Trade agreements in time of crisis : did the Eurasian Economic Union have a positive impact on its members?

MSc Economics

Track: Monetary Policy and Banking

Master Thesis

Trade Agreements in Time of Crisis:

Did the Eurasian Economic Union

have a Positive Impact on its Members?

Ainur Kushmukhan

July 15, 2017

Supervisor:

dhr. prof.

dr. F.J.G.M.Klaassen

Second Reader:

dr. W.E.Romp

Abstract

The financial crisis of 2014, that developed in Russia and spread to the post-Soviet space, created doubts about the advantages of close economic ties with Rus-sia, and in particular of The Eurasian Economic Union. This thesis investigates the effects of the membership of the union on bilateral exports, by using a dynamic panel data gravity model approach on a data of 110 country-pairs in the years of 1996-2016. The results show that the impact of the EAEU is not as small, if the underlying trends are controlled for by including the time trend fixed effects. The estimated effect on exports is thus 18.5%.

Statement of Originality

This document is written by Student Ainur Kushmukhan who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Contents

Abstract 1

Statement of Originality 2

1 Introduction 5

2 Literature Review 6

2.1 The Eurasian Economic Union . . . 6 2.2 The Gravity Model . . . 8

3 Data and Method 9

3.1 Model . . . 9 3.2 Data . . . 11 3.3 Estimation . . . 14 4 Results 16 4.1 Estimation Results . . . 16 4.2 Sensitivity Analysis . . . 21 5 Conclusion 23 References 28 Appendix A 29

List of Tables

1 The EAEU countries descriptive statistics from 2016 and the EACU acces-sion dates . . . 14 2 Non-EAEU countries descriptive statistics from 2016 and dates of imposing

sanctions on trade . . . 14 3 LSDV and GMM system estimation results . . . 17 4 Sensitivity of EAEU estimate to generalizations of model (1) . . . 22

List of Figures

1 Residuals from the LSDV estimation of model (1) averaged across 10 EAEU country-pairs (solid lines) and 100 other country-pairs (dashed lines) . . . . 19 2 Residuals from the LSDV estimation of model (1) under the restriction

β1 = 0 averaged across 10 EAEU country-pairs (solid lines) and 100 other

1

Introduction

In 2014 Russia experienced a currency crisis manifested by a collapse of Russian ruble that led the country into a current deep economic recession (Hansl et al., 2015) and that spread to several neighbor states (Stepanyan et al., 2015). The crisis developed as a result of Russian-Ukraine geopolitical conflict and subsequent international economic sanctions against Russia, as well as the fall in price of oil, the nations biggest export (International Monetary Fund, 2015). Russian ruble thus lost 41% of its value over the course of 2014 only.

Contagion of the crisis had a heavily deteriorating effect on the post-Soviet economies, since Russia is closely tied with neighbor states through investments, trade and remit-tances. Belarus, Azerbaijan, Kazakhstan and Kyrgyzstan, among many who depend on Russia as a major trading partner, had to devalue their currencies in order to increase the competitiveness of their economies in a situation of sharp depreciation of Russian ruble. Migrant labor remittances from Russia that constitute substantial shares of GDPs of Tajikistan, Kyrgyzstan, Uzbekistan and Armenia decreased dramatically and hit hard the economies of the countries as well as livelihood of ordinary citizens.

The strong economic interconnectedness of the countries in the post-Soviet space is conditioned by historical ties and continues to be supported by a number of regional or-ganizations, including the Commonwealth of Independent States (CIS) and the Eurasian Economic Union (EAEU). The economic contagion from Russia has thus raised questions about the sustainability of the EAEU in particular. Established in 2015 on the basis of the Eurasian Customs Union (EACU, 2010) and Single Economic Space (SES, 2012), the EAEU provides for regional economic integration and free movement of goods, services, capital and labor between the member states, namely Russia, Kazakhstan, Belarus, Kyr-gyzstan and Armenia. Even though the cooperative agreements signed by the member states are focused on increasing the trade volumes and strengthening the resilience to external shocks, they also imply political and economic risks, such that by deepening their mutual integration the countries become more prone to the up- and downturns of the economies of the other members.

The crisis of 2014 has thus embodied the potential problems of the region, and heavily deteriorated trade of the member states both within and outside of the EAEU. The question however rises whether the trading agreements signed within the union have in fact amplified or weakened the negative effect of the decreased commodity prices and political tensions. This thesis hence aims at investigating the role of the Eurasian integration in the trade flows of the EAEU member states.

Since 2010 when the EACU was established, the organization has already received many critical reviews. Particularly, several researchers consider the EAEU as a way for Russia to regain geopolitical influence in the region, despite of the purely economic goals

on paper (Aslund, 2013; Popescu, 2014; Dreyer & Popescu, 2014; Tarr, 2016; Strzelecki, 2016). Moreover, some critics emphasize economic risks of the common external tariffs imposed on the member states, as they almost doubled for Armenia, Kazakhstan and Kyrgyzstan, and may lead to trade diversion losses for these countries (Aslund, 2013; Tarr, 2016; International Crisis Group, 2016).

The existing researches on the role of the EAEU are largely descriptive assessments of political and regulatory issues of the EAEU structure and its economic prospects (Block-mans, Kostanyan & Vorobiov, 2012; Popescu, 2014; Dreyer & Popescu, 2014; Bogulavska, 2015; Kansikas, 2015; Strzelecki, 2016). The aspects affecting the trade flows and their dynamics have received significantly less attention (Vinokurov et al., 2015; Mazhikeyev & Edwards, 2015; Tarr, 2016). In the light of the recent geopolitical crisis in Ukraine, subsequent sanctions on Russia and the drop in oil prices, it is of particular interest to fill in this gap. The present thesis therefore investigates the trade flows between the EAEU members and the region’s largest trading partners that imposed sanctions on Russia (Ger-many, Italy, Japan, Netherlands, Ukraine and the USA) over the period of 1996-2016. The analysis implements the gravity model of trade (discussed extensively in section 2.2), and in addition to studying the feasibility of the EAEU objectives, it also sheds light on the trade determinants for this set of countries and on the impact of various fixed effects included in the model.

The set up of the paper is as follows. The next section discusses the literature rel-evant to the EAEU issues as well as the studies of the trade effects using the gravity model. Section 3 describes the data and methodology used. In section 4 the results of the estimation are presented and discussed. Section 5 concludes.

2

Literature Review

This chapter starts by surveying the literature on the effects the EAEU has on its mem-bers. Subsequently, section 2.2 discusses the implementation of the gravity model in empirical studies of trade.

