Export Diversification and Output Volatility - Empirical Evidence from Central and Eastern European Countries

Transcription

1 Export Diversification and Output Volatility - Empirical Evidence from Central and Eastern European Countries Bahar Kartalciklar University of Southern California Job Market Paper: November 2015 Abstract This paper investigates the effects of trade liberalization on output volatility. Traditional comparative advantage theory such as Heckscher-Ohlin and the New Trade Theory pioneered by Krugman and Melitz are at odds when predicting the effects of trade liberalization on export diversification. The traditional view is that trade openness results in greater trade specialization and generates output volatility by exposing countries to external shocks. The new trade theory on the other hand implies that trade liberalization leads to growth in the extensive margin of exports. This expansion in the extensive export margin leads to greater export diversification. More diversified markets mean that with more trade liberalization; countries are less exposed to external shocks. Matching highly disaggregated export data with aggregate- and industry-level production data of Central and Eastern European Countries that joined the EU in 2004, I revisit the openness and volatility link with a particular focus on the effects of the extensive margin of exports. The entry of the Central and Eastern European Countries into trade relationships with the West after the end of the Cold War represents a unique natural experiment. My statistical analysis shows that although more openness increases volatility, this effect can be stabilized through sufficiently diversified export baskets. I find that trade liberalization episodes result in major growth of the extensive margin of exports. I demonstrate that this export expansion has a consistent and significantly negative effect on both per-capita GDP and sectoral output volatility. I also find that the geographical diversification of exports is a more dominant factor in reducing output volatility, than increase in the product variety. 1

2 1 Introduction Output volatility is one of the main factors that slow down economic growth (Ramey and Ramey, 1995; Aizenman and Marion, 1999). Although trade openness is known to foster economic growth (Jeffrey A. Frankel, 1999; Calderon et al., 2005), the relationship between trade openness and output volatility is a topic with little consensus. A widely held view is that greater openness to trade leads to higher GDP volatility by exposing countries to external shocks. According to this view, openness results in higher volatility through two separate channels. First, higher openness is associated with higher specialization, which leaves countries vulnerable to sector-specific shocks that result in greater termsof-trade volatility. If the sectors that the country specializes in are more prone to terms-oftrade fluctuations, greater openness will bring about greater output volatility, hurting economic growth and welfare (Jansen, 2004; Bejan, 2006; di Giovanni and Levchenko, 2009). The second channel is through increased exposure to country-specific (foreign demand) shocks. Openness to trade increases output volatility by transmitting demand shocks to other countries (Bacchetta et al., 2009). Traditional comparative advantage theory suggests trade liberalization and greater openness lead to specialization in production and exports of a country. In other words, given that the openness of a country is measured as the ratio of total trade to the GDP, increased openness mainly comes from the growth in the intensive margin of trade (i.e., the growth in the trade of goods that have been previously traded). When all the growth in total trade is assumed to originate from the intensive margin, as it does in the traditional models, terms-of-trade fluctuations appear to be the major source through which openness affects output volatility. Recently, there has been vast development in a newer trade theory where firms face fixed costs of exporting (Melitz, 2003; Anderson and Wincoop, 2004; Helpman et al., 2008). In this newer trade theory, higher openness does not necessarily increase specialization of exports. Many empirical studies, which are based on the new trade models with monopolistic competition that incorporate fixed trade costs and sunk market-penetration costs, show the extensive margin of trade (i.e., trade in the varieties that have not been traded before) plays as important a role as the intensive margin of trade in contributing to trade growth (Hummels and Klenow, 2

3 2005; Broda and Weinstein, 2006; Kehoe and Ruhl, 2013). Hummels and Klenow (2005) compare the implications of Armington model based on a more traditional theory, which has no extensive margin, with the implications of Krugman s monopolistic competition model in which each country produces an endogenous number of varieties. They find that 60% of the greater exports of larger economies is due to the extensive margin. Kehoe and Ruhl (2013) show a significant growth in the extensive margin characterizes trade-liberalization episodes and periods of structural change. What is more, they find the extensive margin contributes significantly to trade growth during episodes of rapid economic growth and development. 1 It is also shown when there are market-penetration costs, the response of firms in changing their exporter status to temporary distortions such as business cycles is weak, which suggests the extensive margin of exports does not co-move with the temporary fluctuations in the economy. 2 The main purpose of this paper is to contribute to this line of research by studying the link between export diversification and output volatility, both at the aggregate and sectoral levels. To this purpose, I use two separate panel data sets on the trade and production of eight Central and Eastern European countries (CEECs), that joined the European Union in The first data set includes disaggregated bilateral trade data along with GDP data, covering the period Looking at aggregate volatility may mask the real impact of the extensive margin because the correlation between the traded and non-traded sectors in the economy would also appear in the aggregate volatility. In the sectoral data set I match the export data to the production data of the traded sectors at the 6-digit product-level. This allows me examine the immediate effect of a change in the extensive margin on the volatility of the production. I begin my analysis by constructing an extensive margin measure, which evaluates the variety of the export basket of a country depending on its weight in the world trade. I construct this measure using highly disaggregated bilateral trade data specifically, disaggregated at the 6-digit level of Harmonised System (HS) Classification for the period Second, using panel regressions, I investigate the relationship between export diversification in terms of product-market pairs and per-capita GDP volatility, while controlling for open- 1 Kehoe and Ruhl (2013) show that the share of the least traded goods in Korean exports to the United States went from 10% to 60% during Currently, a similar pattern is present in Chinese exports. 2 Ruhl (2008); Alessandria and Choi (2007) 3

