance and
port performance is simply the ratio of the two parameter estimates; the result is the amount port
performance must increase to offset a given distance increase and keep the decision maker
indifferent about changing their port decision.
The estimated trade-offs between proximity and port performance (evaluated at each
parameter's mean) is presented in Table 6.9 The first row lists the increase in performance that
would compensate an average shipper to travel 100 more miles to reach the port. For the average
shipper in the smallest group, the port has to increase its performance by 1.2 percent to
compensate traveling 100 more miles. The tradeoff between distance and port performance is
9
We follow Hastings, et al. (2006) for computing trade-off between attributes in the context of mixed logit model;
where proximity equivalent to border is calculated by
.
17
fairly steady across the three remaining groups although Group 2 requires a somewhat larger
increase in port performance to compensate for additional distance.
The amount a port needs to improve its performance to compensate a shipper for crossing
a border is summarized in the second row of Table 6. The trade-off decreases as shippers become
larger. Finally, the distance equivalents of a generic border crossing in the shipping market for
wine is reported in the third row of Table 6. On average, the smallest shippers are willing to
travel approximately 625 more miles to avoid a border; this value decreases with the size of the
shippers. The final row of Table 6 reports the trade-off for an average shipper with shipment
values one standard deviation above average.10 An increase in the value of a shipment decreases
the trade-off between distance and border in the smallest group and increases this value in the
largest group.
Trade-off between Distance and Border across the Countries
Table 7 reports the distance equivalents estimated from the mixed logit results when
border effects are allowed to by asymmetric (Table 5). Overall, borders dissuade outbound
shipments but encourage inbound shipments, precisely because these shipments are heading to a
port (as reflected in the negative sign of the calculated trade-offs). For all but the largest shippers
leaving the home country for a port in another country is costly, as reflected in the positive
distance equivalents. This might reflect other non-formal trade barriers such as language,
customs, or social connections. For example, the average Italian shipper with less than 5 TEU
considers crossing the Italian border to reach a port in another country as the equivalent of 836
miles in land transportation. This pattern is the same for all domestic movements regardless of
the size of the shipper.
10
The trade-off at one standard deviation above the mean is calculated by
18
.
On the other hand, shippers outside of Italy, France, and Spain consider crossing into any
of those countries as the equivalent of fewer land transport miles as reflected in the negative
distance equivalents. The results suggest that the ports in our sample are strong draws for
domestic and foreign shippers. This might be caused by port-specific infrastructure to handle
wine shipments. Another explanation is that those outside a country are slightly less concerned
into that particular because they have options, that is, an Austrian shipper can go to Italy,
Germany, or France, and therefore any particular country’s border is less of an issue.
4. CONCLUSIONS
As pointed out by Disdier and Head (2008), the large literature that investigates the impact of
distance and national borders on bilateral trade flows has yielded a wide range of estimated
impacts. They attribute the increasing variance in estimated effects to three issues: sampling
error, structural heterogeneity, and methodological differences. This paper offers an estimate of
distance and national borders on the flow of a single commodity in the early stages of the supply
chain using unique data characterized by structural heterogeneity and a methodology that
accommodates that heterogeneity.
We empirically model the choice of European wine shippers on what port to use when
shipping wine to the United States. Our data describe the nature of the shipment, i.e., its size and
value, the shipper's location, the port's location, and port characteristics. Furthermore, we know
the distance and characteristics of the other ports in Europe that weren't selected. Therefore, the
transactions we investigate can be domestic or international and therefore distance and national
borders might be expected to influence port choice. The data used here have several advantages
but most importantly being able to precisely measure actual and potential distance travelled.
19
Focusing on a relatively homogeneous product reduces the potential for attributing the same
distance and border effects across very different types of decision makers.
Using a mixed logit that accommodates heterogeneous preferences, we first investigate
how distance, port performance, and crossing a national border influence port choice across
shippers of different size assuming a common border effect across all countries and that border
effects are symmetric on both sides of the border. We find that port performance is an important
influence on port choice but so too is distance and crossing national borders. We then estimate
the implied tradeoff between performance and distance, performance and crossing a border, and
distance and crossing a border, the latter often referred to as the distance equivalents of a border.
We find that the border effect varies from approximately 90 miles to 560 miles of equivalent
land travel depending on the size of the shipper.
We next relax the assumption that all border effects are the same and that border effects
are symmetric by identifying particular borders that have to be crossed to reach a port and
whether one is leaving or entering a particular country. We find that the border effects are not the
same across countries nor are they the same when you are entering or leaving a country.
