COLLEGE OF EUROPE BRUGES CAMPUS EUROPEAN ECONOMICS DEPARTMENT IN TRADE WE TRUST The EU home bias puzzle in trade Supervisor : Thierry Mayer Thesis presented by Anya Margaret Ogorkiewicz for the Degree of Master of European Studies Academic Year 2005-2006 Statutory Declaration “I hereby declare that the thesis has been written by myself without any external unauthorised help, that it has been neither presented to any institution for evaluation nor previously published in its entirety or in parts. Any parts, words or ideas, of the thesis, however limited, and including tables, graphs, maps etc., which are quoted from or based on other sources have been acknowledged as such without exception.” 11, 547 words. ii Abstract It is this paper’s contention that the lack of trust between EU countries explains in part the invisible barrier that impedes intra-European trade. An initial summation of the effects both trust and distance have on international trade leads to a discussion of the literature surrounding the phenomenon Obstfeld and Rogoff (2000) dubbed, “the home bias puzzle”. Trust is the oft-ignored underlying variable in economic science. Paramount to economic transactions, trust reduces transaction costs and abridges distances in international trade. Before empirically investigating the effects of the home bias in European bilateral trade flows, we appraise the literature portending to the effects cultural variables have on international trade flows. The analytical chapter assesses the gravitational pull of the Eurozone. It opens on the theoretical trade framework of the gravity model, which we use in our empirical strategy to analyze the European home bias in trade. Consistent with Nitsch’s (2000) study, we calibrate the base gravity model with variables indicating bilateral trust, average origin country trust and average destination trust. After running OLS, GLS, 2SLS regressions on two different databases we conclude that the levels of trust Europeans have towards each other do influence the seeming deficiency of interMember State trade in the EU. If one is to consider mistrust as a market imperfection problem, our study indicates a potential scope for EU-level social policies in the name of consolidating the Single Market. iii Keywords Home Bias Trust Gravity Model European Union International Trade iv Table of Contents Statutory Declaration .................................................................................................... ii Abstract .......................................................................................................................... iii Keywords ....................................................................................................................... iv 1 Introduction ............................................................................................................ 1 2 Breaching the Distance - Trading in Trust .......................................................... 3 2.1 Trust and Distance ........................................................................................... 3 2.1.1 Trust: the underlying variable in economic science ................................ 3 2.1.2 What is Distance? .................................................................................... 7 2.2 3 4 The home bias puzzle, a descriptive approach ................................................. 9 2.2.1 Definitions and Determinants ................................................................ 10 2.2.2 Is the home bias overstated? .................................................................. 11 2.2.3 The impact of trust, culture and institutions on trade flows .................. 15 The Gravitational Pull of the Euro zone ............................................................ 18 3.1 The Underlining theory of the Gravity Model ............................................... 18 3.2 Empirical Strategy ......................................................................................... 20 3.2.1 Data Analysis ......................................................................................... 20 3.2.2 Estimation .............................................................................................. 25 3.2.3 Interpretation .......................................................................................... 27 Conclusion ............................................................................................................ 34 Bibliography ................................................................................................................. 35 Annex 1: Trust Matrix................................................................................................. 39 Annex 2 : Descriptive Statistics .................................................................................. 43 Annex 3: Estimation Output ....................................................................................... 45 v « It can be plausibly argued that much of the economic backwardness in the world can be explained by the lack of mutual confidence » Kenneth Arrow “Gifts and Exchanges” Philosophy and Public Affairs, 1972, p. 357 “Borders? I have never seen one but I heard they exist in the minds of some people.” Thor Heyerdahl, explorer and archaeologist vi 1 Introduction « Nous ne coalisons pas les Etats, nous unissons les hommes »1 declared Jean Monnet at the dawn of the European Union. Half a century later, Europeans are united by a single currency, a single market and a judicial system that punishes deviation from the principle of an “ever closer Union” enshrined in EU law. Scholars still detect however, trade frictions in the EU in the absence of borders, tariffs, and exchange rate risk. This issue forms the subject of this paper. This invisible trade barrier was first termed as the “border effect” to emphasize the cross-border aspect of the phenomenon that impedes trade over national frontiers. Empirical studies illustrate that, given both their level of economic activity and the distance separating trading partners, countries tend to trade less with foreign lands than expected. Subsequently scholars have enlarged the scope of this puzzle, calling it the ”home bias effect”, which implies that countries may prefer to trade with culturally and institutionally similar countries. High stakes are involved in resolving the home bias puzzle: if the cognitive distances between countries can be shrunk, the subsequent increase in bilateral trade would benefit all trading partners. This paper investigates the bilateral trade patterns within the EU15 – the 15 European countries that participate in the European Single Market Programme and share a common currency, eliminating from the outset two “natural” causes of the home bias effect: exchange rate risk and national tariffs. These and similar rationalizations fall in line with today’s orthodox theories. It is this paper’s contention however, that “transaction cost economics” is more adept at assessing the credence of the explanation we propose: trust or the lack thereof. This paper seeks to determine whether it is the lack of trust between EU countries that explains at least in part the invisible barrier that impedes intra-European trade. The first chapter examines the definitions and determinants of both trust and distance as they relate to trade, before introducing the home bias puzzle. Beyond the puzzle’s definition and “common causes”, it is necessary to investigate the literature assessing its impact as well as the literature arguing that the home bias puzzle is wildly overstated; before then reviewing the relevant literature concerning the impacts of trust and culture on international trade flows. 1 Discours Washington, 30 avril 1952 1 The second chapter comprises an analytical investigation of the effects of trust on bilateral EU15 trade levels, by first specifying the basic structure of the gravity model, the underlining theory; and the relevant equations. After describing our empirical strategy, the paper concludes with an interpretation of the results obtained. 2 2 Breaching the Distance - Trading in Trust Our main inquiry relates to the extent that trust may breach the distances between trading partners. The first section begins by defining the two most important variables to our study: trust and distance. The second part introduces the home bias puzzle through the framework of transaction cost economics, and is concluded by an overview of previous literature studying the impact of cultural variables, such as trust, on international trade flows. 2.1 Trust and Distance To quote Kenneth Arrow (1972), “virtually every commercial transaction has within itself an element of trust, certainly any transaction conducted over a period of time.”2 International trade is so rife with unforeseen and insecure contingencies that, without a certain degree of reciprocal trust, trade between two dissimilar parties would surely be much lower. Trust is however, rarely taken into account in orthodox trade theories. Rather, it is often abusively assimilated to cultural priors which cannot be easily assessed through orthodox theory. Trust must therefore be disentangled from cultural variables, before it is possible to analyze the underlying notion behind trust or lack thereof. If individuals trust each other, it is because there is a certain affinity or connection between them that is conditioned by the subjective distance, agents judge between themselves. The notion of distance therefore, constitutes a natural bridge between the concept of trust and that of the home bias puzzle. 2.1.1 Trust: the underlying variable in economic science What is trust? According to the Merriam-Webster Dictionary, trust is defined as: (1 a): an assured reliance on the character, ability, strength, or truth of someone or something. 3 Trust is a bilateral concept that refers to a subjective relationship between two or more agents. In economics, trust is paramount to economic transactions. According to Sandelien (2003) economic trust is “one agent’s sentiments of expectations towards another agent’s positive behavior with respect to oneself, in a situation involving 2 3 Kenneth Arrow, (1972). Gifts and Exchanges. Philosophy and Public Affairs, p. 357 http://www.m-w.com/ 3 risk.”4 It is, for example, the determining element of the confidence individuals have in their money, that “businesses must believe that their banks look after their money. Individuals must be convinced that the government will protect their property right. These – and many other types of trust – are prerequisites for a modern society”. 5 Bornhorst et al. (2004) touch the essence of economic trust when they state that preceding each economic transaction, agents must first select a transaction-worthy partner. This choice of partners is notably determined by “the agent’s beliefs about the prospect of building trust and reciprocity with potential partners [and] experience will play a role as well.” 6 In behavioral economics, trust is depicted in a game with multiple interactions, where the agents discount their future gains by a certain factor. The rational explanation for trust is that “individuals are inclined to trust each other due to pay-offs in the long run.” 7 Both agents use the “trigger strategy,” where they choose to cooperate up to the point where one of the agents fails to do so, switching both agents’ strategies to noncooperation. The frequent interaction aspect of this game portrays a learning curve where trust increases with each period. Without an initial level of trust however, the interaction cannot begin, which suggests that a different source of trust exists. The sociological model of trust by comparison, strives to explain the preliminary faith individuals show towards each other. Trust is viewed as a public good that is reflected in the individual’s identity, whether cultural, moral or based on social obligations. In behavioral economics, agents that display this type of trust are altruistic, for an altruistic agent increases his welfare by increasing the welfare of the other agents. The norms and values that guide the individual’s decision are based on a certain lack of diversity amongst players. Such trust exists when a player identifies with the other’s ethic, religious, political, familial, (etc…) insignia and acts accordingly. Bornhorst et al. (2004) recognize that diversity has a “substantial impact on agents’ initial beliefs regarding partners as well as the evolution of their interactions over multiple transactions.” 8 This diversity, reflected through a person’s cultural affinity, determines the initial level of trust. 4 Sandelien (2003) p. 3 ibid. p.1 6 Bornhorst et al. (2004) p.1 7 Sandelien supra note 4 8 Bornhorst supra note 6 5 4 2.1.1.1 Hofstede’s four cultural dimensions Linders et al. (2005) note that cultural distance is conventionally assessed through Hofstede’s (1980) four dimensions of national culture. 9 Hofstede (1980) analyzed survey data from 116,000 IBM employees in 40 countries and ranked each country on a scale from 0 to 100 indicating how people from different cultural backgrounds perceive the following four societal dimensions: (a) Power distance –How different people accept an unequal distribution of power and status as a way of organizing social systems. (b) Uncertainty avoidance – The degree to which people are willing to trade a high risk/high reward situation for low risk/ low reward. It signifies “the extent to which people are uncomfortable with uncertain, unknown or unstructured situations.” 10 (c) Individualism versus Collectivism – A society’s emphasis on the role of individualism and collectivism: notably experienced through the type of wage bargaining in place in a given society. (d) Masculinity versus Femininity – The extent to which a society assesses the importance of one type of stereotypical value over another. Hofstede’s (1980) measures of cultural distance rely on the degree of uncertainty avoidance that characterizes a given population. His uncertainty avoidance index (UAI) – previously employed in assessing trade flows for perceived cultural distances – may explain the degree of trust underlying international trade patterns. 