4 ness and terms-of-trade volatility, along with other sources of income volatility that have been established in the literature. Next, I decompose the impact of diversification into its product and market components by employing product and market diversification measures one at a time in my regressions. In addition to aggregate volatility, I also examine the link between export diversification and output volatility at the sectoral level. To this purpose, I combine the trade data with the sectoral output data at the 6-digit product level. Then, through a set of panel regressions, I explore the link between sectoral export diversification and sectoral output volatility, while controlling for export openness, productivity and government funding levels (of each sector). My findings show that the diversification of exports in terms of product-market pairs plays a significant role in reducing output volatility, both at the aggregate and sectoral levels. At the same time, openness to trade and terms-of-trade fluctuations raise output volatility as predicted in the existing literature. When per-capita GDP volatility is considered, the export diversification measure consistently enters the regressions with a negative and highly significant coefficient. This result is robust to different estimation techniques, though my preferred model is the instrumental variables (IV) regression that accounts for the endogeneity between trade openness and output volatility. The estimates of the benchmark IV regression suggest that for the average country, a 10% increase in the extensive margin of exports leads to 4.6% decline in per-capita GDP volatility. Furthermore, using the aggregate data set, I also show diversification in terms of destination markets plays a more important role in reducing aggregate volatility compared to diversification due to different products, whose effect, while reducing volatility, is insignificant. When I study the sectoral data set, I find the extensive margin of exports remains a significant factor in reducing sectoral output volatility. Although the stabilizing effect is weaker at the sectoral level, the extensive margin measure has a coefficient that is consistently negative and significant in various estimation techniques. I present additional evidence showing the stabilizing effect of diversification is strongest during trade-liberalization episodes, both at the aggregate and sectoral levels. When I control for the accession of the CEECs to the EU, I find that even though the effect of the extensive margin is ambiguous prior to the EU accession, it becomes stronger and negative afterward. The coefficient of the extensive margin effect is always greater (in absolute value) for the post-eu 4

5 period, which results in a negative overall effect during the period following the EU accession. This paper contributes to the existing literature in several different ways. First, even though an ample amount of empirical literature investigates the effects of trade openness on output volatility, very little work has focused on the direct effect of export diversification. 3 This paper bridges this gap by directly relating the extensive margin of exports to both aggregate and sectoral output volatility. Second, existing research either uses traditional concentration indices (Haddad et al., 2013; Cavallo et al., 2008) or a simple count (Buch et al., 2009; di Giovanni and Levchenko, 2009) of varieties to measure export diversification. In this paper, my main diversification measure is the extensive margin measure developed by Hummels and Klenow (2005), which weights the export varieties of a country depending on their importance in world trade. This way of measuring diversification prevents the extensive margin from being overestimated simply because a country exports a single good to a specific partner in high volumes. Third, the data set in this paper consists of countries that underwent a major, exogenously induced trade-liberalization episode. The entry of Eastern and Central European countries into the world-trading regime mainly dictated by post-cold War political events represents a unique natural experiment on the role of trade liberalization on the volatility. Previous empirical studies utilize data sets (cross-sectional or panel data) that include a large set of countries over a long time horizon. Such large data sets lack the ability to draw specific conclusions concerning the relationship between the liberalization process and the structural change in the exports and the output volatility. The selection of countries and the time frame covered in this study provide a unique opportunity to conduct a natural experiment on the relationship between openness, diversification and volatility, because, following their accession to the EU in 2004, the CEECs experienced a substantial increase in their trade openness along with a significant expansion in their export variety. This characteristic of the data gives us the opportunity to examine the actual effect of extensive margin growth on output volatility in the economies that are experiencing an increase in their openness. The remainder of the paper is organized as follows. Section 2 presents some preliminary findings relating trade openness to export diversification, and gives descriptive information for 3 Among the few are Bejan (2006), Cavallo et al. (2008), Buch et al. (2009), and Haddad et al. (2013). 5