Measured in distance equivalents, it is expensive to leave Italy, France, and Spain but it is cheap
to enter the Netherlands, Spain, and Belgium. We find that estimated border effects never exceed
a thousand miles although the vast majority of the estimated border effects are statistically
significant. Thus, even within the free-trade Euro zone, non-trivial border effects still exist. We
do not have the data to speculate as to why this is the case but this is an obvious direction of
possible future inquiry.
While the border effects we estimate are substantially smaller than those estimated in
aggregated gravity models, we appeal to the taxonomy of Disdier and Head (2008) and point out
20
that our study is focused on a structurally heterogeneous group of decision makers (wine
shippers) and we use a different methodology (mixed logit) than would be used in a traditional
gravity model of bilateral trade. Furthermore, we point out that the wine shipments included in
this sample are only on the first stage of the overall shipment process from producer to endconsumer. Therefore, it is not surprising that our estimated border effects are substantially
smaller than those obtained using aggregated trade that includes the entire shipping journey from
producer to end-user.
21
REFERENCES
Anderson, James E. 2010. “The Gravity Model.” Working paper National Bureau of Economic
Research Working paper 16576.
Anderson, James E, and Eric van Wincoop. 2003. “Gravity with Gravitas: A Solution to the
Border Puzzle.” American Economic Review 93(1): 170–192.
Bardají, Isabel, and Samir Mili. 2009. “Prospective Trends in Wine Export Markets – Expert
Views from Spain Prospective Trends in Wine Export Markets – Expert Views from
Spain.” International Association of Agricultural Economists: 16–22.
Behar, Alberto, and Anthony J Venables. 2011. “Transport Costs and Costs and International
Trade.” In A Handbook Of Transport Economics, eds. André de Palma, Robin Lindsey,
Emile Quinet, and Roger Vickerman. Cheltenham: Edward Elgar, p. 97–115.
Behrens, Christiaan, and Eric Pels. 2012. “Intermodal competition in the London–Paris
passenger market: High-Speed Rail and air transport.” Journal of Urban Economics 71(3):
278–288.
Beugelsdijk, Sjoerd, and Ram Mudambi. 2013. “MNEs as border-crossing multi-location
enterprises: The role of discontinuities in geographic space.” Journal of International
Business Studies 44(5): 413–426.
Blonigen, Bruce A, and Wesley W. Wilson. 2007. “Port Efficiency and Trade Flows.” Review of
International Economics 16(1): 21–36.
Brun, Jean-Francois, Carrere Celine, Patrik Guillaumont, and Jaime De Melo. 2005. “Has
Distance Died? Evidence from a Panel Gravity Model.” The World Bank Economic Review
19(1): 99–120.
Chandes, Jérôme, Dominique Estampe, Dean Isli, Romain Berthomier, Laure-anne Courrié,
Lingli Han, and Sylvain Marquevielle. 2003. “of Actors in the Wine Supply Chain.” Supply
Chain Forum: An International Journal 4(1): 12–27.
Chen, Natalie. 2004. “Intra-national versus international trade in the European Union: why do
national borders matter?” Journal of International Economics 63(1): 93–118.
Clark, X, D Dollar, and a Micco. 2004. “Port efficiency, maritime transport costs, and bilateral
trade.” Journal of Development Economics 75(2): 417–450.
Coe, David T, Arvind Subramanian, and Natalia T Tamirisa. 2007. “The Missing Globalization
Puzzle: Evidence of the Declining Importance of Distance.” IMF Staff Papers 54(1): 34–58.
22
Combes, Pierre-Philippe, Thierry Mayer, and Jacques-François Thisse. 2008. Economic
geography: The integration of regions and nations. Princeton.
Cullinane, Kevin, and Dong-Wook Song. 2006. “Estimating the Relative Efficiency of European
Container Ports: A Stochastic Frontier Analysis.” Research in Transportation Economics
16(06): 85–115.
Depken, Craig A., and Robert J. Sonora. 2002. “International price volatility: Evidence from
U.S. and Mexican cities.” The North American Journal of Economics and Finance 13(2):
179–193.
Disdier, Anne-Célia, and Keith Head. 2008. “The Puzzling Persistence of the Distance Effect on
Bilateral Trade.” Review of Economics and Statistics 90(1): 37–48.
Engel, Charles, and John Rogers. 1996. “How Wide is the border.” American Economic Review
86(5): 1112–1125.
