The Impact of Culture on Intra-Industry Trade in the European
Monetary Union
Tamara Doeve S2384027
MSc International Economics and Business
Faculty of Economics and Business Economics, University of Groningen Course code: EBM868A20
June 19th, 2018
Supervisor: Abdul Erumban Co-assessor: T. M. Harchaoui
Abstract
This paper examines determinants of bilateral intra-industry trade (IIT) in the European Monetary Union (EMU), with an emphasis on the role of culture in trade integration. Trade integration achieved by IIT is perceived by the EMU as a main objective to obtain more benefits of the Euro. Therefore, the sample contains 15 EMU countries over the years 2002 till 2016. Culture is indexed by Hofstede’s dimensions and a measure for cultural differences is calculated by a mean-based index of Kogut and Singh (1988). The empirical analysis is based on the gravity equation, and the Grubel-Lloyd index for IIT. After controlling for country specific variables the results are inconclusive for the three machinery sectors used. Showing that effects of culture might be different across sectors.
JEL Code: C33, E02, E52, J51
1.
Introduction
Nearly 17 years have passed since the introduction of the Euro and almost 24 years since the completion of a Single Market in the European Union (EU). In 2007, Slovenia joined the European Monetary Union (EMU), and in the preceding years Cyprus, Malta, Slovakia, Estonia, Latvia and Lithuania adopted the Euro. The principal economic argument for broadening the free trade area is to exploit the benefits of economies of scale. The trade integration helps to establish a common risk sharing system and convergence of EMU business cycles, increasing the benefits for the EMU (Gächter & Riedl, 2014; Glick & Rose, 2016). Shin and Whang (2003) emphasize that IIT is more important than inter-industry trade in harmonizing European business cycles. Therefore, intra-industry trade (IIT) intensification has been the main symbol of the European integration process (Affortunato, et al., 2013). The empirical findings of IIT are first noted by Balassa (1966) and later by Grubel and Lloyd (1975). IIT is defined as a two-way exchange of goods within the same standard industrial classifications. Baldwin and Evenett (2015) describe the development of IIT as follows; due to the steam revolution, the benefits of scale economies rose. Accordingly, nations specialised along comparative advantages and international trade had an economic boost in the 19 th century. Baldwin and Evenett (2015) refer to this as the first unbundling, as geographical unbundling of economies between production and consumption coincided with the agglomeration of production activities in large scale factories of industrial zones. After the ICT revolution in the mid-1980s, coordination costs related to communication declined. Consequently, stages of production previously performed in close proximity are dispersed geographically, the second unbundling. As a result, production stages became more separated in different tasks and skills and IIT increased, this process is called fragmentation (Barfield & Matthew, 2012; Baldwin & Evenett, 2015).
Kilavuz, et al., 2013). Previous literature admits the potential influence of culture but only control by including country group dummies or using fixed effects (Loertscher & Wolter, 1980; Clark & Stanley, 1999). Clark and Stanley (1999) note that cultural differences are embedded in other coefficients such as distance and language. According to Felbermayr and Toubal (2010), those cultural proximity variables capture cultural familiarity but also reflect other trade-creating factors, such as the cost of communication. Therefore there is need for a better measurement of culture.
Goods are costly to transfer, both because of geographical distance and cultural, language and institutional differences among countries (Kónya, 2006). Most of the early gravity model studies consider only geographical distance, the effects of intangible trade barriers has been an important recent extension of the gravity model (Möhlmann, et al., 2009). Melitz (2008) stresses the importance of culture in communication and making business deals. Large cultural differences are expected to have a negative effect on bilateral trade flows. This paper defines cultural difference as the degree to which shared norms and values differ from one country to another (Hofstede, 2011). Large cultural differences lead to more uncertainty about business partners. This raises the risk of incomplete contracts, and harms trust in the trading partner therefore increases the transition costs of a trade (Brouthers & Brouthers, 2000; Melitz, 2003; Hofstede, 2011). Evidence on prevailing cultural difference in the EMU are put forward by Polonsky et al. (2001) and Liñán and Fernandez-Serrano (2013). As cultural differences in the EMU are present and could substantially hamper business cycle synchronisation, I research the effect of cultural difference on IIT in the EMU. I expect that cultural differences have a negative effect on IIT, because cultural differences increase the costs of doing business.
Previous literature seem to neglect the importance of cultural differences in IIT to the EMU. Consequently, by calculating a measurement of culture and applying it in the EMU, this research adds to the previous literature on IIT in the EMU. I investigate trade flows between 15 EMU countries over the years 2002 till 2016. In practice, IIT is calculated by the Grubel-Lloyd index, and used as the independent variable in the pooled ordinary least squares (OLS) regression. The control variables can be justified by carefully investigating previous literature. Following Linders (2005) I calculate a cultural distance measure based on a mean-based index of Kogut and Singh (1988). By using Hofstedes dimensions of national
culture I obtain a cultural measurement. After performing the pooled OLS, this paper finds that results differ among the sectors. Suggesting that cultural differences play a different role in each sector. Contradictory to previous findings, this thesis finds insignificant results for some of the standard gravity model variables. A lack of data on potential important factors can contribute to this varying result.
The remainder of this paper is structured as follows, in the next section I briefly review the importance of IIT for the EMU. Next, I discuss the early insights of Ricardo’s theory of comparative advantage and a simplified version of the Hecksher-Ohlin theory. I briefly discuss further developments of the international trade theories and empirical findings on determinants of international trade and IIT, emphasizing the effect of culture in international trade. The next section discusses the technical specification of the gravity model. Subsequently, the data chapter provides a clarification on the data selection and a description of the data. Thereafter, I present and discuss the regression results. Lastly, I conclude, present the limitations of my research and provide suggestions for further research.
2
Literature
2.1 Importance of Intra Industry Trade in the EMU
As internal barriers to trade are eliminated, the EU has become progressively more integrated over the past three decades. Allington et. al (2005) emphasize the importance of the completion of the Single Market Project in 1992 and the start of the EMU in 1999. The former removed the remaining physical, administrative and technical barriers to integration, and stimulated competition. The latter introduced a common currency and eliminated exchange rate variations between the Eurozone countries. The main benefits of a currency1 area are gaining free factor mobility and a stable exchange rate. It is expected that the single currency deepens integration by removing exchange rate risk, lowering uncertainty and making cross-border business more profitable. Free factor mobility refers to the ability to move factors of production, labour and capital freely between countries within and across
1 Table A.1 in Appendix A shows the countries that have adopted the Euro, sorted by year.
industries. This allows the factors of productions to move to sectors where they can be more productive (De Grauwe, 2016).
