The empirical consequences of trade sanctions for directly
and indirectly aected countries
Jonas Frank, M. Sc.
∗
University of Hohenheim
July 29, 2016
Preliminary Version, Please Do Not Cite
Abstract
Economic sanctions are a popular diplomatic tool for countries to enforce political demands abroad
or to punish non-complying countries. There is an ongoing debate in the literature if this tool
is eective in reaching these goals. This paper adds to the literature by treating sanctions like a
negative form of trade agreements. In order to quantify the direct eects of sanctions on the trade
ows between countries I make use of a gravity equation controlling for country pair, importeryear, and exporter-year xed eects. The estimates reveal that there is a signicant decrease in the
value of trade after the introduction of sanctions. In a second step, trade diversion is introduced
as a potential instrument for countries to soften the negative impact of sanctions. However, the
estimates reveal no evidence for trade diversion.
Keywords: gravity, international trade, trade sanctions
∗
Corresponding email: [email protected]
1
Introduction
Trade sanctions and embargoes have always been a popular instrument of diplomatic behavior against
ill-behaving states and continue to be popular today. Currently, there is an open debate on whether
or not economic sanctions should be applied against Turkey, to keep the regime from moving further
away from democratic principles. Also, after the annexation of the Crimea by the Russian Federation,
the European Union, the United States of America, and several other states were quick to implement
sanctions against Russia. Russia, in turn, reacted by implementing multilateral trade sanctions on its
own, specically, a total ban on food imports from the EU, USA, Norway, Australia, and Canada.
Are economic sanctions an eective tool of enforcing the sender-countries' goals? This question has
been widely discussed in the literature with very mixed results and, so far, inconclusive evidence. By
taking a look at the world right now, sanctions do not seem to be very eective. Russia has not taking
any steps to undo the annexation of the Crimea, North Korea keeps testing missiles, and Cuba did
not change their regime, even though these countries are subjected to drastic sanctions from many
countries across the globe.
This paper adds to the sanctions literature by linking it to the trade-creation and trade-diversion literature. In a rst step, I quantify the eect of sanctions and potential counter-sanctions on international
trade empirically by means of gravity equations with dierent specications. I argue that the implementation of sanctions can be treated similarly to the formation of a regional trade agreement between
two countries, but in the opposite direction. Instead of abolishing taris and streamlining standards
to facilitate trade, it is possible to treat a sanction like the introduction of an innitely high tari
that prevents countries from trading specic goods or from trading all together. The results show that
sanctions indeed have a signicant negative impact on the aggregate trade values between countries.
Since I am interested in consequences for indirectly aected countries, in a second step control for the
possibility of trade diversion. If sanctioned and sanctioning countries are able to simply divert their
trade to other partners at little cost, then this can potentially oset the negative sanctions mechanism.
Also, if the targeted country can divert trade more easily than the sender, it may lead to cases where
the punishing country suers more economic damage than the country that is supposed to be punished.
However, there seems to be no evidence supporting that hypothesis in the data.
The remainder of the paper is structured as follows. First, some evidence from the literature is given,
regarding both, sanctions and trade creation and trade diversion. Next, the data and empirical methods
are explained. Chapter 4 presents the results and chapter 5 concludes.
1
2
Brief summary
During the last two decades several researchers tried their hands at explaining the consequences of
economic sanctions on trade from various angles. In 2003, Hufbauer and Oegg measured the damage
of sanctions on US trade in 1995 and 1999, using the
naive
gravity specication. The impact of sanc-
tions on aggregate trade over the time frame from 1960 until 2000 has been analyzed by Caruso (2003),
employing a similar estimation strategy. They control for various aspects that aect trade but do not
take the multilateral resistance terms into account, which most likely biases their results. A detailed
and instructive summary of economic sanctions and their goals in the 20th century is given by G. Hufbauer, Schott, Elliott, and Oegg (2009). The Threat and Imposition of Sanctions database (TIES) by
Morgan, Bapat, and Kobayashi (2014) also contains specic records of cases of sanction threats and
impositions.
Other authors focused specically on the eect of sanctions on targeted countries, for
example the consequences for Iran after the oil boycott Dizaji and van Bergeijk (2013). Crozet and
Hinz (2016), on the other hand, concentrate on the costs of imposing and maintaining sanctions for
the sender countries.