2.1.1.2 Sandelien’s five factors that determine trust Sandelien (2003) identifies five factors that determine trust and regroups the factors in three categories: predictability, controllability and interdependence. Within the first category, he cites two factors: (a) reputation/history of interaction/information and (b) similarity/identity/moral bonds. (a) Reputation: refers to a history of past interaction that provides information about the other party’s consistency over a long period of time. Trust can be Linders et al. (2005) note that Hofstede had subsequently uncovered a fifth cultural dimension, “longterm orientation” but due to the difficulty of empirical applicability and scholars questioning its added value, it has been left out of our discussion of cultural dimensions. 10 ibid. pp. 3 9 5 institutionalized through a middleman acting as a “bridge of trust”11 where the two agents trust the middleman more than each other. (b) Similarity: of ethnicity, religion, social belonging, and gender are “subjective sentiments of perceived equality”12. This trust -inducing factor is based on the assumption that someone similar will not act opportunistically. In essence, a similarity in religion or social belonging may outwardly signal that certain norms and values are shared, which in turn may create an initial level of trust. As an example, Sandelien (2003) cites the Japanese business culture where “a history of trusting business norms [established] a ‘culture of trust’ in the culture pattern.”13 The controllability category consists of the dyadic aspects of contracts and sanctions. (c) Contracts/sanctions: determine trust by establishing control and transparency, most notably when sanctions are automatic and non-biased towards either party. The final category of interdependence presents two final variables: (d) Altruism: The situation in which one party has interest in the well-being of the other party: often present in family relationships. According to the author, emphasizing common goals can also be assimilated to this notion. (e) Asymmetry: One party has an upper hand or outright power over the other party. Sandelien (2003) doubts whether asymmetry is a determinant factor of “trust” despite it leading to cooperation, since asymmetry corresponds rather to a situation of fear or bondage towards the stronger party. The notion of trust thus overlaps cultural variables, since it draws from both rational and sociological sources. Orthodox theory struggles with the concept of trust, since the neoclassical world operates on the strong assumption that transactions are costless actions, and market imperfections are at most ephemeral in nature. According to Sandelien however, mistrust is a market imperfection problem, and one that cannot be resolved by equilibrating supply and demand. 11 Sandelien, supra note 4, p. 7 ibid. 13 ibid. 12 6 Building on Coases’s (1937) foundations that firms were created in order to internalize transaction costs, Willliamson (1998) developed the theory of transaction cost economics: the “science of contracts”. Since contracts are inevitably incomplete, trust can complete contracts by diminishing transaction costs. How does trust diminish transaction costs? Search costs, the costs of matching supply and demand, can be diminished if a company operates within a trustworthy network of business relations, or if it trusts in another’s reputation. Trust reduces negotiation costs, the costs of concluding a contract, because when the two parties to the contract trust each other, their contracts need to be less exhaustive on prices, qualities and sanctions. Furthermore, it diminishes control costs, the costs of monitoring an agreement, when a trusting relationship between buyers and sellers guarantees quality, delivery time, etc. Trust as reflected in Sandelien’s three categories acts as a safeguard against opportunism, and breaches the cognitive distance that separates agents involved in an economic transaction. Distance is indeed an essential concept to our paper; after assessing this notion as a multidimensional concept we will discuss whether distance has been abridged in a globalized world. 2.1.2 What is Distance? Both the impact of trust and the existence of a home bias are conditioned on distance, whether perceived or real. Distance is however, a broader concept than Hofstede’s uncertainty avoidance index suggests, with subjective aspects present within all notions of trust, the home bias puzzle and the gravity equation that will be used to estimate the effect of trust on the home bias puzzle. 2.1.2.1 Distance - a multidimensional concept Distance is a fluid concept. For instance, international metropolises are closer to each other than to secondary cities despite the actual distance covered “as the crow flies”.14 Beyond actual mileage, Gatrell’s (1983) classification distinguishes between four different types of distances: - Time Distance: the time required to travel between two entities; 14 An example of which is the distance between Paris and London as compared to Paris and Strasbourg. Respectively 343 and 397 kilometers separate these cities, and yet by train it takes 2 hours 30 minutes to reach London from Paris as compared to the 4 hours and 30 minutes it takes to reach Strasbourg from the same point of origin, and this despite London being separated by borders both political and geographical. Sources: http://www.eurostar.com/ and http://www.voyages-sncf.com/ 7 - Economic Distance: the costs of traveling from one area to another; - Cognitive Distance: the subjective perception of distance; and - Social Distance: the distance between networks or social classes could be assimilated to Hofstede’s cultural distance. These four types of distances, independent of actual longitude and latitude, can thus be expanded or contracted according to both perceptions and technological advances, such as transportation improvements. The next paragraph examines distance reducing factors and whether globalization has had an impact on cognitive distances. 2.1.2.2 Do networks and ICT contract distances? Travel time and cost influence the cognitive perception of distance - and thus of trust - by making foreign entities physically more accessible. Two other relevant variables however, influence the perception of distance: the cultural networks existent in the foreign land and the penetration of information and communication technology (ICT) – both consequences of globalization. Rauch (2001) argues that unlike standardized goods bought on organized markets, the price of differentiated goods, of variable quality, has to be negotiated. The risks of moral hazard are substantial, raising the stakes in choosing the appropriate trading partner. National and regional institutions reduce costs associated with researching business opportunities and potential partners. In global environments however, where language and cultural barriers subsist, ICT is said to further diminish information costs. However, the Internet cannot yet check if your chosen partner is trustworthy. Leamer and Storper (2001) suggest that the Internet makes long-distance “conversations” not “handshakes” – in sum, ICT can’t start business relationships but can only maintain them. Brun et al (2002) tested the idea that communication only strengthens already existent relationships, and found that, long distance trade between new partners is unaffected by ICT. More specifically, they observed that distance is increasing between poor countries, but elasticity to trade between rich countries is decreasing. Globalization is therefore marginalizing poor countries, whereas it reduces the cognitive distance between rich countries. One solution is to induce trade within cultural networks where risk of opportunism is lessened. 8 According to Rauch (2001), “an economic network can be defined as a group of agents that pursue repeated and enduring exchange relations with one another.” 15 Trade within co-ethnic networks, such as those spread through globalized migration, heavily depend on two key factors: information and trust. Information flows concerning markets, participants and the trustworthiness of external partners to the network, reduce search costs and allow for international arbitrage. Trust is determined through frequent interactions, knowledge transfers about past conducts, and various forms of conduct monitoring. Network-maintenance however, demands frequent interactions and broad information channels. Consequently, less interaction and poorer information, due to the breadth of distance separating partners in the same network, reduce trust levels. Furthermore, studies have discerned that whereas within networks trust is increased, between co-ethnic networks, the sentiment of trust is lower. This is because co-ethnic networks are predominately local in character, indicating that instead of increasing trade over long distances they may actually inhibit it. If the distance effect were substantially affected by communication and technological improvements, then empirical studies would show that the distance elasticity has reduced over time. Yet Rose (1999) demonstrated that despite significant technological progress, the elasticity of distance to trade has not significantly diminished over the last twenty years. Neither ICT nor trade within established networks significantly bridge the distance between trading partners. From the viewpoint of empirical investigation, distance is a difficult concept to gauge. The CEPII database makes different measures available for indicating distance between national borders, distance between capitals, and the distance between the two largest cities, conditioned or not by the percentage of urban population. The correct measure of distance is essential when studying the home bias effect. 2.2 The home bias puzzle, a descriptive approach Trefler (1995) and McCallum (1995) have demonstrated the breadth of the home bias phenomenon. Obstfeld and Rogoff (2000) identified it as a major puzzle in economics today, and Guiso, Sapienza, and Zingales (2004) initiated work on the effect trust and culture have on the home bias. We will initially define the home bias and 15 Rauch (2001) cited in Sandelien supra note 4 9 provide the most common justifications for it, before reviewing the afore-mentioned studies. 2.2.1 Definitions and Determinants 2.2.1.1 Definitions Deardorff's Glossary of International Economics defines a “home bias” as “a preference, by consumers or other demanders, for products produced in their own country compared to otherwise identical imports.” 16 Trefler (1995) conceived the term, proposing it as an explanation to his “mystery of missing trade” where he observed that the amount of trade is far less than predicted by the Heckscher-Ohlin model, particularly regarding trade in factors of production. An assimilated notion to the home bias is the “cultural bias”, which indicates the consumers’ preferences for products produced in their own cultural sphere. Guiso, Sapienza, and Zingales (2004) define “cultural biases” as “the role customary beliefs have on trust”17, culture being, to paraphrase Einstein, “the collection of prejudices acquired by age eighteen18”. Obstfeld and Rogoff (2000) identified the “home bias puzzle” as one of the six major puzzles in international finance today, sparking a debate attempting to explain the phenomenon. Before assessing whether the home bias has been overstated, it is necessary to review its common explanations. 2.2.1.2 Determinants of home bias Hillberry (2000) offers four possible explanations for the home bias effect between the U.S. and Canada that can be extended to account for all instances of home bias. He first suggests that the home bias effect may have its roots in the hidden costs of trading internationally, which are individually too minute to be taken into account but, when aggregated, could amount to a substantial barrier to trade. These additional expenses may include exchange rate costs, transportation costs and the cost of additional administrative work. This implies a large scope for aggressive governmental action in terms of coordinating regulatory, transportation, monetary policy issues, etc. The plausibility of this explanation is nevertheless quite low since this hypothesis is not borne out by empirical testing: “ it appears that hindrance costs would have to have an 16 Available at http://www-personal.umich.edu/~alandear/glossary/ Guiso, Sapienza, and Zingales (2004) p. 3 18 ibid. p. 2 17 10 effect equivalent to a tariff of between 60 and 200 percent. These costs seem implausibly large, given that they appear to have escaped detection (…)”.19 The second justification draws from the theory of comparative advantage that illustrates trade between two countries. In the case of the U.S. and Canada, these two similar-good-producing entities don’t trade with each other because there is an insufficient amount of mutually advantageous trade to be captured. The home bias therefore, isn’t a harmful aspect of international trade, but a natural one that may arise between any similar good producing countries. The third explanation states that home bias exists due to the fact that the composition of international trade is different from that of domestic trade. This strikes at the heart of econometric models that study the bias, since the core assumption of the gravity model is that countries export the same basket of goods. Here we’re considering the case where the composition of the basket changes. This explanation has received considerable empirical support. Hillberry relates the main findings, notably that the composition of trade does depend on the existence of a border, the distance between the regions and the “output composition” in the destination region. Finally, Hillberry touches upon the explanation which is of the most interest to us: The home bias may exist because consumers have national preferences and they may thus be willing to pay a premium in terms of transport costs - non negligible in Canada - to buy from distant countrymen than to buy from nearby foreigners (for reasons which are not taken into account by orthodox economic theory). For policymakers, this implies a limit to the competition domestic producers face from imports, since the price differential must be substantial for consumers to substitute domestic products for imports. Hillberry notes that cultural preferences may hold true for retail consumption, but for wholesalers or manufacturers, one expects them to be less sensitive to the existence of cultural preferences that prevail beyond the border. Beyond these four explanations there exists a fifth explanation, considered below: what if that the home bias is greatly overstated? 