GLOBALIZATION and SIMILARITIES in CORPORATE GOVERNANCE: a CROSS-COUNTRY ANALYSIS Tarun Khanna, Joe Kogan, and Krishna Palepu

GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE: A CROSS-COUNTRY ANALYSIS Tarun Khanna, Joe Kogan, and Krishna Palepu

Abstract—Some scholars have argued that globalization should pressure International Corporate Governance Network, the Interna- firms to adopt the most efficient form of corporate governance; others maintain that such convergence will not occur because of path depen- tional Accounting Standards Committee, and the Interna- dence. We find robust evidence that economically interdependent coun- tional Organization of Securities Commissions, to name but tries have similar corporate governance laws protecting stakeholders. In a few. The OECD and the World Bank have issued guide- contrast, we find virtually no relationship between corporate governance practices and globalization in a battery of estimations at the country, lines for global principles of good corporate governance and industry, and firm levels. We conclude that globalization may have promote the dissemination of these guidelines (OECD, induced the adoption of some common corporate governance standards 2000; World Bank, 2001). This cottage industry of efforts but these standards may not have been implemented. received a fillip in the wake of recent financial crises, when I. Introduction at least some fingers were pointed at corporate governance problems. conomists have studied convergence for a long time. Theorists have joined the debate with gusto. Some, es- EMost of this work has focused on convergence in pousing faith in the efficiency-enhancing aspects of compe- national income levels (Solow, 1956; Baumol, 1986; Ro- tition (especially in the capital markets), aver that there will mer, 1986; Barro & Sala-i-Martin, 1992; Mankiw, 1995). be complete convergence. This position is exemplified by Recently, however, scholars have also directed attention to Hansmann and Kraakmann (2001) in a paper entitled “The the microeconomic foundations of such convergence. In End of History for Corporate Law.” Others take the polar particular, the question being asked is whether or not in- opposite view that the irresistible force of global competi- creased global integration of markets is prompting conver- tion will meet the immovable object of path dependence. gence in the institutional foundations of economies (North, Even if there could be agreement on what constitutes an 1994). The idea that increasingly stringent global competi- optimal corporate governance system, there are too many tion should increase the likelihood of convergence to effi- complementarities in economic systems for unstinting evo- cient institutional arrangements has some intuitive appeal lution toward the optimal corporate governance system and has been investigated in a historical context (Ben- (Aoki, 1994; Bebchuk & Roe, 1999).1 Further, vested inter- David, 1993; Williamson, 1996). ests might well oppose such an evolution (Olson, 1971). One set of institutional arrangements that has been scru- Finally, advocates of adherence to various sets of minimal tinized is related to the governance of corporations. The standards as being the only feasible outcome are implicitly number of standards-setting bodies and multilateral institu- signaling belief in partial convergence (Eichengreen, 1999). tions that have been set up in recent times to foster good These arguments virtually run the gamut of possibilities. corporate governance worldwide is testimony to the height- Yet there is no empirical work on whether there is conver- ened interest in this particular issue. These include the gence in corporate governance systems of any sort, nor any on the association of globalization with this phenomenon. Received for publication June 10, 2003. Revision accepted for publication March 4, 2005. Presumably this is partly because of the nontrivial nature of *Harvard Business School; Esquela de Administracioń, Pontifica Uni- the question. At some level, it is hard to dispute that there is versidad Cato´lica de Chile, and Lehman Brothers; and Harvard Business convergence. For example, there is probably some conver- School, respectively. We are grateful, for helpful conversations, to Michael Kremer, Dani gence to acceptance of the idea that resource providers Rodrik, Andrei Shleifer, John Sutton, and Louis Wells, and, for sugges- should be protected, but much less to whether the resource tions, to two anonymous referees as well as seminar participants at the providers in question should be primarily shareholders or HBS/Tsinghua conference on Global Corporate Governance (Shanghai, July 2001), the Aspen Institute for Social Business/William Davidson include other stakeholders as well (Berglof & von Thadden, Institute Conference (Aspen, September 2001), the Harvard/MIT Growth 1994; Shleifer & Vishny, 1997; Tirole, 2001). Similarly, and Development Seminar (October 2001), the University of Michigan there is probably near-complete convergence to the idea that Business School Corporate Strategy Seminar, the Asian Development Bank conference on Corporate Governance (November 2001), the Hito- good information is needed for good corporate governance, tsubashi University Economics Seminar (Tokyo, November 2001), the but much less convergence on whether financial markets or Australian Graduate School of Management Seminar (Sydney, November 2001), the Hong Kong University of Science and Technology Economics banks are the best targets for such information. Finally, Seminar (November 2001), the Columbia Law School Law & Economics Hansmann and Kraakman (2001) argue that several fea- Seminar (December 2001), the Instituto Tecnolo´gico Autońomo de Me´x- tures of the modern corporate form are nearly universally ico Business School Seminar (February 2002), the Pontificia Universidad Cato´lica de Chile Business School Seminar (March 2002), the Harvard Business School Accounting and Control Seminar (April 2002), the 1 Though not focusing on corporate governance per se, Rodrik (2000) William Davidson Institute Conference in Santiago, Chile (December points out that successful countries have often tailored their institutions to 2002), and the Columbia Law School Globalization Conference (2004). their idiosyncratic contingencies. Further, countries that have followed the The HBS Division of Research and Pontificia Universidad Cato´lica de prescription in vogue (for example, Latin America implementing Wash- Chile generously funded this research. All errors remain our own. ington Consensus reforms) have often failed.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 70 THE REVIEW OF ECONOMICS AND STATISTICS

adopted,2 but it is hard to imagine that there is convergence of globalization of capital markets, but depart from the on the actions taken by managers even in the convergent theoretical literature’s usual focus solely on capital markets legal structures that such firms represent.3 to consider competitive forces emanating from global mar- Scholars have made an attempt to account for such kets for products and talent. subtleties by distinguishing between convergence in “form” An important feature of our estimation technique—ex- and convergence in “function.” The latter arises because amining similarities in governance among pairs of countries different institutional arrangements in different countries rather than treating the individual country governance re- have sufficient plasticity to enable them to achieve similar gime as the unit of analysis—is that it avoids normative economic ends (Gilson, 2001). There is also ambiguity as to assumptions regarding the optimality of a particular gover- the system toward which convergence might occur. The nance regime, a stance vindicated by the continued theoret- syste`me du jour is the U.S. shareholder-centered one, ical debate on this issue. We choose to let the data speak though other systems have found favor in the not very rather than impose our priors on the data. distant past (Berglof & Perotti, 1994; Porter, 1992). But our analysis has two important limitations worth Empirical issues compound these theoretical conun- highlighting. First, lacking time series data, we are unable to drums. First, looking purely at rules on the books as indi- test convergence in a literal sense. Rather, our tests are cators of corporate governance practices is unlikely to be about correlations between globalization and similarity in sufficient, even though it is one indicator (Pistor, Raiser, & corporate governance practices at a point in time. It is thus Gelfer, 2000). It could be that formal rules are circumvented possible that convergence is occurring but that the de facto in the all too common absence of good enforcement—a practices today do not allow us to see this process. Second, familiar phenomenon in many developing countries. Much we do not say much about causality. We cannot distinguish less common, but quite possible, is the idea that particular between the possibility that the flows of global factors and firms might well exceed the rules of their country of domi- products force changes in corporate governance and the cile in a bid to gain economic advantage (Blass & Yafeh, possibility that the adoption of uniform corporate gover- 2001; Khanna & Palepu, 2004). Thus attention to intracoun- nance systems gives globalization a boost. try variation is necessary. Second, consistent cross-country We summarize our results as follows. We find strong data on the multiple aspects of corporate governance— evidence that de jure similarity in governance is correlated board structure, compensation practices, disclosure, and with several of our proxies for globalization. These proxies transparency, to name but a few—have traditionally been are not limited to those that measure capital market integra- difficult to come by. tion. Further, the de jure results are not driven by similarity with U.S. corporate governance. Rather, pairs of economi- We attempt to fill this empirical void partially by analyz- cally interlinked countries display similarity to each others’ ing a variety of new data sets. A number of private-sector systems, especially if both countries are economically de- organizations have recently started collecting firm-level veloped. Finally, we find virtually no evidence of de facto data on corporate governance practices. This phenomenon is similarity. An interpretation is that, even though countries itself evidence of convergence in that it demonstrates a might mimic the tenets of each others’ systems, their im- belief on the part of at least some organizations that insti- plementation is subject to significant lags. tutional investors are willing to pay for corporate gover- The rest of the paper is organized as follows. Section II nance information. We focus primarily on year-2000 data briefly reviews the literature on convergence in corporate obtained from Credit Lyonnais Securities Asia (CLSA) on governance. Section III introduces the governance ratings, corporate governance practices covering 24 developing covering selection effects, and reliability issues. It also countries in Asia, Latin America, and eastern Europe, and motivates and describes the independent variables we will data on laws protecting shareholders and creditors from La 4 use in our analysis. Section IV discusses the methodology of Porta et al. (1998). Finally, we develop a host of measures the country-level analysis and presents the results. Section V confirms the country-level results using additional data 2 They emphasize full legal personality, limited liability, shared owner- sources and tests at the industry and firm levels. Section VI ship by investors, board structure, and transferable shares. A similar argument is made regarding the organization of nation-states by Meyer et concludes. al. (1997). They conduct the interesting thought experiment: How would world society treat a hypothetical country that newly emerged on the world stage? They aver that several aspects of societal organization in II. The Literature on Convergence such a nascent country would mirror arrangements commonly found elsewhere, that is, that there would be convergence. in Corporate Governance 3 Khanna, Palepu, and Srinivasan (2004) find some evidence that disclosure regimes of foreign companies (a facet of governance practices) Globalization entails a lifting of barriers to the mobility converge in response to exposure to U.S. markets. But this convergence is of capital, products, and labor, leading to an intensification far from complete, and other facets of managerial practices do not appear to converge as much (Khanna & Palepu, 2004). of competition for these factors across borders by firms and 4 These data, and supplementary data that we use from De´minor and Proxinvest, have not been used in academic work, to the best of our strong relationship between the market value of Russian firms and the knowledge. Black’s (2001) data source differs from our own. He finds a corporate governance rating they received from UBS Brunswick Warburg.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 71

countries. Just as U.S. states competed for most of the in Japan depended for its functioning on the existence of twentieth century for franchise tax revenues by offering the particular labor market practices (Aoki, 1994) and patterns best terms for incorporation (Easterbrook & Fischel, 1991; of equity cross-holdings among firms (Berglof & Perotti, Romano, 1993),5 nations could compete for firms and re- 1994). Sunk adaptive costs and positive network externali- sources by creating the most efficient corporate governance ties from the use of the same governance practices by all environment.6 firms in a country make switching more costly and prevent Proponents of the convergence hypothesis tend to high- unilateral governance changes by any one firm or institution. light the role of global capital flows in eliminating ineffi- Even in the absence of path dependence, complementa- cient forms of governance.7 They aver that convergence is rities induce multiple optima, so that nations, even unbur- hastened by a realization that alternatives to U.S.-style dened by historical constraints, may well choose different shareholder-centered governance have generally not suc- bundles of practices that yield equivalent long-run corporate ceeded. Several multilateral bodies are spurring on this governance.8 Believers in such a view of the world disagree process by urging the adoption of common standards with the idea that the U.S. model is the one optimal (OECD, 2000; World Bank, 2001). governance model. Not very long ago the Japanese model An aspect of the convergence debate recognized but not held pride of place in the minds of academics and practi- emphasized in the literature is the distinction between de tioners (Porter, 1992). In particular, the U.S. model was jure and de facto convergence. De jure convergence is the criticized at the time for three failures relative to the Japa- adoption of similar corporate governance laws across coun- nese system—agency costs arising from differences in man- tries. De facto convergence, on the other hand, refers to a agers’ and shareholders’ agendas, time-horizon costs due to convergence of actual practices. Put simply, nations may the reliance of distant shareholders on short-term perfor- formally adopt other countries’ corporate governance sys- mance, and industrial organization costs caused by the tems, but the acceptance of the enshrined principles may bureaucracy in vertical organizations that could have been significantly lag their codification. This may be for several lessened in the presence of strong financial intermediation reasons, including a lack of understanding of what is im- (Roe, 1996). It is more likely that there are tradeoffs plied by good corporate governance, absence of comple- between the systems and that each has costs and benefits mentary institutions needed to implement the principles, or suited to different circumstances (Bhide, 1993). Consistent simply poor enforcement (Pistor et al., 2000). One contri- with the idea that there is no universally optimal model, bution of our paper is to empirically demonstrate a distinc- research on the effects of corporate governance on perfor- tion between these two types of convergence across a large mance has produced ambiguous results (see, for example, sample of countries. Demsetz & Lehn, 1985; Thomsen & Pederson, 1996; Coles, There are three categories of explanations for why de McWilliams, & Sen, 2001). facto convergence may lag de jure convergence: path de- The force of path dependence leads some theorists, pri- pendence, multiple optima, and rent-seeking by interest marily legal theorists interested in the foundations of finan- groups. According to the path dependence perspective, ini- cial markets, to draw a distinction between convergence in tial historical conditions matter in determining the corporate form and convergence in function. Formal convergence governance structures that are prevalent today. Central to predicts a convergence of legal rules and institutions, the idea of path dependence are complementarities, also whereas functional convergence predicts adaptations within called indivisibilities (Bebchuk & Roe, 1999; Bratton & different existing institutions to perform the functions of McCahery, 1999; Schmidt & Spindler, 2002). Corporate good corporate governance. Proponents of functional con- governance results from a system of complementary insti- vergence would argue that different institutions are equally tutions, legal rules, and practices where improving any one capable of performing corporate governance functions such 9 element independently may actually hurt efficiency. Thus as ensuring management accountability. To the extent that the main bank monitoring system that historically held sway it is possible, it is much less costly to improve corporate performance by working within the current institutions than 5 There is some disagreement on whether competition between states has by creating new institutions. Thus, the initial response to the led to a race to the top or a race to the bottom. For example, Bebchuk, competitive forces of globalization should be functional Cohen, and Ferrell (2002) showed empirically that state competition has rather than formal convergence (Gilson, 2001). worked poorly in takeover regulation. 6 Although most industrial nations do not permit reincorporation across countries such as the U.S. allows between states, firms are able to opt into 8 See Kauffman (1993) for an early articulation of this from evolutionary particular regulatory regimes by listing on foreign stock exchanges. In the biology, and Milgrom and Roberts (1990) for an adaptation to economics. case of Centros, the court ruled that a Danish company could incorporate 9 See Merton (1968, “Chapter 1: Manifest and Latent Functions”) for an in the United Kingdom even though it had no intention of doing business early description of functional analysis in sociology. Crane, Bodie, and in the United Kingdom and simply wanted to avoid the minimum capital Perold (1995) is a more recent example of functional analysis as applied requirement for Danish incorporation. See European Court of Justice, to financial systems. Kaplan (1994) has provided some econometric March 9, 1999, C. 21/297, as cited in Gilson (2001). evidence for functional convergence for a particular aspect of corporate 7 Khanna and Palepu (2004) have demonstrated the role of global talent governance. Statistically, poorly performing CEOs appear equally likely markets and global product markets in fostering some corporate gover- to be dismissed in the United States, Germany, and Japan, despite the very nance convergence in the global software industry. different formal systems in place.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 72 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 1.—SUMMARY STATISTICS No. of Variable Countries No. of Pairs Mean Std. Dev. Min. Max. Variable Deﬁnition

Dependent Variables

De jure governance distance 49 1,176 5.26 2.08 0.00 10.00 Sum of abs. differences along 13 dimensions of legal protections De facto governance distance 24 276 5.91 2.69 1.89 16.74 Norm. sqrt of sum of squared differences along 7 dimensions of governance practices

Economic Integration

Wage correlation 45 990 0.35 0.29 Ϫ0.63 0.94 Correlation coefﬁcient between average monthly wages by occupation Capital market correlation 50 1,225 0.28 0.18 Ϫ0.12 0.84 Correlation coefﬁcient between weekly % changes in national stock indices FDI partner 51 1,275 0.03 0.09 0.00 0.93 Share of total FDI in country A originating in B and vice versa FDI magnitude 49 1,176 0.03 0.14 0.00 1.80 FDI partner multiplied by FDI/GDP for each country Trade partner 54 1,431 6.95 13.66 0.00 194.19 Share of total imports and exports in country A originating in B and vice versa Trade magnitude 54 1,431 0.01 0.03 0.00 0.40 Pairwise total trade divided by GDP of each country Geographic distance 54 1,431 8.08 4.81 0.07 19.84 Distance in 1,000s of kilometers between the capitals of two countries Common language 54 1,431 0.32 0.47 0.00 1.00 Equals 1 if the two countries have a language in common.