2.1

The Eurasian Economic Union

Several researchers examined the effects of the Eurasian Economic Union on the economies of its member states. The more in-depth studies however concentrate on the problems present in the EAEU that may affect trade, rather than the membership itself. In this way, Tarr (2016) argues that in order to achieve the objective of deep integration, the EAEU has to improve trade facilitation and reduce the substantial internal non-tariff barriers. The latter are substantially represented by strict and distinct sanitary conditions of the member states, and sanctions imposed on Russia. The sanitary conditions on food and

agricultural products are thus often used to ban imports of specific goods, partially re-ducing the trade flows. Whereas the non-unified application of counter-sanctions (Belarus and Kazakhstan refused to impose counter-sanctions on the Western countries along with Russia) led to ban on import of some products from Belarus, as they were perceived to be Western exports. According to Tarr, so far the Eurasian Economic Union has failed to facilitate the trade and harmonize regulations, imposing more costs than benefits. Tarr’s findings hence motivate the negative expectations regarding the influence of the EAEU on the trade flows.

Similarly, Vinokurov et al. (2015) present an extensive study of the effects of reducing non-tariff barriers (NTB) in the EAEU. The authors implement the gravity model to evaluate the trade costs of the non-tariff barriers, and their results imply that there is indeed a negative relationship between the NTBs and the country pairs volume of exports. Furthermore, effect of the decrease in non-tariff barriers is higher for Russian and Kazakh exporters than for Belarusian exporters. The paper also concludes that NTB most significantly impacts the exports in chemical, rubber and plastic industry, textile and sewing industry, as well as food production. Interestingly for this thesis, the authors suggest that “the [...] slowing of trade flows (of capital goods, in particular) [in 2012] was largely due to the influence of NTBs rather than a slowdown of economic growth in Russia.” Their gravity model results in theoretically expected coefficients, in particular, the volume of exports is negatively correlated with the distance between the country-pair, and positively with the existence of common border and the GDPs.

In a recent publication by Khitakhunov, Mukhamediyev & Pomfret (2017) interna-tional sanctions during the Ukranian crisis are discussed with respect to their effect on the EAEU. The authors point out that even though the implementation of sanctions in general harms both parties, the effect on third countries is ambiguous; so while they may potentially suffer from the decrease in trade flows, third countries may equally well capture the gains of trade diverted from the banned countries. In case of the EAEU, the member-states refused to join Russia in its sanctions against Western countries, and this in turn created a possibility for them to pass banned agricultural goods from the European countries to Russia, while Russia can potentially export its products through the EAEU partners, in this way avoiding the costs of sanctions. In fact, Western food embargo led to an increase in smuggling activity from the EU to Russia through Belarus borders, which in turn caused Russia to reinforce the trade restrictions with Belarus. (Khitakhunov et al, 2017) The authors thus indicate that the sanctions are indirectly imposed on the whole EAEU, while the EACU provides with export opportunities for both the sanctions-sending states and the targeted nations. This thesis therefore tries to empirically determine which effect is predominant.

2.2

The Gravity Model

The gravity model has been used extensively over the past 50 years as one of the most robust methods in international trade analysis. It was first applied by Jan Tinbergen in 1962 who, in analogy with Newtons theory of gravitation, explained the trade flows be-tween two countries as being proportional to their sizes, measured by GDP, and inversely proportional to the distance between them. Since then the model has been abundantly extended and augmented by variety of factors that may affect trade, including economic integration.

Apart from empirical studies, several scholars also presented theoretical investigations of the gravity model and interpreted the causal relationships between the variables. (Lin-nemann, 1966; Bergstrand, 1985; Deardorff, 1998; Krugman & Obstfeld, 2005) Today it is thus widely accepted that gravity model has a solid theoretical justification.

The model was often implemented with the use of cross-sectional data (Aitken, 1973; Bergstrand, 1985; Frankel, Stein & Wei, 1995), however panel data prevails in more recent research papers, as it results in more accurate estimates (De Grauwe & Skudelny, 2000; Wall, 2000; Glick & Rose, 2001) and allows for the use of country-pair specific effects. (Bun & Klaassen, 2002) Furthermore, many studies agree that dynamic model is preferred to the standard static specification, because trade is a dynamic process and variables affecting it do not always manifest in a single period. (Bun & Klaassen, 2002; Martinez-Zarzoso & Horsewood, 2005; Magee, 2007; Olivero & Yotov, 2012)

Bun and Klaassen (2002) in particular argue that dynamics of the trade flows are strongly significant, since they account for the impact of distribution networks histori-cally formed between the trading countries, as well as for consumers loyalty for certain historically traded commodities. Taking into account the long history of economic part-nership of the members of the EAEU, trade dynamics are considered to be an important extension to the gravity model in this thesis. In their modeling the authors also test the fact that income affects trade with a lag, as suggested by Goldstein and Kahn (1985). They however conclude that income affects trade rather quickly, and tends to 0 in a span of two years. In this paper, due to practicality and the diminishing effect of the income lag, it will not be specified in the baseline model.

According to Magee (2007) many weaknesses of the gravity model estimation may be addressed with the use of fixed effects. In particular, a natural trading partner hypothesis proposed by several authors (Krugman, 1991; Haveman & Hummels, 1998; Soloaga & Winters, 2001) stating that high levels of trade may be due to historical and cultural factors rather than trading agreements, can be controlled for by including country-pair fixed effects. The author also uses importer-year and exporter-year fixed effects to capture aggregate shocks to the individual countries. Overall, implementation of fixed effects eliminates the sensitivity of gravity model estimates to the choice of variables. This

thesis thus aims to correct for omitted variable bias by including the fixed effects.

Another problem that may arise in the gravity model analysis and that according to Bun and Klaassen (2007) may be controlled with the use of fixed effects is trend misspecification in trade flows over time. The authors argue that “trending behavior of trade flows may [...] be affected by variables not included in the specification and trends may vary across country-pairs, both due to country-specific and country-pair specific factors”. To avoid the omitted trending variables bias the study implements country-pair specific time trend fixed effects. The solution will be adopted in this thesis to prevent the possible trend misspecification. The detailed discussion of the fixed effects is deferred to the next section.

3

Data and Method

This chapter outlines the methodological framework used to investigate the research ques-tion of the present thesis and explains the choice of the data.

3.1

Model

The variable to be explained is EXP ORTijt, defined as the logarithm of real non-oil

exports from country i to country j at time t. Oil exports were subtracted from the model in order to isolate the trade flows from the effect of the oil price fluctuations. At its basic the gravity model explains the trade flows through the distance between the countries and their economic sizes. In this paper the economic size is proxied by the logarithm of a country’s real GDP. GDPit and GDPjt thus measure the logarithms

of the real GDPs of countries i and j, respectively, in period t. The distance between the countries is a time-invariant variable and in the present model is absorbed by the country-pair fixed effects.