2.2.2 Is the home bias overstated? After McCallum’s (1995) study on trade levels between Canada and the USA, many have countered his results by attempting to prove that his bias was overestimated. 19 Hillberry, (2000) 11 As Hillberry notes, if McCallum is correct, such a barrier to trade is equivalent to a tariff between 60 and 200 percent, a tremendous obstruction to trade. Before assessing whether the home bias effect is overstated, it’s necessary to review McCallum’s pioneering work. 2.2.2.1 The home bias effect Home bias effect is estimated by means of the “gravity model”, which relates the volume of bilateral trade to the distance between two countries and their economic size. At the dawn of the North-American Trade Agreement, which allowed a free movement of goods between Canada and the US, McCallum, using 1988 data for ten Canadian provinces and 30 US states, obtained the following regression, standard errors being in parenthesis: (1) log ˆ( xij ) 1.21log( y i ) 1.06 log( y j ) 1.42 log( d ij ) 3.09border (0.03) (0.03) (0.06) (0.13) Where x ij is the volume of bilateral exports, yi is the GDP of the region of origin, y j is the GDP of the destination region, d ij is the distance between region, and border is a dummy variable indicating whether the two trading entities share a common border. His results indicate that when the GDP of one of the trading partners increases by 1%, the trade volume increases more than proportionally. As expected, sharing a common border increases trade (by 851%). Moreover, two regions distanced by 500 miles, trade 267% more than if 1000 miles separated them, ceteris paribus. Controlling for trade amongst entities themselves (ie: Canadian province with Canadian province, or American state with American state) McCallum finds that “trade between two provinces is more than 20 times larger than trade between a province and a similar sized state, each pair having the same distance [separating them].”20 While the idea that national borders impeded international trade is not surprising, the magnitude of the border effect most certainly is. Obstfeld and Rogoff (2000) have hailed this observation as “a dramatic suggestion of segmentation.”21 20 21 Sundelien, supra note 4, p. 26 Obstfeld, Rogoff (2000) p. 342 12 Following McCallum (1995), Wei (1996) reconstructed national trade data for OECD countries (to account for the “border effect”) by defining national trade as the difference between national production and exports. His estimations found that countries trade just 2.5 times more with themselves than with others. Helliwell (1998), using Wei’s definition of home trade, re-estimated McCallum’s (1995) regressions however, with data from 1988 to 1996, and found that trade between two provinces remains approximately 12 times larger than trade between a province and a similar sized state. Nitsch (2000) estimates the home bias by specifying a dummy variable “Home” that takes the value of 1 for intranational trade flows and zero for international flows. He recalibrates McCallum’s (1995) regression with data for bilateral trade in the European Union and obtains the following regression: (2) log (ˆ xij ) 1.92 Home 0.67 log( yi ) 0.71log( y j ) 1.07 log( d ij ) (0.20) (0.04) (0.04) (0.08) Unlike McCallum, he finds that a 1% increase in GDP of a trading partner increases the trade volume by less than 1%. He finds however, the European home bias to be extensive: within the European Union, “a country exports about 6.8 (=exp[1.92]) times as much to itself as it exports to a foreign country of similar economic size and distance”. 22 Adding dummy variables to measure border adjacency and common language, his regression becomes: (3) log ˆ( xij ) 2.59Home 0.74 log( yi ) 0.75 log( y j ) 1.82 log( d ij ) 0.42 Adjacency 0.48language (0.23) (0.04) (0.04) (0.09) (0.16) (0.25) … indicating that after controlling for a common language and border, the home bias effect in the EU jumps: a given country exports about 13.3 (=exp[2.59]) times as much to itself as it exports to a foreign country of similar size and distance that does not share the home countries language or an adjacent border. The subsequent sections of this paper build on Nitsch’s findings, with particular emphasis on his observation that such an important home bias effect exists within such a highly integrated trading entity as the European Union. However, many scholars 22 Nitsch (2000) p. 1099 13 argue that the home bias has been overstated by either mismeasurements or poor data analysis. Their analysis contributes to the scholarly debate about the home bias puzzle. 2.2.2.2 How is the home bias overstated? In addition to contributing to the methodology 23 , Wei (1996) specifies a “remoteness” variable, taking into account the relative distance of trading partners, capturing third country effects 24 . Wei (1996) finds that the border effect between OECD countries does exist, although it is of a magnitude of 2.5 – considerably lower than McCallum’s results. Anderson and von Wincoop (2001) include Wei’s remoteness variable in their study, as well as a “mutual resistance variable”, i.e. the respective price levels that are dependant on trade barriers such as tariffs. They argue that economically smaller countries, such as Canada, are subject to bigger border effects than large economies for, unlike large economies, a small variation in international trade can lead to a large difference in trade within the small country. They find that a 1% drop in international trade leads to a 60 % increase in Canadian inter-province trade, but only to a 25% increase in American inter-state trade. They challenge McCallum’s results, alleging that they were due to an inherent bias, brought on by the variables he omitted, and the relative small size of the Canadian economy as compared to the American one. The measure of distance plays a central role in home bias estimations: Distance is used in the gravity model as a proxy for freight, time, and other costs. Moreover, the border effect is sensitive to both internal and external distance specification. Head and Mayer (2002) recapitulate the measures employed in recent literature and argue that distance mismeasurements overstate the home bias phenomenon. To measure internal distances, Wei (1996) uses one fourth of the distance to the nearest neighbor. This type of measure relies on the distance from one country’s economic center to another25, and was strongly criticized in Nitsch (2000) who argued that it assumed that capital cities are of the same distance to their respective borders. Instead, Nitsch proposes an area-based measure based on the radius of a disk. Moreover, scholars have also defined distance as sub-unit based weighted averages, taking natural geographical units, and weighting them by either their population or their wealth. By constructing a border effect that doesn’t rely on national data. ie: taking into account if a bigger trade entity is closer. 25 An economic capital being a capital city or one with a concentrated amount of economic activity. 23 24 14 Head and Mayer (2002) conclude that the closer the nations are to each other the more these existing measures inflate distances. They estimate a new distance, a variation on the “sub unit based weighted averages” measure, in an attempt to neutralize the border effects. Their findings indeed shrink the border effect, notably in the EU during the years 1993-1995 where they find a border effect of 3.12, but they do not succeed in eliminating it, proving the point that although all measures of distance suffer from “distance inflation”, the border effect is not illusory. While border effects therefore, do exist, their exact scale is subject to scholarly debate. If the European Single Market Programme, sans tariffs and exchange rate risks, is nonetheless subject to trade frictions, scholars have begun testing cultural hypothesis such as the effects of public opinion, of cultural distances and of trust as explanatory variables for this seeming lack of European trade. This paper ultimately seeks to analyze the effect of trust on inter-European trade levels. Guiso, Sapienza, and Zingales (2004) find that trust has a robust impact on European trade flows. The concluding section overviews the relevant literature, concerning the effects trust and other cultural variables have on international trade flows. 2.2.3 The impact of trust, culture and institutions on trade flows Culture seems at first a vague notion: difficult to measure and difficult to empirically test. The causality between culture and economics was also deemed hazy; it was assumed to fluctuate over a lifetime, or at least alter according to personal experiences. Becker (1996) noted however: “Because of the difficulties of changing culture and its low depreciation rate, culture is largely a ‘given’ to individuals throughout their lifetimes.” 26 An invariant notion of culture allows analysts to instrument it with other hereditary aspects, like ethnicity or religious denomination – as Guiso et al. (2006) noted, “[This] prevents cultural explanations from becoming simple ex post rationalizations”27 Concerning the home bias more specifically, Grossman (1996) argues for “a model with imperfect information, where familiarity declines rapidly with distance.”28 26 Becker (1996) p. 16 Guiso et al. (2006) p. 3 28 Grossman (1996) 27 15 Huang (2005) specifies such a model by assuming that risk aversion influences the degree of familiarity populations feels towards each other, given the same amount of information – “the same distance ‘looks longer’ in the eyes of uncertainty averse countries.”29 Using Hofstede’s (1980) international survey data and controlling for the previously documented causes of trade frictions,30 the author confirms that high riskaverse countries trade disproportionately less over long distances than low risk-averse countries. In determining international trade flows, cultural factors are as important as geographic ones. Guiso et al. (2004), using the available the Eurobarometer data, identify three elements of trust: the average level of a given country’s trust, the average level of other countries’ trust towards that given country, and the dyad-specific level of trust of one country towards another. They instrument their trust variable by using the cultural determinants of trust (genetic distance and religious commonality) and find that a 1% increase in trust increases inter-European trade levels by 30%. Disdier and Mayer (2005) follow up Guiso et al.’s (2004) work, again using the Eurobarometer data, and study the impact of “cultural affinity” on the trade flows between EU15 and the 10 new member states. They find that the causality goes one way: although an increase of imports does not significantly influence public opinion, a rise in public opinion does influence the level of imports. These studies have ascertained that cultural distances and low levels of trust have a negative impact on trade flows. Linders et al. (2005) however find the reverse result. In their empirical investigation, they include measures of cultural effects (such as common language, common religion) as well as institutional and cultural distances. They find that although the institutional distances have a negative effect on trade, cultural distances have a positive effect on trade. The authors explain this seemingly counter-intuitive result by making a distinction between trade with culturally distant countries and host production in culturally distant countries. They argue, “firms prefer trade to host-country production in culturally distance countries.” 31 Although we expect to find cultural distances to have a negative effect on trade levels, we are now aware that this result is not necessarily established. Nonetheless, cultural variables have 29 Huang (2005) p. 3 His explanatory variables include: origin and destination dummies, country-pair characteristics, fixed year effects dummies and UAI: Hofstede’s (1980) uncertainty avoidance indicator. 31 Linders supra note 9, p. 19 30 16 an incontestable impact, be it positive or negative, on trade flows and explain in part the home bias effect scholars observe in world trade patterns. The above discussion has illustrated that contrary to classical assumptions, trust is an important variable to be taken into account when analyzing the home bias puzzle. Even in a highly integrated area such as the European Union, trade frictions exist, although its magnitude is subject to scholarly debate. The following section assesses the gravitational pull of the euro zone, and how far trust goes in explaining the home bias effect. 