6 the key variables of interest. The CEECs have undergone major trade liberalizations and other structural transformations that affect the degree of openness of their economies during the period Here, I document that a significant spike in trade openness characterizes this period, accompanied by a major growth in the extensive margin of exports, for the CEECs. I find that while the terms-of-trade volatility also increases, the volatility of GDP per capita declines. Section 3 reviews the prior literature. Section 4 describes the data and the empirical strategy that I adopt. Section 5 reports and discusses the main regression results. Section 6 includes some robustness checks of the relationship between extensive margin and volatility. Section 7 concludes. 2 Trade Openness and Export Diversification Although in general, trade liberalization and greater trade openness is expected to lead to higher output volatility, a theoretical ambiguity exists. On the one hand, greater trade openness leads to greater specialization according to comparative advantage, which leaves countries more vulnerable to terms-of-trade fluctuations and foreign demand shocks. On the other hand, the new trade theory of Krugman and Melitz predicts that trade liberalization will lead to a greater variety of products being exported, given the lowering of trade costs. Echoing this theoretical ambiguity, the empirical research yields mixed results. Some of the existing literature relates openness to higher output volatility, mainly through terms-oftrade shocks (Easterly and Kraay, 2000; Jansen, 2004; Cavallo et al., 2008). On the other hand, other research find that trade liberalization leads to less volatility, especially for industrialized countries (Bejan, 2006; Krishna and Levchenko, 2013). The prediction that trade liberalization and openness lead to higher specialization is based on comparative-advantage theory. Research based on traditional comparative advantage theory assume the growth in total trade following a liberalization episode comes fully from the intensive margin growth, which results in higher specialization of the economy. Recent theoretical and empirical studies, however, argue the extensive margin of trade plays an equivalent role in increasing the aggregate volume of trade (starting from Melitz (2003)). Thus, a trade-liberalization episode leading to higher openness does not necessarily increase 6

7 Table 1: Descriptive statistics: Export diversification, variety: product-market pair. Variable Obs. Mean Std. Dev. Min Max EM v Theil-between v HI v Num. of Varieties Notes: Table reports the descriptive statistics export diversification measures between the periods pre- ( ) and post- EU ( ). Each HS-6 category defines a product and each product-market combination is defined as a variety. Diversification measures are calculated for varieties. EM is the extensive margin, Theil-Between is the between component of the Theil index and HI is the Herfindahl-Hirschman index of concentration. Numbers are based on the author s calculations from UN COM- TRADE data. specialization of exports; on the contrary it can result in a more diversified export basket (Melitz, 2003; Anderson and Wincoop, 2004; Broda and Weinstein, 2006; Helpman et al., 2008; Debaere and Mostashari, 2010; Kehoe and Ruhl, 2013). Kehoe and Ruhl (2013) in particular show a significant and robust link exists between total trade growth and growth of the extensive margin. Moreover, these authors document that significant growth in the extensive margin characterize trade-liberalization episodes and periods of structural-change. Table 1 reports descriptive statistics for various export diversification measures along with the simple count of varieties exported for the aggregate data set. The percentage change in export diversification in terms of product-market pairs between the pre- ( ) and post- EU ( ) periods, for all the countries in the sample are reported in Table 2. Contrary to the traditional view, the CEECs clearly experienced a substantial growth in their extensive margin following their accession to the EU. The data reveal the growth in the extensive margin ranges between 14% (Hungary) and 114% (Poland). On average, the extensive margin of exports grew by 47.66% during the post-eu period. Traditional concentration measures also confirm these findings. Both Theil index between component and the Herfindahl index display a negative trend pointing to a declining concentration of exports. The change in per-capita GDP volatility, average trade openness, and terms-of-trade volatility over the periods and are reported in Table 3. Note that accession to the EU is a significant liberalization episode for the CEECs. The data reveal accession to the EU is associated with a major spike in the openness of CEECs. The growth rate of openness ranges from 51 % (Slovenia) to 141 % (Slovakia) with an average of almost 97 %. At the same time, the majority of the countries experienced a decline in per-capita GDP volatility, though unlike 7

