Evaluating Trade Patterns in the CIS Paper prepared for the first World Congress of Comparative Economics Rome, Italy, June 26, 2015 Yugo Konno, Ph. D. 1 Senior Economist, Mizuho Research Institute Ltd., Japan 1. Introduction According to traditional trade theory represented by Heckscher-Ohlin model, trade is determined by differences in factor endowments between countries, hence trade will occur in different products and industries. However, following the work of Grubel-Lloyd (1975) which revealed the existence of intra-industry trade (IIT), defined as the simultaneous exports and imports of goods which are grouped in the same industry, an abundant theoretical and empirical literature has evaluated this phenomenon. Theoretical literature on IIT has suggested that increased specialization, imperfect competition, and economies of scale lead to intra-industry trade being predominant between countries of similar structure and factor endowments, whereas inter-industry trade (INT) is determined by differences in factor endowments. Theoretical differences in the types of trade have stimulated a vast empirical research trying to identify the nature and the determinants of each type of trade. However, empirical research with this kind of approach on the CIS countries is extremely rare. Therefore, this paper aims to contribute a better understanding of recent trade developments in the CIS, focusing on Russia. In order to meet this aim, this paper is structured as follows. The next section outlines the trade pattern of Russia by decomposing the total trade into some types of trade. Section 3 empirically analyzes the determinants of each type of trade. Section 4 summarizes the main findings of this paper and presents future research issues of this study. 1 yugo.konno@mizuho-ri.co.jp The views expressed in the paper are the author's and do not necessarily reflect those of Mizuho 1 Research Institute.
2. Trade pattern of Russia (1) Three types of trade According to recent literature, intra-industry trade (IIT) is further divided into horizontal IIT (HIIT) and vertical IIT (VIIT). HIIT occurs when similar products are simultaneously exported and imported mainly due to product differentiation. VIIT is defined by Grubel-Lloyd (1975) as the simultaneous export and import of goods in the same industry but at different stages of production. VIIT results from vertical disintegration of production due to varying factor intensities within an industry. Thus, the bilateral trade flows can be broken down into the three types: horizontal intra-industry trade (HIIT), vertical intra-industry trade (VIIT), and inter-industry trade (INT). The most widely used method to decompose IIT into horizontal and vertical parts is based on ratio of unit values of exports to those of imports. However, this method has been criticized by the randomness in the choice of threshold ratio, which is used to determine whether IIT in an industry is vertical or horizontal. Moreover, by this method, it is difficult to capture the aforementioned VIIT as defined by Grubel-Lloyd (1975). Therefore, in this paper, a newer method proposed by Kandogan (2003) is employed. It uses values of exports and imports at two different levels of aggregation. The higher level of aggregation defines industries, and the lower level of aggregation defines different products in each industry. Trade in products (p) of an industry (i ) is aggregated to the trade in that industry: Equation (1). X i = X ii p M i = M ii p (1) Firstly, the total amount of IIT in each industry (IIT i ) is calculated from the Equation (2), using trade data at higher level of aggregation. As the equation shows, IIT i is the trade overlap - simultaneous import and export - of goods within an industry (i), which forms a part of the famous Grubel-Lloyd index. Then, the amount of trade overlap in each product (p) of an industry (i) is calculated, using trade data at lower level of aggregation: Equation (3). This is the trade of similar products, i.e. horizontal IIT (HIIT) in the industry (i). The remaining part of the IIT in the industry (i) is the vertical IIT (VIIT i ), which consists of trade of different products and/or products at different stages of production within the industry: Equation (4). The amount of inter-industry trade in an industry (INT i ) can be obtained by subtracting IIT i from the total trade in the industry (X i + M i ): Equation (5). 2