In his model Viner (1950) shows that, after forming a regional trade agreement (RTA) with a new
partner, the welfare of members of the RTA does not necessarily has to improve. It is possible that
trade is simply shifted away from another, more ecient supplier and not newly generated due to
realized gains from trade. This would lead to ineciency in world production. This model started a
debate on trade creation and trade diversion, both, empirically and theoretically. An overview until
the year 2010 is given by Freund and Ornelas (2010).
3
Data and empirical methods
The data set at hand covers the years from 1990 to 2006. The information on sanctions stems from
TIES. It contains detailed information on the implementation, duration, and type of specic sanctions
between countries until the year 2005. The authors dierentiate between 10 types of sanctions, which I
have grouped into three categories, following G. C. Hufbauer and Oegg (2003), namely extensive, moderate, and limited sanctions, respectively. Extensive sanctions contain only total economic embargoes
and blockades, e.g. those against North Korea and Cuba. Sanctions regarding partial economic embargoes, specic import and export restrictions and suspension of trade agreements are combined within
moderate sanctions. Finally, limited sanctions refer to travel bans, termination of foreign aid and asset
freezes. If a country has multiple sanction types in place, I only count the most severe. Information
on bilateral trade on the country level is provided by Feenstra (XX) and the common gravity controls,
like distance, GDP, common language, and trade agreements, stem from CEPII (Head, Mayer, and
Ries (2010),Head and Mayer (2013)). I end up with a sample size of 292,621 observations and 27,319
(non-singleton) country pairs.
The typical gravity specication with time xed eects is given below:
ln(mijt ) = β1 ln(Yit ) + β2 ln(Yjt ) + β3 ln(distij ) + β4 sancijt + β5 tdijt + γXijt + αt + ϵijt ,
2
(1)
with
i's
mijt
and
giving the value of imports of target
j 's
GDP in year
t,
respectively.
countries. The sanction-dummy
from country
j
at year
t,
sancijt
i
distij
from sender
j
in year
t. Yit
Yjt
and
denote country
is the geographical distance between both trading
takes the value of 1 if there is an active sanction in country
i
and zero otherwise. In the regression I will also use dummies for the three
intensity measures from G. C. Hufbauer and Oegg (2003) instead of the sole sanction dummy. Trade
diversion (tdijt ) takes the value of 0 if an active sanction takes place in country
year
t.
i
from country
j
in
The dummy variable is 1 if a country is either the sender or the target of sanctions but the
trade partner is not directly aected. It is also zero if both countries do not actively apply sanctions.
Xijt
incorporates the vector of usual dummy variables for common language, contiguity, and colonial
ties. The time xed eect is included via
αt .
Finally, the error term is denoted by
ϵijt .
This approach, however, most likely suers from omitted variable bias due to unobserved country-pair
characteristics that aect trade ows other than the ones already included, mainly due to unobserved
multilateral resistance. It could be possible, for example, that country pairs that have strong historical
ties potentially also have higher trade value and are less likely to implement strict sanctions.
In the next specication I will therefore include country-pair xed eects,
αij :
ln(mijt ) = β1 ln(Yij ) + β2 ln(Yjt ) + β3 SAN Cijt + β4 T Dijt + γXijt + αt + αij + ϵijt
(2)
The country-pair xed eects capture all time invariant country-pair specic inuences on trade, both,
observable and unobservable.
Specications (1) and (2) do not take into account shocks that independently aect the importer or
the exporter in a given year, such as changes in legislature due to election outcomes which could either
be a boost or a hindrance to trade. To account for this possibility, equation (3) includes exporter-year
and importer-year xed eects
αit
and
αjt ,respectively.
ln(mijt ) = β1 SAN Cijt + β2 RT Aijt + αij + αt + αit + αjt + ϵijt
(3)
The xed eects now capture, additionally to GDP and population, also factor endowments, openness,
and other time specic shocks, some of which are quite hard to observe. The only explanatory variables
uncontrolled by the xed eects left are the dummies for sanctions and RTAs.
An alternative way to avoid mis-specication is rst dierencing.