8 Table 2: Change in the average export diversification (in %), between pre- ( ) and post-eu ( ) periods, variety: product-market pair Country EM v Theil-Between v HI v Czech Rep Estonia Hungary Latvia Lithuania Poland Slovakia Slovenia Overall Notes: Table reports the percentage change in the average export diversification measures between the periods pre- ( ) and post-eu ( ). Each HS-6 category defines a product and each product-market combination is defined as a variety. Diversification measures are calculated for varieties. EM is the extensive margin, Theil-Between is the between component of the Theil index and HI is the Herfindahl-Hirschman index of concentration. Numbers are based on the author s calculations from UN COMTRADE data. Table 3: Change in the average openness and volatility of the economic indicators (in %), between pre- ( ) and post-eu ( ) periods. Country GDP/cap volatility Trade openness terms-of-trade volatility Czeck Rep Estonia Hungary Latvia Lithuania Poland Slovakia Slovenia Overall Notes: Table reports the percentage changes in the per-capita GDP volatility, average openness, and terms-of-trade volatility, between the periods pre- ( ) and post-eu ( ). Trade openness is measured as (Exports + Imports)/GDP. Volatility is measured as the relative standard deviation of the relevant variable (standard deviation as a percentage of the mean). Numbers are based on the author s calculations from UN COMTRADE and World Bank data. openness, the change in volatility varies greatly among countries. In Czech Republic, output volatility drops by only 3.6 %, whereas the decline is 50.3 % in Estonia and 70 % in Hungary. On the other hand, overall, terms-of-trade volatility increases by 35%. In summary, the data presented in Table 3 reveal that the trade-liberalization episode in CEECs resulted in greater trade openness along with higher terms-of-trade volatility but lower GDP volatility. This result contradicts the traditional view that increased openness leads to increased volatility. Although I leave the detailed empirical analysis to the following sections, considering the substantial growth in the extensive margin of exports during this period, I argue at this point that the main factor behind the decline of the output volatility in the CEECs is the greater trade openness combined with greater export diversification. 8

9 Table 4: Descriptive statistics: Sectoral export diversification, variety: product-market pair. Variable Obs. Mean Std. Dev. Min Max EM vs Theil-Between vs HI vs Num. of Varieties s Notes: Table reports the descriptive statistics of the export diversification measures in terms of variety (product-market pair). The diversification measures are calculated by economic activity (sector). The data are obtained from UN COMTRADE and Eurostat Economy and Finance database. The logic behind this argument is twofold. First, greater openness increases a country s exposure to different types of external shocks. On the one hand, the country will be vulnerable to terms-of-trade shocks, which are regarded as negative product-specific shocks. On the other hand, greater openness increases a country s exposure to different markets, which will increase output volatility through spillover effects, especially if the majority of a country s trading partners experience high volatility. However, if a country s exports are sufficiently diversified in terms of both products and markets, the effect of openness can be reversed because the high diversity in the variety of exports will provide insurance against both product- and market-specific shocks. Second, while penetrating new markets requires large sunk costs, firms do not stop exporting a product to a country when the economy experiences a temporary change because the change in the expected future profits can be ignorable. Thus, the extensive margin does not respond to the business-cycle fluctuations and acts as a stabilizer against temporary shocks in the economy. Ruhl (2008) formalizes this approach through a quantitative general equilibrium model in which firms face uncertainty about future profits and sunk costs. He shows that although the extensive margin responds to permanent changes such as trade liberalization, its response to business cycle-fluctuations is weak. Now I move to the sectoral analysis of the data. Table 4 reports the descriptive statistics for the export diversification measures for the sectoral data set. Although the extensive margin measure and the Theil index have similar averages to the aggregate data, the Herfindahl index points to a much larger concentration of exports at the sectoral level. Additionally, sectoral data display slightly higher standard deviations. Among the three measures, the Theil index exhibits the highest standard deviation, which is not surprising, because unlike the extensive 9

10 Table 5: Descriptive statistics: Selected variables, by economic activity, before ( ) and after ( ) EU. Variable Obs Mean Std. Dev. Min Max Panel A: Before-EU ( ) Output volatility Export openness Output per worker Fixed capital consumption per worker Taxes paid to government Panel B: After-EU ( ) Output volatility Export openness Output per worker Fixed capital consumption per worker Taxes paid to government Notes: Each variable is calculated by economic activity (sector), based on the NACE classification, using the sectoral data set. The table reports the descriptive statistics of the selected variables, for the pre- ( ) and post-eu ( ) periods, separately. The data are obtained from the Eurostat Economy and Finance database. margin measure and the Herfindahl index, the Theil index can take values greater than 1. Table 5 presents the descriptive statistics of the other variables derived from the sectoral data, for the two sample periods. When we compare the sectoral output volatility between preand post-eu periods, we do not see a clear decline (1.1 percentage point), as in the aggregate data. However, note that considerable heterogeneity exists among sectors in terms of volatility change during this period. On the other hand, the post-eu period represents a major increase in export openness. Average export openness of a sector almost triples after EU accession. A similar trend is observed in labor productivity, which more than doubles during the post-eu period. In terms of export diversification growth, the data are similar to (though slightly lower than in) the aggregate data. Table 6 reports the average change in the export diversification by economic activity, after the accession of CEECs to EU. The average sectoral extensive margin grew by almost 41% between pre- and post-eu periods. Both the Theil and Herfindahl indices confirm these findings, though the Herfindahl index points to a much smaller decline in sectoral export concentration. To examine the link between export diversification and output volatility at the sectoral level, Table 7 documents the percentage changes in selected variables, for the first three economic activities that experienced the largest decline in output volatility, between the periods 10