The gravity model can then be augmented with the explanatory variables of interest. This thesis in particular includes the first lag of the dependent variable, the real effective exchange rates, and dummy variables for the EAEU membership, war between Russia and Ukraine and sanctions on trade. The former dummy (EAEUijt) is of particular interest

to this paper, as it accounts for an impact the EAEU might have on the trade. It thus takes the value 1, if both countries i and j are members of the EAEU at time t, or else it is 0. As discussed in section 2.2, lagged exports are important to control for, due to the historically established trade relationships in the region. To allow for the dynamics the model includes the first lag of the exports, EXP ORTijt−1. In order to correct for volatility

of the exchange rates during the studied period the variables REERit and REERjt are

added. They represent real effective exchange rates in US dollars of countries i and j, respectively, at time t. In addition, the countries’ real GDPs per capita, denoted by

GDP CAPitand GDP CAPjt, were used to account for the population sizes of the countries

i and j at time t respectively, while avoiding correlation with the real GDPs values, as well as for the income levels of the countries. Given the fact that the studied period covers a war between Russia and Ukraine, it is assumed that it heavily deteriorated the trade between the two countries, hence a dummy variable warijt that takes the value 1 in the

years of 2014-2016 for Russia and Ukraine, and 0 otherwise, is used. As emphasized by Khitakhunov et al. (2017), bilateral sanctions enacted by Russia on one side, and the EU, the US and Japan on another, had a negative impact on the trade between the countries and economies of the EAEU members. The sanctions are therefore accounted for in the model, to separate their expected negative effect from the rest of the coefficients, and the EAEUijt in particular. The dummy variable Sanctijt hence takes the value 1, if the

exports, wholly or partially, from country i to country j fell under sanctions in a year t, and 0 otherwise.

There are indeed many more trade determinants than those already included in the present model, and to account for such, fixed effects are added. Fixed effects are preferred to random effects, as it is expected that potential omitted variables are correlated with the regressors in the model (such as the EAEU membership and GDPs). To control for all country-pair invariant explanatory variables (for instance, state of the world economy, business or financial cycles) φt is used, and αij corrects for all possible time-invariant

trade determinants (such as historically established trading networks, common language, common border and distance). As suggested by Bun and Klaassen (2007), country-pair specific trend variables are controlled for by including τij· t – country-pair combined with

trend fixed effects. These are of particular importance, because trends in the trade flows can lead to substantial bias if they accumulate over time without being accounted for. For instance, trade liberalization between the countries is a gradual process, individual to each country pair; the non-tariff barriers discussed in Vinokurov et al. (2015) and Tarr (2016) in this way are still at different stages of elimination within the EAEU. The process of integration of the countries within the economic union and the post-Soviet space in general may thus also present an ambiguous trend, because the recent geopolitical crisis as well as a contagion of financial crisis from Russia could have had a deteriorating effect on the economic and political relationships between the states. It is however rather beyond the scope of this research to construct proxies for all possible trending regressors that affect the trade within the country-pairs, so in order to account for a time trend this model, in accordance with Bun and Klaassen (2007), includes time as an explanatory variable, allowing it to have heterogeneous impact across country-pairs.

The model thus takes the following form

EXP ORTijt= αij + φt+ τij· t + β1EXP ORTijt−1+ β2GDPit+ β3GDPjt

+ β4GDP CAPit+ β5GDP CAPjt+ β6REERit+ β7REERjt

+ β8EAEUijt+ β9W arijt+ β10Sanctijt+ εijt,

(1)

where εijt is allowed for heteroskedasticity across country-pairs and time. Clustering of

the standard errors is used to account for the possible correlation within the country-pairs. Bun and Klaassen (2002) point out that εijt is likely to correlate with the regressors, and

in particular the GDP variables, because exports and imports form a part of countries’ GDPs. This endogeneity is, however, usually ignored in literature, because the trade variable makes up only a small part of GDP. Taking into account that the dependent variable in the present model concerns even a smaller portion of total trade (non-oil exports), this thesis follows the same assumptions.

3.2

Data

The paper uses annual bilateral exports data from 11 countries over the period 1996-2016, thus forming a panel data set of N = 110 country-pairs and T = 20 years, which gives a potential total of 2310 observations. The data on exports was obtained from the United Nations Commodity Trade Statistics Database, by excluding oils and distillation products that may be affected by the price volatility (a list of the eliminated commodities can be found in appendix A).1 The UN Comtrade database lacks exports observations for Russia and Ukraine for the year of 2016, these were hence obtained from the Eurasian Economic Commission Statistics database. Bilateral trade flows were also not available for Belarus, Kyrgyzstan, Ukraine and Armenia for certain years before the establishment of the EAEU, some of these observations were retrieved from the national customs statistics, others were discarded. The model thus uses 2305 real exports observations. The most extensive records of the bilateral exports are on a value basis in current US dollars. In order to adjust them to the real values, the data was first converted to local currency to remove the exchange rate effect, where yearly average exchange rates were retrieved from the World Bank DataBank and the OECD Statistics. Then the values of non-oil exports were deflated using GDP deflator, obtained from the IMF World Economic Outlook (WEO) database, to remove the effect of prices. Consequently the value of export of each country in the base year of 2011 was converted to international dollars using the implied PPP conversion rate (from the IMF WEO database). These data points were then extended backwards and forwards using the growth rate of the deflated export values, in this way

1_{The values of commodities subtracted from the overall exports are classified under Article 27}

(min-eral fuels, min(min-eral oils and products of their distillation; bituminous substances; min(min-eral waxes) of the Harmonized System classification of goods (HS code).

the series is presented in common constant 2011 international dollars.2

As has been mentioned, establishment of the Eurasian Economic Union was gradual, and the regional trading block, in particular, was formed several years before the Treaty on the EAEU was signed. Since the aim of this thesis is to assess the trading effects of the union, the dates of accession to the Eurasian Customs Union rather than the EAEU itself were considered as the determinants for the dummy variable EAEUijt. So even though

the economic union came into existence in 2015, the EACU eliminated intra-bloc tariffs for Belarus, Kazakhstan and Russia already in 2010.

The data on real effective exchange rates (REERit and REERjt) was obtained from

the Bruegel Working Paper (2012). Their database presents the consumer price index (CPI) - based annual real effective exchange rates that measure the real value of a coun-try’s currency against the basket of currencies of major trading partners of the country. An increase in REER implies an appreciation of the currency (increase in value of ex-change rate (ER)), and vice versa. It is hence expected that the variable REERit will

demonstrate negative elasticity with respect to exports, since a decrease in REER, imply-ing real depreciation of a currency, makes exports from this country more attractive to the foreign market. On the other hand, REERjt is expected to have a positive coefficient,

since an appreciation of the currency of the importing country makes foreign goods and services more affordable to the domestic consumers, and potentially leads to an increase of imports, which is an increase of exports from country i to country j.