IIT i = (X i + M i )- X i -M i (2) HIIT i = X ii + M ii X ii M ii p (3) VIIT i = IIT i -HIIT i (4) INT i = (X i + M i )-IIT i (5) In order to perform calculations based on the above equations, trade data is obtained from the United Nations COMTRADE database. In this paper, data at 1-digit level of SITC (Standard International Trade Classification) Rev. 3 is used to define industries, and that at 4-digit level is used to define products 2. Using these data, Russia s bilateral trade with eight major trading partners (Germany, Italy, USA, China, Japan, Ukraine, Belarus, and Kazakhstan) is analyzed for the period 2000-2013 3. (2) Trade pattern of Russia Figure 1 presents trends in values of inter-industry, horizontal and vertical IIT of Russia with the eight major trading partners. It clearly shows that inter-industry trade (INT) has made up the bulk of Russia s trade flow. In contrast to this, intra-industry trade, especially horizontal one, has been much inferior to INT in values. Horizontal and vertical IIT as an average share in Russia s total trade in the period is 9.9% and 20.1%, respectively. Looking at trends in each type of trade, values of all the three types of trade significantly increased for almost all years in the period, especially up to 2008 4. The annual average growth rate for each type of trade in 2000-2013 was INT: 15.7%, VIIT: 12.0%, HIIT: 9.7%. 2 The SITC Rev. 3 has the basic 10-section structure at 1-digit level. The sections are subdivided into 67 divisions at 2-digit level, 261 groups at 3-digit level, 1,033 groups at 4-digit level, and 3,121 headings at 5-digit level. The number of industries analyzed in this paper is nine (SITC 0-8) because SITC 9: commodities and transactions not classified elsewhere in the SITC, is omitted from the analysis. 3 According to the database, in 2000-2013, the values of Russia s exports and imports to/from the eight countries as a share of Russia s total exports and imports averaged 35.4% and 54.1%, respectively. It should be noted however, that data on Russia in the database has the following two shortcomings. Firstly, it lacks data reported by Russia on trade with Belarus in 2000-2011 and with Kazakhstan in 2010-2011. These lacked data are substituted by data reported by the trading partners. Secondly, Russia s export of natural gas in the gaseous state (SITC 3432) to Germany and Italy are not reported. 4 The value of Russia s inter-industry trade increased at an annual average of 23.6% in 2000-2008. 3
Trends in values of each type of trade, measured at the industry level, are presented in Figure 2. The proportion of INT is extremely large in Fuels, Machinery and transport equipment, and Misc. manufactured articles. On the other hand, in Manufactured goods, Chemicals, and Food & live animals, intra-industry trade, especially vertical IIT accounts for a relatively large portion of the total trade. In this connection, Kandogan (2003) finds that in CEE (central and east European countries), IIT, especially HIIT, is relatively active in industries where there is significant production differentiation such as manufacturing, machinery and transport equipment sector, while IIT is stagnant in industries with standardized products such as natural resources. Thus, Russia s trade flows are in marked contrast to those of CEE on the point that IIT in machinery and transport equipment sector is stagnant in Russia, while it is considerably active in CEE 5. Figure 3 presents trends in each type of trade compiled into two groups of trading partners: CIS-3 (Ukraine, Belarus, and Kazakhstan) and the other five countries (Germany, Italy, USA, China, Japan). It shows clearly that VIIT and HIIT account for a relatively large portion in trade with CIS-3 compared to those in trade with the other countries. Figure 1. Trends in Russia s international trade by type (bil. USD) Source: Author s calculations based on United Nations COMTRADE database 5 In CEE, nearly the one third of the trade flow in machinery & transport equipment sector was horizontal intra-industry trade at the end of the 1990s (Kandogan, 2003). 4
Figure 2. Trends in Russia s trade by type in each industry Note: Unit is billion USD. Source: Author s calculations. 5
Figure 3. Trends in Russia s trade by type with different partners Note: Unit is billion USD. Source: Author s calculation. 3. Determinants of trade pattern of Russia (1) Empirical methodology In this section, a gravity-type model is estimated in order to identify the main determinants of Russia s trade flows with the eight major partner countries in the period 2000-2013. Many prior empirical studies on IIT show that each type of trade: HIIT, VIIT, and INT, has different determinants. Based on these findings, in this paper, the determinants of Russia s each type of trade are investigated separately. Accordingly, dependent variables take the logarithm of each type of trade which are subdivided into partner (eight) industry (nine) pairs in the period 2000-2013, resulting in 1,008 observations. Independent variables include, first of all, the elements of the most basic gravity model: the size of the Russian economy in the year t (lngdp it ), the corresponding economic size of the trading partner j (lngdp jt ), and the geographic distance between Russia and each of its trading partners (lndist ij ). Other independent variables are also included in the model: the absolute difference in per capita GDP between Russia and the partner country (lnpcgdp ijt ), the inward foreign direct investment stock in Russia (lnfdi it ), and a dummy variable which takes one when the partner country is a member of the CIS (CIS). To estimate the gravity-type model, panel regression method is used. The full estimated model in this paper is therefore: lntradeijt = β0 +β1 lngdpit +β2 lngdpjt +β3 lndistij +β4 lnpcgdpijt +β5 lnfdiit +β6 CIS +μi j +νijt (6) 6