11 Table 6: Average change in the export diversification by economic activity (in %), between pre- ( ) and post-eu ( ) periods, variety: product-market pair Country EM vs Theil-Between vs HI vs CzechRep Estonia Hungary Latvia Lithuania Poland Slovakia Slovenia Overall Notes: The table reports the average percentage change in the export diversification measures in terms of variety (productmarket pair), between the periods pre- ( ) and post-eu ( ). EM vs is the extensive margin of the sectoral exports, Theil-Between is the between component of the Theil index and HI is the Herfindahl-Hirschman index of concentration. The diversification measures are calculated by economic activity, based on the NACE classification. The data are obtained from the UN COMTRADE database. before and after EU. The results show two stylized facts. First, the sectors with the highest decline in output volatility almost always increased their export diversification in terms of product-market categories, the growth rate ranging between 7% and 135%. Second, no clear link exists, positive or negative, between export openness and output volatility at the sectoral level. The change in the export openness of the sectors ranges between -94% to 368%. This observation tells us that a significant degree of heterogeneity exists in terms of the response of the sectors to higher export openness. 3 Literature Review Most of the early empirical papers on the subject focus on the relationship between termsof-trade variability and output volatility (the sector-specific-shocks channel). These papers base their analysis on the traditional Ricardian comparative-advantage models that consider only the intensive margin of trade while ignoring the extensive margin. Even if they include some measure of the export diversification in their analysis, it is usually considered as a control variable, and no clear conclusion is made stating that higher diversification is associated with either higher or lower volatility. A simple two-country - two-good Ricardian model developed by Razin et al. (2002, 2003) predicts trade openness may lead to discrete boom-bust cycles of investment, which causes the terms-of-trade to oscillate and the economy to destabilize. With the help of an optimal mone- 11

12 Table 7: Change in the selected variables (in %), in the sectors with the largest decline in output volatility, between pre- ( ) and post-eu ( ) periods. Economic Output Export Theil- Country activity volatility openness TOT risk EM vs between vs Czech Rep. C C29-C C22-C Estonia C31-C C16-C J Hungary C J M Latvia C13-C M J Lithuania C16-C C J Poland C13-C C C31-C Slovakia C29-C C13-C D Slovenia B A C Notes: The table reports the percentage changes in the output volatility, average export openness, average TOT risk, average extensive margin, and average Theil-between index, between the periods pre- ( ) and post-eu ( ), for the selected sectors. Only data for the the first three sectors that experienced the highest output volatility decline are reported. The names of the economic activities are reported in the Appendix Table A-2. tary policy model, Galí and Monacelli (2005) find increased openness leads to greater output volatility, while reducing the volatility of the exchange rate and having an ambiguous effect on consumption. Using a similar monetary model, Guender (2006) shows increased openness results in higher volatility through the stronger effects of terms-of-trade changes. He further predicts that whereas domestic inflation targeting has a U-shaped effect on output volatility, CPI-inflation targeting results in a strict positive relationship between output variance and openness. On the other hand, by means of a simple theoretical model, Karras (2006) demonstrates the effects of openness on output volatility are ambiguous, both theoretically and empirically. In his paper where he relates the degree of openness of a country to government size, Ro- 12

13 drik (1998) finds greater exposure to external risk results in greater income volatility. However, although he controls for the terms-of-trade volatility as the external risk, he does not take into account the degree of concentration of exports. Easterly and Kraay (2000) examine income, growth and volatility in small states. They find that greater income volatility of small states comes from their greater openness to international trade, which causes large fluctuations in their terms-of-trade. Using a general equilibrium Ricardian model of trade based on Eaton and Kortum (2002), Burgess and Donaldson (2012) investigate the effect of declining transportation costs on the regional income volatility in India. They find that as the openness increases, responsiveness of real income to productivity shocks depends on the price volatility. Kim (2007) distinguishes between openness and external risk in their effect on volatility and finds that once the effect of term-of-trade and exchange rate risk is isolated, openness ceases to be a significant determinant of volatility. di Giovanni and Levchenko (2009) relate output volatility to trade openness using industrylevel data and find that sectors more open to international trade are more volatile. They also document that trade increases specialization, which also plays a role in increasing volatility. Although they do not directly link the extensive margin to output volatility, they investigate the link between openness and volatility of number of firms in an industry. They find that openness induces higher volatility of the entry and exit of the firms and the overall effect of openness on output volatility remains positive. In the existing literature, even though export diversification is considered as a control variable in some of the studies, a direct link between diversification and volatility has not been established. One of the underlying reasons for this shortcoming is that these papers are based on the conventional comparative-advantage models, which fail to account for the fixed and sunk costs the firms face in making export decisions. When firms do not face any marketpenetration costs, the change in the aggregate trade is mistakenly assumed to be all intensive margin growth. However, as mentioned earlier, recent theoretical and empirical trade literature implies the extensive margin is also a significant contributor of trade growth. 4 Research has shown increased openness results in a greater extensive margin, which does not respond 4 Kehoe and Ruhl (2013); Hummels and Klenow (2005); Broda and Weinstein (2006) 13