Fotheringham, A.S., and M.E. O’Kelly. 1989. Spatial Interaction Models: Formulations and
Applications. Dordrecht, The Netherland: Kluwer Academic Publishers.
Garcia, Fernanda a., Martin G. Marchetta, Mauricio Camargo, Laure Morel, and Raymundo Q.
Forradellas. 2012. “A framework for measuring logistics performance in the wine industry.”
International Journal of Production Economics 135(1): 284–298.
Hastings, Justine S, Thomas J Kane, Douglas O Staiger, Steven Berry, Philip Haile, Fabian
Lange, Sharon Oster, Miguel Villas-boas, and Sofia Villas-boas. 2005. “Parental
Preferences and School Competition: Evidence From A Public School Choice Program.”
Working paper National Bureau of Economic Research Working paper 11805.
Helliwell, John F. 1996. “Do National Borders Matter for Quebec’s Trade?” Working paper
National Bureau of Economic Research Working paper 5215.
———. 1998. How much do national borders matter? Brookings Institution Press.
Hensher, David A, and William H Greene. 2002. “The Mixed Logit Model: The State of
Practice.” 2002(June): 1–49.
Hess, Stephane, and John W. Polak. 2005. “Mixed logit modelling of airport choice in multiairport regions.” Journal of Air Transport Management 11(2): 59–68.
Hillberry, Russell, and David Hummels. 2005. “Trade Responses to Geographic Frictions : A
Decomposition Using Micro-Data.” Russell The Journal Of The Bertrand Russell Archives.
Hummels, David. 1999. “Have international transportation costs declined?” University of
Chicago
23
———. 2001. “Toward a Geography of Trade Costs.” Purdue University
Leamer, Edward E., and James Levinsohn. 1995. “International trade theory: the evidence.” In
Handbook of international economics, eds. GM Grossman and K Rogoff. New York:
Elsevier.
Malchow, Matthew B, and Adib Kanafani. 2004. “A disaggregate analysis of port selection.”
Transportation Research Part E: Logistics and Transportation Review 40(4): 317–337.
Marquez, Leorey, Simon Dunstall, John Bartholdi, and Alejandro Maccawley. 2012. “‘ Cool or
Hot ’: A Study of Container Temperatures in Australian Wine Shipments.” Australasian
Journal of Regional Studies 18(3): 420–443.
Mccallum, John. 1995. “National Borders Matter : Canada-U . S . Regional Trade Patterns.”
McCallum J 85(3): 615–623.
Nitsch, Volker. 2000. “National borders and international trade: evidence from the European
Union.” Canadian Journal of Economics/Revue Canadienne d`Economique 33(4): 1091–
1105.
Parsley, David C., and Shang-Jin Wei. 2001. “Explaining the border effect: the role of exchange
rate variability, shipping costs, and geography.” Journal of International Economics 55(1):
87–105.
PIERS. 2007. July 1, 2006 - June 30, 2007- PIERS TI U.S. Waterborne Imports Dataset [Data
file]. UBM Global Trade, Newark, NJ.
Revelt, David, and Kenneth Train. 1998. “Mixed Logit with Repeated Choices: Households’
Choices of Appliance Efficiency Level.” Review of Economics and Statistics 80(4): 647–
657.
Tang, Loon Ching, Joyce M. W. Low, and Shao Wei Lam. 2008. “Understanding Port Choice
Behavior—A Network Perspective.” Networks and Spatial Economics 11(1): 65–82.
Tiwari, Piyush, Hidekazu Itoh, and Masayuki Doi. 2003. “Shippers’ Port and Carrier Selection
Behaviour in China: A Discrete Choice Analysis.” Maritime Economics & Logistics 5(1):
23–39.
Tongzon, Jose L. 2009. “Port choice and freight forwarders.” Transportation Research Part E:
Logistics and Transportation Review 45(1): 186–195.
Train, Kenneth E. 2003. Discrete Choice Methods with Simulation. Cambridge: Cambridge
University Press.
Turrini, Alessandro, and Tanguy van Ypersele. 2010. “Traders, courts, and the border effect
puzzle.” Regional Science and Urban Economics 40(2-3): 81–91.
24
Veldman, Simme, Lorena Garcia-Alonso, and José Ángel Vallejo-Pinto. 2011. “Determinants of
container port choice in Spain.” Maritime Policy & Management 38(5): 509–522.
Wanke, Peter F., Rafael Garcia Barbastefano, and Maria Fernanda Hijjar. 2011. “Determinants
of Efficiency at Major Brazilian Port Terminals.” Transport Reviews 31(5): 653–677.