where TRADE takes any one of HTTI, VIIT, and INT, while μ ij effect and error term, respectively. and ν ijt denote the individual Equation (6) is estimated using panel data. Data sources of dependent variables are as follows. GDP and per capita GDP are obtained from the WDI (World Development Indicators) database published by the World Bank. The data of distance and FDI are obtained from the database of CEPII and that of UNCTAD, respectively. Expected signs of the independent variables and the reasons for those are as follows. Countries are expected to trade more if their economic size is larger, and the distance between them is smaller. Therefore, both β 1 (lngdp it ) and β 2 (lngdp jt ) are expected to be positive, while β 3 (lndist ij ) is expected to be negative, for all types of trade. Regarding dissimilarity in per capita GDP (lnpcgdp ijt ), different sign for each type of trade is expected. The Heckscher-Ohlin trade theory explains trade (mainly inter-industry trade) by differences in relative factor endowments, which are often measured by dissimilarities in per capita GDP. Therefore, a positive sign is expected for INT. Contrary to this, a negative sign is expected for HIIT and for VIIT to a lesser degree, since the intensity of intra-industry trade will be highest when the two trading countries are identical both for economic development and market size (Andresen, 2003). Foreign direct investment (lnfdi it ) is expected to have either positive or negative effect on trade. For example, investing in production facilities abroad by multinational enterprises may create the possibility to exchange products at different stages of production, thereby contributing to IIT, particularly VIIT. However, some studies associate FDI as a substitute for international trade since firms may serve foreign markets directly rather than through trade, which may lead to a decrease in trade. Therefore, the sign for this variable is unclear ex ante. The CIS dummy contains mainly two kinds of determinants of trade, concurrently. One is the existence of Free Trade Agreements (FTA) in the region. Russia and the CIS-3 (Ukraine, Belarus, and Kazakhstan) have tariff-free access to each other under various bilateral FTA after the middle of the 1990s. FTA, which lower the transaction cost, is expected to promote trade between the countries, in general, and intra-industry trade, in particular. The other kind of determinants of trade contained in this dummy is an existence of a common border. Two countries sharing a border will have a lower transport cost, and therefore, have a greater trade. Russia shares a border with all the CIS-3 countries. Above all, the expected sign for this dummy variable is positive for all types of trade. 7
(b) Results Table 1 reports the results of model specification tests and the estimation results. The results of the model specification tests show that the variance of the individual effect (μ ij ) is statistically significantly different from zero (the Breusch-Pagan test), and that a significant correlation between the individual effect and the independent variable cannot be detected (the Hausman test) in all cases. These results lead us to conclude that the random-effects model is more appropriate than either of the pooled-ols and the fixed-effects model. The usual random-effects estimator, however, assumes a homoscedastic error structure. Therefore, tests for the existence of panel level heteroscedasticity were conducted and the results rejected the hypothesis of homoscedasticity in all cases. Accordingly, in order to obtain consistent and efficient estimators, the model was estimated by Feasible GLS (FGLS), correcting for heteroscedasticity. The points to be drawn from the estimation results are as follows. Regarding the independent variables comprising the most basic gravity model, both coefficients on Russia s GDP (lngdp it ) and on trading partners GDP (lngdp jt ) are significant and positive, while the coefficients on distance (lndist ij ) are significant and negative in all cases, as was expected. Additionally, it is seen that Russia s GDP has a relatively larger effect on INT, while GDP of trading partners and the distance have a greater impact on IIT (HIIT and VIIT). As for the influence of dissimilarity in per capita GDP (lnpcgdp ijt ) on Russia s trade flows, coefficients on this variable are significant and negative for all types of trade, particularly for VIIT. The result on HIIT is in line with the expectation, suggesting that a similarity in the level of economic development between Russia and the trading partners has a positive impact on Russia s HIIT. Nevertheless, the results on the other two independent variables defy the prediction: the coefficient for INT is negative, and coefficient for VIIT is much lower than that for HIIT. Inward FDI to Russia (lnfdi it ) is estimated to be insignificant for either of the three types of trade. The insignificance of this variable can be due to a shortcoming of data: this independent variable is based on the total amount of FDI which Russia received from all over the world, while the independent variables are based on the bilateral trade flows of Russia. This issue is a future research task of this study. 8