14 to business cycle-fluctuations as strong as the terms-of-trade. Therefore, any attempt to explain the association between openness and output volatility without the extensive margin of trade would be, if not misleading, incomplete. Ruhl (2008) develops a model that incorporates international real business-cycle theory, in which firms face uncertainty about future profits and sunk export costs. Under a general equilibrium setting, they show the response of firms in changing their exporter status to temporary changes such as business cycles is weak. These types of changes emerge as price changes in the intensive margin of trade. However, a permanent change such as a reduction in tariffs induces more firms to begin exporting, thus increasing the extensive margin. These findings suggest openness to trade brings greater terms-of-trade volatility, but also leads to growth in the extensive margin, which does not respond to short-term fluctuations. Therefore, the extensive margin can play a stabilizing role against aggregate volatility. 5 Tenreyro et al. (2014) develop a multi-sector variation of the Eaton and Kortum (2002) model with stochastic shocks to assess the significance of the two mechanisms, sectoral specialization and market diversification, in determining the role of trade liberalization in output volatility. When exposure to country-specific shocks is an important source of volatility, openness to trade can lower output volatility by reducing exposure to domestic shocks and allowing countries to diversify the sources of demand and supply across countries. Additionally, the paper shows that higher sectoral specialization does not necessarily lead to higher volatility. The direction of the GDP volatility depends on the intrinsic volatility of the sector that the country specializes in, as well as the covariance of the sectoral shocks. Note that Tenreyro et. al. s (2014) model differs from the new trade models in the sense that openness leads to higher specialization of production while it accounts for the market diversification (a broad range of intermediate goods are imported from the rest of the world). Therefore, it fails to capture the full effect of the extensive margin as defined earlier. Although it predicts a stabilizing impact of market diversification on volatility, it lacks the ability to make any statements concerning the effect of product diversification. Jansen (2004) investigates the relationship between income volatility and terms-of-trade 5 Alessandria and Choi (2007) construct a similar, standard international real business-cycle model with export entry costs. They find that the effect of the extensive margin is negligible for aggregate quantities at businesscycle frequencies. 14

15 shocks in small and developing economies. He shows that both higher openness and higher export concentration characterize these countries. After controlling for several structural determinants, he finds a strong positive relationship between terms-of-trade volatility and export concentration. In a second set of regressions, he confirms that higher terms-of-trade volatility leads to higher income volatility, which leads to the conclusion that higher export concentration leaves small economies vulnerable to terms-of-trade shocks, which in return increases income volatility. However, this approach does not establish a direct link between volatility and trade diversification. In a similar study, Bejan (2006) finds mixed results for developing and developed countries. She argues that in developing countries, openness increases volatility through the terms-of-trade channel, whereas export concentration has no significant effect. In developed countries, although both terms-of-trade volatility and export concentration play an important role in increasing volatility, openness decreases volatility because it offers more possibilities for hedging against country-specific shocks. In another study, Cavallo et al. (2008) treat terms-of-trade volatility as a conditioning factor on the relationship between openness and volatility, while introducing export concentration as another control variable into the regression. They find that increased volatility due to higher openness mainly comes from terms-of-trade volatility. Once the effect of terms-of-trade volatility is controlled for, openness plays a stabilizing role in output volatility. He finds no evidence suggesting export diversification affects output volatility. Buch et al. (2009) relates export openness to firm-level output volatility and finds export status is a significant factor in determining the firm-level output volatility. Exporter firms have lower volatility and most of the reduction in volatility comes from the market diversification of the firms. Bacchetta et al. (2009) study the possible role of geographical concentration of exports and exposure to demand shocks in partner countries as a source of economic volatility. They decompose the partner countries volatility of demand into two components: a variance and a covariance component. Using panel data regressions for different country samples, they find geographical diversification of exports plays an important role in reducing the exposure to external shocks. Next, they run a regression of output volatility on terms-of-trade volatility, partner countries volatility, and a set of control variables, and they find partner countries volatility has a positive and significant impact on exporters GDP volatility. In particular, they 15

