UNIVERSITY OF GRONINGEN
FACULTY OF ECONOMICS AND BUSINESS
The effects of the Deep and Comprehensive Free Trade Area between
the European Union and Ukraine on trade
2
ABSTRACT
In recent years, there has been a rising interest of researchers in the effects of free trade agreements. In this thesis, the trade effects of the Deep and Comprehensive Free Trade Area (DCFTA) between the European Union and Ukraine are studied. The gravity model of trade is applied, and the Ordinary Least Squares (OLS) and the Poisson Pseudo Maximum Likelihood (PPML) estimators with fixed effects to account for multilateral resistance terms and unobserved time-invariant heterogeneity across country pairs are used. The estimations show robust evidence of the positive and significant impact of the agreement on trade. This suggests that the effect is in line with the objective of the DCFTA to enhance trade between the EU and Ukraine. There is no evidence of trade diversion found, meaning no evidence of economic efficiency losses associated with the agreement. Besides, according to the findings, the DCFTA had heterogeneous effects on the signatories. The results of the research can be used by policymakers to reflect on the effectiveness of the agreement in terms of its effects on trade.
3
TABLE OF CONTENTS
INTRODUCTION ... 4
LITERATURE REVIEW ... 6
Impact of trade agreements on trade ... 6
Trade creation and trade diversion ... 10
Heterogeneous effects of trade agreements on signatories ... 11
DATA AND METHODS ... 13
The model ... 13
Data ... 16
Estimation methods and econometric issues ... 17
EMPIRICAL RESULTS ... 19
Descriptive analysis ... 19
Regression results ... 21
DISCUSSION ... 29
Discussion and recommendations for future research ... 29
Limitations ... 33
CONCLUSIONS ... 34
REFERENCES ... 36
4
INTRODUCTION
A surge in the number of free trade agreements (FTAs) since the 1990s was accompanied by an increased interest of the researchers in the effects of those on economies. By definition, FTAs are meant to reduce trade barriers by eliminating or reducing customs tariffs in bilateral trade, and consequently increase trade between countries. From a broader theoretical perspective, international trade is expected to be beneficial for countries. According to the conventional theory of comparative advantage (Ricardo, 1817), free trade allows countries to increase overall consumption by letting them focus on production and export of a good for which they have a comparative advantage while importing the other good. At the same time, new trade theory (Krugman, 1979) pays attention to the economies of scale, suggesting that the increasing returns to scale lead to product differentiation and intra-industry trade between the countries. "New" new trade theory predicts a positive impact of trade openness on overall aggregate productivity as a result of moving resources away from less productive to more productive firms (Melitz, 2003; Melitz & Trefler, 2012). While the mechanisms and the assumptions of the classical theories differ, all of them suggest a positive impact of trade on economies.
Much empirical research has been done aimed to study the consequences of trade liberalization. For example, Pavcnik (2002) investigated how trade liberalization affected plant productivity in the case of Chile and found evidence of within-plant productivity increase due to trade liberalization. In addition, the study suggests aggregate productivity improvements due to the shift of the resources to more productive firms, which supports the theory of Melitz. Trefler (2004) found that the Canada-U.S. Free Trade Agreement led to a significant increase in aggregate productivity of the industries that experienced the largest tariff cuts. However, together with the long-run efficiency gains, the author also points out the negative impact on employment and less productive plants in the short-run. Firm‐level productivity was also found to rise due to trade liberalization in the case of India (Topalova & Khandelwal, 2011).
As can be seen, there is evidence of trade agreements affecting productivity, firm exit and entry, and employment. In the origin of those changes is trade liberalization and thus changes in trade itself. However, as mentioned by Kohl (2014) and World Bank (2005), among others, and discussed in detail in the following section, not all trade agreements actually have a significant impact on trade, and some trade agreements have more impact than others. Therefore, if one is interested in the effects of a specific agreement, it is important to study it individually and find out to what extent the agreement affects trade.
5
In recent years, one of the goals of the EU policymakers has been to effectively enhance trade with other countries. As of 2019, the EU had 41 trade agreements with 72 countries (European Commission, 2019), which, among others, are aimed to provide opportunities for European businesses to trade with countries outside the EU at lower costs. One of the most recent trade agreements signed by the EU is the Deep and Comprehensive Free Trade Area (DCFTA) with Ukraine.
Trade between Ukraine and the EU is important for both parties. The European Union, being a huge economy consisting of rather developed countries, is the largest trade partner of Ukraine. It accounts for about 42% of Ukraine’s trade as of 2018 (State Statistics Service of Ukraine, 2020). Ukraine is also a valuable partner for the EU. It is a developing country fully located in Europe. Ukraine has more than 1300 km of common borders with the European Union, and it is big both by area and population. More specifically, Ukraine is larger by area than any of the EU countries (World Bank Open Data, 2020), and it is the 6th country by population – after Germany, France, United Kingdom, Italy and Spain – compared to the EU28 member states (United Nations Population Fund, 2020). At the same time, the population density of Ukraine, although smaller than that of the EU, is rather high – about 70 people per square kilometre (State Statistics Service of Ukraine, 2020). This means over 40 million neighbouring potential consumers for the EU producers. As of 2018, Ukraine is the 21st biggest trade partner of the EU, accounting for about 1% of its trade (Eurostat, 2020). The partnership between the EU and Ukraine has been getting stronger in recent years, and in 2014 an Association Agreement (2014) was signed aimed to strengthen economic and political links between the parties. Besides, by the European Parliament resolution (2014), the EU pointed out to the possibility of Ukraine to become a member of the European Union in the future. Considering these facts, studying trade between these countries is particularly interesting.
6
To the best of my knowledge, at the moment there is no scientific paper in reputable journals that assesses the ex-post trade effects of the EU-Ukraine trade agreement on the parties. This might be the case because the agreement only entered into force in 2016 and timewise there hasn’t been enough data to assess its impact on trade. Narrowly targeted research trying to predict the effects of the DCFTA on the Dutch economy was done by Oomeset al. (2017). Their estimations show that in the long run the agreement could increase Dutch exports to Ukraine by nearly three times and Dutch imports from Ukraine by about two times. Considering the importance of studying the trade effects of individual trade agreements and the importance of trade relations between the EU and Ukraine, the study of the ex-post effects of the DCFTA between the EU and Ukraine on trade is relevant from the research perspective. The results obtained are also useful from the policy perspective. Knowing what the effects of the DCFTA on trade have been, can help policymakers to reflect on the effectiveness of the policy in regards to whether the intervention was justified and whether its objectives have been achieved, which in turn can serve as a background for finding ways to improve future policies. To give an example, the European Commission (2020) publishes reports of ex-post evaluations of the EU trade policies. This indicates the practical importance of assessing the actual effects of agreements.
Thus, with this research, I aim to contribute to the literature by investigating the effects of the DCFTA between Ukraine and the EU member states on bilateral trade. In addition to estimating the total trade flow effect, I also check for trade diversion resulting from the agreement and estimate the effects on each of the EU countries individually. An extensive discussion on the research questions is presented in the following section.
The research is organized as follows. In the following section the literature on how trade agreements influence trade is reviewed. Based on the literature the hypotheses for the research are stated. Afterwards, I propose a model for the research. Further, data and methodology are discussed. The results are presented thereafter. The thesis ends with a discussion of the results in the context of the literature and a conclusion.