Table 1. Results of model specification tests and the estimation results Panel data - FGLS Dependent variable HIIT VIIT INT Independent variable lngdp it 0.556 *** 0.524 ** 1.046 *** (4.15) (2.44) (7.04) lngdp jt 0.961 *** 1.014 *** 0.322 *** (9.64) (8.77) (4.10) lndist ij -1.661 *** -1.424 *** -0.633 *** (-15.54) (-10.68) (-7.75) lnpcgdp ijt -0.337 *** -0.986 *** -0.373 *** (-4.69) (-9.48) (-5.82) lnfdi it -0.025 0.087-0.070 (-0.26) (0.53) (-0.62) CIS 3.400 *** 1.400 *** -0.224 (12.77) (4.52) (-1.01) constant 8.052 *** 13.022 *** 5.552 *** (8.29) (9.56) (6.43) Number of obs. 1008 1008 1008 Number of groups 72 72 72 Wald (chi 2 ) 3255.78 *** 606.09 *** 584.16 *** (All variables-0 except const.) Breusch-Pagan test 4448.28 *** 3060.65 *** 4908.60 *** (random effect / pooled-ols model) Hausman test 2.76 2.91 1.12 (fixed effect / random effect model) Panel level heteroskedasticity test 1640.09 *** 1756.54 *** 1035.49 *** Notes: Figures in parentheses report z-statistics. ***: Significant at the 1% level; **: Significant at the 5% level; *: Significant at the 10% level. Panel level heteroskedasticity test is the likelihood-ratio test, the null hypothesis of which is that the error variance was homoscedastic (Poi and Wiggins, 2001). Source: Author s estimation 9
Concerning the CIS dummy, it is estimated to be significant and positive for IIT, especially for HIIT. It may indicate that both the existence of FTA and a common border between Russia and CIS-3 are very essential conditions for Russia s HIIT, though it is not clear which of those is more essential to that trade flow. On the other hand, the estimated coefficient for INT is found to be negative and statistically insignificant. It suggests that neither the existence of FTA nor a common border significantly affects Russia s INT with Ukraine, Belarus, and Kazakhstan. 4. Concluding remarks The main findings of this paper can be summarized as follows: 1. This paper analyzed Russia s trade pattern with eight major trading partners (Germany, Italy, USA, China, Japan, Ukraine, Belarus, and Kazakhstan) for the period 2000-2013, by decomposing total trade into three parts: inter-industry trade (INT), vertical intra-industry trade (VIIT), and horizontal intra-industry trade (HIIT). It was found that Russia s HIIT increased nearly ten percent annually in the period, though INT was surpassing both in growth rate and share in the total trade. It was also found that intra-industry trade (VIIT and HIIT) accounts for a relatively large portion in trade with the CIS countries. 2. Then, a gravity-type model was estimated in order to identify the main determinants of each type of trade. The estimation results shows that both the economic size of Russia and the trading partner have positive effects on all types of trade, while the effects of distance are negative. The estimation results also suggest that dissimilarity in per capita GDP between Russia and a trading partner has a negative impact on HIIT, and that the existence of FTA and a common border lead to an increase in HIIT and VIIT. All of these are in line with the previous expectation. 3. However, the estimation results failed to meet the expectation in some cases. A similarity in the level of economic development is suggested to have a much larger positive impact on VIIT than on HIIT. The similarity is also suggested to have a positive impact on Russia s INT. Both of these are contrary to the expectation. Moreover, relationship is ambiguous between inward FDI and all the three types of trade. The unsuccessful part of the estimation results can be related to the following shortcomings of this paper. Firstly, regarding independent variables, the level of trade data aggregation used to define industries, i.e., one-digit level of SITC, may be too high. This high level of aggregation may lead to an overvaluation of VIIT and an undervaluation of INT. Secondly, the estimated model should include a wider variety of dependent variables. All of these are future research issues of this study. 10
References Andresen, A. Martin (2003) Empirical intra-industry trade: what we know and what we need to know [http://www.sfu.ca/~andresen/papers/empirical_iit_lit_review.pdf]. Caetano, J. and Galego, A. (2007) In search for the determinants of intra-industry trade within an enlarged Europe, South-Eastern Europe Journal of Economics 2, pp. 163-183. Grubel, H. and Lloyd, P. (1975) Intra-industry trade: the theory and measurement of international trade in differentiated products, McMillan: London. Iwasaki, Ichiro and Suganuma, Keiko (2013) A gravity model of Russian trade: the role of foreign direct investment and socio-cultural similarity, Russian Research Center Working Paper Series No. 40. Kandogan, Yener (2003) Intra-industry trade of transition countries: trends and determinants, William Davidson Institute Working Paper No. 566 [http://deepblue.lib.umich.edu/bitstream/handle/2027.42/39952/wp566.pdf?sequence=3]. Poi, B. and Wiggins, V. (2001) STATA: How do I test for panel-level heteroskedasticity and autocorrelation? [http://www.stata.com/support/faqs/statistics/panel-level-heteroskedasti city-and-autocorrelation/]. 11