77637 Banking on Politics: When Former High-ranking Politicians Become Bank Directors ´as Braun and Claudio Raddatz Matı New data are presented for a large number of countries on how frequently former high-ranking politicians become bank directors. Politician-banker connections at this Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 level are relatively rare, but their frequency is robustly correlated with many important characteristics of banks and institutions. At the micro level, banks that are politically connected are larger and more pro�table than other banks, despite being less lever- aged and having less risk. At the country level, this connectedness is strongly nega- tively related to economic development. Controlling for this, the analysis �nds that the phenomenon is more prevalent where institutions are weaker and governments more powerful but less accountable. Bank regulation tends to be more pro-banker and the banking system less developed where connectedness is higher. A benign, public- interest view is hard to reconcile with these patterns. Banking sector development, institutions, political economy. JEL codes: G15, G21, P16 There is ample evidence that access to external �nancing is critical for the level and ef�ciency of investment, productivity, and economic growth at the �rm and the aggregate level. Yet �rms in different countries do not have the same access to �nance.1 This raises two important questions: Why do some countries lack a well developed �nancial system if it is so bene�cial? And how do �rms react to �nancial sector underdevelopment? A recent strand of �nancial devel- opment literature aims at answering both questions from a political economy standpoint. On the �rst question, the literature complements theories of �nancial devel- opment based on stable and largely predetermined factors (such as the origins of a country’s legal system, pattern of colonization, religion and culture, and social capital endowment) with a role for dynamic political economy Matı´as Braun (matias.braun@uai.c) is director of strategy and partner at IM Trust and professor of economics and �nance at Universidad Iban ˜ ez. Claudio Raddatz (corresponding author; craddatz@ worldbank.org) is a senior economist at the World Bank in the Macroeconomics and Growth Unit of the Development Economics Research Group. Braun gratefully acknowledges �nancial support from Fondecyt Chile [grant number1060015.] 1. See Levine (2006) for an extensive review of the literature on the subject. THE WORLD BANK ECONOMIC REVIEW, VOL. 24, NO. 2, pp. 234– 279 doi:10.1093/wber/lhq007 Advance Access Publication June 23, 2010 # The Author 2010. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org 234 ´as Braun and Claudio Raddatz Matı 235 considerations (LaPorta and others 1997, 1998; Acemoglu and Johnson 2005, Stulz and Williamson 2003, and Guiso, Sapienza, and Zingales 2004). Private interests and politics appear to be relevant determinants of �nancial develop- ment, as suggested, for instance, by Rajan and Zingales (2003), Pagano and Volpin (2001, 2005), and Braun and Raddatz (2007, 2008). One channel through which this could occur is the regulatory effect of the interaction between politicians and �nancial sector �rms.2 That regulators come from or end up in the regulated industry—the revolving door phenomenon—has long been recognized as a potential determinant of regulation.3 And indeed, the Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 empirical work, although still scarce, points to its having large social costs (see Kwhaja and Mian 2005 and Dal Bo ´ and Rossi 2006.) As for how �rms react to �nancial sector underdevelopment, several recent papers have documented that politically connected �rms seem to get preferen- tial access to credit (Cull and Xu 2005; Khwaja and Mian 2005) and better treatment by the government. These links between politics and business seem quite widespread (Faccio 2006) and seem to add considerable value to �rms (Fisman 2001). This article focuses on banks. Because of their critical role in allocating credit, the behavior of banks, unlike that of most other types of �rms, affects the entire economy. A new dataset linking more than 10,000 politicians (cabinet members, �nancial sector regulators, and central bank governors) and some 60,000 members of bank boards in a large number of countries is used to compare the names of bankers and of politicians to search for matches. The frequency of these matches is then used to compute measures of the connection between politicians and bankers to explore the role of political connectedness. Banks, like any other �rm, may use these connections to improve their pos- ition, perhaps by affecting banking regulation. This would be more likely to happen where institutions are weak and the government is relatively powerful yet less accountable. It may also carry large social costs, through more restricted access to credit. The article examines the extent to which banks are politically connected, where this connectedness is more prevalent, and whether it is associated with better outcomes for banks. This private-interest view of the presence of former politicians on bank boards is, of course, not the only possibility. Links between politicians and bankers may be a way of fruitfully sharing ability, knowledge, and experi- ence between the public and private sectors. These links could imply better outcomes for the �rm without negative social effects. Banks could simply be lobbying to make a legitimate case to government of�cials or could consider these links more as consumption than as investment (see, for instance, Ansolabehere, de Figueiredo, and Snyder 2003). The merit of these two 2. Financial sector incumbents are de�ned as the people and �rms that form part of the �nancial sector at a given time, as opposed to those interested in entering the sector. 3. See Dal Bo´ (2006) for a review of regulatory capture. 236 THE WORLD BANK ECONOMIC REVIEW perspectives is ultimately an empirical question. In that sense, the stylized facts provided in this article may shed some light on which interpretation is more likely. Several stylized facts stand out. At the micro level, politically connected banks are different from unconnected banks: they are larger, more pro�table, less leveraged, and less risky. When aggregated at the country level in various ways, bank connectedness is found to be strongly negatively related to GDP per capita. After controlling for this and for other traditional elements, countries where banks are more connected are shown to rank higher on cor- Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 ruption and government regulatory power and lower on accountability. Overall regulation is less market friendly, bank regulation is generally more pro-banker, and the �nancial system is less developed. This article is closely related to the recent literature showing that politically connected �rms appear to fare better than the rest (see, for example, Faccio 2006; and Faccio, Masulis, and McConnell 2005.) This article adds to this work in three main ways. It focuses on banks, an important contribution because of the likely effect bank connectedness may have on the entire economy through credit allocation. Rather than determining whether political connections improve outcomes for the connected �rms, it delves deeper into the country characteristics and policy choices associated with these kinds of connections. And it looks at former politicians as well as incumbents. The article is also related to the literature on the search for political experi- ence by boards of directors (see, for instance, Agrawal and Knoeber 2001 and Goldman and others 2009). Similarly, it is related to recent work on the relationship between connections and development, including banking sector development from a historical perspective (Haber 1991; Maurer 2002; Maurer and Gomberg 2005; Milanovic, Hoff, and Horowitz forthcoming; and Razo forthcoming). In this article, the assembly of the new dataset has allowed con- sistent exploration of the issue across a large number of countries. The article compares politically connected banks to banks that are not con- nected and correlates several country-level measures of connectedness with variables capturing the quality of institutions, bank regulation, and �nancial development. Section I describes the data and the matching procedure used to identify banker-politicians. It also discusses ways of aggregating the results into a country-level connectedness variable. Section II shows how connected banks differ from unconnected ones and explores the characteristics of countries where the phenomenon is more frequent. Section III presents conclusions and implications. I. MEASURING THE CONNECTION BETWEEN BANKERS AND POLITICIANS This section describes the methodology used to measure the connection between bankers and politicians, presents summary statistics from the resulting ´as Braun and Claudio Raddatz Matı 237 dataset, and introduces aggregate measures of the degree of connection across countries. Building the Data The data on names of politicians came from the Country Reports of the Economist Intelligence Unit, which were revised twice yearly for each country for 1996–2005. This review yielded 72,769 names of cabinet members and central bank governors. These names were complemented by a smaller set of Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 593 names of �nancial sector supervisors obtained from the 2000, 2002, 2003, and 2004 editions of How Countries Supervise their Banks, Insurers, and Securities Markets (Central Bank Publications, various years). These two data sets together provide extensive coverage for cabinet members and �nancial sector supervisors in 154 countries over 10 years (see supplemental appendix table S1, column 3, available at http://wber.oxfordjournals.org/). Once cleaned (as explained below), the data yielded an average of 72 politicians in each country, which is around 7 a year. There is some variation across countries, but it is small: 40 –100 names of politicians were found for 70 percent of the countries. The names of bank board members are from Bankscope (Bureau van Dijk 2006), which has data on the most recent board composition of both listed and unlisted banks in nearly all countries. The data were collected for 2006, so the board composition is typically from December 2005. Once duplicates were identi�ed among the 109,645 board member names found for 4,618 banks, 64,169 unique board member names remained. Although Bankscope is the most comprehensive source of bank data around the world, its coverage is not necessarily complete. It is close to universal, however, as evidenced by the cor- respondence between the average number of banks with board composition data in Bankscope in 2001 and the total number of commercial banks reported by Barth, Caprio, and Levine (2003) for the same year (see column 5 in sup- plemental appendix table S1). Although there is some variation across countries, the difference between the number of banks in the two datasets falls within a 20 percent range in about 70 percent of countries. The banks for which board data are available account, on average, for 72 percent of the assets in each country; in only about a fourth of the countries is the fraction below 60 percent (see column 6). Because data on bank directors are from the December 2005 issue of Bankscope and data on politicians cover the period 1996–2005, matches between the two datasets typically consist of former politicians who later sit on a bank board. This is the convention followed in the rest of the article, which refers to these individuals as “former politicians.