The Intersection of History and Trade; Essays in Trade and Trade Policy.

Introduction

Institutions have long been identified as worthy of investigation in economics. Institu-

tions have often been cited as instrumental in determining rates of economic growth since

Adam Smith made them a central theme in The Wealth of Nations, yet, understanding

their theoretical and direct impact has always been difficult. The break through in both the

theoretical and empirical measurement of institutions came from Acemoglu et al. (2005).

Acemoglu et al. provided a framework for why institutions matter. Their paper shows that

institutions shape and incentivize market actors, this in turn organizes productivity, and

results in the observed differences in economic growth between countries. In recognizing

the importance of modern-day and historical institutions on growth results in an important

corollary.

Undeniably international trade generates institutions that are worth studying as well.

These institutions take the form of long-term commercial agreements between nations, both

modern and historical trade, that exert some effect trade values and flows of specific goods.

The central questions this dissertation grapples with is what are the effects of historical

trading institutions on trade and society? The study of international trade institutions was

formalized by Eichengreen and Irwin (1995). Eichengreen and Irwin’s paper focused on

studying the impact of previous trade institutions and international borders on future trade

patterns. Their specific focus was not only on historical trade institutions’ creation, but also

on their dissolving, and how those impacts ripple out through time.

Research into historical trade institutions have examined the impact of free trade agree-

ments, currency unions, trading blocks, and international borders. Although theoretically

sound, the empirical examination of these historical institutions are met with two important

challenges. The first is the question of if these institutions have truly developed exogenously

to international trade. To elucidate the issues of endogeneity more succinctly, the literature

has grappled with the question: does high trade flows between nations result in new political

trade agreements or does politically motivated trade agreements result in high trade flows.

Unfortunately, a large number of empirical studies of historical trading institutions have

assumed a priori that the formation of international agreements is exogenous in order to

facilitate their work (Barro and Tenreyro, 2007). Thus, more robust studies have endeavored

to find “natural experiments” in which trade agreements, monetary unions, or international

borders have formed exogenously to trade rationales.

The second, and more easily remedied, issue in empirical studies of historical trading

institutions comes from the empirical methodology employed. Traditionally, these empirical

studies have applied a gravity trade model to understanding the impact of organizations

forming on trade between nations, or the welfare implications of the formation of these bod-

ies. As the gravity trade literature has grown, a number of bias and misspecification issues

have been identified. The two most important empirical developments that have shifted the

literature are Anderson and van Wincoop (2003) and Silva and Tenreyro (2006). Ander-

son and van Wincoop (2003) illustrated a theoretically consistent gravity trade model and

identified the omission of a variables to properly capture the dynamics of international bor-

ders. Anderson and Van Wincoop coined these variables as Multilateral Resistance Terms,

and demonstrated how their omission injected bias into any gravity trade model. Silva and

Tenreyro prove that a log-linearization empirical specification results in biasing the result-

ing coefficients. The bias, the authors show, is a function of heteroskedasticity present in

most empirical trade and error corrections don’t solve the innate higher order statistical

distribution problems that result in incorrectly estimated standard errors.

The work in chapters one and two of this dissertation seeks to resolve the aforementioned

problems in empirical research focused on historical trading institutions. First, to solve the

exogeneity problem, this research utilizes Russia’s post-Soviet trade flows. The work endeav-

ors to show that the formation of the Soviet Union and it’s main trading bloc the Council for

Mutual Economic Assistance (also known as Comecon), were generated exogenously to trade

motivations. Instead these historical institutions were created for political reasons. Utiliz-

ing this natural experiment allows for an empirically clean investigation into the effects of

historical trading institutions on modern day trade flows. Secondly, following best empirical

practices outlined by the likes of Anderson and Van Wincoop and Silva and Tenreyro, the

model employed is a theoretically consistent Poisson Pseudo Maximum Likelihood Model

(PPML).

In addition to generating theoretically and empirically consistent work, chapters one and

two further push the literature forward by not simply showing that historical institutions

affect current day trade flows, but by measuring how those effects decay over time. In doing

so these chapters are able to show that historical trading institutions, well over 80 years old,

have a positive effect that decays in magnitude as time marches further away from their

removal.

Chapter one measures the lagged impact of previous membership in the Soviet Union

and Comecon on trade volume (imports plus exports) to and from Russia. The chapter’s 19

year data set is employed to generate three-year panel coefficients on each of the historical

legacy variables. These coefficients are used to examine the change in the impact of the

legacy of previous trading institutions over time. Further, this chapter develops a “back of

the envelope” estimator of institutional legacy decay on trade flows.

Chapter one finds large and persistent legacy effects on trade between Russia and former

Soviet and Comecon countries. The estimates of institutional legacy decay of the impact of

former membership in the Soviet Union were found to be 18 years while former membership

in Comecon is 11 years. The results point to the effect of the membership of the Soviet Union

is stronger than that of Comecon. The result is most likely driven by the impacts of borders

on trade. The absence of borders in the past between former members of the Soviet Union

established strong trade patterns that hold even today. The strength of this conclusion is

demonstrated by both the steeper slope of a plot of the coefficients for the USSR indicator

variable and the decay measure.

Chapter two builds off the previous chapter by applying the same technique to all the of

the components of trade, i.e. Total Trade (TT), Intra-Industry Trade (IIT), Inter-Industry

Trade (INT), Horizontal Intra-Industry Trade (HIIT), and Vertical Intra-Industry Trade

(VIIT). The chapter utilizes data on Russia’s trade with 183 countries from 1996-2018, via

a correctly specified PPML gravity trade model with measures for historical borders and

trade union membership, five different regressions are run, one for each component of trade.

Chapter two is unique in that it employs a method of decomposing total trade into it’s parts

that is both empirically free from researcher bias as to how to differentiate HIIT from VIIT,

and consistent with theoretical literature that defines HIIT and VIIT.

In building off chapter one’s results showing historical trading institutions impact on

trade, chapter two illustrates that the impact of historical institutions decays in heteroge-

neous ways. Specifically, the pattern of INT, HIIT, or VIIT decay is dependent upon the

underlying theoretical foundations which govern each component of trade. It can be observed

that variables controlling for historical trading institutions are generally positive both when

aggregating for all years as well as disaggregating the data into 3-year chunks, though not

uniformly decreasing.

Two distinct narratives emerge when looking into how the impact of historical borders

and trade unions affect current day trade flows. Firstly, historical borders present a stronger

and more lasting impact. In chapter one, the coefficients on historical borders from total

trade showed a monotonic convergence to statistical zero. Chapter two demonstrates this

convergence is mainly driven by INT not IIT. Although IIT’s coefficients on historical borders

aren’t converging to statistical zero, looking specifically at it’s components shows that VIIT

is converging to statistical zero while the HIIT seems to be on a much longer convergence

path.

The story of the impact of historical trade unions on current trade flows is much less clear.

In the previous chapter, the coefficients measuring the impact of a historical trade union was

positive and arguably converging to statistical zero in a monotonic patter. Conversely, when

disaggregating total trade the data argues that the pattern illustrated by the coefficients on

Comecon is driven by the monotonic convergences of INT and possible oscillating convergence

of IIT. These results further strengthen the argument in chapter one of that the impact of

historical trade unions is much weaker than historical borders. Investigating further into

IIT shows that while VIIT is always positive and slowly converging to statistical zero, HIIT

is driving the oscillating pattern on IIT by presenting a much more aggressive harmonic

pattern.

Chapter three returns to broad question of how does historical trading institutions effect

on current day trade and society, but focuses on the latter question of society. Particularly,

the question of the institution and advocacy for Free Trade policy within the science of

economics. Although in modern times it feels as if free trade (under some specific assumptions

and conditions) is one of the few points of agreement among economists this was not always

the case.

During the end of the 19th century a number of American students went abroad to

Germany to receive their doctorates in economics under the tutelage of the German Histor-

ical School. These economists came back and founded the American Economic Association

(AEA) and were instrumental in leading the American Progressive Movement. The Progres-

sive movement (1870-1920) was key in altering what it meant to be a professional economist

and how the federal government interacted with its citizens. As budding economists steeped

in the traditions of the German Historical School, it impregnated the Progressives with beliefs

of the importance of science in solving political and social problems. Central to their ideol-

ogy was the rejection of classical laissez-faire ideology and free trade policy. Even though

the Economic Progressives controlled nearly all aspects of the profession, crafted local and

federal economic policy, and were the leaders of every higher education institution of note

the American economics discipline still embraces free trade and not protectionism.

Progressives are a complex group of thinkers often seen as more homogeneous in their

beliefs than they were. Because of this misconception of ideological uniformity, it would be

tempting to believe that there is a moment when newer methodologies, which could articulate

the gains and losses of free trade, won over members. In truth, research paradigms/schools do

not dissolve overnight. Instead the story of the American Progressive Movement’s inability

to sway the academic discourse around free trade is emblematic to the inherent flaws present

at the “hard core” of their research program.

Free Trade, and the Economic Progressives’ goals to replace it with “scientific protection-

ism” is a metaphysical manifestation of the innate problems in their hard core. Their theories

lacked science, and although they employed the rhetoric, they were unable to budge when

confronted with newer more scientifically based theories. In the face of the growing neo-

classical movement and the fallout of WWI the Economic Progressive movement collapsed

under its own weight.

The binding thread throughout these chapters is understanding the important and nu-

anced ways in which historical institutions, whether formalized through international agree-

ments or implicit within the core of an academic discipline affect the present. Historical

institutions do not simply guide or act as way markers, but instead reach from beyond the

here and now, and shape the future that we all march towards. Like a shadow, there is no

way to be rid of them, only to appreciate and understand the unique shapes they take in

the light.

Chapter 1

The Institutional Memory of Trade

Flows: Russia as a Natural

Experiment

Travis Freidman

Bruce Elmslie

Sofia Kuznetsova

Abstract:

This paper uses a multilateral resistance gravity model to examine the historical legacy of

trading institutions utilizing the impact of membership in the ex-Soviet Union and Comecon

on current day Russia’s bilateral trade flows. The use of long-term data from 1998 to 2016

allows for examination of changes in the legacy effects over time. The main finding of the

paper is that historical patterns do matter in the determination of current trade flows of

Russia. Even though there is a declining trend in volume of trade flows between Russia and

former members of the Soviet Union and Comecon, overall historical patterns developed by

these institutions remain highly significant 26 years after the collapse of the Soviet Union.

We provide the first estimates of the legacy left by past institutions with an “institutional

legacy decay” measure.

1.1 Introduction

We are living in unprecedented times for international agreements. The United Kingdom,

after three years of negotiation officially left the European Union at the end of January 2020

(Bennett, 2020). Three days after being sworn into office, President Trump pulled the United

States out of the Trans-Pacific Partnership. In another executive order, many have called

President Trumps decision to not full judges to the Appellate Court of the World Trade

Organization as a final step in killing the trade organization (Johnson, 2019). Given the

number of trade institutions that we are seeing either exited by individual members or being

disbanded, it is worth asking how long does the impact of previous membership in a trading

institution last?

Institutions have been studied widely in economics for a long time. Yet, it was Acemoglu

et al. (2005) who provided a framework and makes a case for why institutions matter. In

short, Acemoglu et al. argue that institutions shape and incentivize market actors, this

in turn organizes productivity and results in the observed differences in economic growth

between countries. There is a large literature addressing the impact history and institutions

have on trade (Mitchener and Weidenmier, 2008; Karnups, 2008; Estevadeordal et al., 2003;

Brodzicki and Uminski, 2018) and economic growth (North, 1995; Zukowski, 2004; Campos

et al., 2016). In looking specifically at trade, the literature has found strong legacy effects

of previous institutional trading arrangements in virtually all cases that have been studied

(Eichengreen and Irwin, 1995; Anderson and Smith, 2007; Stack et al., 2019). However, the

ability to confidently determine the extent of trade persistence has been hampered by the

question of the endogeneity of the development of trade, customs, and monetary unions.

Empirical studies of free trade agreements, currency unions, or trading blocks tradition-

ally apply a gravity trade model to understanding the impact of these organizations forming

on trade between nations or the welfare implications of the formation of these bodies. The

channels in which endogineity becomes an issue for the empirical models is twofold. First,

is an implicit a priori assumption in the literature, about the exogenous formation of in-

ternational agreements. As Barro and Teenreyro state “The implicit assumption in various

empirical studies is that currency unions (or, more generally exchange rate arrangements)

are randomly formed among countries” (Barro and Tenreyro, 2007, p. 3). Second, is a reverse

causality issue. Specifically Wolf and Ritschl (2011); Baldwin and Jaimovich (2012); Keller

and Shiue (2014) give different examples of how the application of national arrangements,

made because of politics may cause higher trade flows, or higher trade flows may result in

politics generating a trade agreement. These issues become even more thorny when trying

to take historical factors into account. The work in this paper is able to successfully bypass

these issues because it exploits the natural experiment of the creation, and eventual disband-

ing of the USSR and Comecon. As will be argued later on in this paper, the historical trading

institutions of the USSR border and the Comecon trading bloc were created independent of

trade promotion reasons.

Eichengreen and Irwin (1995), the seminal paper which identifed the importance of his-

tory in gravity trade models, find strong endogeneity in the formation of the Ottawa Agree-

ment in 1932, the Reichsmark Bloc, and the Ouchy Accords. Similarly, Nitsch and Wolf

(2013) find that the development of the Euro followed a trend of increased economic inte-

gration. In general, Wolf and Ritschl (2011, p. 310) conclude that “to a large extent such

arrangements [currency and trading blocs] are endogenous to the pre-existing pattern of

trade.” To the extent that Wolf and Ritschl are correct about the development of trading

institutions, the issue of historical legacy can be difficult to determine since the formation of

the bloc may have resulted from pre-existing comparative advantages that cannot be picked

up by the gravity equation methodology.

This paper utilizes Russia’s post-Soviet trade flows to address the persistence and en-

dogineity questions. We argue that both the development of the Soviet Union and its main

trading bloc Comecon were developed independently from any particular trade-related ra-

tionale, allowing for a more robust test of institutional persistence in trade flows. We utilize

a gravity model with data on 108 countries from 1998 to 2016, and measure the lagged im-

pact of previous membership in the Soviet Union and Comecon on trade volume (imports

plus exports) to and from Russia. This 19 year data set is used to develop an estimate of

the “decay” of institutional persistence of trade flows. More specifically, we run a Poisson

pseudo-maximum likelihood (PPML) gravity model with multilateral resistance controls in

three-year panels and for each year, to examine the change in the impact of the legacy of

previous trading institutions over time. We find large and persistent legacy effects on trade

between Russia and former Soviet and Comecon countries. Our estimates show greater

persistence of the legacy effect than most previous studies of other trading institutions.

1.2 Background

1.2.1 History & Trade Flows

Institutional legacy has been found important in the determination of trade flows by

several authors. de M´enil and Maurel (1994) analyze the breakup of the Austro-Hungarian

in 1919 and find that the “dissolution of the Empire did not result in the immediate reversal

of the trade patterns of the former union. Even after their dramatic post-war contraction,

trade flows between the successor states remained significantly much larger than would have

been predicted by economic, demographic and geographic factors alone” (ibid, p. 564-

5). Eichengreen and Irwin (1998) investigate the influence of pre-WWII trade (1928 and

1938) on post-war trade flows. They find strong, but diminishing effects for 1949 and 1954,

but no logically consistent effects by 1964. And with specific reference to countries that

had once been part of the British Empire, they find that “Former British colonies traded

disproportionately more with one another in 1949. . . because of the effects of history” (ibid,

p. 55). But again, this effect disappears by 1954 and 1964. In general, Eichengreen and

Irwin persuasively argue that history is fundamental for the determination of trade flows in

any gravity approach.

Anderson and Smith (2007) attempt to validate the results of seminal papers on the

hysteresis of past trading institutions. They use a panel data set and a lagged trade variable

specification from Eichengreen and Irwin (1998) and find strong evidence that historical

patterns do matter in the estimation of trade flows in Canadian trade. Using a fixed effects

approach to estimate the gravity equation, they show that importer and exporter time fixed

effects can capture the effects of history without the use of a lagged dependent variable

approach. They make the case that researchers need to put time and effort into ensuring that

the gravity trade model is correctly specified.1In thinking about the correct specification of

the gravity trade model, the authors argue that accounting for the “border puzzle” is much

more empirically important than accounting for hysteresis. The importance of the border

puzzle will be discussed more in-depth below.

As the empirical literature has evolved, the application of history’s effect in empirical

models has been deployed in a more nuanced way. Instead of simply thinking of history’s

effects as simply the lag of trade, newer investigations, such as Gowa and Hicks (2013),

Brodzicki and Uminski (2018), and Stack et al. (2019) seek to specify the gravity trade

model with variables that appropriately calibrate the model to take important historical

factors that still affect current (and future) trade volumes into account. Gowa and Hicks

(2013) look at trade volume and the effects of trade blocs on trade during the intervening

years between World War I and World War II. They take into consideration that the trade

blocs that were formed Post World War I, had different political aims(all of which shared the

goal of trying to curb intense global economic downturn) depending upon which major power

1Chit¸u et al. (2014) show the importance of a history effect in patterns of bilateral financial investment.

The authors support the idea of a historical legacy effect, in which patterns of country holdings seven decades

ago continue to impact current portfolios.

formulated them when specifying their gravity model. They find that, contrary to recent

literature, none of the great power trading blocs affected trade in positively or negatively.

Brodzicki and Uminski (2018) include variables that account for the historical metro-

polis of Poland to understand foreign trade persistence and development. Using a PPML

gravity model they find that there is evidence of trade flows being a function of the historical

partitions and metropolises of Poland. Similarly, Stack et al. (2019) look at global trade flows

of sugar and account for colonization’s part in developing this market. In demonstrating that

colonial ties dictate current global sugar trade, they show that the geographical direction

those colonial ties originate from can have either positive or negative effects on growth and

trade broadly.

1.2.2 The Gravity Model & Border Puzzle

Another key literature involves the “border puzzle”: after controlling for distance, re-

gions within countries trade much more with each other than do regions across countries

(McCallum, 1995; Anderson and van Wincoop, 2003; Ishise and Matsuo, 2015). In order to

fully understand this particular problem in the gravity trade literature, a discussion of the

gravity trade model is needed.

The empirical framework for the gravity trade model was introduced in Tinbergen (1962).

The gravity trade model uses the metaphor of Newton’s Law of Gravity to explain trade flows.