16 find that the correlation among the various trading partners business cycles (measured by the covariance component) is an important factor in determining the domestic volatility. In a recent study that is possibly the closest to this one, Haddad et al. (2013) investigate the variation of the effect of trade openness on output volatility depending on the concentration level of exports. Using a comprehensive set of export diversification measures, in terms of both products and markets, they run a set of regressions by which they aim to find a cutoff value for the export-concentration level below which the effects of openness on output volatility becomes negative. They find trade openness has a decreasing effect on output volatility in countries where the export concentration (Herfindahl index of product concentration) is below On the other hand, they do not find significant evidence suggesting market concentration plays a role in determining the effect of openness on volatility. The majority of the prior research that relates trade openness to output volatility focuses either on aggregate volatility or industry level volatility. In addition, most of the studies employ data sets that consist of either cross-section or panel data of a large number of countries that are at different stages of development or liberalization episodes. In this study, I attempt to establish a direct association between output volatility and export diversification, by employing measures capturing the extensive margin for products, markets and product-market pairs, both at the aggregate and sectoral levels. By focusing on a set of countries that underwent a major trade-liberalization episode during the time frame chosen, I analyze the effects of the extensive margin of exports on the volatility of per-capita GDP and sectoral output. This study is unique in the sense that the choice of the country set and time frame constitutes a natural experiment by which I examine the actual link between export diversification and output volatility in the economies that go through a major trade-liberalization episode (EU accession) so that their openness to trade increases considerably. My findings support, in part, the conventional view that openness and terms-of-trade fluctuations contribute to output volatility, both at the aggregate and sectoral levels. However, I find strong evidence that the effect arising from the diversification of exports counteracts this effect. Furthermore, I document that growth in the extensive margin of exports in terms of destination markets is the driving force behind the stabilizing effect of export variety growth on output volatility. An important mechanism behind this result may be that 16

17 exporting to a larger number of destinations helps diversify the risks emerging from demand shocks. Additionally, Malik and Temple (2009) find export concentration in terms of products is a strong determinant of terms-of-trade volatility. Therefore, the effect of product diversification on output volatility may be observed through its effect on terms-of-trade volatility. On the other hand, I find that whereas the effect of diversification appears to be ambiguous during the pre-eu period, it is significantly stabilizing during the post-eu period. In this sense, I argue that as the economies that go through trade liberalizations diversify their exports, they insure against terms-of-trade and other external risks that are transmitted through trade. 4 Data and Methodology In this study, I use two different cross-sectional time-series data sets. The first comprises percapita GDP and trade data of a balanced panel of eight Central and Eastern European Countries that joined the EU in 2004, covering the period To be specific, my sample includes Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia, and Slovenia. The second data set includes output and trade data at the sectoral level, for the same set of countries and time frame. I use these two data sets to test and compare the effect of export diversification on both aggregate and sectoral output volatility. Studying the case of CEECs for the period is interesting because during this period, the CEECs underwent an extensive trade-liberalization episode, which resulted in a staggering 162% increase in the average trade openness from 1995 to This startling growth of trade is accompanied by a large increase in the export variety. From 1995 to 2012, the average extensive margin of exports in terms of product-market pairs grew by 94%. The data show the average relative standard deviation of per-capita GDP declined 38%. Therefore, the data sample utilized in this study represents a natural experiment whereby we can observe the actual impact of growth in trade openness on output volatility, and determine the role of extensive margin growth in the volatility change. The trade data I use are from the United Nations Comtrade database, reported in thousand of US dollars, in 6-digit HS 1988/92 classification, covering the exports of the CEECs to 244 partner countries, in 5,030 unique product codes, resulting in 388,352 unique product-market 17

18 pairs, which I treat as a variety in this study. GDP per-capita data are obtained from the World Banks s World Development Indicators database, measured in 2011 purchasing power parity (PPP) dollars for comparability. I convert nominal variables into real variables by deflating each variable by the US real GDP deflator. I measure the aggregate output volatility based on the rolling relative standard deviation of per-capita GDP over a three-year window. The main advantage of this approach is that the rolling-volatility measure yields a longer time-series dimension, which produces more reliable results. Additionally, by using the relative standard deviation measure (standard deviation as a percentage of the mean), I remove the scale effect and render the volatility measure comparable across cross sections. On the other hand, the main disadvantage of measuring output volatility this way is that, by definition, the measure is autocorrelated. However, prior literature has used similar overlapping period measures, making my results comparable to the previous work in the field. My central variable of interest is the interaction between openness and export diversification. I use the extensive margin measure developed by Hummels and Klenow (2005) as my main diversification measure. The extensive margin is based on the product-varieties-growth measure developed by Feenstra (1994), which states the growth in product variety of a single country over time. However, instead of measuring variety growth over time, Hummels and Klenow (2005) adopt a cross-sectional comparison approach. They construct the extensive margin of exports relative to a reference country, so that one can compare the two countries at a point in time. When the rest of the world is chosen as the reference country, the extensive margin becomes a weighted count of the varieties of country j relative to the varieties of the rest of the world. In the spirit of Helpman et al. (2008), in my main specification, I define a variety as a product-market pair. To be clearer, I consider a computer as a product, and a computer exported to the United States as a variety. Specifically, the extensive margin of country j s exports to country m in a given year is defined as EM jm,t = i I jm x kmi,t (1) x kmi,t i I where x kmi,t is the value of country k s (rest of the world) exports of product i to country m at 18