Control Variables

Diff. in log GDP/capital 54 1,431 1.03 0.79 0.00 3.53 Absolute difference in the log of GDP/capita between the two countries Common legal origin 54 1,431 0.27 0.44 0.00 1.00 Equals 1 if the two countries have the same legal origin according to LLSV data

Even if there were a single universally optimal corporate III. Data governance system, political resistance would pose a major We use primarily two data sources on corporate gover- obstacle to governance reform. Because the benefits of nance practices across a number of countries. Laws-on-the- improved corporate governance are not distributed evenly, books indicators collected by La Porta et al. (1998) (LLSV) an increase in social welfare that exceeds switching costs represent indicators of de jure governance. A firm-level would not guarantee the adoption of better practices. This is survey completed by Credit Lyonnais Securities Asia an idea dating back at least to Olson (1971). Interest groups (CLSA), when aggregated to the country level, provides such as labor unions, banks, controlling shareholders, and ratings on de facto governance. Even though much of our lawyers may sabotage governance reform (Bebchuk & Roe, analysis ends up using the LLSV data, we describe the 1999; Coffee, 1999). U.S.-style governance based on indi- CLSA data in detail. This is because it is new to academic vidualism may also conflict with values and anti-Western analysis, and because our inability to detect de facto con- sentiment found in many developing countries (Branson, vergence must be judged within the constraints of the 2001). quality of the CLSA data. Our independent variables are Despite these ongoing debates, there is exceedingly little proxies for country-level exposure to global capital, prod- empirical work, and none that confronts the nexus between uct, and labor markets. Table 1 provides summary statistics globalization and corporate governance convergence that is on all the variables. the topic of this paper.10 A. Dependent Variable: Corporate Governance Ratings 10 Guilleń (2000) approaches the convergence hypothesis quantitatively. He examines descriptive summary statistics of aggregate national corpo- Legal Rules, Country-Level Data: We use novel data rate governance indicators over time for a panel of countries. With a few assembled by La Porta et al. (1998) on laws regarding exceptions, he finds that there have been no shifts in the proportion of shareholder and creditor protection for 49 developed and stockholding by institutional investors, debt-equity ratios, long-term in- centives as a percentage of CEO remuneration, and hostile takeover developing countries. Each of eight aspects of shareholder activity. protection and five aspects of creditor protection can be

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 73

coded as binary variables.11 A value of 1 for a variable Further, because the firm-level data are about actual indicates that, say, the country’s company law and commer- practices, rather than laws on books, these data address de cial code or bankruptcy and reorganization law incorporate facto rather than de jure convergence. A disadvantage of that form of shareholder or creditor protection. Because these data, relative to the de jure measures, is that the CLSA corporate governance laws typically apply to all firms and data cover only developing countries, whereas the laws-on- industries, these variables are coded only at the country the-books indicators span both developed and developing level. markets (in Europe, Asia, and Latin America). An advantage of these data is that they are collected By averaging the scores across all firms within a country, objectively. There is little uncertainty regarding whether or we obtain de facto governance scores at the country level. not a particular law exists within a country. La Porta et al. The aggregation of firm-level practices to the country level (1998) also identify the origin of the legal system of each of is validated by the existence of a strong country component their countries—English, French, German, or Scandina- in firm governance scores. We have shown, using Hotelling vian—which we use as a measure of path dependence in generalized means tests, that there exist statistically significant differences between the means of firm governance some of our estimations. practices in pairs of countries. Furthermore, we found that firms located in the same country tend to have similar Credit Lyonnais Securities Asia (CLSA): In April 2001, governance practices, and this effect was statistically sig- CLSA released a large study of corporate governance enti- nificant at the 1% level. tled “Saints & Sinners: Who’s Got Religion?” which re- The most significant limitation of the CLSA data is that viewed corporate governance at the firm level in 25 emerg- they are based on subjective opinions. Although research ing markets. A firm’s corporate governance score is based analysts rather than firms complete a detailed and method- on responses to 57 yes-no questions about various aspects of ical questionnaire, for some of the questions the analysts governance. CLSA’s research analysts completed the ques- must rely on information provided to them by the firms they tionnaire for 495 companies over a six-week period ending are rating. Because firms are asked to provide information March 2001. Of these 57 questions, 70% are based on about their corporate governance, a substantial amount of objective facts, and the remaining 30% require some inter- misreporting can be expected. Misreporting can lead to pretation by the analyst but still must be resolved as a yes or noisy data, and, even worse, if firms with poor governance no answer. In cases where information was not available or were more likely to misreport, then the ratings would not evidence was lacking, analysts were instructed to answer measure the strength of governance at all. The reliance on negatively. In as much as a lack of corporate governance intermediaries such as consulting firms and investment information should be correlated with poor corporate gov- banks to interpret the data and generate the ratings may also ernance, this practice should not lead to large errors. The lead to biases. These intermediaries are hardly disinterested CLSA questionnaire is divided into seven sections covering parties, and in some cases they are rating firms that may be fiscal discipline, accounting transparency and disclosure, their clients. board independence, board accountability, responsibility, We handle these reliability issues in two ways. First, the equitable treatment of shareholders, and social awareness. thesis of CLSA’s “Saints & Sinners: Who’s Got Religion?” The score for each of the seven sections is simply the is that corporate governance is correlated with performance. percentage of questions receiving a yes answer in that We have confirmed that CLSA’s governance ratings are 12 section. correlated with firm P/E ratios. Although we are not able Note that the absolute value of the CLSA score for a to determine the direction of causation, the correlation of company, therefore, effectively measures proximity to the corporate governance ratings with investors’ valuations of Anglo-Saxon system of governance (where, for example, the firm suggests that the ratings are capturing a real and relevant variable. Second, as described in appendix A, we there is greater disclosure to outside parties and boards are generate our own ratings of egregious corporate governance composed of independent outsiders), rather than proximity problems based on public sources only, avoiding some of to alternative relationship-based governance systems. the biases present in the other ratings. Based on business newspaper and magazine articles, we compiled a list of 11 The eight shareholder protections are: one-share–one-vote rule, proxy vote by mail, shares not blocked before meeting, cumulative voting or firms with reported instances of minority shareholder ex- proportional representation, mechanism for dealing with grievances of propriation in the last two and a half years, and other kinds oppressed minorities, preemptive rights to buy new stock issues by of corporate governance problems. Because compiling such existing shareholders, low percentage of share capital required to call extraordinary shareholder meeting, and mandatory dividend payment. The data is highly labor-intensive, we focused the collection five creditor protection are: restrictions on going into reorganization, no efforts only on firms in India, the emerging market for automatic stay on secured assets in reorganization, secured creditors first which CLSA provided data on the largest number of firms. in line, management does not stay in reorganization, and mandatory minimum percentage of total share capital to avoid dissolution. More detailed descriptions of these protections are available in La Porta et al. 12 The Spearman’s rho was 0.14, and this coefficient was significant at (1998). the 5% level.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 74 THE REVIEW OF ECONOMICS AND STATISTICS

We find that the CLSA rating is correlated with our index of markets is transmitted through such flows. For example, minority shareholder exploitation (but not as strongly with trade flows or simply the threat of trade flows from abroad other corporate governance problems). It appears as though puts pressure on domestic firms to be more efficient. Sim- the CLSA index does capture major governance problems. ilarly, the potential for capital flows should make countries We also check CLSA’s sample selection procedure. Ac- and firms more likely to reform governance in order to cording to CLSA, the sample of firms was selected based on attract this capital. Our measures of actual flows include two criteria: size and investor interest. “[B]eyond the largest pairwise trade in products and pairwise foreign direct in- stocks in each market, our coverage would be biased to- vestment (FDI). Our measures of potential flows include wards companies that we see as likely to be of interest to geographic distance and the presence of a common lan- institutional investors and in making this decision, an inher- guage. ent sampling bias creeps in.” If investor interest were A variable denoting the origin of the legal system is used correlated with corporate governance, then the portion of to control for path dependence. If path dependence is indeed the sample that was selected after the largest firms would be a force in the evolution of corporate governance within a biased toward well-governed firms, and larger firms in the country, then the starting point should have an effect on the CLSA sample would appear to have poorer corporate gov- current corporate governance laws and practices. La Porta et ernance. We test the determinants of selection empirically al. (1998) found that the origin of the legal system is a by focusing on the selection of firms in India, to ease the significant determinant of current legal structure. We treat data-gathering process. We use Bloomberg to generate a the origin of the legal system as a proxy for the starting pool of firms from which the CLSA analysts could have point because it was determined exogenously many years selected. Appendix B describes a probit analysis of the ago.14 If the origin of the legal system is the only historical determinants of CLSA firm selection in this pool. This factor relevant in determining current legal systems, then analysis confirms that firms were selected based on size and our test for similarity becomes a test for convergence as investor interest, as CLSA claims. well.15 We motivate and describe the independent variables Finally, for the bulk of our analysis, we aggregate the in greater detail below. firm-level CLSA governance ratings for all firms within a country, along each of the dimensions captured by the data. Wage Equalization: Under conditions of perfect labor We use the average ratings thus generated as country-level mobility, workers of a particular skill level, ability, or indicators of the particular facets of corporate governance occupation should receive the same wage across countries (disclosure, transparency, board independence, and so on), after controlling for differences in the cost of living. Until the analogs to the country-level de jure indicators in the recently, no systematic data on wages across countries were LLSV data. This process averages out at least some of the available. Freeman and Oostendorp (2001) have refined the idiosyncratic subjectivity in responses, though obviously it ILO international database of wages to create a data set of does not remove country-level response biases. monthly average pay for male workers coded for 161 occupations across 154 countries over the years 1983 to B. Independent Variables Used in Estimations 1998.16 Because each country had data available for different years and different occupations and the data are pre- Our country-level regressors try to capture the degree of sented in current local currency units, we cannot simply capital market, product market, and labor market integration compare wage rates by occupation across countries. Instead, between pairs of countries. Factor price equalization pro- we examine the structure of wages by calculating the vides one theoretically defensible measure of the extent to correlation of wages by occupation across country pairs.17 If which two markets are integrated.13 Though we do not wages are equalized, then we expect a correlation coeffi- actually have such measures, we use two approximations. cient of 1; for example, if doctors earn more than lawyers in The first pertains to a measure of the correlation in wage one country, then they would earn more than lawyers in the structure in pairs of countries, the second to comovement in equity prices. Of course, such correlations might well be 14 Rajan and Zingales (2003) have argued against such an interpretation, induced by common unobservables that have nothing to do however. with true market integration. 15 To see this, assume that the origin of the legal system entirely determines the starting point of the corporate governance environment. Accordingly, we also examine quantity rather than price The LLSV and CLSA data represent the current corporate governance measures of integration, using measures of both actual and environment. In this way, we implicitly have data on corporate governance potential flows between countries. Competitive pressure in for each country at two points in time. This information allows us to estimate whether corporate governance between pairs of countries is converging. 13 Similarly, a lack of equalization in the prices of capital and labor 16 We are grateful to Remco Oostendorp for sharing these data with us. would indicate the presence of barriers to the mobility of these factors. 17 The actual procedure is to first take an average of wages by occupation However, the converse of this statement is not necessarily true. We know for all available years from 1983 to 1998 for each country. The correlation from international trade theory that factor price equalization may occur coefficient is calculated on a pairwise basis for all occupations for which when factors are not mobile but products are. [See, for example, Samuel- both countries have reported data. This method has obvious limitations in son (1948) for an early version of this idea.] that different pairs of countries have different occupations in common.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 75

other country as well. A drawback of this proxy is that wage tion, pressures in a vertical relationship are also possible. structure may itself be an indicator of corporate governance We develop two related measures. The first, Trade Partner, separate from the effects of cross-country labor mobility; is appropriate if trade bears the same importance in the certain rankings of wages by occupation may result from a political process regardless of size, perhaps because of its shareholder-centered system, and other rankings from a potential for growth in the future. The second, Trade Mag- labor-centered system. nitude, assumes that the importance of trade to political decision-makers is proportional to its current share in the Capital Market Comovement: A rough measure of economy. Trade Partner is the sum of the ratios IMPij, economic integration is the correlation of weekly stock IMPji, EXPij, and EXPji, where IMPij is the share of total market index changes in percentage terms between two imports into country i originating from country j, and EXPij countries. By definition, movement in stock prices is caused is the share of total exports of country i that are going to by either a change in fundamentals or a change in discount country j. Although this measure uses the same numerator rate. If an economic shock occurs in one country and the twice, it provides the most complete measure of whether the two countries are economically interdependent, then the two countries are major trading partners from either coun- shock will also affect the other country. Because fundamen- try’s perspective. Trade Magnitude takes into account the tals in both countries are affected, the stock prices of both fact that international trade plays a larger role in some countries should move in the same direction. Alternatively, economies than others. For example, trade varies with the if two countries have the same set of investors, then the degree of openness. Furthermore, for a given level of discount rates in the two countries will move together and openness, trade is more important for a small country than the stock markets will be correlated. In this case, comove- for a large country. Trade Magnitude is defined as the sum ment becomes a proxy for integration of two countries into of pairwise total trade divided by the GDP of the first 18 a single capital market. Stock price data come from the country and the same total trade divided by the GDP of the Morgan Stanley stock indices, and, for a few countries second country. The pairwise trade data come from the where a Morgan Stanley index was not available, from the World Trade Analyzer for 1996. national indices. Most countries had data available for the period January 1993 to August 2001, but in a few cases Foreign Direct Investment: Pairwise FDI can proxy for shorter time series were used. pressures for convergence of corporate governance in sev- We intentionally measure correlation rather than causa- eral ways. Multinationals might operate more efficiently tion for this variable for reasons that the following examples than local firms because of their access to home-country clarify. First, suppose Argentinean companies were to start institutions—capital, talent, or technology, for example. raising money in Chile; external shocks to Chilean investors Competitive forces from the multinationals would then would then affect both Chilean and Argentinean markets. force local firms to abandon inefficient corporate gover- Although there is a causal link between the shock to Chilean nance structures (Foley, 2002). Mirroring our constructs in investors and Argentina’s capital market, there is no causal the case of the pairwise trade variables, we use two vari- link between the Chilean and Argentinian stock markets. ables for foreign direct investment, called FDI Partner and Second, suppose heavy snows in the mountains in Chile FDI Magnitude. Of these, FDI Partner indicates to what disrupt trucking routes with Argentina. Snow thus causes a degree inward investment in one country originates from decline in Chilean stock markets and in the Argentinean another country. It is defined as the sum of FDI and FDI , stock markets; however, there is no causation between ij ji where FDI is the percentage of total FDI in country i Chilean and Argentinean stock markets. Again, the relevant ij j. proxy for economic integration is a correlation coefficient. owned by firms in country Because FDI plays a larger role in some economies than others, FDI Magnitude weights the two percentages by the degree to which FDI is present in the Trade: Flows of goods between two countries are in- ϫ dicative of openness between these countries, at least in two economies. The variable is defined as FDIij (average ϩ ϫ product markets. The flow measures include both intra- FDI flows/GDP)i FDIji (average FDI flows/GDP)j. industry and interindustry trade; although competitive pres- Data for pairwise FDI are taken from the United Nations sures should be greater in cross-border horizontal competi- World Investment Directory, volumes 1–7.