The foremost cost of joining a currency union stems from losing the independent monetary policy. This can be explained by the Mundell (1963) Fleming (1962) model. It is a trilemma between a fixed exchange rate, free movement of factors of production and an independent monetary policy. When countries in the EMU gain free factor mobility and a stable exchange rate, they have to give up their independent monetary policy. Setting a domestic interest rate different from the EMU’s interest rate results in excess demand or supply of the domestic currency, and the exchange rate will adjust accordingly and violating the stable exchange rate. Suppose there are two countries in a currency area, country A and country B. Country A is confronted with a negative demand shock, causing unemployment. As country A is in a monetary union, it cannot adjust its own interest rate to regain competitiveness to boost export, and raise the employment rate. The central bank can choose to lower its interest rates and increase the competitiveness of country A, but it simultaneously pushes inflation upwards in country B. A second option for the central bank is to keep the interest rate constant and force country A to lower its wage rates to become competitive again. Henceforth, giving up the independent monetary policy can be costly when countries in the union experience different shocks and business cycles. Scholars believe that the ECB is better able to set interest rates when the business cycles of EMU countries synchronize (Fidrmuc, 2004; Glick & Rose, 2016).
Several researchers believe that an increase in trade integration facilitates synchronization of EMU business cycles, hence increasing the benefits of the EMU (Clark & Wincoop, 2001; Cerqueira & Martins, 2009). Empirical research on the EMU suggests an increase in trade of up to 50 percent since the introduction of the Euro (Glick & Rose, 2016). Another study points out that the intensity of IIT rose by five to ten percent on average (Baldwin, 2006). Next to this, Shin and Wang (2003) investigate business cycle harmonization and conclude that IIT is more important than inter-industry trade. Also, Gächter and Riedl (2014) emphasize that a rise in trade integration increases labour mobility, the establishment of common risk sharing systems and convergence of EMU business cycles. Consequently, trade
integration is one of the main objectives of the EMU to increase the benefits of the monetary union (Polonsky, et al., 2001).
However, scholars differ in opinion whether highly integrated markets are beneficial for stability in a common currency area. Krugman (1991), views that trade liberalization facilitates more specialization in countries or regions according to comparative advantage theory. This causes regions to be more specialized in a specific sector. Therefore, they are more prone to asymmetric shocks and economic distress, increasing the costs of a currency union. A contradictory view is proposed by Frankel and Rose (1998), they argue that a beneficial currency area can be developed ex ante, due to the endogenous nature of a currency union. A currency union might currently not be optimal, but due to the benefits of a stable exchange rate and freely moving factors of production, the trade volume should grow. A rise in trade integrates markets and increases the fruits of a common currency. Fidrmuc (2004) does confirm the OCA endogeneity hypothesis, while emphasizing the role of trade specialization, specifically IIT. Overall, trade integration has a positive effect on the optimality of the EMU. Accordingly, the EMU’s goal to completely integrate the different economies is justified (Polonsky, et al., 2001). As there are no more internal barriers to trade, such as trade tariffs or different currencies, the EMU should be close to complete integration into one market (Allington, et al., 2005).
2.2 International Trade Theory and Intra Industry Trade 2.2.1 Traditional Trade Models and Comparative Advantages
To understand the patterns of IIT in the EMU, it is important to review the most prevalent trade theories. International trade theories can assist in predicting international trade patterns and its determinants. David Ricardo’s theory of comparative advantage, developed at the beginning of the 19th century, has played a major role in modern thinking about trade (Helpman, 1999). The Ricardian trade theory considers two countries, two goods, and assumes that only labour is used to produce goods and services. The essence of the theory is that whilst country A has an absolute advantage in producing both goods, it exports one product and imports the other product from country B. This theory discusses the causes of international trade based on differences in labour productivity under perfect competition. Capital abundant countries are able to allocate more capital per worker than capital poor
countries, an important source of variation in labour productivity. However, what causes the differences in labour productivity of countries remains unanswered (Feenstra & Taylor, 2014).
Building on the comparative advantage theory, a general equilibrium model of international trade has been developed by Heckscher (1919) and Ohlin (1924). In their framework they emphasize the roles of labour, capital and land in agriculture and industry sectors with constant returns to scale. The availability of these factors of production shapes a country’s pattern of specialization and trade, based on comparative advantages. It shows that when keeping technology across countries identical, differences in factors of production result in international trade based on comparative advantage by factor endowments. Countries that are relatively capital-abundant will export capital-intensive goods, countries that are relatively labour-abundant will export labour-intensive goods (Helpman, 1999; Barfield & Matthew, 2012). Leamer (1995) emphasizes that the model explains many features of the patterns of international trade, and is therefore essential to consider. However, there is still a gap between the theory and data (Helpman, 1999). In reality, technology is not identical across countries and much trade is not driven by factor endowment differences only. As large volumes of trade flows between countries with similar factor proportions and significant trade overlap was observed within industries (Balassa, 1966).
2.2.2 Product Differentiation and Economies of Scale
Traditional trade theory helps to explain the patterns of trade after the first unbundling. Nonetheless, the underlying assumptions of traditional trade models are quite rigid and might not hold in all situations, specifically not after the second unbundling described by Baldwin and Evenett (2015). According to the theories discussed above, trade takes place between countries that have different labour productivity or have different factor endowments. This type of trading is referred to as inter-industry trade, trade between heterogeneous products. However, international trade in similar products cannot solely be explained within the framework of comparative advantages just described.
The first observations of IIT are noted by Balassa (1966), and later by Grubel and Lloyd (1975), and have spurred the interest of explaining IIT patterns. The New Trade Theory developed, emphasizing that in reality many industries do not seem to be characterized by
either constant returns to scale or perfect competition, as proposed by traditional trade theory. New Trade Theory emphasizes economies of scale and product differentiation, and became complementary to the explanations provided by factor endowments. Helpman and Krugman (1985) use this feature of increasing returns to scale, allowing for sectors that differ in their sources of scale economies, and in market structure.