LITERATURE REVIEW Impact of trade agreements on trade
7
address the endogeneity of FTAs. Baier and Bergstrand (2007) looked into bilateral trade flows of 96 potential trading partners between 1960 and 2000. It was found that generally FTAs have a positive impact on bilateral trade. More specifically, estimates show that on average bilateral trade increases by 58% as a result of an FTA. Besides, when the authors account for the phase-in period of FTAs by lagging the effects, they find that trade between the countries involved increases approximately twofold (by 114%) in 10 years after the implementation of an FTA.
Although the objective of trade agreements is to enhance trade, the effects can also be insignificant or even negative; as noted by the World Bank (2005) design and implementation of the agreement play a role in explaining this. Kohl (2014) assessed the effect of 166 Economic Integration Agreements (EIAs) throughout 1950-2010 on trade flows and found that more than a half (63.9%) of the agreements had no significant impact on bilateral trade, only about a quarter (26.5%) had positive impact (with a maximum of 50% increase in trade estimated) and the remaining 9.6% were estimated to have a negative effect on trade. According to the author, institutional quality and design of the agreements along with countries’ participation in the WTO can help to partially explain this.
First of all, by institutional quality of an agreement, Kohl (2014) means the extent to which the necessary details are included in the agreement (for example, dispute settlement mechanism and liberalisation schedule). The author found a positive relation between the institutional quality and the effects on trade. Secondly, turning to the institutional design, it can be defined as the scope of the agreement in terms of policy areas (provisions) covered (for example, customs administration, investment and public procurement). Kohl (2014) found that institutional design combined with binding commitments has positive relation with the effects of agreements on trade.
8
In contrast to the findings of Kohl (2014) about a rather small share (26.5%) of agreements having positive effects on trade, Baier et al. (2019) found positive effects for more than a half (57%) of the agreements in their study, which used data on 70 countries in 1986-2006. The difference in the results of these studies might, among others, be explained by different estimation methods employed. For example, Baier et al. (2019) used the Poisson Pseudo Maximum Likelihood estimation instead of the OLS in their research. The reason why the PPML estimation should be preferred in this type of research is discussed in the following section. Furthermore, Soete and Van Hove (2017) studied the effects of EIAs involving the EU in 1988-2013. It was found that 55% of the agreements in the sample had a positive effect on trade, and 40% were estimated to have insignificant effects.
Since the effects of FTAs clearly differ from one agreement to another, much research has been done to examine individual agreements. Various studies look into the trade effects of agreements involving the EU in particular. Egger and Larch (2011) assessed the effects of the Interim Agreements and the Europe Agreements of the 1990s between the EU15 and 10 countries in Central and Eastern Europe (CEECs). The agreements were aimed to (partially) eliminate tariffs between the parties on bilateral level and therefore to strengthen economic links between the countries. The authors use panel data for 167 economies over the period from 1990 to 2001, which covers two years before the first agreement entered into force and two years after the last one. It was found that the agreements had a significantly positive effect on trade between the EU15 and the CEECs. The Europe Agreements increased bilateral trade on average by 30%. In addition, the authors estimated that the agreements led to an increase in GDP and welfare of the countries involved. An increase in welfare was larger for the countries in Central and Eastern Europe than for the EU15.
There are also studies of trade agreements of the European countries with countries in other regions. Bensassi et al. (2012), for example, investigated the effects of Euro-Mediterranean trade agreements on trade between four EU countries (France, Germany, Italy and Spain) on one side and four North African countries (Algeria, Egypt, Morocco and Tunisia) on the other side considering the period of 1995-2008. Their findings suggest a positive effect of the agreements on export of North African countries to their EU partners. It should be noted that this research also looks at the effects of the agreements on extensive (number of exporting firms) and intensive (value of individual shipments) margins separately and finds that the increase in exports from Algeria and Tunisia was mainly driven by the intensive margin, while for Egypt and Morocco both margins were important. The authors mention that different trade patterns of the countries might explain this.
9
led to an increase in both the probability to export (extensive margin) and the value of exports (intensive margin). Notably, the effect was larger for the exports from the EU (about 10% increases in both margins) than exports from Korea. The authors mention that this can be explained by the fact that the European side had larger opportunities for gains from the alleviation of uncertainty at the destination market. In addition, the authors found evidence of anticipation effects of the agreement. These effects were also noticed by Magee (2008), discussed further, who found that trade increases by about 26% within the 4 years before an average agreement enters into force. Although trade agreements are aimed to enhance trade between countries, literature suggests that their effects can also be insignificant or even negative. As shown above, literature provides evidence of the positive effects of various agreements involving the EU on trade. The effects of the Deep and Comprehensive Free Trade Area between Ukraine and the EU are to be investigated in this study. Based on the literature the first hypothesis to be stated is the following:
Hypothesis 1: The DCFTA had positive impact on trade between Ukraine and the EU.
The DCFTA between the EU and Ukraine was preceded by unilateral trade preferences (the Autonomous Trade Preferences) granted by the EU to Ukraine. Turning to unilateral trade preferences, only a few studies investigated their effects on trade accounting for unobserved heterogeneity between countries. One of these studies is Péridy (2005), which looked at the effects of the trade preferences granted by the EU to Mediterranean countries in the period from 1975 to 2001. The author found that the trade preferences raised exports from the Mediterranean countries to the EU by about 20-27% depending on the model specification. Another study was done by Persson and Wilhelmsson (2007), who also provide evidence of the positive effects of unilateral trade preferences on trade. The authors investigated the effects of trade preferences that the EU offered to developing countries over the period 1960-2002 and found that they generally led to increased exports from developing countries to the EU. The effects varied depending on the preference system with the largest increase of exports of about 30% for African, Caribbean and Pacific countries. Keeping in mind the fact that trade preferences are aimed to create better conditions for trade, in this study, it can be expected to find positive effects of the ATP on exports from Ukraine to the EU.
10
associated with greater effects on trade. Therefore, because of the lower envisaged degree of integration, I expect the effects of the ATP to be smaller than the effects of the DCFTA.
Trade creation and trade diversion
When one talks about the effects of trade agreements on trade, the question of the economic efficiency gains arises automatically. If a trade agreement indeed enhances trade between the signatories, it is important to know where the additional trade comes from. In this context, Viner (1950) introduced the concepts of trade creation and trade diversion. The author argued that the new trade results from a shift of discrimination of sources of supply. Trade creation was described as a situation in which new trade is a result of a better international resource allocation. According to Viner, this happens when a part of domestic production is replaced by imports from more efficient producers in a partner country. At the same time, trade diversion refers to a situation in which new trade results from a shift of imports: when producers in the rest of the world are more efficient than in a country with which a trade agreement was signed, however because of the lowered tariffs with the latter, imports from that country increase (in expense of imports from the rest of the world). Thus, when trade switches to lower-cost producers, economic efficiency is improved (trade creation). On the other hand, trade diversion can be harmful not only to more efficient producers in the rest of the world but also to a country that turns to less efficient suppliers.