17 3 The Gravitational Pull of the Euro zone Sir Isaac Newton defined in the 17th century the Law of Universal Gravitation, which contends that the attraction between two given objects i and j is proportional to the product of their masses, and inversely proportional to the square of the distance between them. Sandelien (2003) affirms that the first use of the theory of gravity in a social science context is credited to Carey who in 1858 applied it to the group behavior phenomenon. In trade economics, Beckerman (1956) was the first to transpose the gravity model to study trade patterns. Yet no theoretical foundation for the gravity model was found until Anderson’s 1979 breakthrough. In this second chapter, we will highlight the impact of the trust variables as we analyze the gravitational pull of several contiguous countries of the euro zone. In the first part, we will provide a theoretical basis for the gravity equation, and state the equation we wish to estimate. The second part will proceed with the regressions and conclude on the credence of our explanatory variable: trust. 3.1 The Underlining theory of the Gravity Model The gravity equation is a spatial interaction model that illustrates the two main factors that influence bilateral trade: the size or scale of the given economies and the distance between them. The scale impact alludes to the empirical observation that countries that carry a large economic weight attract more economic activity than (economically speaking) lightweight countries32. The distance effect conveys the idea that the further apart the two economic bodies lie; the lower are the observed trade flows. The interaction between any two countries is the bilateral trade flow, subscripted by i and j, and expressed as the multiple of the two given GDPs conditioned by the distance separating the two trading partners. Anderson (1979) was first to link the gravity equation, empirically validated, to a theoretical international trade framework. Subsequent scholars have revised and augmented this initial breakthrough; as for instance did Deardorff (1995). Deardorff (1995) employs a two-country Heckscher-Ohlin model supplemented by three primal assumptions. His first assumption is that each good is 32 This economic attraction could be explained by the difference in importance of business opportunities or size of home markets. 18 exported/produced by only one country 33 , which stands in for the orthodox product differentiation assumption. His second assumption is that trade incorporates both goods and factors of production. Given his third assumption, positive transportation costs, the prices of the production factors are not equalized, as neoclassical theory would have it. Instead the transportation costs between country i and country j (“ t ij ”) are of the “iceberg” type, meaning that despite that a fraction (“ t ij 1 ”) of the shipped goods were lost in transportation, the importer country’s consumers’ pay the full price “ p i t ij ” for the shipment. Deardorff (1995) specifies Cobb-Douglas type preferences, meaning that consumers in both countries spend a fixed ratio i of their respective incomes on a given country’s good. Let xi be the output of country i, let Yi be the nominal income of country i, and let Y W be the world income. Then, Yi pi xi i Y j i Y W (4) j …which rewritten gives: i Yi YW . If Tij denotes the nominal value of exports from country i to country j, then in the absence of transportation costs, the nominal value of country i’s exports is thus: (5) Tij i Y j Yi Y j YW With positive transportation costs, the nominal value of exports becomes: (5.1) Tij Yi Y j t ijY W …which closely mirrors the basic gravity equation: (6) Exportsij (GDPi GDPj ) ( DISTij ) The two notable differences between Anderson’s and Deardorff’s theoretical foundation are that the gravity model’s elasticity of exports to GDP, denoted by alpha, is not present in Deardorff’s (1995) framework due to the Cobb-Douglas preferences he specifies. This can be remedied by incorporating constant elasticity substitution (CES) 33 This is equivalent to the assumption used by Anderson (1979) that consumers differentiate goods by country of origin. 19 preferences, which also allow for the fiction variable, the elasticity of bilateral trade to distance, beta, to immerge. The natural logarithm of the previous equation, provides a linear relationship: (7) ln Exportsij 1 ln GDPi 2 ln GDPj ln DISTij Adding an error term, u i allows for the gravity equation to be estimated: (8) ln Exportsij 1 ln GDPi 2 ln GDPj ln DISTij u i Equation (8) is the basic framework of the gravity model – it is identical to mcCallum’s regression in equation (1). To this structure, we add explanatory variables, the most common being a dummy indicating a shared border. The gravity model this paper estimates will most notably include variables indicating trust. 3.2 Empirical Strategy We first describe our data, its sources and run basic descriptive analysis on our samples; before specifying our model, underlining hypothesis, and our estimation methods. A discussion of the results our model yields concludes this study. 3.2.1 Data Analysis 3.2.1.1 Sources Our data comes from five sources. The bilateral trade data, converted to the relevant units and exchange rates, comes from the United Nations COMTRADE database. Our GDP data 34 has been extracted from the Eurostat “National Accounts” database - it is in millions of ECU (in current prices). The distance indicator, as well as our dummy variables for common language and common border, has been retreived from the CEPII data library. Our dummy variable “religion” was complied from the CIA World Fact Book. Finally, our trust data was made available courtesy of www.gesis.org, a site that provide Eurobarometer data. 34 As well as the data for every county’s total exports, used to calculate internal trade à la Wei (1996). 20 3.2.1.2 Bilateral Trade Description of variables and our sample From the United Nations COMTRADE database, we have extracted the annual exports of the EU15 countries. We have transformed this data from thousands of US$ (in current prices) to millions of ECU (in current prices)35 to be harmonized with our GDP data. To construct internal trade flows, we followed Wei’s (1996) approach and subtracted the total exports of every country from its total production, and this for every year and country in our series. GDP and Exports The GDP data in current prices was retrieved from Eurostat – besides specifying the relevant years and countries, this data didn’t need any further transformation. Data for exports were used in constructing internal trade flows. Distance and Dummies for common languages, adjacent borders and religion The religion variable indicates a shared (majority or state) religion, as recorded in the CIA World Fact Book. Instead of the variable indicating a shared official language, we chose to use the one indicated a shared ethnic language for it shows whether a given population can understand the foreign language despite the language’s official status in the home country. Taking into account our previous criticism on distance mismeasurements, we use the distance indicator that we consider the most realistic -“distwces”- given that it is weighted by the geographic dispersion of the population within each country. Trust We initially wanted to study the interaction between trust and the home bias in the 21st century but appropriate trust data for those dates are unavailable either because they are under embargo or because the Eurobarometer surveys did not include the relevant questions for those years. Given that the most recent data for trust ends in 1997, we have requested data for 1993, 1994, 1995, 1996 and 1997. Upon delivery, we noticed several discrepancies, notably that three different questions have been asked concerning the level of trust Europeans have towards each other. In 1994, the survey asked, “Which, if any, European Union country or countries do you think can be politically trusted more than others?” to which respondents list trustworthy countries. The 1993 and 1996 survey asked, “I would like to ask you a question about how much trust you have in people from various countries. For each, 35 The past rates have been found on http://www.xe.com/ict/. 21 please tell me whether you have a lot of trust, some trust, not very much trust or no trust at all?” The 1995 and 1997 survey asked, “I would like to ask you about how much trust you have in people from various countries. For each, please tell me whether you tend to trust them or tend not to trust them?” The 1994 data has been dropped since it explicitly refers to political trust and not trust towards a foreign population. Following Guiso et al, we have encoded the 1993 and 1996 data as follows: 1 (=no trust at all), 2 (=not very much trust), 3 (=some trust) and 4 (=a lot of trust). The 1995 and 1997 survies that gave a binary choice to respondents has been coded as 1 (=tend not to trust them) and 2 (=tend to trust them). Although the questions are similar, they do not provide for identical answers, so we cannot combine the four years together. Therefore, we will compare the surveys two by two. Sample In 1993, twelve countries were surveyed. To compare this study to the 1996 one, we dropped from the later the countries which were not surveyed in 1993. The 1995 and the 1996 surveys included all EU15 countries. However, since our bilateral trade data aggregates Belgium and Luxembourg, the latter country has been dropped from all our samples. 3.2.1.3 Descriptive statistics Before running the descriptive statistics on all our variables, the trust variable deserves particular attention. As we mentioned, we aggregated the raw trust response data following Guiso et al.’s encoding, computing the mean value per country and per year. For each year, we have built a symmetric trust matrix, included in our annex. Just as the previous scholars did, we notice substantial variations in our data. For 1993 and 1996, the maximum values were respectively 3.547 (the level of Danish trust towards countrymen) and 3.588 (the Swedish trust towards countrymen). In each year, those countries were the most trusting towards foreigners. The minimum values for 1993 and 1996 were respectively 2.133 (the level of Greek trust towards Germans) and 2.00 (the level of Portuguese trust towards Greeks). Once again, each of these origin countries registered the weakest level of trustiness. For 1995 and 1997, the maximum values were respectively 1.990 (the level of Danish trust towards countrymen) and 1.976 (the level of Dutch trust towards Swedes). Unlike in the previous tables, the Danes in 1995 aren’t the most trusting country (average of 1.821) for France, the Netherlands and Sweden register larger values. In 22 1997, as expected, Sweden is once again the country that registers the highest levels of trust towards foreigners. The minimum values for these years are respectively 1.276 (the level of Greek trust towards the Germans) and 1.298 (the level of Greek trust towards the British), and once again Greece is the country that registers the lowest average levels of trust. There seems to be a cursory correlation between trusting and being trusted. Just as Guiso et al (2004) observe, and as reflecting in the main diagonal of each trust matrix, citizens trust their own countrymen more, the main exception being Scandinavian countries, which, as in the case of Sweden in 1997, trust its countrymen by a value of 1.949, but trust Danes and Austrians more (respectively 1.959, and 1.957). “This is hardly consistent with differences in trust being driven by differences in information concerning the other countries citizen (which should decay with distance), but can be explained by opinions on trustworthiness being highly affected by cultural sterotypes.”36 Another interesting result is in the levels of bilateral trust between the UK and France, hereditary enemies. In all the years we surveyed, they still rank each other’s countrymen as the least trustworthy of all Europeans. Finally, we observe that the average level of trust seems to be decreasing over the years, although we cannot definitely conclude on this, due to our small period sample. We have consolidated the data for 1993 and 1996 in one database, and the data for 1995 and 1997 in another. The results for our descriptive statistics for all variables are in the annex. We ran the descriptive statistics on all non-string variables for both databases, first differentiating by year, then by year and by country. There are 121 observations for the years 1993 and 1996 (making a total of 242 observations in whole), and 196 observations for the rest of the years (392 observations in total). The results are in the annex. For all periods, the minimal registered level of bilateral trade corresponds to Greek exports towards Ireland, which also coincides with the maximum recorded measure of distance in our database. Despite the same breadth of distance separating them, Ireland exports more to Greece than vice versa. For all years, the minimum distance recorded is the internal Greek distance, and the maximum level of trade corresponds to internal German trade (with the sole exception of 1993, where internal French trade is the highest). To check whether there has been a computing error in our 36 Guiso et al (2004) p. 11 23 samples, we verify that the mean GDP of origin countries and of destination countries are identical for all years. We repeat the test for the average level of bilateral trust for origin and destination countries, and notice that for 1993, the means of average trust levels are not identical. Furthermore, 1993 is the only year that records a higher standard error for average destination trust than for average origin trust. There is a probable computing error in our 1993 sample, which brings us to reject the 1993 data. For 1995, 1996 and 1997, we notice that the standard deviation in average destination trust is lower than the standard deviation in average origin trust, indicating that although European countries record different levels of national trust, they most concur in their opinions of other European’s trustworthiness. This falls in line with Guiso et al’s hypothesis that “opinions on trustworthiness [are] highly affected by cultural sterotypes.”37 We have rejected the 1993 sample for possible errors, but since we computed the 1996 data to be compared to the 1993 one (because both use the same base for their measures of trust), we must reject the 