Specifically, the theoretical argument states that trade flows between any two locations are

positively correlated with their combined GDP (analogous to size in the Newtonian model),

and negatively correlated to distance between the two countries (which mirrors the distance

between two physical particles in Newtons law).2After it’s empirical formulation the gravity

2Although it is undeniable that the empirical formulation of trade flows, distance, and GDP were first

purposed by Tinbergen in 1962, there were similar models such as Savage and Deutsch (1960)’s probabilistic

formulation, that were around at the same time. As to who first conceptualized employing the metaphor

of Newton’s Law to trade flows is a much more debated question. Elmslie (2018) makes a case that Adam

Smith in the Wealth Of Nations used a gravity trade model framework in analyzing the gains from trade as

trade model quickly became “the most empirically successful” model in economics (Anderson

and van Wincoop, 2003, p. 170).

Although successful, the gravity trade model suffered from a lack of theoretical grounding

for it’s formulation beyond the parallels to Newton’s Law, which resulted in biased estimation

results. This problem came to ahead in McCallum (1995), where the author estimated the

trade flows of the United States and Canada via a gravity trade model. McCallum found that

the presence of an international border between the two countries results in 2200% increase in

intranational trade for Canadian Providences. The surprising result from McCallum begged

the question of why it is that the presence of a border results in dramatic diversion between

international trade and intranational trade. This result became known as the border puzzle.

The border puzzle compelled researchers in the literature to ask if borders do really pro-

duce such dramatic effects and/or if the underlying empirical model of gravity was flawed.

Anderson and van Wincoop (2003) solved this puzzle by asserting two claims: (i) the gravity

theory suffered from an omitted variable bias that the authors term as multilateral resistance

terms (MRT) (ii) if one takes into account MRTs, then it is possible to construct a theoreti-

cally consistent and and free of bias model. What made Anderson and van Wincoop’s MRT

model innovative, by comparison to simple remoteness variables proposed by others in the

literature, is that their variable decomposes trade resistances into their component parts.

Trade resistance, as argued by Anderson and van Wincoop, between any two countries

(iand j) can be decomposed into three specific effects: (i) bilateral trade barriers between

region i&j, (ii) i’s resistance to trade with all regions in the world, and (iii) j’s resistance to

trade with all regions in the world. The previously proposed remoteness variables, Anderson

and van Wincoop argue, only captures distance from bilateral trading partners (effect (i)

from above), while the MRT capture all three affects. In applying MRT’s understanding to

their theoretical model they are able probe and empirically test three implications: (I) trade

a function of trade volume.

barriers decrease trade more between large countries than small countries, (II) trade barriers

increase trade within small countries more than large countries, and (III) trade barriers raise

the ratio of trade within country 1 relative to trade between country 1 and 2 where the

smaller country is 1. Mapping this to the United States and Canadian trade results from

McCallum (1995), Anderson and van Wincoop show that researchers would observe a border

effect (though not nearly as large as previously estimated) given testable implications (II)

and (III).

Based on the results of Anderson and van Wincoop, researchers understand that borders

pose a more nuanced effect on international trade flows. In thinking about our research

question with the lenses of the border puzzle and historical effects on trade, we expect to

find that the historical legacy of being a part of the ex-Soviet Union will be stronger than

those of being a former member of Comecon, because within country exchange during the

Soviet era would have been stronger than any international trade all else equal. The a priori

prediction will be especially important for Russia and the other ex-Soviet countries since the

Soviet Union pursued economic planning based on autarky until 1956 (Korbonski, 1970).

This prediction is supported by the border effects literature.

1.2.3 Border Effects

The border effects literature seeks to exploit natural experiments of impact that the

generation and disbandment of national borders have on trade flows. Border effects have

been studied in a wide variety of settings that include, cultural identity (Falck et al., 2012),

war (Che et al., 2015), and the reintegration of economies (Felbermayr and Gr¨oschl, 2014;

Nitsch and Wolf, 2013). Each study finds evidence of long term persistence. Regarding

war, Che et al. (2015) study the impact of the Japanese invasion and 8-year occupation

of China on current trade and other bilateral economic relationships. The authors exploit

differences in the negative impact of the occupation on regions within China and find that a

1% decrease in their measure of intensity (civilian casualties) increases imports from China

to Japan by 14.7% in 2001. Regarding the elimination of borders, Felbermayr and Gr¨oschl

(2014) find that by 1993, the historical border between the Confederate South and the North

(the Mason-Dixon Line) reduced trade by 13% to 14%. However, some of this could be the

result of endogeneity issues. Additionally, in a study that is similar to ours, Nitsch and Wolf

(2013) argue that the reunification of Germany provides a natural experiment regarding the

importance of previous borders. Even given the extraordinary resources devoted to ensure a

rapid reunification, the authors find that it will take between 33 and 40 years for the impact

of the previous border to be statistically eliminated.

Other studies that investigate the effects of history and borders on bilateral trade in

terms of the disintegration of states, unions and trading blocs including the Soviet Union

are Djankov and Freund (2002b,a); Fidrmuc and Fidrmuc (2003); De Sousa and Lamotte

(2007). Djankov and Freund (2002b,a) use a gravity equation to examine trade flows among

and between 9 Russian regions and 14 former USSR republics during the period of 1987-1996.

They find that trade flows between Russian regions and former members of the Soviet Union

were significantly impacted by past linkages. In the beginning of this period, the regions did

not trade more with each other than they did with republics. In contrast, after the collapse

of the Soviet Union, during the period 1994 to 1996, it is shown that Russian regions traded

significantly more with each other than with former Soviet Union republics and that trade

had been reoriented more within Russian regions. The result indicates that Russian regions

tend to trade extensively with former members but over time there is an increasing home

bias in Russia as well as in the new republics.

Djankov and Freund (2002b); ?find a classic border effect considering trade within and

between regions of the Soviet Union. A limitation of their analysis deals with institutional

legacy. The short-term data utilized in the studies (only 5 years after the collapse of the

Soviet Union) does not allow for a longer term analysis of the hysteretic nature of the impact

of past institutional trading arrangements on later trade flows.

Continuing with the investigation of border effects of Eastern Europe, Fidrmuc and

Fidrmuc (2003) examine three disintegrated unions - Yugoslavia, Czechoslovakia, and the

Soviet Union (represented by Russia, Ukraine and Belarus). To capture different trade

relations in gravity equations, they include indicator variables for formal preferential trade

areas, common border or language, and successor states of former federations in Europe,

with data covering the period 1990 to 1998. The results suggest that the trade effects of

former institutions decline rapidly over the 8 years, however, trade relations between former

members remain significant to 1998. These results are inconsistent with previously mentioned

work that finds strong persistence trade patterns after a political disintegration.

De Sousa and Lamotte (2007) attempt to determine why the legacy effects found by

Fidrmuc and Fidrmuc (2003) dissipate quickly relative to other findings. Utilizing controls

suggested by (Anderson and van Wincoop, 2003) and a 1993 to 2001 data set that includes

all countries created by the political disintegration of the Soviet Union, Czechoslovakia and

Yugoslavia, de Sousa and Lamotte find more persistence. The most persistence was found

for the former Yugoslavian states. In 1993 the former Yugoslavian states traded 29 times

more with each other than expected while the still traded 23 times more by 2001. The form

Czechoslovakia states demonstrated the least persistence. The authors found that the results

of Firdmuc and Firdmuc were biased by the limited number of former Soviet, Czechoslovakian

and Yugoslavian states covered in their study, more than their lack of multilateral resistance

controls.

One major difference between our study and those of Fidrmuc and Fidrmuc (2003) and

De Sousa and Lamotte (2007), who also address the hysteresis question using ex-Soviet and

ex-Soviet satellite countries, is in the general empirical strategy. We focus on Russia’s trade

legacy with its ex-Soviet and Comecon member states, while the other studies address legacy

by analyzing the effect of begin a member of a formerly larger state in general. Firdmuc and

Firdmuc investigate if ex-Soviet, Czechoslovakian, and Yugoslavian states trade more with

each other in general than would be predicted by gravity considerations alone. Complicating

this question is nature in which goods flowed throughout the USSR and Comecon states.

In his overview on trade of ex-soviet and Comecon states Pelzman states “the distortions

created by intra-CMEA pricing policy, industrial specialization, and single minded depen-

dence on the Soviet Union as the dominant market, resulted in the formation of industrial

structures inappropriate to these economies.”(Pelzman, 1991, p. 311)3In an earlier article

Pelzman also points out that this distortion is not limited to Comecon-USSR trade flows,

even trade within the USSR between Russia and the eastern republics was distorted for

strategic reasons.(Pelzman, 1980) Trade was used as a tool of planning for the Soviet Union

first and Comecon states second. As such much of the trade moved between the Comecon

countries and Russia. Therefore, the trade links would be best established between Russia

and these other states rather than between the states in general. Utilizing this theoretical

strategy results in substantial differences between our results and those of De Sousa and

Lamotte (2007).

The second contribution of this paper is that most of the literature finds that even though

the trade impact of former institutions dissipate over time, overall trade patterns between for-

mer members of various trading and political institutions remain significant for long periods

after the dissolution of the institutions. This paper expand on this literature by considering

the development and break up of institutions that were developed independently from any

trade-related rationale. The exogenous development of state borders and trade agreement

provides a natural experiment from which to address the legacy question. Moreover, we add

to the literature on the Soviet Union and its satellites by increasing the length of time in

the study and using all ex-Soviet states and all Comecon member countries allowing for a

comparison of borders effects and the effects of a trading union. This allows for an exam-

3Note that CMEA stands for the Council for Mutual Economic Assistance also known as Comecon.

ination of the institutional legacy of trade flows free of the question of the endogeneity of

institutional development and allows us to estimate the half-life of the legacy effect.

Moreover, with the exception of De Sousa and Lamotte (2007), the above studies that

focus on the breakup of the Soviet Union were conducted without controlling for multilateral

resistance as well as utilizing a log-log specification of the gravity trade model. Anderson

and van Wincoop (2003) demonstrates that significant bias is possible in estimates of border

effects from traditional gravity models due to omitted variable bias and Silva and Tenreyro

(2006) illustrate that not using a PPML specification injects bias into the coefficients of

estimators. Our estimates are relatively consistence across regressions with and without

controls. Our institutional decay measures, with controls, demonstrate somewhat more

persistence than the improperly specified gravity model due to controlling for these potential

model biasing errors.

1.3 The Soviet Union and Comecon as Natural Exper-

iments

Can the Soviet Union and Comecon be considered as natural experiments for tests of

the legacy effects of previous trading institutions? We argue in the affirmative since these

institutions were founded for non-trade related reasons, and there is no evidence of strong

trading relations between Russia and the other countries studied prior to the development of

the Soviet Union or Comecon. Thus, from the point of view of the determinants of current

trade flows they can be considered exogenous shocks.

1.3.1 Motivation to the Formation of the Soviet Union

The literature suggests that there were many reasons for the formation of the Soviet

Union that began in 1922 with the unification of the Russian, Transcaucasian, Ukrainian,

and Byelorussian republics, and by 1940 included 15 sub-national Soviet republics existing

until 1991 (see Table 2.2 for a list of countries). In looking at maps of the former Czarist

Russian Empire, one can see that no significant portion of USSR was not part of the Russian

Czar regime. One could think of the unification of these nations with the Russian state is

as if California broke off from the United States then was readmitted.

A fundamental factor driving the unification was ideology (Sherman, 1994). During the

rule of Joseph Stalin, the most widely disseminated book known as “the Short Course”

(C.P.S.U., ed, 1939), claims that the main goal of the formation of the USSR was the con-

solidation of the Soviet power and a victory for the working class. To construct socialism

required “welding the Soviet republics closer together in a single federal state”. (ibid) This

rationale for the development of the Soviet Union is also supported in the work of Sakwa

(1999). Furthermore, this ideology followed the idea that the removal of all the political dif-

ferences and further consolidation of the members into the entirely cohesive Socialist state

and society was meant as a strong counter balance to Western capitalism. The foundation of

this effort was the elimination of nationalistic sentiments that could block the full develop-

ment of worker solidarity. The communist ideology driving national decisions was prevalent

well before Joseph Stalin.

In 1921, the Tenth Congress of the Revolutionary Communist Party (Bolsheviks) was

held with the charge of determining “The Immediate Tasks of the Party in the National

Question”. The Commission was led by Vladimir Lenin and the report from the commission

was developed by Joseph Stalin (Stalin, 1953). The report gives specificity to the ideology

behind the planned development of the Soviet Union. It states, “history tells us that the

only way to abolish national inequality, the only way to establish a regime of fraternal co-

operation between the laboring masses of the oppressed and non-oppressed nations, is to

abolish capitalism and establish the Soviet system.” (ibid, 38) Those in power in Soviet

Russia believed that the state, if managed properly could be a vehicle to united the working

masses. Further, in an effort for the state to manifest unification of the workers, leaders

believed it needed to be a transnational mission.

In an interview with the Russian newspaper Pravda, number 261, on November 18, 1922,

Stalin made this clear when he stated that, “the union of the Soviet republics in a single

union state will undoubtedly create a form of all-round military and economic co-operation

that will greatly facilitate the economic progress of the Soviet republics and convert them

into a citadel against attacks by international capitalism.” (ibid, 141) Thus by welding states

together the Soviet government was strengthening the bonds between workers as a complete

union of the proletariat via abolishing nationalist ties. “The state union of the individual

Soviet republics was considered as the only way of salvation from imperialist bondage and

national oppression” (Grosul, 2007, translation from Russian by S. Kuznetcova).

It goes without saying that the past cultural, economic, military, and historic linkages as

well as external political reasons also played a role in the formation of the Union of Soviet

Socialist Republics, but there is no evidence that direct trade-related rationales existed for

the development of the Soviet Union. Trade statistics from the period leading to the Soviet

Union back up this claim. From 1899 to 1913, no countries that would become part of

the Union with Russia were listed among the top 18 import or export partners of Russia

(Vyacheslav, 2011, 30).4

4The original tables from 1915 are available as “Overview of Russia’s Foreign Trade with European and

Asian Borders in 1914,” Tables 5 and 6 at http://istmat.info/node/213. Translation from Russian by S.

Kuznetcova.

1.3.2 Formation of Comecon

The Council for Mutual Economic Assistance (CMEA) also referred to as Comecon,

was the main trading bloc of the Soviet Union. Comecon was an economic organization

that existed from 1949 to 1991 under the leadership of the Soviet Union and comprised the

countries of the Eastern Block along with a number of socialist states elsewhere in the world

(former members of Comecon are listed in Table 2.2). The official purpose of Comecon was

to coordinate planning, promote country and regional specialization, increase trade among

member states (Korbonski, 1970), and “to improve economic and military cooperation” (New

York Times, 1988). Increased trade flows was among the motivations for the formation of

Comecon, but this was not based on any pre-existing strong or rapidly increasing trade

relations.

From 1926 to 1928, for example, Czechoslovakia (which in our listing of countries includes

the Slovak Republic and the Czech Republic) accounted for only 4% of Soviet trade. No

other Comecon country had large enough trade with the Soviet Union to be listed in the

Soviet’s own publication of economic statistics (Soviet Union Information Bureau, 1929).

Moreover, between 1928 and 1938, Holzman (1976, 1985) reports that overall imports from

Eastern Europe to the Soviet Union fell by about half, while exports from the Soviet Union

to Eastern Europe fell by about one-quarter. As we move back in time to the period from

1899 to 1913, no Comecon country was a major exporter or importer of Russia. In 1913,

for example, Austria-Hungary represented 4.3% of exports and 2.6% of imports with no

significant trend from 1899. This was the largest representative of what would partly (as

Hungary without Austria) become part of Comecon (Vyacheslav, 2011 and details from

footnote 2).

The most direct reason for Comecon’s development was ideological. Comecon was founded

in response to the Marshall Plan “to reinforce the bonds between the Soviet Union and the

”people’s democracies” of Eastern Europe” Brine (1992). So, it can be considered as the so-

cialist alternative and reply to the formation of the Organization for European Cooperation

in Western Europe. After World War II Comecon was seen as an effective instrument to

spread communism to the countries of the Eastern European block with the USSR being the

dominant member. (ibid.) Therefore, Comecon was formed mainly for reasons unconnected

to previously strong or growing ties to trade. Even by 1956, six years after its formation in

1949, there was insignificant intra-Comecon trade (Korbonski, 1970, 957). The literature on

the development and operation of Comecon demonstrates that it was a poorly designed and

managed trading institution that resulted in little true trade creation and was mostly trade

diverting (Zickel, ed, 1989; Pelzman, 1977; Holzman, 1985; Biessen, 1991). Additionally, and

most importantly, Comecon was meant mainly as a control devise for the Soviet Union over

Comecon members (Pelzman, 1977).

While Comecon had no initial economic advantages motivating its existence, it did suc-

ceed in dramatically increasing trade flows between members, creating new trade patterns

(Hewett, 1976; Holzman, 1985; Pelzman, 1977). It was, therefore, successful in its efforts

to promote trade and specialization through central planning (Biessen, 1991). For exam-

ple, Hewett (1976), using a gravity approach finds that within Comecon trade was 20-times

larger than it was predicted to be without the trading bloc. Additionally, (Zickel, ed, 1989,

601) reports that “in 1960 the Soviet Union sent 56% of its exports to and received 58% of

its imports from Comecon members. From that time, the volume of this trade has steadily

increased. . . ”

Because both the Soviet Union and the Comecon were constructed for non-trade related

reasons, they are good candidates to be considered as natural experiments for the deter-

mination of institutional legacy effects in trade flows. Since no evidence exists suggesting

that pre-institutional trade flows were greater than would be expected by a gravity analysis,

any lingering effects of these institutions on current trade flows can be attributed to legacy

effects with a high degree of confidence.

1.4 Data

The data set developed in this paper is a panel of Post-USSR Russian imports from and

exports to 108 countries for the 19-year period from 1998 to 2016. The choice of the countries

for the empirical analysis (listed in the appendix Table 2.2) is based on the data availability

for all variables and for all years.

The variables and data sources used to build the variables are listed in Table 2.1. The

dependent variable is bilateral imports plus exports to and from Russia. To measure the

trade flows between Russia and its trading partners, import and export figures were taken

from the World Bank’s World Tables at market prices in U.S. dollars. The trade data is then

converted into constant chained 2005 dollars. Country-pairs are used to reflect the bilateral

relationship between Russia and its trading partners. To estimate the volume of trade flows

between Russia and its trading partners we use three “groups” of independent variables.

First, to predict bilateral trade flows we use the traditional gravity variables of economic

size of the countries and distance the between them. To investigate the influence of an

economy’s size on trade, GDP measurements at market prices in U.S. dollars are also obtained

from the World Bank database. Nominal GDP data is then converted into constant chained

2005 dollars. The distance variable is the distance between Moscow and the capital city

of Russia’s trading partner measured in kilometers, and was generated using online maps.