The data on the GDPs were obtained from the IMF WEO Database. The data is presented in constant 2011 purchasing power parity (PPP) international dollars, which was achieved by extending the 2011 GDP in current PPP international dollars using the growth rates of GDP in constant local currency unit (LCU). The PPP here brings the data in a common numeraire, which improves comparability, and helps to control for the influence of the exchange rates in each country. While constant 2011 international dollars account for the general price inflation. It is expected that both GDPit and GDPjt will

show significant positive coefficients. First, the GDP of destination country j captures the income effect, so as the economy’s income increases, the demand for goods and services in this country increases as well, which subsequently leads to higher imports (exports from i to j). Second, according to Anderson and van Wincoop (2003), under the assumption of competitive trade of goods differentiated by the country of origin (Armington assumption), countries with higher GDP will export more intensively, because their relative export’s prices decrease under market clearing condition. At the same time, if prices do not adjust, under the assumption of frictionless trade and factor price equalization, as in Helpman and Krugman (1985), the countries with higher GDPs will adjust by increasing the varieties of their products, which then leads to increase in exports. (Benedictis & Taglioni, 2011)

2_{An international dollar has the same purchasing power over GDP as the U.S. dollar has in the United}

Therefore, it is assumed that the GDP of exporting country i positively affects the trade. According to theoretical specifications by Bergstrand (1989) and Deardorff (1998), GDP and per capita GDP should both be included in the gravity model. While GDPs are standard to the specification, they suggest that high per capita income countries have high capital labor ratios and tend to produce and consume more of capital-intensive goods, therefore an increase in exporter’s per capita GDP will tend to increase the exports, if the trade flow consists of capital-intensive goods, and decrease it if the goods are labor-intensive; whereas, an increase in importer’s per capita GDP will have a positive effect if the import consists of luxury goods, and a negative effect if the trade flow is mostly necessity goods. This thesis thus uses data on GDP per capita in current PPP international dollars from the IMF World Economic Outlook Database. Equivalently to the data on GDP, in order to convert per capita GDP to constant international dollars, 2011 was chosen as a base year, and the growth rates of GDP per capita in LCU was applied forwards and backwards to the GDP per capita in current PPP international dollars of each country in the year of 2011 to extend the series.

Tables 1 and 2 show summaries of some economic factors in the year of 2016 as well as important dates, used in the model, for the two country groups. The countries in Table 1 are all members of the EAEU, and the dates of their accession to the Eurasian Customs Union are presented in the last column. It should be noted that the years of the accession are used in the model as a dummy variable EAEUijt, and as seen from Table 1

the dummy takes the value 1 rather at the end of the sample period. In accordance with Bun and Klaassen (2007), the estimated EAEUijt may thus exhibit an upward bias, as it

potentially picks up unobservable integration trends accumulated throughout the sample. This thesis therefore checks for the presence of trends in residuals in Section 4. The EAEU countries are also observably smaller developing economies in comparison to the second group of developed countries, with an exception of Ukraine. Such heterogeneity may potentially cause diverging trends in the residuals from the model, and hence should be noted.

Table 2 shows non-EAEU countries that trade the most with Russia and that intro-duced sanctions on the latter as a response to the geopolitical crisis in Ukraine. The dates of sanctions on trade also fall at the end of the sample period, and as the years of sanctions enter the model as a dummy Sanctijt, it is expected that they will have a

Table 1: The EAEU countries descriptive statistics from 2016 and the EACU accession dates

2016 Countries GDP PPP (constant 2011 international $ bln) GDP per capita PPP (constant 2011 international $) REER EACU accession date Russia 3522.37 24556.42 86.85 1 January 2010 Armenia 23.90 7991.76 105.42 2 January 2015 Belarus 158.48 16685.96 72.04 1 January 2010 Kazakhstan 418.34 23309.64 74.76 1 January 2010 Kyrgyzstan 19.93 3264.16 111.23 12 August 2015

Table 2: Non-EAEU countries descriptive statistics from 2016 and dates of imposing sanctions on trade

2016 Countries GDP PPP (constant 2011 international $ bln) GDP per capita PPP (constant 2011 international $)

REER Sanctions on trade imposed

USA 17213.80 53244.32 113.26 28 March 2014

Germany 3689.78 44599.40 90.98 31 July 2014

Italy 2071.41 34144.77 93.76 31 July 2014

Japan 4855.50 38262.11 100.57 24 September 2014

The Netherlands 805.91 47323.12 93.11 31 July 2014

Ukraine 327.22 7698.92 66.60 29 March 2014

3.3

Estimation

A dynamic panel data approach is used to analyze the trade and its determinants in this thesis. A common practice in the previous studies is to use fixed effects in estimating the gravity model. Cheng and Wall (2005), in particular, used country-pair specific and time fixed effects in explaining the real exports. This subsequently popularized FE model is in fact the special case of (1) where τij = 0 for all country pairs. As stated in Bun

and Klaassen (2002a), if there are indeed no omitted trending variables in reality, then the estimated coefficient of the EAEU, β8 in (1), will be on average equal to its value

under the τij = 0 restriction, but with a larger standard error. It is hence concluded that

unrestricted τij does not produce a downward bias.

Following a standard approach in gravity model literature, this thesis implicitly as-sumes cointegration of the variables and ignores potential non-stationarity in model (1).

It is likely from an economic perspective that the EXP ORTijt, GDPs and per-capita

GDPs all exhibit non-stationarity, but the present model uses a rather limited time pe-riod sample (T =20), and so the estimator is implicitly approximated by the distribution for infinite N , and finite time dimension T .

The FE models of panel data are often estimated using the least-squares dummy variable (LSDV) approach, also known as the Within or FE estimator. However, while for static models with non-stochastic Xijt LSDV is strictly consistent and unbiased, the

introduction of dynamics raises econometric problems. The bias of the LSDV estimator in dynamics models is generally referred to as Nickell’s bias (1981). Nickell (1981) shows that with lagged dependent variables as regressors, strict exogeneity of the explanatory variables no longer holds. The within-transformation and first-differencing hence lead to correlation between the regressors and error term, producing biased and inconsistent results, and the bias only reduces if T tends to infinity, N on the other hand does not influence it. The general agreement is that the LSDV estimation performs well with time-dimension of 30 − 40 (Judson & Owen, 1998; Bun & Klaassen, 2002b), since in the present thesis T = 20, which is fairly large, the bias may be negligible, and the results of the LSDV estimation are of interest.

Another complication arises from the within transformation required to wipe out τij· t

in model (1), since subtracting only country-pair specific means over time from each variable when transforming αij will not affect τij · t. Bun and Klaassen (2002a)

sug-gest “projecting all variables on the null-space of the matrix of dummy/time variables corresponding to all [αij], τij · t and [φt]”, which in fact implies the within

transforma-tion according to Wansbeek and Kapteyn (1989). In this way all fixed effects will be transformed in the present thesis as well.