Magee (2008) used a dataset of 133 WTO countries in the period of 1980-1998 and estimated the trade diversion and trade creation effects of the trade agreements between the countries. The author emphasized that a correct interpretation of trade diversion of Viner (1950) is a decrease in trade of the countries that introduced an FTA with the rest of the world accompanied by an increase of trade between the signatories. Therefore, to conclude whether an agreement was creating or trade-diverting, Magee firstly looks at whether intra-block trade rose as a result of the agreement and then classifies it as trade-creating or trade-diverting depending on how extra-block trade has changed. The author didn’t find any robust and significant (at 1% level) evidence of trade diversion on a large scale. However, for some individual trade agreements, the results show a decrease of extra-block trade. In the end, taking into account changes in intra-extra-block trade, the evidence of trade diversion was only found for two agreements in the dataset: the 1986 European Communities expansion and the European Communities trade deals with Romania and Bulgaria in 1993. On the other hand, Cheong et al. (2015a), using data on 216 countries in 1980-2010, found that when one accounts for multilateral resistance terms while estimating the trade diversion effects, trade diversion and trade creation effects appear to be comparable in dollar terms.
11
resistance. First of all, the authors found that the trade agreements enhanced intra-block trade, therefore supporting the idea that FTAs lead to better economic integration between countries. Furthermore, estimates show that in the case of three countries (Slovenia, Czech Republic and Slovak Republic) the trade increased at the expense of trade with the third countries. Thus, for several countries robust evidence of trade diversion was found.
Magee (2016) studied the trade effects of the customs union between Turkey and the European Community, which entered into force in 1995. Similarly to Cheong et al. (2015a), the author accounted for multilateral resistance terms by the inclusion of fixed exporter-time and importer-time effects in their model. The results show a twice larger trade-creating effect in comparison to the diverting effect.
Besides, Magee (2017) analysed all potential free trade agreements between 122 countries regarding their trade creation and trade diversion effects using trade data of 2012/2013. Only about a quarter of imports were estimated to suffer from diverting effects. Moreover, the author noted that with the rising number of trade agreements, the share becomes lower, leading to a conclusion that the concern about trade diversion is to become less relevant. Magee (2017) also provided estimates for potential trade-creating and trade-diverting effects in the case of a free trade agreement between Ukraine and the EU. The results, which were provided for Ukraine, show the net trade creation effect of about USD 11.29 per capita. In addition, it was found that countries signing agreements with the European Union (and the United States) are generally expected to have larger trade creation than trade diversion effects.
Overall, according to Viner (1950), trade creation means an increased economic efficiency, and trade diversion is an undesirable effect. The literature suggests that usually the trade-creating effect dominates. Therefore, the second hypothesis to be tested in this research is the following:
Hypothesis 2: Implementation of the DCFTA didn’t result in trade diversion. Heterogeneous effects of trade agreements on signatories
Several researchers pointed out that countries can be affected differently by an FTA. World Bank (2005) estimated the effects of 17 trade agreements on individual members and found no evidence that countries benefit equally from the agreements. Baier et al. (2019) found that two thirds of the heterogeneity occurs within trade agreements rather than across them.
12
member states. It should be noted that Soete and Van Hove (2017) estimate the effects on extensive (number of goods traded) and intensive (value traded of each good) margins separately. They found that FTAs generally lead to a decrease in the intensive margin and an increase in the extensive margin. The impact on different countries varied, suggesting heterogeneous effects of the agreements on signatories. For example, as a result of FTAs signed by the EU, trade of Ireland increased by almost 150% at the extensive margin, but the trade of Spain – by only 26%. Notably, for the vast majority of the countries the effect of FTAs on total trade flows was insignificant. Nilsson Hakkala et al. (2019) focused on assessing the effects of the EU FTAs on the trade of Finland. First of all, they found that the EU trade agreements in the period of 1988-2017 on average had positive and significant effect on trade of the EU with partner countries. Secondly, looking at the period from 1995 (since Finland joined the EU) until 2017, the authors found that the overall effect of the agreements on exports from Finland was statistically insignificant. Nevertheless, FTAs led to a substantial increase in exports in the industries where Finland has a comparative advantage. Regarding Finnish imports, the effects on them were heterogeneous, but the overall effect was negative.
The research of Magee (2008), discussed in the previous subsection, also provides some evidence of heterogeneous effects of trade agreements on the signatories. The author noticed that the trade effects of FTAs are larger when the signatories are larger and located nearby. Overall, literature provides evidence of heterogeneous effects of trade agreements on countries; therefore, the following hypothesis can be stated:
Hypothesis 3: The DCFTA had heterogeneous effects on the signatories.
13
(when the average GDP per capita of the country pair is smaller). Lastly, the estimations of Baier et al. (2018) show that common language and common religion increase trade-flow EIAs elasticity. On the other hand, EIAs affect trade less in the case of common legal origins and colonial history; however, the coefficients for these two variables were significant for EIAs in general, but not significant for FTAs specifically. Similarly, Vicard (2011) showed that sharing a common language leads to a larger trade increase resulting from a trade agreement, while common colonial history was estimated to lead to the reverse; the latter might be explained by non-trade reasons to sign the agreement.
Table 1. Some sources of heterogeneity across country pairs within trade agreements Source of heterogeneity Effect on trade Authors
Distance Smaller Vicard (2011)
Cheong et al. (2015b) Baier et al. (2018) Baier et al. (2019)
Economic size Larger Vicard (2011)
Baier et al. (2019) Similarly in terms of
economic size
Larger Vicard (2011)
Cheong et al. (2015b) Similarly in terms of income Larger Cheong et al. (2015b)
Common language Larger Vicard (2011)
Baier et al. (2018) Common colonial history Smaller Vicard (2011)
Baier et al. (2018)
Common religion Larger Baier et al. (2018)
Common legal origins Smaller Baier et al. (2018)
Table 1 summarizes the findings of different authors on some of the sources of heterogeneity within trade agreements. It can also be noted that according to Baier et al. (2019), almost half of the within-agreement heterogeneity is actually attributed to asymmetric effects within country pairs; however, looking at these effects is beyond the scope of this study. As mentioned earlier,Lakatos and Nilsson (2017) also provided some evidence of within-agreement heterogeneityin the effects of the EU-Korea trade agreement.
DATA AND METHODS The model
14
the force (F) directly proportional to their masses (𝑀1 and 𝑀2) and inversely proportional to the squared distance between them (𝑟2). G is a constant.
𝐹 = 𝐺𝑀1𝑀2 𝑟2
(1)
Tinbergen (1962) applied this physical law to economics and suggested that export from country i to country j (𝐸𝑖𝑗) is directly proportional to the economic masses of the countries 𝑌𝑖𝛼𝑌𝑗𝛽 (usually measured by GDP) and inversely proportional to the distance between them. G is a constant. This is represented in equation (2). 𝐸𝑖𝑗 = 𝐺𝑌𝑖 𝛼𝑌 𝑗 𝛽 𝐷𝑖𝑠𝑡𝑖𝑗𝛾 (2)
Traditionally, the gravity model of trade (hereafter – the gravity model) is estimated in the log-log form. In its basic version it looks like equation (3), where 𝜀𝑖𝑗 is the error term, and a constant G from equation (2) becomes a part of the intercept 𝛽0.
ln𝐸𝑖𝑗 = 𝛽0+ 𝛽1ln𝑌𝑖 + 𝛽2ln𝑌𝑗+ 𝛽3ln𝐷𝑖𝑠𝑡𝑖𝑗 + 𝜀𝑖𝑗 (3) After the basic gravity model of trade was introduced, various additional explanatory variables were added to it by researchers. The most commonly used are variables for common border (related to physical distance between countries), common language and common history (related to socio-economic distance).