1993-1996 database in its entirety. Table 7 illustrates that for the EU14 38 countries in our 1995-1997 database, 13.2% of our sample countries share a common border, 44.3% share a common religion and 6.2% share a common language. We test the between/within coefficients of our 1995 and 1997 periods – the results are in the annex. We have 28 observations in each period “T”, 14 different individuals “n” in each T, which makes for a population “N” of 392. We observe that for the bilateral trade variable, the dispersion within each period is not as large as the dispersion between periods. This is expected since the countries surveyed record bilateral trade ranging from 24 million ECUs to over one million ECUs. Between 1993 and 1997, bilateral trade in our sample countries has accrued. For bilateral trust, we make the contrary observation: the dispersion within each period is larger than the dispersion between periods. This confirms that cultural stereotypes, on which trust is based, do not vary as much over the years as they vary between the countries sampled. The above descriptive analytics have illustrated our variables and samples. Below, we will first define the model we wish to estimate and our estimation methods, before studying the results we find. 37 38 Ibid. p.11 EU14=EU15-Luxembourg 24 3.2.2 Estimation 3.2.2.1 Our Model and Hypothesis Our results depend on the assumption that although all European populations may have replenished their ranks over the past decade, the distribution of the population within each country hasn’t significantly altered from 1995 to 2005. We state this hypothesis because we use nominal data for all our trade variables, with the sole exception of the distance measure that is conditioned by the distribution of individuals in their countries as of 2005. We set out to test whether the home bias effect can be mitigated by the trust Europeans feel towards each other. We will use a log/log functional form, consistent with the gravity model, which measures the elasticity of change in trade in respects to the percentage variation in the regressors. Our base model is the first one tested by Nitsch (2000): (9) log( xij ) 1 Home 2 ln( yi ) 3 ln( y j ) 4 log( d ij ) 5 Border 6 com _ lang eijt …where the dependant variable is the log of exports and the explanatory variables are Home which captures the home bias effect, the log of origin and destination countries’ GDPs, the log of the distance variable, and dummy variables indicating an adjacent border and a common language, as well as an error term. To dissociate the effects of trust from the effects of increased information; we follow Guiso et al (2004) in deeming common borders, distance and commonality between languages, as proxies for the information Europeans have upon each other. To equation (9) we introduce the bilateral level of trust: (10) log( xij ) 1 Home 2 ln( y i ) 3 ln( y j ) 4 log( d ij ) 5 Border 6 com _ lang 74Trust eijt For sensitivity analysis purposes, we modify the estimated equation by controlling for average level of origin trust and destination trust: (11) log( xij ) 1 Home 2 ln( y i ) 3 ln( y j ) 4 log( d ij ) 5 Border 6 com _ lang 7 Trust 8 ave _ o 9 ave _ d eijt 25 Because of potential measurement errors in our trust variable (as well as the likelihood that it captures the effect of omitted cultural variables), we will instrument it with a measure of shared religion. Shared religion makes for an appropriate instrumental variable for we argue that it is highly relevant given that we determined trust to be partly based on sociological factors such as religion and, moreover, opinions on trustworthiness may be based on cultural stereotypes of which religion plays a prominent role. It is also fulfills the instrument exogeneity condition, for although religion is related to trust, the former does not influence bilateral trade levels. 39 Nonetheless, to verify that religion is partly correlated to trust, we will regress bilateral trust on religion: (12) bil _ trust 1 Home 2 ln( y i ) 3 ln( y j ) 4 log( d ij ) 4 Border 4 com _ lang 4 Re ligion eijt … before conducting our Instrument Variable estimation of equation (11). 3.2.2.2 Estimation Methods To carry out our study, we will use four different estimation methods. Consistent with Nitsch (2000), we first pool our data, and estimate its entirety with the Ordinary Least Squares (OLS) method, computing heteroskedasticity-robust statistics. Pooled data however, does not control for period effects. As Wooldridge (2000) notes, “the errors in such an equation are almost always serially correlated because of an unobserved effect. Random effects estimation corrects the serial correlation problem and produces asymptotically efficient estimators”.40 Therefore, by first specifying the periods in our panel data, we carry out Generalized Least Squares (GLS) estimations, assuming that the unobserved effect is uncorrelated with all our explanatory variables. Subsequently, we lift this assumption and test fixed effect estimations. In discussing whether to use fixed or random effects, Wooldridge indicates that for data that cannot be consider as random draws from a large population - as for instance, country specific data – it’s judicious to use fixed effects estimations. To paraphrase Guiso et al, “Ireland is not Catholic because it traded more with Rome than with London” 40 Wooldridge (2000), p. 625 39 26 Finally, we instrument our trust variable with a dummy indicating 1 if the country shares their religion and 0 otherwise, and complete our study with Instrument Variable (IV) estimations. 3.2.3 Interpretation The full list of all our regressions is in the annex. Regression (1) details the standard gravity model augmented by a home dummy variable that captures intranational trade, a border dummy that indicates whether a common border exists and a language dummy that denotes a shared language. The estimated equation yields (the standard errors are in brackets beneath the coefficients): (13) ln( xˆ ij ) 0.66 ln( y i ) 0.65 ln( y j ) 1.27 ln( d ij ) 0.834 Home 0.103Border 0.049language eijt (0.0212) (0.022) (0.042) (0.184) (0.0922) (0.108) With an R2 of over 0.99, the empirical fit of the gravity model is perfect. The home bias effect is of a similar magnitude to Wei’s (1996) estimation: in our first model, a country exports about 2.3 (=exp[0.834]) times as much to itself as to a foreign country with an adjacent border and a similar size, distance and language. Yet this regression seems erroneous for not all the coefficients of the variables in this standard gravity have the expected signs. More precisely, we believe our distance variable is correlated with our GDP variables because the distance variable is weighted by the population, a measure of economic size that Nitsch (2000) used as an instrumental variable for GDP. We therefore, choose a different distance that measures the distance between capitals, distcap. It is unweighted, and isn’t potentially correlated to any other explanatory variables. Regression (2) reestimates equation (9) with a new measure of distance: (14) ln( xˆ ij ) 1.928Home 0.647 ln( y i ) 0.642 ln( y j ) 1.22 ln( d ij ) 0.309 Border 0.024lang eijt (0.296) (0.488) (0.459) (0.11) (0.151) (0.135) The empirical fit of the model is once again perfect. The sign on the common language coefficient is curious since it indicates that a shared language negatively influences bilateral trade. Otherwise the change in distance measurements significantly influences the home bias effect. Instead of Wei’s low effects, we obtain results similar to Nitsch’s magnitude of 6. In our model, a country exports about 6.6 (=exp[1.928]) 27 times as much to itself as to a foreign country with an adjacent border and a similar size, distance and language. The GDP coefficients situate around 0.64, meaning that when the GDP of one trading partner increases by 10%, the trade volume increases less than proportionally by 6.4%. A 10% increase in distance decreases bilateral trade more than proportionally by 122%. The language and border variables are statistically significant at respectively the 10% and 2% level. For the language variable, we can argue following Linders et al. (2005) that it may decrease bilateral trade for countries prefer to implant branches in a country that speaks the same language instead of trade with it, but we view this explanation as shaky at best. If we add the bilateral trust variable to our equation, we obtain: (15) ln( xˆ ij ) 1.99 Home 0.208Trust 0.66 ln( y i ) 0.647 ln( y j ) 1.208 ln( d ij ) 0.308Border 0.0005lang eijt (0.186) (0.183) (0.26) (0.24) (0.052) (0.106) (0.125) Once again the gravity model fit is ideal. The sign on the language coefficient is now positive, but the coefficient on the trust variable is negative (which could vindicate Linders et al’s assumption that less bilateral trade due to a greater trust could indicate that the countries have chosen to implement mutual branches in each other’s countries instead of partaking in long-distance trade). The home bias coefficient increases: a country exports about 7.315 (=exp[1.99]) times as much to itself as to a foreign country, once adjacent borders, similar sizes, distances, languages and trust levels are taken into account. The GDP coefficients, being similar to the ones found in the previous regression, carry the same interpretation. We notice that the distance coefficient in Regression (2) captures some of the trust effects of this regression since it is now lower. If we control for the average level of origin country trust and the average level of destination country trust, we obtain: (16) ln( xˆij ) 2.16 Hm 2.03Trst .72 ln( yi ) .68 ln( y j ) .93 ln( dij ) .32 Brdr .21lang .09ave _ o 3.98ave _ d eijt (.155) (.340) (.025) (.025) (.051) (.092) (.128) (.422) (.325) After controlling for average origin and destination trust, regression (4) shows a positive relationship between trust and trade. A given country exports 7.61 times more to a country in whose citizens it trusts than to one it distrusts, ceteris paribus. The home bias effect soars: controlling for all previous variables, Europeans countries export 28 more than 8.5 times to themselves than to foreigners. The coefficients on the GDP variables haven’t significantly changed since the previous regressions: a 1% increase in the GDP of a trading partner increases bilateral trade by less than proportionally. Unlike in regression (3), a 1% increase in the bilateral distance separating trading partners decreases trade also, less than proportionally. The average trustiness of the origin country is found to be statistically significant at the 10% level; a common language – that now has a positive impact on bilateral trade – is significant at the 1% level. Although these results seem encouraging, Wooldridge (2000) sounds a note of caution and underlines that pooled data does not control for period-specific effects and that this unobserved effect is often serially correlated with the errors in our sample. We have tried correcting this by robustifying our standard deviations, but nonetheless, we will now include a random-effects specification to correct the potential serial correlation problem. Assuming that the unobserved effect is uncorrelated with all our explanatory variables, the GLS estimation yields: (17) ln( xˆij ) 2.79 Hm 1.35Trst .70 ln( dij ) .44 Brdr .76 ln( yi ) .79 ln( y j ) .25lang .27ave _ o 1.59ave _ d eijt (.15) (.29) (.05) (.09) (.08) (.02) (.12) (.89) (.42) This regression (5) yields results that are more in line with the previous literature. The coefficients on the distance measure, the GDPs, the common border and common language variables are nearly identical to the previous ones estimated. For the measure of average origin and destination, we note that their coefficients follow our previous assumptions. Indeed, a 10% increase of origin country trustiness increases bilateral trade by 2.7% - this result is significant at the 5% level. The bilateral trust coefficient has sharply decreased: a given country exports now 3.86 times [and not 7.61 times as in regression (4)] more to a country in whose citizens it trusts than to one it distrusts, ceteris paribus. The home effect in this non-pooled sample is the largest one yet recorded: a European country exports about 16.3 (=exp[2.79]) times as much to itself as to a foreign country, once adjacent borders, similar sizes, distances, language and trust levels are taken into account. This result is consistent with McCallum’s (1995) home bias results for the US and Canada. However, in carrying out this GLS regression, we have assumed that there is a period-constant effect that is uncorrelated with the explanatory variables. Wooldridge (2000) indicates that for data such as ours – that can’t be assumed to be random draws 29 from a large population – fixed effect regressions yield more efficient results. In the following regression (6), we therefore lift the assumption that the unobserved effect is uncorrelated with the explanatory variables. (18) ln( xˆij ) 2.8Hm 1.36Trst .70 ln( dij ) .44 Brdr .60 ln( yi ) .79 ln( y j ) .25lang 2.65ave _ o 1.59ave _ d eijt (.15) (.29) (.05) (.09) (.29) (.02) (.12) (1.48) (.42) Our model now explains about 88% of the variations in bilateral trade. Between the random-effects estimation and the fixed-effects one, not many variables have registered any change: the bilateral trust levels, the distance measure, the home bias measure, the GDP of the destination country, the average level of trust emanating from the destination country and the language dummy register only minute variations to their coefficients (and none to their standard errors). On the other hand, the coefficient on the measure of average origin country trust carries now the wrong sign and the coefficient of the GDP of the origin country has decreased. We believe that there is a problem of in our model, where the error term and key explanatory variables are correlated. Endogeneity will yield biased and inconsistent estimators for the OLS estimations