� There are a few caveats with this terminology. First, the entire history of each individual is unknown. Thus, some of them may have been bankers before 1996 (the �rst observation of the politician dataset). Second, how long a director has been sitting on the board is 238 THE WORLD BANK ECONOMIC REVIEW also unknown. For instance, the data do not show whether a politician who is on a board in 2005 and whose term in government ended in 2004 was already sitting on the board in 2003. Third, matches between politicians who are in their political positions in 2005 correspond to cases where the politicians sim- ultaneously hold both positions. Fourth, a given issue of Bankscope reports the latest director data available. In more than two-thirds of cases, this corresponds to December 2005, but in a few cases the data are from earlier years. Thus, to be more precise, “former politicians� refer to individuals who were politicians at some point during 1996–2005 and who were on a bank board in December Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 2005. Finding matches between politicians’ and bankers’ names involved four steps. First, the strings containing the names were standardized by converting them to lowercase and removing punctuations and titles (Sir, PhD, and so on). Second, duplicate entries were removed by identifying observations that were simply different spellings of the same name (for instance, with and without the middle initial). Third, the datasets containing names of politicians were pooled and duplicate observations across the datasets were identi�ed. Once the names had been cleaned in this way, the names in the politician and banker datasets were compared to obtain the matching observations. At each step, a record-linkage algorithm was used to �nd matching names. The algorithm forms all possible pairs of names within each country and ranks the pairs on three standard measures of string similarity from the record-linkage literature: bigram, Levenshtein, and longest common subse- quence.4 The bigram metric counts the number of consecutive matching pairs of characters between two strings. The Levenshtein measure counts the minimal number of edits required to convert one string into the other. Allowable edit operations are the deletion of a single character, the insertion of a single character, and the substitution of one character for another. The longest common subsequence counts the number of consecutive characters that are present in two strings, and keeps the largest number. All three methods are based on the way names are written. If the difference between the way a name sounds and the way it is written varies across countries, so that mistakes are more prevalent in some countries than in others, these methods could be differentially effective and could potentially induce bias. For these reasons, the algorithms were used only to restrict the sample of potential matches, as described below. Ultimately, the matches were visually identi�ed. When two strings containing names are compared, each of these criteria results in a value between 0 and 1 that measures the likeliness of the two names being the same. All pairs with a minimum value of 0.8 in at least one of the three methods were retained and visually checked to determine 4. The record linkage software used was Merge Toolbox, a Java-based tool created by the members of the Safelink project (see Schnell, Bachteler, and Bender 2004). ´as Braun and Claudio Raddatz Matı 239 whether they matched. While alternative ways could have been used to restrict the set of pairs to be visually checked, this relatively restrictive way was chosen so as to err more on the side of failing to �nd true matches than of falsely identifying matches. This was also the basic principle used for the visual veri�cation. After step two, the data contained 10,829 politicians and 62,981 bankers in 146 countries. Step three yielded 218 matching names across these two lists (see column 4 in table S1). The mean number of matches per country was 1.4, and the median was 1.0. At 0.34 percent, the share of bankers who Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 are politicians is quite small and unimpressive. The dearth of matches reflects in part the restrictive way that the matches were identi�ed. On the other hand, the fraction of politician-bankers does not seem as small in the context of the size of the populations from where they were drawn (see below). Having high-ranking politicians on the board of banks is not the only way banks can be politically connected. Non-cabinet level politicians can also play an important role connecting banks. And there are more subtle forms of con- nection: a politician can be connected to a bank by having relatives or associ- ates on the board (Faccio 2006) or by supporting the appointment of directors or chief executive of�cers. There are also less subtle ways, such as outright bribery and corruption. Politicians sitting on bank boards seem to be a rela- tively rare form of connection compared with some other channels, to judge by country case studies and anecdotal evidence.5 However, these other types of connections are much more dif�cult to document systematically across countries. Rather than arguing that a direct presence on the board is the only or the most important way politicians and bankers relate, the article considers the presence of high-level politicians on bank boards as a proxy for the general connection between politicians and bankers. As long as people do not completely specialize in one particular form of connection, the different ways of connecting are likely to be positively correlated. Since the analysis here looks only at the top posts in both politics and banking, the results are likely just the tip of the iceberg. Instead of focusing on absolute magnitudes, the article looks at how vari- ations in the importance of politicians sitting on bank boards links to several bank and country characteristics. There are two sources of variation in the data: variation between countries with matches and those without (the exten- sive margin) and variation in the number of matches for the countries with at least one match (the intensive margin). The 72 of 154 countries for which no matches were found were dropped from the sample for most of the analyses, for several reasons. Most important is concern about the reliability of the data for many countries with zero 5. See, for instance, Fisman (2001) for an account of Suharto’s Indonesia. 240 THE WORLD BANK ECONOMIC REVIEW matches. For instance, while 60 percent of countries with some matches meet the International Monetary Fund’s Special Data Dissemination Standard (IMF 2009), only 20 percent on those with zero matches do (many of these countries are not generally included in systematic cross-country analyses).6 Second, many countries with zero matches have very few banks. A third had fewer than three banks in Bankscope in 2005, compared with just 4 percent among countries with matches. And the median number of banks with data is 5 in the no-matches group but 16 in the group with at least one match. Third, the zeroes give little information on whether the selection of bankers is biased Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 toward former politicians. Under reasonable assumptions, the probability of �nding zero matches between bankers and politicians is high even if banker selection is seriously biased toward picking politicians.7 In contrast, �nding even one match provides considerable information on the likely bias of the selection, since a match is typically a low probability event under the null hypothesis of unbiased matching. Nevertheless, results are also presented for analyses that include countries with zero matches but more than two banks (as an arbitrary cutoff for considering the zero as reliable), and many of the corre- lations documented below remain unaffected. Of course, this argument could be stretched to restrict the sample to countries with more than one or two matches because �nding a small number of matches may simply occur by chance, something that is less likely if a more substantial number of matches are found. Although �nding a single match is a very low probability event that is unlikely to occur by chance in most countries, the article returns to this issue below to show that, even though the sample size drops quite rapidly, the results are not very different when the sample is further restricted. Measuring Connectedness at the Aggregate Level There are several ways to aggregate the information on individual matches to measure and compare the connectedness between banking and politics in different countries. Each method has pros and cons and is more or less appro- priate under different assumptions about the process that generates the matches between politicians and bankers. Instead of focusing on a single measure, the analysis is conducted with �ve different metrics (table 1). Three measures are straightforward, and two are more elaborate because they address some short- comings of the other three. The �ve measures are computed twice: once for matches found for all Bankscope banks ( public, private, and mixed), and once only for matches for fully private banks. 6. For instance, 63 percent of these countries were not included in the cross-country analysis of bank regulation by Barth, Caprio, and Levine (2003). Some were included in later rounds of the survey, but coverage is incomplete. 7. See the appendix for a description of the distribution of matches under an unbiased selection process. T A B L E 1 . Measures of the Degree of Connectedness across Countries between Banks and Politicians All Banks Fully Private Banks Only World Bank Country FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE Country Code (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gabon GAB 100 100 10 7.75 0.1 - - - - - Georgia GEO 100 100 8 8.48 0.08 100 100 8 8.48 0.08 Myanmar MMR 100 - 19 11.93 - 100 - 19 11.93 0 Angola AGO 67 66 8 9.76 0.02 - - - - - Burundi BDI 67 64 8 8.61 0.03 33 9 3 7.71 0.8 Madagascar MDG 67 68 6 9.23 0.04 100 29 14 10.05 0.08 Cameroon CMR 50 84 4 8.86 0.27 50 84 4 8.86 0.27 Malta MLT 50 53 5 6.16 0.94 0 0 0 - - Rwanda RWA 50 52 5 8.51 0.1 50 52 4 8.37 0.4 Belarus BLR 45 84 4 8.59 0.04 38 62 4 8.52 0.04 Qatar QAT 43 61 5 6.41 0.4 33 11 2 5.76 1 Uzbekistan UZB 40 89 5 9.73 0.02 40 89 5 9.73 0.02 Peru PER 38 29 2 8.14 0.09 29 12 1 7.88 0.13 Bangladesh BGD 35 23 2 10.44 0 35 11 2 10.27 0.01 Morocco MAR 33 17 3 9.39 0.02 50 17 4 9.68 0.01 Sierra Leone SLE 33 12 7 7.94 0.18 50 12 11 8.34 0.11 United Arab ARE 32 44 3 7.77 0.06 50 23 7 8.55 0.04 Emirates Kuwait KWT 27 20 2 6.46 0.51 43 20 3 6.96 0.31 Matı Tunisia TUN 27 44 3 8.3 0.04 0 0 0 - - Burkina Faso BFA 25 16 2 7.98 0.64 50 16 5 8.97 0.21 El Salvador MRT 25 37 2 7.45 1.09 0 0 0 - - Mauritania OMN 25 32 2 6.42 0.58 33 32 3 6.67 2 Oman SLV 25 - 2 6.62 2.15 50 - 5 7.63 0.9 Chile CHL 23 30 1 8.09 0.09 25 30 2 8.11 0.09 Jordan JOR 22 5 2 6.64 0.64 22 5 2 6.64 0.64 Hungary HUN 21 37 2 7.64 0.09 27 37 2 7.85 0.09 Colombia COL 20 14 1 8.58 0.06 0 0 0 - - Serbia & SDN 20 - 1 7.19 - 22 - 2 7.22 0.3 Montenegro Sudan YUG 20 23 1 8.7 0.05 25 23 2 8.87 0.05 Philippines PHL 18 34 2 9.37 0.01 20 34 2 9.52 0.01 ´as Braun and Claudio Raddatz Zambia ZMB 18 17 3 8.06 0.15 22 17 3 8.24 0.1 Armenia ARM 17 11 3 7.15 1.47 20 11 4 7.36 1.17 (Continued ) 241 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 TABLE 1. Continued 242 All Banks Fully Private Banks Only World Bank Country FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE Country Code (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Lebanon LBN 17 23 1 6.25 1 17 23 1 6.25 1 Moldova, Rep. MDA 17 5 4 7.37 0.23 25 5 6 7.82 0.12 Yemen YEM 17 20 4 9.2 0.1 33 20 14 10.43 0.05 Iceland ISL 15 60 2 5.52 1.63 14 37 2 5.31 9.21 Dominican DOM 14 25 1 7.29 1.24 0 0 0 - - Republic Lithuania LTU 14 7 3 6.96 1.61 14 7 3 6.96 1.61 Macedonia, MKD 14 37 1 5.57 4.96 20 37 2 5.94 5 FYR Korea, Rep. KOR 12 21 1 8.18 0.06 9 4 0 7.15 0.93 Romania ROM 12 42 2 8.1 0.1 0 0 0 - - Croatia HRV 11 23 3 7.17 0.13 0 0 0 - - Hong Kong, HKG 11 16 1 6.81 0.12 10 14 1 6.87 0.12 China THE WORLD BANK ECONOMIC REVIEW Latvia LVA 11 24 2 6.19 0.6 7 20 1 6.05 1.21 South Africa ZAF 11 2 1 8.34 0.05 12 2 1 8.46 0.06 Taiwan, China TWN 11 9 0 - - 4 5 0 - - Thailand THA 11 24 1 8.13 0.08 8 23 1 8.11 0.39 Nepal NPL 9 15 1 7.61 0.94 10 15 1 7.68 0.88 Belgium BEL 8 10 1 7.42 0.13 7 10 1 7.01 0.29 Canada CAN 8 1 0 7.13 0.41 10 1 0 7.37 0.32 Egypt, Arab EGY 8 6 1 9.15 0.05 8 2 1 9.36 0.14 Rep. Israel ISR 8 1 1 5.88 5.21 14 1 1 6.48 2.61 Turkey TUR 8 25 1 7.72 0.1 10 25 1 7.9 0.1 Brazil BRA 7 15 1 9.4 0.02 6 2 1 8.96 0.05 Cyprus CYP 7 3 1 4.65 11.57 8 3 1 4.87 10 Nigeria NGA 7 6 0 8.49 0.1 4 1 0 7.91 0.6 Russian RUS 7 58 1 9.51 0.01 6 7 1 9.17 0.02 Federation Uganda UGA 7 27 1 8.31 0.36 7 27 1 8.31 0.4 China CHN 6 16 0 11.58 0 0 0 0 - - Finland FIN 6 - 0 6.29 3.34 8 - 1 6.45 2.51 Austria AUT 5 13 0 6.24 0.57 5 13 0 6.28 0.57 Netherlands NLD 5 49 0 7.37 0.16 4 36 0 7.12 0.32 India IND 4 22 0 10.2 0.01 4 1 0 10.29 0.02 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Mexico MEX 4 3 2 10.19 0.01 0 0 0 - - Pakistan PAK 4 1 0 8.73 0.29 0 0 0 - - Portugal PRT 4 1 0 5.68 6.13 4 1 0 5.75 5 Denmark DNK 3 10 0 5.56 1.55 3 10 0 5.61 1.55 Indonesia IDN 3 0 0 8.34 0.31 3 0 0 8.58 0.31 Luxembourg LUX 3 11 0 3.54 14.88 1 4 0 2.91 - Norway NOR 3 0 0 5.1 11.46 0 0 0 - - Poland POL 3 5 0 6.86 1.94 4 5 0 7.18 1.16 Singapore SGP 3 - 0 5.88 4.93 3 - 0 5.93 4.93 Australia AUS 2 2 0 6.73 2.16 0 0 0 - - Argentina ARG 1 0 0 7.09 1.33 0 0 0 - - France FRA 1 4 0 5.4 8.3 1 4 0 5.45 7.61 Germany DEU 1 3 0 6.56 0.26 1 1 0 5.69 1.31 Italy ITA 1 8 0 6.47 0.34 1 2 0 5.65 1.7 Japan JPN 1 0 0 5.95 4.57 1 0 0 5.98 4.57 Spain ESP 1 15 0 5.67 6.11 1 15 0 5.75 5 Switzerland CHE 1 0 0 4.77 12.11 1 0 0 4.96 10 United GBR 1 0 0 6.9 0.28 2 0 0 6.98 0.3 Kingdom United States USA 1 8 0 7.82 0.07 1 7 0 7.32 0.1 Albania ALB 0 0 0 - - 0 0 0 - - Algeria DZA 0 0 0 - - 0 0 0 - - Andorra ADO 0 0 0 - - 0 0 - - - Antigua & ATG 0 0 0 - - 0 0 - - - Barbuda Aruba ABW 0 0 0 - - 0 0 0 - - Matı Azerbaijan AZE 0 0 0 - - 0 0 0 - - Bahamas BHS 0 0 0 - - 0 0 0 - - Bahrain BHR 0 0 0 - - 0 0 0 - - Barbados