18 Of course, correlated changes in fundamentals may have other causes. Geographic Distance and Common Language: Geo- For example, if two economies were affected by the same severe weather graphic distance and common language predict the likeli- patterns, then their stock markets would be correlated even if there were hood of flows between countries. We expect countries that no economic linkages between them. In the case of price fluctuations due to changes in discount rates, the measure is also imperfect, because it is are located closer together to have lower transport costs and possible that each stock market has separate investors with correlated smaller information acquisition costs, facilitating the flows preferences. Because comovement represents only one of the many of goods and factors. Similarly, flows between countries are proxies we are testing, we do not attempt to separate correlations due to fundamental and nonfundamental factors, using only a simple correlation aided by the presence of a common language between the coefficient for the proxy. two countries. These two variables have been extensively

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 76 THE REVIEW OF ECONOMICS AND STATISTICS

used in the literature on gravity models in international nomic integration: wage correlation, capital market co- trade. Of course, geography and language might well proxy movement, FDI partner, FDI magnitude, trade partner, trade for other underlying variables (Acemoglu, Robinson, & magnitude, geographic distance, and common language. We Johnson, 2001; Diamond, 1999). Data are from the atlas and use the quadratic assignment procedure (QAP) as described from the CIA World Factbook. in appendix C to calculate standard errors for such dyadic data (Krackhardt, 1988). Because the independent variables IV. Similarity at the Country Level are correlated and, in many instances, meant to proxy for the same type of integration, we test the effect of each of these We test the hypothesis that the corporate governance separately. We distinguish between the effects of integration environment in a particular country will be affected by that and historical path dependence by controlling for common country’s integration into worldwide capital markets, prod- origin of legal system. We also control for differences in uct markets, and labor markets. We make no assumptions level of development on the grounds that governance is regarding which country has the optimal governance system expected to differ between developed and developing coun- and which other countries will try to emulate it, and instead tries. Less developed countries do not have the resources or test a general version of the hypothesis. Economic linkages expertise to refine corporate governance laws and practices. between any two countries are hypothesized to create com- Because the level of economic integration may differ be- petitive pressure in each of the two countries to adopt the tween developed-developing and the developed-developed optimal elements of its partner’s governance structure. and developing-developing pairs, this control eliminates a Thus, pairs of countries with greater economic linkages potential omitted variable bias. should have more similar governance structures.19 We perform the estimations first using the de jure governance data and then using the de facto data. We explore the de jure B. De Jure Analysis results in detail to understand which pairs of countries are We find a strong correlation between economic integra- driving the results. tion and de jure governance similarity. Table 2 displays the baseline regressions. All of the globalization proxies are of A. Methodology the expected sign, and five of the variables (wage correla- We use the thirteen indicators of legal rules from the tion, trade partner, trade magnitude, geographic distance, LLSV data to construct a measure of de jure governance and common language) are significant at the 10%, 5%, or similarity between two countries i and j as follows: even 1% level. The predicted magnitude of the effect, expressed as the percentage-of-a-standard-deviation change L in the dependent variable resulting from a standard- ϭ ͸͉ Ϫ ͉ deviation change in the independent variable, ranges from Governance Distancei, j Ci,l Cj,l , lϭ1 8% to 29% for the variables that were significant. Common origin of the legal system is of the expected sign and where Ci,l represents a particular binary legal characteristic significant at the 1% level in all regressions. Difference in l of country i. The analogous measure of de facto similarity development is of the expected sign in all but one of the using the CLSA data is regressions, but is not significant. We have also performed this analysis for shareholder rights and creditor rights sep- L arately. We find a link between integration and similarity for ϭ ͸͑ Ϫ ͒2 Governance Distancei, j ͱ Ci,l Cj,l , both of types of governance, though the results for share- lϭ1 holder rights are stronger. Which pairs of countries are driving the results? We first where Ci,l represents a 1-to-10 rating of the strength of test whether the results are driven by integration with the protection in country i for dimension l. United States and similarity to its governance. We create For each possible pair of countries, we regress these two new independent variables to replace each independent dependent variables on eight measures of pairwise eco- variable used in the baseline regression. The first independent variable equals the proxy for integration when the pair 19 Our methodology is distinct from the gravity equation of international contains the United States and equals 0 otherwise. The trade (see, for example, Anderson, 1979; Leamer and Levinsohn, 1995; Evenett and Keller, 2002), which predicts that the volume of trade second independent variable equals the proxy for integra- between two countries is proportional to the product of the two countries9 tion when the pair does not contain the United States and GDPs. We do, however, use some similar independent variables in our equals 0 otherwise. regressions, such as geographic distance and common language. Our approach is similar in spirit to Mukand and Rodrik’s (2002) model, in Contrary to expectations, the regression analysis of the which countries, in the course of their search for optimal policies, may effects of these new independent variables (panel B1) re- adopt the policies of “leaders.” In their empirical implementation, using veals that U.S. pairs are not driving the results. For U.S. data from postsocialist countries, they probe the relationship between adoption of western European institutions and geographic distance (a pairs, only three of the eight variables are of the expected proxy for institutional distance) from western Europe. sign, and only one of these three is significant (and merely

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 77

TABLE 2.—DYADIC ANALYSIS OF DE JURE COUNTRY DATA

A. Baseline Dyadic Regressions

Common- Diff.-in- Origin-of- Indep.- Log-GDP/ Legal- Expected Variable Cap. System No. of No. of Independent Variable Sign Coeff. Coeff. Coeff. R-Squared Countries Pairs Magnitude Deﬁnition of Variable

Wage correlation ϪϪ1.1707** Ϫ0.0243 Ϫ1.3702*** 0.1116*** 40 780 Ϫ18% Correlation coefficient between average (1.6%) (44.6%) (0.0%) (100.0%) monthly wages by occupation Capital market correlation ϪϪ0.1831 0.1405 Ϫ1.2867*** 0.0905*** 45 990 Ϫ2% Correlation coefficient between weekly % (39.6%) (84.4%) (0.0%) (100.0%) changes in national stock indices FDI partner ϪϪ1.4039 0.0463 Ϫ1.3075*** 0.0938*** 47 1,081 Ϫ6% Share of total FDI in country A originating in (13.0%) (69.0%) (0.0%) (100.0%) B and vice versa FDI magnitude ϪϪ0.4752 0.0551 Ϫ1.2222*** 0.0801*** 45 990 Ϫ4% FDI partner multiplied by FDI/GDP for each (22.4%) (66.2%) (0.0%) (100.0%) country Trade partner ϪϪ0.0121* 0.0626 Ϫ1.2510*** 0.0923*** 49 1,176 Ϫ8% Share of total imports and exports in country (4.2%) (68.0%) (0.0%) (100.0%) A originating in B and vice versa Trade magnitude ϪϪ7.6401** 0.0540 Ϫ1.2458*** 0.0947*** 49 1,176 Ϫ10% Pairwise total trade divided by GDP of each (1.6%) (68.0%) (0.0%) (100.0%) country Geographic distance ϩ 0.1240*** 0.0594 Ϫ1.2000*** 0.1669*** 49 1,176 29% Distance in kilometers between the capitals of (100.0%) (70.0%) (0.0%) (100.0%) the two countries Common language ϪϪ0.6032** 0.1105 Ϫ1.0612*** 0.1030*** 49 1,176 Ϫ14% Equals 1 if the two countries have a language (1.8%) (81.8%) (0.0%) (100.0%) in common Each row represents a separate univariate dyadic regression utilizing one of eight proxies for pairwise economic integration as the explanatory variable. All regressions control for pairwise differences in log GDP per capita and for common origin of legal systems. The dependent variable in all regressions is the distance in corporate governance between two countries, which is calculated as the sum of differences along thirteen dimensions of creditor and shareholder rights with data from La Porta et al. (1998). The sample is all possible pairs of countries for which we have governance and pairwise economic integration data. The expected sign in the second column is based on the hypothesis that more economic integration leads to a smaller “distance” in governance. Parentheses contain percentile distributions of the actual coefficient within coefficients under the null hypothesis generated from a QAP simulation with 500 draws. A percentile less than 5% or greater than 95%, for example, indicates a coefficient that is significant at the 10% level. Magnitude, in the columns so labeled, is calculated as the coefficient on the independent variable multiplied by that variable’s standard deviation and then divided by the standard deviation of the dependent variable. *Significant at 10% level; **significant at 5% level; ***significant at 1% level.

B1. Dyadic Regressions Separating U.S. and Non-U.S. Pairs

Common- Independent- Independent- Difference- Origin-of- Variable Variable in log- Legal- Magnitude— Expected Coeff.— Coeff.—non- GDP System No. of No. of Magnitude non-U.S. Independent Variable Sign U.S. pair U.S. Pair Coeff. Coeff. R-Squared Countries Pairs —U.S. Pair Pair

Wage correlation ϪϪ0.0072 Ϫ1.3085** Ϫ0.0260 Ϫ1.3956*** 0.1183*** 40 780 Ϫ0% Ϫ20% (50.6%) (1.6%) (46.0%) (0.0%) (100.0%) Capital market correlation Ϫ 0.3182 Ϫ0.2502 0.1357 Ϫ1.2878*** 0.0911*** 45 990 Ϫ2% Ϫ2% (62.4%) (37.0%) (83.8%) (0.0%) (100.0%) FDI partner Ϫ 0.4037 Ϫ4.0855*** 0.0377 Ϫ1.2726*** 0.1038*** 47 1,081 1% Ϫ12% (61.0%) (0.2%) (61.2%) (0.0%) (100.0%) FDI magnitude Ϫ 0.7065 Ϫ1.8873*** 0.0447 Ϫ1.2098*** 0.0900*** 45 990 4% Ϫ10% (77.4%) (0.4%) (63.4%) (0.0%) (100.0%) Trade partner ϪϪ0.0001 Ϫ0.0268** 0.0297 Ϫ1.2199*** 0.1017*** 49 1,176 Ϫ0% Ϫ13% (47.4%) (0.6%) (61.2%) (0.0%) (100.0%) Trade magnitude Ϫ 2.2442 Ϫ11.2781*** 0.0424 Ϫ1.2431*** 0.1011*** 49 1,176 2% Ϫ13% (62.8%) (0.0%) (64.0%) (0.0%) (100.0%) Geographic distance ϩ 0.1433* 0.1233*** 0.0574 Ϫ1.2048*** 0.1673*** 49 1,176 13% 29% (95.4%) (100.0%) (68.4%) (0.0%) (100.0%) Common language Ϫ 0.3259 Ϫ0.6783*** 0.0975 Ϫ1.0571*** 0.1092*** 49 1,176 3% Ϫ16% (70.4%) (0.4%) (80.4%) (0.0%) (100.0%) Each row represents a dyadic regression where the independent variable from the baseline regressions in panel A has been replaced by two independent variables, which separate the effects between U.S. pairs and non-U.S. pairs. The ﬁrst independent variable equals the underlying proxy for integration when the pair contains the United States, and 0 otherwise. The second independent variable equals the proxy for integration when the pair does not contain the United States, and 0 otherwise.

B2. Dyadic Regressions Separating All Pairs and Non-U.S. Pairs (Stronger Test to Reject U.S. Convergence)

Independent- Independent- Difference- Common- Variable Variable in-log Origin-of- Magnitude— Expected Coeff.— Coeff.—Non- GDP Legal-System No. of No. of Magnitude— non-U.S. Independent Variable Sign All Pairs U.S. Pairs Coeff. Coeff. R-Squared Countries Pairs All Pairs Pairs

Wage correlation ϪϪ0.0072 Ϫ1.3013 Ϫ0.0260 Ϫ1.3956*** 0.1183*** 40 780 0% Ϫ19% (49.6%) (11.2%) (44.8%) (0.0%) (100.0%) Capital market correlation Ϫ 0.3182 Ϫ0.5684 0.1357 Ϫ1.2878*** 0.0911*** 45 990 3% Ϫ5% (63.4%) (32.6%) (83.0%) (0.0%) (100.0%) FDI partner Ϫ 0.4037 Ϫ4.4893* 0.0377 Ϫ1.2726*** 0.1038*** 47 1,081 2% Ϫ13% (62.0%) (4.8%) (62.6%) (0.0%) (100.0%) FDI magnitude Ϫ 0.7065 Ϫ2.5938** 0.0447 Ϫ1.2098*** 0.0900*** 45 990 5% Ϫ13% (76.6%) (1.8%) (64.2%) (0.0%) (100.0%) Trade partner ϪϪ0.0001 Ϫ0.0267* 0.0297 Ϫ1.2199*** 0.1017*** 49 1,176 0% Ϫ13% (53.0%) (4.2%) (61.0%) (0.0%) (100.0%) Trade magnitude Ϫ 2.2442 Ϫ13.5223 0.0424 Ϫ1.2431*** 0.1011*** 49 1,176 3% Ϫ15% (64.6%) (6.4%) (63.0%) (0.0%) (100.0%) Geographic distance ϩ 0.1433* Ϫ0.0200 0.0574 Ϫ1.2048*** 0.1673*** 49 1,176 33% Ϫ5% (95.2%) (35.2%) (71.4%) (0.0%) (100.0%) Common language Ϫ 0.3259 Ϫ1.0042 0.0975 Ϫ1.0571*** 0.1092*** 49 1,176 8% Ϫ23% (67.8%) (15.6%) (80.4%) (0.0%) (100.0%) Each row represents a dyadic regression where the independent variable from the baseline regressions in panel A has been replaced by two independent variables, which contrast the effects in all pairs and non-U.S. pairs. The first independent variable equals the underlying proxy for integration. The second independent variable equals the proxy for integration when the pair does not contain the United States, and equals 0 otherwise. A statistically significant coefficient for this second variable indicates that there is a statistically significant difference between the effect of this variable for U.S. and non-U.S. pairs.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 78 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 2.—(CONTINUED)