Nowadays, the gravity model is widely used to assess the ex-post effects of trade agreements on trade (for example, Soete & Van Hove, 2017;Kohl et al., 2016;Egger & Larch, 2011; Head et al., 2010; Magee, 2008 and Baier & Bergstrand, 2007), therefore it is also employed in this study. The basic equation I estimate is the following:
𝑙𝑛𝑋𝑖𝑗𝑡 = 𝛽0+ 𝛽1𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑗𝑡+ 𝛽3𝑙𝑛𝐷𝐼𝑆𝑇𝑖𝑗 + 𝛽4𝐶𝑂𝑁𝑇𝐼𝐺𝑖𝑗 + 𝛽5𝐿𝐴𝑁𝐺𝑖𝑗+ 𝛽6𝐻𝐼𝑆𝑇𝑖𝑗 + 𝛽7𝐸𝑈𝑖𝑗𝑡 + 𝛽8𝐷𝐶𝐹𝑇𝐴𝑖𝑗𝑡+ 𝜀𝑖𝑗𝑡
(4)
15
and 0 otherwise), 𝐿𝐴𝑁𝐺𝑖𝑗 is a dummy variable for common official language between countries i and j (1 in case of common language and 0 otherwise), 𝐻𝐼𝑆𝑇𝑖𝑗 is a dummy variable for common history between countries i and j (1 in case of common history and 0 otherwise), 𝐸𝑈𝑖𝑗𝑡 is a dummy variable that takes the value of 1 when both countries i and j are members of the EU in year t, 𝐷𝐶𝐹𝑇𝐴𝑖𝑗𝑡 is a dummy variable that takes the value of 1 when the DCFTA is in place between countries i and j in year t, 𝜀𝑖𝑗𝑡denotes the error term.
According to the gravity model, contiguity, common language, common history and larger GDP of the countries are expected to have positive impact on trade between the countries. Distance is expected to have a negative relationship with trade. A positive coefficient estimate of the DCFTA dummy would suggest that trade rises when the Deep and Comprehensive Free Trade Area is in place between the countries.
Further, a variable for the Autonomous Trade Preferences is added to the model:
𝑙𝑛𝑋𝑖𝑗𝑡 = 𝛽0+ 𝛽1𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑗𝑡+ 𝛽3𝑙𝑛𝐷𝐼𝑆𝑇𝑖𝑗 + 𝛽4𝐶𝑂𝑁𝑇𝐼𝐺𝑖𝑗+ 𝛽5𝐿𝐴𝑁𝐺𝑖𝑗 + 𝛽6𝐻𝐼𝑆𝑇𝑖𝑗+ 𝛽7𝐸𝑈𝑖𝑗𝑡+ 𝛽8𝐷𝐶𝐹𝑇𝐴𝑖𝑗𝑡+ 𝛽9𝐴𝑇𝑃𝑖𝑗𝑡 + 𝜀𝑖𝑗𝑡
(5)
where 𝐴𝑇𝑃𝑖𝑗𝑡 is a dummy variable that takes the value of 1 when the Autonomous Trade Preferences are granted by country j to country i in year t.
To test for trade diversion, a diversion dummy similar to the one used by Magee (2008) was added to the equation (5):
𝑙𝑛𝑋𝑖𝑗𝑡 = 𝛽0+ 𝛽1𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑗𝑡+ 𝛽3𝑙𝑛𝐷𝐼𝑆𝑇𝑖𝑗 + 𝛽4𝐶𝑂𝑁𝑇𝐼𝐺𝑖𝑗+ 𝛽5𝐿𝐴𝑁𝐺𝑖𝑗 + 𝛽6𝐻𝐼𝑆𝑇𝑖𝑗+ 𝛽7𝐸𝑈𝑖𝑗𝑡+ 𝛽8𝐷𝐶𝐹𝑇𝐴𝑖𝑗𝑡+ 𝛽9𝐴𝑇𝑃𝑖𝑗𝑡+ 𝛽10𝐷𝐼𝑉𝐸𝑅𝑆𝐼𝑂𝑁𝑖𝑗𝑡+ 𝜀𝑖𝑗𝑡
(6)
16
𝑙𝑛𝑋𝑖𝑗𝑡 = 𝛽0+ 𝛽1𝑙𝑛𝐺𝐷𝑃𝑖𝑡+ 𝛽2𝑙𝑛𝐺𝐷𝑃𝑗𝑡+ 𝛽3𝑙𝑛𝐷𝐼𝑆𝑇𝑖𝑗 + 𝛽4𝐶𝑂𝑁𝑇𝐼𝐺𝑖𝑗+ 𝛽5𝐿𝐴𝑁𝐺𝑖𝑗 + 𝛽6𝐻𝐼𝑆𝑇𝑖𝑗+ 𝛽7𝐸𝑈𝑖𝑗𝑡+ 𝛽8𝐷𝐶𝐹𝑇𝐴𝑖𝑗𝑡× 𝐼𝑗+ 𝛽9𝐴𝑇𝑃𝑖𝑗𝑡+ 𝜀𝑖𝑗𝑡
(7)
In addition to the traditional OLS estimation, the Poisson Pseudo Maximum Likelihood (PPML) estimator is employed at all the stages of the study. This method is suggested by Santos Silva and Tenreyro (2006). The authors recommend using PPML for gravity equation estimation as a good way to deal with trade data heteroscedasticity and zero trade flows. Egger and Larch (2011), Head et al. (2010) and Magee (2008), among others, also use this method for the gravity equation estimation. Equation (8) is the basic gravity equation (as in (3)) in the PPML form. In the PPML the dependent variable is exports, instead of the log of exports as in the OLS estimation. Following Santos Silva and Tenreyro (2006) the function can be interpreted as the conditional expectation of 𝐸𝑖𝑗 given 𝑌𝑖, 𝑌𝑗 and 𝐷𝑖𝑠𝑡𝑖𝑗.
𝐸𝑖𝑗 = exp [𝛽0+ 𝛽1ln𝑌𝑖 + 𝛽2ln𝑌𝑗+ 𝛽3ln𝐷𝑖𝑠𝑡𝑖𝑗] + 𝜀𝑖𝑗 (8) Data
I use data on 46 countries (Appendix A). Among those there are the EU28 and Ukraine – countries that have signed an agreement under study. Also, there are top export destinations of the EU and Ukraine in the dataset. To add those I use top extra-EU trade partners of the EU (Eurostat, 2020) and of Ukraine (State Statistics Service of Ukraine, 2020) by total trade in goods as of 2018. After taking the top 11 non-EU trade partners of Ukraine and the top 14 extra-EU trade partners of the EU, since some countries overlap, I end up with a list of 17 countries, which represent 70,68% of extra-EU trade of the EU and 73,83% of that of Ukraine in 2018.
For bilateral trade, I use the value of exports data (in millions USD) from the International Monetary Fund’s Direction of Trade Statistics (DOTS). Data on countries’ nominal GDP (in current USD) is from the World Bank’s World Development Indicators. These data sources were also used by Baier and Bergstrand (2007), Head et al. (2010) and Kohl et al. (2016) for their gravity models estimations. Following Head et al. (2010) both trade flows and GDPs are not deflated. Data on bilateral distance between the countries (distance in kilometres between the biggest cities in terms of population), common border, common language (common official language) and common history (colonial relationship post 1945) is from the CEPII geographical database GeoDist by Mayer and Zignano (2011), which was also used by Egger and Larch (2011) and Kohl et al. (2016) among others.