BRB 0 0 0 - - 0 0 0 - - Benin BEN 0 0 0 - - 0 0 0 - - Bermuda BMU 0 0 0 - - 0 0 0 - - Bolivia BOL 0 0 0 - - 0 0 0 - - Botswana BWA 0 0 0 - - 0 0 0 - - Brunei BRN 0 0 0 - - 0 0 0 - - Darussalam Bulgaria BGR 0 0 0 - - 0 0 0 - - Cambodia KHM 0 0 0 - - 0 0 0 - - Cape Verde CPV 0 0 0 - - 0 0 0 - - Cayman CYM 0 0 0 - - 0 0 0 - - ´as Braun and Claudio Raddatz Islands Costa Rica CRI 0 0 0 - - - - - - - 243 (Continued ) Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 TABLE 1. Continued 244 All Banks Fully Private Banks Only World Bank Country FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE Country Code (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Coˆ te d’Ivoire CIV 0 0 0 - - 0 0 0 - - Cuba CUB 0 0 0 - - 0 0 0 - - Czech Republic CZE 0 0 0 - - 0 0 0 - - Djibouti DJI 0 0 0 - - 0 0 0 - - Ecuador ECU 0 0 0 - - 0 0 0 - - Estonia EST 0 0 0 - - 0 0 0 - - Ethiopia ETH 0 0 0 - - 0 0 0 - - Gambia, The GMB 0 0 0 - - 0 0 0 - - Ghana GHA 0 0 0 - - 0 0 0 - - Gibraltar GIB 0 0 0 - - 0 0 - - - Greece GRC 0 0 0 - - 0 0 0 - - Grenada GRD 0 0 0 - - 0 0 - - - Guatemala GTM 0 0 0 - - 0 0 0 - - Guyana GUY 0 0 0 - - 0 0 0 - - THE WORLD BANK ECONOMIC REVIEW Haiti HTI 0 0 0 - - 0 0 0 - - Honduras HND 0 0 0 - - 0 0 0 - - Iran, Islamic IRN 0 0 0 - - 0 0 0 - - Rep. Ireland IRL 0 0 0 - - 0 0 0 - - Jamaica JAM 0 0 0 - - 0 0 0 - - Kazakhstan KAZ 0 0 0 - - 0 0 0 - - Kenya KEN 0 0 0 - - 0 0 0 - - Kyrgyz KGZ 0 0 0 - - 0 0 0 - - Republic Lesotho LSO 0 0 0 - - 0 0 0 - - Libya LBY 0 0 0 - - - - - - - Liechtenstein LIE 0 0 0 - - 0 0 0 - - Macao MAC 0 0 0 - - 0 0 0 - - Malawi MWI 0 0 0 - - 0 0 0 - - Malaysia MYS 0 0 0 - - 0 0 0 - - Maldives MDV 0 0 0 - - - - - - - Mali MLI 0 0 0 - - 0 0 0 - - Mauritius MUS 0 0 0 - - 0 0 0 - - Monaco MCO 0 0 0 - - 0 0 - - - Mongolia MNG 0 0 0 - - 0 0 0 - - Mozambique MOZ 0 0 0 - - 0 0 0 - - Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Namibia NAM 0 0 0 - - 0 0 0 - - Netherlands ANT 0 0 0 - - 0 0 - - - Antilles New Zealand NZL 0 0 0 - - 0 0 0 - - Niger NER 0 0 0 - - 0 0 0 - - Panama PAN 0 0 0 - - 0 0 0 - - Papua New PNG 0 0 0 - - 0 0 0 - - Guinea Paraguay PRY 0 0 0 - - 0 0 0 - - Saint Kitts & KNA 0 0 0 - - 0 0 - - - Nevis Saint Lucia LCA 0 0 0 - - - - - - - Samoa WSM 0 0 0 - - 0 0 0 - - San Marino SMR 0 0 0 - - 0 0 - - - Saudi Arabia SAU 0 0 0 - - 0 0 0 - - Senegal SEN 0 0 0 - - 0 0 0 - - Seychelles SYC 0 0 0 - - 0 0 0 - - Slovakia SVK 0 0 0 - - 0 0 0 - - Slovenia SVN 0 0 0 - - 0 0 0 - - Sri Lanka LKA 0 0 0 - - 0 0 0 - - Suriname SUR 0 0 0 - - - - - - - Swaziland SWZ 0 0 0 - - 0 0 0 - - Sweden SWE 0 0 0 - - 0 0 0 - - Syrian Arab SYR 0 0 0 - - - - - - - Rep. Tanzania TZA 0 0 0 - - 0 0 0 - - Matı Togo TGO 0 0 0 - - 0 0 0 - - Trinidad & TTO 0 0 0 - - 0 0 0 - - Tobago Ukraine UKR 0 0 0 - - 0 0 0 - - Uruguay URY 0 0 0 - - 0 0 0 - - Venezuela, RB VEN 0 0 0 - - 0 0 0 - - Vietnam VNM 0 0 0 - - 0 0 0 - - Zimbabwe ZWE 0 0 0 - - 0 0 0 - - Total 10 12 1 7.58 1.54 9 7 1 7.52 1.4 Note: Countries are sorted in decreasing order by fraction of banks with data on board of directors that had a former politician on their boards; countries with the same values are listed alphabetically. FRACBANKS is the fraction of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole popu- ´as Braun and Claudio Raddatz lation of a country. MAXSHARE is the largest fraction of a country’s population from which politicians and bankers would have to be selected so that the hypothesis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. Not available is coded as “-�. Source: Authors’ analysis based on data described in the text. 245 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 246 THE WORLD BANK ECONOMIC REVIEW FRACTION OF CONNECTED BANKS. The fraction of connected banks (FRACBANKS) is the number of banks with at least one former politician on the board of directors divided by the number of banks for which there are data on board members. The mean fractions of connected banks of 10 percent for all kinds of banks and 9 percent for private banks are much larger than the fraction of matches among individuals documented above. Indeed, when only countries with at least one match are considered, the average share increases to about a �fth of the banks. There is interesting variation across countries. The countries with fewest connected banks are Germany, the United States, Italy, Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Japan, and Switzerland, all with less than 2 percent of banks connected in this way. In contrast, more than two-thirds of the banks are connected in Gabon, Georgia, Myanmar, Angola, Burundi, and Madagascar. The picture is gener- ally the same whether considering all banks or just private banks; the corre- lation between the two groups is 0.86. The rationale behind this �rst aggregation is that what determines a signi�- cant political link for a bank is whether the bank has at least one politician on its board. The higher the fraction of the banks in the system that are con- nected in this way, the larger the degree of connectedness between banking and politics. The issue is not about having a large number of people in both worlds but rather about having people in the right place, even if their number is small. In this sense, FRACBANKS is more naturally interpreted as a measure of the institutional connection between banking and politics, rather than a personal matter related perhaps to the existence of a common set of skills. SHARE OF ASSETS OF CONNECTED BANKS. A simple variation on the FRACBANKS measure consists of computing the share of total banking system assets in banks that have a politician on their board. This metric, the share of assets of connected banks (SHAREASSETS), has the advantage of acknowledging that larger banks might differ from smaller ones in their need or ability to connect to politics. Smaller banks may �nd free-riding on the connections of large banks more pro�table than establishing their own connections. Also, this measure would probably be more relevant when looking at the likely effects of connectedness since it would be a measure of the amount of credit that is subject to these links. This metric is then more likely a proxy for the extent of power—both political and economic—that these relationships might entail. On a more technical note, giving a higher weight to larger banks minimizes the potential problems induced by the smaller coverage for small banks. SHAREASSETS is strongly and signi�cantly correlated with FRACBANKS, both for all banks and for private banks (table 2). For countries with at least one match, the mean share is 25 percent for all banks and 18 percent for private banks. The groups of countries that rank very high and very low are similar to those for the FRACBANKS measure. These results suggest that the difference between large and small banks might not be very relevant. The T A B L E 2 . Correlation among Measures of Connectedness All Banks Fully Private Banks Only Measures FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE FRACBANKS SHAREASSETS FRACBANKERS PREVALENCE MAXSHARE FRACBANKS 1.00 1.00 SHAREASSETS 0.88*** 1.00 0.75*** 1.00 FRACBANKERS 0.92*** 0.82*** 1.00 0.87*** 0.55*** 1.00 PREVALENCE 0.40*** 0.30*** 0.43*** 1.00 0.56*** 0.30 0.57*** 1.00 MAXSHARE 2 0.30*** 2 0.30*** 2 0.31*** 2 0.70*** 1.00 2 0.32** 2 0.17** 2 0.25** 2 0.66*** 1.00*** **Signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Correlations are computed including the countries with zero matches (for those measures that can take the value zero). FRACBANKS is the frac- Matı tion of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. MAXSHARE is the largest fraction of a country’s population from which politicians and bankers would have to be selected so that the hypoth- esis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. Source: Authors’ analysis based on data described in the text. 247´as Braun and Claudio Raddatz Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 248 THE WORLD BANK ECONOMIC REVIEW correlation between the measures computed over all banks and over private banks is also quite high (0.79). FRACTION OF CONNECTED BANKERS. The third measure, fraction of connected bankers (FRACBANKERS), is the ratio of the number of matches to the number of bank directors in the dataset. This metric looks at the extent to which politicians populate bank boards. The average fraction of connected bankers across all countries is around 1 percent and is close to 2 percent for countries with more than one match. These numbers suggest that the phenom- Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 enon is not particularly frequent. The correlation with the �rst two measures is small (0.34 for FRACBANKS and 0.38 for SHAREASSETS for all banks) but statistically signi�cant. Furthermore, the countries at both tails of the measure are similar to those at the tails of the previous two measures. Thus, despite the low level of the variable, its cross-country variation captures a similar concept to the previous two. PREVALENCE. The �rst three measures of connectedness are easy to compute and natural in their interpretation. But they do not take into account that the expected number of banker-politicians may differ across countries even if the selection of bankers is not biased toward former politicians. In particular, countries with more matches might simply be countries with fewer people from which both bankers and politicians are selected. For greater precision, the probability of obtaining a given number of matches was derived under the assumption that the people needed to �ll the politician and banker posts are selected randomly with replacement (at the sample level) from a common pool (see the appendix). Everyone in the pool has the same probability of being selected for either position, and there is no bias in favor of politicians in the selection of bankers. This probability is then used to compute the expected number of matches assuming that the common pool is the entire population of each country (more on this below). This ratio of actual to expected matches (in logs) is called PREVALENCE. The correlation of this metric with the previous ones is not as strong as for the others, particularly with FRACBANKERS, but it is still positive. The countries that rank highest in this connectedness measure are Myanmar, China, Bangladesh, India, and Mexico (see table 1). The countries where the phenomenon is least prevalent include Luxembourg, France, Switzerland, and Norway. For most countries the actual number of matches is many times larger than the expected number because of the assumption that the pool from which directors are selected is the total population of a country. Since it is highly unli- kely that every person has the same probability of being chosen as a politician or a banker, the results for this measure are exaggerated. Nevertheless, the cross-country variation in this measure is the same as it would be if it were assumed that the selection pool for bankers and politicians is a �xed fraction of a country’s population. In fact, it can be shown that the expected number of ´as Braun and Claudio Raddatz Matı 249 matches is proportional to the size of the pool. Therefore,   ELITE ð1Þ PREVALENCE ¼ PREVALENCEðELITEÞ þ log POP where PREVALENCE(ELITE) is the log ratio of actual to expected matches   ELITE considering the true size of the elite, and log is the log ratio of the POP size of the elite as a fraction of the population. Thus, as long as the elite are a �xed fraction of the population across countries, the PREVALENCE measure Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 and true prevalence would differ only in a constant. The measure will be incorrect, however, if there is systematic variation across countries in the elite as a share of population. This could happen if the number of elite is relatively �xed in all countries, so that the elite decline as a fraction of the population from smaller to larger countries. The analysis con- siders the size of the population in each country to control for this possibility. MAXIMUM SHARE OF POPULATION FOR RANDOMNESS. Another possibility is that the size of the elite is related to the educated portion of the population. If one assumes that the pool is the number of people with a tertiary education, the expected �gures are closer to the actual number of matches. This correction incorporates the possibility that PREVALENCE is highest in some countries simply because there are so few people in those countries who are capable of assuming these posts. The correction, however, is not free of problems because it is not obvious that the relevant pool is the group of highly educated people. On the one hand, the pool may be too narrowly de�ned since not all the bankers and politicians have a tertiary education.8 On the other, the pool might not be suf�- ciently small if a certain kind of economic or �nancial skill is shared between politicians working in economic spheres within the government and bankers. Most important, such a correction might confound the interpretation of the results because one variable mixes two concepts—availability of human capital and connectedness—that may have independent (and opposite) effects on many country characteristics (such as real GDP per capita). The �nal measure, maximum share of population for randomness (MAXSHARE), takes into consideration the uncertainty about the size of the pool of individuals from which bankers and politicians are selected. MAXSHARE corresponds to the largest pool (as a fraction of the population) from which bankers and politicians are selected so that the hypothesis that the selection is random could not be rejected at the 5 percent level (for the number of matches found in the data). For most countries, in order not to reject this hypothesis, the size of the pool turns out to be a very small fraction of the 8. See Dreher and others (2009) for data on the educational attainment of presidents. These data show that 30 percent of presidents worldwide since 1975 did not receive a higher education. 