In the framework of Helpman and Krugman (1985), there are several varieties of each commodity. A typical consumer wishes to consume many varieties of a commodity, or desire to consume the same variety of a particular commodity each time. It is irrelevant which type of behaviour a consumer exhibits, as long as in aggregate, all varieties are enjoyed. This love of variety effect is first developed by Dixit and Stiglitz (1977) to justify that a consumer will achieve a higher utility level when more varieties are available. To match these type of preferences, the supply side of the model is characterized by increasing returns to scale and monopolistic behaviour. A producer competes for the same market as any other producer. If he produces a variety that is already being produced, he needs to share the profits with the other producer. If he produces a new variety, that is not available on the market yet, he obtains the monopoly profits. With increasing returns to scale, the average costs decline as output increases. This induces trade specialization, as more profits are obtained when one specializes in one good. Consequently, a firm focusses on one variety, and no variety will be produced by more than one firm. When incorporating transportations costs, industries tend to concentrate in the largest markets to achieve economies of scale. Even if countries have the same factor endowments, external economies of scale and differentiated products in an industry provide an important incentive to trade. Under these assumptions IIT can be explained theoretically, and is mostly influenced by relative market sizes and economies of scale (Helpman & Krugman, 1985).
results for all OECD countries and repeated the test on non-OECD countries with panel data. Their findings are partly inconsistent with Helpman and Krugman’s (1985) model. Hummels and Levinsohn (1995) find that the volume of trade is well explained by the size similarity of the trading partners, and that distance is an important factor in this relationship. Confirming the results of Helpman (1987). Using country-pair dummies, the authors conclude that most of the variation in IIT for all OECD country pairs is explained by country-pair specific factors. Suggesting that another factor than imperfect competition drives the success of the model of Helpman (1987). Cieslik (2005) re-examines the relationship between differences in relative factor endowments ant the share of IIT in bilateral trade flows stemming from the Helpman and Krugman’s (1985) model. He notes that Hummels and Levinsohn (1995) did not derive their estimating equations directly from the Helpman-Krugman framework. Cieslik (2005) adds a control variable, capital-labour ratio, to estimate the relationship between relative factor endowments and IIT shares in bilateral trade flows. The theory of Helpman and Krugman (1985) predicts a negative relationship, as the larger the differences in factor endowments are, the less IIT is expected. Cieslik (2005) confirmed this hypothesis by estimating a panel regression, and considering fixed and random effects estimation methods. Furthermore, Helpman (1999) researched the effects of scale economies, and argues that the existence of scale economies, not the degree, is important for determining the extent of IIT. Clark (2010) investigates the relationship between intra-industry trade and scale economies. He finds that industries with low scale economies, rather than high scale economies, more frequently have high, rather than low, IIT shares. He therefore concludes that theoretical studies should focus on causes of IIT rather than scale economies.
In general, the New Trade Theory is less rigid and better in predicting IIT than standard trade theories, as they include concepts of economies of scale, firm heterogeneity and love of variety of the consumers. Traditional trade theories focus on differences in labour productivity and factor endowments as determinant for trade. To the contrary, the New Trade Theory shift their focus from differences between countries to similarities across countries as determinants of trade patterns. In general, New Trade Theory emphasizes the role of market structure, similarities in demand, economies of scale, income distribution, similar market sizes and trade costs.
2.3 Determinants in International Trade 2.3.1 International Trade
Next to the theoretical studies attempting to explain IIT, empirical studies focus on predicting IIT and finding their determinants. Research on the effects of free trade agreements, tariff rates and currency unions on international trade are mostly calculated with the basic gravity equation derived by Tinbergen (1962) (Soete & Hove, 2017). The gravity model posits that the volume of trade is a function of supply and demand forces in country A and B, and a proxy for transport costs represented by distance between country A and B. It is expected that the larger the distance between two countries, the higher the transport costs are and the less the countries trade. The gravity model was regularly criticized for its lack of theoretical foundation, the model was given a stronger base due to the works of Anderson (1979) and McCallum (1995). Anderson (1979) presented a theoretical foundation for the gravity model based on constant elasticity of substitution (CES) preferences and goods that are differentiated by region of origin.
McCallum (1995) investigates the impact of the Canada-U.S. border on regional trade patterns. When border effects are present, such as transportation costs or tariffs, then it is no longer the case that prices of identical goods have the same price across countries. McCallum (1995) finds that, despite the two countries being similar in terms of culture, language, and institutions, Canadian provinces trade 22 times more with themselves than with the U.S.. It was expected that the border effect would be close to zero, as there is free trade between both countries. This 22 should be close to zero, as this is not the case, it is referred to as the ‘Border Puzzle’. Further evidence on the existence of border effects are put forward by Wei (1996), who focussed on OECD countries. When controlling for country size of origin and destination country, distance, remoteness, language and adjacency, he found a border effect of 2.5. Anderson and van Wincoop (2003) reinvestigated the border effect found by McCallum (1995), where they argue that McCallum (1995) excluded the relative trade barriers with all trading partners. The remoteness of the trading partner should be correctly incorporated, as the more resistant to trade with all others a country is, the more it is pushed to trade with a given bilateral partner. Including the remoteness of the trading partner resulted in a downward adjustment from 22 to 10.7 of the border effect.
Further evidence on the existence of border effects are found in European countries. The success of Europe’s Single Market Programme has been investigated by Head and Mayer (2000). They find a border effects between 10 and 22 in 11 European Union (EU) countries. Their findings suggests that there is almost no relationship between market fragmentation and the barriers that were identified and removed by the Europe Single Market Program. Market fragmentation is defined as the effect of national borders on the pattern of commercial transactions. In their discussion they emphasize that consumers exhibit a bias towards domestic goods. This can be the outcome of protection of domestic suppliers or cultural differences (Head & Mayer, 2000). Chen (2004) also investigates the border effect in the EU and finds a magnitude of 3.7. He notes that the choice of the distance measure, together with the specification of the gravity equation are crucial for assessing the size of the border effect. Combining previous research Rose and van Wincoop (2001) find that more distant countries trade less, as do countries with larger land areas and land-locked countries. Countries with common languages trade more, so do countries with land borders, trade agreements, colonial histories, large markets and richer countries. Their main point is that when countries join a currency union, the trade barriers associated with national borders are halved, significantly raising international trade and welfare.
Another component that affects international trade is poor governance of a country. As poor governance entails negative externalities for transactions, and consequently raises transaction costs (Wei, 2000). Linders et al. (2005) extend the gravity analysis by focusing on the relevance of the quality of governance on bilateral trade. As good legislation initiates more security in trade, and its effectiveness stimulates inter-personal trust and ways of doing business. Environmental uncertainty caused by weak legislation can lead to frictions and uncertainty between trading partners. Their research shows that having a similar institutional framework and familiarity with the rules and norms of the trading partner promotes bilateral trade (Linders, et al., 2005). De Groot et al. (2004) found that countries with comparable governance quality levels generally traded more. Hence, it is expected that institutional distance between countries has a negative effect on bilateral trade flows.
affect bilateral trade by separating the measurements of common language. It is divided into a measure of direct communication and open-circuit communication. The former is a language spoken by at least 4 percent of the population, and the latter is a language spoken by at least 20 percent of the population. Language is interpreted strictly as a tool of communication, but Melitz (2008) admits that it reflects many aspects of culture as well. Melitz (2008) estimated a gravity equation following Anderson and van Wincoop (2003) and conclude that the impact of a common language on foreign trade can originate from direct communication as well as the ability to translate. Consequently, a common language affects bilateral trade positively as it decreases the communication costs and possibly lowers the chance of miscommunication. Linders et al. (2005) argue that by including language, colonial linkage and religion dummies important aspects of cultural distance are covered. Felbermayr and Toubal (2010) claim that cultural proximity variables clearly capture cultural familiarity but also reflect other trade-creating factors. For that reason, they attempt to draw a measure of cultural proximity from the Eurovision Song Contest. An advantage of this measurement is that it is not time invariant, where traditional measurements such as religion and common language are. This measure of cultural proximity positively affects trade, and shows that previous measurements underestimate the effect of culture on trade (Felbermayr & Toubal, 2010).