C. Dyadic Regressions Testing for Regionalization

Independent- Independent- Common- Variable Variable Origin-of- Coeff.— Coeff.— Difference- Legal- Magnitude Expected between within in-log-GDP System No. of No. of between Magnitude Independent Variable Sign regions regions Coeff. Coeff. R-Squared Countries Pairs regions within regions

Wage correlation ϪϪ0.7200 Ϫ3.0419*** Ϫ0.1430 Ϫ1.3116*** 0.1538*** 40 780 Ϫ12% Ϫ49% (9.6%) (0.0%) (17.0%) (0.0%) (100.0%) Capital market correlation Ϫ 1.6371* Ϫ1.3864* 0.1522 Ϫ1.2610*** 0.1454*** 45 990 13% Ϫ11% (95.6%) (3.4%) (85.0%) (0.0%) (100.0%) FDI partner ϪϪ0.2251 Ϫ4.0747*** 0.0188 Ϫ1.2997*** 0.1003*** 47 1,081 Ϫ1% Ϫ17% (48.6%) (0.2%) (56.8%) (0.0%) (100.0%) FDI magnitude Ϫ 0.1467 Ϫ1.6022** 0.0419 Ϫ1.2311*** 0.0843*** 45 990 1% Ϫ11% (62.0%) (1.8%) (64.0%) (0.0%) (100.0%) Trade partner Ϫ 0.0067 Ϫ0.0195*** 0.0278 Ϫ1.2393*** 0.1004*** 49 1,176 3% Ϫ8% (65.4%) (0.2%) (56.2%) (0.0%) (100.0%) Trade magnitude Ϫ 4.0344 Ϫ9.8360*** 0.0417 Ϫ1.2512*** 0.1000*** 49 1,176 3% Ϫ6% (67.8%) (0.0%) (66.0%) (0.0%) (100.0%) Geographic distance ϩ 0.1248*** 0.1334 0.0603 Ϫ1.2005*** 0.1669*** 49 1,176 25% 26% (100.0%) (94.0%) (69.2%) (0.0%) (100.0%) Common language ϪϪ0.4972* Ϫ0.9134*** 0.0896 Ϫ1.0492*** 0.1059*** 49 1,176 Ϫ12% Ϫ21% (5.0%) (0.2%) (73.8%) (0.0%) (100.0%) Each row represents a dyadic regression where the independent variable from the baseline regressions in panel A has been replaced by two independent variables, which separate the effects of integration between regions and within regions. The ﬁrst independent variable equals the underlying proxy for integration when the observation is for two countries located in the same region, and equals 0 otherwise. The second independent variable equals the proxy for integration when the observation is for two countries located in different regions, and equals 0 otherwise. Regions are North America, South America, Europe, Asia, Africa, Middle East, Australia.

D. Dyadic Regressions by Pair Type: Developed-Developed, Developing-Developing, Developed-Developing Pairs

Independent- Independent- Independent- Variable Variable Variable Common- Coeff.— Coeff.— Coeff.— Origin- Developed- Developing- Developed- Difference- of-Legal- Magnitude— Magnitude— Magnitude— Expected Developed Developing Developing in-Log/GDP System No. of No. of Developed- Developing- Developed- Independent Variable Sign pair pair pair Coeff. Coeff. R-Squared Countries Pairs Developed Developing Developing

Wage correlation ϪϪ1.8627** Ϫ1.3117* Ϫ0.9767* Ϫ0.1746 Ϫ1.4108*** 0.1204*** 39 741 Ϫ19% Ϫ19% Ϫ16% (1.2%) (4.4%) (4.6%) (18.4%) (0.0%) (100.0%) Capital market correlation ϪϪ1.0380 2.8380** 1.0447 0.0043 Ϫ1.4559*** 0.1292*** 44 946 Ϫ8% 19% 8% (16.4%) (99.0%) (86.6%) (53.0%) (0.0%) (100.0%) FDI partner ϪϪ4.8584** 5.5091 Ϫ3.6936** 0.0085 Ϫ1.2628*** 0.1042*** 46 1,035 Ϫ30% 5% Ϫ16% (0.6%) (80.0%) (1.4%) (54.8%) (0.0%) (100.0%) FDI magnitude ϪϪ1.6813** 15.3550 Ϫ2.6744** 0.0355 Ϫ1.1861*** 0.0897*** 44 946 Ϫ20% 8% Ϫ18% (1.6%) (92.4%) (1.6%) (61.8%) (0.0%) (100.0%) Trade partner ϪϪ0.0327*** Ϫ0.0020 Ϫ0.0211 Ϫ0.0098 Ϫ1.2474*** 0.1026*** 48 1,128 Ϫ30% Ϫ1% Ϫ13% (0.2%) (44.6%) (12.8%) (49.2%) (0.0%) (100.0%) Trade magnitude ϪϪ11.9237** Ϫ9.6582 Ϫ10.3290* 0.0185 Ϫ1.2331*** 0.0987*** 48 1,128 Ϫ21% Ϫ5% Ϫ12% (1.6%) (21.8%) (4.0%) (55.6%) (0.0%) (100.0%) Geographic distance ϩ 0.1128*** 0.1393*** 0.1230*** 0.0395 Ϫ1.1916*** 0.1714*** 48 1,128 31% 31% 24% (99.6%) (100.0%) (100.0%) (61.0%) (0.0%) (100.0%) Common language ϪϪ1.8627** Ϫ1.3117* Ϫ0.9767* Ϫ0.1746 Ϫ1.4108*** 0.1204*** 39 741 Ϫ44% Ϫ28% Ϫ23% (1.2%) (4.4%) (4.6%) (18.4%) (0.0%) (100.0%) Each row represents a dyadic regression where the independent variable from the baseline regressions in panel A has been replaced by three independent variables, which separate the effects of integration into three categories: developed-developed, developing-developing, and developed-developing. The ﬁrst independent variable equals the underlying proxy for integration when the observation is for two developed countries. The second independent variable equals the proxy for integration when the observation is for two developing countries, and equals 0 otherwise. The third independent variable equals the proxy for integration when the observation is for a developed and a developing country, and equals 0 otherwise.

E1. Robustness Check, Controlling for European Unobservables

Common- Indep.- Diff.-in Log- Origin-of- Both-Countries- Expected Variable GDP/cap. Legal-System European Dummy No. of No. of Independent Variable Sign Coeff. Coeff. Coeff. Coeff. R-Squared Countries Pairs Magnitude

Wage correlation ϪϪ1.0921* Ϫ0.1861 Ϫ1.4385*** Ϫ2.0954*** 0.1567*** 39 741 Ϫ15% (2.8%) (8.8%) (0.0%) (0.0%) (100.0%) Capital market correlation Ϫ 0.9123 0.0938 Ϫ1.3447*** Ϫ1.9077*** 0.1341*** 44 946 8% (81.8%) (72.4%) (0.0%) (0.0%) (100.0%) FDI partner ϪϪ2.4216 Ϫ0.0933 Ϫ1.3279*** Ϫ1.6696*** 0.1293*** 46 1,035 Ϫ7% (6.0%) (25.8%) (0.0%) (0.2%) (100.0%) FDI magnitude ϪϪ1.4557** Ϫ0.1164 Ϫ1.2511*** Ϫ1.7741*** 0.1225*** 44 946 Ϫ8% (1.8%) (21.4%) (0.0%) (0.0%) (100.0%) Trade partner ϪϪ0.0113 Ϫ0.0894 Ϫ1.3029*** Ϫ1.7339*** 0.1292*** 48 1,128 Ϫ6% (13.8%) (29.4%) (0.0%) (0.0%) (100.0%) Trade magnitude ϪϪ7.8452** Ϫ0.0992 Ϫ1.3004*** Ϫ1.7715*** 0.1339*** 48 1,128 Ϫ9% (1.8%) (25.6%) (0.0%) (0.2%) (100.0%) Geographic distance ϩ 0.1086*** Ϫ0.0257 Ϫ1.2120*** Ϫ1.0270** 0.1801*** 48 1,128 25% (100.0%) (47.0%) (0.0%) (1.2%) (100.0%) Common language ϪϪ0.7447*** Ϫ0.0621 Ϫ1.0674*** Ϫ2.0063*** 0.1527*** 48 1,128 Ϫ17% (0.4%) (35.4%) (0.0%) (0.0%) (100.0%) This panel shows the baseline regression with an additional control variable: a dummy variable that is equal to 1 when a dyad consists of two European countries and equal to 0 otherwise. U.S. pairs are excluded from the analysis.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 79

TABLE 2.—(CONTINUED)

E2. Robustness Check, Controlling for Regional Unobservables

Common- Indep.- Diff.-in Log- Origin-of- Both-Countries- Expected Variable GDP/cap. Legal-System in-Same-Region No. of No. of Independent Variable Sign Coeff. Coeff. Coeff. Dummy Coeff. R-Squared Countries Pairs Magnitude

Wage correlation ϪϪ1.2595** Ϫ0.1784 Ϫ1.3040*** Ϫ1.1012*** 0.1557*** 39 741 Ϫ18% (1.8%) (15.0%) (0.0%) (0.0%) (100.0%) Capital market correlation Ϫ 0.8591 0.1076 Ϫ1.2428*** Ϫ1.2613*** 0.1441*** 44 946 8% (82.2%) (73.0%) (0.0%) (0.0%) (100.0%) FDI partner ϪϪ2.5658* Ϫ0.0908 Ϫ1.2125*** Ϫ0.9448*** 0.1316*** 46 1,035 Ϫ7% (3.8%) (27.4%) (0.0%) (0.0%) (100.0%) FDI magnitude ϪϪ1.2766 Ϫ0.1019 Ϫ1.1322*** Ϫ1.0255*** 0.1221*** 44 946 Ϫ7% (6.2%) (27.6%) (0.0%) (0.0%) (100.0%) Trade partner ϪϪ0.0069 Ϫ0.0761 Ϫ1.1993*** Ϫ1.0089*** 0.1266*** 48 1,128 Ϫ3% (28.2%) (29.0%) (0.0%) (0.0%) (100.0%) Trade magnitude ϪϪ4.3135 Ϫ0.0762 Ϫ1.2001*** Ϫ0.9928*** 0.1277*** 48 1,128 Ϫ5% (17.6%) (30.0%) (0.0%) (0.0%) (100.0%) Geographic distance ϩ 0.1194*** 0.0334 Ϫ1.1606*** Ϫ0.1476 0.1700*** 48 1,128 28% (100.0%) (64.8%) (0.0%) (31.4%) (100.0%) Common language ϪϪ0.5812** Ϫ0.0427 Ϫ1.0038*** Ϫ1.0122*** 0.1414*** 48 1,128 Ϫ13% (1.8%) (44.6%) (0.0%) (0.0%) (100.0%) This panel shows the baseline regression with an additional control variable: a dummy variable that is equal to 1 when a dyad consists of two countries located in the same region and equal to 0 otherwise. Regions are North America, South America, Europe, Asia, Africa, Middle East, Australia. U.S. pairs are excluded from the analysis.

at the 10% level). The magnitudes of the estimated coeffi- efficients for the effects of integration across regions. As cients for U.S. pairs for all variables except for geographic shown in panel C, the results for within-region integration distance are relatively small and in the range of 0 to 4%. In are much stronger, in terms of both the significance of the contrast, for the case of the non-U.S. pairs, all the variables coefficients and the magnitudes of the predicted effects. are of the expected sign, and all except for capital market This regionalization of governance is consistent with the correlation are significant, with five of the variables signif- previous result that convergence is not driven by integration icant at the 1% level. For all variables, the magnitude of the with the United States. predicted effect for non-U.S. pairs is equal to or larger than We also analyze the results according to level of devel- the magnitude for the U.S. pairs. Because the U.S. pair opment of the country pairs. Common wisdom is that coefficients are estimated using a small number of observa- convergence will occur because governance practices tions with essentially nondyadic data, these coefficients are spread from developed countries with good governance to not strictly comparable to non-U.S. coefficients. developed and developing countries with poor governance. We repeat the analysis using a more robust test (panel Convergence between two developing countries is expected B2). We generate two independent variables. The first al- to be less likely. We divide the countries in our sample ways equals the proxy for integration. The second equals the evenly into developed and developing countries according proxy for integration for non-U.S. pairs and equals 0 for to GDP per capita and then group the pairs into three U.S. pairs. A statistically significant coefficient for the categories: developed-developed, developing-developing, second independent variable would indicate that there is a and developed-developing. Panel D shows the coefficients statistically significant difference in the effect of that vari- for these regressions. The effects of integration are stron- able between U.S. and non-U.S. pairs. We find three of these gest, in terms of significance and magnitude of the coeffi- second variables to be significant, two at the 10% level and cients, in the developed-developed pairs and weakest in the one at the 5% level. The overall results provide strong developing-developing, as expected. evidence that U.S. pairs are not alone driving the findings of A potential problem with the preceding analysis is the the baseline regressions, and weak evidence that there is no existence of unobservable variables that affect governance similarity to U.S. standards. and are correlated with our measures of economic integra- If U.S. pairs are not driving the results, then to what tion. Though we cannot reject all such unobservables, we do model of governance are countries converging? It appears test for some of the more obvious ones to verify our results. that there is no single model. Rather, countries tend to have For example, it is possible that there exists some unobserv- governance similar to those of regional partners. We dem- able factor that induces correlation among European coun- onstrate this pattern using a methodology similar to that tries even after controlling for legal origin. To test whether described above for the United States. We generate two this unobservable might be responsible for all the results, we independent variables, one for integration within regions repeat the baseline regressions, adding as a control variable and the other for integration across regions, such that the a dummy variable that is equal to 1 when both countries are former variable is 0 when the pairs are not in the same European and equal to 0 otherwise. Inclusion of this dummy region and the latter is 0 when the pairs are in the same control variable should decrease the coefficients on our region. We are thus able to estimate coefficients for the various proxies for economic integration, either because the effects of integration within regions and then separate co- dummy variable indeed is a control for some unobservable

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 80 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 3.—FACTOR ANALYSIS ON PAIRWISE EXPLANATORY VARIABLES Factor Analysis Factor 2: Factor 1: Difference in Factor 3: Variable Globalization Development Common Origin Uniqueness Eigenvalue 2.96 0.87 0.51 Rotated factor loadings: Common legal system 0.09124 0.01408 0.49227 0.74915 Difference in log GNP/capita Ϫ0.04588 0.6387 0.0313 0.58897 Wage correlation 0.17705 Ϫ0.06871 Ϫ0.13838 0.94478 Capital market correlation 0.2691 Ϫ0.67576 Ϫ0.03513 0.46969 FDI partner 0.79849 0.00914 Ϫ0.02247 0.36182 FDI magnitude 0.78307 0.00016 Ϫ0.01538 0.38656 Trade partner 0.84138 Ϫ0.17286 0.09574 0.25303 Trade magnitude 0.79839 Ϫ0.16707 0.09359 0.3259 Geographic distance Ϫ0.26254 0.16065 Ϫ0.1402 0.88561 Common language 0.14661 0.10342 0.52292 0.69436