250 THE WORLD BANK ECONOMIC REVIEW population. As expected, this variable is negatively correlated with the previous ones because it measures the inverse of the underlying concept. The usual groups of countries are at both extremes of the metric. CONNECTEDNESS AND COUNTRY CHARACTERISTICS. Overall, the different measures are signi�cantly correlated, suggesting that they are likely to be different proxies for the same general concept. It is also clear that considering links solely to private banks makes little difference, suggesting that politicians sitting on the boards of state-owned banks do not drive the �ndings for the various measures. Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Countries that rank highest on the connectedness measures9 (Bangladesh, China, Mexico, India, and the Russian Federation) and those that rank lowest (Luxembourg, Switzerland, Cyprus, Norway, and France ) clearly differ in many other respects as well. The most obvious is economic development. Countries where connectedness is more prevalent are signi�cantly poorer than countries where it is less prevalent. Mean GDP per capita is $3,944 for countries with higher than the median share of connected banks, and $18,958 for the others. The share of connected banks in countries with lower than median per capita GDP (28.2 percent) is two and a half times larger than the share in more developed countries (11.4 percent). The picture is about the same for the other measures and when only private banks are considered. The second distinctive feature is that countries where the prevalence of con- nectedness is higher also appear to have less developed institutions. For instance, countries with lower than median connectedness have control of corruption indi- cators (de�ned below) that are one standard deviation higher than countries with higher prevalence. While the share of connected banks is 15.1 percent in countries with higher than median control of corruption, it is 26.5 percent in the rest. Finally, banking sector development differs considerably across the two groups of countries. The ratio of private credit to GDP (from Beck, Demirguc-Kunt, and Levine 2000) is 3 times higher where connectedness measures are lower (76 percent) than where they are higher (25 percent), while the share of connected banks is almost twice as high in countries where banking sector development is low (26.5 percent) than in those where it is high (15.1 percent). Connectedness, then, does not seem to be equally distributed across countries but rather to cluster in countries where things do not work very well. In particu- lar, connectedness is higher where economic development is low and where institutions and the �nancial system are underdeveloped. These are some of the relationships examined more deeply in the following section. I I . T H E CO R R E L AT E S OF CONNECTEDNESS This section explores the correlates of connectedness �rst at the bank level and then at the cross-country level. It shows that the measures of connectedness 9. Giving equal weight to each of the �ve different connectedness measures. ´as Braun and Claudio Raddatz Matı 251 introduced above are robustly correlated to important bank and country characteristics and also to policy choices. Bank Characteristics Connected and unconnected banks can be compared on several characteristics. Here, they are compared on measures of size, pro�tability, leverage, and riski- ness, which were constructed from Bankscope data using bank statements at the end of 2004. Table 3 shows averages for these characteristics for connected and uncon- Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 nected banks, their differences, and whether the differences are statistically sig- ni�cant according to a simple test of means. Clearly, connected banks are larger, more pro�table, and less leveraged than are unconnected banks. They also have a smaller share of net charge-offs to gross loans, suggesting that they take less risk than unconnected banks, although on a worldwide comparison, the difference is not signi�cant. The sign and signi�cance of these differences remain unchanged when only fully private banks are considered. The regressions in table 4 further test whether these correlations hold when connected and unconnected banks are compared within a country. The param- eters of the following parsimonious speci�cation are estimated: ð2Þ Yi;c ¼ a þ b  CONNECTEDi;c þ gSIZEi;c þ uc þ 1i;c where Yi,c corresponds to the �nancial characteristics of bank i in country c, which include measures of size, pro�tability, riskiness, liquidity, and leverage; CONNECTEDi,c is a dummy variable that takes a value of 1 if at least one of the bank’s directors has been a politician or bank regulator, and 0 otherwise; SIZEi,c controls for (log) total assets (except when the left-side variable is itself T A B L E 3 . Differences between Connected and Unconnected Banks, Worldwide Comparison of Average Bank Characteristics (tests of equality of means) Bank Characteristics Connected Unconnected Difference All Bankscope Banks Total assets 9.72 8.60 1.12*** Return on average assets 2.40 1.26 1.14*** Equity/Total assets 14.23 11.44 2.79*** Net charge-off/Average gross loans 0.70 1.24 2 0.54 Fully Private Banks Only Total assets 9.58 8.44 1.14*** Return on average assets 2.46 1.19 1.27*** Equity/Total assets 15.20 11.17 4.02*** Net charge-off/Average gross loans 0.66 1.11 2 0.45 ***Signi�cant at the 1 percent level on a simple test of means. Source: Authors’ analysis based on data described in the text. 252 THE WORLD BANK ECONOMIC REVIEW T A B L E 4 . Differences between Connected and Unconnected Banks, Within-Country Comparison of Bank Characteristics (regression analysis) Total Return on Equity/Total Net Charge-off/Average Assets (1) Average Assets (2) Assets (3) Gross Loans (4) All Bankscope Banks Connected 0.3358** 0.0062** 0.0225** 2 0.0054** (0.1349) (0.0025) (0.0105) (0.0023) Number of observations 3,312 3,285 3,311 1,176 R2 0.635 0.150 0.329 0.294 Fully Private Banks Only Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Connected 0.3131* 0.0079** 0.0284*** 2 0.0050* (0.1600) (0.0031) (0.0108) (0.0026) Number of observations 2,845 2,819 2,845 1,016 R2 0.611 0.145 0.324 0.239 *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Number in parentheses are robust standard errors. All dependent variables are in logs. Ratios that can take negative values are measured as the log of one plus the corresponding ratio. Connected is a dummy variable that takes a value of 1 if a bank has at least one former politician among its board members and 0 otherwise. All regressions included a country �xed effect, and all regressions except for total assets also control for (log) total assets. Source: Authors’ analysis based on data described in the text. a measure of size); uc is a country �xed-effect that controls for cross-country differences in bank characteristics, and 1i,c is a residual term. Since these regressions exploit only within-country differences between connected and unconnected banks, and bank-level data are notoriously noisy, all variables are measured in logarithms to reduce the influence of outliers (variables corre- sponding to ratios that can plausibly take negative values are expressed as the logarithm of one plus the variable).10 As in table 3, the parameters of the benchmark model are estimated separately for all banks and for banks with no public ownership. The coef�cients con�rm that connected banks tend to be the largest banks in a country, with total assets about 34 percent larger than those of the average unconnected bank (see table 4, column 1). Similar results are obtained for other measures of size, such as loans and country ranking (not reported). Connected banks also tend to be more pro�table and to have a return on average assets 0.6–0.8 percent higher than the average unconnected bank (column 2). Leverage is signi�cantly lower among connected banks; the ratio of equity to total assets is 2 percent higher in connected banks than in the average bank, a difference that increases to 3 percent in the sample of privately owned banks (column 3). Connected banks also tend to have a lower 10. This is not a major issue in the overall comparisons in table 3, which compute the average of each characteristic across all connected and all unconnected banks. In contrast, these regressions compare connected and unconnected banks within a country. ´as Braun and Claudio Raddatz Matı 253 proportion of write-offs and impaired loans relative to average gross loans and reserves, suggesting that they take on less risk (column 4). Overall, the results across and within countries show that connected banks are larger, more pro�table, less leveraged, and less risky than unconnected banks, regardless of whether there is any government ownership.11 In addition, to see whether bank characteristics are correlated with the share of former poli- ticians on a bank’s board, equation (2) was reestimated using that share (a measure of the intensity of banks’ political connections) instead of the dummy variable described previously. While the results are similar to those reported in Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 table 4, they are weaker in statistical and economic terms (not reported). Thus, desirable bank characteristics are more strongly correlated with whether a bank has a former politician on its board than with the number of former poli- ticians. It does not seem that politicians cluster in banks with desirable characteristics. Country Characteristics As discussed in section I, a simple look at the data suggested that banks were less politically connected in richer, more �nancially developed countries. The results reported here systematically test whether the degree of connectedness of banks is robustly correlated with important country characteristics and whether those correlations survive when controlling for several straightforward omitted variables in a multivariate setting. Country characteristics, such as development level, institutional quality, extent of pro-banker regulation, and banking sector development, were related to the �ve measures of connectedness by estimating the parameters of the following speci�cation: ð3Þ Yc ¼ a þ b  CONNECTEDNESSc þ g 0 Xc þ 1c where Y is a measure of any of the country characteristics described above for country c, and CONNECTEDNESS is any of the �ve measures of connected- ness discussed in section I: the fraction of connected banks (FRACBANKS), the share of assets of connected banks (SHAREASSETS), the fraction of connected bankers (FRACBANKERS), the (log) of actual to expected number of matches of bankers-politicians (PREVALENCE), and the maximum share of the popu- lation from which bankers and politicians are selected so that the null of random selection cannot be rejected at a 5 percent level of signi�cance (MAXSHARE). The variables in X control for other country characteristics that may be simultaneously related to both Y and CONNECTEDNESS. 11. These �ndings are robust to using the standard Heckman (1979) two-step estimator to control for possible sample selection issues in the set of banks with information on directors (not reported). 