Several alternatives to LSDV estimator exist that deal with the inconsistency and bias of the least-squares in case of a small T . Among such are instrumental variable estimations, developed by Anderson and Hsiao (1981), where lagged values of the levels or the differences of the dependent variable are used as instruments. Arellano and Bond (1991) further argue that there are more instruments available, and derive a Generalised Method of Moments (GMM) estimator (referred to as difference-GMM) that uses first differences to eliminate the fixed effects from the model, solving the endogeneity problem, and subsequently, uses the lagged dependent variables of the transformed model as instru-ments. This estimator is then also extended by Arellano and Bover (1995), and Blundell and Bond (1998) to the system-GMM that applies additional moment restrictions, and uses lagged changes in the dependent variable as instruments for the level of the lagged dependent variable. Both difference- and system- GMM estimators are widely applied in the gravity model analysis as they are consistent for N → ∞ and finite T samples; the system-GMM is however shown to outperform the difference-GMM (Behr, 2003; Soto, 2009), and therefore will be used in the present analysis. The problem that may arise

with estimation of model (1) is the weakness of the export level as an instrument in ex-plaining the export growth. It is however rather difficult to conclude at this point whether LSDV or system-GMM poses a more significant obstacle, and hence both estimators will be implemented and discussed in the empirical analysis.

4

Results

This section presents and discusses the results of estimation of model (1) with regards to the research question. It also assesses the impact of country-pair specific time trend fixed effects added to the model. Section 4.2 is dedicated to sensitivity analysis, it examines the variations of the baseline model (1) and verifies the robustness of the results.

4.1

Estimation Results

Table 3 shows the results of estimating model (1) with both LSDV and GMM system estimators. The columns titled as “No Trend” show the estimations under the restriction τij = 0, so without a country-pair specific time trend included in the model, columns

headed by “Trend”, on the other hand, show the outputs for the unrestricted model. Static model results shown in columns (1) and (2) are obtained by the LSDV estimation under the restriction β1 = 0. As has been discussed in the previous section, robustness of

the results of two types of estimations, system-GMM and LSDV, is ambiguous, since both methods have certain potential bias that may emerge in the present analysis. The output from the system-GMM estimator shown in column 5, however, immediately raises a ques-tion about the robustness of these results, as it assigns a negative coefficient to the lag of exports. Even though this result is not significant at 5% level (the level used throughout the thesis), it is highly expected that EXP ORTijt−1 will show a positive relationship with

the current values of exports: first of all from an economic intuition perspective dynamics appear to be important for the trade, due to the established trade relations and integra-tion (for theoretical foundaintegra-tions for dynamic gravity model, see Olivero & Yotov, 2012), and second, several empirical studies show that the lagged values are indeed positively correlated with the trade (Bun & Klaassen, 2002b; Martinez-Zarzoso & Horsewood, 2005; Magee, 2007). The system-GMM therefore doesn’t seem to produce very robust results for this data sample, large standard errors also support this. As it has been mentioned, using level of export as an instrument for the export growth, as well as level of GDP -for the GDP growth, may cause bias, and it appears that this is indeed the case here. The LSDV estimator on the other hand produces significant positive coefficient for the lagged value of exports in the model estimated without the trends, and a positive, but not significant result, in the estimation of the model with trend specification. In the light of this, the problem of finite T in the LSDV estimator is minor in comparison to the bias

of the system-GMM, and hence the thesis will further rely on the results of the LSDV estimation.

Table 3: LSDV and GMM system estimation results

Static LSDV GMM system

(1) (2) (3) (4) (5) (6)

No Trend Trend No Trend Trend No Trend Trend

EXP ORTijt−1 β1 - - 0.39∗ 0.21 -0.17 -0.13

(0.13) (0.14) (0.74) (0.58) EAEUijt β2 -0.24 0.26 -0.11 0.17 -0.23 0.47 (0.17) (0.15) (0.12) (0.13) (2.87) (11.26) GDPit β3 -1.60 1.47 -0.62 1.61 2.23 7.07 (0.98) (1.72) (0.69) (1.36) (56.14) (361.83) GDPjt β4 0.66 0.81 0.34 0.77 1.04 5.26 (0.66) (0.99) (0.40) (0.84) (19.45) (59.29) GDP CAPit β5 1.83∗ -0.73 0.76 -1.08 -1.56 -4.22 (0.92) (1.58) (0.65) (0.89) (43.14) (382.76) GDP CAPjt β6 0.63 1.12 0.50 0.88 1.38 -1.03 (0.68) (0.85) (0.45) (0.81) (28.49) (10.01) REERit β7 -0.49∗ -0.22 -0.48∗ -0.34 -0.41 -0.15 (0.22) (0.24) (0.16) (0.21) (4.25) (7.99) REERjt β8 0.50∗ 0.61∗ 0.23 0.43 0.01 0.07 (0.19) (0.27) (0.14) (0.27) (3.36) (5.65) W arijt β9 -0.49 -0.17 -0.41 -0.23 -0.19 -69.99 (0.27) (0.33) (0.22) (0.35) (5.24) (166.46) Sanctijt β10 -0.28∗ 0.24 -0.08 0.24 0.04 6.75 (0.10) (0.16) (0.08) (0.14) (3.47) (10.44) Observations 2305 2305 2194 2194 2194 2194 Adjusted R2 _{0.978 0.973 0.978 0.980 -}

-Clustered standard errors in parentheses

The variable of specific interest for this thesis is EAEUijt as it examines the impact of

the membership in the economic union on the trade flows within the countries. However, all six columns of estimations show insignificant at a 5% confidence level results for this coefficient. Even though the output is inconclusive, one may still observe a rather inter-esting pattern in the coefficients of the EAEUijt: the estimated effects are −0.24, 0.26,

−0.11, 0.17, −0.23 and 0.47, which stand for the elasticities of the trade with respect to the membership in the EAEU of [exp(β2) − 1 =] −21.3%, 29.7%, −10.4%, 18.5%, −20.5%

and 59.99%, respectively. Indeed, addition of the time trend increases the observed im-pact of the EAEUijt, and even reveals a positive effect it may have on the bilateral trade

flows between the member states when the country-pair specific time trend fixed effects are controlled for. The result may also point at the fact that in contrast to Bun and Klaassen (2007) unobservable time trends in the present data sample have a downward pressure on the EAEU dummy. If this is true, it should be the case that the longer time period sample results in a smaller estimate of the EAEUijt, since the underlying trends

accumulate more. Indeed, as the time sample shrinks, the coefficients for the EAEU be-come -0.036, 0.087, and 0.231, respectively. While it is difficult to justify this from an economic point of view, as one would expect an increasing trade integration throughout the time, the result may suggest that the EAEU is not as detrimental to the trade as could be concluded otherwise. Yet, to assess the relevance of the time trend fixed effects in model (1), the residuals of the estimated models, both under the restriction τijt = 0

and with the time trends, are studied. Noteworthy, the restricted model still accounts for some omitted time trending variables by including φt, but it only considers the trends

common to all country-pairs, so τij · t controls for the differences in trends between the

country pairs. Furthermore, the omitted country-pair specific time trends can be said to cause the downward bias of the EAEU effect only if the trends are downwards for the members of the economic union relative to the non-members, hence to check whether the EAEUijtcaptures the underlying downward trends the model is also re-estimated without