2.3.2 Intra-Industry Trade
One of the first empirical findings on IIT was published by Balassa (1966), a vast literature has developed on the subject. Balassa (1966) investigates the effects of the elimination of trade tariffs upon the formation of the EU. He posits that trade among the countries is characterized by IIT rather than inter-industry specialization. Loertscher and Wolter (1980) investigate a sample of bilateral trade flows among OECD-countries to identify simultaneous differences in the extent of IIT among industries and among countries. They conclude that a relative decrease of prices for transport and communication services, due to the removal of trade barriers, tend to be accompanied by an increase in IIT. Other determinants tested are the development level of the country, market size, distance, customs unions, language group, adjunct countries, and cultural group dummies. Despite most variables being significant, and showing the expected signs, the explanatory power of the analysis was low, according to Balassa (1986). Henceforth, Balassa (1986) analyse developed and developing countries in the manufacturing sector. He used cross-country trade data and divided the countries into
three groups; a developing country sample, a developed country sample, and a pooled sample. In the pooled sample the regression coefficients of GNP per capita, GNP, an adjusted distance variable, trade orientation and a Singapore dummy are all statistically significant. The adjusted distance variable is a weighted average of the inverse of the distance between two countries and divided by the GNP of the trading country. Trade orientation is indirectly measured by residuals from hypothetical values of per capita exports. The separation in two 2 country groups does not affect the explanatory power of the regression equation.
One important distinction in the IIT literature arose between horizontal and vertical intra-industry trade. The former emerges when different varieties of a product are of similar quality. The latter exists when different varieties of a product are of different qualities (Greenaway, et al., 1995). Greenaway et al. (1995) investigated the determinants of IIT, and the determinants of horizontal and vertical ITT separately. In their measurements they include a variable to proximate for scale economies, market structure competitiveness, importance of multinational for a certain industry. For total IIT all variables are significant, indicating that vertical and horizontal IIT are systematically related to features of production and industrial structure. In particular, horizontal IIT is more likely to be driven by scale economies and imperfect competition. In contrast, vertical IIT is more likely to be driven by differences in endowments at different stages of the value chain (Aturupane, et al., 1999). Kawecka-Wyrzykowska (2009) empircally investigates IIT in the EMU, and concludes that IIT has been dominated by vertical IIT. She distinguishes low and high quality vertical IIT and posits that low quality vertical IIT is primarily present, and high quality vertical IIT is increasing. As horizontal IIT is a small part of total IIT in the EMU, and it is empirically difficult to distinguish between horizontal and vertical IIT, I will focus on IIT in general (Greenaway, et al., 1995; Kawecka-Wyrzykowska, 2009).
Clark and Stanley (1999) identify country and industry level determinants of IIT between the United States and developing countries. As independent variable they use IIT measured by the Grubel-Lloyd index (1975). The higher the value in this index, the more trade is associated with IIT. In their estimations they control for random group effects. The results show that the size of the trading partner exerts a positive effect, and distance a negative effect on IIT.
2 The values for per capita exports have been derived from a regression equation including variables
representing the availability of mineral resources and proximity to foreign markets (Balassa, 1986).
Further industry specific determinants that affect IIT positively are the advertising intensity, the offshore assembly provisions use, and the industry size. The seller concentration, sectoral dispersion of industry sales, and capital intensity have negative effects on IIT between industrial nations and developing countries.
Clark and Stanley (1999) research IIT and note that geographical distance can also reflect such factors as seasonal trade, border trade, regional economic integration, language and cultural differences, and market familiarity. Aturupane et. al (1999) studied country and industry determinants of IIT as well, and found strong support for country effects, but little for the industry specific variables. More recent research on IIT is conducted by Shahbaz and Leitao (2010), who consider country specific determinants of IIT between Pakistan and her trading partners. Differences in GDP per capita between Pakistan and her partners and geographical distance have a negative effect on the degree of IIT. Additionally, they added a proxy for economies of scale and the variety of differentiated products by including the average difference of GDP per capita Pakistan and its trading partners, that have a positive effect on IIT.
In summary the following variables are the major determinants identified in the literature to explain IIT in the EMU. For the standard gravity equation, GDP of origin and destination country are frequently used proximate market size of the origin country and destination country (Anderson & Wincoop, 2003). Geographical distance and effective distance are used as a distance measure, the choice of distance measure is very important (De Groot, et al., 2004; Clark, 2010). Other geographical variables such as adjacent countries and remoteness are important in investigating trade and IIT effects (Clark & Stanley, 1999; Anderson & Wincoop, 2003). Many studies have extended the basic gravity equation indicating whether the trading partners share a common language, religion or a measurement for institutional distance (Linders, et al., 2005; Melitz, 2008). Most studies find that these variables positively affect the magnitude of trade flows. These variables only capture cultural familiarity, as the trading partners will have more knowledge of each other’s culture and find it more convenient to communicate and share information (Möhlmann, et al., 2009). Culture has been conspicuously ignored, or less considered in the literature. I attempt to fill that gap by going beyond cultural familiarity by focusing on cultural distance.
2.4 Culture in International Trade
To understand the effects of culture on IIT better, this section explains the effects of cultural distance on international trade. In behavioural economics, an increasing number of researches are focussing on establishing relationships between cultural variables and economics. Especially after 1980, due to the publication of Hofstede’s study about national cultures (Taras, et al., 2009). The most reoccurring phenomenon is increasing transaction costs when cultural distance raises (Williamson, 1979). Hofstede (2011) argues that an increase in cultural differences increases the uncertainty about the business partner, thus impeding the realization of business deals. Additionally, Brouthers and Brouthers (2000) note that people from two different cultures have different understandings of the same situation, increasing the risk of incomplete contracts. Differences in perceptions complicate interactions and harm trust, factors that generally facilitate the interaction process and lower the costs of trade. Also, it is difficult for multinational companies to transfer skills and competences to culturally distant people. As a consequence, there are lower levels of foreign direct investments when countries are dissimilar culturally (Pejovich, 2003). This suggest that cultural distance between countries reduces the amount of trade between them. Therefore, it is not surprising that past research shows that cultural differences increases the costs of doing business (Anderson & Wincoop, 2003; Huang, 2007).