254 THE WORLD BANK ECONOMIC REVIEW Economic Development The results show a strong negative correlation between the degree of connect- edness and GDP per capita, whether considering all banks or only those that are fully private (table 5). The correlation is particularly strong when no T A B L E 5 . Connectedness and Development Controls: Log Population, Log Fraction of Population with Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 No Controls Tertiary Education Number of Number of Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) All Bankscope Banks FRACBANKS 2 2.044*** 79 0.136 2 0.450 78 0.794 (0.721) (0.376) SHAREASSETS 2 0.994** 76 0.047 0.146 75 0.788 (0.479) (0.250) FRACPOLITICIANS 2 23.35*** 79 0.192 2 5.644 78 0.796 (6.435) (3.742) PREVALENCE 2 0.481*** 79 0.383 2 0.157** 78 0.805 (0.0588) (0.0600) MAXSHARE 0.163*** 79 0.184 0.0319* 78 0.795 (0.0214) (0.0179) Fully Private Banks Only FRACBANKS 2 2.673*** 64 0.215 2 0.848* 63 0.814 (0.678) (0.433) SHAREASSETS 2 1.425*** 61 0.061 0.167 60 0.796 (0.490) (0.271) FRACPOLITICIANS 2 20.72*** 64 0.26 2 8.004*** 63 0.827 (3.230) (2.195) PREVALENCE 2 0.534*** 64 0.436 2 0.203*** 63 0.829 (0.0530) (0.0717) MAXSHARE 0.197*** 63 0.153 0.0562** 62 0.801 (0.0340) (0.0266) *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: The dependent variable is the average 1995– 2005 log real GDP per capita (from Heston, Summers, and Aten 2006). Standard errors (in parentheses) are robust to heteroskedasti- city. FRACBANKS is the fraction of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. MAXSHARE is the largest fraction of a country’s population from which politicians and bankers would have to be selected so that the hypothesis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. Source: Authors’ analysis based on data described in the text and table. ´as Braun and Claudio Raddatz Matı 255 F I G U R E 1. Connectedness and Development Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Note: Figure shows the relation between (log) average real 1995– 2005 GDP per capita (from Heston, Summers, and Aten 2006) and (log) ratio of actual to expected number of matches between bankers and politicians (PREVALENCE), controlling for (log) fraction of population over age 25 with a tertiary education and log population. The displayed coef�cient is the value for the PREVALENCE measure of connectedness in the multivariate regression against log real GDP per capita. Country observations are labeled according to the World Bank’s codi�cation system (see Table 1.) Source: Authors’ analysis based on data described in the text. additional controls are included (columns 1–3), but it survives after controlling for log population and for the fraction of the population with tertiary edu- cation (columns 4 –6), especially when focusing on fully private banks. Educational attainment is particularly relevant as a standard measure of a country’s stock of human capital (which most theories relate to a country’s per capita GDP), but it is also important as a proxy for the size of the pool of elite from which politicians and bankers are selected (see section I). The results are statistically stronger for the more complex measures of connectedness: PREVALENCE and MAXSHARE. This suggests that these measures have greater economic content than the simpler ones. Nonetheless, results are quali- tatively similar, whatever the measure. Furthermore, �gure 1 shows that the negative correlation between connected- ness and development is not driven by a few outliers but reflects a robust pattern of the data. The relation between connectedness and GDP per capita (from Heston, Summers, and Aten 2006) is economically large. For instance, the differ- ence in (log) GDP per capita between Morocco and France is commensurate with their difference in PREVALENCE. Although causality cannot be attributed 256 THE WORLD BANK ECONOMIC REVIEW to this strong cross-country correlation without a good instrument, it is clear that the degree of connectedness is not neutral but is associated with a country’s level of development. The regressions discussed below show that connectedness is also associated with other country characteristics that have been causally related in the literature to level of development, even after controlling for the direct link between development and connectedness documented here. Institutions Correlating the �ve measures of connectedness with cross-country measures of Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 institutional quality (from Kaufmann, Kraay, and Mastruzzi 2004) shows that connectedness is signi�cantly higher in countries with less developed insti- tutions for preventing corruption and limiting the power of the government over its citizens (voice and accountability; table 6). The relation holds for all banks and for private banks. For all measures, the relation between connected- ness and institutional quality is signi�cantly negative even after controlling for GDP per capita and population size (columns 4 –6 and 10 –12). This is reassur- ing because of the widely documented link between institutions and develop- ment and because the measures of connectedness may be correlated with population size. With these correlations, it is not surprising that the estimated coef�cient changes according to the unconditional speci�cation. However, all the coef�cients maintain their sign and statistical signi�cance, which shows that the relations between connectedness and a country’s level of development and population size are not qualitatively driving the �ndings.12 It is even clearer than for the case of overall development that a few outliers do not drive the relations between connectedness and institutional quality (�gure 2). The magnitude of the estimated coef�cient is also economically rel- evant: a one standard deviation increase in SHAREASSETS (equivalent to the difference between Luxembourg and Philippines) is associated with a 0.4 decline (around half a standard deviation) in the control of corruption, corre- sponding to 25 percent of the difference in control of corruption between the two countries. Also, as shown in the bottom panel of �gure 2, the difference in control of corruption between Angola and Spain is commensurate with their difference in PREVALENCE. Regulation The results so far have shown that prevalence is systematically related to under- development and weak institutions. The next test is for a systematic relation between connectedness and banking sector regulation. As discussed in the introduction, the political economy literature typically associates the links 12. To capture nonlinearities, speci�cations were also estimated that included a quadratic term for log GDP per capita in addition to log GDP per capita and log population. The results, available on request, are qualitatively and quantitatively similar to those obtained with the baseline control set. This exercise was repeated for all the regressions reported in this section, with similar results. T A B L E 6 . Connectedness and Institutions Control of Corruptiona Voice and Accountabilitya No Controls Controls: Log Real GDP, Log Population No Controls Controls: Log Real GDP, Log Population Number of Number of Number of Number of Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) All Bankscope Banks FRACBANKS 2 2.377*** 79 0.21 2 1.230*** 79 0.72 2 2.168*** 79 0.23 2 1.264*** 79 0.58 (0.435) (0.371) (0.556) (0.440) SHAREASSETS 2 1.575*** 76 0.15 2 1.012*** 76 0.73 2 1.539*** 76 0.17 2 1.076*** 76 0.62 (0.379) (0.285) (0.368) (0.304) FRACPOLITICIANS 2 25.19*** 79 0.26 2 13.30*** 79 0.72 2 22.42*** 79 0.27 2 13.19*** 79 0.58 (3.691) (3.897) (4.308) (3.962) PREVALENCE 2 0.473*** 79 0.43 2 0.263*** 79 0.73 2 0.393*** 79 0.38 2 0.330*** 79 0.63 (0.0575) (0.0636) (0.0491) (0.0718) MAXSHARE 0.174*** 79 0.24 0.0613*** 79 0.71 0.143*** 79 0.21 0.0644*** 79 0.56 (0.0242) (0.0201) (0.0203) (0.0222) Fully Private Banks Only FRACBANKS 2 2.317*** 64 0.20 2 0.573 64 0.73 2 2.335*** 64 0.28 2 1.005 64 0.56 (0.548) (0.530) (0.712) (0.746) SHAREASSETS 2 1.691*** 61 0.12 2 0.790*** 61 0.74 2 1.734*** 61 0.15 2 0.926** 61 0.59 (0.404) (0.283) (0.458) (0.394) FRACPOLITICIANS 2 15.60*** 64 0.19 2 0.782 64 0.72 2 14.51*** 64 0.22 2 2.784 64 0.54 (3.943) (3.699) (4.448) (4.916) Matı PREVALENCE 2 0.474*** 64 0.43 2 0.146** 64 0.74 2 0.367*** 64 0.35 2 0.250*** 64 0.60 (0.0543) (0.0673) (0.0555) (0.0814) MAXSHARE 0.207*** 63 0.21 0.0717** 63 0.73 0.171*** 63 0.19 0.103*** 63 0.58 (0.0377) (0.0290) (0.0265) (0.0313) **Signi�cant at the 5 percent level; ***Signi�cant at the 5 percent level FRACBANKS is the fraction of banks with Bankscope data on board of direc- tors that had a former politician on their boards. Note: Standard errors (in parentheses) are robust to heteroskedasticity. FRACBANKS is the fraction of banks with Bankscope data on board of direc- tors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. MAXSHARE is the largest fraction of a ´as Braun and Claudio Raddatz country’s population from which politicians and bankers would have to be selected so that the hypothesis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. a. Average for 1996– 2002, from Kaufmann, Kraay, and Mastruzzi (2004). 257 Source: Authors’ analysis based on data described in the text and table. Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 258 THE WORLD BANK ECONOMIC REVIEW F I G U R E 2. Connectedness and Institutions Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Note: Figures show the relation between control of corruption (average 1996– 2002 from Kaufmann, Kraay, and Mastruzzi 2004) and the fraction of total banking system assets owned by connected banks (SHAREASSETS; top panel) and the (log) ratio of actual to expected number of matches between bankers and politicians (PREVALENCE; bottom panel), controlling for (log) real GDP per capita (adjusted for purchasing power parity) and log population. The displayed coef�cients are the values for the two connectedness measures in the multivariate regression against control of corruption. Source: Authors’ analysis based on data described in the text. ´as Braun and Claudio Raddatz Matı 259 between regulators and regulated �rms with private interests that depend criti- cally on both parties having something to gain from colluding. Regulation that favors incumbents in the banking system is the obvious service that politicians can exchange for a seat on a bank’s board. Barth, Caprio, and Levine (2003) use �ve dimensions of �nancial regulation to show how countries regulate their �nancial systems: restrictions on bank activities, entry regulation, supervisory powers, private monitoring and self- regulation, and capital requirements. They assign an index to each of these broad ways of regulating banks that corresponds to the �rst principal com- Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 ponent of the answers to surveys conducted by regulators in each country. To address the ambiguity inherent in some of these dimensions, these indexes were used to construct an overall measure of the pro-banker leaning of �nancial regulations across countries. For instance, it is unclear whether restric- tions on bank activities are pro- or anti-incumbents. On the one hand, restric- tions constrain the ability of banking incumbents to expand into new lines of business. On the other hand, restrictions constrain other institutions from expanding into the banking business. Similarly, whether giving responsibility for supervision and monitoring to the public or private sector is pro- or anti- bankers depends on what type of monitors are more easily captured. Instead of taking an arbitrary stance on whether each of these �ve dimensions of �nancial regulation is pro- or anti-banking incumbents, cross-country data on the degree of rents in a country’s banking sector (measured as the average net interest margin, also from Barth, Caprio, and Levine 2003) are used to build a de facto index by regressing these rents on the �ve individual indexes. Burnside and Dollar’s (2000) methodology was used to construct an index of the pro- banker intensity of regulation by weighting each index by its estimated elasticity to rents. The intention is to let the data speak: if a given dimension of regulation is more pro-banker, an increase in its index should be associated with higher rents (and vice-versa). The regression yields the following result NIM ¼ :30 ÂENT À :32 ÂCAP þ :51 ÂACT À 1:0 ÂPRIV À :05 ÂOSP; ð:20Þ ð:31Þ ð:33Þ ð:38Þ ð:24Þ ð4Þ 2 R ¼ 0:28 where NIM is a country’s banking sector average net interest margin, and ENT, CAP, ACT, PRIV, and OSP are the �ve principal component indexes as described above: entry restrictions, capital requirements, activities restrictions, private monitoring, and overall supervisory power (all standardized to have zero mean and unit variance so that the magnitude of the coef�cients reveal the rela- tive importance of each dimension). According to the regression, average net interest margins are positively correlated with restrictions on entry and activity and negatively correlated with capital requirements, the extent of private moni- toring, and the power of the supervisor. In terms of magnitude and signi�cance, private monitoring has the largest correlation with margins, followed by 260 THE WORLD BANK ECONOMIC REVIEW restrictions on activities, capital requirements, and entry restrictions. Surprisingly, the index of supervisory power has a negligible correlation with margins, in both magnitude and signi�cance. In addition to this index, the Kaufmann, Kraay, and Mastruzzi (2004) index of regulatory quality was also used (the index measures the incidence of market-unfriendly policies such as price controls or inadequate bank