Taking the rst dierence of the
estimation equation (4) takes care of all country pair xed eects and yields a dierence-in-dierences
estimator, that measures the changes on trade value, if and when a country pair implements sanctions
(and stops them again). The same interpretation holds for the RTA-dummy.
∆ln(mijt ) = β1 ∆sancijt + β2 ∆rtaijt + αit + αjt + δϵijt
3
(4)
4
(Preliminary) Results
This section presents the (preliminary) results of the empirical estimation. The rst subsection consists
of results solely seeking to quantify the trade destruction eect that sanctions possibly have on the
previously introduced model specications.
The trade diversion dummy is therefore left out of the
specications. The second part estimates the eects of potential trade diversion in the face of sanctions,
also estimating dierent models.
4.1
Trade destruction
In table 1 the results of the four gravity specications are given. Each specication includes year xed
eects and standard errors are robust and clustered at the country-pair level. Model 1 gives the results
for estimation specication (1) and does not include dyad, sender-year, and target-year xed eects.
The basic gravity controls are of expected size and level of signicance. Both GDPs are highly significant and around 1, distance has the opposing eect. RTAs, common language, colonial background,
and contiguity all boost the value of trade signicantly.
The sanction dummy, however, appears to
have a highly signicant positive inuence on international trade. This result is very counter-intuitive.
In model 2, based on equation (2), dyad xed eects are included together with the time xed eects.
This means that now all time-invariant country-pair specic explanatory variables are accounted for,
including distance and language. Because of these xed eects, all of the unobservable variables that
potentially aect trade value are controlled for, allowing for a less biased result. It can be seen that
the inuence of GDP is still large and highly signicant, but the size of the eect has decreased dramatically. This is in line with previous studies. Also, the positive eects of regional trade agreements
have decreased and are now around a more reasonable 24 percent increase in trade value. Under this
specication, sanctions seem to have a highly signicant inuence on the value of trade. If one country
imposes sanctions on another country, the value of trade decreases by about 10 percent.
In model 3, additionally to the dyad xed eects, sender-year and target-year xed eects are included
at the cost of GDP measures. Here, equation
(3)
supported by the steady increase of the adjusted
by model specication 3.
further improves the accuracy of the estimates, also
R2 .
The results are quite similar to those obtained
Sanctions decrease the value of trade by around 10 percent, while trade
agreements increase trade value by about 23 percent. Finally, model 4 makes use of a rst-dierence
approach given by equation
(4) to eliminate the pair xed eects at the cost of one year of observations,
yielding a dierence-in-dierences estimator. This also means that the adjusted
R2
is not comparable
to the other specications. The implementation of trade sanctions results in a negative trade value
eect and its size is similar to specications 2 and 3. A change in trade agreements seems to have an
insignicant eect on the value of trade. This rather implausible result may be due to the fact that
most of the trade agreements within the sample have been active over the whole period of observation
and some countries could not be observed every year.
Table 2 presents the ndings for dierent levels of sanction intensities, limited, moderate, and extensive
sanctions. The model specications are similar to those in table 1 to allow for comparability. Again,
4
specication 1 seems to suer from omitted variable bias, rendering the use of moderate sanctions
highly signicant and positive for the value of international trade. Controlling for country-pair xed
eects changes these ndings towards more plausible results.
Whereas limited sanctions appear to
have no signicant eect on trade whatsoever, carrying out moderate sanctions decrease trade value
by around 10 percent and applying extensive sanctions destroys trade all together.
Adding sender-
year and target-year xed eects renders limited and extensive sanctions insignicant, while moderate
sanctions remain stable in signicance and similar in size. One potential reason for this may be that
the number of extensive sanctions is low and mostly remain steady over the whole sample period which
means there are not many changes to observe. RTAs in specications 4 suer from the same issues as
in table 1.
Concluding from the evidence presented here, it seems that moderate sanctions are the main driver of
the overall sanctions, decreasing trade value by about 6 to 10 percent.
4.2
Trade diversion
In order to check for trade diversion it is unfortunately not possible to make use of the well-specied
estimation equations (3) and (4). Trade diversion is a country- and time-specic unilateral variable,
so only the potentially mis-specied models 1 and 2 can be used. In table 3, therefore, only the results
for these empirical specications are shown.