17
data available on 3 years after the DCFTA started to work (2016-2018). In the period of 2014-2016 there were trade preferences (the ATP) in force between Ukraine and the EU which also should be included in the model. Following Egger and Larch (2011), who focused on Europe Agreements and used data covering the period two years before the first agreement was enacted and two years after the last one, I use a time period starting two years before the ATP started to work. Thus, for each country pair data for 2012-2018 is included in the dataset. Consequently, the final dataset contains 14466 observations (24 observations are excluded because of the missing export values).
Estimation methods and econometric issues
Primarily, the gravity model is estimated using the OLS. This is a traditional way of estimating the gravity equation introduced by Tinbergen (1962) and still used by various researchers such as Egger and Larch (2011), Head et al. (2010) and Magee (2008). Furthermore, various fixed effects, that help to deal with different challenges of the gravity model estimation, are added to the model. Firstly, I include time fixed effect dummies to capture the differences in global trade over the years. These dummies are also employed by Magee (2008) and Baier and Bergstrand (2007) to account for shocks that affect global trade flows in a specific year.
18
be correlated with the likelihood of signing an FTA, fixed effects model is more suitable than the random effects model, as the latter would assume no correlation between these variables. In addition, Egger (2000) used the Hausman test to check whether random or fixed effects are more suitable and concluded that the fixed effect model is more appropriate for gravity model estimations. It should be mentioned that country pair fixed effects also capture the effects of observed time-invariant variables. In the case of this research these are bilateral distance, common border, common language and common history.
Thirdly, the importance of accounting for multilateral resistance terms was emphasized by Baldwin and Taglioni (2006). According to Adam and Cobham (2007), multilateral trade resistance can be defined as “the barriers to trade that each country faces with all its trading partners”. As suggested by Baier and Bergstrand (2007) and also employed by Soete and Van Hove (2017), Egger and Larch (2011) and Head et al. (2010), I include fixed exporter-time (exporter-year) and importer-time (importer-year) effects to control for multilateral resistance terms. According to Egger & Larch (2011), these fixed effects “wipe out” multilateral resistance terms because of perfect collinearity. Exporter-time and importer-time fixed effects control for country-specific characteristics, whether they change over time or not – for example, the economic and political situation in the country (Egger and Larch, 2011). Since both observable and unobservable country-specific characteristics are captured by these fixed effects, the GDP variables drop out from the regression. In addition, these fixed effects also absorb the time dummies (as also mentioned by Baldwin and Taglioni, 2006), because fixed exporter-time and importer-time effects add up to total year fixed effect. It should be mentioned that while testing for trade diversion, country-year fixed effects cannot easily be included since in the presence of these fixed effects the diversion dummy gets dropped out (Magee, 2008). This is because FTA and diversion dummies add up to the country’s change in overall exports represented by country-year fixed effects. Therefore, when the trade diversion is estimated, these fixed effects are not included in the regression.
19
the PPML for gravity equation estimation is that it helps to handle zero trade flows (Egger & Larch, 2011, Yotov et al., 2016), although in this research the dataset doesn’t contain those.
To conclude, this study uses a panel approach of the gravity model and incorporates several fixed effects to avoid possible biases. The most commonly used specification in the recent literature was suggested by Baier and Bergstrand (2007). It controls for fixed importer-time, exporter-time and country pair effects. Apart from the OLS, the model is estimated using the PPML estimator.
EMPIRICAL RESULTS Descriptive analysis
In Table 2, the descriptive statistics of the variables used in the research is presented. As the standard deviation is much larger than the mean, it is clear that trade and GDP variables have a great variation in values. This means that the dataset contains countries of various sizes by GDP, and there are country pairs with much and with very little trade. The smallest values of trade can be mainly found for country pairs consisting of small and distant countries, for example Malta-Moldova and Malta-Moldova-Luxembourg1. Looking at the bilateral distance it can be concluded that the dataset contains both neighbouring and very distant countries. The means of the indicator variables show that 6% of the country pairs share a border, 5% have a common official language, 1% have/had a colonial relationship after 1945. 36% of the observations include country pairs consisting of the EU members, 1% – the country pairs in the DCFTA relationship, and for 0.4% of observations unilateral abolition of tariffs within the framework of the ATP takes place.
Table 2. Descriptive Statistics
Variable* Obs Mean Std.Dev. Min Max
Bilateral trade 14466 5316.3 20422.51 0.0002 480688.7
GDP of origin 14812 1.43e+12 3.11e+12 7.75e+09 2.05e+13
GDP of destination 14812 1.43e+12 3.11e+12 7.75e+09 2.05e+13 Bilateral distance 14812 3949.904 3612.436 6.686 18549.61 Contiguity 14812 .062 .242 0 1 Common language 14812 .045 .208 0 1 Common history 14812 .012 .11 0 1 EU 14812 .367 .482 0 1 ATP 14812 .004 .061 0 1 DCFTA 14812 .011 .106 0 1
* The units of measurement discussed in Data
1
20
All non-indicator variables were tested for normality using the skewness and kurtosis measures. The results are presented in Table 3. For a normal distribution skewness should be equal to 0 and kurtosis – equal to 3. The distribution of the variables is clearly far from normal. Since none of the variables take negative values, a log transformation can be performed to address the issue. Table 3 shows that for the log-transformed variables the values of skewness and kurtosis are closer to the desired values, meaning that these variables are more normally distributed. At the same time, the skewness/kurtosis test for normality (Appendix B) with the Prob>chi2 values being equal to zero still rejects the null hypothesis of the variables being normally distributed.
Table 3. Skewness and kurtosis values before and after the log transformation
Variable Skewness Kurtosis
Bilateral trade 11.084 169.193
Log Bilateral trade -.527 3.807
GDP of origin 4.163 21.474
Log GDP of origin -.06 2.468
GDP of destination 4.163 21.474
Log GDP of destination -.06 2.468
Bilateral distance .995 3.017
Log Bilateral distance -.527 3.52
Besides, the test for heteroscedasticity was conducted. The Breusch-Pagan test (Figure 1) rejects the null hypothesis of homoscedasticity since the p-value is lower than 0.05. Hence, there is heteroscedasticity present, meaning that robust standard errors should be used for the estimations. In addition, following Head et al. (2010), Egger and Larch (2011) and Yotov et al. (2016), I cluster the error term by country pair, which allows the within-group correlation while keeping the assumption of zero correlation across groups. In this way, I aim to account for heteroscedasticity across country pairs.
At the same time, Santos Silva and Tenreyro (2006) suggest that because of the presence of heteroscedasticity OLS estimations can still be biased and inconsistent while estimating the gravity equation. Because of that, the PPML estimation, as recommended by the authors, is used in this research along with the OLS.
Figure 1. Breusch-Pagan/Cook-Weisberg test for heteroscedasticity
21
distance variables have a relatively strong correlation of -0.55. This is because EU countries are located in the same region and are rather close to each other. To check whether this is an issue I also estimate the preferred specification of the model without the EU dummy.