All standard gravity variables are estimated in terms of natural logarithms. Based on the

theoretical foundation of the gravity equation, it is expected that greater distance between

trading partners reduces the volume of trade and that countries with higher levels of income

tend to trade more with Russia.

Second, to follow the historical patterns of trade between Russia and its trading partners

we include two indicator variables. The first variable indicates if the country was a member

of the Soviet Union (1 for former members and 0 otherwise). The second variable defines if

the country was a non-Soviet part of a trading block Comecon (1 for former members and 0

otherwise).5This variable excludes countries of the Soviet Union to avoid strong collinearity

with the USSR indicator variable. Given the discussion in the previous section we expect

these variables to be exogenous. Both the Soviet Union and Comecon were constructed for

ideological rather trade-related reasons. The inclusion of these two indicator variables in

the gravity model, enables us to compare two effects of disintegration: the border effect

(the absence of borders in the USSR) and the effect of a trading agreement (Comecon

membership). The long-term nature of the data also allows us to observe the change in

these effects over time.

Another variable often used in this type of study is a measure of linguistic distance (Fidr-

muc and Fidrmuc, 2003; Hutchinson, 2005). In our case, however, there is a high correlation

between linguistic similarity and the makeup of the ex-Soviet states. Therefore, for complete-

ness, we chose another widely utilized variable in these studies, that of economic/political

freedom (e.g. Depken and Sonora, 2005; Sonora, 2014; Wall, 1999). Our measure is the In-

dex of Economic Freedom provided by the Heritage Foundation. This index combines ten

measures of economic freedom into a single composite index. Measures of freedom represent

various quantitative and qualitative factors in business, labor, monetary, trade, investment,

and financial freedoms, as well as legislative factors such as freedom from corruption, fiscal

freedom, property rights enforcement, and government spending. The index rates countries

on a scale of one to five, where numerical scores correspond respectively to the level of eco-

nomic freedom of a country represented as repressed; mostly unfree; moderately free; mostly

free; and free. Lower index numbers represent lower levels of economic freedom of a country.

This index allows for the examination of not only trade freedom of countries but to determine

5Although Comecon is treated as a uniform organization, in truth there was some heterogeneity even in

active members. For example, Albania was an official member, but it stopped active participation in the

Comecon starting in 1961, while Cuba and Vietnam only joined in 1970. This paper has altered the indicator

variable, to reflect this non-uniformity and found it does not change the results in significant ways.

the overall degree of countries’ openness for trade flows. While potentially of interest in its

own right, for our present purposes, the variable is used as an additional control and is not

highlighted. The results are similar when the variable is excluded from the analysis.

1.5 Methodology and Empirical Analysis

In order to examine the effects of hysteresis on bilateral trade flows between Russia and

its trading partners, a traditional gravity model is employed. As stated earlier, the basic

gravity trade model relates trade flows to GDP and distance between trading partners. We

are arguing that a well-specified gravity trade model needs to take historical institutions into

account, and must be altered to include variables that account for hysteresis, border effects,

and a PPML estimation method. In order to achieve this goal, this paper employs three

functional forms of estimation. Before discussing the three different model specifications it

is worth while to discuss why a PPML estimation was chosen.

The reason for the employment of the PPML model specification is that it solves two

innate problems of the gravity trade model. First, the PPML solves the problem of zero

values in trade flows, as the natural log of zero is undefined. The literature has traditionally

solved this by either dropping zero observations, or by adding some extremely small value to

the non-existent trade flows to allow for estimation. By using a PPML specification we can

include zero values without injecting bias into the results. The second reason for employing a

PPML is that it solves a much more pressing issue, heteroskedasticity in trade data. As Silva

and Tenreyro (2006) makes clear, when the data is heteroskedastic a log-linearization of the

model will result in biased and inconsistent results. This is true irrespective of the application

of MLR terms. The authors demonstrate that heteroskedasticity results as a function of

Jensen’s inequality and “the expected value of the logarithm of random variable depends

on higher-order moments of its distribution. Therefore, if the errors are heteroskedastic,

the transformed errors will be generally correlated with the covariates.” (ibid, p. 653) Given

their findings, Silva and Tenreyro advocate for the use of PPML model specification to solve

both the zero trade-flows and heteroskedasticity in trade data.

Following Silva and Tenreyro’s advice,6we employ a PPML specification with a panel of

total trade volume over all the years of the data set, taking the functional form of:

TiR,t =expβ1lnGDPi,t +β2lnDistiR +β3USSR ·t+β4Comecon ·t+β5Fi,t+

β6CNT G +β7lnMLRexp,t +β8lnMLRimp,t∗εiR,t (1.1)

The dependent variable TiR,t is the total volume of trade that combines exports and

imports between country iand Russia at each year interval. The standard gravity predictor

variables are each ith country’s real GDP (GDPi) and the distance between Moscow and

the capital of Russia’s trading partner (DistiR). There are two hysteresis variables which are

interacted with a year indicator variable, USSR and Comecon. The reason for this interacted

variable is that the indicator variables of USSR and Comecon are constant over time, but

performing cross-sectional analysis reveals the effects of these variables are changing over

time. Therefore, the interacted variables will allow for the change of the effects over time.

The Fi,t variable is the Index of Economic Freedom of trading partners of Russia. The

variable CNTG is also an indicator variable to control for countries that have a contiguous

border with Russia. Lastly MLR terms are reduced form multilateral resistance terms and

εiis the error term. The purpose of the first regression is to test out the validity of the

impact of the hysteresis and robustness of using a gravity specification in this manner. The

results are reported in Table 2.6 and will be discussed below.

Before moving to the second functional form, it is worth discussing the empirical con-

6In addition to the case made in their paper, Xiong and Chen (2014) show that PPML and not other

proposed models such as a Tobit or Heckman model result in the best possible estimations of the gravity

model.

struction of the MLR variables throughout all the of the estimations. As was discussed in

the boarder puzzle section, multilateral resistance terms are a theoretical construct, and

as such must be generated. The original MRT’s described in Anderson and van Wincoop

(2003) were a custom non-linear least squares program that generated values of the MRT

after repeated simulations until convergence. Luckily the broader literature has developed

two easily deployable empirical solutions to construct MRT for researchers (Yotov et al.,

2016).

The first empirical solution is referred to as a “remoteness index,” and is what is em-

ployed under the both specifications. The remoteness index is a reduced form of the custom

built MRT’s that Anderson and van Winkoop introduced. These remoteness indexes are

output and expenditure weighted averages of bilateral distance. They are constructed via

the following two equations:

REM EXPi,t =ΣjDistij /Ej,t

Yt(1.2)

REM IMPj,t =ΣiDistij /Yi,t

Yt(1.3)

Where Ej,t is the value of importer expenditure, obtained by summing up the value of all

trade exported by country jin year t. Similarly, Yi,t is the value of exporter output by

country iin year t. In equation (2.3), the variable Ytis sum of all Ej,t in a year then utilizing

the max value of that year. Conversely, in equation (2.4), Ytrepresents sum of all Yj,t in a

year then utilizing the max value of that year.

The second empirical way to obtain MLR terms is via exporter(and importer) paired-

time fixed effects variables, where an indicator variable is created for when country itrades

with country jin time t. These fixed effects capture the “special” underlying factors that

resulted in these two countries trading in this particular time. The problem with deploying

this solution with our data is that all trade is to and from Russia, thus creating indicator

variables that are multicollinear and will absorb all variation in the data.

Given our interest in the time-trend effect of the USSR and Comecon variables, we

employed the PPML specification with a year-by-year regression, which takes the functional

form of:

TiR,t =expβ1lnGDPi,t +β2lnDistiR +β3USSR +β4Comecon +β5Fi,t+

β6CNT G +β7lnMLRexp,t +β8lnMLRimp,t∗εiR,t (1.4)

This model is similar to specification (2.6), with one notable difference, our hysteresis dum-

mies are no longer interacted with time. Under specification (2.7), a regression is run for

each year, as such, interacting the indicator variables with a time variable would not have

an econometric impact. The results from this specification can be found on Table 2.7 in the

appendix.

Finally, we altered specification (2.7) and experimented with regressions at multiple year

intervals (i.e. 5 year, 4 year, 3 year, etc.). Multiple year interval regressions were also

performed to overcome a common issue in the gravity trade literature, specifically how to

adjust to trade policy changes over the time of the data. If there is dramatic changes to a

trade policy from year 1 to year 2, then it will likely generate influential outlier data points

depending the total span of the data. A simple solution to the policy change issue is to use

panel data over multiple year intervals (Yotov et al., 2016). Since the regressions are over

multiple years the hysteresis dummies are again interacted with a time indicator variable.

Reproduced below, in Table 2.8, are the results from a 3-year interval specification. The

results are consistent whether looking at 2, 3, 4, or 5 year intervals. Three years was chosen

because it showed the smoothest paths of our variables of interest.

1.6 Results

Beginning first with the panel of all the years (Table 2.6), we note that the coefficients

represent the average impact of each variable on Russia’s trade volume. Looking at the

results we see that GDP, distance, and contiguous borders all have the expected signs and

are statistically significant at the highest levels, while economic freedom is negative and

insignificant. Excluding the statistical significance, the negative coefficient is somewhat

unexpected given that economic freedoms increase as the variable increases, indicating more

economic openness which should result in more increased trade. The indicator variable for

USSR and Comecon in the aggregated data are time interacted indicators, meaning that

the coefficients need to be interpreted with care as they represent an average effect of the

previous institution on trade for 1/19 of the overall panel, given that there are 19 years in

the data. The difficulties associated with interpreting an indicator variable that is mostly

decreasing over time means that we will comment only on the variables in the year-by-year

results. The important take away from this model with regards to our historical institution

variables is that they are positive and statistically significant, meaning that it is clear these

historical institutions are impacting current day trade flows to and from Russia.

Table 2.7 breaks the data by year and reports results from the yearly regressions. Note

that the coefficients on GDP are stable and consistent with expectations, while distance

is negative it oscillates being statistically significant. The variable for economic freedom

displays a lot of “noise” due to it oscillating signs and not being statistically significant. In

contrast to economic freedom, contiguous boarders are constantly positive with only a few

years retaining any statistical significance.

The coefficient on our main variables of interest (USSR and Comecon) are also stable and

declining over time as expected from the general literature on the legacy effects of trading

institutions. However, the coefficients on distance and economic freedom vary widely from

year to year indicating a substantial amount of noise in the yearly data. This is common in

the trade gravity literature (Yotov et al., 2016). To diminish the impact of these year-to-year

variations, a standard practice is to create multi-year panels (ibid). We did multi-year panels

with 2, 3, 4, and 5 year periods. The results are consistent across all panels. We chose to

report the 3-year panel estimates given that it allows for the use of all years except for 2016

and the coefficients on distance are more stable than with the 2-year panel. In these panels,

we utilize a time control that allows for the stacking of the data when utilizing an indicator

variable.

Turning to the 3-year panel in Table 2.8, it shows that Russia trades more with larger

countries and less with countries that are geographically further away, as expected. Further

contiguous borders are positive and statistically significant, at varying levels, throughout

all of the 3-year intervals. Interestingly, the economic freedom variable is negative and not

statistically different from zero in all the panels. While no evidence of collinearity problems

with the multilateral resistance variables exists, the result should be interpreted with caution

since trade freedom makes up part of each county’s score on the index.

Our main variables of interest are the indicator variables for the USSR and Comecon.

The coefficients demonstrate strong and significant positive effects on Russia’s trade with

former members of the USSR and Comecon across all specifications and models. To interpret

the importance of these variables, we must look to the yearly regressions under the PPML

specification. Referring to Table 2.7, the coefficients on USSR and Comecon respectively for

1998 are 2.263 and 0.843. These coefficients indicate that Russia traded approximately 8.611

times (eβ−1) more than expectation based on country GDPs and distance from Russia with

former Soviet countries and approximately 1.32 times more with ex-Comecon countries than

expected.7

7Note that this Comecon estimate is in line with the result found by Hewett (1976) in 1970. He found that

membership in Comecon increased trade compared with gravity expectations by 200% in gravity estimates

that did not utilize proper controls thus biasing his results.

Looking across the Table 2.7 from 1998 to 2016, a consistent pattern occurs on the relative

magnitude of the legacy impacts of membership in the Soviet Union versus Comecon. The

effect is about twice as large for ex-Soviet countries. To help aid in examining the pattern,

a plot of the coefficients and confidence intervals of both of the variables can be found in

Figure 1.1. The coefficients have been put into percentage impact upon total trade form.

Looking at this figure it is clear that there is a monotonically decreasing to zero patter, in

both variables though, it is much stronger in the USSR variable. Further, we can see that

Comecon is much more impacted by the statistical “noise” that comes from doing a year-

by-year specification. Irrespective, it is clear that impact of the institutions are decreasing

as we move away from the initial dissolution of the USSR and Comecon.

Turning to the 3-year panel results (Table 2.8), we see that our institution indicators are

a lot more stable. Specifically, the coefficients on the USSR are still statistical significant at

the highest levels and Comecon stays statistically significant for longer. Further in looking

at the plot for the coefficients (Figure 1.2) the pattern of monotonically decreasing is much

more evident. It is clear from the 3-year panel that the noise present from the previous

specifications is smoothed out, showing that the impact of previous institutions decays and

trends towards zero as we move forward in time.

The pattern of results on our institutional variables is consistent with existing literature

and the estimation of the border effect by Anderson and van Wincoop (2003) and others.

Borders significantly impede trade. While intra-Soviet trade statistics are not available, it

is reasonable to assume that intra-Soviet regional trade was stronger (controlling for gravity

variables) than it would have been for trade between the Soviet Union and Comecon member

states(Pelzman, 1980). Given that stronger trade ties would have been created for regions

within the Soviet Union, the legacy impact is expected to be stronger.

As figure 1.2 illustrates, institutional legacy effects ex-USSR and ex-Comecon states

differently. Although both USSR and Comecon are positive and decaying monotonically, the

institutional legacy of being a former member of the USSR is decaying much more rapidly.

Utilizing the coefficients in 1998 and 2016 and the 19 year time period, we can estimate a

“back of the envelope” figure which describes the “decay” of these historical institutions. We

find that the institutional impact of being a member of the ex-Soviet Union to decay by half

in 18.07 years while the impact of the institution of being a Comecon member state decays

by half in 11.31 years. This measure takes into account only the magnitude of the coefficient

and the rate of decline. Analyzing the magnitude of the institutional impact over time we

find that over the 19 year period the impact of being a former member-state of the Soviet

Union resulted in Russia trading with its other member-states, 8.611 times more in 1998

and 1.98 times more in 2016. For ex-Comecon, Russia traded 1.32 times more than expected

by trade gravity in 1998 and 0.3 times more by 2016. Of course the exact numbers change

depending on the end year selected, but the main pattern is consistent for the institutional

decay calculations across years.8

The main result from the empirical work is that trade patterns developed during previous

institutions impact the volume of trade in a manner that demonstrates strong persistence.

Twenty-six years after the collapse of the Soviet Union and Comecon trade flows between

former members remain significantly greater than gravity alone predicts because of the his-

torical patterns established in the past. Our results show substantially longer persistence

than is found by De Sousa and Lamotte (2007). By 1998, the beginning year of our study,

they find that ex-Soviet states trade about 7.5 times more with each other than expected,

while we show that Russia traded about 8.611 times more than expected based on trade

gravity. This difference is most likely explained by the difference of our empirical strategy.

Sousa and Lamotte ask, how much more are ex-Soviet countries trading with each other.

But during the Soviet era, a large share of trade moved thorough Russia. Russia was the

8For example, utilizing the same initial year of 1998, the USSR (Comecon) institutional decay estimates

are 16.05 (11.16), 16.56 (9.41) for the years 1998 to 2013 and 1998 to 2014 respectively.

trade center of gravity around which the bulk of trade flowed. This is true for the satellite

states in Comecon as well. Thus, the expectation is that the legacy effects are larger for

trade between Russian and ex-Soviet and ex-Comecon states than between those states in

general.

1.7 Conclusion

This paper examines the effect of historical legacy of previous institutions on the volume of

Russia’s bilateral trade using the gravity model approach. The effects of historical legacy are

represented by the former membership of the Soviet Union and trading agreement Comecon.

We reach the following conclusions.

First, the results indicate that we cannot fully understand bilateral trade flows without

considering the impact of past institutional trading arrangements. The effects of historical

legacy of the Soviet Union and Comecon are still impacting Russia’s bilateral trade flows

even 26 years after their respective collapse. In other words this finding suggests that former

networks persist, overlap new borders and encourage trade between successor states.

Second, consistent with expectations and all other studies of the legacy effects of trading

institutions, we find declining trends in trade intensity between Russia and the former mem-

bers of the USSR and Comecon. These findings are consistent with the results of Fidrmuc

and Fidrmuc (2003); De Sousa and Lamotte (2007). Institutional disintegration does not

lead to an immediate trade disintegration.

We also find that the effect of the membership of the Soviet Union is stronger than the

effect of Comecon, most likely because of the impacts of borders on trade. The absence of

borders in the past between former members of the Soviet Union established strong trade

patterns that hold even today. The strength of this conclusion is demonstrated by both the

steeper slope of a plot of the coefficients for USSR and Comecon, as well as a back of the

envelope “decay” calculation. We find that the legacy decay for the impact of the previous

institution on trade is 18 years for the USSR and 11 years for Comecon.

This paper provides the first estimates of the persistence of the legacy effect on trade flows

generated by former trading institutions. We demonstrate this both via our figures of the

coefficients and our institutional decay measure. Our institutional decay measure estimates

indicate that former trading arrangements cast a long shadow on current trade relations.

Much as (Krugman, 1987, 47) described, trade patterns are like rivers; once established,

even if by purely institutional arrangements, the path becomes self-reinforcing as the flows

dig their path deeper and deeper.

Bibliography

Acemoglu, Daron, Simon Johnson, and James A. Robinson, “Institutions as a

Fundamental Cause of Long-Run Growth,” in Philippe Aghion and Steven N. Durlauf,

eds., Handbook of Economic Growth, Vol. 1A, Handbooks in Economics, 2005, pp. 385–

472.

Anderson, James E. and Eric van Wincoop, “Gravity with Gravitas: A Solution to

the Border Puzzle,” American Economic Review, 2003, 93 (1), 170–192.

Anderson, Michael A. and Stephen L. S. Smith, “How Does History Matter? Hys-

teresis in Canadian Trade,” North American Journal of Economics and Finance, 2007, 18

(3), 279 – 291.