Regression on Factors (4) (1) (2) (3) Magnitude All Available Excludes Magnitude, excluding countries U.S. all countries U.S. Factor 1: Globalization Ϫ0.2701** (1.5%) Ϫ0.4913*** (0.0%) Ϫ13% Ϫ23% Factor 2: Difference in development 0.1583 (80.8%) 0.0062 (53.1%) 6% 0% Factor 3: Common origin Ϫ0.9393*** (0.0%) Ϫ0.9393*** (0.0%) Ϫ30% Ϫ30% Number of countries 0.1108*** (100.0%) 0.1469*** (100.0%) Number of pairs 35 34 R-Squared 595 561 A factor analysis of the eight pairwise proxies for economic integration identifies three factors that are easily interpretable as globalization, difference in development, and common origin. These factors are then used in a multivariate dyadic regression. The dependent variable in all regressions is the distance in corporate governance between two countries, which is calculated as the sum of differences along thirteen dimensions of creditor and shareholder rights with data from La Porta et al. (1998). The sample is all possible pairs of countries for which we have governance and all eight pairwise integration measures. Parentheses contain percentile distributions of the actual coefficient within coefficients under the null hypothesis generated from a QAP simulation with 500 draws. A percentile less than 5% or greater than 95%, for example, indicates a coefficient that is significant at the 10% level. Magnitude, in the columns so labeled, is calculated as the coefficient on the independent variable multiplied by that variable’s standard deviation and then divided by the standard deviation of the dependent variable. *Significant at 10% level; **significant at 5% level; ***significant at 1% level.

or because the dummy variable is itself a proxy for integra- these proxies measure globalization of a particular kind tion. We exclude U.S. pairs from the analysis, in view of the of market—for example, a capital market—this is a tough prior results. As shown in panel E1, this dummy variable is case to make in practice. For example, FDI might be a of the expected sign and statistically significant at the 5% or proxy for any of capital, product, or labor market integra- 1% levels in all regressions. Despite this control, seven of tion. If correlation of capital markets is due to common the eight proxies for economic integration are of the ex- economic factors affecting the performance of firms in both pected sign, and five of these are statistically significant. countries, then this explanatory variable is a proxy for Though similarity in governance is present among Euro- economic integration in general. pean countries, this analysis shows that it is not the only Table 3 shows a factor analysis on the ten explanatory case of similarity correlated with integration in our data. variables, retaining three factors. Factor 1 represents glob- European countries are not the only ones whose governance alization, with heavy loadings on FDI, trade, and geograph- might be affected by some regional unobservable factor. We ical distance. Factor 2 represents differences in develop- generate a more general dummy control variable, which is ment, with loadings on differences in log GNP per capita equal to 1 if both countries in the dyad are in the same and capital market comovement. The presence of the latter region and equal to 0 otherwise. As shown in panel E2, this variable in this factor suggests that the correlation within regional dummy variable is always of the expected sign and emerging markets or within developed markets is higher statistically significant at the 1% level for seven of the eight than the correlation between a developed country and a regressions. Despite the inclusion of the regional dummy developing country [see Morck, Yeung, and Yu (1999) for a variable, seven of the eight proxies for integration are still of more detailed discussion of this pattern]. Factor 3 represents the expected sign, and four are statistically significant. path dependence and contains common origin and common Without time series data, we are unable to control for all language. Whereas common language was intended to be a possible unobservables, but the checks do assuage these proxy for the ease of mobility of labor between two coun- concerns. tries, it is easy to see how it would be highly correlated with common origin of the legal system and included in that Factor Analysis: Our proxies for globalization are often factor. The wage correlation variable does not receive heavy highly correlated. Further, though it may seem that some of weightings in any of the three retained factors. The regres-

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 81

TABLE 4.—DYADIC ANALYSIS OF CLSA (DE FACTO)DATA Expected Independent- Difference-in Common-Origin-of- No. of No. of Independent Variable Sign Variable Coeff. Log-GDP Coeff. Legal-System Coeff. R-Squared Countries Pairs Wage correlation ϪϪ0.0620 (49.2%) 0.7314 (81.2%) Ϫ1.2697 (6.4%) 0.0589 (59.2%) 21 210 Capital market correlation Ϫ 2.4122 (77.8%) 0.7242 (86.6%) Ϫ0.9956 (7.8%) 0.0539 (63.0%) 23 253 FDI partner ϪϪ4.3005 (13.4%) 0.7556 (90.2%) Ϫ0.3887 (25.4%) 0.0430 (70.8%) 23 253 FDI magnitude Ϫ 6.2462 (89.4%) 0.9343 (93.0%) Ϫ0.4008 (25.2%) 0.0753 (90.8%) 22 231 Trade partner ϪϪ0.0153 (26.0%) 0.7235 (87.8%) Ϫ0.7849 (12.0%) 0.0464 (71.2%) 24 276 Trade magnitude Ϫ 4.0250 (72.6%) 0.6578 (85.6%) Ϫ0.8583 (10.8%) 0.0453 (72.6%) 24 276 Geographic distance ϩ 0.0068 (56.0%) 0.6926 (86.8%) Ϫ0.8204 (10.2%) 0.0429 (64.0%) 24 276 Common language ϪϪ1.2015* (4.6%) 0.8764 (90.4%) Ϫ0.3103 (31.6%) 0.0804 (88.2%) 24 276 Each row represents a separate univariate dyadic regression utilizing one of eight proxies for pairwise economic integration as the explanatory variable. All regressions control for pairwise differences in log GDP/capita and for common origin of legal systems. The dependent variable in all regressions is the distance in corporate governance between two countries, which is calculated as the sum of differences along seven dimensions of corporate governance practices using CLSA firm-level data aggregated to the country level. The sample is all possible pairs of countries for which we have governance and pairwise economic integration data. The expected sign in the second column is based on the hypothesis that more economic integration leads to a smaller distance in governance. Parentheses contain percentile distributions of the actual coefficient within coefficients under the null hypothesis generated from a QAP simulation with 500 draws. A percentile less than 5% or greater than 95%, for example, indicates a coefficient that is significant at the 10% level. Magnitude, in the columns so labeled, is calculated as the coefficient on the independent variable multiplied by that variable’s standard deviation and then divided by the standard deviation of the dependent variable. *Significant at 10% level; **significant at 5% level; ***significant at 1% level.

sion of legal-rules differences on these three factors finds all between de facto and de jure results. Here, complementary coefficients of the expected sign, with common origin sig- institutions are weaker, political rent-seeking possibilities nificant at the 1% level and globalization significant at the are greater, and enforcement of the rules is inadequate. As 5% level. We also reran this analysis focusing on dyads that hard as it is to change the legal regime defining the rules of do not involve the United States. The common-origin vari- corporate governance, modifying actual practices to con- able is as strong a predictor of convergence as before. form to these legal reforms under the political and economic Additionally, globalization is strengthened as a predictor, in circumstances found in most developing countries is an magnitude and significance (1% level). This buttresses our even greater challenge. Our results in table 5 confirm the finding of de jure similarity in governance among pairs of hypothesis that there is a large gap between de facto and de economically linked countries (especially when the United jure similarity in developing countries. For this sample of States is excluded). countries, none of the coefficients in the de facto regressions are significant, and six of the eight coefficients are of the C. Comparison with de Facto Analysis wrong sign. In contrast, in the de jure analysis all but one of the coefficients are of the right sign, and five of the coeffi- In contrast to the de jure results, we find almost no cients are significant. Thus, our results still hold in the evidence of de facto similarity between economically linked apples-to-apples sample of countries. pairs of countries. As shown in table 4, of the eight mea- We summarize the results of the country-level analysis sures that we test, the coefficients on three are of the wrong thus. We find evidence that economic interdependence and sign. Of the remaining five, only common language is similarity in de jure corporate governance practices between significant, and merely at the 10% level. Common origin of pairs of countries are correlated, especially intraregionally, the legal system is of the expected sign in all regressions, and especially among more developed countries. Estimation but, surprisingly, is not significant, indicating that this mea- using regressors derived from a factor analysis confirms that sure of path dependence is a poor predictor of governance globalization and similarity in corporate governance are enforcement. Difference in level of development is also of correlated even after controlling for common origins. There the expected sign in all regressions but not significant. is no such correlation between economic interdependence We are unable to implement the type of detailed analysis and de facto corporate governance practices, however. An for the CLSA data that we used for the LLSV data, because interpretation of these results is that economically integrated the CLSA data cover only developing countries. We at- countries have rules on books regarding corporate gover- tempted a factor analysis of the economic integration prox- nance that are more similar than they would be if the ies for just the developing countries, but the three factors countries were not as integrated; however, this similarity of were not easily interpretable. rules on books does not transfer into similarity in practice. The de facto and de jure results are not strictly compa- This interpretation is consistent with the findings in transi- rable, because they are derived from different samples of tion economies (Pistor et al., 2000) and regarding Mexican countries. Most importantly, the de jure results are based on firms (Siegel, 2004) as well as regarding the importance of a sample that includes both developed and developing the process by which laws are implemented (Berkowitz, countries, whereas the de facto sample only contains devel- Pistor, & Richards, 2003). oping countries. To address this problem, we perform an apples-to-apples comparison by running the regressions on V. Robustness Checks just the nineteen countries for which we have both de facto and de jure data. These nineteen countries are all developing A problem with aggregating corporate governance ratings countries, where we expect to see the biggest difference at the country level is that we ignore convergence along

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 82 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 5.—DYADIC ANALYSIS (APPLES VERSUS APPLES COUNTRIES) Independent Expected Independent- Difference-in-Log- Common-Origin-of- No. of No. of Variable Sign Variable Coeff. GDP Coeff. Legal-System Coeff. R-Squared Countries Pairs A. CLSA (de Facto) Governance Indicators Wage correlation ϪϪ0.0328 (41.6%) 1.0441 (87.4%) 0.2554 (67.6%) 0.0836 (71.8%) 16 120 Capital market correlation Ϫ 1.1823 (65.2%) 1.1437** (97.6%) 0.4062 (85.8%) 0.1217 (85.2%) 18 153 FDI partner Ϫ 3.1219 (69.6%) 1.0744 (94.8%) 0.4530 (91.0%) 0.1143* (95.2%) 19 171 FDI magnitude Ϫ 8.3139 (94.0%) 1.3584** (98.8%) 0.4530 (84.0%) 0.1973** (99.2%) 18 153 Trade partner Ϫ 0.0208 (77.0%) 1.0703* (96.6%) 0.3844 (81.0%) 0.1211* (95.4%) 19 171 Trade magnitude Ϫ 13.1201 (93.4%) 1.0480* (96.4%) 0.3552 (81.8%) 0.1604* (97.0%) 19 171 Geographic distance ϩ 0.0434 (83.0%) 1.1555** (97.8%) 0.5828 (92.0%) 0.1264* (97.0%) 19 171 Common language Ϫ 0.0693 (46.8%) 1.0841* (96.8%) 0.4359 (85.8%) 0.1130 (92.4%) 19 171 B. LLSV (de Jure) Rights Differences Wage correlation Ϫ 0.1348 (58.4%) Ϫ0.4304 (6.6%) Ϫ1.7124** (0.6%) 0.1575** (99.0%) 16 120 Capital market correlation ϪϪ0.5082 (33.4%) Ϫ0.5086** (1.4%) Ϫ1.8450*** (0.0%) 0.1852** (100.0%) 18 153 FDI partner ϪϪ0.2321 (48.0%) Ϫ0.3933 (5.4%) Ϫ1.7771*** (0.2%) 0.1668*** (99.8%) 19 171 FDI magnitude ϪϪ7.6380* (4.2%) Ϫ0.2818 (15.6%) Ϫ1.4137*** (0.2%) 0.1498*** (99.8%) 18 153 Trade partner ϪϪ0.0446** (0.8%) Ϫ0.3509 (7.4%) Ϫ1.6340*** (0.2%) 0.2047*** (100.0%) 19 171 Trade magnitude ϪϪ13.0249** (1.0%) Ϫ0.3522 (7.4%) Ϫ1.6818*** (0.2%) 0.2135*** (100.0%) 19 171 Geographic distance ϩ 0.1180*** (100.0%) Ϫ0.2179 (21.0%) Ϫ1.4187*** (0.4%) 0.2668*** (100.0%) 19 171 Common language ϪϪ0.7876* (3.4%) Ϫ0.3201 (9.0%) Ϫ1.6041*** (0.0%) 0.1980*** (100.0%) 19 171 These panels repeat the regressions of table 2 and table 4, but restrict the sample in both panels to a common set of countries.

industry and firm lines. In as much as most national laws governance scores, while country explains 17% to 57%. The apply to all firms and industries, de jure convergence is contrast in explanatory power is largest for the De´minor rat- inherently a country-level phenomenon. On the other hand, ings, where country explains 29 times more variance than de facto convergence can occur at the country, industry, or sector. The relatively small effect of industry casts doubt on firm level. We show, using CLSA and other firm-level data industry-level explanations for the absence of de facto similar- sets, that de facto convergence is occurring at neither the ity in corporate governance. industry level nor the firm level. We also investigate a For another robustness check, we tested separately the number of political and economic factors at the country argument that more global industries have higher overall level that may explain why convergence is happening only governance scores. This is effectively a test for similarity to between certain pairs of countries. U.S. corporate governance standards. For each industry, we compiled several measures of how global an industry is by A. Similarity in Governance at the Industry Level aggregating data up from the firm level. Two of these measures were based on emerging-market data (namely, the It has been argued that a different type of governance is propensity to list abroad and to export), and two other appropriate for different types of industries. In skilled-labor- measures were based on U.S. and western European multi- intensive industries, it may be economically optimal to give national activity. The advantage of the latter two variables is labor greater control in the governance of the firm in order to that they are exogenous because they are not related to the induce labor to invest in acquiring firm-specific skills. Alter- dependent variable, the corporate governance score of firms natively, corporate governance convergence may occur only in in emerging markets. For all four measures, we were able to global industries, which would be more greatly affected by reject the hypothesis that more global industries have gov- capital market, product market, and labor market integration. ernance closer to the U.S. system of governance. We test for the role of industry in determining governance using the data from CLSA as well as three other sources of B. Similarity in Governance at the Firm Level firm-level governance ratings as described in appendix D: De´minor, Proxinvest, and Fortune. Data from De´minor and Our sector categories in the industry-level analysis are Proxinvest, two independent consulting firms, provide gov- quite broad, resulting in significant heterogeneity between ernance ratings on the top 300 firms in Europe. Fortune firms in the same sector. For this reason, we conducted magazine has rated firms worldwide in its World’s Most firm-level analyses as well. Mirroring some of our industry- Admired Companies Index. level analyses, we test whether firms that are more exposed If optimal governance structure differed by industry, then we to global markets have higher governance scores, after would expect a regression of firm level governance indicators controlling for country and industry effects. We use three on industry and country dummy variables to find a significant indicators of exposure to global capital markets, one indi- role for industry. Appendix D displays the ANOVA decomposi- cator of exposure to global product markets, and one indi- tion of the four firm-level governance ratings (CLSA, De´mi- cator of exposure to global labor markets. These variables nor, Proxinvest, and Fortune) into country and sector effects. are: institutional investor holdings, U.S. common stock or Industry explains only 2% to 13% of the variance in corporate ADR listings, the use of global intermediaries, export in-