Table 4. Correlation matrix
B il ater al tra de GD P of origin GD P of de sti na ti on B il ater al dist anc e C onti guit y C omm on langua ge C omm on hist ory EU AT P DC F TA Dive rsion dumm y Bilateral trade 1.00 GDP of origin 0.27 1.00 GDP of destination 0.32 -0.02 1.00 Bilateral distance 0.00 0.27 0.27 1.00 Contiguity 0.26 -0.01 -0.01 -0.24 1.00 Common language 0.16 0.06 0.06 0.02 0.22 1.00 Common history 0.01 -0.01 -0.01 -0.05 0.15 0.18 1.00 EU -0.01 -0.19 -0.19 -0.55 0.10 -0.03 -0.03 1.00 ATP -0.01 -0.03 -0.02 -0.04 0.02 -0.01 -0.01 -0.05 1.00 DCFTA -0.02 -0.04 -0.04 -0.08 0.04 -0.02 -0.01 -0.08 -0.01 1.00 Diversion dummy -0.03 -0.15 0.02 -0.18 0.02 -0.01 -0.01 0.30 -0.04 -0.06 1.00 Regression results
Testing Hypothesis 1 using the OLS model
22
and the significance levels indicated in the table didn’t change compared to the previous specification.
Further, various fixed effects get included in the model. Column (3) provides results including fixed exporter-time and importer-time effects, which allow to control for multilateral resistance terms. Due to the inclusion of the fixed effects, GDP variables get omitted. Bilateral distance coefficient suggests a larger negative relationship between the distance and trade than the previous two specifications. At the same time, the effects of common border and EU membership of the countries are considerably smaller in this model. The ATP dummy coefficient changed the sign to negative and became insignificant. The DCFTA dummy coefficient decreased and became insignificant. Column (4) provides the empirical results including exporter-importer fixed effects, which capture unobserved heterogeneity across the country pairs. The variables that don’t change for the country pairs over the period – such as distance between the countries, common border, common language and common history – drop out. The EU membership coefficient didn’t drop out in this specification, because in 2013 Croatia joined the EU. It can be seen that the coefficient of the GDP of origin shrank significantly in comparison to that in Columns (1) and (2) (from about 1.04 to about 0.24), while the coefficient for the GDP of the destination country only decreased a bit (from 0.89 to about 0.75). Thus, this specification suggests the higher impact of the GDP of destination on bilateral trade, while the models in Columns (1) and (2) show that the GDP of origin is relatively more important. The coefficients of the ATP and the DCFTA are insignificant in this specification. Finally, in Column (5) the fixed effects are included to control for both multilateral resistance terms and heterogeneity across the country pairs. This specification provides positive and significant coefficients for the ATP and the DCFTA. Since the model with the higher R-squared is preferred, the preferred specification is the one with fixed exporter-time, importer-time and exporter-importer effects. In this model we can see that the DCFTA coefficient 0.297 is significant at 1% level. This suggests that the DCFTA increased bilateral trade by 35% (𝑒0.297 = 1.35). In addition, the positive and significant (at 10% level) ATP coefficient means that exports from Ukraine to the EU increased by 25% (𝑒0.220 = 1.25) in the presence of the ATP. This specification was also estimated without the EU dummy (Appendix C) and the results show that in this case the correlation between the EU dummy and bilateral distance doesn’t have a noteworthy impact on the variable of interest.
23
Table 5. Gravity model estimates of the DCFTA trade effects using OLS estimation
(1) (2) (3) (4) (5)
VARIABLES OLS OLS OLS OLS OLS
ln GDPi 1.038*** 1.038*** - 0.238*** - (0.0186) (0.0186) (0.0599) ln GDPj 0.893*** 0.893*** - 0.752*** - (0.0177) (0.0178) (0.0548) ln DISTij -0.963*** -0.960*** -1.364*** - - (0.0395) (0.0395) (0.0583) CONTIGij 0.927*** 0.926*** 0.430*** - - (0.128) (0.128) (0.125) LANGij 0.265* 0.268* 0.257* - - (0.158) (0.158) (0.150) HISTij 0.491 0.495 0.461 - - (0.349) (0.349) (0.400) EUij 0.621*** 0.629*** 0.270** 0.218*** 0.0201 (0.0622) (0.0626) (0.110) (0.0673) (0.129) ATPij 0.528** -0.0758 -0.0179 0.220* (0.205) (0.392) (0.0599) (0.127) DCFTAij 0.519*** 0.524*** 0.142 0.0358 0.297*** (0.151) (0.152) (0.247) (0.0454) (0.0886) Constant -37.94*** -38.00*** 16.76*** -20.22*** 6.196*** (0.714) (0.718) (0.479) (2.230) (0.0467) Observations 14,466 14,466 14,466 14,466 14,466 R-squared 0.761 0.761 0.855 0.980 0.983
Year FE YES YES NO YES NO
Exporter-time FE NO NO YES NO YES
Importer-time FE NO NO YES NO YES
Exporter-Importer FE NO NO NO YES YES
The dependent variable is the (natural log of the) bilateral trade flow from i to j. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Notably, Baier and Bergstrand (2007) concluded that the use of panel data with fixed country-time and country-pair effects, as well as differenced panel data with country-time effects, provides “the most plausible estimates of the average effect of an FTA on a bilateral trade flow”. Similarly, Baldwin and Taglioni (2007) tested the models with various fixed effects and preferred the model with time-invariant country pair and time-varying nation dummies for the gravity equation estimation. Thus, the preferred specification in this study is in line with the findings of researchers. Testing Hypothesis 1 using the PPML model
24
distance, contiguity, common history and EU membership are significant and have expected signs in these specifications. The common language coefficients are positive, but insignificant. The ATP coefficient in Column (2) is negative, in contrast to the expectations, but insignificant. Notably, the DCFTA coefficients in these two estimations have positive signs as expected, but are insignificant. However, when I include fixed exporter-time and importer-time effects (Column (3)), or fixed exporter-importer effects (Column (4)), or all together (Column (5)), the DCFTA coefficient becomes significant while remaining positive. The ATP coefficient is also positive when additional fixed effects are included, but is only significant in Column (4) taking the value of 0.08. The EU coefficient is positive and significant in Columns (1) to (4), but becomes negative and insignificant when all the fixed effects are included. According to the PPML estimations the effects of GDP of origin and GDP of destination on trade are very similar.
Table 6. Gravity model estimates of the DCFTA trade effects using PPML estimation
(1) (2) (3) (4) (5) VARIABLES PPML PPML PPML PPML PPML ln GDPi 0.788*** 0.788*** - 0.552*** - (0.0324) (0.0325) (0.0530) ln GDPj 0.806*** 0.806*** - 0.488*** - (0.0388) (0.0388) (0.0659) ln DISTij -0.546*** -0.547*** -0.596*** - - (0.0553) (0.0553) (0.0397) CONTIGij 0.741*** 0.741*** 0.597*** - - (0.194) (0.194) (0.0941) LANGij 0.0214 0.0211 0.148 - - (0.143) (0.143) (0.108) HISTij 0.400** 0.400** 0.375* - - (0.170) (0.170) (0.222) EUij 0.248** 0.247** 0.732*** 0.254*** -0.0598 (0.107) (0.107) (0.126) (0.0302) (0.200) ATPij -0.272 0.0466 0.0771*** 0.105 (0.203) (0.365) (0.0289) (0.0703) DCFTAij 0.0380 0.0370 0.445** 0.126*** 0.307*** (0.174) (0.175) (0.223) (0.0395) (0.102) Constant -31.32*** -31.30*** 14.29*** -19.24*** 10.38*** (1.738) (1.742) (0.344) (2.375) (0.0677) Observations 14,466 14,466 14,466 14,466 14,466 Pseudo R-squared 0.848 0.848 0.918 0.996 0.997
Year FE YES YES NO YES NO
Exporter-time FE NO NO YES NO YES
Importer-time FE NO NO YES NO YES
Exporter-Importer FE NO NO NO YES YES
25
According to the Pseudo R-squared, models in Columns (3), (4) and (5) have a better fit, than models in Columns (1) and (2). The preferred model is the one with exporter-time, importer-time and exporter-importer fixed effects included (Column (5)). In the preferred specification, the ATP coefficient is not significant at 10% level, and the DCFTA coefficient is positive and highly significant. The presence of the DCFTA is estimated to increase bilateral trade by 36% (𝑒0.307 = 1.36). This is similar to the result of the preferred specification of the OLS estimation, suggesting that for the particular case the OLS also provides plausible results.