According to Möhlmann et al. (2009), it is well established that cultural distance raises the costs of international trade, as cultural differences make it harder to understand, control, and predict the behaviour of others. However, there is a theoretical argument for a positive relationship between trade and culture. Linders et al. (2008) argue that initially the relationship is negative, but in very cultural dissimilar countries, an indirect positive relationship can occur. When firms choose their entry modes the costs of the different entry possibilities are analysed. Cultural distance can raise the costs of FDI substantially, as FDI requires resource commitment and considerable investment in the local market. Moreover, FDI requires closer interactions between a wide variety of local stakeholders such as employees, unions, suppliers, and government institutions than international trade. At a certain point the costs of FDI might be higher than the costs of trade, contributing to a positive effect of cultural distance on trade. Helpman (2004) investigates the choice of entry mode and concludes that when entry costs become too significant, firms choose to trade rather
than to enter the country. This indirect positive effect starts to be of a considerable effect when the countries are culturally highly distant.
There are a view studies that investigate cultural differences in the European Union. Polonsky et al. (2001), focussed on consumers’ perspective concerning ethical questions. The study sample university students across four Northern EU countries and four Southern EU countries. They find that Southern EU consumers are more eager to actively benefit from illegal and questionable activities than Northern EU consumers. Liñán and Fernandez-Serrano (2013) also find different cultural values in Europe regarding entrepreneurship measured by the Global Entrepreneurship Monitor and the Schwartz Value Survey. They identify four clusters based on cultural differences in the EU: firstly, the English- speaking countries, secondly, Eastern European countries, thirdly, Mediterranean Countries, and lastly, North and Central European Countries. The figure below presents the cultural differences in the EMU based on the six dimensions of Hofstede (2011) and the mean based measure of Kogut and Singh (1988). The figure shows that cultural differences are present in the EMU, but there is3 not a country that has a significant higher average cultural distance compared to the other EMU countries.
Figure 1: Average Cultural Distance between the other EMU Countries
Source: Hofstedes’ Cultural Dimensions (2015)
3 The choice of distant measure and the corresponding calculations are further explained in the data section.
In a world of perfect competition, open borders and the same currency, border effects should be close to zero (McCallum, 1995). All studies discussed in the previous section demonstrate a significant positive border effect, but differences in magnitude (Head & Mayer, 2000; Chen, 2004). As a result, these studies show that the markets are less than perfectly integrated, which could be due to an omitted variable bias. Head and Mayer (2000) argue that consumers exhibit a bias towards domestic goods, due to protection of domestic suppliers or cultural differences. Linders et al. (2005) posit that the unobserved barriers to trade are often related to asymmetric or incomplete information and uncertainty in the exchange. In order to increase the benefits of the EMU, trade barriers have to be minimal to enhance IIT (Polonsky, et al., 2001). Literature on the effect of cultural differences on IIT have been minimal in the EMU. As there is evidence of cultural differences in the EMU, presented in Figure (1), this study investigates the effects of cultural differences on IIT in the EMU.
3.
Methodology
The gravity model is used to study the impact of culture on bilateral IIT, and is estimated by a pooled OLS. The model has its advantage, as it has been very consistent (Rose, 2000). Furthermore, it has been the exclusive instrument in studying the effects of barriers of trade (Melitz, 2008). IT I ij,k = D ij Y Yi j (1) Equation (1) is the standard gravity model, where i and j denote the exporting and importing country, respectively, and k refers to the kth industry. IITij,k represents the degree of IIT between a country pair in a specific country. Yi is the GDP of the exporting country and Yj of the importing country, Dij stands for the distance between both countries. The gravity model posits that country ’s imports from country j i increase with the trading partners’ combined economic mass, and decrease as geographic distance, a proxy for transportation costs, increases. By rewriting Equation (1) to make the estimates less sensitive to extreme observations, and transforming it into an logarithmic estimation equation I obtain:
og og og og
log l
(
IITij, k)
= β0+ β1log l (Y )i + β2log l(
Yj)
− β3log l(
Dij)
+ εij,k (2)is the error term, that captures the residuals of the equation. Many possible influences on
εij,k
bilateral trade are not captured by the model. Therefore, several control variables are added to account for bilateral trade costs. I employ the knowledge from the previous studies to determine the explanation variables in this study, as well as the variable of interest, cultural distance:
og og og og og
log l
(
IITij,t)
= β0+ β1log l (Y )it + β2log l(
Yjt)
+ β3log l(
Dij)
+ β4log l(
DYijt)
+ (3)The sign β3 is made positive, as this notation is more convenient for calculating and interpreting the regression results. The variables are the logarithm of GDP of country i and country , j DYijt is the logarithm of the difference in GDP per capita between country i and is a dummy for common border, a dummy for language, is an indicator .
j CBij Lanij LLi
for the origin country being landlocked, a measure for institutional distance between country and is , and a measure for remoteness of the destination country. The
i j IQij Remijt
variable of interest is cultural distance, represented by CDij.
Table 1: The Expected Sign and Effect of the Independent Variables on IIT
Variable Definition Expected Sign Expected Effect on IIT
Yit GDP of origin country - Positive
Yjt GDP of destination country - Positive
Dij Distance + Negative
DYijt Difference in GDP per capita + Negative
CBij Common border - Positive
Lanij Common language - Positive
LLi Landlocked + Negative
Remij Remoteness + Negative
IDij Institutional distance + Negative
CDij Cultural distance + Negative
Notes: The dependent variable becomes negative when transforming it into logarithms. Therefore, the expected sign and expected effect have opposite signs.
Table 1 provides an overview of the expected effects of the control variables and the variable of interest. The standard gravity model predicts trade flows by country size and distance. It is expected that the larger the countries are, the more trade flows they have. Furthermore, the more distant the countries are the higher the trade costs, and the fewer bilateral IIT is expected (Clark & Stanley, 1999; Shahbaz & Leitao, 2010). The difference between exporter and importer’s GDP per capita are included to account similarity in demand and consumer tastes. Literature insinuates that the smaller the differences are, the more IIT the country pair exhibits (Loertscher & Wolter, 1980; Hummels & Levinsohn, 1995). Also, it is expected that adjacent countries trade substantially more than non-border countries, and their IIT is expected to be higher (Clark & Stanley, 1999; Linders, et al., 2005). Often literature includes a dummy for the origin country being landlocked, the expected effect of an origin country being landlocked is negative. It is expected that a country that has access to the open sea is better able to trade (Wei, 2000; Melitz, 2008).