Baldwin, Richard and Dany Jaimovich, “Are Free Trade Agreements Contagious?,”

Journal of International Economics, 2012, 88 (1), 1 – 16.

Barro, Robert and Silvana Tenreyro, “Economic Effects of Currency Unions,” Economic

Inquiry, 2007, 45 (1), 1 – 23.

Bennett, Asa, “How Will the Brexit Transition Period Work?,” The Daily Telegraph,

January 2020.

Biessen, Guido, “Is the Impact of Central Planning on the Level of Foreign Trade Really

Negative?,” Journal of Comparative Economics, 1991, 15 (1), 22–44.

Brine, Jenny J.,COMECON: The Rise and Fall of an International Socialist Organization,

Vol. 3, Transaction Publishers, 1992.

Brodzicki, Tomasz and Stanislaw Uminski, “A Gravity Panel Data Analysis of Foreign

Trade by Regions: The Role of Metropolises and History,” Regional Studies, 2018, 52 (2),

261–273.

Campos, Nauro F., Menelaos G. Karanasos, and Bin Tan, “From Riches to Rags, and

Back? Institutional Change, Financial Development and Economic Growth in Argentina

since 1890,” Journal of Development Studies, 2016, 52 (2), 206 – 223.

Che, Yi, Julan Du, Yi Lu, and Zhigang Tao, “Once an Enemy, Forever an Enemy?

The Long-Run Impact of the Japanese Invasion of China From 1937 to 1945 on Trade and

Investment,” Journal of International Economics, 2015, 96 (1), 182–198.

Chit¸u, Livia, Barry Eichengreen, and Arnaud Mehl, “History, Gravity and Interna-

tional Finance,” Journal of International Money and Finance, 2014, 46, 104–129.

C.P.S.U., ed.,History of the All-Union Communist Party (Bolsheviks) Short Course, In-

ternational Publishers, 1939.

de M´enil, Georges and Mathilde Maurel, “Breaking Up a Customs Union: The Case

of the Austro-Hungarian Empire in 1919,” Review of World Economics, 1994, 130 (3),

553–575.

De Sousa, Jos´e and Olivier Lamotte, “Does Political Disintegration Lead to Trade

Disintegration? Evidence from Transition Countries,” Economics of Transition, 2007, 15

(4), 825–843.

Depken, Craig A. and Robert J. Sonora, “Asymmetric Effects of Economic Freedom

on International Trade Flows,” International Journal of Business and Economics, 2005, 4

(2), 141.

Djankov, Simeon and Caroline Freund, “New Borders: Evidence from the Former

Soviet Union,” Weltwirtschaftliches Archiv, 2002, 138 (3), 493–508.

and , “Trade Flows in the Former Soviet Union, 1987 to 1996,” Journal of Comparative

Economics, 2002, 30 (1), 76–90.

Eichengreen, Barry and Douglas A. Irwin, “Trade Blocs, Currency Blocs and the

Reorientation of World Trade in the 1930s,” Journal of International Economics, 1995, 38

(1-2), 1–24.

and , “The Role of History in Bilateral Trade Flows.,” in “The Regionalization of the

World Economy,” NBER Project Report series., 1998, pp. 33 – 57.

Elmslie, Bruce, “Retrospectives: Adam Smith’s Discovery of Trade Gravity,” Journal of

Economic Perspectives, 2018, 32 (2), 209 – 222.

Estevadeordal, Antoni, Brian Frantz, and Alan M. Taylor, “The Rise and Fall of

World Trade, 1870–1939,” The Quarterly Journal of Economics, 2003, 118 (2), 359–407.

Falck, Oliver, Stephan Heblich, Alfred Lameli, and Jens S¨udekum, “Dialects, Cul-

tural Identity, and Economic Exchange,” Journal of Urban Economics, 2012, 72 (2-3),

225–239.

Felbermayr, Gabriel and Jasmin Gr¨oschl, “Within U.S. Trade and the Long Shadow

of the American Secession.,” Economic Inquiry, 2014, 52 (1), 382–404.

Fidrmuc, Jan and Jarko Fidrmuc, “Disintegration and Trade,” Review of International

Economics, 2003, 11 (5), 811–829.

Gowa, Joanne and Raymond Hicks, “Politics, Institutions, and Trade: Lessons of the

Interwar Era,” International Organization, 2013, 67 (3), 439–467.

Grosul, Vladislav IAkimovich,Obrazovanie SSSR: 1917-1924 gg., Moskva: ITRK, 2007.

(in Russian).

Hewett, Edward A., “A Gravity Model of CMEA Trade.,” in J.C. Brada, ed., Quantitative

and Analytical Studies in East-West Economic Relations., Bloomington: International

Development Research Center, 1976, pp. 1–15.

Holzman, Franklyn D.,International Trade Under Communism: Politics and Economics,

Basic Books, 1976.

, “Comecon: A ‘Trade-Destroying’ Customs Union?,” Journal of Comparative Economics,

1985, 9(4), 410–423.

Hutchinson, William K., “‘Linguistic Distance’ as a Determinant of Bilateral Trade,”

Southern Economic Journal, 2005, pp. 1–15.

Ishise, Hirokazu and Miwa Matsuo, “US-Canada Border Effect Between 1993 and 2007:

Smaller, Less Asymmetrical, and Declining,” Review of World Economics, 2015, 151 (2),

291–308.

Johnson, Keith, “How Trump May Finally Kill the WTO,” Foreign Policy, December

2019.

Karnups, Viesturs Pauls, “Latvia in the International Economy: Historical Dimensions,”

Transformations in Business & Economics, 2008, 7(1).

Keller, Wolfgang and Carol H. Shiue, “Endogenous Formation of Free Trade Agree-

ments: Evidence from the Zollverein’s Impact on Market Integration,” The Journal of

Economic History, 2014, 74 (4), 1168–1204.

Korbonski, Andrzej, “Theory and Practice of Regional Integration: The Case of Come-

con,” International Organization, 1970, 24 (4), 942–977.

Krugman, Paul, “The Narrow Moving Band, the Dutch Disease, and the Competitive

Consequences of Mrs. Thatcher,” Journal of Development Economics, 1987, 27 (1-2), 41

– 55.

McCallum, John, “National Borders Matter: Canada-US Regional Trade Patterns,” The

American Economic Review, 1995, 85 (3), 615–623.

Mitchener, Kris James and Marc Weidenmier, “Trade and Empire,” The Economic

Journal, 2008, 118 (533), 1805–1834.

New York Times, “Warsaw Pact and Comecon: Formal Ties,” New York Times, Jan 03

1988.

Nitsch, Volker and Nikolaus Wolf, “Tear Down This Wall: On the Persistence of Borders

in Trade,” Canadian Journal of Economics, 2013, 46 (1), 154–179.

North, Douglass C., “Institutions and Economic Development,” Taiwan Economic Re-

view, 1995, 23 (1), 1 – 24.

Pelzman, Joseph, “Trade Creation and Trade Diversion in the Council of Mutual Economic

Assistance 1954-70,” The American Economic Review, 1977, 67 (4), 713–722.

, “Economic Determinants of Soviet Foreign Trade with Eastern Europe,” European Eco-

nomic Review, 1980, 14 (1), 45 – 59.

, “The New Soviet-East European Economic Relationship: Post CMEA,” Economic Sys-

tems, 1991, 15 (2), 309 – 321.

Sakwa, Richard,The Rise and Fall of the Soviet Union, 1917-1991, Routledge, 1999.

Savage, I. Richard and Karl W. Deutsch, “A Statistical Model of the Gross Analysis

of Transaction Flows,” Econometrica, 1960, 28, 551 – 572.

Sherman, Howard J., “Rise and Fall of the Soviet Union,” International Journal of Po-

litical Economy, 1994, 24 (1), 5–18.

Silva, J.M.C. Santos and Silvana Tenreyro, “The Log of Gravity,” The Review of

Economics and Statistics, 2006, 88 (4), 641–658.

Sonora, Robert J., “All Economic Freedom is Not Created Equal: Evidence from a Gravity

Model,” Contemporary Economic Policy, 2014, 32 (1), 30–41.

Soviet Union Information Bureau,The Soviet Union: Facts, Descriptions, Statistics,

Soviet Union Information Bureau, 1929.

Stack, Marie M., Rob Ackrill, and Martin Bliss, “Sugar Trade and the Role of

Historical Colonial Linkages,” European Review of Agricultural Economics, 2019, 46 (1),

79–108.

Stalin, Joseph V.,Works: v.5, 1921-1923, Lawrence & Wishart, Ltd, 1953.

Tinbergen, Jan,Shaping the World Economy: Suggestions for an International Economic

Policy, New York: Twentieth Century Fund, 1962.

Vyacheslav, M.,Krushenie Rossii, 1917, Moscow: ACT: Astrel, 2011. in Russian.

Wall, Howard J., “Using the Gravity Model to Estimate the Costs of Protection,” Review-

Federal Reserve Bank of Saint Louis, 1999, 81, 33–40.

Wolf, Nikolaus and Albrecht O. Ritschl, “Endogeneity of Currency Areas and Trade

Blocs: Evidence from a Natural Experiment,” Kyklos, 2011, 64 (2), 291–312.

Xiong, Bo and Sixia Chen, “Estimating Gravity Equation Models in the Presence of

Sample Selection and Heteroscedasticity,” Applied Economics, 2014, 46 (24), 2993–3003.

Yotov, Yoto V., Roberta Piermartini, Jos´e-Antonio Monteiro, and Mario Larch,

An Advanced Guide to Trade Policy Analysis: The Structural Gravity Model, World Trade

Organization, 2016.

Zickel, Raymond E., ed.,Soviet Union: a country study, 2nd ed., Federal Research

Division, 1989.

Zukowski, Ryszard, “Historical Path Dependence, Institutional Persistence, and Transi-

tion to Market Economy: The Case of Poland,” International Journal of Social Economics,

2004, 31 (9-10), 955 – 973.

1.9 Tables & Figures

Table 1.1: List of Countries

Albania** Denmark Kazakhstan* Romania**

Algeria Dominican Republic Kenya Rwanda

Argentina Ecuador Korea Rep. Saudi Arabia

Armenia* Egypt, Arab Rep. Kyrgyz Republic* Senegal

Austria El Salvador Lao PDR Singapore

Azerbaijan* Estonia* Latvia* Slovak Republic**

Bangladesh Ethiopia Lebanon Slovenia

Belarus* Finland Lithuania* Spain

Belgium-Luxembourg France Malaysia Sri Lanka

Belize Georgia* Malta Sweden

Bolivia Germany Mexico Switzerland

Bosnia and Herzegovina Ghana Moldova* Tajikistan*

Brazil Greece Mongolia** Tanzania

Bulgaria** Guatemala Morocco Thailand

Cambodia Guinea Nepal Tunisia

Cameroon Honduras Netherlands Turkey

Canada Hong Kong, China New Zealand Turkmenistan*

Chile Hungary** Nicaragua Uganda

China Iceland Nigeria Ukraine*

Colombia India Norway United Arab Emirates

Congo, Rep. Indonesia Oman United Kingdom

Costa Rica Iran, Islamic Rep. Pakistan United States

Cote d’Ivoire Ireland Panama Uruguay

Croatia Israel Peru Uzbekistan*

Cuba** Italy Philippines Venezuela

Cyprus Japan Poland** Vietnam**

Czech Republic** Jordan Portugal Zambia

Note: * indicates former members of the Soviet Union, ** indicates former members of Comecon

Table 1.2: Measurement and Data Sources

Description Measurement The Source

TR,i The bilateral trade flows: Exports (X) Chained 2005 The World

& Imports (M) between Russia & country ithousand dollars Bank

GDPiCountry i’s real GDP Chained 2005 The World

dollars Bank

DistR,i The distance between Moscow & the Kilometers Online maps

capital city of the trading partner

FiThe Index of Economic Freedom Scale of The Heritage

1–Repressed 1 to 5 Foundation

2–Mostly unfree

3-Moderately free

4-Mostly free

5-Free

USSR Indicator variable: former members of the 1–if former

Soviet Union. These members are: Armenia, member of the

Azerbaijan, Belarus, Estonia, Georgia, Soviet Union

Kazakhstan, Kyrgyz Republic, Latvia, 0-otherwise

Lithuania, Moldova, Tajikistan,

Turkmenistan, Ukraine, Uzbekistan

Comecon Indicator variable:former members of 1-if former

Comecon. They are: Albania, Bulgaria, member of

Cuba, Czech Republic, Hungary, Mongolia, Comecon

Poland, Romania, Slovak Republic, Vietna* 0-otherwise

*Soviet Union countries are excluded from because of the existence of USSR indicator.

Table 1.3: PPML Panel Results Covering 1998 to 2016

All Years

Log of GDP 0.804∗∗∗

(0.0164)

USSR Membership 0.0794∗∗∗

(0.00567)

Comecon Membership 0.0247∗∗∗

(0.00556)

Log of Distance -0.861∗∗∗

(0.0427)

Economic Freedom Index -0.00106

(0.0291)

Contiguous Borders 0.595∗∗∗

(0.0762)

Exporter Remoteness Index -0.0279

(0.0200)

Importer Remoteness Index -0.0546∗∗

(0.0172)

Constant 1.199

(0.685)

Observations 4104

R20.473

Standard errors in parentheses

∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

Table 1.4: PPML Results by Year

98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16

Log of GDP 0.833∗∗∗ 0.811∗∗∗ 0.785∗∗∗ 0.806∗∗∗ 0.769∗∗∗ 0.751∗∗∗ 0.764∗∗∗ 0.773∗∗∗ 0.817∗∗∗ 0.816∗∗∗ 0.887∗∗∗ 0.874∗∗∗ 0.838∗∗∗ 0.848∗∗∗ 0.834∗∗∗ 0.806∗∗∗ 0.810∗∗∗ 0.810∗∗∗ 0.822∗∗∗

(0.0394) (0.0364) (0.0551) (0.0505) (0.0477) (0.0715) (0.0707) (0.0632) (0.0628) (0.0666) (0.0615) (0.0543) (0.0567) (0.0543) (0.0581) (0.0589) (0.0585) (0.0555) (0.0493)

USSR Indicator Varible 2.263∗∗∗ 2.035∗∗∗ 1.923∗∗∗ 1.749∗∗∗ 1.532∗∗∗ 1.309∗∗∗ 1.337∗∗∗ 1.247∗∗∗ 1.202∗∗∗ 1.253∗∗∗ 1.229∗∗∗ 1.187∗∗∗ 1.139∗∗∗ 1.205∗∗∗ 1.151∗∗∗ 1.134∗∗∗ 1.111∗∗∗ 1.137∗∗∗ 1.092∗∗∗

(0.285) (0.295) (0.352) (0.309) (0.319) (0.321) (0.316) (0.324) (0.323) (0.322) (0.323) (0.296) (0.324) (0.297) (0.322) (0.318) (0.325) (0.319) (0.301)

Comecon Indicator Varible 0.843∗∗∗ 0.738∗∗ 0.752∗∗ 0.696∗∗ 0.508∗0.366 0.276 0.428 0.405 0.334 0.513∗0.359 0.327 0.375 0.334 0.312 0.214 0.198 0.263

(0.223) (0.246) (0.240) (0.241) (0.241) (0.245) (0.235) (0.221) (0.256) (0.234) (0.244) (0.218) (0.256) (0.246) (0.270) (0.248) (0.262) (0.267) (0.250)

Log of Distance -0.897 -0.490 -0.187 -0.914 -12.29 -1.112 -1.133∗-6.226 -3.567 -24.43∗-4.765∗∗ -8.513∗∗ -2.182∗-4.621 -1.965∗-6.049 -3.543 -1.464∗-1.690∗∗

(0.731) (0.806) (1.131) (1.414) (21.85) (0.843) (0.547) (4.563) (29.01) (11.64) (1.450) (2.987) (0.923) (3.109) (0.865) (4.698) (2.269) (0.584) (0.542)

Economic Freedom Index -0.00481 0.0556 0.0164 -0.0240 0.00974 -0.0489 -0.0397 0.0535 -0.0307 -0.0469 -0.112 -0.106 -0.0216 -0.0504 0.00503 0.0493 0.0599 0.000167 -0.0593

(0.0749) (0.0944) (0.0991) (0.0808) (0.0848) (0.106) (0.116) (0.136) (0.0904) (0.102) (0.0987) (0.0807) (0.0902) (0.0831) (0.100) (0.0878) (0.0913) (0.0869) (0.0779)

Contiguous Borders 0.182 0.267 0.245 0.304 0.373 0.390 0.337 0.286 0.248 0.311 0.220 0.242 0.468 0.517∗0.481∗0.538∗0.566∗0.593∗∗ 0.528∗

(0.224) (0.260) (0.284) (0.258) (0.258) (0.250) (0.234) (0.245) (0.234) (0.227) (0.227) (0.206) (0.263) (0.214) (0.232) (0.239) (0.240) (0.222) (0.209)

Exporter Remoteness Index 0.0517 -0.447 -0.802 0.00838 11.32 0.200 0.197 5.390 2.552 23.70∗4.019∗∗ 7.749∗1.457 3.900 1.261 5.382 2.919 0.937 1.143

(0.848) (0.929) (1.239) (1.509) (21.76) (0.930) (0.640) (4.666) (28.96) (11.74) (1.554) (3.062) (1.033) (3.206) (0.953) (4.788) (2.344) (0.655) (0.641)

Importer Remoteness Index 0.0149 -0.404 -0.734 -0.0404 11.37 0.122 0.105 5.122 2.513 23.33∗3.547∗7.334∗1.158 3.660 0.994 5.151 2.653 0.635 0.777

(0.661) (0.741) (1.041) (1.324) (21.93) (0.753) (0.458) (4.479) (29.03) (11.57) (1.385) (2.911) (0.838) (3.027) (0.768) (4.604) (2.152) (0.463) (0.448)

Constant -2.712 9.344 19.43 -0.137 -298.4 -2.821 -2.642 -132.9 -64.53 -585.7∗-94.58∗-185.4∗-34.09 -99.49 -28.62 -127.6 -68.00 -21.39 -26.43

(19.11) (21.15) (29.22) (35.49) (575.8) (22.77) (15.91) (116.6) (750.0) (290.7) (36.87) (73.23) (24.33) (81.39) (21.36) (113.0) (53.59) (13.65) (14.23)

Observations 216 216 216 216 216 216 216 216 216 216 216 216 216 216 216 216 216 216 216