-.4

-.2

0

.2

1996 2000 2004 2008 2012 2016

(a) no EAEUijt, no trend

-.4

-.2

0

.2

1996 2000 2004 2008 2012 2016

(b) EAEUijt, no trend

-.4

-.2

0

.2

1996 2000 2004 2008 2012 2016

(c) no EAEUijt, trend

-.4

-.2

0

.2

1996 2000 2004 2008 2012 2016

(d) EAEUijt, trend

Figure 1: Residuals from the LSDV estimation of model (1) averaged across 10 EAEU country-pairs (solid lines) and 100 other country-pairs (dashed lines)

Subfigure (a) of Figure 1 thus shows the residuals of the model without the EAEU dummy and under τijt = 0, where the solid line refers to the residuals averaged across the

10 pairs of only EAEU members, and the dashed line shows 100 other country-pairs. It appears that the residuals of both groups oscillate around zero, while there are few troughs and peaks plotted by the solid line, general pattern doesn’t seem to exhibit any specific trend. However, when the EAEUijt is included in the model (see Subfigure

(b)) one can see that at the end of the sample period, specifically in the last two years, the residuals exhibit less negative trend than in Subfigure (a). Yet, it should be noted that using a dynamic model in analyzing the residuals for underlying trends may be misleading, as a lag of the dependent variable may pick up the trends, making the figures inconclusive. In order to verify the conclusions regarding the omitted trending variables, the equivalent plots for the static model (under the restriction β1 = 0) are presented in Figure 2. The

two subfigures (a) and (b) of Figure 2 show that the inclusion of the EAEUijt indeed

explains the downward trends at the end of the period studied, and hence the EAEU effect is biased downwards.

Addition of the country-pair time trends to the model (1) is therefore justified, and Subfigures (c) and (d) of both Figures 1 and 2 plot the residuals of the estimation with the inclusion of trends. Noteworthy, Subfigures (c) show that the residuals of the EAEU country-pairs, when the trends are accounted for and when there is no EAEU dummy,

actually exceed the residuals of the other country-pairs at the the end of the sample period. Seeing that, the model without the trends indeed seems to underestimate the EAEU effect on export. Residuals of the model (1) without the restriction τijt = 0 are

thus shown in Subfigure (d). One may observe that the residuals series of the EAEU country-pairs converge with the residuals of other 100 country-pairs at the beginning and at the end of the sample period (see Figure 2). Particularly in the last year of the sample, the solid line that exhibited sharp negative trend in Subfigures (a) and (b), in the last subfigure increases distinctly (see Figure 2) and seems to converge to zero together with the dashed line.

(a) no EAEUijt, no trend (b) EAEUijt, no trend

(c) no EAEUijt, trend (d) EAEUijt, trend

Figure 2: Residuals from the LSDV estimation of model (1) under the restriction β1 = 0

averaged across 10 EAEU country-pairs (solid lines) and 100 other country-pairs (dashed lines)

Such results are in line with the findings of Bun and Klaassen (2007), and confirm rel-evancy of inclusion of the time trend fixed effects. The EAEUijt, as a result of accounting

for the trends, is no longer misused and does not capture the downward trends. It can be thus concluded that the effect of the EAEU on trade was biased downwards, and is not in fact that negative, as may be inferred otherwise: it changes from −10.5% to 17.4%, when one controls for the trending variables. In other words, a member of the EAEU has on average 17.4% higher non-oil exports to the countries in the sample, than a non-EAEU member. Despite these specific values of elasticities are not significant under 5% level, this thesis emphasizes the need of accounting for time trends in the further research of

the effects the EAEU has on trade.

The inclusion of the trend fixed effects, as shown in Figure 1, corrects for the down-ward trend, and consequently produces more robust results. The differences between the coefficients presented in the columns (3) and (4) are hence due to the added τijt· t.

Apart from the EAEUijt coefficient, the results for GDPit and GDP CAPit also change

their signs when the trends are included in the model. The output for both is however insignificant, and with the trend specification the standard error in fact increases even more. One should also note that the signs of these two change interrelatedly, so that estimated GDPit is negative in the column (1), and becomes positive in column (2), and

the reverse happens with the estimated coefficient of GDP CAPit. The change in the sign

of GDPit actually stresses the relevance of the time trend specification, as one expects

that GDPs will have a positive effect on exports, and the estimation with the trend fixed effects gives a more likely result. Negative coefficient of per capita GDP of an exporter, as discussed in Section 3.2, suggests that the traded goods are mostly labor-intensive. Taking into account that the present analysis excludes oil exports, it is probable that the labor-intensive goods prevail over the capital-intensive in the sample trade flows.

Both real effective exchange rates show expected signs under the LSDV estimation. In particular, β7 of exporter’s REERit is negative, which implies that real depreciation

of a currency indeed makes exports more attractive to the importers. Elasticity of the exports with respect to the REER of exporter is therefore −28.8%. On the other hand, elasticity of exports to REERjt is 53.7%, which conforms to the fact that appreciation of

the currency in the importing country will lead to an increase of its imports. Even though both results are not statistically significant, results with the expected signs indicate the right direction of the research. Interestingly enough, β10 estimated with the inclusion

of trends shows a significant positive output, which suggests that the sanctions on trade exhibit positive effect on exports. This result is however insignificant, and one cannot conclude that the sanctions indeed result in higher exports between the countries, but the inclusion of Sanctijt will nevertheless be checked for robustness in the next section.

The dummy for war between Russia and Ukraine exhibits negative effect under both specifications, and the result is as expected.

4.2

Sensitivity Analysis

This section analyses robustness of the estimation results presented in Section 4.1. Several generalizations of the model are thus examined to study their effects on the estimates and to verify the sufficiency of the specification (1). In particular: 1) the model is re-estimated without per capita GDPs; 2) war and sanctions dummies are alternately excluded; 3) lags of real effective exchange rates, GDPs and per capita GDPs are included; 4) alternative control for the trends is considered.

Table 4: Sensitivity of EAEU estimate to generalizations of model (1)

Model Generalization EAEUijt

0: Baseline 0.16

(0.14)

1: No per capita GDPs specification 0.16 (0.13)

2: No Sanctions specification 0.16

(0.13)

3: No War specification 0.17

(0.13)

4: No War and Sanctions specification 0.16 (0.13)

5: Long-run effects (2 lags) 0.20

(0.14)

6: Unrestricted country trends ζit+ ζjt added 0.03

(0.13)

Clustered standard errors in parentheses

∗ _{p < 0.05}

Table 4 shows the estimates of EAEUijtunder different generalizations of the baseline

model. The baseline estimation shows that per capita GDPs do not significantly influence the exports of the countries in the sample. As an explanatory variables GDP CAPit and

GDP CAPjt mainly account for the populations of the countries i and j, respectively, and

may potentially control for some factors of economy size not fully disclosed in the GDPit

and GDPjtvariables. Insignificance of the variables in the estimation of the baseline model

may point at a weak ‘correlation’ between GDPs and per capita GDPs. The model is therefore re-estimated without the GDP CAPit and GDP CAPjt, by imposing restriction

β5 = β6 = 0. Except for the fact that variable GDPjt becomes significant at 5% level

(from 0.80 in model (1) to 1.72∗), as only the GDPs variables now contain the information on economic sizes, all other estimates are hardly affected. In particular, the variable of interest, EAEUijt, shows the same coefficient in both specifications.