Melitz (2008) has stressed the importance of common language. And so, it is expected that countries that share the same language trade more (Loertscher & Wolter, 1980; Melitz, 2008). Anderson and van Wincoop (2003) incorporate remoteness in their study, to adjust for the ability of the other country to trade. It is predicted that the more resistant to trade the destination country is, the more it is pushed to trade with a given bilateral partner. As remoteness increases with the ability of a country to trade, the expected effect on IIT is negative. It is expected that the more similar the institutions are the more trade the countries will have. Therefore the expected sign of institutional distance is negative (Linders, et al., 2005). The last variable is cultural distance, and is expected to have a negative effect. Because, cultural differences make it harder to understand, control, and predict the behaviour of the trade partner (Möhlmann, et al., 2009). As only Eurozone countries are considered, a measurement of openness, common religion, common currency and free trade agreements, used in most research, are excluded. To sum up, the paper will investigate the following 4 hypotheses; cultural differences have a negative effect on IIT in the EMU.
4 Papers that include openness, common currency and/or free trade agreements are Rose (2000), Melitz (2008)
and Anderson (2011).
4.
Data and Descriptive Statistics
The sample used in this paper is a dataset which contains data of the years 2002 till the most recent complete data 2016. The year 2002 is taken as a starting point of the dataset, as this is the year that the Euro is introduced in several countries. The cross-section includes 15 Eurozone countries. Due to data limitations on Cyprus, Lithuania, Latvia and Malta, these5 countries are excluded from the dataset. The dataset includes only countries that have the Euro as currency, Austria, Belgium, Germany, Estonia, Spain, Greece, Ireland, Italy, Finland, Luxemburg, the Netherlands, Portugal, Slovenia and Slovakia. Table A.1 in Appendix A provides an overview of all the countries and the year that they introduced the Euro.
In practice, international trade flows are classified in various ways, I apply SITC 3-digit categories organized by the Standard International Trade Classification (SITC) Revision 3. The SITC provides values of exports and imports by partner countries and by commodity measured in current USD dollars, obtained from the Eurostat database. As the SITC data is used to calculate the degree of IIT, a ratio, SITC data is not converted into Euros. To have a measure for IIT the Grubel-Lloyd index is used, as it is a widely used method to calculate IIT.
This index expresses IIT as a share of total bilateral trade in a particular industry :k
6
IITij,k = 1 − |X −Mi i|
(X +M )i i (4)
The Grubel and Lloyd index measures IIT of a particular product or sector for a country pair. Where Xi is the export from country i to country , and j Mi the imports of country i from country j in a specific sector. If ITI ij,k is equal to one, there is only IIT for that product or sector, as the country exports the same quantity of a good as it imports. Consequently, there is only inter-industry trade when IITij,k is close to 0 (Grubel & Lloyd, 1975). One of the critiques on the Grubel-Lloyd index that this paper handles is the aggregation problem. As different types of goods and services are lumped together in the same sector, the corresponding IIT does not show a compatible picture. By applying SITC 3-digit categories I
5 Table A.1 in Appendix A presents the Eurozone countries and year of entry.
6 Several other measures exist, but I will not account for them here. The interested reader is referred to, for
example, Balassa (1979).
try to overcome the aggregation problem. The more disaggregated the category is the better the Grubel Lloyd index is in predicting the true IIT level.
Figure 2: Average IIT in the EMU over Time, sorted by Category
This paper investigates IIT in 3 different categories in the EMU manufacturing sector. These 3 categories are used as they represent different types of machinery industries, which allows to check for consistency across the results. Next to this, these categories have a significant amount of data available, therefore useful in the estimation. Figure 2 above provides the average IIT over the years of these categories. The first category is motor vehicles for the transport of persons other than public-transport vehicles, it has the lowest average IIT shares of this selection in the EMU. The second category is tubes, pipes and hollow profiles, and tube or pipe fittings, of iron or steel. This category has the highest average IIT rates, as it is an intermediary product, and not a final product (Baldwin & Evenett, 2015). The third category is household type, electrical and non-electrical equipment, represented by the green line. In the remainder of this paper these categories are referred to as motor vehicles, iron or steel pipes and household equipment, respectively. More descriptive statistics are provided later in this section.
of effective geographical distances, as it takes into account underlying variables that affect trade costs. These variables could be correlated with variables used to calculate geographical distance (Clark & Stanley, 1999). Following the direct communication argument of Melitz (2008) the country dummy for common language is equal to 1 when a language is the main language in both countries. Information on geographical distance, adjacent borders, and common language are taken from the CEPII database.
Discussion on choosing the appropriate measurement for culture in the literature remains. In 1980 Hofstede’s publication of national cultures started to make it possible to draw relationships between cultural variables and economics as culture is indexed. Later, the Global Leadership and Organizational Behaviour Effectiveness Research (GLOBE) or GLOBE project was developed (House, et al., 2004). The main limitation of the latter approach is the focus on organizational leadership, rather than national culture. As the GLOBE surveys managers only, but limits conclusions to national differences in managerial cultures (House, et al., 2004). Despite such differences, the resulting measures of culture correlate strongly between the frameworks. Consequently, whichever choice one makes does not have a significant effect on measured national culture. The use of GLOBE is advised when the nuances and layers it offers are necessary, but if not, the universal language of Hofstede may be a more elegant choice (Maseland & Hoorn, 2017). For this reason, and following Linders et al (2005), this paper will use the six dimensions of Hofstede (2011) outlined in Appendix A. The dataset from Hofstede’s website is used. It contains dimension scores from 2015 in the 0-100 continuum. Following Linders et al. (2005), different dimensions of national culture will be used to indicate culture, and cultural distance is calculated by a mean-based cultural distance measure developed by Kogut and Singh (1988):
CDij = 61 ∑6 c=1 C /V
(
(
ci− Ccj)
2 c)
(5)Where Cci represents country i ’s score on the c th cultural dimension, Ccj represents country ’s score, and j Vc the variance of this dimension across all countries. This measure is often used in international business research (Linders, et al., 2005; Brouthers & Brouthers, 2000). By adding a measurement for cultural distance, my research differs from other IIT
studies that incorporate culture (Loertscher & Wolter, 1980; Greenaway, et al., 1995; Melitz, 2008).
Furthermore, it is expected that high quality institutions boost international trade flows (Linders, et al., 2005). Kaufmann and Mastruzzi (2003) developed indicators of governance quality based on an unobserved components analysis of several hundreds of variables measuring perceptions of governance drawn from 25 sources constructed by 18 organizations. Kaufmann and Mastruzzi (2003) identified six dimensions of governance infrastructure quality along which countries are different. Namely, voice and accountability, political stability, government effectiveness, regulatory quality, rule of law and control of corruption. More detailed information on these dimensions are outlined in Appendix A, table A.4. Using the method of Kogut and Singh (1988), and following Linders et al. (2005), institutional distance is calculated as:
IDij = ∑6 n Vn I −I (ni nj) 2 (6) Where Ini represents country ’s score on the i n th institutional dimension, Inj represents country ’s score and j Vn the variance of this dimension across all countries considered in this estimation.