The models with fixed exporter-time, importer-time and exporter-importer effects are preferred in both the OLS and the PPML estimations. In conclusion, the preferred specifications of the OLS and the PPML estimations suggest that the implementation of the DCFTA increased trade between Ukraine and the EU member states by 35-36%, therefore supporting the hypothesis about the positive impact of the agreement on trade. This is comparable to the results of the research on the Europe Agreements by Egger and Larch (2011). They estimated that the Europe Agreements led to about 30% increase in bilateral trade between the Central and Eastern European countries and the EU15. On the other hand, the effect of the ATP on trade while being positive and significant in the preferred OLS specification, is insignificant in the PPML one, which doesn’t allow to conclude on robust significant impact of the ATP on trade.
Testing Hypothesis 2 using the OLS and the PPML models
In Table 7 results of trade diversion estimation are presented. Column (1) uses the OLS estimation without extra fixed effects apart from time fixed effects. The DCFTA coefficient is positive and significant, suggesting positive impact of the DCFTA on trade. Trade diversion coefficient is also positive, meaning that the DCFTA also led to an increase in trade of the countries involved with the third countries. Column (2) adds country pair fixed effects to the specification. The DCFTA coefficient, while still being positive, becomes insignificant. Trade diversion coefficient is positive and significant.
26
Looking at the R-squared values, it can be concluded that the preferred specification is the one that controls for the country pair fixed effects in both the OLS and the PPML estimations. The DCFTA coefficients in these specifications are positive, suggesting that the DCFTA induced trade between the countries involved. Nevertheless, in the OLS estimation, the DCFTA coefficient is not significant. The diversion coefficients are positive and significant meaning that the agreement also led to more trade of the signatories with the rest of the world. Consequently, based on the interpretation of trade creation and trade diversion of Magee (2008), the results support the Hypothesis that the DCFTA didn’t result in trade diversion. However, since the DCFTA coefficient is insignificant in the preferred specification of the OLS, the results of these estimations don’t provide robust evidence of trade creation.
Table 7. Gravity model estimates of the DCFTA trade effects including diversion
(1) (2) (3) (4)
VARIABLES OLS OLS PPML PPML
ln GDPi 1.046*** 0.214*** 0.783*** 0.590*** (0.0188) (0.0599) (0.0326) (0.0538) ln GDPj 0.889*** 0.749*** 0.806*** 0.525*** (0.0177) (0.0549) (0.0387) (0.0631) ln DISTij -0.956*** - -0.547*** - (0.0393) (0.0550) CONTIGij 0.931*** - 0.737*** - (0.128) (0.192) LANGij 0.269* - 0.0215 - (0.158) (0.143) HISTij 0.500 - 0.395** - (0.350) (0.171) EUij 0.568*** 0.207*** 0.278*** 0.233*** (0.0618) (0.0671) (0.105) (0.0301) ATPij 0.513** -0.0251 -0.275 0.0926*** (0.204) (0.0599) (0.203) (0.0289) DCFTAij 0.667*** 0.0653 -0.0149 0.183*** (0.158) (0.0476) (0.177) (0.0424) DIVERSION 0.259*** 0.0553*** -0.107* 0.0897*** (0.0536) (0.0208) (0.0599) (0.0169) Constant -38.14*** -19.52*** -31.17*** -21.39*** (0.719) (2.250) (1.713) (2.240) Observations 14,466 14,466 14,466 14,466 R-squared 0.762 0.980 0.848 0.996
Year FE YES YES YES YES
Exporter-Importer FE NO YES NO YES
27
Testing Hypothesis 3 using the OLS and the PPML models
The effect of the DCFTA on trade estimated above shows the average impact on the country pairs involved. It can also be interpreted as an impact on the trade of Ukraine with the EU countries, because the DCFTA takes place between Ukraine-EU28 and EU28-Ukraine country pairs. Next, I estimate the impact of the DCFTA on the trade with Ukraine of each of the EU member states separately. As concluded earlier, the preferred specification of the gravity model includes exporter-time, importer-time and exporter-importer fixed effects, and therefore, this specification is used for the estimations. The fixed effects mentioned, capture the effects of countries’ GDP, bilateral distance, common border, common language and common history, so these variables get dropped out from the regression.
Figures 2 and 3 present the results of the OLS and the PPML estimations in the percentage change in bilateral trade due to the DCFTA in place between the countries. The regression coefficients are provided in Appendix D. Looking at the figures, it is immediately clear which countries gained more than average, which countries gained less and for which countries the effect of the DCFTA on trade was negative.
28
Figure 2. Effect of the DCFTA on individual countries using OLS estimation
(% change in trade flows)
The dashed line shows the average effect estimated. *** p<0.01, ** p<0.05, * p<0.1
Figure 3 presents the results of the PPML estimation. There are 21 countries with significant results, and the level of significance changed for some countries compared to the OLS estimation. Cyprus remains the only country with a significantly negative coefficient, and the PPML estimation suggests that the trade of Cyprus with Ukraine decreased by 80%. The countries that experienced much higher than average increase in trade with Ukraine are Denmark (66%), Latvia (68%), Luxembourg (68%), Netherlands (67%) and Sweden (87%). The results for Malta show more than 7 times increase in trade with Ukraine due to the DCFTA, and, in contrast to the OLS estimation, the coefficient is significant. Overall, it can be concluded that the Hypothesis about the heterogeneous effects of the DCFTA on the signatories is supported.
29
Figure 3. Effect of the DCFTA on individual countries using PPML estimation
(% change in trade flows)
The dashed line shows the average effect estimated. *** p<0.01, ** p<0.05, * p<0.1 DISCUSSION
Discussion and recommendations for future research
In the previous section, the results for each hypothesis were presented. Two estimation methods with various fixed effects were used throughout the study. First of all, a standard OLS model with year fixed effects was used. Furthermore, fixed exporter-time and importer-time effects were added to account for multilateral resistance terms. Lastly, fixed exporter-importer effects were incorporated in the model to rule out possible omitted variable bias. Apart from the OLS model, the PPML estimation was done employing the same fixed effects. The PPML model is recommended for the gravity equation estimation, because, according to Santos Silva and Tenreyro (2006), it is better in dealing with heteroscedasticity of trade data. It was concluded that the preferred model specification includes fixed exporter-time, importer-time and exporter-importer effects.The results are considered robust if preferred specifications of both the OLS and the PPML estimations are statistically significant with the same sign.