Since coefficient of the dummy for sanctions on trade had an unexpected positive sign, the robustness check requires excluding alternately Sanctijt, W arijt, and both. The

second generalization (2) without the dummy for sanctions does not make any difference to the EAEU dummy nor does it change substantially any other estimated variables in comparison to the baseline model. It may be the case that the time fixed effects in the generalization already account for the influence of the sanctions quite well. Yet, there is also no suggestion of an over-specification of model (1). When the dummy for war is excluded, the EAEU estimate does not change much either, other regressors also show similar to the baseline scenario results. Moreover, one would expect that in the absence of W arijt, dummy Sanctijt would pick up a negative effect of the former, as these two

dummies are directly related. However, this is not the case, as the coefficient for Sanctijt

almost doesn’t change and remains positive. In order to verify that there is no over-specification by controlling for both war and sanctions, the model is estimated without the two dummies. Again, it has no substantial effect on the regressors, and the EAEU variable, in particular, remains just the same as in the baseline model.

Subsequently, the model is extended to control for the dynamics of the GDP, per capita GDP and REER variables. Two lags are added to GDPit, GDPjt, GDP CAPit,

GDP CAPjt, REERit and REERjt, while EXP ORTijt−1 is extended by one more lag.

There is no evidence that dummy variables for war and sanctions require lags, and EAEUijt is a stationary variable. The lagged regressors attenuate the specification of

model (1), and two past values of the GDPs and real effective exchange rates seem rele-vant for this sample (it is also a standard approach in the literature, see Bun and Klaassen, 2002a; Bun and Klaassen, 2007). Results under this specification do not show substantial difference from the estimation of model (1): the coefficient for the EAEU is equal 0.17 in the baseline specification, and 0.20 in the re-estimated model.

Another robustness check deals with the alternative trend specification, as suggested by Baltagi et al. (2003), and which was also used in Bun and Klaassen (2007). The model is generalized to account for individual country unrestricted time-varying effects ζit+ ζjt, instead of country-pair specific τij· t. The last row of Table 4 demonstrates that

the EAEU estimate decreases slightly, but remains in line with the baseline model. The reason for this may be that ζit + ζjt even though flexible in the it and jt dimensions,

are more restricted country-pairwise, so the specification does not necessarily extends the model (1). Also, the present trend specification in (1) already accounts for the omitted trending variables quite well, as it increases the EAEU estimate from -0.11 to 0.17. The country-pair trends and corresponding estimates are therefore concluded to be robust in this research.

5

Conclusion

This thesis investigated the effects of a membership of the Eurasian Economic Union on trade by using a dynamic panel data gravity model. The sample dataset included

information on 110 country-pairs over the years of 1996-2016. The initial estimation showed negative effects of the EAEU on bilateral exports. However, the residuals from that model were shown to exhibit trends different across country-pairs. The member-states of the EAEU thus particularly have downward trends, and they were partially captured by the EAEUijt in the end of the sample period. This in turn resulted in a

downward bias of the estimated EAEU effect. In order to control for the underlying trends, a method proposed by Bun and Klaassen (2007) was used, and the model was extended by including country-pair specific time trend fixed effects τij · t. The EAEU

effect hence changed from -10.4% in the initial estimation to 18.5%. Both results are insignificant under 5% level, but if it would not be the case, the latter value would imply that the members of the EAEU on average have 18.5% higher exports relative to the non-members. The resulted estimate was verified to be robust to various other specifications employed.

The final result, even though insignificant, suggests that the critics regarding the EAEU may be biased and underestimate the true effect the membership has on the trade flows. Since τij · t in the model accounts for the deviations from the global trends of

integration, the thesis concludes that the EAEU members may exhibit less than normal trends. Such hypothesis seems plausible taking into account the unique historical back-ground and economic development of the post-Soviet states. It is therefore important to take this into consideration in the future researches, and account for the omitted trending variables.

The thesis also investigated the question of how the geopolitical crisis in Ukraine and subsequent sanctions on trade affected the exports. Again, insignificantly at a 5% level, it was concluded that the conflict between Russia and Ukraine on average leads to a 20.5% decrease in exports between the two. Whereas, the effect of sanctions was estimated to be positive, which may be due to the misspecification of the model, and may require further investigation. The presence of the dummy for sanctions however does not change the coefficients of other variables, and therefore the issue is disregarded.

Overall, the results of the thesis suggest that the Eurasian Economic Union is not necessarily detrimental to the trade of the member-states. The crisis of 2014 has indeed damaged the economies of the post-Soviet countries, but the objective of the EAEU to increase the trade volumes is not necessarily unattainable. One therefore should not underestimate the effects of the union. Even though, the values presented in the present paper are inconclusive, they suggest the direction for further analysis and improvement of the model is left for future research.

References

Aitken, N. D., (1973). The Effect of the EEC and EFTA on European Trade: A Tem-poral Cross-Section Analysis,” American Economic Review, American Economic Association, vol. 63(5), pages 881-892, December.

Anderson, J. E., and van Wincoop, E., (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review, 93(1): 170-192.

Anderson, T. W. and Hsiao, C., (1981). Formulation and estimation of dynamic models using panel data, Journal of Econometrics, 18, issue 1, p. 47-82.

Arellano, M. and Bover, O., (1995). Another look at the instrumental variable estimation of error-components models, Journal of Econometrics, 68, issue 1, p. 29-51.

Aslund, A. (2013). Ukraines Choice: European Association Agreement or Eurasian Union? Peterson Institute for International Economics, PB13-22.

Baltagi, B.H., Egger, P. and Pfaffermayr, M. (2003). A generalized design for bilat-eral trade flow models, Economics Letters, Vol. 80, pp. 391-397.

Benedictis, L.D., and Taglioni, D., (2011). The Gravity Model in International Trade, Springer, 2011.

Bergstrand, Jeffrey H., (1985). The Gravity Equation in International Trade: some Microeconomic Foundations and Empirical Evidence, Review of Economics and Statistics, volume 67, number 3, August, 474-481.

Bergstrand, J., (1989). The Generalized Gravity Equation, Monopolistic Competi-tion, and the Factor-Proportions Theory in International Trade, The Review of Economics and Statistics, 71, issue 1, p. 143-53.

Blockmans, S., Kostanyan, H., Vorobiov, I. (2012) Towards a Eurasian Economic Union: The Challenge of Integration and Unity. CEPS Special Report No. 75. Blundell, R. & Bond, S. (1999). GMM estimation with persistent panel data: an

application to production functions,” IFS Working Papers W99/04, Institute for Fiscal Studies.

Boguslavska, K. (2015). The First Steps of the Eurasian Economic Union: Disputes, Initiatives and Results. Center for Security Studies.

Bun M., Klaassen F. (2002a). Has the Euro Increased Trade?. Tinbergen Institute Discussion Paper 02-108/2.