R20.791 0.795 0.763 0.772 0.710 0.714 0.681 0.637 0.612 0.559 0.614 0.627 0.517 0.592 0.515 0.536 0.523 0.568 0.683

Standard errors in parentheses

∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

Table 1.5: PPML Three Year Panel Results

98-00 01-03 04-06 07-09 10-12 13-15

Log of GDP 0.735∗∗∗ 0.765∗∗∗ 0.785∗∗∗ 0.864∗∗∗ 0.844∗∗∗ 0.804∗∗∗

(0.0315) (0.0355) (0.0382) (0.0374) (0.0328) (0.0356)

USSR Membership 0.554∗∗∗ 0.269∗∗∗ 0.153∗∗∗ 0.108∗∗∗ 0.0837∗∗∗ 0.0632∗∗∗

(0.0829) (0.0345) (0.0220) (0.0177) (0.0131) (0.0117)

Comecon Membership 0.175∗0.0874∗∗ 0.0493∗∗ 0.0339∗0.0250∗0.0120

(0.0701) (0.0273) (0.0172) (0.0142) (0.0110) (0.00990)

Log of Distance -0.988∗∗∗ -0.955∗∗∗ -0.994∗∗∗ -1.017∗∗∗ -0.868∗∗∗ -0.737∗∗∗

(0.152) (0.0871) (0.0908) (0.159) (0.106) (0.101)

Economic Freedom Index -0.0568 -0.0374 0.0176 -0.0807 -0.0252 0.0362

(0.0596) (0.0541) (0.0640) (0.0601) (0.0540) (0.0548)

Contiguous Borders 0.483∗∗∗ 0.399∗∗ 0.313∗0.294 0.495∗∗∗ 0.581∗∗∗

(0.135) (0.147) (0.146) (0.156) (0.147) (0.155)

Exporter Remoteness Index 0.0776 0.0293 0.00585 0.0294 0.00850 -0.0308

(0.144) (0.0386) (0.0256) (0.160) (0.0861) (0.0527)

Importer Remoteness Index 0.0200 -0.0175 -0.0389 -0.00371 -0.0149 -0.0525

(0.116) (0.0344) (0.0215) (0.152) (0.0743) (0.0421)

Constant 0.316 0.759 1.783 -0.460 -0.645 0.283

(3.460) (1.373) (1.237) (4.316) (2.408) (1.370)

Observations 648 648 648 648 648 648

R20.705 0.717 0.610 0.560 0.531 0.499

Standard errors in parentheses

∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

USSR

Indicator

Varible

Comecon

Indicator

Varible

ed Py

| | |

ofr

tyr

[Tt

—*—

1998

—*—

1999

—*—

2000

—*—

2001

—*—

2002

—*—

2003

—*—

2004

—*—

2005

—*—

2006

—*—

2007

—*—

2008

—*—

2009

2010

—*—

2011

2012

—*—

2013

—*—

2014

—*—

2015

—*—

2016

Figure 1.1: Coefficients & Confidence Intervals for Year-by-Year Panel Regressions

USSR

Membership

Comecon

Membership

Hho

——

1998-2000

—*®—

2004-2006

—*—

2010-2012

——

2001-2003

—*®—

2007-2009

—*—

2013-2015

Figure 1.2: Coefficients & Confidence Intervals for Three-Year Panel Regressions

Chapter 2

Russia, Intra-Industry Trade, and

Historical Institutions

Travis Freidman

Abstract:

This paper is the first of it’s kind to examine how components of Intra-Industry Trade

(IIT) are impacted by historical trading institutions. Utilizing a Kandogan decomposition

system of equations of IIT and a Poisson Pseudo-Maximum Likelihood (PPML) gravity trade

model to show the decay path of historical institutions of trade flows between Russia and 183

countries from 1996-2018. In understanding the long reaching effects of the historical USSR

border and Comecon trading bloc, this paper finds that all components of IIT are positively

impacted by the historical trading institutions. In particular this paper find that Vertical

Intra-Industry Trade (VIIT) shows signs of the impact of these institutions decaying in a

monotonic fashion, while Horizontal Intra-Industry Trade (HIIT) indicates that the effect of

historical institutions is much longer.

2.1 Introduction

Institutions have been shown to be important in understanding economic growth out-

comes (Acemoglu et al., 2005). Further, there has been much agreement in the literature

that international trade also generates institutions worth studying, notably formalized by

Eichengreen and Irwin (1995). This paper in particular is interested in studying the impact

of previous trade institutions and international borders on past institutional trade patterns.

The specific focus is not only on historical trade institutions’ creation, but also on their

dissolving, and how those impacts ripple out through time. This paper has chosen to look

at current day Russia trade flows to understand the impact of the dissolving of the USSR

border and Comecon Trading bloc.

Besides having two trading institutions originate and dissolve in less than 100 year period,

studying Russia allows this paper to avoid the endogeneity issues innate to historical trading

institutions research (Barro and Tenreyro, 2007; Wolf and Ritschl, 2011). The creation of

the USSR and Comecon were, as Freidman et al. (2020) and others have argued, created

exogenous to trade flows. The result of USSR and Comecon’s exogeneity allows for the study

of the impact of historical trade institutions without fear of endogeneity of the dependent

and variables measuring historical institutions.

Although previous studies such as Djankov and Freund (2002a,b), Fidrmuc and Fidrmuc

(2003), and De Sousa and Lamotte (2007) have utilized Russia to study and demonstrate the

impact of USSR border and/or Comecon on modern day trade flow, these authors were never

able to demonstrate exactly at how these institutions decayed over time. Freidman et al.

(2020) built off this previous literature by examining more countries over a longer time frame

with a well-specified gravity trade model. Freidman et al. was able to show two important

results. First, that the historical institutions of the USSR border and Comecon were still

impacting current day trade flows. Second, they were able to visualize how these historical

institutions decay over time. Their work showed both historical institutions decayed in a

monotonic fashion, and that historical borders exert a stronger and longer lasting impact on

modern day trade flows.

While Freidman et al. showed how historical institutions decay over time, a limitation

of that study was that it only focused on total trade. Thus, the central question of this

paper is to ask, are there heterogeneous effects of previous trade institutions? That is to

say, is inter or intra-industry trade more (or less) impacted by old institutions? Further are

the components of Intra-Industry trade (i.e. Horizontal and Vertical Intra-Industry Trade)

uniformly impacted by historical trade institutions? This paper’s main contribution is to

examine if the past trading institutions of USSR borders and Comecon affect all the of the

components of total trade, (i.e. Inter-Industry Trade, Intra-Industry Trade, Horizontal Intra-

Industry Trade and Vertical Intra-Industry Trade) and how those institutions may uniquely

decay over time. This paper is able to show that the historical borders and trading unions

are most binding in Inter-Industry Trade and Vertical Intra-Industry Trade.

Intra-Industry Trade (IIT) has been an important area of study for international trade

economists since it’s characterization by Balassa (1966). It would take nearly a decade be-

fore the literature truly exploded post Grubel and Lloyd (1975). This was due in no small

part to the creation of the Grubel-Llyod index to explicitly measure IIT flows from Inter-

Industry Trade (INT) flows. Numerous theoretical models (e.g. Krugman 1979; Lancaster

1980; Falvey 1981; Helpman 1981; Helpman and Krugman 1985) have allowed for the un-

derstanding, and empirical prediction of IIT. The growth of the theoretical literature has

also pointed to the need to understand that IIT is actually composed of two different com-

ponents Horizontal Intra-Industry Trade (HIIT) and Vertical Intra-Industry Trade (VIIT)

(Greenaway et al., 1994).

The importance of examining HIIT and VIIT is more than just a theoretical consid-

eration. Numerous studies have sought to understand the empirical determinants of IIT.

Importantly Greenaway et al. (1994, 1995) illustrate that empirical studies that don’t dis-

tinguish between HIIT and VIIT are likely to have extremely biased coefficients for product

differentiation and scale economies. The bias due to the fact that HIIT and VIIT are driven

by different forces and adjustment costs. Broadly HIIT occurs between countries of similar

factor endowments trading similar goods of different varieties. Conversely, VIIT is done

between nations of different factor endowments at different points in the global production

process Falvey and Kierzkowski (1987); Jambor (2014).

Further, and more importantly to the work in this paper, identifying what factors empir-

ically move IIT, HIIT and VIIT at the country level is still a major debate in the empirical

literature (Lloyd and Grubel, eds 2003; Thorpe and Zhang 2005; Zhang and Clark 2009).

This paper argues that given the long reach of historical trading institutions, there is merit

to including them in empirical studies that look to measure these total trade components.

Including variables that empirically measure previous trading institutions may provide more

robust empirical measurements and help to overcome modeling fit issues.

Although the literature is resolute in the importance of decomposing IIT (Lloyd and

Grubel, eds, 2003), the agreement of how to define both HIIT and VIIT has been elusive

(Fontagn´e and Freudenberg, 1997). The heterogeneity in the definition of both HIIT and

VIIT has resulted in a plethora of methods to empirical characterize the horizontal and

vertical components of IIT. This paper employs a method that still is underutilized in the

literature.1Kandogan (2003b) demonstrates a way to decompose IIT, INT, HIIT, and VIIT

from Total Trade (TT). This method is both empirically free from researcher bias as to how

to define HIIT from VIIT as well as being theoretically consistent with the wider literature.

Of the papers that have utilized the Kandogan decomposition system of equations, only

a handful have employed it via a gravity model specification (Al-Mawali, 2005; Turkcan and

1At the time of writing this paper fewer than 150 papers employ this methodology despite being nearly

two decades old.

Ates, 2009; Leit˜ao et al., 2014). Importantly, in surveying the literature none of these papers

control for 2 important empirical specifications needed to make a theoretically consistent

gravity trade model: (i) multi-lateral resistance terms and (ii) a Poisson Pseudo-Maximum

Likelihood (PPML) estimation method. Drawing from the seminal works of Anderson and

van Wincoop (2003) and Silva and Tenreyro (2006, 2011), shows researchers that ignoring

iand ii produces empirically biased estimations. This paper’s first contribution to the

literature is that it is the first to estimate all the components of trade via the Kandogan

methodology and, with a properly specified gravity trade model. It is important to note

that this paper differs from a similar study Yotov et al. (2016), in that although Yotov does

also utilize both Kandogan method and a gravity trade model, they do not account for the

aforementioned proper gravity specification. Another point of similarity between this paper

and Yotov, is that although both papers look at Russian trade flows, this paper is interested

in the impact of historical trading institutions on current day trade.

The remainder of the paper is organized as first a background discussion on the theory of

IIT, literature on empirical modeling of IIT, and why USSR and Comecon can be considered

exogenous to trade. Next will be a description of the data, discussion of the explicit model

used in this paper, followed by a discussion of the results.

2.2 Background

In order to properly contextualize the work presented, the focus of this section will be in

answering the following questions: (i) what factors drive country-wide horizontal and vertical

IIT, (ii) what advantages does a gravity trade model specification brings to empirically

understanding IIT, and (iii) how the legacy of historical trade institutions can impact current-

day trade flows. In endeavoring to answer question i, a discussion of the theory of IIT, as

well as it’s horizontal and vertical components is necessary to understand the empirical

determinants of IIT.2

2.2.1 Intra-Industry & It’s Components

Mounting data evidence beginning in the early 1960s showed the existence of IIT. Al-

though they did not provide the theoretical foundation, much of the groundwork and pop-

ularization of how to identify IIT can be traced to Grubel and Lloyd (1975). In their work

Grubel and Lloyd defined IIT as “the simultaeous export and import of goods from the

same industry” (ibid, p. xii). Grubel and Lloyd illustrated that international trade theory,

dominated by Heckscher-Ohlin (H-O), needed to be adapted and modernized. Specifically,

a relaxation of the restrictions of perfect competition, constant returns to scale, constant

technology, and perfectly substitutable goods was needed in order to explain the mounting

empirical evidence of IIT.

As a response to the need for a framework to understand IIT, three theoretical expla-

nations emerged. The literature posited that IIT arises due to increasing returns to scale

(Krugman 1979, Lancaster 1980, Krugman 1980, Krugman 1981, & Helpman 1981), im-

perfect markets (Dixit and Grossman 1982, Eaton and Kierzkowski 1984, & Helpman and

Krugman 1985), or product varieties (Falvey 1981, Falvey and Kierzkowski 1987, & Flam

and Helpman 1987). Due to the complexity of IIT, there is no single class of model that

can explain the entire phenomenon. Since each model explains only part of IIT, each model

is considered a partial equilibrium model (Al-Mawali, 2005). Irrespective of the underlying

market assumption, each of these class of models provided a theoretically consistent way to

understand why trade within the same industry between similarly factor endowed nations

takes place.

2Since it’s discovery, the literature on new (and new-new) trade theory has grown nearly exponentially.

In an effort to keep the background section focused on the prime question, this paper suggest looking at

several outstanding surveys of the literature. Specifically, Lloyd and Grubel, eds (2003) for an overview of

seminal papers and Greenaway and Milner (2005, 2006) on how the literature has evolved.

As a result of the growth of the theoretical models in the 1980s, a body of empirical

literature emerged demonstrating the need for a distinction between horizontal and vertical

IIT (Lloyd and Grubel, eds, 2003, p. xiii). The first empirical studies to draw attention

to the need to disaggregate IIT were Greenaway et al. (1994), Greenaway et al. (1995),

and Torstensson (1996). Although Grubel and Lloyd (1975) did identify and gave an initial

definition of horizontal and vertical IIT, the aforementioned studies were the first to propose

and employ an empirical methodology to decompose IIT into HIIT and VIIT. Often it is easy

to paint a narrative about theory and empirical literature evolving in a sequential order. In

the case of IIT it would be incorrect to believe that the recognition of decomposition evolved

independently from the theoretical literature.

As the understanding of HIIT and VIIT has progressed, so has the definitions and the

theoretical foundations of the components of IIT. Most of the theoretical literature has

focused on understanding and predicting the horizontal component of IIT (Al-Mawali, 2005).

The focus on HIIT arises from empirical studies which did not disaggregate the horizontal

and vertical components of IIT, and chose the more common theoretical models of IIT based

off monopolistic competition (Thorpe and Leit˜ao, 2013). The theoretical underpinning of

HIIT comes from arguments from which monopolistic competition is arises.

The first theoretical basis for HIIT comes from “neo-Chamberlinian models” formalized

by Dixit and Stiglitz (1977) and Krugman (1979, 1980, 1981). These models assume that

consumers are motivated to consume as many different varieties of the same good (i.e. a “love

of variety”), which in turn drives IIT. The other theoretical foundation for HIIT comes from

“neo-Hotelling” models formalized by Lancaster (1979, 1980). In these models consumers

prefer a specific variety (i.e. a “diversity of taste”) and as result, promotes IIT between

countries. Irrespective of which theoretical grounding an author adheres to, the literature

generally defines HIIT as trade in different varieties of the same good.3

3For example Greenaway et al. (1994), Al-Mawali (2005), Thorpe and Zhang (2005), Thorpe and Leit˜ao

In contrast to HITT, there is less uniformity on both the motivation and theoretical

grounding for the vertical component of IIT. The diversity of both theoretical and definition

of VIIT is due in no small part to the fact that HIIT best explains trade between developed

nations and VIIT characterizes trade between developed and developing nations (Al-Mawali,

2005). To put it another way, why would two nations of radically different factor endowments

trade in goods of the same industry? Much like a matryoshka doll, in trying to understand IIT

the literature had arrived at the same questions that had given birth to New Trade Theory.

The literature has come up with two theoretical foundations to explain the phenomenon of

VIIT, the result of which is two different definitions.

The first theoretical foundation for VIIT is to assume a H-O model with perfect com-

petition. Utilizing this market structure results in defining VIIT as trade in goods of the

same industry but of different quality.4The differences in individual countries skill intensity

results in different production functions which, in turn, drives countries to organize via com-

parative advantage. Countries thus specialize in producing the same good but of different

quality as dictated by their factor endowments (Falvey 1981, Falvey and Kierzkowski 1987,

& Flam and Helpman 1987). Thus the differences in prices of the good across countries are

indication of the differences in quality.

The second theoretical foundation returns to an oligopolistic market structure to explain

VIIT. Under this specification the theory argues that VIIT is defined by trade of goods in

the same industry, but at different stages of production.5Here differences in factor endow-

ments are important at the sub-industry level as fixed costs in R & D result in specialization.

Different nations specialize in production of the final homogeneous good at either the inter-

mediate or final stage given their factor endowments (Jaskold-Gabszewicz and Thisse 1980,

(2013), and Aggarwal and Chakraborty (2017) use this definition.

4For example, this definition is used by Greenaway et al. (1994, 1995), Aturupane et al. (1999), Ekanayake

et al. (2009), Turkcan and Ates (2009), and Jambor (2014).

5For example Grubel and Lloyd (1975), Al-Mawali (2005), & Thorpe and Zhang (2005) use this definition.

Dixit and Grossman 1982, and Shaked and Sutton 1984). Thus, IIT involves vertical spe-

cialization as the intermediate and final good are part of the same industry and therefore

traded to complete production. Given the preponderance of definitions (as well as theoret-

ical foundations) of HIIT and VIIT, the following diagram is produced to help in keep the

“family tree” of IIT organized for the rest of this section.6

Intra-Industry Trade

Horizontal Intra-Industry Trade

Monopolistic

Competition

“Love of

Varieties”

Krugman

Different varieties

“Diversity of

Taste”

Lancaster

Vertical Intra-Industry Trade

Perfect

Competition

Comparative

Advantage

Falvey

Kierzkowski

Quality differentiated

Fixed Costs

in R & D

Globalization

Gabszewicz Thisse;

Dixit Grossman;

Shaked Sutton

Stages of production

While the theoretical literature has focused on explaining the foundation for and the

existence of IIT, empirical studies have turned their attention to testing the determinants

of IIT (Ekanayake et al., 2009). The majority of the empirical studies do break out HIIT

and VIIT from IIT, but are often focused on a particular industry7or set of industries

6This diagram is adapted from both Fontagn´e and Freudenberg (1997) and Al-Mawali (2005).

7See for example Turkcan and Ates (2009), Leit˜ao and Shahbaz (2012), Jambor (2014), Leit˜ao et al.

(2014), Konno (2016), Jambor and Leit˜ao (2016), or Lee (2018).

(ibid). A further complication is that due to the varied theoretical models, and the diverse

market assumptions that each model makes, results in a plethora of variables used (Lloyd

and Grubel, eds, 2003; Thorpe and Zhang, 2005). Despite the large volume of empirical

studies, there are some consistent factors that affect IIT and it’s components.

It is important to note that when considering the determinants of IIT, HIIT, and VIIT

the literature has identified that empirical variables can be further segregated into country

and industry-level factors. This paper is chiefly concerned with country-level factors and

thus, focus will be on those determinants. For an outstanding list of country and industry-

level factors see Zhang and Clark (2009) or Ekanayake et al. (2009) for a good discussion of

only industry factors.