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 83

tensity, and proportion of foreign board members. Appendix find the greatest differences from country peers for those firms E explains these variables in more detail. where the top shareholder is a U.S. firm or a U.K. firm but a As with some of our industry-level analyses, the firm- regression analysis found no statistically significant effect.22 level analyses effectively look for correlations between exposure to global markets and de facto similarity to U.S. C. Additional Explanatory Variables in the Country-Level corporate governance standards. We find little evidence of a Analyses relationship between firm-level exposure to globalization and corporate governance. Regression results from the One explanation for why convergence may not happen CLSA data are presented in appendix F. Although signifi- despite competitive pressures is that vested interests capture cant, the coefficient for U.S. listing is small.20 The small political processes and prevent the adoption of needed reforms. size of the effect is consistent with Siegel’s (2004) finding We test this hypothesis first using data on the level of democ- that ADR listings did not constrain Mexican companies racy in each country, hypothesizing that reform is more likely with ADRs from expropriating minority shareholders. The in democratic countries. We use three proxies for democracy institutional-ownership coefficient is significant at the 10% (democracy index, political constraints index, and Gini coeffi- level in some of the regressions, and the coefficient is also cient) and test whether integration has a larger effect when the small. The global-intermediary variables, representing pair of countries are more democratic. The empirical results whether the firm has had any connection with the big five here are inconclusive. In as much as there is substantial accounting firms or major Western investment banks, are disagreement in the academic literature regarding whether not significant. The export-intensity variable is not signifi- reform is more feasible under democracy or autocracy, the lack cant when we control for fixed country effects. The coeffi- of a clear relationship in this case is not surprising. We repeat cient for foreign management and directors is significant at the analysis using a clearer proxy for the power of vested the 1% level, but the effect it represents is small. We have interests—union membership. We expect unions to oppose repeated this analysis for European countries using the corporate governance reform intended to strengthen share- De´minor data set and obtained similar results. holder rights or creditor rights; for this reason, countries with a We have ignored the issue of causation so far, implicitly larger percentage of workers in unions should be less likely to assuming that variables such as institutional investor hold- reform. Surprisingly, the empirical analysis finds the opposite ings are exogenous and that higher institutional holdings relationship. Similarity in governance due to integration was would put more pressure on firms to change their gover- significantly greater in pairs of countries with high union nance practices. Instead, it is possible that a firm’s corporate membership. In a separate analysis, we test whether reform is governance affects the size of institutional investor hold- more likely in countries that have recently experienced a ings. Knowing that they will be unable to affect corporate currency crisis or received international aid. We find no sig- governance of the firms they invest in, institutional inves- nificant evidence in support of either relationship. tors may decide to only invest in firms that already have good corporate governance. Because the coefficient on VI. Conclusion institutional investor holdings was small and not highly Theorists have debated the extent of convergence in corpo- significant, however, we are able to reject either direction of rate governance for some time now. Opinions span the spec- causation as a significant force. Similarly, it may be that trum from believers that global competition will force (and has firms with better corporate governance tend to list in the already forced) such convergence (Hansmann & Kraakman, United States rather than that listing in the United States 2001), through more measured views that functional conver- causes firms to improve corporate governance; but the gence rather than formal convergence will occur (Gilson, coefficient here was small as well. As long as both direc- 2001), to the idea that path dependence precludes the possibil- tions of causation lead to a coefficient of the same sign, our ity of meaningful convergence (Bebchuk & Roe, 1999). Others negative results still hold.21 have drawn attention to the importance of distinguishing rules- We also test for convergence along a more direct channel, on-the-book convergence from effective convergence, which that of pressure by controlling shareholders. We expected to we refer to in this paper as de jure and de facto convergence respectively (Pistor et al., 2000). 20 Because a couple of the questions in the transparency section of the CLSA questionnaire related to requirements that would be imposed by the SEC or by exchange listing rules, the improvement in governance is 22 By searching Bloomberg and the Internet, we are able to identify the probably accounted for by compliance with these regulations rather than origin of the top shareholder for 235 of the 482 firms in the CLSA sample. indicative of a voluntary effort by firms to improve governance in order to Although most firms in our CLSA sample are locally owned, there are 13 attract U.S. investors. firms where the top shareholder was a U.K. or U.S. firm. Of these 13 firms, 21 If, on the other hand, poorly governed firms rather than well-governed 10 had a corporate governance rating greater than their country peers, and firms were more likely to list in the United States because they needed to the remaining 3 had a lower rating. However, a fixed-country-effects improve investors’ perceptions of their governance practices, then we regression of weighted corporate governance ratings on a dummy variable might obtain a negative result when in fact there were opposing forces representing whether or not the top shareholder was a U.S. or U.K. firm along both directions of causation. yielded a positive but insignificant coefficient.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 84 THE REVIEW OF ECONOMICS AND STATISTICS

Yet these debates about the role of global competition in dards but that there is little evidence that these standards fostering convergence are generally not informed by sys- have been implemented. tematic empirical inquiry. We conjecture that this is partly because of the difficulty of defining and operationalizing the notion of convergence. We take a small step in the direction REFERENCES of remedying this lack of literature. In particular, we amass Acemoglu, Daron, James A. Robinson, and Simon Johnson, “The Colonial Origins of Comparative Development: An Empirical Investiga- several new data sets on de facto corporate governance tion,” American Economic Review 91 (2001), 1369–1401. practices of firms in several developed and developing Anderson, J., “A Theoretical Foundation for the Gravity Equation,” markets. We contrast the de facto results from these data American Economic Review 69:1 (1979), 106–116. Aoki, Masahiko, “The Contingent Governance of Teams: Analysis of sets with a new use of the now ubiquitous La Porta et al. Institutional Complementarity,” International Economic Review data on shareholder and creditor rights, which we interpret 35:3 (1994), 657–676. Baker, H. Kent, and Richard B. Edelman, “The Signaling Effects of as indicators of de jure similarity, as Pistor et al. have AMEX Listing,” Akron Business and Economic Review 22:2 justified convincingly in their analysis of transition econo- (1991), 124–138. mies. We try to be careful about what we mean by “glob- Barro, Robert J., and Xavier Sala-i-Martin, “Convergence,” The Journal of Political Economy 100:2 (1992), 223–251. alization.” In other words, we distinguish—with useful Baumol, William J., “Productivity Growth, Convergence, and Welfare: empirical implications, as it turns out—globalization as What the Long-Run Data Show,” The American Economic Review exposure to the U.S. system currently in ascendancy around 76:5 (1986), 1072–1085. Bebchuk, Lucian Arye, Alma Cohen, and Allen Ferrell, “Does the Evi- the world, from globalization as economic interdependence dence Favor State Competition in Corporate Law?” California Law between any pair of countries. Review 90 (2002), 1775–1821. Bebchuk, Lucian Arye, and Mark J. Roe, “A Theory of Path Dependence Our results, based on cross-sectional data, test the hy- in Corporate Ownership and Governance,” Stanford Law Review pothesis that similarity in corporate governance between 52:1 (1999), 127–170. two countries is correlated with economic integration be- Ben-David, Dan, “Equalizing Exchange: Trade Liberalization and Income Convergence,” The Quarterly Journal of Economics 108:3 (1993), tween those countries. This test for similarity is also a test 653–679. for corporate governance convergence under certain condi- Berglof, Erik, and Enrico Perotti, “The Governance Structure of the tions. First, complete convergence to one optimal standard Japanese Financial Keiretsu,” Journal of Financial Economics 36:2 (1994), 259–284. must have not already occurred, for if such convergence had Berglof, Erik, and Ernst-Ludwig von Thadden, “Short-Term versus Long- occurred, we would actually find no correlation between Term Interests: Capital Structure with Multiple Investors,” The Quarterly Journal of Economics 109:4 (1994), 1055–1084. similarity in governance and integration. The heterogeneity Berkowitz, D., K. Pistor, and J. F. Richard, “Economic Development, in governance across countries demonstrates that this con- Legality, and the Transplant Effect,” European Economic Review dition is met. Second, common origin of the legal system 47:1 (2003), 165–195. Bhide, Amar, “The Hidden Costs of Stock Market Liquidity,” Journal of must be a sufficient control for the path dependence. If Financial Economics 34:1 (1993), 31–51. common origin does indeed sufficiently control for the point Black, Bernard S., “Agents Watching Agents: The Promise of Institutional at which governance laws and practices started hundreds of Investor Voice,” UCLA Law Review 39 (1992a), 811–893. “The Value of Institutional Invest Monitoring: The Empirical years ago, then any common deviation from this starting Evidence,” UCLA Law Review 39 (1992b), 895–939. point by economically interlinked countries is indicative of “Does Corporate Governance Matter? A Crude Test Using Rus- sian Data,” University of Pennsylvania Law Review 149 (2001), a convergence process. 2131–2150. Our conclusions challenge conventional wisdom. We find Blass, Asher, and Yishay Yafey, “Vagabond Shoes Longing to Stray: Why robust evidence of de jure similarity among economically Foreign Firms List in the United States,” Journal of Banking & Finance 25 (2001), 555–572. interdependent countries. Interestingly, this is not driven by Branson, Douglas M., “The Very Uncertain Prospect of ‘Global’ Conver- similarity to U.S. standards. Rather, pairs of economically gence in Corporate Governance,” Cornell International Law Jour- interdependent countries appear to adopt common corporate nal 34:2 (2001), 321–362. Bratton, William W., and Joseph A. McCahery, “Comparative Corporate governance standards, especially if the pair of countries in Governance and the Theory of the Firm: The Case Against Global question are in the same geographic region and are rela- Cross Reference,” Columbia Journal of Transnational Law 38 (1999), 213–298. tively developed economies. In contrast to these de jure Buxbaum, Richard M., “Institutional Owners and Corporate Managers: A results, we find virtually no evidence of de facto similarity Comparative Perspective,” Brooklyn Law Review 57 (1991), 1–53. in corporate governance in a battery of estimations at the Coffee, John C., Jr., “The Future as History: The Prospects for Global Convergence in Corporate Governance and Its Implications,” country, industry, and firm levels. This is consistent with Northwestern University Law Review 93 (1999), 641. either the proposition that complementarities result in dif- Coles, Jerilyn W., Victoria B. McWilliams, and Nilanjan Sen, “An Exam- ination of the Relationship of Governance Mechanisms to Perfor- ferent national systems appropriately having different cor- mance,” Journal of Management 27:1 (2001), 23–50. porate governance systems, or the proposition that global- Crane, Dwight B., Zvi Bodie, and Andre F. Perold (Eds.), The Global ization is not strong enough to overcome local vested Financial System: A Functional Perspective (Boston: Harvard Business School Press, 1995). interests. We conclude that globalization may have induced Credit Lyonnais Securities Asia, “Saints & Sinners: Who’s Got Religion?” the adoption of some common corporate governance stan- Hong Kong: CLSA (2001).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 85

Demsetz, Harold, and Kenneth Lehn, “The Structure of Corporate Own- nous Stock Price Movements?” Journal of Financial Economics ership: Causes and Consequences,” The Journal of Political Econ- 58:1–2 (1999), 215–260. omy 93:6 (1985), 1155–1177. Mukand, Sharun, and Dani Rodrik, “In Search of the Holy Grail: Policy Diamond, Jared, Guns, Germs, and Steel: The Fates of Human Societies Convergence, Experimentation, and Economic Performance,” JFK (New York: W.W. Norton & Co., 1999). School of Government, Harvard University, working paper (2002). Easterbrook, Frank H., and Daniel R. Fischel, “The Incorporation Debate North, Douglass C., “Economic Performance through Time,” The Amer- and State Antitakeover Statutes” The Economic Structure of Cor- ican Economic Review 84:3 (1994), 359–368. porate Law (Cambridge, MA Harvard University Press, 1991). OECD, “Corporate Governance in OECD Member Countries: Recent Eichengreen, Barry, Toward a New International Financial Architecture: Developments and Trends,” OECD report (2000), http://www. A Practical Post-Asia Agenda (Washington, DC: Institute for oecd.org/daf/corporate-affairs/governance/. International Economics, 1999). Olson, Mancur, The Logic of Collective Action (Cambridge: Harvard Evenett, Simon J., and Wolfgang Keller, “On Theories Explaining the University Press, 1971). Success of the Gravity Equation,” The Journal of Political Econ- Opler, Tim C., and Jonathan Sokobin, “Does Coordinated Institutional omy 110:2 (2002), 281–316. Shareholder Activism Work? An Analysis of the Council of Insti- Foley, C. Fritz, “The Effects of Having an American Parent: An Analysis tutional Investors,” Ohio State University research paper (1998). of the Growth of U.S. Multinational Affiliates,” Harvard Business Pistor, Katharina, Martin Raiser, and Stanislaw Gelfer, “Law and Finance School working paper (2002). in Transition Economies,” Economics of Transition 8:2 (2000), Freeman, Richard B., and Remco H. Oostendorp, “The Occupational 325–368. Wages around the World Data File,” International Labour Review Porter, M. E., “Capital Disadvantage: America’s Falling Capital Invest- 140:4 (2001), 379–401. ment System,” Harvard Business Review 70:5 (1992), 65–82. Gillan, Stuart L., and Laura T. Starks, “Corporate Governance Proposals Pound, John, “The Rise of the Political Model of Corporate Governance and Shareholder Activism: The Role of Institutional Investors,” and Corporate Control,” New York University Law Review 68 Journal of Financial Economics 57 (2000), 275–305. (1993), 1003–1071. Gilson, Ronald J., “Globalization of Corporate Governance: Convergence Rajan, Raghuram G., and Luigi Zingales, “The Great Reversals: The of Form or Function,” American Journal of Comparative Corpo- Politics of Financial Development in the 20th Century,” Journal of rate Law 49, 329–357 (2001). Financial Economics 69:1 (2003), 5–50. Guilleń, Mauro F., “Corporate Governance and Globalization: Is There Rodrik, Dani, “Institutions for High Quality Growth: What They Are and Convergence across Countries?” Advances in Comparative Inter- How to Acquire Them,” Studies in Comparative International national Management 13 (2000), 175–204. Development 35:3 (2000). Hansmann, Henry, and Reinier Kraakman, “The End of History for Roe, Mark J., “Chaos and Evolution in Law and Economics,” Harvard Corporate Law,” Georgetown Law Journal 89 (2001), 439–468. Law Review 109 (1996), 641–668. The Genius of American Corporate Law Kaplan, Steve, “Top Executive Rewards and Firm Performance: A Com- Romano, Roberta, (Washington, DC: AEI Press, 1993). parison of Japan and the U.S.,” Journal of Political Economy, 102 Romer, Paul M., “Increasing Returns and Long-Run Growth,” The Jour- (1994), 510. nal of Political Economy 94:5 (1986), 1002–1037. Kauffmann, S., The Origins of Order (New York: Oxford University Samuelson, Paul, “International Trade and the Equalisation of Factor Press, 1993). Prices,” Economic Journal 58(1948), 163–184. Khanna, Tarun, and Krishna Palepu, “Globalization and Convergence in Saudagaran, Shahrokh, “An Empirical Study of Selected Factors Influ- Corporate Governance: Evidence from Infosys and the Indian encing the Decision to List on Foreign Stock Exchanges,” Journal Software Industry,” Journal of International Business Studies 35:6 of International Business Studies 19:1 (1988), 101–127. (2004), 484–507. Schmidt, Reinhard H., and Gerald Spindler, “Path Dependence, Corporate Khanna, Tarun, Krishna Palepu, and Suraj Srinivasan, 2004. “Disclosure Governance and Complementarity,” International Finance 5:3 Practices of Foreign Companies Interacting with U.S. Markets,” (2002), 311–333. Journal of Accounting Research 42:2 (2004), 475–508. Shleifer, Andrei, and Robert W. Vishny, “A Survey of Corporate Gover- Kissane, Mary E., “Global Gadflies: Applications and Implications of nance,” Journal of Finance 52:2 (1997), 737–783. U.S.-Style Corporate Governance Abroad,” New York Law Journal Siegel, Jordan I. “Can Foreign Firms Bond Themselves Effectively by of International & Comparative Law 17 (1997), 621–675. Submitting to U.S. Law?” Journal of Financial Economics 75(2): Krackhardt, David, “Predicting with Networks: Nonparametric Multiple 319–359. Regression Analysis of Dyadic Data,” Social Networks 10 (1988), Solow, Robert M., “A Contribution to the Theory of Economic Growth,” 359–381. The Quarterly Journal of Economics 70:1 (1956), 65–94. La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert Thomsen, Steen, and Torben Pedersen, “Nationality and Ownership Struc- W. Vishny, “Law and Finance,” The Journal of Political Economy tures: The one hundred Largest Companies in Six European Na- 106:6 (1998), 1113–1155. tions,” Management International Review 36 (1996), 149–166. La Porta, Rafael, Florencio, Lopez-de-Silanes, and Andrei Shleifer, 1999a. Tirole, Jean, “Corporate Governance,” Econometrica 69:1 (2001), 1–35. Corporate Ownership around the World. Journal of Finance 54(2): Williamson, Jeffrey G., “Globalization, Convergence, and History,” The 471–517. Journal of Economic History 56:2 (1996), 277–306. Leamer, E., and J. Levinsohn, “International Trade Theory: The Evi- World Bank, “Corporate Governance: A Framework for Implementation,” dence,” in G. M. Grossman and K. Rogoff (Eds.), Handbook of World Bank paper (2001), http://www.worldbank.org. International Economics 3 (Amsterdam: Elsevier, 1995). Mankiw, Gregory N., “The Growth of Nations,” Brookings Papers on Economic Activity 1995:1 (1995), 275–310. Merton, Robert C., “Presidential Address: A Simple Model of Capital APPENDIX A Market Equilibrium with Incomplete Information,” Journal of Finance 42 (1987), 483–510. Constructing the Scandals Index Merton, Robert K., Social Theory and Social Structure (New York: Free Press, 1968). We created an index of egregious corporate governance violations by Meyer, John W., J. Boli, G. Thomas, and F. Ramirez, “World Society and combing the public media over the last two and a half years for instances the Nation State,” American Journal of Sociology 103 (1997), of such violations. Because of the time required to manually scan news- 144–181. paper articles for scandals for each company, we focused our search on the Milgrom, Paul, and John Roberts, “The Economics of Modern Manufac- 80 Indian companies in the CLSA database. We chose India because it is turing,” American Economic Review 80 (1990), 511–528. the country for which CLSA provides data on the most firms and there is Morck, Randall, Bernard Yeung, and Wayne Yu, “The Information Con- a large range of corporate governance practices in India, with CLSA tent of Stock Markets: Why Do Emerging Markets Have Synchro- ratings ranging from 37 to 93 on a 100-point scale. We recorded for each