Wolf, Holger C. 1997. “Patterns of Intra- and Inter-state trade.” National Bureau of Economic
Research 5939.
25
Table 1: Country of origins included in the analysis
Country of Origin
ITALY
FRANCE
SPAIN
GERMANY
PORTUGAL
NETHERLANDS
IRELAND
TEUs
38,266
18,713
6,567
3,204
1,236
648
237
Country of Origin
GREECE
AUSTRIA
SLOVENIA
BELGIUM
SWITZERLAND
ROMANIA
HUNGARY
26
TEUs
223
216
205
114
93
84
78
Table 2: Sample Ports and Wine Shipments.
Port Name
(Country)
Leghorn
(Italy)
La Spezia
(Italy)
Le Havre
(France)
Barcelona
(Spain)
Antwerp
(Belgium)
Genoa
(Italy)
Fos
(France)
Rotterdam
(Netherlands)
Valencia
(Spain)
Naples
(Italy)
Algeciras
(Spain)
Bremerhaven
(Germany)
Lisbon
(Portugal)
Sines
(Portugal)
Piraeus
(Greece)
Gioia Tauro
(Italy)
Number of wine shipments
(Percentage of Total Shipments)
10,200
(22%)
8,523
(18%)
7,619
(16%)
3,965
(8%)
3,902
(8%)
3,559
(7%)
3,348
(7%)
2,587
(5%)
1,117
(2%)
528
(1%)
414
(0.8%)
344
(0.7%)
326
(0.7%)
271
(0.6%)
186
(0.4%)
181
(0.4%)
27
Table 3: Descriptive Statistics of the Sample
Mean
Std. Dev.
Min
Max
Distance
226.30
194.63
0
1453.82
Border cross (1=Yes)
0.22
0.42
0
1
Performance (log of port TEUs)
9.08
0.76
6.60
10.63
Distance of European port of origin to US port of destination
4414.09
775.03
3224.12
6905.23
Home port (1=Yes)
0.44
0.50
0
1
Lagchoice (1=Yes)
0.52
0.50
0
1
Value x distance
14.22
284.90
-994.56
8359.68
Valuex border
0.03
0.55
-1.03
28.80
Notes: Sample includes 32,079 observations. Shipment value normalized to have a mean of zero and a standard deviation of one.
28
Table 4: Mixed Logit Estimation Results Assuming Symmetric Border Effects
Group 1
Group 2
Group 3
Group 4
TEUs<5
5=<TEUs<50
50<=TEUs<300
300<TEUs<600
Distance
Mean
SD
-0.005 (0.0002)
0.002 (0.0002)
-0.013 (0.000)
0.007 (0.0003)
-0.014 (0.0005)
0.016 (0.0007)
-0.009 (0.0005)
0.14 (0.001)
Border cross
Mean
SD
-3.134 (0.12)
2.339 (0.14)
-2.151 (0.15)
3.220 (0.15)
-1.263 (0.09)
3.490 (0.12)
-1.635 (0.12)
2.854 (0.18)
Performance (log)
Mean
SD
0.421 (0.02)
0.500 (0.05)
0.579 (0.06)
1.496 (0.08)
1.097 (0.05)
2.926 (0.10)
0.662 (0.06)
0.760 (0.05)
Distance port to US port
-0.001
-0.002
-0.003
-0.004
(0.000)
(0.0001)
(0.0001)
(0.0002)
-0.006
-0.259
0.427
0.811
(0.03)
(0.05)
(0.05)
(0.06)
Lagchoice
1.489
1.138
0.689
0.535
(0.05)
(0.03)
(0.03)
(0.05)
Val * distance
-0.00004
-0.0003
-0.0002
0.002
(0.0001)
(0.0002)
(0.0001)
(0.0003)
Val * border
-0.042
0.221
0.164
-1.039
(0.06)
(0.05)
(0.06)
(0.12)
Number of observations
162048
149104
140656
61456
Number of shippers
3072
2479
147
34
Notes: Numbers in parentheses are standard errors. All parameters in bold are statistically significant at the 5%
level.
Home port
29
Table 5: Mixed Logit Results Allowing for Asymmetric Border Effects.