Across the literature there are six country-wide variables that are consistently used as

metrics for the determinants of IIT and it’s components. In no particular order, the first

factor is the size, and specifically the size of the market created by the bilateral trade between

nations. Lancaster (1980) argues that as the size of the domestic economy grows so will the

number of different products supplied by the home market thanks to scales economies. On

the demand side, as the economy grows so will the demand for foreign varieties of goods.

Thus, the combined size of the two bilateral trading partners is important and is proxied

by GDP. In studies that examine multiple trading partners size is measured by the average

of the GDPs of the two nations (Thorpe and Zhang, 2005; Al-Mawali, 2005) or if the study

is focusing on one country’s trade then it is proxied by the trading partner’s GDP (Zhang

and Clark, 2009; Thorpe and Leit˜ao, 2013; Lapi´nska, 2016). Given the theoretical grounding

for this variable, the literature asserts that HIIT, VIIT will be positively correlated with

economic size.

The second determinant from the empirical literature is factor endowments. As articu-

lated by Helpman and Krugman (1985), factor endowments are important due to the nature

of international trade. As countries with dissimilar factor endowments trade more, the trad-

ing partners become more similar. According to Helpman and Krugman, this increase in

country similarity will result in an increase in demand for foreign varieties of the same good.

Similarities in factor endowments are positively correlated to HIIT and IIT. Conversely if the

two countries are dissimilar in their endowments then they are likely representing countries

in different stages of development, leading to increased VIIT between the partners. Draw-

ing from Helpman (1987), the literature proxies factor endowment via the difference in per

capita GDP between the trading partners.8Smaller differences equate to more similar factor

endowments between trading partners, thus more HIIT and IIT.

Importantly there exists a debate on whether per capita GDP proxies for factor endow-

ments or consumer tastes and preferences (Thorpe and Leit˜ao, 2013). Other authors have

argued that the difference in per capita GDP is a metric for similarity in per capita income

(Al-Mawali, 2005; Thorpe and Zhang, 2005; Thorpe and Leit˜ao, 2013). Thankfully, in the

context of effect on the components of IIT, per capita income has similar effects as factor

endowments. The argument being that increased similarity results in overlapping of con-

sumer tastes and preferences, as well as levels of development. Meaning that a negative

relationship exists between per capita GDP differences between countries and HIIT & IIT,

while a positive relationship exists with VIIT.

The third determinant of country wide IIT is distance. Krugman (1979, 1980) argues that

distance not only proxies for transportation cost of goods, but also for cultural similarity

between bilateral trading partners. Zhang and Clark (2009) additionally argue that larger

cultural divides make it more costly to import foreign goods, thus compounding distance’s

effects. Unlike the other factors, distance is easily measured between the two bilateral trading

partner’s capitals and is negatively correlated with IIT, HIIT, and VIIT.

A number of studies have been an interested in disentangling some of the cultural divides

captured by distance. Studies have looked at similar language (Kandogan, 2003b; Lapi´nska,

8I.E. Difference in Per Capita GDP= abs|GDPi

P opi−GDPj

P opj|

2016; Aggarwal and Chakraborty, 2017), but most have focused on common borders as a way

to disentangle distance effects. In the studies that look at the effect of a common border find

they exert a positive effect on IIT, HIIT, and VIIT (Balassa and Bauwens, 1988; Ekanayake,

2001; Lapi´nska, 2016; Konno, 2016; Aggarwal and Chakraborty, 2017). The aforementioned

authors argue that common borders reduces the cost captured in geographic distance, but

don’t provide a theoretical justification as to why.

A fourth standard empirical determinant is openness of the economy to international

trade. Although increased trade orientation of a country would lead to an increase in both

inter and intra-industry trade, Falvey (1981) argues that openness also corresponds to re-

duction in protectionist measures and thus increases all three IIT variables. Following Stone

and Lee (1995), most of the literature proxies this via the residuals from a per capita trade

on per capita income and population regression.

The penultimate determinant of IIT is trade imbalance. Due to the way that the Grubel-

Llyod index is constructed, as trade imbalance rises, the index reflects lower IIT (Grubel and

Lloyd, 1975). As a nation develops a trade deficit (excess), the lower the amount of similar

products exported (imported) can make up of total trade. Therefore trade imbalance is seen

to be negatively correlated with all three components of IIT, and as such a number of studies

create a varible to measure trade imbalance (Stone and Lee, 1995; Clark and Stanley, 1999;

Thorpe and Zhang, 2005; Zhang and Clark, 2009). In this paper the issue of bias from trade

imbalance will not be an issue. As enumerated on in the next section, the way in which this

paper makes account of IIT from INT does not rely upon defining a threshold to which to

designate an industry as IIT. Given the methodology utilized there is no need to include a

control for trade imbalance in order to control for omitted variable biasing.

The final determinant of IIT is Foreign Direct Investment (FDI). FDI has an ambiguous

affect on IIT in the literature (Gray 1988, Leit˜ao and Shahbaz 2012, & Thorpe and Leit˜ao

2013). Gray (1988) considered FDI as a substitute to trade, while Markusen (1984) and

Helpman (1987) see FDI inflows of a nation’s trading partner to be positively affecting their

bilateral IIT. To further complicate matters, FDI can be viewed as technology transfer from

one nation to another via capital goods thus leading to increases in IIT (Zhang and Clark,

2009). Alternatively, FDI could also reflect increased VIIT as increased FDI could reflect

the increase in multinational corporations expanding supply chains.

In reviewing the literature is it clear that there is a multitude of theoretical models to

inform how you define HIIT and VIIT. Due to the lack of uniformity of a generalizable

theoretical model, there is a plethora of empirical variables utilized. From looking at the

literature there is six consistent factors size, factor endowments, distance, openness, trade

imbalance, and FDI. A discussion of which empirical variables are used and how they connect

to the aforementioned theoretical framework will be done in the empirical mode section.

2.2.2 Decomposing & Modeling Intra-Industry Trade

Due to the various theoretical groundings for IIT and it’s components, there are no

uniform government statistics delineating what is and isn’t IIT. Therefore any empirical work

must first decompose IIT, as well as HIIT and VIIT, from available trade data. The Grubel-

Lloyd index, developed in Grubel and Lloyd (1975), is uniformly employed in delineating

inter from intra-industry trade.

The general form of the index takes can be expressed by the equation

GLi= 1 −|(Xi−Mi)|

Xi+Mi

(2.1)

where home’s IIT for any industry iis measured in terms of their exports of that industry

(Xi) and imports of that industry (Mi). The measure takes a value between one and zero,

where values closer to one signify higher degree of IIT within said industry.

While there is much agreement on the utilization of the Grubel-Lloyd index, there is

less agreement when it comes to disentangling HIIT and VIIT from IIT. Empirical studies

decomposing HIIT and VIIT have employed two general methods, the first method was

introduced by Greenaway et al. (1994). Greenaway et al. decompose the Grubel-Llyod index

results into HIIT and VIIT by stating vertically differentiated goods are ones in which the

unit values at the SITC 5-digit level differ more than ±15 per cent. The second method comes

from Abd-el-Rahman (1991) and is expanded upon by Fontagn´e and Freudenberg (1997)

and Fontagn´e et al. (2006). The methodology introduced by the aforementioned papers

meticulously categorizes each trade flow as either horizontal (based off different varieties

definition) or vertical (based off quality differentiated definition), and computes the share of

total trade by each category.

The major concern and criticism of this methodology, called the unit value dispersion

method, is in how it establishes a link between observed unit values and uses that to deter-

mine the quality ranking of each good. The unit value dispersion method assumes products

are vertically differentiated when the differential between export and import value is larger

than a certain threshold. The criticism is that the value threshold is “arbitrary” (Lloyd

and Grubel, eds, 2003; Kien and Thao, 2016) and could lead to inflating measurements of

VIIT (Kandogan, 2003a; Zhang and Clark, 2009; Thorpe and Leit˜ao, 2013). The over mea-

surement of VIIT is argued to occur via conflating true VIIT with IIT in quality vertically

differentiated products (Schuler, 1995). In short utilizing a definition of quality differentiated

products leads to possibly over-counting a component of IIT.

A method that has gained traction in the literature comes from Kandogan (2003b).9

Here the author begins by utilizing the stages of production definition for VIIT. Kandogan

utilizes two different levels of aggregation from the SITC, where higher levels of aggregation

define industries(2-digit SITC) and lower levels (4-digit SITC) designate products within

9It is important to note that the methodology and system of equations is first proposed in Kandogan

(2003a). In this he uses a slightly altered formula but the concept and overall structure is consistent when

he publishes the second paper.

that industry. The method allows Kandogan to look at the values of exports and imports

without needing data on quantity nor setting any value based threshold to delineate HIIT

from VIIT. With this methodology the author outlines a system of equations to empirical

decompose international trade.

Kandogan begins with the following equation to define what an industry is

Xi=X

Xip Mi=X

Mip (2.2)

where Xis exports, Mis imports, idenotes a particular industry, and pis a specific product

in said industry. Given equation 2.2, the total amount of trade in each industry (T Ti)

can be found summing up exports and imports in each industry via equation 2.3. Intra-

industry trade for each industry (IITi) is found by matching exports and imports at higher

levels of aggregation for each industry and subtracting from total industry trade, illustrated

by equation 2.4. The amount of trade of similar products, horizontal intra-industry trade

(HIITi), is calculated via the amount of matched trade in each product of an industry at

lower levels of aggregation, thus giving us equation 2.6. Finally, vertical intra-industry trade

(V IITi), is found from trade of different products/products at different stage of production

in an industry, by subtracting equation 2.4 from equation 2.6 to get equation 2.7.

T Ti=X

Xip +Mip =Xi+Mi(2.3)

IITi=T Ti− |Xi−Mi|(2.4)

INTi=T Ti−IITi(2.5)

HIITi=X

Xip +Mip − |Xip −Mip|(2.6)

V IITi=IITi−HIITi(2.7)

The Kandogan system of equations creates a uniform way to decompose all the compo-

nents of trade and avoids over-biasing any of the measures. Further this system of equations

can be used to describe the heterogeneous effects of trade by industry or can be summed by

category (i.e. IIT =PiIITior HIIT =PiHIITi, etc.) to get country trade.

Interestingly, Kandogan shows the efficacy of his system of equations via a gravity trade

model. Kandogan argues that the beauty of a gravity trade model is that you can add

variables to account for empirical factors predicted by both H-O and the Increasing Returns

models. He goes on to show that even including both sets of variables, total trade can not

be accounted for as the factors affecting each model move counter to each other. He then

re-runs his models with only his INT and IIT variables constructed from his index and find

that INT is better predicted by the H-O theory variables while IIT is better predicted by

the Increasing Returns theory variables. What is important to point out is that Kandogan

doesn’t make a claim as to why the gravity model was an ideal empirical specification for

employing his decomposition method.

Kandogan’s agnostic stance to model specification has resulted in a plethora of different

model employed with his index. In the few papers that utilize his index, only one uses a

gravity model, specifically Konno (2016). The remainder of the papers employ a plethora of

methods such as a Tobit with random effects (Zhang and Clark, 2009), a pooled OLS models

with random effects (Thorpe and Leit˜ao, 2013), and a GLS (Kien and Thao, 2016).10 Even

expanding the survey of literature to non-Kandogan decomposition methods we find that

only a handful of papers utilize a gravity model specification (Al-Mawali 2005, Sohn 2005,

Turkcan and Ates 2009, Leit˜ao and Shahbaz 2012, and Leit˜ao et al. 2014). Interestingly all

of the aforementioned studies which do employ a gravity specification don’t necessarily make

10There is a great heterogeneity in empirical models used to examine IIT and it’s components. A brief

literature survey of non-Kandogan decomposition methods results in empirical models specified as OLS

(Thorpe and Zhang, 2005; Lapi´nska, 2016), Logit (Ekanayake et al., 2009), log-normal hurdle model (Lee,

2018), dynamic GMM pannel (Jambor and Leit˜ao, 2016), and panel corrected standard error model (Jambor,

2014).

a case for why it might be an ideal model.

Given the volume of theoretical models and empirical variables that have been ascribed

to IIT and it’s components, the gravity trade model offers a flexible framework to specify

and test the components of IIT. The basic gravity framework draws a parallel to Newton’s

Law of Gravity to explain trade flows. Specifically, the theoretical argument states that

trade flows between any two locations are positively correlated with the combined GDP

(analogous to size in the Newtonian model), and negatively correlated to distance between

the two countries (which mirrors the distance between two physical particle in Newtons law).

When it was originally formulated the gravity trade model lacked a theoretical foundation

as well as plagued by a particular methodological issue called “the border puzzle”.11 After

Anderson and van Wincoop (2003) gravity models were shown to be theoretically consistent,

able to predict general equilibrium effects, and could solve the border puzzle if properly

specified (Yotov et al., 2016). Given Anderson and Van Wincoop’s result, the utilization of

gravity exploded in the empirical literature, though to the question of how generalizable the

results were took another important contribution.

Arkolakis et al. (2012) argue that the gravity model is unique in that it is able to estimate

the the welfare gains from trade across a large class of models. Arkolakis et al. makes four

micro assumptions (i) one factor of production, (ii) Dixit-Stiglitz preferences, (iii) linear

cost functions, and (iv) perfect or monopolistic competition. The authors further restrict

their model with three macro-level restrictions of (a) balanced trade, (b) aggregate profits

are constant share of aggregate revenue, and (c) import demand is CES. These assumptions

allow the authors to be consistent with several micro founded models, specifically Armington-

CES, H-O, Monopolistic Competition, Heterogeneous Firms, Ricardian, Sectoral Ricardian,

Sectoral Armington-CES, and Dynamic Factor Accumulation models.

11Briefly, the border puzzle refers to the empirical result of after controlling for distance, regions within

countries trade much more with each other than do regions across countries (McCallum, 1995; Anderson and

van Wincoop, 2003; Ishise and Matsuo, 2015).

The result of Arkolakis et al’s assumptions is that the authors are able to show that

despite their micro-welfare results, all of the aforementioned models welfare predictions can

be simplified into a single equation asking what are the changes to real income as a function

of foreign shock from trade costs. In developing an equation to express their overarching

welfare effects of trade, the authors show that the gravity equation model is a common

estimator for all of the aforementioned models since “by its very nature, it captures by how

much aggregate trade flows, and therefore consumption, reacts to changes in trade costs”

(ibid, p. 119). The assertion of gravity being a consistent estimator for a whole class of

models is bolstered by Allen et al. (2020).12 Allen et al. illustrates sufficient conditions for

the existence and uniqueness of the trade equilibrium for the same group of trade models

utilizing the standard gravity constraints.

There is clear theoretical evidence for the gravity trade model being a consistent and

theoretically justified model choice for IIT. The gravity model is able to be isometrically

equivalent to a number of micro-founded trade models as well as provide a general equilibrium

framework for them. The result of this equivalences is that gravity is able to ask the same

big “gains from trade” question that each of the various IIT models do. Furthermore, the

gravity model has a flexible empirical framework such that variables can be added which

help to explain various theoretical models without conflict. Anecdotally, in looking at the

previous section, two well documented empirical movers of IIT, size and distance, are also

two of the core theoretical and empirical factors of the gravity trade model. Thus given

the evidence presented, a gravity model with a Kandogan system of equations would be an

optimal choice to both theoretically and empirically model Russian trade flows.

The one caveat to this endorsement is the assumption that the gravity trade model

needs to be specified correctly. Correct specification in the context of a gravity trade model

12Note that the conclusions of Allen et al. (2020) first appear in Allen et al. (2014) and have only recently

been formalized in the aforementioned publication. Ergo, their results are usually cited by the latter, not

the former paper.

requires, at the very least, that a model contains Multi-Lateral Resistance variables as ar-

ticulated in Anderson and van Wincoop (2003). In looking across the papers that utilize a

gravity trade model to explain IIT, none of them control for border effects in a theoretically

consistent way. The empirical process of how to control for the boarder effect will be further

enumerated on in the empirical model section. The work presented in this paper represents

the first time a Kandogan decomposition is applied with a correctly specified gravity model.

Yet, a well specified gravity model is not just a function of multilateral resistance terms. To

put it another way, controlling for border effects are a necessary not a sufficient condition

for a well specified gravity model. As argued in Freidman et al. (2020), in order to achieve

a well specified gravity model, the importance of historic institutions and their impact on

future trade flows of the country of interest must be taken into account.

2.2.3 History’s Effect on Trade Flows

Institutions have been studied widely in economics for a long time, and have been found

important in the determination of trade flows by several authors. Despite economics’ long

fascination with institutions, Acemoglu et al. (2005) was the pivotal work in providing a

framework and argument for why institutions matter. In short, Acemoglu et al. argue

that institutions shape and incentivize market actors, this in turn organizes productivity

and results in the observed differences in economic growth between countries. Since it’s

publication, Acemoglu et al. has inspired interest in the impact history and institutions

have on trade (Estevadeordal et al., 2003; Mitchener and Weidenmier, 2008; Karnups, 2008;

Brodzicki and Uminski, 2018) and economic growth (North, 1995; Zukowski, 2004; Campos

et al., 2016). In looking specifically at trade, the literature has found strong legacy effects of

institutional trading arrangements in virtually all cases that have been studied(Eichengreen

and Irwin, 1995; Anderson and Smith, 2007; Stack et al., 2019; Freidman et al., 2020).

Institutions, in the economic context, are generally thought of as property rights, rule

of law, competitive markets. Yet, upon reflection one realizes that international trade also

generates institutions that have impacts on how society organizes itself. Eichengreen and

Irwin (1995), the seminal paper on the matter, makes the argument that historical trading

institutions impacts current trade flows. Utilizing a gravity trade model with a lagged

trade variable, Eichengreen and Irwin investigate the influence of pre-WWII trade (1928 and

1938) on post-war trade flows. The authors find strong, but diminishing, effects for 1949 and

1954. With specific reference to countries that had once been part of the British Empire,

they find that “Former British colonies traded disproportionately more with one another in

1949. . . because of the effects of history” (ibid, p. 55). In general, Eichengreen and Irwin

persuasively argue that history is fundamental for the determination of trade flows in any

gravity approach leading to the literature of investigating historical trading institutions.

Anderson and Smith (2007) attempt to validate the results of seminal papers historical

trading institutions’ impact on current day trade flows. The authors use a panel data set

and a lagged trade variable specification from Eichengreen and Irwin (1998) and find strong

evidence that historical patterns do matter in the estimation of trade flows in Canadian trade.