Anderson and van Wincoop (2003) incorporate implicit barriers to trade using iterative custom nonlinear least squares programming. However, many researchers have used a reduced-form version of the custom treatment, where the multilateral resistance terms are approximated by the remoteness indexes (Yotov, et al., 2016). Therefore this study controls for remoteness of country by including a measurement for remoteness, following Wei (1996) and Anderson (2011):
Remj = ∑
i Y m
Dim
(7)
Where Remj is the continuous variable between zero and one for remoteness, Dim is the average distance between country i the other trading partners other than country , and j ym the average GDP of the trading partners other than country .j
Table 2: Summary Statistics of the Variables Variabl e Number of observations Mean Standard Deviation Media n Maximu m Minimum GDP* 3150 26.24 2 1.496 26.154 28.680 23.210 DY* 3150 9.236 1.132 9.449 11.191 4.605 D* 3150 7.008 0.775 7.220 8.121 4.087 CD 3150 11.45 9 3.122 11.104 19.972 4.652 REM 3150 0.222 0.081 0.191 0.517 0.119 ID 3150 12 12.883 7.379 62.689 0.363 IIT1* 2463 -1.929 2.014 -1.188 -0.000 -12.842 IIT2* 2738 -1.396 1.651 -0.771 -0.000 -12.097 IIT3* 2913 -1.474 1.532 -1.023 -0.000 -11.513
Notes: The variables are GDP, the logarithm of GDP of the origin and destination country, DY the difference of
GDP per capita between the origin and destination country, D is the logarithm of geographical distance, CD is a
measurement for cultural distance, REM is a measurement for remoteness of the destination country, ID is a
measurement for institutional distance. IIT 1 is IIT in the motor vehicles sector, IIT2 is IIT in the steel or iron
pipes sector, and IIT3 is IIT in the household equipment sector. Furthermore, variables indicated with ‘*’, its
descriptive statistics are presented in logarithms.
Table 2 presents the summary statistics of the data as specified in the regressions. The sample is not corrected for outliers to accurately reflect the full sample. The lowest GDP stems from Estonia in 2002, and the highest from Germany in 2016. The largest GDP per capita differences stem from Luxembourg and Slovakia in 2007. In general, Luxembourg has a relatively high GDP per capita, therefore this observation is not very striking. The smallest difference is between Slovenia and Greece in 2012. The average cultural difference between EMU countries is 11.459, with a standard deviation of 3.122. The score per dimension is between 0 and 100, an average difference of 11.459 is significant. Portugal and Belgium exhibit the largest cultural differences, and the smallest differences are between Finland and Ireland. Institutional differences are also scored between 0 and 100, and the institutional difference measure has higher standard deviations than cultural differences. An outlier is the country pair Greece and Finland, who are institution wise very different, with a measure of 62.689. Ireland and Germany have very similar institutions. Observing the statistics of IIT
shows that all the sectors miss observation. Hence, for a country pair with a missing observation, the country pair is removed from the regression.
5.
Results
5.1 Econometric Issues
Before discussing the results of Equation (3), I will first discuss the econometric problems, and how I solved them. This section considers problems related to multicollinearity, endogeneity and heteroscedasticity.
For this panel analysis a pooled OLS regression is used. In the total dataset there are some variables on bilateral IIT per category missing, this country pair is than omitted from the regression in that category of that year. Furthermore it is important to investigate the correlation coefficients of the explanatory variables. Table 3 shows the correlation coefficients for all variables used in the regression. Distance is highly correlated with the indicator of a common border, as expected. Therefore, this paper checks for multicollinearity after the standard regression, as a robustness check. Clark and Stanley (1999) note that distance can also reflect information that is captured in other variables, such as border trade, language difference and cultural differences. Despite this, there are no correlation coefficients, next to distance and common border, with a correlation coefficient above 0.5.
Table 3: Correlation Coefficients of the Dependent Variables
LL -0.205 0.481 0.058 0.013 -0.281 -0.198 1.000 REM -0.005 -0.108 -0.114 -0.125 0.352 0.317 -0.232 1.000
ID -0.023 0.188 -0.157 -0.163 0.294 0.171 -0.051 0.080 1.000
Notes: *GDP is measured in millions of dollars. Also the distinction between GDP (per capita) of origin country
and destination country is removed, as the import and export countries are identical. The variables are GDPi, the
logarithm of GDP of the origin country, DY the difference of GDP per capita between the origin and destination
country, CB is the common border, Lan is an indicator variable for common language, D is the logarithm of
geographical distance, CD is a measurement for cultural distance, LL is a dummy variable for the origin country
being landlocked, REM is a measurement for remoteness of the destination country, ID is a measurement for
institutional distance.
A general econometric issue when working with panel data is the potential endogeneity of variables. Due to unobserved heterogeneity and omitted variables, coefficients and their confidence intervals will be biased. Baier and Bergstrand (2007) explain that the endogeneity of variables can be caused by simultaneity bias, omitted variable bias or measurement errors. As GDP is measured as a function of net exports, and is thus potentially endogenous to bilateral trade flows and IIT (Glick & Rose, 2016). Nonetheless, Baier and Bergstrand (2007) think this can largely be ignored as the exports mentioned in this literature are concerned with total exports, instead of bilateral exports, which tend to be a small fraction of any country’s multilateral exports.
Baier and Bergstrand (2007) argue that the most important source is the omitted variables bias and can be best treated by using panel data and fixed effects. When using panel data one can use either fixed effects or random effects to control for variables one cannot observe or measure, as it accounts for individual heterogeneity. However, when using fixed effects, time-invariant variables, such as indicator variables and culture, disappear from the equation. To overcome this problem, I estimate the models by implementing fixed effects using a least squares dummy variable model. As suggested by Baldwin and Taglioni (2007) I estimate Equation (3) with country pair dummies to control for country pair fixed effects. Another option is to use random effects, it assumes that the variation across entities is random and uncorrelated with the independent variable. After estimating both models, I conducted a Hausman test to determine whether random effects are to be preferred over fixed effects. Table B.1 in appendix B shows the results. In all the models the leas squares dummy variable is preferred, hence I estimate the models accordingly.
Another potential econometric problem is heteroscedasticity, different variances across the bilateral partners. It does not affect the consistency of the least squares estimator, but it affects the standard errors. The standard errors become too small and overstate the reliability of the least squares estimates. By applying the Breusch-Pagan test for heteroscedasticity I find that the null hypothesis for homoscedasticity the three models can be rejected. Hence the datasets are heteroskedastic, I will adjust for this by adding robust standard errors. After these adjustments, I obtain the results presented in Table 3.