19 time t, I jm is the set of observable products in which country j has positive exports to country m (i.e. x jmi > 0), and I is the set of all possible products in which the rest of the world has positive exports to country m. Therefore, the numerator of this expression is the aggregate value of the rest of the world exports to country m in the products in which country j exports to country m. In the current sample, country j s product set is always a subset of the products exported by the rest of the world. The denominator, on the other hand, is the aggregate value that the rest of the world exports to country m in the set of all possible products. Hummels and Klenow (2005) summarize the extensive margin for varieties of country j s exports in a given year by taking the geometric average of EM jm across all of its export partners: EM vj,t = m M j (EM jm,t ) wjm,t (2) where v denote each variety, M j is the set of partner countries of country j and w jm is the logarithmic mean of the shares of m in the total exports of j and the rest of the world, respectively. w jm s are normalized to 1 so that M j w jm = 1. In addition to the extensive margin for varieties, I further decompose EM vj into its product and market components, and include them in the regressions one by one. This approach gives me the opportunity to further analyze distinct effects of each dimension (product and market margins) independently. 6 Note that the majority of the existing literature uses one or several of the concentration indices that are used in the income-distribution literature (Herfindahl, Theil, and Gini indices). Although these indices give a relatively good idea about the diversification level of exports, they all depend on the value of the country s own exports, which may cause an overestimation problem. Unlike the traditional concentration measures, the extensive margin defined in equation (2) does not depend on country j s export volume; thus, it prevents the extensive margin from being overestimated simply because country j exports a single good to a specific partner in high volumes. Nevertheless, following the existing empirical literature on trade diversification, I also use standard concentration indices for robustness checks, including the 6 The extensive margin of country j s exports for products in a given year is defined as the ratio of rest-of-the world (ROW) exports in the product basket of country j to the ROW exports in all possible products. The extensive margin for markets is defined in a similar way where it is the ratio of ROW exports to the destination market basket of country j, to the ROW exports to all possible markets. Specifically, EM pj = x ki i I j i I x ki and EM mj = x km m M j x km. m M 19

20 Herfindahl-Hirschman index and the between component of the Theil index. My second variable of interest in the first part of my analysis is the trade openness. Consistent with the existing literature, I compute openness as the ratio of total trade (exports plus imports) to the GDP. The GDP data come from the World Banks s World Development Indicators database. In addition to the export diversification and openness measure, I include several control variables, because a number of factors determine aggregate volatility. Financial openness is measured by an index of restrictions on capital account transactions (Chinn and Ito, 2006). Both trade-openness and financial-openness variables reflect a country s exposure to the world markets. External volatility is captured by terms-of-trade risk and exchange rate risk. The terms-of-trade is computed as the ratio of the export-value index to the import-value index, for which the data are obtained from the World Banks s World Development Indicators database. Following Rodrik (1998), I measure the terms-of-trade risk as the volatility of termsof-trade (relative standard deviation of terms-of-trade) multiplied by openness. Exchange-rate volatility is measured as the relative standard deviation of the real effective exchange rate with 18 trading partners. The data for exchange rate are obtained from Eurostat s Economy and Finance database. Other control variables for the aggregate volatility analysis include foreign direct investment volatility, inflation volatility, government spending, population and GDP per capita (PPP). In determining the role of export diversification on aggregate output volatility, I estimate a set of regressions of the following general form: GDP Volatility it = β 0 + β 1 Openness it + β 2 Openness it EMv it + β 3 Openness it TOT Volatility it + θ X it + ɛ it (3) where i and t denote country and year, respectively. The dependent variable is the ratio of the standard deviation of GDP per capita (PPP) to the mean. Openness it is the ratio of total trade to GDP. Openness it EMv it is the interaction between openness and export diversification, whereas Openness it TOT Volatility it is the interaction between openness and terms-oftrade volatility. X is the vector of control variables, and ɛ it is the disturbance term. Control variables include financial openness, foreign direct-investment volatility, inflation volatility, real effective exchange-rate volatility, government spending, population and GDP per capita 20