30
Before the results of the study are discussed, it is useful to cast a glance at the statistics of trade between the EU and Ukraine in recent years. Foremost, Figure 4 gives an idea of how trade between the EU and Ukraine changed over the period considered in the research. It can be seen that there was a great decrease in exports to Ukraine in 2014 and 2015. In 2015 the exports of the EU to Ukraine dropped by about 40% compared to 2012. Notably, the imports of the EU from Ukraine decreased only slightly in that period. The decrease in trade might be explained by an unstable political situation in Ukraine,namely the Ukrainian Revolution of (February) 2014 and the Russian military intervention to Ukraine that started in February 2014 and continues as of May 2020. Starting from 2016, both exports and imports started to grow substantially with similar pace. However, while the EU imports from Ukraine were about 22% higher in 2018 compared to 2012, the exports to Ukraine haven’t reached the level of 2012 by that moment.
Figure 4. Trade of the EU28 with Ukraine in 2012-2018 (Eurostat, 2020)
Additionally, to get a better idea about the structure of trade between Ukraine and the EU it is interesting to take a look at the change patterns of different trade items displayed in Figure 5. It can be seen that the EU mostly exports to Ukraine manufactured goods, while importing much food and raw materials. Talking about exports to Ukraine, it is clear that the exports of manufactured goods have suffered the most from the slump; the subsequent increase in exports was also mainly associated with those products. Turning to imports from Ukraine, the increase after 2016 comes from various product groups: manufactured goods, food and raw materials. On the other hand, imports of energy products from Ukraine shrank compared to 2012 and 2013. So it can be seen that after the EU experienced a dramatic decline in exports to Ukraine, it is getting the position at the market back. At the same time, the imports from Ukraine increased compared to 2012, which is attributed to both high and low value-added products.
31
Figure 5. Trade of the EU28 with Ukraine in 2012-2018 byproduct group (Eurostat, 2020)
This study was aimed to identify the effects of the Deep and Comprehensive Free Trade Area between the EU and Ukraine on trade. First of all, the trade between Ukraine and the EU is estimated to increase by 35-36% in result of the DCFTA according to the preferred specifications of the OLS and the PPML estimations. Thus, a part of the increase in trade between the EU and Ukraine from 2016 (Figure 4) can be explained by the agreement. This finding is in line with the main objective of the agreement to enhance trade between the countries. Notably, the estimated increase in trade is quantitatively similar to the one resulting from Europe Agreements as estimated by Egger and Larch (2011). As discussed in the literature review, not all trade agreements actually induce trade. According to Kohl (2014), only about a quarter of the agreements have positive effects on trade. Soete and Van Hove (2017) and Baier et al. (2019) estimated positive effects for slightly more than half of the agreements. The fact that the DCFTA is estimated to have positive effects on trade might be partially explained by the high level of economic development of the signatories and their WTO membership. Kohl et al. (2016) found that these factors have positive impact on institutional design and legal enforceability of agreements negotiated, which, in turn, leads to greater trade effects. Thus, the results show that the actual effects of the agreement are in line with the aim of the DCFTA to enhance the trade relations between the EU and Ukraine. These results can be used by policymakers to reflect on the effectiveness of the policy in regards to the extent by which the effects of the DCFTA on trade fulfil the objectives of the agreement. The first recommendation for the future research is to study whether the changes in trade due to the trade
32
liberalization have led to changes in productivity and welfare. For example, Egger and Larch (2011), found that an increase in trade between the EU15 and the CEEC due to the Europe Agreements generated positive GDP and welfare effects. The effects were larger for the CEEC than for the EU15. Regarding the effects of the Autonomous Trade Preferences, there is no robust evidence of their significant impact on trade. This might partially be explained by a low envisaged degree of integration. According to Soete and Van Hove (2017), Kohl and Trojanowska (2015), Baier et al. (2014) and Magee (2008), a greater envisaged degree of integration leads to larger effects from trade agreements.
Next, trade diversion effects resulting from the DCFTA were estimated. In the preferred specifications of both the OLS and the PPML models the diversion coefficients are positive and significant suggesting that the trade agreement didn’t result in a decrease in trade with the third countries. According to Viner (1950), this means that there is no evidence of efficiency losses associated with the agreement.
Finally, the effects of the DCFTA on each of the EU countries individually were estimated. The results confirm the hypothesis of the heterogeneous effects of the agreement on the signatories. While on average the DCFTA induced trade between the EU and Ukraine by 35-36%, the effect on different countries varies. The only country which, according to the estimations, experienced a decrease in trade resulting from the agreement is Cyprus. The significant coefficients for all the other countries were positive with the coefficient for Malta being exceptionally high, however only significant in the PPML estimation. According to both the OLS and the PPML estimations Denmark, Luxembourg, Netherlands and Sweden experienced more than 60% increase in trade with Ukraine resulting from the DCFTA. Although, identifying the sources of heterogeneity within the agreement is beyond the scope of this study, it makes sense to cast a glance at it. The possible reasons for heterogeneous effects of trade agreements, as discussed in the literature review, were investigated by Baier et al. (2019), Baier et al. (2018), Cheong et al. (2015b) and Vicard (2011). It was found that less distant countries experience greater increase in trade. In addition, studies show that economic size, income, common language, common colonial history, common religion and common legal origins also play a role. However, running the preferred model specification together with the interaction variables for the effect of distance and the effect of economic size didn’t provide any significant results2. Thus, the question about the sources of heterogeneity within this
2
33
agreement remains open, and the second recommendation for the future research is to tackle it. This also includes heterogeneity within country pairs, the importance of which was emphasised by Baier et al. (2019).Understanding why some countries gained more than others could help to manage and predict the gains from agreements in the future.
It is worth paying attention to the country that experienced a decrease in trade resulting from the DCFTA according to the estimations. Both the OLS and the PPML estimations show a significant decline in trade for Cyprus. In an attempt to find out what might explain this, I looked at the structure of trade by product group using data for both exports and imports between Cyprus and Ukraine provided in Appendix F. It can be noticed that in 2014-2015 exports of manufactures, chemicals, machinery and transport equipment from Cyprus to Ukraine (as well as imports from Cyprus to Ukraine) dropped sharply. On the other hand, a very dramatic decrease of exports from Ukraine to Cyprus is associated with fuel. The products experiencing significant declines in exports and imports took a great share of trade before 2015. The fact that the DCFTA didn’t raise trade between Cyprus and Ukraine might partially be explained by the nature of products traded, because, as a result of Russian military intervention in Eastern Ukraine in 2014, a historical coal mining and a heavily industrialised region of Ukraine ended up being under occupation that still takes place as of May 2020. Besides, as displayed in Figure 5, overall imports of energy products by the EU from Ukraine also decreased compared to 2012 and 2013. Hence, the third recommendation for the future research is to check how the structure of trade between countries influenced the effects of the agreement. Urata and Okabe (2014) used a dataset on 67 countries/regions in 1980-2006 and found that the effects of trade agreements generally differ across product categories. Bensassi et al. (2012) studied the effects of Euro-Mediterranean agreements on different product categories and partially confirmed the hypothesis that the effects of the agreements were different across sectors. Disaggregated trade data was also used by Lakatos and Nilsson (2017) to identify the effects of the EU-Korea trade agreement on products subject to liberalization in contrast to other products. Notably, the authors found that not only exports of liberalized goods increased, but also exports of goods already traded duty-free. Lakatos and Nilsson (2017) mention that this finding may suggest positive spill-over effects and positive effects of uncertainty reduction on the whole.
Limitations