Group 1
Group 2
Group 3
Group 4
TEUs<5
5<TEUs<50
50<TEUs<300
300<TEUs<600
-0.0045 (.0002)
-0.0084 (.0004)
-0.021 (.001)
-0.006 (.001)
0.003 (.0002)
0.006 (.0004)
0.018 (.001)
0.012 (.001)
Home port
-0.048 (.033)
-0.034 (.038)
-0.046 (.044)
1.012 (.068)
Performance (log)
0.380 (.029)
0.258 (.034)
0.124 (.039)
1.014 (.064)
From Italy border
-3.780 (.184)
-2.810 (.220)
-2.813 (.301)
-4.355 (.595)
From Spain border
-0.912 (.130)
-0.784 (.163)
-0.569 (.188)
2.006 (.424)
From France border
-3.107 (.111)
-2.425 (.135)
-3.227 (.164)
-1.552 (.315)
From Germany border
-1.278 (.199)
-0.476 (.307)
-2.254 (.541)
3.264 (.713)
To Italy border
0.460 (0.126)
0.967 (.147)
1.156 (.184)
-2.022 (.613)
To Spain border
1.604 (.113)
1.845 (.118)
2.176 (.140)
1.451 (.336)
To France border
1.500 (.104)
1.271 (.116)
1.583 (.164)
-0.490 (.263)
To Belgium border
1.296 (.117)
1.864 (.128)
2.298 (.154)
0.719 (.176)
To Netherlands border
1.588 (.119)
2.032 (.131)
2.793 (.163)
0.989 (.170)
-0.0002 (.0001)
-.00008 (.0001)
-0.0001 (.0002)
0.001 (.0003)
0.104 (.030)
0.111 (.035)
-0.058 (.044)
-0.952 (.128)
-0.001 (.0001)
-0.002 (.0001)
-0.003 (.0002)
-0.004 (.0003)
1.624 (.051)
1.54 (.028)
1.167 (.028)
0.600 (.052)
-0.002 (.0003)
-0.002 (.0002)
0.0002 (.0003)
-0.006 (.009)
Distance
Mean
SD
Val x distance
Val x border
Distance port to US port
Lagchoice
Distance x border
Notes: Numbers in parentheses are standard errors. All parameters in bold are statistically significant at 5%.
30
Table 6 Trade-offs between Performance, Distance, and Border Crossing
Shipment
value above
average
Between performance and distance (percentage)
NO
Between performance and border (percentage)
NO
NO
Group 1
Group 2
Group 3
Group 4
1.2
(0.000)
-7.44
(0.48)
626.47
(36.90)
2.3
(0.002)
-3.71
(0.42)
161.50
(14.20)
1.2
(0.000)
-1.15
(0.09)
92.31
(7.59)
1.4
(0.001)
-2.47
(0.26)
172.81
(68.46)
557.64
(35.50)
148.62
(14.28)
93.44
(8.50)
359.95
(39.65)
Between distance and border (miles)
YES
Notes: Values in parentheses represent standard errors. All values in bold significant at the 5% level. Shipment values are
standardized to have zero mean and standard deviation of one.
31
Table 7: Distance Equivalents of Entering and Exiting EU Countries for Wine Shipments
From Italy border
From France border
From Spain border
From Germany border
To Italy border
To France border
To Spain border
To Belgium
To Netherlands
Group 1
Group 2
Group 3
Group 4
836.40
(52.56)
201.88
(28.40)
687.36
(33.99)
282.67
(44.08)
-101. 71
(28.08)
-331.70
(26.21)
-354.81
(28.64)
-286.77
(27.99)
-351.39
(29.82)
336.21
(31.81)
93.85
(19.31)
290.09
(21.07)
59.98
(36.63)
-115.66
(18.68)
-152.04
(15.76)
-220.69
(18.38)
-222.96
(19.55)
-243.146
(20.51)
134.92
(16.01)
27.32
(8.75)
154.78
(11.19)
108.08
(24.97)
-55.46
(9.02)
-75.94
(7.74)
-104.38
(9.18)
-110.21
(9.68)
-133.96
(10.59)
748.99
(210.55)
-345.08
(119.58)
267.09
(80.05)
-561.48
(194.02)
347.88
(138.02)
84.34
(49.74)
-249.61
(85.87)
-123.61
(43.64)
-169.91
(52.14)
Note: Values in parentheses are standard errors. All values in bold significant at the 5% level.
32
Figure 1: Spatial Distribution of European Ports of Origin and Wine Shippers
1a) U.S.-bound containerized wine traffic of major European
ports (TEUs)
1b) Spatial distribution of shippers, weighted by the size
of wine shipment between 7/2006 and 6/2007
33
Figure 2: Distribution of Wine Shippers by Annual TEUs Shipped
34