supervi- sion), and the correlation between connectedness and each of the �ve individ- ual dimensions of regulation was checked (see table A1 in the appendix). Table 7 shows the relation between the measures of connectedness and the Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 index of pro-banker regulation (columns 1 –6) and the index of overall regulat- ory quality (columns 7–12) for all banks and for private banks only, both unconditionally and after controlling for log real GDP per capita and log popu- lation. With a few exceptions, there is a positive relation between connected- ness, however measured, and the index of pro-banker regulation. There is also a strong negative correlation between connectedness and the index of regulat- ory quality (the correlations with MAXSHARE have the opposite sign, as expected). The results are especially strong when connectedness is measured among private banks only, demonstrating again that politicians sitting on boards in public banks do not drive the �ndings. Again, the economic magni- tude of the effect is large: moving from the 10th to the 90th percentile of PREVALENCE is associated with a one standard deviation increase in the index of pro-bank regulation, an increase roughly commensurate with the difference between the index in Lithuania and Spain. Similarly, the same increase in PREVALENCE is associated with more than a one standard devi- ation decline in the index of regulatory quality, commensurate with the differ- ence between Egypt and Japan. The correlations with the regulatory index are not driven by a few outliers (�gure 3), although the relation is not as strong as that with country character- istics. This is due partly to the smaller sample for regulatory variables, but also to the dif�culty of aggregating the indicators into a measure of pro-banker regulation. To check the robustness of the results, the pro-banker index was also built using the simpler indexes reported by Barth, Caprio, and Levine (2003) for each dimension of regulation instead of the principal component indexes. The results are qualitatively similar, but signi�cance is lost in several cases. Finally, the results were checked using data from Barth, Caprio, and Levine (2006) to construct a simple index (rather than a principal components index) based on surveys in 2001 and 2003, which increases the cross-sectional dimension of the data. As before, the results are qualitatively similar, but the signi�cance is lost except in the unconditional regressions and the conditional regressions using MAXSHARE and PREVALENCE.13 13. Results are available on request. T A B L E 7 . Connectedness and Regulation Pro-Banker Regulation Indexa Regulatory Qualityb Controls: Log Real GDP, Log Controls: Log Real GDP, Log No Controls Population No Controls Population Number of Number of Number of Number of Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) All Bankscope Banks FRACBANKS 5.055*** 51 0.25 1.733 51 0.49 2 2.175*** 79 0.29 2 1.401*** 79 0.68 (1.456) (2.142) (0.456) (0.362) SHAREASSETS 3.818*** 48 0.26 2.360** 48 0.57 2 1.593*** 76 0.24 2 1.190*** 76 0.70 (0.888) (0.963) (0.339) (0.332) FRACPOLITICIANS 54.51*** 51 0.33 28.71 51 0.52 2 23.82*** 79 0.38 2 17.35*** 79 0.72 Matı (18.31) (25.83) (3.833) (3.721) PREVALENCE 0.491*** 51 0.25 0.362** 51 0.53 2 0.349*** 79 0.38 2 0.241*** 79 0.67 (0.0968) (0.170) (0.0475) (0.0739) MAXSHARE 2 0.216*** 51 0.14 2 0.0709 51 0.49 0.116*** 79 0.18 0.0352** 79 0.61 (0.0501) (0.0435) (0.0178) (0.0157) Fully Private Banks Only FRACBANKS 4.444** 46 0.20 1.877 46 0.50 2 2.170*** 64 0.28 2 1.004** 64 0.64 (1.729) (1.586) (0.485) (0.501) SHAREASSETS 4.561*** 43 0.21 3.497*** 43 0.60 2 1.716*** 61 0.19 2 1.048** 61 0.65 (1.463) (1.219) (0.418) (0.469) FRACPOLITICIANS 55.28*** 46 0.31 39.28** 46 0.55 2 14.65*** 64 0.26 2 4.803 64 0.62 ´as Braun and Claudio Raddatz (14.28) (18.33) (3.581) (3.765) (Continued ) 261 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 262 TABLE 7. Continued Pro-Banker Regulation Indexa Regulatory Qualityb Controls: Log Real GDP, Log Controls: Log Real GDP, Log No Controls Population No Controls Population Number of Number of Number of Number of Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) PREVALENCE 0.612*** 46 0.32 0.389** 46 0.54 2 0.360*** 64 0.40 2 0.178** 64 0.64 (0.108) (0.170) (0.0437) (0.0787) THE WORLD BANK ECONOMIC REVIEW MAXSHARE 2 0.267*** 46 0.16 2 0.0997* 46 0.50 0.135*** 63 0.14 0.0407 63 0.60 (0.0555) (0.0511) (0.0286) (0.0268) *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: A higher number indicates more pro-banker regulation or better regulatory quality. Standard errors (in parentheses) are robust to heteroskedasti- city. FRACBANKS is the fraction of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. MAXSHARE is the largest fraction of a country’s population from which politicians and bankers would have to be selected so that the hypothesis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. a. Index built from Barth, Caprio, and Levine (2003) using the methodology of Burnside and Dollar (2000). b. Average for 1996– 2002 from Kaufmann, Kraay, and Mastruzzi (2004). Source: Authors’ analysis based on data described in the text and table. Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 ´as Braun and Claudio Raddatz Matı 263 F I G U R E 3. Connectedness and Pro-Banker Regulation Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Note: Figures show the relation between the index of pro-banker regulation and the fraction of total banking system assets owned by connected banks (SHAREASSETS; top panel) and the (log) ratio of actual to expected number of matches between bankers and politicians (PREVALENCE; bottom panel), controlling for (log) real GDP per capita (adjusted for purchasing power parity) and log population. The displayed coef�cients are values for the two connectedness measures in the multivariate regression against pro-banker regulation. Source: Authors’ analysis based on data described in the text. 264 THE WORLD BANK ECONOMIC REVIEW Financial Development The evidence presented above suggests that the connectedness of bankers and politicians is signi�cantly and robustly correlated with how the banking sector operates and is regulated. Insofar as these differences have no impact on the ef�ciency of the �nancial system, the issue would simply be a matter of diverse preferences across countries. The importance rises, however, if the connection between bankers and politicians is correlated with the ability of the system to allocate funds ef�ciently. This section tests whether connectedness is related to the degree of development of the banking system. The speci�cation is the same Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 as above, with Y now being each country’s log ratio of bank credit to the private sector to GDP. Also as before, univariate and multivariate regressions are presented that control for per capita GDP and population size and for other standard determinants of �nancial development. The coef�cient of all measures of connectedness is negative (except, of course, for MAXSHARE, which is an inverse measure of connectedness) and almost always signi�cant in univariate and multivariate regressions regardless of whether connectedness is measured over all banks or private banks only (table 8). In fact, as before, the results are stronger when connectedness is measured over private banks only. Thus, connectedness is associated with a lower degree of banking sector development. The relation is large in economic terms: moving from the 10th to the 90th percentile of prevalence is associated with a ratio of private credit to GDP 45 percentage points higher, an increase roughly commensurate with the difference between Philippines and Japan. Again, a few outliers do not drive the results (�gure 4). The negative correlation between the measures of connectedness and �nan- cial development is not driven by the traditional measures used to explain �nancial development across countries, such as the degree of protection of creditors, the quality of accounting practices, and investment opportunities measured using the decade’s effective GDP growth rate (columns 7–9).14 Both creditor rights and accounting quality enter positively as expected (although not signi�cantly). Robustness The results have shown that the connectedness of banks, however measured, is negatively correlated with economic development, the existence of less corrupt and more accountable institutions, and development of the banking sector and is positively correlated with the extent to which regulation favors bank incum- bents. As mentioned, without a good instrument for connectedness, causal inferences cannot be made, but it has been shown that these reduced form relations are not trivially driven by some obvious third variables that may be 14. When the decadal growth rate of per capita GDP is included, log real per capita GDP is dropped. T A B L E 8 . Connectedness and Financial Development Controls: Log Population, Creditor Controls: Log Population, Log Real Rights, Accounting Standards, per No Controls GDP per Capita Capita GDP Growth Number of Number of Number of Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) All Bankscope Banks FRACBANKS 2 2.905*** 70 0.276 2 0.844 70 0.63 2 3.275*** 59 0.382 (0.512) (0.526) (0.575) SHAREASSETS 2 2.189*** 67 0.219 2 1.039** 67 0.65 2 1.961*** 56 0.333 (0.381) (0.404) (0.544) FRACPOLITICIANS 2 33.95*** 70 0.419 2 15.13** 70 0.657 2 34.57*** 59 0.421 Matı (5.164) (6.581) (7.436) PREVALENCE 2 0.412*** 70 0.268 2 0.229** 70 0.651 2 0.466*** 59 0.413 (0.0703) (0.0870) (0.0849) MAXSHARE 0.150*** 70 0.128 0.0358 70 0.621 0.125*** 59 0.262 (0.0292) (0.0226) (0.0288) Fully Private Banks Only FRACBANKS 2 2.594*** 58 0.301 2 0.857* 58 0.69 2 2.958*** 49 0.528 (0.466) (0.441) (0.437) SHAREASSETS 2 2.216*** 55 0.175 2 1.155*** 55 0.705 2 1.677** 46 0.319 FRACPOLITICIANS 2 21.18*** 58 0.408 2 9.147*** 58 0.715 2 23.94*** 49 0.558 ´as Braun and Claudio Raddatz (3.467) (3.323) (4.027) (Continued ) 265 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 266 TABLE 8. Continued Controls: Log Population, Creditor Controls: Log Population, Log Real Rights, Accounting Standards, per No Controls GDP per Capita Capita GDP Growth Number of Number of Number of Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) PREVALENCE 2 0.441*** 58 0.327 2 0.193** 58 0.701 2 0.416*** 49 0.506 (0.0744) (0.0813) (0.0910) MAXSHARE 0.195*** 58 0.138 0.0617** 58 0.681 0.154*** 49 0.344 THE WORLD BANK ECONOMIC REVIEW (0.0428) (0.0306) (0.0433) *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Standard errors (in parentheses) are robust to heteroskedasticity. The dependent variable is the (log) ratio of average 1995– 2005 private credit to GDP (from Beck, Demirguc-Kunt, and Levine 2000). FRACBANKS is the fraction of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. MAXSHARE is the largest fraction of a country’s popu- lation from which politicians and bankers would have to be selected so that the hypothesis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. Source: Authors’ analysis based on data described in the text and table. Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 ´as Braun and Claudio Raddatz Matı 267 F I G U R E 4. Connectedness and Financial Development Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Note: The �gures show the relation between the ratio of average 1995– 2005 private credit to GDP and the fraction of total banking system assets owned by connected banks (SHAREASSETS; top panel) and the (log) ratio of actual to expected number of matches between bankers and politicians (PREVALENCE; bottom panel), controlling for (log) real GDP per capita (adjusted for purchasing power parity) and log population. The displayed coef�cients are the values for the two connectedness measures in the multivariate regression against private credit to GDP. Source: Authors’ analysis based on data described in the text. 268 THE WORLD BANK ECONOMIC REVIEW simultaneously related to the connectedness measures and any of the country characteristics analyzed, such as a overall development or population size. The regressions reported here address some further robustness concerns. As discussed in section I, although the PREVALENCE measure takes no stance on the share of the population from which bankers and politicians are selected, it assumes that the share is constant across countries. This is a reason- able assumption, but it may be that the elite are not a �xed share of the popu- lation but rather a �xed number of people. If so, PREVALENCE, one of the most robust measures, could simply be capturing the relation between cross- Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 country differences in the size of the elite as a fraction of the population over several country characteristics. This is partially controlled by including the log population in the speci�cations, which does not eliminate the �ndings of the unconditional regressions. Nevertheless, it is also possible that the size of the elite is not �xed but proportional to the share of the highly educated popu- lation. To check for this possibility, the log share of the population with ter- tiary education was added to each speci�cation.15 The regressions show that differences in the size of the elite as a fraction of the population do not drive the documented negative correlation of connectedness with institutions and �nancial development or the positive correlation with pro-banker regulations. Although this is mainly a concern for the PREVALENCE measure, results are also reported using the share of assets of connected banks to show that control- ling for this additional variable does not change these results either. Results for other variables are similar and available on request. Another concern with the connectedness measures is that, empirically, they are negatively correlated with the number of banks reporting to Bankscope. This number is an endogenous variable that may clearly be correlated with banking sector development, but since the measures of connectedness may be mechanically related to this number by construction, the documented corre- lations could be spurious. To check for this possibility, the measures were recomputed using only the 10 largest banks in a country as measured by total assets at the end of 2004 (columns 4–6 in table 9). All banks were included for countries with fewer than 10 reporting banks. This reduced by two orders of magnitude the cross-country variance in the number of banks used in calcu- lating the measures of connectedness, and the resulting measures are not signi�- cantly correlated with the number of banks. Nevertheless, the results obtained with these measures are quantitatively and qualitatively similar to those obtained when all banks are included. Thus the signi�cantly larger number of banks reporting in richer and more developed countries is not behind the docu- mented correlations. 15. As shown in section I, PREVALENCE computed using the total population equals PREVALENCE computed using only the elite plus the log of elite share of the population. Assuming that this log share is proportional to the share of the population with tertiary education, true PREVALENCE would be PREVALENCE with all population less the fraction of the population with a tertiary education. T A B L E 9 . Robustness Exercises Controls: Tertiary Education, Computing Connectedness on 10 Dropping Countries with Fewer Using Robust Regression; Controls: Log Real GDP Per Log Real GDP Per Capita, Log Largest Banks only; Controls: than Two Matches; Controls: Controls: Log Real GDP Per Capita, log Population, and Population Log real GDP, Log Population Log Real GDP, Log Population Capita and Log Population Former Socialist Countries Dependent Number of Number of Number of Number of Number of Variable and Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) Dependent variable: control of corruptiona SHAREASSETS 2 0.988*** 75 0.73 2 0.956*** 65 0.71 2 1.066*** 52 0.72 2 1.119*** 76 0.72 2 0.828*** 73 0.76 (0.318) (0.283) (0.330) (0.287) (0.301) PREVALENCE 2 0.251*** 78 0.74 2 0.234** 65 0.68 2 0.335*** 52 0.73 2 0.271*** 79 0.70 2 0.243*** 76 0.77 (0.0661) (0.105) 0.711 (0.0890) (0.0767) (0.0643) Dependent variable: pro-banker regulation indexb SHAREASSETS 2.193** 48 0.57 2.081* 39 0.52 2.725** 36 0.55 2.427*** 48 0.52 1.779** 48 0.62 (0.952) (1.073) (1.024) (0.896) (0.871) PREVALENCE 0.347** 51 0.55 0.599* 39 0.50 0.634** 36 0.53 0.322* 51 0.47 0.256* 51 0.59 (0.164) (0.300) (0.234) (0.174) (0.149) Dependent variable: �nancial developmentc SHAREASSETS 2 1.035** 67 0.65 2 1.082** 56 0.62 2 1.347** 44 0.67 2 1.122*** 67 0.62 2 0.697* 64 0.72 Matı (0.408) (0.423) (0.500) (0.419) (0.353) PREVALENCE 2 0.228** 70 0.65 2 0.425*** 56 0.64 2 0.311** 44 0.64 2 0.216** 70 0.62 2 0.182)** 67 0.73 (0.0870) (0.125) (0.134) (0.0947) (0.0792) *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Standard errors (in parentheses) are robust to heteroskedasticity. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. a. Average for 1996– 2002, from Kauffman, Kraay, and Mastruzzi (2004). b. Index built from Barth, Caprio, and Levine (2003) using the methodology of Burnside and Dollar (2000). ´as Braun and Claudio Raddatz c. Log ratio of average 1996– 2002 private credit to GDP (from Beck, Demirguc-Kunt, and Levine 2000). Source: Authors’ analysis based on data described in the text. 269 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 270 THE WORLD BANK ECONOMIC REVIEW The analysis was restricted to countries with at least one match, but it could be argued that that was not restrictive enough and that the �nding of one or two matches might be an overinterpretation. To check this, the analysis was restricted to countries with at least two matches (making two matches the base- line). The results follow the same pattern as before (columns 7 –9), indicating that countries with more than one match drive the correlations. Further restricting the sample to include only countries with at least three matches (31 countries) yields qualitatively similar results, but some of the coef�cients are not signi�cant at a 10 percent level because of the reduction in the sample size Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 (31 countries; not reported). The regressions reported in columns 10–15 check for the influence of out- liers on the results. Columns 10–12 take an agnostic approach and simply use a robust regression technique to reduce the influence of outliers.16 As before, there is no important change in the results. Columns 13 –15 control for the potential influence of socialist countries. Although �gures 1 –4 and the regressions reported in columns 10–12 show that a few countries do not drive the correlations, they also show that the group of formerly socialist countries tends to be at the extreme of the distribution of connectedness. Thus, the corre- lations reported may come from the difference between former socialist countries and the rest of the sample. To check for this without unnecessarily reducing the sample, a dummy variable was added that takes a value of 1 for formerly socialist countries and 0 otherwise. Reassuringly, the sign and magni- tude of all the reduced-form coef�cients remain unaffected (the dummy for for- merly socialist countries is typically signi�cant and in the expected direction, for example, with lower �nancial development). Finally, because the quality of the information in many countries with zero matches cannot be trusted and the �nding of a zero match provides very little information on the process driving the selection of bankers and politicians, the analyses were conducted again after dropping countries with zero matches. Countries with zero matches are very heterogeneous, and there is no good way of separating the zeroes resulting from data quality from the true zeroes. While this seems to be the right approach, it would be troubling if the pattern of results changed qualitatively or was even reversed when the zeroes were con- sidered. That was not the case (table 10). As a mild way of separating zeroes resulting from poor data from true zeroes, only the countries with zero matches and more than two banks were included. The unconditional regressions always result in signi�cant coef�cients of the same sign as those reported previously, and the regressions controlling for log real GDP per capita and log population size also show a similar pattern to those previously reported. The only major difference is that the coef�cients for the degree of pro-banker regulation are no longer statistically signi�cant for any measure. This is not so surprising considering that the relation with 16. Stata command rreg. T A B L E 1 0 . Robustness Exercise Including Countries with Zero Matches Control of Corruptiona Pro-banker Regulation Indexb Financial Developmentc Number of Number of Number of Measure Coef�cient (1) Observations (2) R2 (3) Coef�cient (4) Observations (5) R2 (6) Coef�cient (7) Observations (8) R2 (9) FRACBANKS 2 1.429*** 131 0.08 2.806** 74 0.07 2 1.788*** 110 0.10 (0.256) (1.345) (0.407) SHAREASSETS 2 0.819** 126 0.04 2.051** 71 0.07 2 1.247*** 107 0.07 (0.317) (0.928) (0.362) FRACBANKERS 2 12.56*** 131 0.10 33.95** 74 0.11 2 22.51*** 110 0.17 (2.448) (13.11) (4.278) PREVALENCE 2 0.301*** 130 0.09 0.240* 73 0.03 2 0.211** 109 0.04 (0.0585) (0.124) (0.0929) FRACBANKS 2 0.633** 126 0.64 0.0525 72 0.54 2 0.734* 108 0.58 (0.300) (1.380) (0.404) SHAREASSETS 2 0.552** 123 0.64 1.207 69 0.56 2 0.815** 105 0.59 (0.254) (0.976) (0.327) FRACBANKERS 2 6.257** 126 0.64 5.338 72 0.54 2 10.12** 108 0.60 (2.956) (13.92) (4.425) PREVALENCE 2 0.118* 126 0.64 0.0550 72 0.54 2 0.145 108 0.58 Matı (0.0639) (0.158) (0.0891) *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Standard errors (in parentheses) are robust to heteroskedasticity. All regressions include the observations with zero matches between bankers and politicians in countries with more than two banks, and all regressions control for log real GDP per capita and log population. FRACBANKS is the fraction of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. a. Index of control of corruption, average 1996– 2002, from Kauffman, Kraay, and Mastruzzi (2004). ´as Braun and Claudio Raddatz b. Built by the authors using data from Barth and others (2003). c. Log ratio of average 1996– 2002. Private credit to GDP (from Beck, Demirguc-Kunt, and Levine 2000). Source: Authors’ analysis based on data described in the text. 271 Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 272 THE WORLD BANK ECONOMIC REVIEW regulation is the most dif�cult to pin down and was the weakest among those reported in the baseline results. Including many diverse countries with the same value of connectedness (zero) clearly reduces the variance of the explana- tory variable and its ability to account for this country characteristic. III. CONCLUDING REMARKS This article builds an extensive dataset to measure the extent to which banks are politically connected across countries. The measure is based on the fact Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 that some high-ranking politicians end up on bank boards of directors. Of course, this represents just one way of establishing relationships between bankers and politicians. It may not even be the most important one, but it is likely to be correlated with other forms. Although formal tests are not pre- sented and causality is not established, the article presents pieces of reduced- form evidence that hold together better as a private interest story than as a public interest story. First, connected banks do better than unconnected ones: they are larger and more pro�table, and these characteristics are not related to higher risk taking. These results are consistent with those for nonbank �rms documented in the political economy literature. While a public interest view is still possible (say if politicians were attracted to good banks), in that case poli- ticians would be expected to cluster in the best banks, which should result in a strong relation between the share of politicians on a bank’s board and the bank’s performance. But no such relationship was found: once a bank is con- nected, having more politicians on the board is not associated with better performance. Second, connectedness is more prevalent where deals between bankers and politicians are likely to be less costly and more influential. Connectedness corre- lates positively with corruption but negatively with government accountability. Third, these politician-banker relationships are associated with poorer out- comes for society in the form of lower overall and �nancial development. A likely mechanism is regulatory capture, a conjecture supported by the �nding that bank regulation is more pro-banker and of lower quality where these links are more important. If that is the direction of causality, a permissive insti- tutional context allows banks to achieve better regulatory treatment by con- necting themselves to politicians. These links allow banks to achieve higher pro�ts without taking more risk or boosting ef�ciency, in the process incurring high social costs in the form of inhibited �nancial sector development and reduced access to �nancing for many �rms. Restricting these types of connec- tions could limit the ability of incumbent �nanciers to tilt regulations in their favor and impede �nancial sector development. It is important, however, not to draw direct, partial equilibrium policy conclusions from this exercise. If this particular avenue of connection is absent, incumbents might instead pressure regulators some other way, such as through outright bribes, that could be even more detrimental to the institutional framework. ´as Braun and Claudio Raddatz Matı 273 REFERENCES Acemoglu, Daron, and Simon Johnson. 2005. “Unbundling Institutions.� Journal of Political Economy 113 (5): 949–95. Agrawal, Anup, and Charles Knoeber. 