Using a fixed effects approach to estimate the gravity equation, they show that importer and

exporter time fixed effects can capture the effects of history without the use of a lagged

dependent variable approach. Making the case that researchers need to put time and effort

into ensuring that the gravity trade model is correctly specified.

Since the above-mentioned papers, the application of history’s effect in empirical models

has been deployed in a more nuanced way. Newer research into historical institutions from

Gowa and Hicks (2013), Brodzicki and Uminski (2018), and Stack et al. (2019) seek to

appropriately calibrate the gravity trade model to take important historical factors that still

affect current (and future) trade volumes into account. Gowa and Hicks (2013) look at trade

volume and the effects of trade blocs on trade during the intervening years between World

War I and World War II. They take into consideration that the trade blocs that were formed

Post World War I, had different political aims(all of which shared the goal of trying to curb

intense global economic downturn) depending upon which major power formulated them

when specifying their gravity model. They find that, contrary to recent literature, none of

the great power trading blocs affected trade in positively or negatively.

Brodzicki and Uminski (2018) include variables that account for the historical metro-

polis of Poland to understand foreign trade persistence and development. Using a PPML

gravity model they find that there is evidence of trade flows being a function of the historical

partitions and metropolises of Poland. Similarly, Stack et al. (2019) look at global trade flows

of sugar and accounts for colonization’s part in developing this market. In demonstrating

that colonial ties dictate current global sugar trade, Stack et al. show that the geographical

direction those colonial ties originate from can have either positive or negative effects on

growth and trade broadly.

Clearly the channels in which historical trade institutions affect current day trade are

broad. Arguably, some of the most interesting institutions are ones which exist for a short

period allowing for study of their limited effects. In the context of international trade,

institutions such as trade agreements, currency unions, and international borders represent

opportunities to explore the effects of intuitions which arise and potentially decay. What is

of particular interest for this paper is the effects of borders, both creation and dissolution.

The border effects literature seeks to exploit natural experiments of impact that the

generation and disbandment of national borders have on trade flows. Border effects have

been studied in a wide variety of settings that include cultural identity (Falck et al., 2012),

war (Che et al., 2015), and the reintegration of economies (Felbermayr and Gr¨oschl, 2014;

Nitsch and Wolf, 2013). Each study finds evidence of long-term persistence. Regarding

the elimination of borders, Felbermayr and Gr¨oschl (2014) find that by 1993, the historical

border between the Confederate South and the North (the Mason-Dixon Line) reduced trade

by 13% to 14%. However, some of this could be the result of endogeneity issues.

The question of the exogenous development of trade, customs, and monetary unions

permeates the historical trade institutions literature. The channels in which endogeneity

becomes an issue for the empirical models is twofold. First, by an a priori assumption about

formation of these international agreements. As (Barro and Tenreyro, 2007, p. 3) state “The

implicit assumption in various empirical studies is that currency unions (or, more generally

exchange rate arrangements) are randomly formed among countries” or second, because of

reverse causality issues. Specifically Wolf and Ritschl (2011), Baldwin and Jaimovich (2012),

and Keller and Shiue (2014) give different examples of how the application of national ar-

rangements may either increase trade flows or high trade flows depending on what politically

stimulated the formation of these agreements. Given the questions of endogeneity surround-

ing hysteresis literature, the literature has gone to great lengths to find natural experiments

to prove trade intuitions are exogenous (Nitsch and Wolf, 2013).

This paper and others13 argue that the Soviet Union (USSR) and Comecon can be con-

sidered as natural experiments for tests of the legacy effects of previous trading institutions.

Not only are these institutions founded for non-trade related reasons, there is no evidence

of strong trading relations between Russia and the other countries studied prior to the de-

velopment of the USSR or Comecon.

There were many reasons for the formation of the Soviet Union that began in 1922 with

the unification of the Russia, Transcaucasia, Ukraine, and Byelorussia. By 1940 included

15 sub-national Soviet republics existing until 1991 (see Table 2.2 for a list of countries).

A fundamental factor driving the unification was ideology (Sherman, 1994). Specifically,

the Bolsheviks believed that in order to have socialist utopia the ideas of personal property,

national identity, and individuality needed to be abolished. By uniting all the soviet na-

tions under one communist government they could progress their people towards socialism

13See discussion in the introduction on papers Djankov and Freund (2002a,b), Fidrmuc and Fidrmuc

(2003), De Sousa and Lamotte (2007), and previous work by Freidman et al. (2020).

(C.P.S.U., ed, 1939; Stalin, 1953; Sakwa, 1999). Further, there is no evidence that direct

trade-related rationales motivated the formation of USSR as prior to USSR formation, no

future member states were listed among the top 18 import or export partners of Russia (Vy-

acheslav, 2011). It is easy to be persuaded that the USSR wasn’t created for trade purposes,

but Comecon presents a more difficult case to be made.

The Council for Mutual Economic Assistance(CMEA) also referred to as Comecon, was

the main trading bloc of the Soviet Union from 1949 to 1991. Although all members were

equal, the organization was under the leadership and control of the Soviet Union. Come-

con comprised the countries of the Eastern Block along with satellite states.14 The official

purpose of Comecon was to coordinate planning, promote country and regional specializa-

tion, increase trade among member states (Korbonski, 1970), and “to improve economic and

military cooperation” (New York Times, 1988). Increased trade flows was among the moti-

vations for the formation of Comecon, but like the USSR’s foundation, this was not based

on any pre-existing strong or rapidly increasing trade relations (Freidman et al., 2020).

The most direct reason for Comecon’s development was, similar to the USSR, ideology.

Comecon was founded in response to the Marshall Plan and to counter the OECD (Brine,

1992). Comecon was seen as an effective instrument to spread communism to the countries

of the Eastern European block with the USSR being the dominant member (ibid). The

political value of Comecon outweighed any economic desires as evidenced by the fact that in

1956, six years after its formation, there was insignificant intra-Comecon trade (Korbonski,

1970, p. 957). The literature on the development and operation of Comecon demonstrates

that it was a poorly designed and managed trading institution that resulted in little true

trade creation and was mostly trade diverting (Pelzman, 1977; Holzman, 1985; Zickel, ed,

1989; Biessen, 1991). Additionally, and most importantly, Comecon was meant mainly as a

control device for the Soviet Union over Comecon members (Pelzman, 1977). Thus, it can

14Former members of Comecon are listed in Table 2.2

be concluded that the dissolving of the USSR borders and Comecon trading bloc presents

a unique, and exploitable natural experiment from which to interrogate the importance of

historical intuitions on current day trade flows.

This paper is not unique in arguing for utilizing USSR and Comecon as natural exper-

iments. Other studies that investigate the effects of USSR on bilateral trade are Djankov

and Freund (2002a,b), Fidrmuc and Fidrmuc (2003), and De Sousa and Lamotte (2007).

Djankov and Freund (2002a,b) use the gravity trade model to examine trade flows among

and between 9 Russian regions and 14 former USSR republics during the period of 1987-

1996. The authors are able to shown that Russian regions traded significantly more with

each other than with former Soviet Union republics. A limitation of their analysis deals

with understanding the full arc of the legacy of historical trading institutions since they only

employ data from 5 years after the collapse of the Soviet Union.

Fidrmuc and Fidrmuc (2003) examine three disintegrated unions Yugoslavia, Czechos-

lovakia, and the Soviet Union (represented by Russia, Ukraine and Belarus). In order to

capture different trade relations, they include variables for formal preferential trade areas,

common border or language, and successor states of former federations in Europe. With

data covering the period 1990 to 1998, their results suggest that the trade effects of for-

mer institutions decline rapidly over the 8 years, however, trade relations between former

members remain significant to 1998.

De Sousa and Lamotte (2007) attempt to determine why the legacy effects found by

Fidrmuc and Fidrmuc (2003) dissipate quickly relative to other findings. Utilizing controls

suggested by Anderson and van Wincoop (2003) and a data set from 1993 to 2001, they

were able to include all countries created by the political disintegration of the Soviet Union,

Czechoslovakia and Yugoslavia, de Sousa and Lamotte find more persistence. The authors

found that the results of Firdmuc and Firdmuc were biased by the limited number of former

Soviet, Czechoslovakian and Yugoslavian states covered in their study, more than their lack

of multilateral resistance controls.

As discussed in the introduction, these papers primarily focused on showing the connec-

tion of historical trading institutions and Russia’s current trade flow. This paper is unique

in that it focuses on measuring how long these institutions last with a well specified gravity

trade model.

2.3 Description of Data

The data utilized in this work is created from merging four different data sets. The

data encapsulates bilateral trade flows to and from Russia in the years from 1996 to 2018.

The specific data sets used are trade data from the United Nations’ COMTRADE database,

GDP and FDI data from the World Bank, bilaterat distance between Russia and their trading

partners from the Centre d’´

Etudes Prospectives et d’Informations Internationales (CEPII),

and the Index of Economic Freedom from The Heritage foundation.

The data comprises export and import trade to Russia and 183 countries as listed in

Table 2.2 in the appendix. Specifically, from the COMTRADE database trade flows at

both the SITC15 2-digit and 4-digit commodity code level. Further this data is reported in

nominal USD, so the values were transformed into constant 2010 USD using CPI data from

the World Bank. Performing this action also makes the trade data the same denomination

as the data from the World Bank.

CEPII GeoDist database has number of distance variables constructed in differing levels

of complexity. Of the available variables, CEPII’s simple distance variable had the largest

number of distance pairs between Russia and different country partners. The distance vari-

able from CEPII is calculated utilizing the great circle formula, which find the distance

between the most important cities (in terms of population) for each country. (Mayer and

15SITC revision 3 codes are employed in this paper as the COMTRADE database does not report revision

4 codes.

Zignago, 2011)

The Heritage Foundation’s Economic Freedom Index makes up the last data set that

makes up the base of the data used in this paper. The foundation ranks countries based off

12 qualitative and quantitative metrics16 which broadly fall into the categories of (i) rule of

law, (ii) government size, (iii) regulatory efficiency, or (iv) open markets. Each of the twelve

categories are scored from 0 to 100, where higher scores indicate more “freedom.” This paper

employs the overall score, which is the average of the individual scores of each of these 12

categories.

2.4 Empirical Model

Before reviewing the empirical model employed, it is worth taking a moment to discuss

how this paper constructs the dependent variables utilized in the analysis. This paper is

concerned with looking at the effects of historical trade institutions on the Total Trade

(TT), Intra-Industry Trade (IIT), Inter-Industry (INT), Horizontal Intra-Industry Trade

(HIIT), and Vertical Intra-Industry Trade (VIIT). As discussed in section 2, the methodology

this paper uses to identify the different trade flows is via the Kandogan decomposition

method. As identified by both Konno (2016) and Kandogan (2003b) the United Nations’

Standard International Trade Classification(SITC) provides a uniform and internationally

standardized way of classifying traded goods.

The SITC classifies commodity flows into varying levels of aggregation of goods, where

higher aggregation numbers equate to more granular classification of commodities. The

aggregation levels are 1-5 with the following descriptions of 1 denoting sections, 2 is divisions,

3 is groups, 4 is subgroups, and 5 is items. Thus given the level of aggregation employed

each good traded will be assigned a commodity code with digits corresponding to the level

16The twelve individual categories are: property rights, government integrity, judicial effectiveness, gov-

ernment spending, tax burden, fiscal health, trade freedom, investment freedom, and financial freedom.

of aggregation i.e. if you were looking at aggregation level 3 commodities would be classified

with 3-digit codes.

Provided in Table 2.1 below, is a sample of SITC codes at all five levels of aggregation.

All nine basic 1-digit categories are provided for reference, the rest of the table goes into

deeper levels of aggregation of commodity groups zero and seven. From the table it is easy

to see that either 5 or the 4-digit SITC code commodities would be good candidates for

defining products. The 4-digit code was selected as difference between each commodity can

be thought of as a distinct good. Looking between 1, 2, and 3-digit commodity codes, the

2-digit level of aggregation is selected as it aggregates goods into reasonable approximations

of industries.

Using the SITC codes to define industries and products, the equations of 2.3-2.7 are

constructed from the trade data. Since the equations return inter and intra-industry trade

by industry, summing up all the terms by varible and industry gives us our dependent

variables of TT, IIT, INT, HIIT, and VIIT. The base empirical model specification takes

the form of

T radeiR,t =expβ1lnGDPi,t +β2lnDistiR +β3CNT G +β4Fi,t+

β5lnDGDP P Ci,t +β6lnF DIi,t +β7USSR ·t+

β8Comecon ·t+β7lnMLRexp,t +β8lnMLRimp,t∗εiR,t (2.8)

Where trade takes on the form of one of the dependent variables, indicating the type of trade

between Russia and some country iin year t.

Before discussing the independent variables, it’s important to note that all of the models

are estimated with a Poisson pseudo-maximum likelihood (PPML) specification. A PPML

model specification solves two problems of the gravity trade model. First, the PPML solves

the problem of zero values in trade flows. The traditional specification of gravity models

is to put the dependent variable in logarithmic form, thus resulting in zero trade flows as

undefined. Using the PPML specification allows researchers to include zero values. Second,

the PPML solves a much more pressing issue, heteroskedasticity in trade data.

As Silva and Tenreyro (2006) makes clear, when the data is heteroskedstic a log-linearization

of the model will result in biased and inconsistent results. The authors demonstrate this

result as a function of Jensen’s inequality and “the expected value of the logarithm of a ran-

dom variable depends on higher-order moments of its distribution. Therefore if the errors

are heteroskedastic, the transformed errors will be generally correlated with the covariates.”

(p. 653, ibid) Given their findings in both Silva and Tenreyro (2006, 2011), the authors ad-

vocate for the use of PPML model specification as the industry standard for gravity trade

models.17

Turning to the dependent variables of equation 2.8, the variables of lnGDPi,t and lnDist

correspond to the the standard gravity variables of the natural log of GDP of country iin

year t, and the natural log distance(as described in section 3) between Russia and some

country i. As described in section two, these also correspond to standard empirical IIT

variables. Since we are only concerned with looking at Russia trade we will only use the

trading partners GDP to proxy the size of the combined market. A priori, it is expected

that all trade will be positively correlated GDP while distance will be will be negatively

correlated with trade.

The variables of CN T G and Fi,t are variables for an indicator variable if a country shares

a contiguous border with Russia and the overall score from the Index of Economic Freedom

(discussed above). The 14 countries whom Russia shares a border with are North Korea,

China, Norway, Finland, Ukraine, Kazakhstan, Poland, Georgia, Mongolia, Latvia, Estonia,

Azerbaijan, Belarus, and Lithuania. Of the 16 countries only North Korea, China, Norway,

17Additionally Xiong and Chen (2014) provides further evidence that a PPML specification, and not other

proposed models such as a Tobit or Heckman model, result in the best possible estimations of the gravity

model given the above outlined pervasive issues.

and Finland were not part of either the USSR or Comecon.

The Index of Economic Freedoms will be this papers measurement of economic openness,

where all measures of trade should be positively correlated with the index. The indicator of

continuous borders is a standard gravity variable that has also been used in several empirical

IIT studies mentioned in section 2. Drawing from the gravity and IIT literature, the common

borders indicator should also be positively correlated with all forms of trade.

As discussed in section 2, controlling for differences in factor endowments and investment

are key to well crafted models. The variable of lnDGDP P Ci,t is the log of the absolute

difference in GDP per capita between Russia and some country iin time t. While the

variable lnF DIi,t is the log of the net inflows of Foreign direct investment to some country

iin time t. As asserted by the background section we can expect the absolute difference in

GDP between Russian and it’s trading partners to be negatively correlated with HIIT and

IIT, while positively correlated with VIIT. Given the multiple interpretations of FDI from

the empirical literature, there is no a priori assumption of the correlation between FDI and

any of trade variables.

The variables of USSR and Comecon are the primary variables of interest for this paper.

These variables are constructed indicator variables articulating if a trading partner is a

former member of the USSR or Comecon respectively. In Table 2.2 you can find a list of

which countries were label as ex-USSR and ex-Comecon. These indicator variables are the

proxy for the historical institutional effects of dissolving borders and dissolving currency

unions respectively. In Freidman et al. (2020), a positive correlation between total trade and

these indicator variables were shown. Yet, to the best of this papers knowledge, there are

very few paper which look at the effects of historical borders or trade unions on different

components of trade, as discussed in the background section most focus on total trade. In

looking to the empirical literature on IIT, HIIT, and VIIT there is also a lack of investigation

into nuanced effects that current day borders and trade unions have on IIT. Because of this

lack of discussion in the literature, this paper must make some inferences to predict a priori

behavior of these indicator variables.

To begin with our measure for border effects, there are three papers which give some

indication on what to expect from the USSR variable. First, Chen (2004) investigates the

border effects within the European Union, investigating net exports at the “pooled”, country,

and industry level. The author is able to show that technical barriers to trade and product-

specific information costs increases border effects. The second paper comes from Wolf (2009)

who examines German economic unification from 1855 to 1933 using a well specified gravity

trade model. Wolf is able to show via the geography of trade costs that it took till the end

of the Weimar Republic in 1933 for economic unification to happen, even though Germany

officially unified in 1871. Although the country showed signs of integration as early as prior

to the outbreak of WWI, integration wasn’t uniform. Wolf argues that the results are driven

by cultural heterogeneity saying that it took at least one generation to break these barriers

down. The final paper is Jambor (2014), where the author investigates what drives HIIT and

VIIT in the agri-food trade of new members states of the European Union. Jambor argues

that economic integration of countries fosters the growth of IIT in agricultural products.18

The author goes on to argue and show that integration should have positive effects on HIIT

and VIIT. Interestingly, they note that the growth of IIT because of integration is primarily

driven by increases in VIIT.

Taken together,these three paper paint a picture of what can be expected of the USSR

variable. Wolf (2009) demonstrates that the process to integrate new territories into a nation

takes at least a generation. Given the length of time the member states were incorporated

to the USSR we should expect this effect to be strong and take at least the same amount of

time to dissipate. Secondly, Jambor (2014) shows that when new territories are integrated

18Jambor (2014) makes the argument that IIT agri-trade is boosted by economic integration of new

member states based off the well know empirical work of McCorriston and Sheldon (1991) and Qasmi and

Fausti (2001).

that those territories see an increase in IIT primarily driven by VIIT. Although Jambor was

looking at agricultural trade, the author’s result might be more generalization as work done

by Aturupane et al. (1999) investigates reintegrating eight former Central and Eastern Bloc

states to the European community. Aturupane et al. also found that IIT increases between

the European Union and these states, which is also driven by VIIT. Taken together this

indicates that we should see strong postive effects on VIIT from historical borders, which

may tend to dissipate over time as these states become disintegrated. Finally, the result

from Chen (2004) indicates that as states integrate that borders produce higher costs of

importing different foreign varieties of goods. Given this result, USSR is expected to be

positively related to HIIT because as these states reemerge as independent they’ll have a

stronger affinity to the varieties of their former home state. Unlike VIIT there seems to be

no evidence that this effect will taper off.