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 86 THE REVIEW OF ECONOMICS AND STATISTICS

firm whether it had any of the following three types of scandals: share- Exchange were selected for inclusion. These results confirm the claim that holder expropriation, tax evasion, and price fixing. CLSA selected large firms as well as firms that should be of interest to We searched the India Business Insight database within Lexis-Nexis for U.S. institutional investors. articles about each of these 80 firms from the last two and a half years. According to Lexis-Nexis, this database has over 70,000 articles covering 30,000 companies, with coverage beginning on January 11, 1999. We manually scanned headlines of articles about our 80 firms for relevant stories and APPENDIX C then read any such articles to determine if there indeed was a scandal and what type of scandal it was. For each firm we recorded the total number of QAP Estimation articles found about that firm, which will serve as a control for the media attention devoted to that firm. Three types of scandals were recorded: An OLS regression with dyadic variables produces unbiased coefficients, but because the pairwise observations are not independent, OLS 1. Shareholder Expropriation equals 1 if any shareholders have ac- standard errors can be too small, leading to possible misinterpretations of cused management of taking actions such as selling off assets, spurious correlations. We use the quadratic assignment procedure (QAP), siphoning off funds, or insider trading, where management bene- a technique developed for the analysis of dyads in networks (see, for fited at the expense of shareholders or majority shareholders ben- example, Krackhardt, 1988), to test the statistical significance of the efited at the expense of minority shareholders. (Eight cases found.) regression coefficients. Our pairwise-dependent and -independent vari- 2. Tax Evasion equals 1 if the company has been raided or investigated ables each form the lower triangle of a matrix where the rows and columns for tax evasion or if a warrant has been issued in response to a are countries. The QAP method involves scrambling the data so that the failure to pay taxes. Does not include instances where certain each dependent variable is randomly reassigned to a different observation. deductions were ruled inappropriate by the courts or by the income The OLS regression is run on the scrambled data, and new coefficient tax department. (Three cases found.) estimates are generated. Repeating this process 500 times generates a 3. Price Fixing equals 1 if the company has been accused by the sampling distribution of the coefficient estimate under the null hypothesis. government of operating in a cartel to raise the prices of its In order to preserve the interdependences in the rows and columns in products. (Four cases found, but all were related to one instance of generating the null distribution, the scrambling is accomplished by ran- price-fixing by four firms in our sample.) domly permuting rows and columns of the matrix; the rows of the dependent variable are rearranged in a new, randomly assigned order, and Our search also found other types of scandals, but many of these occurred the columns are rearranged in the same order. To assess statisticalsignifi- only once and were not as clear-cut cases of poor corporate governance. cance, we compare the original coefficient estimate with the null distri- Probit regressions found that the shareholder expropriation index was bution. If this coefficient falls in the upper or lower 2.5% of the distribu- correlated with the CLSA data but the tax evasion and price fixing indices tion, then this coefficient is significant at the 5% level. Recent studies were not. using QAP have carried out permutation and estimation steps from several hundred to over 1,000 times. Varying the number of repetitions from 500 has very little effect on our results. APPENDIX B

Selection Issues in the CLSA Data APPENDIX D The CLSA sample contains 80 Indian firms with scores ranging from 37 to 93 on a 100-point scale. To examine the selection process that ANOVA Decomposition of Country and Industry Effects resulted in these firms showing up in the CLSA data, we select all Indian firms listed in Bloomberg with more than $20 million in market capital- Table D1 shows the percentage of total variance in corporate gover- ization, producing a list of 301 firms. This pool contains 73 of the CLSA nance ratings explained by country and sector dummy variables using firms, with 5 left out because no market capitalization information was partial sum-of-squares analysis of variance. The corporate governance available and 2 left out because market capitalization was less than $20 ratings we use come from CLSA, Fortune, De´minor, and Proxinvest. million according to Bloomberg. Table B1 shows a probit regression In April 2001, CLSA released a large study of corporate governance testing the determinants of firm selection by CLSA. The coefficient of log entitled, “Saints and Sinners: Who’s Got Religion?” which reviewed assets, a proxy for firm size, is positive and significant at the 1% level, corporate governance at the firm level in 25 emerging markets. A firm’s indicating that large firms are more likely to be included. U.S. listing and corporate governance score is based on responses to 57 yes no questions percentage of outstanding shares held by institutional investors are also about various aspects of governance. The CLSA questionnaire is divided significant at the 1% level. Exports/sales is positive, indicating that firms into seven sections covering fiscal discipline, accounting transparency and that sell to foreign markets are more likely to be selected, but the disclosure, board independence, board accountability, responsibility, eq- coefficient is not significant. Having a big five auditor increases the uitable treatment of shareholders, and social awareness. The score for each probability of selection and is significant at the 10% level. Not shown in of the seven sections is simply the percentage of questions receiving a yes the table is the fact that all Indian firms listed on the New York Stock answer in that section. The overall corporate governance score for the firm is a weighted average of the firm’s scores in these seven sections. Fortune magazine publishes an annual global list of most admired TABLE B1—PROBIT ANALYSIS OF THE DETERMINANTS companies. To construct this list, firms from the Global 500 list of largest OF CLSA FIRM SELECTION IN INDIA companies are separated by sector, and sectors dominated by firms from one country are removed. Every firm is sent a survey asking it to rate all Variable (1) of the other firms in its sector on a scale of 1 to 10 on eight characteristics: quality of management; quality of products and services; innovativeness; Log assets 0.396*** (0.074) long-term investment value; financial soundness; ability to attract, de- U.S.-listed (common stock or ADR) 0.624*** (0.207) velop, and retail talent; community responsibility; and use of corporate Institutional ownership (% of outstanding shares) 4.79*** (1.72) assets. A firm’s “admired” score is simply the average of the scores over Exports/Sales 0.521 (0.381) these eight characteristics. The variable Sales/Max Sales is included in the Big five accounting firm 0.535* (0.279) Fortune regressions to control for the portion of the admired score Ϫ Constant 4.90*** (0.754) attributable to being the largest firm in the sector. Number of observations 208 De´minor, an independent consultancy firm in Brussels, has rated 250 of Log likelihood 103.9 the largest 300 European firms on over 70 governance indicators, which 2 Pseudo R 0.2140 are grouped into four governance categories: rights and duties of share- *10% significance; **5% significance; ***1% significance. holders, absence of takeover defenses, disclosure, and board structure.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 87

Because the board structure rating is not publicly available, the table uses performance has been addressed by academics theoretically and studied the average of the first three. empirically, with mixed results, by a number of researchers (Buxbaum, ProxInvest is a French proxy voting service, which calculated 1991; Black, 1992a, 1992b; Pound, 1993; Kissane, 1997; Opler & Soko- corporate governance ratings for 250 top European companies. It bin, 1998; Gillan & Starks, 2000). publicly discloses ratings for only the top-rated 100 of these companies. ProxInvest ratings are matched with De´minor subsector categories for this 2. U.S. Listing table. We track whether each company’s securities are traded in the U.S., either as American Depositary Receipts (ADR) or as underlying stock. Firms list in foreign capital markets to gain access to a different investor TABLE D1 base, to aid product visibility, and to improve the predictability of an employee stock ownership plan (Merton, 1987; Saudagaran, 1988; Baker Fortune & Edelman, 1991). CLSA Global Weighted Most De´minor 3. Capital Market Intermediaries: Auditors and Investment Banks Governance Admired Average ProxInvest Index Rating Rating Rating We code two binary variables for the regression analysis: whether the Country 37% 17% 57% 39% firm has a big five auditor and whether it recently placed an offering Sector 6% 13% 2% 12% through a global investment bank. It is argued that the use of global rather Sales/Max Sales in sector — 11% — — than local intermediaries (such as big five accounting firms, top invest- Number of countries 16 13 13 9 ment banks, and consulting firms) will facilitate convergence in corporate Number of sectors 18 26 10 10 governance (Hansmann and Kraakman, 2001). Number of firms 470 294 250 88 4. Export Intensity

Our proxy for a firm’s exposure to global product market competition is its export-to-sales ratio. We assume firms that do not report geographic segmentation data in their annual report have no exports, although that is APPENDIX E not necessarily the case. Because the lack of disclosure of geographic segmentation can be correlated with poor governance, assuming nondis- Description of Explanatory Variables at the Firm Level closing firms have zero exports can bias the coefficient upward.

The firm-level empirical analysis will use three indicators of capital 5. Foreign management and directors market globalization as well as two indicators of product market and labor market globalization: institutional investor holdings, U.S. common stock Although we do not know the percentage of all employees that are or ADR listings, the use of global intermediaries, export intensity, and foreign, we can estimate this figure for senior management and directors, proportion of foreign board members. All indicators are derived from as it is this class of employees that should have the strongest relationship Bloomberg firm data. Here we briefly explain each variable and cite the with corporate governance. A sampling of the management and board of prior academic literature that has each one. directors was taken using the first page of the Bloomberg management screen, which lists up to eighteen executives or directors. These names 1. Institutional Investors were manually scanned for instances of foreign first and last names. Although the list was scanned for any person whose name was clearly not Bloomberg compiles data on the percentage of shares outstanding held from the region in which the firm is located, almost all of the foreign by institutional owners, mostly U.S. and U.K. mutual funds, based on persons found were American. Firms in South Africa and the Philippines documents that these organizations are required to file with the U.S. were not included in this sample, because of the difficulty of distinguish- government. The role of shareholder activism in affecting governance and ing foreign names for these countries.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 88 THE REVIEW OF ECONOMICS AND STATISTICS

APPENDIX F

FIRM-LEVEL REGRESSION ANALYSIS (1) (2) (3) (4) (5) (6) (7) (8) Capital market Trades in U.S. as 4.32*** 3.60*** 2.37 3.41** an ADR or as (1.33) (1.24) (1.56) (1.45) common stock Institutional 4.56 21.39* 23.53* 13.41 ownership (11.48) (11.05) (13.73) (12.06) (fraction of outstanding shares) Auditor is KPMG, Ϫ1.34 Ernst, Price, (2.36) Andersen, or Deloitte Recent offering by 2.65 major U.S. (2.17) investment bank

Product market Exports/Sales 15.32*** 4.45 Ϫ0.30 (3.07) (2.87) (3.69)

Labor market Fraction of 20.88*** 11.83*** 15.11*** management or (4.52) (4.15) (4.24) directors with foreign ﬁrst and last name

Control vars. Log of assets in 0.12 Ϫ0.88* Ϫ0.97 0.22 Ϫ0.67 Ϫ0.90** Ϫ1.49*** Ϫ1.82*** millions of USD (0.39) (0.50) (0.62) (0.38) (0.46) (0.41) (0.53) (0.57) Holdings of largest 0.79 shareholder (3.75) (fraction of outstanding shares)

Dummy vars. Number of country 18 15 18 17 16 dummy vars. (not including constant) Number of sector 17 17 17 17 17 dummy vars. (not including constant) Constant 52.39*** 62.92*** 59.59*** 54.09*** 64.24*** 63.32*** 66.11*** 48.10*** (3.02) (9.83) (12.88) (2.88) (9.69) (3.08) (10.16) (12.24)