above. In an attempt to correct this, we introduce an instrumental variable, “religion”, that we believe is correlated with the explanatory cultural variables, but not with the bilateral trade flows. When discussing our hypothesis above, we have already argued that it is a relevant and efficient instrument. The following regression (7) is our IV estimation that instruments trust with religion: (19) ln( xˆij ) 2.63Hm 2.96Trst .63 ln( dij ) .45Brdr .76 ln( yi ) .80 ln( y j ) .30lang 0.14ave _ o 3.05ave _ d eijt (.29) (1.77) (.08) (.10) (.02) (.03) (.13) (1.71) (1.69) For the first time in our study, the average trust variable is statistically significant at the 10% level. A given country exports about 20 times (=exp[2.96]) more to a country in whose citizens it trusts than to one in whose citizens it distrusts, ceteris paribus. This significant result confirms our original hypothesis that trust plays a significant role in bilateral trade flows. Controlling for trust, distance, GDPS, adjacent borders, common languages, we find that the home bias effect in the EU15 is substantial: we find that a European country exports about 14 (=exp[2.63]) times as much to itself as to a foreign country, ceteris paribus. These results fall in line with the studies conducted by McCallum (1995) and Nitsch (2000). Furthermore, the common 30 language and the average trust levels for destination countries are significant at respectively 2% and 1% levels. As far as the EU14 is concerned, the home bias exists and trust is a significant force in bilateral trade flows. Yet within the EU14, some countries are outliers in regards to their shared languages or geographical borders. Since languages in Europe are dispersed we will ignore the outliers that do not share a common language. But as for borders, we remarks that about half of our sample countries either do not share any common border or share a common border with only one other country. (Table 9 and 10 in annex) For these reasons, we will now discard Denmark, Finland, Greece, Great Britain, Ireland and Sweden from our sample. A further justification to this choice is that these countries, being geographically distant from continental Europe, are likely to use different type of transport costs41 that could carry an inherent bias to our home bias estimations. To test whether dropping outliers has a significant effect, we will run the regressions again on the remaining sample – Austria, Belgium, Germany, Spain, France, Italy, Portugal and the Netherlands. Moreover, this sample eliminates two natural causes of home bias, for these eight countries share a common currency42 and are members of the Schengen zone, which eliminates border checks on people and merchandise. A summery of the descriptive statistics on our new sample shows that all our countries share at least a border with two other countries. 43 Moreover, Table 12 verifies that 62% share the same religion (Catholicism in this case); 31% share a contiguous border and 16% share the same language. The descriptive statistics differentiated per year, Table 13 in the annex, do not raise any red flags. Although logarithmic functional forms are less sensitive to outlying observations, we drop the outlier country from our database and see whether the OLS fixed-effects estimates change by a “large” margin. Our estimation (8) yields the following: 41 Indeed, these countries are either islands or could proxy as such. Islands use water freight, which carries a different cost than ground transportation that landlocked or contiguous countries use. For the purpose of this paper, a discussion regarding which of these modes of transportation is cheaper or more effective is of no interest to us. 42 Or at the very least, for the years surveyed, the wholesale goods were priced in ECUs although the retail goods were still in national currencies. 43 Portugal is our sole exception. Taking Portugal out of the sample would have subsequently required us to take Spain out as well, leaving us with too few countries in our sample. 31 (20) ln( xˆij ) 2.55Hm 1.44Trst .63 ln( dij ) .43Brdr 1.16 ln( yi ) .66 ln( y j ) .169lang 1.92ave _ o 3.05ave _ d eijt (.26) (0.52) (.102) (.11) (.74) (.04) (.14) (2.0) (0.71) We compare equation (20) to equation (18) above and detect that the following coefficients have only registered minute changes: the distance measure, the dummy indicating a shared common border and the GDP of the destination country. Our current model fits the data better (with an R2 of 0.92, as compared with equation (18)’s R2 of 0.88). On the other hand, in this smaller more homogeneous sample we still detect a substantial home bias. It’s not as large as previously measured in equation (19) where it was of a magnitude of 16.4; but its current magnitude of 12.8 is noteworthy. The importance of trust has diminished within this sample, possible due to the long period of interaction between these Member States. The levels of average trust in the country of destination and of origin carry a negative sign, which may support both the notion that average country trustworthiness and average country trustiness move in parallel, as well as Linders et al’s hypothesis that a greater amount of trust does not necessarily lead to more bilateral trade, but to more branches implanted in respective countries, and thus to less bilateral trade. Finally, we notice that the coefficient on the dummy variable indicating a common language is smaller in this regression. We believe that this is because these countries can be roughly divided into Germanic countries and Latin countries – there are no linguistic outliers in our sample – and a latin-language speaking person, for instance, needs only to learn a Germanic language to comprehend the gist of all the European languages in this sample population. Once again we instrument our trust variable with religion, to check whether our trust variable is a “noisy” measure. We first regress Trust on Religion to confirm whether they are positively correlated. Regression (9) yields the following results: (21) Tˆrust 0.33Home .07 ln( y i ) .01 ln( y j ) .06 ln( d ij ) .01Border .20Language .01 Re ligion eijt (0.06) (0.01) (0.01) (0.02) (0.03) (0.03) (0.02) Religion is significant at the 5% level. We run the last regression of our paper, which yields: (22) ln( xˆij ) 26.5Hm 140Trst 1.22 ln( dij ) 1.16 Brdr 0.21 ln( yi ) 1.1 ln( y j ) 8.7lang 103.7ave _ o 110.9ave _ d eijt (148.1) (711.5) (3.55) (8.27) (5.5) (2.64) (45.7) (524.4) (552.2) 32 As compared to equation (19), this result doesn’t fit with the rest of empirical analysis despite religion being a relevant instrument for trust. We note that all our variables are significant at levels between 6% and 10%. Moreover all our variables carry a negative sign, with the exception of bilateral Trust and the GDP of the country of destination. The magnitude of trust has increases by over 4000%. Although none of the previous literature have recorded such results, the magnitude of the trust coefficient in the study by Guiso et al (2004) has also greatly increased when instrumented by religion. In view of their results, they concluded, “our measure of trust is a noisy measure of the true trust between two countries (…) One potential concern with our instrumental variable regression is that our instruments may be only weakly correlated with trust. If this is the case then the two stage least squares regressions will be biased and the standard errors misleading.”44 We accept that the trust variable, derived from the Eurobarometer surveys, is a noisy measure of “true trust”. Statisticians need to perfect a measure of “true trust”, perhaps beginning by defining what variables lie behind the trust countries have towards each other. At the very least, our estimations have confirmed that if only the GDPs and the distances between countries were the sole relevant factors in bilateral trade, then the dummy variable representing intranational trade would register minute coefficients instead of the immense ones it registers in our models. While, for instance in equation (20), the distance is not significant to bilateral trade levels, the extent of the home bias and bilateral trust are. In this chapter we have set out to test whether the home bias effect can be mitigated by the trust Europeans feel towards each other. On the contrary, we found that the home bias effect and bilateral mistrust compound each other, which indicates that the levels of trust Europeans have towards each other influences the seeming deficiency of inter-Member State trade in the EU. This is not necessarily a pessimistic observation. If trust or the lack thereof lies behind the EU home bias puzzle in trade, then we can discard the assumption that it is national preferences that impede European trade. If one is to consider mistrust as a market imperfection problem, this leaves considerable scope for aggressive intergovernmental (read EU-level) action in thwarting this last frontier to the Single Market. The next step in this line of research is in testing the impact of EU social policies - such as, for instance, the student-exchange program “ERASMUS” – on breaching the trade distances between European countries. 44 Guiso et al (2004) p. 20 33 4 Conclusion In this paper we showed that trust influences European bilateral trade flows. We began by examining the definitions and determinants of both trust and distance as they relate to trade. After determining that trust is the underlying variable in economic transactions, we introduced the notion of distance as a multidimensional concept, which can be expanded and contracted according to certain variables. We considered trust to be such a variable, capable of contracting long distances when trust is strong, or on the contrary, making a distance “look longer in the eyes” when trust is absent. The discussion on distances introduced the “home bias puzzle” which we first analyzed in a descriptive manner. After defining the concept and the classical determinants of the home bias puzzle, we gave voice to the literature that estimates the home bias puzzle to be substantial as well as the literature arguing that the home bias puzzle has been wildly overstated. We concluded our review of the literature by focusing on the literature portraying the impact of trust and other cultural variables on international trade flows. The analytical chapter that followed first introduced the theoretical framework of the gravity model, which we used in assessing the impact of trust on European bilateral trade flows. We defined our empirical strategy in three parts: we first conducted data analysis, then defined the augmented gravity model and estimation methods, and finally we interpreted our results. We found that the trust Europeans have towards each other, influence the seeming deficiency of inter-Member State trade in the EU. This is not a pessimistic observation since it justifies EU-wide action in thwarting the last hindrance to a Single Market. The founding father of the European Union, Jean Monnet, already perceived trust as being the last hurdle in uniting Europe when he declared, « l’Europe [...] est le résultat de la confiance que nous avons en nousmêmes.»45 45 Discours, Strasbourg, 15 juin 1953. 34 Bibliography Articles J. E.ANDERSON, (1979). A Theoretical Foundation for the Gravity Equation. The American Economic Review, 69 (1), pp. 106-116 J. E.ANDERSON,; E.VAN WINCOOP, (2003). Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review 93, 170-92. K.ARROW, (1972). Gifts and Exchanges. Philosophy and Public Affairs, pp. 343-362 G.BECKER (1996). Preferences and Values. in Becker Gary (ed.), “Accounting for Taste” Harvard University Press: Cambridge. W.BECKERMAN (1956). Distance and the Pattern of Intra-European Trade. Review of Economics and Statistics, 28(3), pp. 156-163. F.BORNHORST, A.ICHINO, K.SCHLAG (2004). Trust and Tustworthiness among Europeans : South-North Comparison. CEPR Discussion Papers, number 4378. J.BRUN, C.CARRERE, P.GUILLAUMONT, J.MELO (2002) Has Disatnce Died ? Evidence from a Panel Gravity Model. CEPR Discussion Paper no. 3500 R.COASE (1937). The Nature of the Firm. Economica, 4, pp. 386-405 A. V. DEARDORFF, (1995). Determinants of bilateral trade: Does gravity work in a neoclassical work?. NBER working paper 5377 A.C. DISDIER, T. MAYER (2005). Je t'aime, moi non plus: Bilateral Opinions and International Trade. CEPR DP 4928. A.GATRELL, (1983). Distance and Space: A Geographical Perspective. Oxford, Clarendon Press. G.GROSSMAN (1996). Comments on Alan V. Deardroff, Determinants of bilateral trade: Does gravity work in a neoclassical world? In: Jeffrey A Fankel (Ed.). The regionalization of the World Economy. Chicago: University of Chicago for NBER L.GUISO, P. SAPIENZA AND L. ZINGALES (2004). Cultural biases in economic exchange. NBER Working Paper 1105, December. L.GUISO, P. SAPIENZA, L. ZINGALES (2006). Does Culture Affect Economic Outcomes? NBER Working Paper No. 11999 J.HELLIWELL, (1998). How much do National Borders Matter? The Brookings Institution Press, Washington D.C. R.HILLBERRY, (2000). Interpreting ‘Home bias’ in U.S.