Turning to the Comecon variable, there is also two paper which help inform the prediction

of it’s effects on HIIT and VIIT. The first paper comes from De Sousa (2012), in which the

author examines net exports using a PPLM gravity trade model for 203 countries over a

60 year period. In De Sousa’s paper, he finds that the currency union effect19 dissipates

over time. Specifically after an initial positive effect, currency unions decline in impact to

zero after 35, and then oscillating between negative and positive, and finally after 50 years

the currency union effect becomes statistically equal to zero. Although Comecon was not a

currency union, the particular point to take from this is the fragility of these institutions,

meaning we should expect a weaker effect of Comecon on all the types of trade and that its

effects should decay faster.

The final paper to help predict the direction and effect of Comecon is Ekanayake et al.

(2009). In their paper the authors investigate HIIT and VIIT of US-NAFTA trade from

19First documented by Rose (2000), any two countries that share a currency trade three times as much as

they would if they had differing currencies.

1990 to 2007. The authors find that IIT increases over their data period and that it is driven

by VIIT. They argue that increased specialization via division of labor between parties

resulting in a range of quality within the same industry, the authors call this “qualitative

division of labor.” From Ekanayake et al.’s results we can argue that Comecon will be

positively correlated with all the IIT, VIIT, and HIIT. This paper expects that the strongest

correlation between VIIT and our measure of trading bloc, but given the results of De Sousa

(2012), these effects will be weaker than border effects and will dissipate faster.

Lastly, the final two terms of lnMLRexp,t and lnMLRimp,t are multilateral resistance

(MLR) terms used to overcome the “border effects” implicit in every gravity trade model. As

described by Anderson and van Wincoop (2003), trade resistance between any two countries

(iand j) can be decomposed into three specific effects: (i) bilateral trade barriers between

region i&j, (ii) i’s resistance to trade with all regions in the world, and (iii) j’s resistance to

trade with all regions in the world. Simple remoteness variables, Anderson and van Wincoop

argue, only captures distance from bilateral trading partners (effect (i) from above), while

the MRT capture all three affects. Although extremely important, MLR’s are a theoretical

construct, and as such must be generated. The border literature has developed two easily

deployable empirical solutions to construct MLR for researchers (Yotov et al., 2016).

The easiest and by far simplest MLR terms are the exporter (and importer) paired-

time fixed effects variables, where an indicator variable is created for when country itrades

with country jin time t. These fixed effects capture the “special” underlying factors that

resulted in these two countries trading in this particular time. As one can see this lines up

nicely with the three effects described by Anderson and van Wincoop (2003). The problem

with deploying this solution with the current model specification is that of all trade is to

and from Russia, thus creating indicator variables that are multicollinear. The colinearity

problem is compounded with the total of 183 countries, resulting in indicator variables absorb

all variation in the data; making this an untenable solution.

The second empirical solution is referred to as a “remoteness index,” and is what is

utilized in equation 2.8. The remoteness index is a reduced form version of the custom

built MLRs that Anderson and van Wincoop introduced. These remoteness indexes are

output and expenditure weighted averages of bilateral distance. They are constructed via

the following two equations:

REM EXPi,t =ΣjDistij /Ej,t

Yt(2.9)

REM IMPj,t =ΣiDistij /Yi,t

Yt(2.10)

Where Ej,t is the value of importer expenditure, obtained by summing up the value of all

trade exported by country jin year t. Similarly, Yi,t is the value of exporter output by

country iin year t. In equation (2.9), the variable Ytis sum of all Ej,t in a year then utilizing

the max value of that year. Conversely, in equation (2.10), Ytrepresents sum of all Yj,t in a

year then utilizing the max value of that year.

This empirical model is applied three different specifications. The first is a regression

with all the available years in the data set, which can be used to test the basic assumption

of our historical institutions validity on modern trade. The second specification is regressing

the model with only a year’s worth of data at a time (i.e. t=1996 or 1997, etc.) Using the

second specification will make the full time trend of our indicator variables visible, but with

the cost of picking up a lot of ”noise” from the data. Thus the final, and most important,

specification is in 3-year intervals(ie regression over years 1996-1998 then 1999-2001 etc). As

was shown in Freidman et al. (2020), using a multi-year regression allows for the time trend

to be made visible but filters out the noise found in the second specification.20

20Multiple year interval regressions are also a best practice suggestion from the literature (Yotov et al.,

2016). Specifically to overcome trade policy changes over the time of the data. If there is dramatic changes

to a trade policy from year 1 to year 2, then it will likely generate influential outlier data points depending

the total span of the data. A simple solution the policy change issue is to use panel data over multiple year

intervals.

Finally it is important to note that this paper doesn’t included a variable outlined as one

of the standard empirical IIT variables, namely trade imbalance. The reason for this choice

is that the motivation for that particular variable is a direct result in how the Grubel-Llyod

index is created. Without controlling for trade imbalance an empirical study is subject to

over-inflating IIT. Utilizing the Kandogan system of equations, this particular issue will not

affect the measures of TT, IIT, INT, HIIT, or VIIT. Thus trade imbalance has been omitted

from the list of empirical variables for this paper.

2.5 Results

2.5.1 All Years Panel Results

Looking at Table 2.3 will show all the regression results for each of the 5 dependent

variable under the “all years” specification. Generally speaking it seems that overall the

data is well explained by the empirical model.

Looking across the variables at GDP we find that it is positive and statistically consistent

across all of the models at the highest levels. Interestingly, distance is the correct sign in

all the models but is not statistically significant with the same strength as GDP. Distance

is not statistically significant for INT, but is at the 95% level for TT and at the highest

levels for the remaining models. The coefficients across all the models shows that these two

characteristics have large impacts on trade, which is consistent with the literature.

Turning to contiguous border, we see that it is positive, with a large magnitude and

statistically significant at the highest level across all the models. By contrast, economic

freedom seems to varying effects. In the models of TT, INT, and HIIT, it has a positive

effect and only statistically significant in the former two. While economic freedoms is negative

in the IIT and VIIT models, with it only being statistically significant in the latter.

A counter point could be made a more persuasive variable would be one for political

freedom, not economic freedom, given the political history of Russia. To answer this potential

counter claim, the Freedom House’s Freedom in the World index was also employed in place

of the economic index in all the possible regressions. Political freedom was found to not

impact the statistical significance of any of the other independent variables across all the

specifications and dependent variable. Further, political freedom seemed to only inflate the

coefficients of the other independent variables, while decreasing the goodness of fit measure.

As such economic freedom is what is reported in this paper.21

Examining the results on differences in per capita GDP, Table 2.3 shows that all the

different measure of trade are positively correlated with this measure. Further, the table

reports that DGDP P C is significant at the highest levels for TT and INT, 5% level for

HIIT while not significant for IIT and VIIT. This finding is somewhat counter-intuitive

given that a positive correlation would be most associated with VIIT, the reason for this

result could be driven by the aggregation of all the years in the data set. Also interesting is

that the measure for FDI is both consistently positive and significant at the highest levels for

all the different trade flow measures. Given that the literature has a mixed opinion of what

a priori expectations to have, it seems more prudent to focus on the 1 and 3-year results.

Turning to the variables of interest on Table 2.3 we see that both the variables across

all 5 models are positive and statistically significant. It is worth noting that the Comecon

indicator variable is weakly statistically significant in all but the VIIt model, while the ex-

USSR indicator variable is statistically significant at the highest level in all but the INT

Model. The results form the variables of interest echos the broad findings of Freidman et al.

(2020), where historical impact of borders is stronger and longer than that of trade unions.

21Tables of all the regressions utilizing the political freedom variable are available upon request to the

author.

2.5.2 Yearly Panel Results

In looking at the yearly results (Tables 2.4-2.8), the first thing that is clear is that there is

a substantial amount of “noise” present. Specifically, focusing on GDP, we find that only in

the models regressing HIIT (Table 2.7) and VIIT (Table 2.8) show GDP to be both always

positive, and statistically significant in all years. Distance does not stay negative for all

years in any model and is never consistently statistically significant in any model. In the

models not mentioned explicitly, the aforementioned coefficients osculate from their proper

signs with varying, often on the lower-end, levels of statistical significance.

The coefficients on contiguous borders and economic freedom show even more volatility.

In all but the model using IIT as the dependent variable (Table 2.5), contiguous border is

not consistently positive over all years. Further, contiguous borders is overall statistically

significant in only the IIT model, while to lesser degrees in the other yearly models. In the

model using VIIT (Table 2.8), contiguous borders starts off negative then moves positive to

then end negative. Across all models, the coefficient for economic freedom has no consis-

tent sign nor truly ever becomes consistently statistically significant. The most indicative

examples of this sign oscillation is the models using IIT (Table 2.5) and HIIT (Table 2.7).

Similar the other variables, GDP per capita also shows high volatility. On the regres-

sions for TT and IIT we see that DGDP P C is negative and has only a few years that are

statistically significant, while the coefficients on INT, HIIT, and VIIT starts negative and

oscillates to positive with very few years being statistically significant. Although IIT has

the right sign, the lack of both similar signs for HIIT/VIIT as well as the lack of statistical

significance across all models gives more weight to the argument that the 3-year intervals

are going to present a more stable story. Interestingly the coefficients on FDI show much

more stability. Across all of the models FDI is consistently positive, though it’s statistical

significance is present at mainly the 5% level in a minority of years in no clear pattern.

Lastly, turning to the variables of interest, it can be seen that USSR and Comecon have

not been immune to picking of the statistical noise by moving to the by year regression

specification. Turning first to the TT model (Table 2.4), USSR oscillates between positive

and negative with only four particular years being statistically significant. Comecon in the

TT model also begins positive then oscillates between negative and positive with few years

showing it being statistically significant.

Turning to the disaggregated total trade variables, we see that in the model looking

at IIT (Table 2.5) the USSR variable is always statistically significant and every year has

a positive coefficient. Conversely, the coefficient for Comecon starts positive then moves

negative with none of the years being statistically significant. Moving to the yearly model

of INT (Table 2.6) we find that both USSR and Comecon coefficients start off positive than

become negative with only a few years showing any level of statistical significance. Looking

at HIIT and VIIT models (Tables 2.7 and 2.8), the coefficients on the USSR variable is

positive and statistically significant across both models. Similarly, the Comecon variable is

also mainly positive but only exhibiting a few years of statistical significance.

Despite the large amount of variation in the yearly data, there is still some value to

be gleamed. Specifically, by looking at all the coefficients on USSR and Comecon we can

see an overall time trend emerge. Since looking at these coefficients on the tables can be

cumbersome, plots of both variables’ coefficients and confidence interval are provided in

Figures 2.1-2.5. In the TT (Figure 2.1), IIT (Figure 2.3) , and VIIT (Figure 2.5) show a

relatively smooth monotonically decreasing to zero trend for the USSR indicator variable.

Conversely, the models for INT and HIIT (Figures 2.2 and 2.4) demonstrate an oscillating

and inverted “v” pattern respectively. For the Comecon variable only the VIIT (Figure 2.5)

shows a monotonically decreasing to zero pattern, the remaining figures simply exhibit a lot

of noise.

2.5.3 3-Year Panel Results

Given the large amount of noise in the yearly regressions, the 3-year panel models (Tables

2.9-2.13) are employed to understand the underlying effects of historical trading institutions

with more precision. In all of the models GDP is positive a statistically significant in all the

models. Interestingly, the distance coefficient is only consistently negative in the TT (Table

2.9) and IIT (Table 2.10) models. In the remaining the coefficient for distance, it starts off

negative and in the last 3-year intervals becomes positive. Further, in no model is distance

statistically consistent across all the 3-year intervals.

Similar to distance, the coefficients for contiguous borders and economic freedom display

non-uniform behavior. Contiguous borders in all but the VIIT model (Table 2.13), exhibits

positive coefficients of varying levels of statistical significance. Economic freedom is much

more spurious, with only the model of INT (Table 2.11) exhibiting coefficients with the same

sign over all the 3-year sets.

Unlike the previous variables, DGDP P C and F DI both exhibit some general pattern.

First, differences in GDP is overall negative in all the models, with the exception of HIIT

in which it oscillates between negative and positive. Further all but HIIT exhibit varying

degrees of statistical significance. Given theory, one would expect that IIT and HIIT to

have a negative correlation to DGDP P C, but VIIT should be positively correlated. It is

unclear what might be driving this result, but the coefficients are not consistently statistically

significant making the result suspect. FDI, on the other hand, is consistently positive across

all the models, maintaining the same sign across all specifications. Further across all the

specifications it is generally statistically significant to varying degrees.

Lastly, this section concludes by looking at the variables of interest for this paper. Start-

ing with the USSR indicator variable across all models the coefficients are positive, save

for the TT (Table 2.9) and INT (Table 2.11). In the aforementioned exceptions, the USSR

coefficient becomes negative after starting positive. Looking to statistical significance, the

tables illustrate that overall USSR is strongly statistically significant. In the cases of IIT

(Table 2.10), HIIT (Table 2.12), and VIIT (Table 2.13) USSR is statically significant at the

highest levels, while in the other models it starts off strongly significant and tapers off to

not statistically different from zero by end end of the time period.

Comecon also exhibits an overall positive effect, with the exception of both TT (Table 2.9)

where it starts positive and turns negative and IIT (Table 2.10) where it oscillates positive

to negative. On the question of statistical significance there are three general patterns that

emerge. The first pattern is statistical significance which decays into non-significance by the

end of the period (models of TT and INT), no statistical significance (models IIT and HIIT),

and strong and consistent statistical significance (model VIIT).

Similar to the yearly models, figures containing the coefficients and the confidence in-

tervals for both of the indicator variables (Figures 2.6-2.10) can be found in the appendix.

In looking first at USSR, we find that TT (Figure 2.6) and VIIT (Figure 2.10) exhibit a

monotonically decreasing pattern. IIT (Figure 2.7) and HIIT (Figure 2.9) exhibit a some-

what increasing pattern, while INT (Figure 2.8) exhibits an oscillating convergence pattern.

Comecon also exhibits signs of an oscillating convergence pattern for the models of TT, IIT,

and INT(Figures 2.6-2.8). Interestingly, when decomposing IIT into it’s horizontal and ver-

tical components, it can be seen that both HIIT (Figure 2.9) and VIIT (Figure 2.10) seem

to be both on slower and much longer convergence paths, with VITT exhibiting more of a

monotonic convergence pattern.

2.6 Conclusion

In parsing through the results of this empirical work two broad important conclusions

can be made. First, this paper has built a much more robust and theoretically consistent

modeling specification than either previous IIT studies employing gravity and Yotov et al.

(2016). Reflecting this robust specification, is more nuanced and complex results. It can be

observed that variables controlling for historical trading institutions are generally positive

both when aggregating for all years as well as disaggreaging the data into 3-year chunks.

Further the standard empirical variables for IIT and it’s components were shown to be

broadly consistent with their a priori assumptions.

Given the robust and consistent results of this well-specified gravity trade model, an

argument is made for utilizing a similar specification for the handful of IIT papers which

employ a gravity trade model since their results will undoubtedly be biased. An important

caveat on the non-historical institutions variables must be given. This paper was not focused

on interrogating the determinants of IIT, HIIT, and VIIT. Instead it is much more interested

in understanding history’s impact on trade and as such this paper must be seen as an

illustration of how to properly specify IIT gravity models but not definitive argument. In

short this paper only included the variables that literature as deemed absolutely necessary.

Secondly, turning to the prime question of this paper “how long does the impact of

previous institutions last?” there emerges two distinct narratives, one about borders and one

about trade union’s historical impact on trade. Building off the general results of Freidman

et al. (2020), historical borders (as proxied by USSR) presents a much stronger and lasting

impact. TT demonstrates a general monotonic convergence path. When separating out the

different types of trade we see that the convergence in TT is mainly driven by INT and

thus the comparative advantages of different nations. Numerous papers have argued the

importance of disaggregating IIT, in this study it proves pivotal to see that the historical

border effect in VIIT is converging to statistical zero while the HIIT seems to be a much

longer convergence path. The latter observation of the two paths is more strongly illustrated

by the yearly regressions. The result of the VIIT shows that even in the context of IIT, states

will tend back to their comparative advantage, but it is the historical borders which impedes

this process. Further, these historical ties seem to strengthen the interest in importing, now

foreign, varieties of goods once restricted by national borders.

The story of trade unions is much less clear. When aggregating all the years the historical

impact of a trade union (proxied by Comecon) is positive, and even when put in 3-year

intervals Comecon is still positive and arguably converging to statistical zero in a monotonic

patter. Yet when disagregating TT the data argues that most of that story is driven by the

monotonic convergences of INT and possible oscillating convergence of IIT, again mirroring

the story of historical borders. These results make the case that the impact of historical trade

unions is much weaker than historical borders, something echoed by the sparse literature on

the subject. Investigating further into IIT shows that while VIIT is always positive and

slowly converging to statistical zero, HIIT is driving the result of the Comecon’s harmonic

effect on IIT by presenting a much more aggressive harmonic pattern. Again these results

both are weaker, but mirror the impacts seen by historical borders.

What the data seems to be arguing is that as countries of different stages of development,

and thus radically different factor endowments, see the strength of historical borders become

less important as they regress to their factor endowed motivated specializations. This story

seems to be true both for historical borders and trade unions, though is much stronger (both

in magnitude and statistical significance) in historical borders. Conversely the impact of a

historical border is much slower to dissipate among similarly endowed countries and thus

makes a case that historical ties compound the trade of goods of different varieties, which is

undoubtedly driven by past cultural connections. In the context of historical trade unions, it

is unclear if these effects are dissipating at all or going through cyclical waves of importance.

These results build off the previous work in historical trading institutions by showing that

historical border effects are much more binding than that of trade unions.

Lastly, and importantly this paper builds on the broad results of papers looking to

understand the length of the impact of previous trading institutions on current trade volume.

It is not only one of a few which look to metric the length of institutions but importantly

shows that these decaying effects are exhibited at all levels of trade, from total trade to

vertical intra-industry trade. Understanding how and why the impacts of historical trade

institutions decay is becoming increasingly paramount given that the first decade of the

21st century has resulted in a number of international institutions dissolving. Being able to

advise policy makers on the impact of such a decision helps economists to better articulate

the answer questions regarding the gains from trade, and more importantly the cost of

disengaging in the international community.

The Intersection of History and Trade; Essays in Trade and Trade Policy.

Students also viewed