R-squared 0.0294 0.443 0.5326 0.0560 0.4696 0.641 0.4526 0.4648 Adj. R-squared 0.0222 0.3847 0.4574 0.0515 0.4184 0.0587 0.3908 0.3961 Observations 407 402 275 426 421 351 346 335 The table explores the effects of various types of global integration on a firm’s governance score for firms in 24 emerging markets. The dependent variable is an overall governance score computed by averaging each firm’s scores, as computed by CLSA, along seven dimensions of governance: fiscal discipline, accounting transparency and disclosure, board independence, board accountability, responsibility, equitable treatment of shareholders, and social awareness. In some cases not all seven dimensions are used, because one of the dimensions is based in part on an explanatory variable. One of the 10 questions in the equitable-treatment-of-shareholders section is whether foreign portfolio managers own more than 20% of the stock; this question is related to the institutional-ownership independent variable. Regressions that estimate the effect of this independent variable exclude this section of the survey in the dependent variable. Similarly, a question in the accountability section asks about the presence of foreign directors, which is related to our labor market variable. Standard errors shown in parentheses. *Significant at 10% level; **significant at 5% level; ***significant at 1% level.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 GLOBALIZATION AND SIMILARITIES IN CORPORATE GOVERNANCE 89

APPENDIX G—PAIRWISE ECONOMIC INTEGRATION Argentina* Australia Austria Belgium Brazil* Canada Chile* China* Colombia* Czech* Denmark Ecuador Egypt Finland France Germany Greece* Hong Kong* Hungary* India* Indonesia* Ireland Israel Italy

Argentina* 15 11 37 90 28 90 37 44 11 31 59 54 15 59 62 22 47 9 21 28 17 23 65 Australia 15 22 43 26 48 42 72 13 22 24 17 49 46 56 74 10 87 9 56 83 28 27 62 Austria 11 22 80 23 42 18 34 12 96 59 8 38 61 84 99 54 60 97 20 33 50 39 92 Belgium 37 43 80 56 57 46 59 32 84 80 48 56 80 99 99 75 81 79 78 49 86 95 93 Brazil* 90 26 23 56 49 86 44 65 36 30 74 52 28 48 71 33 58 54 16 33 27 20 63 Canada 28 48 42 57 49 65 64 66 32 23 44 26 34 59 68 22 77 21 31 52 47 37 57 Chile* 90 42 18 46 86 65 68 63 6 32 78 5 41 71 78 12 75 4 29 36 22 20 73 China* 37 72 34 59 44 64 68 19 52 41 7 56 43 59 79 38 99 47 43 78 38 35 64 Colombia* 44 13 12 32 65 66 63 19 15 15 90 2 26 51 75 5 23 2 13 9 19 19 64 Czech* 11 22 96 84 36 32 6 52 15 66 17 31 70 91 99 42 52 91 29 21 58 32 96 Denmark 31 24 59 80 30 23 32 41 15 66 11 34 84 83 96 52 65 47 30 35 69 35 81 Ecuador 59 17 8 48 74 44 78 7 90 17 11 1 9 38 72 5 44 1 5 12 14 32 64 Egypt 54 49 38 56 52 26 5 56 2 31 34 1 45 83 88 48 59 16 41 32 41 56 87 Finland 15 46 61 80 28 34 41 43 26 70 84 9 45 81 94 45 69 62 22 41 64 42 79 France 59 56 84 99 48 59 71 59 51 91 83 38 83 81 97 87 88 85 45 64 89 82 95 Germany 62 74 99 99 71 68 78 79 75 99 96 72 88 94 97 93 92 99 70 82 94 91 96 Greece* 22 10 54 75 33 22 12 38 5 42 52 5 48 45 87 93 43 37 22 26 43 47 95 Hong Kong* 47 87 60 81 58 77 75 99 23 52 65 44 59 69 88 92 43 43 82 89 60 79 91 Hungary* 9 9 97 79 54 21 4 47 2 91 47 1 16 62 85 99 37 43 19 19 35 33 93 India* 21 56 20 78 16 31 29 43 13 29 30 5 41 22 45 70 22 82 19 57 25 49 50 Indonesia* 28 83 33 49 33 52 36 78 9 21 35 12 32 41 64 82 26 89 19 57 21 1 64 Ireland 17 28 50 86 27 47 22 38 19 58 69 14 41 64 89 94 43 60 35 25 21 46 83 Israel 23 27 39 95 20 37 20 35 19 32 35 32 56 42 82 91 47 79 33 49 1 46 89 Italy 65 62 92 93 63 57 73 64 64 96 81 64 87 79 95 96 95 91 93 50 64 83 89 Japan 38 88 65 81 56 80 81 93 75 69 62 78 69 77 65 78 69 99 66 66 94 90 80 57 Jordan 45 71 49 79 83 56 2 82 4 35 47 2 68 28 95 96 40 61 20 80 72 40 30 95 Kenya 42 42 26 76 32 50 1 71 1 17 35 0 28 36 85 86 25 63 10 84 57 25 51 80 Malaysia* 49 88 32 62 48 60 38 85 12 30 38 5 53 45 84 92 18 96 23 76 87 79 16 79 Me´xico* 48 16 13 40 54 80 82 26 67 8 11 78 8 10 51 74 7 41 4 13 21 28 10 49 Netherlands 45 37 82 99 63 52 54 68 41 85 87 41 63 82 95 99 83 85 80 46 61 88 79 90 New Zealand 8 97 23 39 21 53 27 58 5 7 27 10 19 34 54 77 10 72 3 30 49 24 14 63 Nigeria 33 6 39 53 73 29 40 60 3 14 14 0 3 11 83 86 13 66 6 72 40 35 17 72 Norway 14 24 50 78 24 72 35 44 14 38 92 3 14 86 80 93 23 51 20 16 19 76 36 76 Pakistan* 26 44 18 40 32 27 14 74 5 17 33 2 24 38 56 76 16 68 16 42 47 19 0 66 Peru* 60 16 14 29 66 50 69 46 81 11 21 56 1 21 40 61 3 29 12 18 12 13 19 52 Philippines* 20 77 16 44 50 46 34 74 3 6 20 2 7 34 65 84 5 92 5 42 75 53 42 58 Poland* 18 10 74 78 40 27 10 51 12 92 76 18 26 70 85 98 38 45 76 20 26 46 20 92 Portugal 22 15 57 83 57 25 20 29 20 22 60 12 18 53 94 96 34 36 19 25 18 52 33 92 Russia* 21 7 72 70 35 28 24 80 16 98 67 28 70 93 75 92 72 63 97 53 33 92 36 84 Singapore* 29 93 51 71 57 68 45 94 13 37 48 8 41 66 94 96 33 99 30 82 93 89 60 89 South Africa* 40 62 35 65 46 36 31 63 15 30 24 14 16 31 72 91 15 76 8 44 30 47 46 77 South Korea* 39 84 36 46 56 70 78 92 44 67 45 61 63 47 63 84 62 98 55 60 87 64 58 68 Spain 63 29 67 87 49 33 73 47 57 74 61 74 56 61 96 94 75 65 64 28 50 74 73 91 Sri Lanka 17 71 21 78 44 48 31 82 24 12 29 5 27 22 51 82 9 90 8 93 71 25 53 67 Sweden 27 54 72 90 33 39 55 50 38 74 96 17 60 96 86 96 57 74 70 33 53 79 57 80 Switzerland 26 47 88 85 44 52 37 43 43 81 69 40 56 65 95 98 58 85 73 42 36 68 84 94 Taiwan* 37 79 44 48 40 67 62 90 34 62 39 37 41 46 65 83 32 99 36 42 81 67 58 62 Thailand* 28 75 32 69 45 52 31 76 8 28 39 4 19 51 69 87 22 91 17 52 75 53 53 71 Turkey* 24 34 60 73 33 30 6 42 5 44 35 6 61 36 84 95 55 55 39 30 24 34 54 90 United Kingdom 40 79 79 97 50 76 61 59 55 89 88 54 75 91 94 96 85 93 81 72 70 99 91 89 United States 86 93 79 94 87 99 98 87 97 82 77 98 95 87 82 85 71 98 77 73 87 97 97 77 Uruguay 94 13 7 23 95 32 59 43 15 9 22 23 1 10 68 63 7 64 5 25 9 11 25 78 Venezuela 60 3 9 31 69 70 62 3 89 2 15 76 1 12 50 67 1 312849466 Zimbabwe 26 62 28 47 49 25 4 45 1 25 35 0 6 43 75 83 9 58 10 55 18 24 52 73 Table displays relative levels of economic integration on a scale from 0 to 99 based on the trade magnitude proxy. Trade magnitude for a pair of countries is defined as the total trade between these two countries divided by the GDP of the first country plus the total trade between these two countries divided by the GDP of the second country. The reported value of 90 between Argentina and Brazil, for example, indicates that the trade magnitude score for this pair of countries was higher than 90% of trade magnitude scores in our sample. In other words, Argentina and Brazil are in the top 10% in bilateral economic integration. Table also identifies countries for which we had CLSA data on corporate governance practices, with an asterisk (*).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021 90 THE REVIEW OF ECONOMICS AND STATISTICS

APPENDIX G—CONTINUED ´xico* Japan Jordan Kenya Malaysia* Me Netherlands New Zealand Nigeria Norway Pakistan* Peru* Philippines* Poland* Portugal Russia* Singapore* South Africa* South Korea* Spain Sri Lanka Sweden Switzerland Taiwan* Thailand* Turkey* United Kingdom United States Uruguay Venezuela Zimbabwe

38454249484583314266020182221294039631727263728244086946026 88 71 42 88 16 37 97 6 24 44 16 77 10 15 7 93 62 84 29 71 54 47 79 75 34 79 93 13 3 62 65 49 26 32 13 82 23 39 50 18 14 16 74 57 72 51 35 36 67 21 72 88 44 32 60 79 79 7 9 28 81 79 76 62 40 99 39 53 78 40 29 44 78 83 70 71 65 46 87 78 90 85 48 69 73 97 94 23 31 47 56 83 32 48 54 63 21 73 24 32 66 50 40 57 35 57 46 56 49 44 33 44 40 45 33 50 87 95 69 49 80 56 50 60 80 52 53 29 72 27 50 46 27 25 28 68 36 70 33 48 39 52 67 52 30 76 99 32 70 25 81 2 1 38 82 54 27 40 35 14 69 34 10 20 24 45 31 78 73 31 55 37 62 31 6 61 98 59 62 4 93 82 71 85 26 68 58 60 44 74 46 74 51 29 80 94 63 92 47 82 50 43 90 76 42 59 87 43 3 45 75 4 1 12 67 41 5 3 14 5 81 3 12 20 16 13 15 44 57 24 38 43 34 8 5 55 97 15 89 1 6935173088571438171169222983730677412748162284489829 2 25 62 47 35 38 11 87 27 14 92 33 21 20 76 60 67 48 24 45 61 29 96 69 39 39 35 88 77 22 15 35 782057841100325621812288146174517403746549823760 69 68 28 53 8 63 19 3 14 24 1 7 26 18 70 41 16 63 56 27 60 56 41 19 61 75 95 1 1 6 77 28 36 45 10 82 34 11 86 38 21 34 70 53 93 66 31 47 61 22 96 65 46 51 36 91 87 10 12 43 65 95 85 84 51 95 54 83 80 56 40 65 85 94 75 94 72 63 96 51 86 95 65 69 84 94 82 68 50 75 78 96 86 92 74 99 77 86 93 76 61 84 98 96 92 96 91 84 94 82 96 98 83 87 95 96 85 63 67 83 69402518783101323163 5383472331562759575832225585717 1 9 99 61 63 96 41 85 72 66 51 68 29 92 45 36 63 99 76 98 65 90 74 85 99 91 55 93 98 64 31 58 662010234803 6201612576199730855648707336173981775 2 10 66 80 84 76 13 46 30 72 16 42 18 42 20 25 53 82 44 60 28 93 33 42 42 52 30 72 73 25 8 55 94 72 57 87 21 61 49 40 19 47 12 75 26 18 33 93 30 87 50 71 53 36 81 75 24 70 87 9 4 18 90 40 25 79 28 88 24 35 76 19 13 53 46 52 92 89 47 64 74 25 79 68 67 53 34 99 97 11 9 24 80305116107914173601942203336604658735357845853549197254 52 57 95 80 79 49 90 63 72 76 66 52 58 92 92 84 89 77 68 91 67 80 94 62 71 90 89 77 78 66 73 88 89 99 76 84 91 74 70 86 67 98 48 66 54 99 83 97 61 91 67 71 98 99 69 81 92 55 57 83 88 1078682510153912341374539218871060700758892974381 89103647838344603613135988735122455238541694811043 99 78 36 29 73 69 17 25 77 26 86 30 23 55 99 55 96 51 79 66 66 96 95 43 88 99 11 7 26 766 429 301811111161235201956125953312035482420429962712 84 82 78 73 30 40 77 87 55 31 60 81 84 82 85 64 62 86 50 92 86 73 67 74 97 93 31 41 61 91513691840 6161827414121172256326604939684719819282115 740 81711776 16112 8186495429537103648056138790493 7 70 15 34 25 11 87 16 16 17 10 27 61 59 64 58 28 42 59 12 97 55 44 32 24 91 80 5 14 17 86 39 46 77 11 55 18 11 17 4 20 9 25 39 61 24 65 30 68 38 50 46 43 34 73 86 4 2 10 67 1 0 26 61 31 27 2 10 4 14 8 4 16 15 17 57 64 3 27 35 37 21 6 40 94 23 75 0 98233862360418272014 177679521893115504088771970983 2 7 48 41 6 30 5 81 4 18 61 9 8 17 18 93 37 14 68 66 10 77 60 47 34 30 86 75 7 8 12 66 37 13 23 20 84 12 64 59 25 4 7 18 55 37 34 48 98 6 65 67 28 27 27 90 74 6 13 22 544513551982119643916679355 531177581276733363887380243 2 99 39 59 99 56 85 72 54 58 61 15 95 37 37 53 68 98 68 91 72 84 98 99 52 95 99 32 13 39 83 21 88 55 12 64 25 29 28 24 17 21 14 34 11 68 68 48 51 55 64 72 41 29 89 90 14 13 99 97 88 73 96 59 62 63 53 42 65 57 89 68 48 77 98 68 48 89 46 58 85 85 57 69 97 58 32 58 61 71 51 51 53 86 26 71 59 30 64 31 66 98 58 68 48 48 18 70 70 39 45 66 90 80 71 49 27 9102279315060012683151061291518918 50430763685844 1 9 67 60 45 66 20 92 49 36 97 38 27 50 77 65 76 72 55 46 70 50 76 53 54 58 94 83 31 29 51 71 70 52 66 35 86 39 48 55 50 35 40 60 67 73 84 64 58 70 43 76 55 65 67 89 91 37 25 52 98 0 38 96 48 73 68 0 44 46 37 88 47 28 33 98 72 85 39 0 53 55 88 44 69 97 36 29 51 99 75 54 95 24 67 47 56 32 43 21 77 34 27 63 99 41 85 45 76 54 65 88 23 77 95 7 4 15 69881643207419132434619302788522957663658674423 83868 6 9 81 92 94 88 42 97 81 87 91 73 40 70 86 90 73 95 89 69 90 85 94 89 69 77 83 93 59 54 89 92 97 81 99 99 93 92 90 80 86 94 98 75 74 80 99 90 97 80 84 83 91 97 95 86 93 90 98 81 5541116231849542337624321458714313736785990492 57 38 0 7 71 41 21 3 14 2 75 2 8 13 3 13 13 32 49 1 29 25 29 4 6 54 98 49 7 831432626115717100 71222239995827951525115989812 7

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest.2006.88.1.69 by guest on 30 September 2021