- Canada Trade. International Economic Review, June/July 2000 pp.6-9 G. HOFSTEDE, (1980). Culture’s Consequences: international differences in workrelated values. Beverly Hills, CA: Sage Publications R.R.HUANG, (2005). Distance and trade: Disentangling unfamiliarity effects and transport cost effects. European Economic Review, Economics Working Paper Archive EconWPA, International Trade Series, No. 051101 35 E.E. LEAMER, , M.STORPER (2001). The Economic Geography of the Internet Age. NBER Working Paper Series 8450 G-J M.LINDERS, A SLANGEN, H.L.F. DE GROOT, S. BEUGELSDIJK (2005). Cultural and Institutional Determinants of Bilateral Trade Flows. Tinbergen Institute Discussion Papers, number 05-074/3. J. MCCALLUM (1995). National Borders Matter: Canada-U.S. Regional Trade Patterns. American Economic Review.1995, 85. pp. 615-623 J.MONNET, (1952, 1953) Discours Washington, 30 avril 1952 ; Discours, Strasbourg, 15 juin 1953. V. NITSCH, (2000) National borders and international trade: evidence from the European Union. Canadian Journal of Economics 33 (4), 1091–1105. M.OBSTFELD, K.ROGOFF (2000) The Six Major Puzzles in International Finance: Is There a Common Cause? in Bernanke, B. S.; Rogoff, K., eds. NBER Macroeconomics Annual, 2001. Volume 15. Cambridge and London. MIT Press, pp. 339-390. J.E.RAUCH, (2001). Business and Social Networks in International Trade. Journal of Economic Litterature, XXXIX, pp. 1177-1203 A.K. ROSE, (1999). One Money, One Market: Estimating the Effect of Common Currencies on Trade. Seminar Paper Institute for International Economic Studies, 678. G. SANDELIEN, (2003). Trust & Trade … Is distance dead? Chr Michelsen Institute, R 2003:4 D. TREFLER (1995). The Case of the Missing Trade and Other Mysteries. American Economic Review 85, (December), pp. 1029-1046. S-j.WEI, (1996). Intra-national versus international trade: how stubborn are nations in global integration? NBER Working Paper No. 5531 O.E. WILLIAMSON (1998). Transaction Cost Economics: How it Works; Where it is Headed. De Economist, 146, pp. 23-58 Books J. WOOLDRIDGE (2000) Introductory Econometrics: A Modern Approach, South Western College Publishing, 2000 Websites CIA Fact Book: http://www.odci.gov/cia/publications/factbook/index.html DEARDORFF's Glossary of International Economics http://www-personal.umich.edu/~alandear/glossary/ Gesis: http://www.gesis.org Merriam-Webster Online Dictionary. http://www.m-w.com/ 36 Datasets The datasets we have used can be accessed at: http://www.ogorkiewicz.org/1995_1997.dta http://www.ogorkiewicz.org/1993_1996.dta http://www.ogorkiewicz.org/1995_1997_sans_outliers.dta 37 ANNEX Annex 1: Trust Matrix Table 1: Trust Matrix for 1993 surveyed population. 1993 BEL DNK FRA DEU GRB GRC IRL ITA LUX NLD PRT ESP AVER1 BEL 3.34413 3.243184 3.113169 2.943258 2.893816 2.481436 2.953717 2.785366 2.926882 3.293756 2.890066 2.878229 DNK 3.108796 3.547475 2.883644 2.917199 3.125843 2.24497 2.951872 2.780679 2.972028 3.284129 2.803621 2.860377 2.978917 2.956719 FRA 2.960986 2.895661 3.198582 2.819966 2.280719 2.703108 2.781285 2.87037 2.890558 2.838778 2.98926 2.523243 DEU 2.885598 3.092574 2.788127 2.757166 2.426263 2.133612 2.587514 2.652834 2.690678 2.845287 2.569175 2.709012 GBR 2.89648 3.336401 2.450358 2.750525 3.324663 2.251068 2.765738 2.608741 2.842572 2.980331 2.718866 2.454846 GRC 2.592385 2.58147 2.466149 2.602075 2.611543 3.209137 2.417127 2.460542 2.669136 2.55102 2.589888 2.454765 2.81271 2.678153 2.781716 2.600436 IRL 2.917035 3.057235 2.713508 2.733392 2.670588 2.403743 3.43527 2.515544 2.77381 2.813747 2.608571 2.636912 ITA 2.557394 2.487474 2.488157 2.542625 2.553261 2.201058 2.584464 2.826979 2.677419 2.40836 2.557908 2.565646 LUX 3.361026 3.272931 3.117834 2.997917 2.9437 2.446358 2.945619 2.745989 3.322034 3.372845 2.917553 2.892487 NLD 3.03681 3.355072 2.940067 2.961701 3.159091 2.242673 2.934109 2.856098 3.093478 3.351682 2.865258 3.002375 PRT 2.682703 2.636771 2.577685 2.661324 2.729193 2.575758 2.593974 2.426108 2.713969 2.772619 3.308108 2.491694 AVER2 2.921343 3.012776 2.790734 2.785196 2.772639 2.464294 2.793538 2.68797 2.86461 2.934217 2.789655 2.727865 2.77328 2.537562 3.028024 2.983201 2.680826 2.733293 2.795403 Source: Eurobarometer 39.0 Mar-April 1993. (Gesis code: ZA2346) Variables: V73-V84. Following Guiso et al. (2004), the answers have been encoded as “Lot of trust”= 4; “Some trust” = 3; “Not very much” = 2; “No trust”= 1 This matrix was done by MSExcel. (1) This average indicates how much a given country is trusted by others (2) This average indicates how much a given country trusts others. ESP 2.712777 2.647059 2.751527 2.735209 2.552991 2.678611 2.571767 2.726384 2.802752 2.698052 2.657583 3.264798 Table 2: Trust matrix for 1995, surveyed population. 1995 AUT BEL DNK FIN FRA DEU GRB GRC IRL ITA LUX NLD PRT ESP SWE AVER2 AUT 1.950106 1.793151 1.803301 1.781337 1.503038 1.832421 1.665816 1.530283 1.569425 1.447942 1.824675 1.842298 1.533239 BEL 1.788927 1.918864 1.80917 1.784363 1.689546 1.618182 1.697105 1.52451 1.699041 1.513304 1.935966 1.733262 1.609816 1.608496 1.818182 1.716582 DNK 1.921023 1.916175 1.990798 1.926052 1.642452 1.862288 1.933761 1.58264 1.878788 1.497727 1.941383 1.953362 1.681876 1.627771 1.959119 1.821014 FIN 1.873918 1.822193 FRA 1.802706 1.935079 1.868571 1.886574 1.941896 1.786236 1.875 1.958206 1.715159 1.736541 1.878363 1.601242 1.79845 1.484737 1.825333 1.855346 1.61753 1.508685 1.893276 1.762932 1.62199 1.717314 1.853631 1.688161 1.943417 1.874299 1.771798 1.799571 1.886364 1.825174 DEU 1.75931 1.801287 1.813187 1.804619 1.650977 1.677658 1.681322 1.642036 1.734167 1.611627 1.836388 1.811988 GRB 1.827798 1.822134 1.900878 1.873995 1.497309 1.537037 1.911885 1.547124 1.753261 1.577465 1.845188 1.892562 GRC 1.43075 1.4214 1.391645 1.475795 1.440639 1.27694 1.337486 1.846389 1.423127 1.466216 1.412568 1.401028 IRL 1.847021 1.863714 1.869144 1.816926 1.713939 ITA 1.640196 1.761765 1.757843 1.754902 1.645098 1.568071 1.643487 1.526006 1.615686 1.762977 1.748773 1.784314 LUX 1.55984 1.837596 1.698298 1.6267 1.684385 1.564669 1.958071 1.76671 1.806452 1.816643 1.770291 1.693473 1.697538 1.731646 1.622449 1.728933 1.69726 1.853242 1.837209 1.66135 1.955882 1.838951 1.674535 1.692966 1.83027 1.734822 1.741333 1.89525 1.745936 1.48153 1.56582 1.52657 1.517949 1.456669 1.749169 1.613821 1.85873 1.770267 1.604902 1.691176 1.776471 1.685444 1.647713 1.71916 1.806835 1.750935 NLD 1.856371 1.932377 1.978747 1.965269 1.641138 1.793717 1.878852 1.615481 1.874279 1.465247 1.973517 1.957317 1.769231 1.671676 1.967342 1.822704 PRT 1.694379 1.656977 1.694444 1.669725 1.624851 1.55971 1.708726 1.577586 1.655422 1.637951 1.751256 1.757327 1.902273 1.589474 1.739857 1.681331 ESP 1.717647 1.59976 1.508516 1.76875 1.826027 1.649573 1.908034 1.756447 1.693997 1.94269 1.701453 1.807834 1.942189 1.6703 1.902954 1.585366 1.913174 1.910188 1.709924 1.614325 1.962264 1.831769 1.787925 1.807918 1.818111 1.809507 1.641211 1.665375 1.721702 1.60865 1.740673 1.586617 1.835301 1.818365 SWE AVER1 1.75469 1.742942 1.731861 1.517202 1.94201 1.912519 1.95935 1.56172 1.664858 1.701923 Source: Eurobarometer 44.0 Oct-Nov 1995. (Gesis code: ZA2689) Variables: V298-V312. Following Guiso et al. (2004), the answers have been encoded as “Trust them”= 2; “No trust”= 1. This matrix was done by MSExcel. (1) This average indicates how much a given country is trusted by others (2) This average indicates how much a given country trusts others. 1.676296 1.646492 1.83373 1.733192 Table 3: Trust matrix for 1996, surveyed population. 1996 AUT BEL DNK FIN FRA DEU GBR GRC IRL ITA LUX NLD PRT ESP SWE AVER2 AUT 3.564639 2.946179 2.948276 2.938731 2.622449 3.090909 2.588795 2.518947 2.550549 2.430584 3.065887 BEL 2.833713 3.101493 2.965144 2.916877 DNK 3.216331 2.951299 3.591365 FIN 3.290466 FRA 2.703883 2.974332 2.928834 2.911471 3.230303 2.827801 2.396246 DEU 2.771888 GRB 2.888753 2.822904 3.061176 2.981273 2.249732 2.247096 3.330653 2.432184 2.725053 2.508869 2.875949 3.066362 2.792521 2.572707 3.032634 2.772525 GRC 2.322733 2.354651 2.297521 2.419512 2.422581 2.033403 2.117137 IRL 2.928779 2.858543 2.918079 2.915888 2.837264 2.704572 2.682162 2.619658 ITA 2.663344 2.658757 2.782609 2.784748 2.738693 2.794177 2.663265 2.379078 2.520373 2.879056 2.746532 2.844244 2.430095 2.653261 2.885452 2.694912 2.75129 2.632124 2.705263 2.464706 2.674528 2.403415 3.2 2.670514 2.945918 3.100616 2.95 2.495565 2.582627 3.049485 2.822908 3.17176 2.779412 2.590319 2.591723 2.991677 2.771563 2.50655 2.983766 2.501584 3.120551 3.303609 2.68623 2.677838 3.408308 2.990965 3.07387 3.299046 3.691296 2.915327 2.890411 3.180659 2.677313 2.924945 2.507592 3.060502 3.141458 2.673055 2.605234 3.348813 3.018666 2.49214 2.673635 2.514439 3.007955 2.864183 2.678097 2.743199 2.992883 2.79596 2.72816 2.857616 2.812515 2.679948 2.757019 2.538102 2.524649 2.636096 2.513363 2.898245 2.855473 2.681818 2.699708 2.906608 2.724081 3.34413 2.461899 2.441145 2.394541 3.49121 2.804612 2.945559 2.24263 2.524706 2.567749 2.510297 2.430309 2.93992 2.783309 2.731458 2.91954 2.872037 LUX 2.946667 2.713542 2.975104 2.942605 2.865854 2.836806 2.408602 NLD 2.901822 PRT 2.128591 2.219355 2.260695 2.180082 2.530011 2.200696 2.303357 2.008086 2.172185 2.311456 2.380645 ESP 2.652108 2.682961 2.747175 2.708783 2.279446 2.647406 2.058111 2.351459 2.528121 2.541469 2.704648 2.819048 2.639115 3.436249 2.835443 2.642103 SWE 1 AVER 3.527406 3.00969 3.360707 3.250288 2.498522 2.47957 2.592 2.644128 3.493976 2.994575 2.690813 2.770221 2.983471 2.822529 2.88705 2.902344 2.468271 2.856269 2.334004 3.270896 3.381503 2.849945 2.805613 3.339175 2.941073 3.23463 3.569196 3.491545 3.040237 3.130035 3.432432 2.889408 2.822024 2.970836 2.943041 2.688811 2.708361 2.87619 3.260494 2.810714 2.28978 3.228805 2.303974 2.242619 2.317356 3.30914 3.334951 2.966581 2.857314 3.588364 3.228615 2.69385 2.542862 2.736742 2.543095 2.963119 2.920477 2.714065 2.706592 3.002318 2.789707 Source: Eurobarometer 46.0 Mar-April 1996. (Gesis code: ZA2898) Variables: V81-V95. Following Guiso et al. (2004), the answers have been encoded as “Lot of trust”= 4; “Some trust” = 3; “Not very much” = 2; “No trust”= 1 This matrix was done by MSExcel. (1) This average indicates how much a given country is trusted by others (2) This average indicates how much a given country trusts others. Table 4: Trust matrix for 1997, surveyed population. 1997 AUT BEL DNK FIN FRA DEU GRB GRC IRL ITA LUX NLD PRT ESP SWE AVER2 AUT 1.974113 1.810127 1.817442 1.822738 1.560043 1.882296 1.584946 1.528409 1.64455 1.513007 1.902045 1.888889 1.543773 1.590703 1.882096 1.712219 BEL 1.779682 1.877633 1.830258 1.823303 1.747078 1.696296 1.682213 1.533993 1.70024 1.536184 1.913502 1.798729 1.617128 1.643447 1.840686 1.731478 DNK 1.900332 1.764505 1.967247 1.907996 1.671964 1.835282 1.889597 1.535211 1.786543 1.503356 1.900932 1.921109 FIN 1.884532 1.797821 1.879581 1.951923 1.768722 1.71888 1.847296 1.587571 1.768879 1.492274 1.837408 1.851276 1.572967 1.491011 1.889872 1.74682 FRA 1.777246 1.881029 1.842604 1.854545 1.943633 1.751848 1.525641 1.614532 1.754345 1.686891 1.905405 1.813018 1.717634 1.767343 1.863208 1.78012 DEU 1.742795 1.731626 1.794149 1.783348 1.686772 1.664441 1.639735 1.626081 1.691084 1.615788 1.812986 1.791917 1.698614 GRB 1.810325 1.77684 1.874251 1.875754 1.80615 1.871767 1.745679 1.564293 1.898678 1.729796 GRC 1.441943 1.43462 1.394207 1.487212 1.471947 1.299046 1.298246 1.899598 1.463804 1.477838 1.436693 1.393868 1.535329 1.593394 1.581645 1.483389 IRL ITA 1.88685 1.866771 1.884858 1.869416 1.50455 1.466 1.927845 1.564626 1.721608 1.8226 1.66129 1.690608 1.640754 1.712687 1.779932 1.736964 1.717182 1.59188 1.653534 1.625 1.6191 1.95339 1.712401 1.877049 1.889362 1.63191 1.606025 1.925234 1.774779 1.68682 1.826324 1.717835 1.771574 1.671177 1.90169 1.799799 1.55875 1.647443 1.757544 1.781285 1.804721 1.628163 1.751973 1.771018 1.706648 LUX 1.804 1.625688 1.820225 1.746544 1.731047 1.767273 1.643939 1.490698 1.650964 1.620038 1.945899 1.859287 1.614173 1.694949 1.791209 1.714424 NLD 1.803708 1.826763 1.972284 1.951422 1.554705 1.756871 1.800217 1.543616 1.817972 1.443169 1.960827 1.955301 1.752212 1.643026 1.976589 1.782498 PRT 1.616039 1.488189 1.602902 1.641484 1.676364 1.492288 1.662005 1.492734 1.558594 1.601513 1.688378 1.679952 1.931034 1.535585 1.674623 ESP 1.765891 1.772114 1.804044 1.775641 1.568558 1.692209 1.479644 1.637026 1.686502 1.724691 1.811594 1.843023 1.677835 1.933972 1.835596 1.731604 SWE 1.957092 1.838057 1.959607 1.943894 1.737288 1.813793 1.933025 1.652903 1.897856 1.647205 1.917749 1.909425 1.720056 1.785687 1.746965 1.814906 1.811479 1.677497 1.672646 1.683899 1.592717 1.716252 1.596733 1.833194 1.677205 1.654897 1.840539 1.724074 AVER1 Source: Eurobarometer 47.0 Jan-Feb 1997. (Gesis code: ZA2935) Variables: V64-V78. Following Guiso et al. (2004), the answers have been encoded as “Tend to trust”= 2; “Tend not to trust”= 1. This matrix was done by MSExcel. (1) This average indicates how much a given country is trusted by others (2) This average indicates how much a given country trusts others. 