5.2 The Regression Results
The fixed effects model in Table 3 column 1 shows significant negative results for the cultural difference measure. Implying that the larger the cultural difference are the more IIT in the motor vehicles sector. Significant results are found for GDP of origin country, common border, language, distance, remoteness and institutional distance. GDP of origin, has a negative sign, as expected. Suggesting that a one percent increase of GDP of origin leads to an increase of 1.110 percent in IIT. Striking is that the coefficients of common border, remoteness and institutional differences have a sign that is different than hypothesized. For iron or steel pipes different results are obtained, displayed in column 5. The coefficient for cultural distance is not significant. Therefore, I cannot conclude anything about the relationship between cultural distance and IIT in the iron or steel pipes sector. The coefficients for the GDP of origin country, the language dummy and remoteness are significant. Where GDP of origin country has the sign as hypothesized, and language and remoteness do not. Surprisingly, investigating the household equipment sector, the coefficient for cultural differences has a positive sign. Implying that an increase in cultural dissimilarities hampers bilateral IIT in the household equipment sector. Variables that are significant are GDP of origin and destination country, common border, distance, and if a country is landlocked. The coefficients for GDP of origin and destination countries imply the larger both economies are the less IIT they have. Moreover, distance affects IIT positively in that sector, which is also not as hypothesized. Having a common border affects IIT positively, which is according to theory.
To verify that the results in Table 4 are robust, I estimate Equation (3) without outliers in the sample. By removing outliers for the continuous variables from the dataset provides results
that are not influenced by extreme observations. The results are presented in column 2, 5, and 7 8. For IIT in the motor vehicle sector it does not change the sign of cultural distance, it is still negative and significant at the one percent level. Implying that the larger the cultural difference are the more IIT in the motor vehicles sector. Three differences appear regarding the significance of the coefficients. Namely, the GDP of origin country, language and remoteness become insignificant. Removing the outliers does alter the significance level of GDP of origin country, common border, language and the intercept. Where GDP of origin country becomes less significant, the common border dummy significant, and the language dummy insignificant. Moreover the signs for GDP of destination country and of the origin country being landlocked change, but remain insignificant. Hence, I cannot conclude anything about the relationship between these variables and IIT in the iron or steel pipes sector. The effect of excluding outliers for IIT in the household equipment does not change the significance of cultural distance, it remains positive. It does increase the magnitude of the effect of cultural distance on IIT in the household equipment sector. Suggesting that an increase of cultural distance by one unit, results in an expected increase of 48.2 percent in IIT in the household equipment sector. Possibly leading to IIT levels above 1, which is impossible by construction of the Grubel and Lloyd index. The significance of the size of the home and destination market, GDP of origin and GDP of destination country, disappears, and the variables become insignificant. By omitting the
7 Outliers are observation outside the range of three standard deviations above or below the mean of the
variable.
Table 4: Regression Results
Variables IIT in Motor Vehicles Iron or Steel Pipes Hous
(1) (2) (3) (4) (5) (6) (7) GDPi - 1.110* - 0.486 - 0.502 - 1.536*** - 1.209** - 1.540*** 0.396 GDPj - 0.392 - 0.369 - 0.267 - 0.107 0.051 - 0.097 0.378 DY 0.075 0.064 0.010 - 0.084 - 0.077 - 0.084 - 0.0 CB 4.694*** 4.590*** 4.681*** 0.129 0.418* 0.129 - 0.5 Lan - 3.588* - 4.942 5.197 4.079*** 2.745 4.067** - 0.3 D 7.675*** 7.691*** 7.783*** 0.079 0.433 0.091 - 0.6 CD - 0.710*** - 0.773*** - 0.781*** 0.037 0.011 0.034 0.078 LL 0.391 0.310 0.703 - 0.586 0.579 - 0.552 0.400 REM - 4.846*** - 2.511 - 2.461 - 9.756*** - 7.685*** - 9.734*** 0.64 ID - 0.480*** - 0.496** - 0.485** 0.020 0.042 0.020 - 0.0 A 0.154 - 0.056 Constant -0.372 -15.624 -18.107 40.571*** 25.378 40.372** -18.5
Notes: The table reports the coefficients and an indicator for statistical significance. The dependent variable is
the logarithm of IIT. Columns 1-9 show the pooled OLS regressions for the three sectors. Column 1, 4, and 7
correspond to Equation (3). Column 2, 5, and 8 are the regression results after removing the outliers and the
corresponding country pairs of that year. Column 3, 6, and 9 represent the results by controlling for differences
in unit labour costs. The variables are GDPi, the logarithm of GDP of the origin country, GDPj the logarithm of
GDP of the destination country. DY the difference of GDP per capita between the origin and destination country,
CB is the common border, Lan is an indicator variable for common language, D is the logarithm of geographical
distance, CD is a measurement for cultural distance, LL is a dummy variable for the origin country being
landlocked, REM is a measurement for remoteness of the destination country, ID is a measurement for
institutional distance, and PD stands for the productivity difference between the countries. *** Statistical significance at the 1% level.
** Statistical significance at the 5% level. * Statistical significance at the 10% level
outliers, the significance level of having a common border increases and remains negative. Also, the effect on IIT of the origin country being landlocked becomes negative and significant at the ten percent level. In general, the results in the sectors are not varying significantly, except for the landlocked dummy in the household equipment sector.
Aturupane et al. (1999) stresses that vertical IIT, which is mostly present in the EMU, is more likely to be driven by differences in endowments. This part controls for it by including a variable to proximate labour productivity. Samples of hours worked of all employees, labour costs of all employees and real value-added are needed to calculate labour productivity for the 15 countries. However, the dataset is incomplete and does only provide data for 5 of the 15 countries. Therefore, unit labour costs is used to proximate labour productivity. Unit labour costs is a measure of cost difference per unit of output, and is calculated as total wages paid divided by the total volume of output. Data on unit labour costs are taken from the Conference Board International Labor Comparisons, specifically the dataset on manufacturing productivity and unit labour costs trends. The dataset contains unit labour costs indexes based8 on the year 2002 for nine EMU countries, and the three sectors. A proxy for productivity differences is calculated as:
A = Unit Labour Costsi− Unit Labour Costsj (8)
The results are presented in Table 4 in columns 3, 6, and 9. The coefficients for the IIT in the 9 motor vehicles sector do not change in sign nor significance compared to column 2. The negative effect cultural distance has on IIT in the motor vehicle sector is still significant. By implementing differences in unit labour costs significance levels do change compared to the regression results in column 4. Language stays positive but becomes less significant. However, cultural distance remains insignificant, and I cannot derive a relationship from it. Adding a productivity measure does not change the positive coefficient of cultural distance in the household equipment sector. The effect is higher than in the traditional model, Equation (3), an one unit increase in cultural distance leads to 36.1 percent decrease in IIT in the household equipment sector. Investigating the differences in unit labour costs, one notes that results are not significant. Hence, I cannot derive an economic relationship for this variable.
8 The dataset can be accessed via https://www.conference-board.org/ilcprogram/
9 Table A4 of Appendix A, presents descriptive statistics of unit labour costs differences.