Column 1 and 3 report the ndings for the trade sanctions dummy, in columns 2 and 4 the sanctions are
split into their three levels. Model specication 1 is used in column 1 and 2, specication 2 is observed
in columns 3 and 4. Model 1 again overvalues GDPs and the eect of trade agreements on the value
of trade. It also returns an implausible positive and signicant eect of sanctions on trade values. In
model 2, the dummies for extensive and limited sanctions, as well as the dummy for sanctions as a
whole, returns negative and signicant results. The trade diversion dummy stays insignicant across
all observations, thus pointing towards the interpretation that no trade diversion takes place across
the sample.
5
Table 1: Trade destruction estimates
Sanction
RTA
Model 1
Model 2
Model 3
∗∗∗
0.323
∗∗∗
-0.093
∗∗∗
-0.109
(0.050)
(0.028)
(0.024)
0.985
∗∗∗
0.216
(0.046)
Ln(GDPs )
1.121
(0.022)
∗∗∗
0.354
(0.005)
Ln(GDPt )
0.894
0.962
0.203
∗∗∗
(0.027)
∗∗∗
(0.014)
∗∗∗
0.631
(0.005)
Common language
∗∗∗
Model 4
∗∗∗
(0.014)
∗∗∗
(0.034)
Contiguity
0.859
∗∗∗
(0.083)
Colonial background
0.966
∗∗∗
(0.076)
Ln(distance)
-1.124
∗∗∗
(0.016)
D.Sanction
-0.080
∗∗∗
(0.025)
D.RTA
0.004
(0.028)
Observations
292621
292621
292621
286294
2
Adjusted R
0.654
0.872
0.881
0.036
Dyad xed eects
No
Yes
Yes
No
Sender-year & target-year xed eect
No
No
Yes
Yes
Cluster robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01
Each regression includes year xed eects
Table 2: Trade destruction evidence by sanction intensity
6
Extensive
Moderate
Limited
RTA
Model 1
Model 2
Model 3
-1.160
∗∗
-0.752
-0.445
(0.712)
(0.335)
(0.282)
0.349
∗∗∗
∗∗∗
(0.025)
0.340
0.107
-0.069
(0.238)
(0.080)
(0.101)
0.983
∗∗∗
1.121
∗∗∗
0.894
∗∗∗
(0.005)
Common Language
-0.104
(0.021)
(0.005)
Ln(GDPt )
∗∗∗
(0.051)
(0.046)
Ln(GDPs )
-0.094
0.961
0.215
∗∗∗
(0.023)
0.355
0.203
Model 4
∗∗∗
(0.027)
∗∗∗
(0.024)
0.632
∗∗∗
(0.021)
∗∗∗
(0.034)
Contiguity
0.865
∗∗∗
(0.083)
Colonial Background
0.965
∗∗∗
(0.076)
Ln(distance)
-1.125
∗∗∗
(0.016)
D.Extensive
-0.365
(0.275)
D.Moderate
-0.063
∗∗∗
(0.024)
D.Limited
-0.123
(0.086)
D.RTA
0.005
(0.028)
Observations
292621
292621
292621
286294
2
Adjusted R
0.654
0.872
0.881
0.036
Dyad f.e.
No
Yes
Yes
No
Sender-year & target-year f.e.