1.81811 1.62326 1.64974 1.949625 1.826444 Annex 2 : Descriptive Statistics Table 5: Descriptive Statistics for whole sample, for 1993 and 1996 _______________________________________________________________________________ -> year = 1993 Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------bilat_trade | 121 148004.9 1258806 16.043 1.38e+07 gdp_o | 121 534871.6 545257.8 43753.2 1844608 gdp_d | 121 534871.6 545257.8 43753.2 1844608 distwces | 121 1150.405 725.777 13.18305 2838.347 bilat_trust | 121 2.769822 .288337 2.133612 3.547475 -------------+-------------------------------------------------------ave_o | 121 2.789112 .1382025 2.464294 3.012776 aver_d | 121 2.774256 .1440133 2.537562 2.983201 _______________________________________________________________________________ -> year = 1996 Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------bilat_trade | 121 43914.77 176784.8 26.68344 1458474 gdp_o | 121 574160.1 574789.2 55562 1948601 gdp_d | 121 574160.1 574789.2 55562 1948601 distwces | 121 1150.405 725.777 13.18305 2838.347 bilat_trust | 121 2.682793 .3169785 2.008086 3.591365 -------------+-------------------------------------------------------ave_o | 121 2.682793 .1626823 2.348036 2.901754 aver_d | 121 2.682793 .1166943 2.508265 2.888266 Table 6: Descriptive Statistics for whole sample, for 1995 and 1997 _______________________________________________________________________________ -> year = 1995 Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------bil_trade | 196 29115.55 140643.7 26.97241 1467235 gdp_o | 196 477984.5 531445.6 51324.5 1929422 gdp_d | 196 477984.5 531445.6 51324.5 1929422 distwces | 196 1277.606 760.5966 13.18305 3364.83 bil_trust | 196 1.725155 .1543394 1.27694 1.990798 -------------+-------------------------------------------------------ave_o | 196 1.725155 .0906594 1.459819 1.825955 ave_d | 196 1.725155 .0861746 1.581279 1.835651 _______________________________________________________________________________ -> year = 1997 Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------bil_trade | 196 31720.4 145762.5 24.00341 1383286 gdp_o | 196 527749.5 559377.6 71717.6 1907246 gdp_d | 196 527749.5 559377.6 71717.6 1907246 distwces | 196 1277.606 760.5966 13.18305 3364.83 bil_trust | 196 1.721852 .151139 1.298246 1.976589 -------------+-------------------------------------------------------ave_o | 196 1.721852 .0823582 1.483764 1.829255 ave_d | 196 1.721852 .0805003 1.595069 1.844063 Table 7: Summery of our dummy variables Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------religion | 196 .4438776 .4981126 0 1 contig | 196 .1326531 .340068 0 1 comlang_eth | 196 .0612245 .2403556 0 1 Table 8: Between/Within coefficients for 1995, 1997 and all countries in sample. Variable | Mean Std. Dev. Min Max | Observations -----------------+--------------------------------------------+---------------bil_trad overall | 30417.97 143048.6 24.00341 1467235 | N = 392 between | 35160.37 3004.999 119087.2 | n = 14 within | 138967.7 -86790.85 1378566 | T = 28 | | gdp_o overall | 502867 545461.5 51324.5 1929422 | N = 392 between | 562988.1 61521.05 1918334 | n = 14 within | 49594.71 350822.7 654911.3 | T = 28 | | bil_trus overall | 1.723504 .1525611 1.27694 1.990798 | N = 392 between | .0881651 1.471791 1.827605 | n = 14 within | .1266435 1.375069 2.15131 | T = 28 | | ave_o overall | 1.723504 .0865133 1.459819 1.829255 | N = 392 between | .0881651 1.471791 1.827605 | n = 14 within | .0157551 1.694382 1.752625 | T = 28 Contingency Tables Table 9: Country and Border ---------------------| contig iso_o | 0 1 ----------+----------AUT | 24 4 BLX | 22 6 DEU | 18 10 DNK | 26 2 ESP | 24 4 FIN | 26 2 FRA | 20 8 GBR | 26 2 GRC | 28 IRL | 26 2 ITA | 24 4 NLD | 24 4 PRT | 26 2 SWE | 26 2 ---------------------- Table 10: Country and Language ---------------------|comlang_eth iso_o | 0 1 ----------+----------AUT | 12 2 BLX | 10 4 DEU | 12 2 DNK | 14 ESP | 14 FIN | 14 FRA | 13 1 GBR | 13 1 GRC | 14 IRL | 13 1 ITA | 14 NLD | 13 1 PRT | 14 SWE | 14 ---------------------- Annex 3: Estimation Output Regression (1) Regression with robust standard errors Number of obs F( 6, 386) Prob > F R-squared Root MSE = 392 =16167.39 = 0.0000 = 0.9947 = .59582 -----------------------------------------------------------------------------| Robust lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------ly_o | .6609168 .0212227 31.14 0.000 .6191902 .7026434 ly_d | .6560164 .0220775 29.71 0.000 .6126091 .6994236 ldis | -1.273898 .0428778 -29.71 0.000 -1.358202 -1.189595 Home | .83452 .1836042 4.55 0.000 .4735306 1.195509 contig | .1031118 .092251 1.12 0.264 -.0782655 .2844891 comlang_eth | -.0495423 .108831 -0.46 0.649 -.2635182 .1644335 ------------------------------------------------------------------------------ Regression (2) Regression with robust standard errors Number of obs F( 6, 386) Prob > F R-squared Root MSE = 392 =11659.42 = 0.0000 = 0.9939 = .63631 -----------------------------------------------------------------------------| Robust lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------home | 1.928083 .1783013 10.81 0.000 1.57752 2.278646 ldis | -1.221418 .0487213 -25.07 0.000 -1.31721 -1.125626 border | .3096226 .1076805 2.88 0.004 .0979089 .5213363 lgdp_o | .6470698 .0242636 26.67 0.000 .5993644 .6947751 lgdp_d | .6421693 .0243476 26.38 0.000 .5942988 .6900398 language | -.0242921 .1254375 -0.19 0.847 -.2709184 .2223342 ------------------------------------------------------------------------------ Regression (3) Regression with robust standard errors Number of obs F( 7, 385) Prob > F R-squared Root MSE = 392 =10290.45 = 0.0000 = 0.9940 = .63609 -----------------------------------------------------------------------------| Robust lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------home | 1.997584 .1862011 10.73 0.000 1.631485 2.363682 trust | -.2088498 .1837325 -1.14 0.256 -.5700944 .1523949 ldis | -1.208975 .0524294 -23.06 0.000 -1.312059 -1.105891 border | .3083327 .1063322 2.90 0.004 .0992682 .5173972 lgdp_o | .6625288 .0266275 24.88 0.000 .6101754 .7148822 lgdp_d | .6478927 .0243743 26.58 0.000 .5999693 .6958161 language | .0005466 .1255882 0.00 0.997 -.246378 .2474711 ------------------------------------------------------------------------------ Regression (4) Regression with robust standard errors Number of obs F( 9, 383) Prob > F R-squared Root MSE = 392 =12026.81 = 0.0000 = 0.9955 = .55081 -----------------------------------------------------------------------------| Robust lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------home | 2.169257 .1553935 13.96 0.000 1.863726 2.474788 trust | 2.032039 .3407697 5.96 0.000 1.362026 2.702053 ldis | -.9300413 .0506991 -18.34 0.000 -1.029725 -.830358 border | .3256374 .0916328 3.55 0.000 .1454712 .5058037 lgdp_o | .7206315 .0251697 28.63 0.000 .6711434 .7701195 lgdp_d | .6830237 .0258302 26.44 0.000 .6322369 .7338105 language | .2078467 .1280135 1.62 0.105 -.0438505 .4595439 ave_o | -.0904678 .4223943 -0.21 0.831 -.9209699 .7400342 ave_d | -3.9743 .3254088 -12.21 0.000 -4.614112 -3.334489 ------------------------------------------------------------------------------ Regression (5) Random-effects GLS regression Group variable (i): ctry_code Number of obs Number of groups = = 392 14 R-sq: Obs per group: min = avg = max = 28 28.0 28 within = 0.9288 between = 0.9302 overall = 0.9288 Random effects u_i ~ Gaussian corr(u_i, X) = 0 (assumed) Wald chi2(9) Prob > chi2 = = 4980.37 0.0000 -----------------------------------------------------------------------------lx | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------home | 2.790289 .1589403 17.56 0.000 2.478772 3.101806 trust | 1.351838 .2964185 4.56 0.000 .7708686 1.932808 ldis | -.709331 .0578595 -12.26 0.000 -.8227335 -.5959285 border | .4435777 .0947216 4.68 0.000 .2579268 .6292287 lgdp_o | .7630332 .0807645 9.45 0.000 .6047378 .9213286 lgdp_d | .7903213 .0251363 31.44 0.000 .7410552 .8395875 language | .2597148 .1202689 2.16 0.031 .023992 .4954375 ave_o | .275094 .8996495 0.31 0.760 -1.488187 2.038375 ave_d | -1.592278 .422366 -3.77 0.000 -2.420101 -.7644562 _cons | -7.044771 2.005167 -3.51 0.000 -10.97483 -3.114717 -------------+---------------------------------------------------------------sigma_u | .31558634 sigma_e | .45097034 rho | .32872915 (fraction of variance due to u_i) ------------------------------------------------------------------------------ Regression (6) Fixed-effects (within) regression Group variable (i): ctry_code Number of obs Number of groups = = 392 14 R-sq: Obs per group: min = avg = max = 28 28.0 28 within = 0.9296 between = 0.8222 overall = 0.8871 corr(u_i, Xb) = 0.1877 F(9,369) Prob > F = = 541.76 0.0000 -----------------------------------------------------------------------------lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------home | 2.801074 .1591954 17.60 0.000 2.48803 3.114118 trust | 1.360973 .2956274 4.60 0.000 .779647 1.942299 ldis | -.7034577 .0582208 -12.08 0.000 -.8179438 -.5889715 border | .4460948 .0948203 4.70 0.000 .2596389 .6325507 lgdp_o | .6085404 .2952991 2.06 0.040 .0278602 1.189221 lgdp_d | .7915117 .0250989 31.54 0.000 .7421568 .8408666 language | .2574582 .1202458 2.14 0.033 .0210053 .4939112 ave_o | -2.651216 1.4833 -1.79 0.075 -5.567997 .2655646 ave_d | -1.587346 .4210558 -3.77 0.000 -2.415316 -.7593764 _cons | -.1413576 4.386081 -0.03 0.974 -8.766207 8.483492 -------------+---------------------------------------------------------------sigma_u | .52694944 sigma_e | .45097034 rho | .57722843 (fraction of variance due to u_i) -----------------------------------------------------------------------------F test that all u_i=0: F(13, 369) = 10.77 Prob > F = 0.0000 Regression (7) Instrumental variables (2SLS) regression Source | SS df MS -------------+-----------------------------Model | 1418.6915 9 157.632389 Residual | 109.609017 382 .2869346 -------------+-----------------------------Total | 1528.30052 391 3.90869698 Number of obs F( 9, 382) Prob > F R-squared Adj R-squared Root MSE = = = = = = 392 550.35 0.0000 0.9283 0.9266 .53566 -----------------------------------------------------------------------------lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------trust | 2.969168 1.775183 1.67 0.095 -.5211857 6.459522 home | 2.632225 .2931127 8.98 0.000 2.055909 3.208542 ldis | -.6363638 .0811599 -7.84 0.000 -.7959398 -.4767878 border | .4500467 .1069736 4.21 0.000 .239716 .6603775 lgdp_o | .7633984 .0269485 28.33 0.000 .7104125 .8163844 lgdp_d | .8021323 .0311932 25.72 0.000 .7408005 .8634641 language | .3088377 .1377965 2.24 0.026 .0379031 .5797723 ave_o | .1481123 1.717096 0.09 0.931 -3.22803 3.524255 ave_d | -3.056916 1.690059 -1.81 0.071 -6.379899 .2660676 _cons | -7.741912 2.554551 -3.03 0.003 -12.76465 -2.71917 -----------------------------------------------------------------------------Instrumented: trust Instruments: home ldis border lgdp_o lgdp_d language ave_o ave_d religion ------------------------------------------------------------------------------ NEW SAMPLE – outliers removed. Descriptive Statistics Table 11: Contingency Table of Countries and Borders ---------------------| border iso_o | 0 1 ----------+----------AUT | 6 2 BLX | 5 3 DEU | 4 4 ESP | 6 2 FRA | 4 4 ITA | 6 2 NLD | 6 2 PRT | 7 1 ---------------------- Table 12: Descriptive statistics of our Dummy Variables Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------religion | 128 .625 .4860252 0 1 border | 128 .3125 .4653337 0 1 language | 128 .15625 .3645189 0 1 Table 13: Descriptive statistics of our non-string variables _______________________________________________________________________________ -> year = 1995 Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------bil_trade | 64 68369.38 228996.6 168.3404 1467235 gdp_o | 64 656645.5 602105.6 87038.4 1929422 gdp_d | 64 656645.5 602105.6 87038.4 1929422 distcap | 64 1046.952 637.6471 68.44467 2314.488 trust | 64 1.716655 .1280362 1.447942 1.957317 _______________________________________________________________________________ -> year = 1997 Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------bil_trade | 64 70809.08 228159.7 211.4558 1383286 gdp_o | 64 694925.6 609313.6 98831.6 1907246 gdp_d | 64 694925.6 609313.6 98831.6 1907246 distcap | 64 1046.952 637.6471 68.44467 2314.488 trust | 64 1.714945 .1243087 1.443169 1.974113 Regression (8) Fixed-effects (within) regression Group variable (i): ctry_code Number of obs Number of groups = = 128 8 R-sq: Obs per group: min = avg = max = 16 16.0 16 within = 0.9494 between = 0.9283 overall = 0.9260 corr(u_i, Xb) = -0.4757 F(9,111) Prob > F = = 231.45 0.0000 -----------------------------------------------------------------------------lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------home | 2.551604 .269733 9.46 0.000 2.01711 3.086098 trust | 1.440578 .5196984 2.77 0.007 .4107606 2.470395 ldis | -.6376644 .1025647 -6.22 0.000 -.8409031 -.4344256 border | .4339471 .1194321 3.63 0.000 .1972844 .6706099 lgdp_o | 1.164393 .7413277 1.57 0.119 -.3045978 2.633383 lgdp_d | .6628016 .0438344 15.12 0.000 .5759408 .7496624 language | .1691973 .1427591 1.19 0.238 -.1136894 .4520839 ave_o | -1.92129 2.002992 -0.96 0.340 -5.890352 2.047771 ave_d | -3.052562 .7088904 -4.31 0.000 -4.457275 -1.647848 _cons | -4.822907 10.73718 -0.45 0.654 -26.09935 16.45354 -------------+---------------------------------------------------------------sigma_u | .4253232 sigma_e | .39264014 rho | .53989295 (fraction of variance due to u_i) -----------------------------------------------------------------------------F test that all u_i=0: F(7, 111) = 5.84 Prob > F = 0.0000 Regression (9) Source | SS df MS -------------+-----------------------------Model | 376.860407 7 53.837201 Residual | 1.97394324 121 .016313581 -------------+-----------------------------Total | 378.83435 128 2.95964336 Number of obs F( 7, 121) Prob > F R-squared Adj R-squared Root MSE = 128 = 3300.15 = 0.0000 = 0.9948 = 0.9945 = .12772 -----------------------------------------------------------------------------trust | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------home | .333029 .0692476 4.81 0.000 .195935 .470123 religion | .0162949 .0258372 0.63 0.529 -.0348567 .0674464 ldis | .0618936 .0236605 2.62 0.010 .0150514 .1087359 border | -.0147735 .0370184 -0.40 0.691 -.0880612 .0585143 lgdp_o | .0769189 .0104711 7.35 0.000 .0561886 .0976491 lgdp_d | .0170364 .0104711 1.63 0.106 -.0036938 .0377667 language | .2067025 .0364114 5.68 0.000 .1346166 .2787885 ------------------------------------------------------------------------------ Regression (10) Instrumental variables (2SLS) regression Source | SS df MS -------------+-----------------------------Model | -12233.675 9 -1359.29723 Residual | 12676.7211 118 107.42984 -------------+-----------------------------Total | 443.046098 127 3.48855195 Number of obs F( 9, 118) Prob > F R-squared Adj R-squared Root MSE = = = = = = 128 0.44 0.9121 . . 10.365 -----------------------------------------------------------------------------lx | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------trust | 140.1187 711.538 0.20 0.844 -1268.92 1549.158 home | -26.53683 148.1148 -0.18 0.858 -319.8445 266.7708 ldis | -1.228823 3.551842 -0.35 0.730 -8.262437 5.804792 border | -1.16677 8.270467 -0.14 0.888 -17.54455 15.21101 lgdp_o | -.216746 5.506419 -0.04 0.969 -11.12096 10.68746 lgdp_d | 1.125183 2.646265 0.43 0.671 -4.115141 6.365508 language | -8.744569 45.70638 -0.19 0.849 -99.25565 81.76651 ave_o | -103.7906 524.4018 -0.20 0.843 -1142.249 934.6677 ave_d | -110.8918 552.2553 -0.20 0.841 -1204.508 982.7241 _cons | 137.5045 706.8239 0.19 0.846 -1262.199 1537.208 -----------------------------------------------------------------------------Instrumented: trust Instruments: home ldis border lgdp_o lgdp_d language ave_o ave_d religion ------------------------------------------------------------------------------
© Copyright 2026 Paperzz