Bun M., Klaassen F. (2002b). The Importance of Dynamics in Panel Gravity Models of Trade. UvA-Econometrics Discussion Paper 2002/18.

Bun, M. J. G., & Klaassen, F. J. G. M. (2007). The Euro Effect on Trade is not as Large as Commonly Thought. SSRN Journal.

Cheng, I. and Wall, H., (2005). Controlling for heterogeneity in gravity models of trade and integration, Review, issue Jan, p. 49-63.

Darvas, Z., (2012). ’Real effective exchange rates for 178 countries: a new database’, Working Paper 2012/06, Bruegel, 15 March 2012

Deardorff, A., (1998). Determinants of Bilateral Trade: Does Gravity Work in a Neoclassical World?, p. 7-32 in , The Regionalization of the World Economy, National Bureau of Economic Research, Inc.

Dreyer, I., and N. Popescu. (2014.) The Eurasian Customs Union: The Economics and the Politics. Brief No. 11 (March 21), European Union Institute for Security Studies, Paris, FR.

Eurasian Economic Commission Statistics Department. The EAEU Statistics. (2017) Online. Retrieved on April 26, 2017 from: http://www.eurasiancommission.org/ru

Frankel, J., Stein, E. and Wei, S.-J., (1995), Trading blocs and the Americas: The natural, the unnatural, and the super-natural, Journal of Development Eco-nomics, 47, issue 1, p. 61-95.

Glick, R. and Rose, A.K. (2002). Does a currency union affect trade? The time series evidence, European Economic Review, Vol. 46, pp. 1125-1151.

Goldstein, M. and Khan, M., (1985), Income and price effects in foreign trade, ch. 20, p. 1041-1105 in Jones, R. W. and Kenen, P. B. eds., Handbook of International Economics, vol. 2, Elsevier.

Grauwe, P. and Skudelny, F., (2000), The impact of EMU on trade flows, Review of World Economics (Weltwirtschaftliches Archiv), 136, issue 3, p. 381-402.

Hansl, B. (2016). Russia Economic Report 33: The Dawn of a New Economic Era. Haveman, J. & Hummels, D., (1998). Trade Creation and Trade Diversion, Journal

of Transnational Management Development, 3:2, 47-72.

Helpman, E. and Krugman, P., (1985). Market Structure and Foreign Trade, Cam-bridge. MIT Press.

IMF Country Report No. 15/211. Russian Federation. 2015 Article IV Consultation - Press Release and Staff Report. August 2015.

Judson, R. A. and Owen, A.L. (1999). Estimating dynamic panel data models: a guide for macroeconomists. 94(3): 605-627. Economics Letters

Kansikas, S. (2017). The Eurasian Economic Union, Russias Integration Policy and the EU Challenge. Journal on Baltic Security, 1(1), pp. 108-116.

Khitakhunov, A., Mukhamediyev, B. and Pomfret, R., (2017), Eurasian Economic Union: present and future perspectives, Economic Change and Restructuring, 50, issue 1, p. 59-77.

Krugman, P.R., (1991). The move to Free Trade Zones. In Symposium on Policy Implications of Trade and Currency Zones, Federal Reserve Bank of Kansas City. Krugman, P.R., & Obstfeld, M. (2005). International economics: theory and policy.

Boston: Pearson Addison Wesley.

Linnemann, H., (1966). An Econometric Study of International Trade Flows. Ams-terdam: NorthHolland.

Magee, C. S. (2007), New Measures of Trade Creation and Trade Diversion, unpub-lished manuscript, Bucknell University.

Martnez-Zarzoso, I. & Horsewood, N. (2005). Regional trading agreements: dynamic panel data evidence from the gravity model.

Mazhikeyev, A., Edwards, T. and Rizov, M., (2015), Openness and isolation: The trade performance of the former Soviet Central Asian countries, International Business Review, 24, issue 6, p. 935-947.

National Statistical portal of the Republic of Belarus. (2017) Online. Retrieved on April 17, 2017: www.belstat.gov.by/en/

Nickell, S., (1981). Biases in Dynamic Models with Fixed Effects, Econometrica, 49, issue 6, p. 1417-26.

OECD Statistics Database. (2017). Online. Retrieved on April 18, 2017 from: stats.oecd.org

Olivero, M. P. and Yotov, Y. V. (2012), Dynamic gravity: endogenous country size and asset accumulation. Canadian Journal of Economics/Revue canadienne d’conomique, 45: 6492.

Popescu, N. (2014). Eurasian Union: the real, the imaginary and the likely. Chaillot Papers, (132).

Soloaga, I. and Wintersb, A. L., (2001). Regionalism in the nineties: what effect on trade?, The North American Journal of Economics and Finance, 12, issue 1, p. 1-29.

Stepanyan, A., Roitman, A., Minasyan G., Ostojic D., and Epstein, N. (2015) The Spillover Effects of Russias Economic Slowdown on Neighboring Countries. International Monetary Fund.

Strzelecki, J. (2016). The Eurasian Economic Union: a time of crisis. The Centre for Eastern Studies.

Tarr, D. G. (2016). The Eurasian Economic Union of Russia, Belarus, Kazakhstan, Armenia, and the Kyrgyz Republic: Can It Succeed Where Its Predecessor Failed? Eastern European Economics, 54(1).

Tinbergen, J. (1962). Shaping the World Economy: Suggestions for an International Economic Policy. New York: The Twentieth Century Fund.

United Nations Statistics Division. UN Commodity Trade Statistics Database. New York: United Nations.

Vinokurov E., Demidenko M., Pelipas I., Tochitskaya I., Shymanovich G., Lipin A., Movchan V., (2015) Estimating the economic effects of reducing non-tariff barriers in the EEU Eurasian Development Bank, St. Petersburg (Report 29) Wall, H.J. (1999) Using gravity model to estimate the costs of protection Review of

the Federal Reserve Bank of St. Louis: January/February, 33-40.

Wansbeek, T. J., & Kapteyn, A. J. (1989). Estimation of the error-components model with incomplete panels. Journal of Econometrics, 41, 341-361.

World Bank. (2017). Database. Online. Retrieved April 20, 2017, from http://data.worldbank.org

World Economic Outlook database. (2017). Online. Retrieved April 25, 2017, from http://www.imf.org/external/pubs/ft/weo/2014/02/weodata/index.aspx

A

Subtracted commodities listed by the HS codes HS code Commodity

27010 Benzol 270720 Toluole 270730 Xylole 270740 Naphtalene

270750 Aromatic hydrocarbon mixtures from coal tar, nes 270760 Phenols

270791 Creosote oils

2709 Petroleum oils, oils from bituminous minerals, crude 2710 Oils petroleum, bituminous, distillates, except crude 2711 Petroleum gases and other gaseous hydrocarbons 2712 Petroleum jelly, petroleum wax, other mineral waxes 2713 Petroleum coke, bitumen and other oil industry residues