No
No
Yes
Yes
Cluster robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01
Each regression includes year xed eects
7
Table 3: Trade diversion estimates
Model 1
Sanction
Model 1
Model 2
∗∗∗
0.333
∗∗∗
-0.081
(0.055)
(0.025)
Extensive
-1.153
-0.745
(0.712)
Moderate
0.360
RTA
(0.025)
0.348
0.120
(0.238)
(0.081)
(0.022)
(0.022)
(0.014)
(0.014)
0.985
∗∗∗
1.120
∗∗∗
0.893
∗∗∗
0.962
∗∗∗
0.859
∗∗∗
(0.083)
Colonial Background
0.964
∗∗∗
(0.076)
Ln(distance)
∗∗∗
0.013
(0.034)
Contiguity
-0.081
0.012
(0.006)
Common Language
∗∗∗
0.011
(0.005)
Ln(GDPt )
(0.334)
0.010
(0.046)
Ln(GDPs )
∗∗
(0.056)
Limited
Trade Diversion
Model 2
-1.124
∗∗∗
0.982
∗∗∗
(0.046)
1.120
∗∗∗
(0.005)
0.893
∗∗∗
(0.006)
0.961
0.215
∗∗∗
(0.023)
0.354
∗∗∗
(0.024)
0.631
∗∗∗
0.214
∗∗∗
(0.023)
0.354
∗∗∗
(0.024)
0.632
∗∗∗
(0.021)
(0.021)
∗∗∗
(0.034)
0.865
∗∗∗
(0.083)
0.963
∗∗∗
(0.076)
-1.125
∗∗∗
(0.016)
(0.016)
Observations
292621
292621
292621
292621
2
Adjusted R
0.654
0.654
0.872
0.872
No
No
Yes
Yes
Dyad xed eects
Cluster robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01
Each regression includes year xed eects
8
5
Concluding Remarks
The goal of the paper was to quantify the eect of trade sanctions on the value of trade using the gravity equation, both well- and mis-specied. It also shed some light on the question if trade sanctions
are potentially oset by the occurrence of trade diversion.
The evidence presented in the previous section shows that, indeed, trade sanctions have a signicant
negative impact on the value of trade, mostly due to moderate sanctions, specically targeting single
sectors. Limited sanctions seem to have no measurable inuence and extensive sanctions only in some
of the specications.
It also shows that trade diversion does not play a role in these specications.
Following the results, it is not possible to blame the ineectiveness of sanctions on a diversion from
one trading partner to another one. On the other hand, this can also be due to the limited scope of
the data set used.
As a next step I want to nd more evidence on the eectiveness of dierent sanction types and,
therefore, incorporate a longer sample period to allow for more heterogeneity regarding changes per
sanction-country pair. Furthermore, I want to use weights to account for the dierent impacts of the
sanction stages. I imagine that then, the eect of extensive sanctions yields clearer results. Also, I
want to incorporate interaction terms to see if countries need some time to nd a new trading partner
and divert trade towards them.
9
References
Caruso, R. (2003). The Impact of International Economic Sanctions on Trade: An Empirical Analysis.
Peace Economics, Peace Science, and Public Policy , 9 (2), 134.
Crozet, M., & Hinz, J. (2016).
Countries' Exports.
Collateral Damage: The Impact of the Russia Sanctions on Sanctioning
Dizaji, S. F., & van Bergeijk, P. a. G. (2013). Potential early phase success and ultimate failure of
economic sanctions: A VAR approach with an application to Iran.
50 (6), 721736.
Freund, C., & Ornelas, E. (2010).
Head, K., & Mayer, T.
(2013).
Helpman & Rogo (Eds.),
Regional Trade Agreements
Journal of Peace Research ,
(No. 5314).
Gravity Equations: Toolkit, Cookbook, Workhorse.
Handbook of international economics
In Gopinath,
(Vol. 4 ed., chap. Gravity Eq).
Elsevier.
Head, K., Mayer, T., & Ries, J.
(2010).
The erosion of colonial trade linkages after independence.
Journal of International Economics , 81 (1), 114.
Hufbauer, G., Schott, J., Elliott, K., & Oegg, B. (2009).
Economic Sanctions Reconsidered
(3rd ed.).
Washington, DC: Peterson Institute for Internatioal Economics.
Hufbauer, G. C., & Oegg, B. (2003). The Impact of Economic Sanctions on US Trade: Andrew Rose's
Peterson Institute for International Economics (Policy Briefs PB03-04).
Magee, C. S. P. (2008). New measures of trade creation and trade diversion. Journal of International
Economics , 75 (2), 349362.
Gravity Model.
Morgan, T. C., Bapat, N., & Kobayashi, Y. (2014, nov). Threat and imposition of economic sanctions
1945-2005: Updating the TIES dataset.
Conict Management and Peace Science , 31 (5), 118.
Rose, A. K. (2004). Trade ? Do We Really Know That the WTO Increases Reply.
Economic Review , 94 (1), 98114.
Viner, J. (1950).
The Customs Union Issue.
The American
New York: Carnegie Endowment for International Peace.
10