2001. “Do some Outside Directors Play a political Role?� Journal of Law and Economics 44: 179– 98. Ansolabehere, Stephen, John de Figueiredo, and James Snyder, 2003. “Why Is There So Little Money in US Politics?� Journal of Economic Perspectives 17: 105 –30. Barth, James, Gerard Caprio, and Ross Levine. 2003. “Bank Supervision and Regulation: What Works Best?� Journal of Financial Intermediation 13: 205–48. Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 ———. 2006. Rethinking Bank Regulation: Till Angels Govern. Cambridge, U.K.: Cambridge University Press. Beck, Thorsten, Asli Demirguc-Kunt, and Ross Levine. 2000. “A New Database on Financial Development and Structure.� World Bank Economic Review 14 (3): 597–605. Updated and accessed through worldbank.org. Braun, Matias, and Claudio Raddatz. 2007. Trade Liberalization, Capital Account Liberalization, And The Real Effects Of Financial Development. Journal of International Money and Finance 26 (5): 730–61. ———. 2008. “The Politics of Financial Development: Evidence from Trade Liberalization.� Journal of Finance 63 (3): 1469–508. Bureau van Dijk. 2006. Bankscope. www.bvdinfo.com/Products/Company-Information/International/ BANKSCOPE.aspx Burnside, Craig, and David Dollar. 2000. “Aid, Policies, and Growth.� American Economic Review 90 (4): 847– 68. How Countries Supervise Their Banks, Insurers and Securities Markets 2000. London: Central Banking Publications, 1999. How Countries Supervise Their Banks, Insurers and Securities Markets 2002. London: Central Banking Publications, 2001. How Countries Supervise Their Banks, Insurers and Securities Markets 2003. London: Central Banking Publications, 2002. Cull, Robert, and Colin Xu. 2005. “Institutions, Ownership, and Finance: The Determinants of Pro�t Reinvestment among Chinese Firms.� Journal of Financial Economics 77: 117–46. ´ , Ernesto. 2006. “Regulatory Capture: A Review.� Oxford Review of Economic Policy 22 (2): Dal Bo 203–25. Dreher, Axel, Michael Lamla, Sarah Lein, and Frank Somogoyi. 2009. “The Impact of Political Leaders’ Profession and Education on Reforms.� Journal of Comparative Economics 37: 169– 93. Economist Intelligence Unit. Various years. Country Reports, 1996– 2005. London: Economist Intelligence Unit. http://countryanalysis.eiu.com/country_reports.html. Faccio, Mara. 2006. “Politically Connected Firms.� American Economic Review 96 (1): 369– 86. Faccio, Mara, Ronald Masulis, and John J. McConnell. 2005. “Political Connections and Corporate Bailouts.� Journal of Finance 61 (6): 2597– 635. Fisman, Raymond. 2001. “Estimating the Value of Political Connections.� American Economic Review 91 (4): 1095–102. ¨ rg Rocholl, and Jongil So. 2009. “Do Politically Connected Boards Affect Firm Goldman, Eitan, Jo Value?� Review of Financial Studies 22 (6): 2331–60 Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2004. “The Role of Social capital in Financial Development.� American Economic Review 94 (3): 526– 56. Haber, Stephen H. 1991. “Industrial Concentration and the Capital Markets: A Comparative Study of Brazil, Mexico, and the United States,1830–1930.� Journal of Economic History 51: 559– 80. 274 THE WORLD BANK ECONOMIC REVIEW Heckman, James. 1979. “Sample Selection Bias As a Speci�cation Error.� Econometrica 47: 153– 61. Heston, A., R. Summers, and B. Aten. 2006. Penn World Table Version 6.2. University of Pennsylvania, Center for International Comparisons of Production, Income, and Prices, Philadelphia, Pa. Available at http://pwt.econ.upenn.edu. IMF (International Monetary Fund). 2009. Special Data Dissemination Standard. http://dsbb.imf.org/ Pages/SDDS/Home.aspx. Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2004. “Governance Matters III: Governance Indicators for 1996, 1998, 2000, and 2002.� World Bank Economic Review 18 (2): 253– 87. Khwaja, Asim Ijaz, and Atif Mian. 2005. “Do Lenders Favor Politically Connected Firms: Rent Provision in an Emerging Financial Market.� Quarterly Journal of Economics 120 (4): 1371–411. Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 ´ pez-de-Silanes, Andrei Shleifer, and Robert Vishny. 1997. “Legal determi- La Porta, Rafael, Florencio Lo nants of external �nance.� Journal of Finance 52 (3): 1131–50. ———. 1998. “Law and �nance.� Journal of Political Economy 106 (6): 1113– 55. Levine, Ross. 2006. “Finance and Growth: Theory and Evidence.� In Handbook of Economic Growth, Vol. 1A., ed. Philippe Aghion, and Steven Durlauf. Amsterdam: North Holland. Maurer, Noel. 2002. The Power and the Money: The Mexican Financial System, 1876–1932. Palo Alto, CA: Stanford University Press. Maurer, Noel, and Andrei Gomberg. 2005. “When the State Is Untrustworthy: Public Finance and Private Banking in Por�rian, Mexico.� Journal of Economic History 64 (4): 1087– 107. Milanovic, Branko, Karla Hoff, and Shale Horowitz. Forthcoming. “Turnover in Power As a Restraint on Investing in Influence: Evidence from the Post-Communist Transition.� Economics and Politics. Pagano, Marco, and Paolo Volpin. 2001. “The Political Economy of Finance.� Oxford Review of Economic Policy 17 (4): 502– 519. ———. 2005. “The Political Economy of Corporate Governance.� American Economic Review 95 (4): 1005– 30. Rajan, Raghuram, and Luigi Zingales. 2003. “The Great Reversals: The Politics of Financial Development in the Twentieth Century.� Journal of Financial Economics 69 (1): 5–50. Razo, Armando. Forthcoming. Social Foundations of Limited Dictatorship: Networks and Private Protection during Mexico’s Early Industrialization. Palo Alto, CA: Stanford University Press. Schnell, Rainer, Tobias Bachteler, and Stefan Bender. 2004. “A Toolbox for Record Linkage.� Austrian Journal of Statistics 33 (1 –2):125– 33. Stulz, Rene, and Rohan Williamson. 2003. “Culture, Openness, and Finance.� Journal of Financial Economics 70: 313–49. APPENDIX. DISTRIBUTION OF THE NUMBER OF MATCHES UNDER R A N D O M D R AW S Consider a population with NP politicians and NB bankers. The intersection of the two groups consists of NPB banker-politicians. Two samples are taken con- secutively and matched from the population of bankers and politicians with replacement at the sample level,17 the �rst consisting of nB NB bankers and the second of nP NP politicians. Let X be a random variable that counts the 17. This means that all individuals from the �rst sample are replaced in the population before taking the second sample, so that an individual from the intersection of the two samples can be drawn twice. ´as Braun and Claudio Raddatz Matı 275 number of matches and is distributed according to:       NPB PNPB Àk NPB À k NB À NPB NP À k À i k i¼0 i nB À k À i nP À k PðX ¼ kÞ ¼    NP NB nP nB Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 The denominator corresponds to the ways the two samples of sizes nP and nB can be chosen from populations of sizes NP and NB. The numerator has various components. The �rst term corresponds to the number of ways in which the k common elements can be chosen among the NPB members of the intersection. The summation that follows counts the number of ways to choose the remaining nP 2 k and nB 2 k terms. The �rst term counts the ways to choose i of those elements from among the rest of the members of the intersection. If the i terms are chosen in this way, they can be in only one of the samples. For instance, assume that one of the remaining nB 2 k components of nB also   NPB À k belongs to NPB. This one term can be chosen in ways, and the 1 remaining nB 2 k terms, which are bankers only, can be chosen in   NB À NPB ways. Given that one of the terms in nB 2 k belongs to the inter- nB À k À 1 section, it cannot be selected in the remaining nP 2 k draws from NP, so those   NP À k À 1 terms can be chosen in only. nP À k This distribution is used to estimate the expected number of matches in a country considering the actual size of the samples of bankers and politicians available from the data, which pin down nP and nB and assuming that both are drawn from a common pool corresponding to a country’s total population. In the notation above, the assumption of a common pool corresponds to assuming that NP ¼ NB ¼ NPB. In this case the probability of �nding k matches simpli�es to     N NÀk N À nP k n Àk n Àk PðX ¼ kÞ ¼  P  b N N nP nB T A B L E A 1 . Connectedness and Detailed Regulation 276 Entry Requirements Capital Requirements Controls: Log Real GDP, Log Controls: Log Real GDP, Log No Controls Population No Controls Population Number of No. of No. of No. of Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Coef�cient Observations R2 Measure (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) All Banks FRACBANKS 2 0.334 52 0.002 2 2.293 52 0.06 2 0.842 52 0.017 0.00387 52 0.129 (1.059) (1.646) (0.725) (1.010) SHAREASSETS 0.0731 49 0 2 0.276 49 0.026 2 0.879 49 0.032 2 0.542 49 0.133 (0.689) (0.792) (0.609) (0.606) FRACBANKERS 3.187 52 0.002 2 12.06 52 0.033 2 8.900 52 0.021 2 1.190 52 0.13 THE WORLD BANK ECONOMIC REVIEW (10.05) (13.12) (5.792) (8.839) CONNECTEDNESS 0.0203 52 0.001 2 0.0795 52 0.026 2 0.222** 52 0.127 2 0.149 52 0.15 (0.0930) (0.187) (0.0847) (0.130) PREVALENCE 2 0.0273 52 0.003 2 0.00193 52 0.023 0.112*** 52 0.09 0.0650* 52 0.154 (0.0589) (0.0664) (0.0323) (0.0367) Fully Private Banks Only FRACBANKS 2 0.178 46 0.000 2 1.485 46 0.044 2 0.180 46 0.001 0.269 46 0.109 (1.171) (1.634) (0.766) (0.997) SHAREASSETS 0.106 43 0.000 2 0.379 43 0.027 2 0.358 43 0.004 2 0.380 43 0.103 (1.268) (1.409) (0.675) (0.702) FRACPOLITICIANS 8.438 46 0.009 0.414 46 0.026 2 6.641 46 0.014 2 7.130 46 0.115 (8.671) (14.16) (7.420) (10.38) CONNECTEDNESS 0.0331 46 0.001 2 0.0972 46 0.031 2 0.238** 46 0.148 2 0.194* 46 0.153 (0.121) (0.218) (0.0947) (0.114) MAXSHARE 2 0.0283 46 0.002 0.00911 46 0.026 0.128*** 46 0.109 0.0914** 46 0.153 (0.0683) (0.0791) (0.0355) (0.0420) Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 Measure Activities Restrictions Private Monitoring No Controls Controls: Log Real GDP, Log No Controls Controls: Log Real GDP, Log Population Population Coef�cient No. of R2 Coef�cient No. of R2 Coef�cient No. of R2 Coef�cient No. of R2 Observations Observations Observations Observations (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) All Banks FRACBANKS 2.812*** 51 0.153 2.655** 51 0.271 2 3.227*** 52 0.226 2 0.994 52 0.391 (0.747) (1.072) (1.018) (1.470) SHAREASSETS 1.211 48 0.05 0.665 48 0.23 2 2.645*** 49 0.265 2 1.724** 49 0.478 (0.751) (0.842) (0.602) (0.662) FRACBANKERS 25.04*** 51 0.137 24.68** 51 0.258 2 35.31*** 52 0.301 2 17.65 52 0.414 (8.104) (11.83) (12.58) (17.24) CONNECTEDNESS 0.325*** 51 0.221 0.298* 51 0.266 2 0.233*** 52 0.125 2 0.169 52 0.403 (0.0787) (0.151) (0.0865) (0.129) PREVALENCE 2 0.120** 51 0.085 2 0.0478 51 0.209 0.103*** 52 0.068 0.0202 52 0.382 (0.0451) (0.0577) (0.0339) (0.0293) Fully Private Banks Only FRACBANKS 3.276*** 46 0.230 3.448*** 46 0.331 2 2.630* 46 0.150 2 0.691 46 0.402 (0.731) (0.719) (1.348) (1.248) Matı SHAREASSETS 1.309 43 0.036 1.086 43 0.210 2 3.480*** 43 0.252 2 2.731*** 43 0.529 (1.051) (1.129) (0.976) (0.702) FRACPOLITICIANS 27.26** 46 0.157 30.72* 46 0.270 2 34.50*** 46 0.254 2 20.47 46 0.438 (12.08) (15.92) (11.71) (15.64) CONNECTEDNESS 0.328*** 46 0.191 0.273* 46 0.237 2 0.340*** 46 0.207 2 0.202 46 0.430 (0.0975) (0.150) (0.0744) (0.130) MAXSHARE 2 0.149*** 46 0.101 2 0.0825 46 0.200 0.132*** 46 0.081 0.0268 46 0.398 (0.0495) (0.0636) (0.0419) (0.0372) 277´as Braun and Claudio Raddatz Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 278 Overall Supervisory Power No Controls Controls: Log Real GDP, Log Population Coef�cient No. of Observations R2 Coef�cient No. of Observations R2 THE WORLD BANK ECONOMIC REVIEW Measure (25) (26) (27) (28) (29) (30) All Banks FRACBANKS 0.310 52 0.002 0.360 52 0.059 (1.022) (1.511) SHAREASSETS 2 0.607 49 0.013 2 0.741 49 0.052 (0.909) (0.924) FRACBANKERS 2 0.842 52 0 2 4.211 52 0.059 (8.829) (13.97) CONNECTEDNESS 0.131 52 0.031 2 0.0293 52 0.058 (0.0942) (0.158) PREVALENCE 2 0.00670 52 0 0.0452 52 0.066 (0.0678) (0.0720) Fully Private Banks Only FRACBANKS 1.119 46 0.021 1.901 46 0.103 (1.011) (1.354) SHAREASSETS 2 0.175 43 0.001 0.0633 43 0.041 (1.509) (1.621) Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010 FRACPOLITICIANS 6.595 46 0.007 16.86 46 0.088 (11.08) (19.25) CONNECTEDNESS 0.188 46 0.048 0.0687 46 0.069 (0.121) (0.188) MAXSHARE 2 0.0385 46 0.005 0.0142 46 0.067 (0.0948) (0.101) *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Standard errors (in parentheses) are robust to heteroskedasticity. The dependent variable are the Barth, Caprio, and Levine (2003) principal component indexes of �ve dimensions of bank regulation: the degree of restrictions to entry, the magnitude of capital requirements, the extent of restric- tions to cross activities, the reliance of self monitoring, and the overall authority of the regulator. All regressions include the observations with zero matches between bankers and politicians in countries with more than two banks, and all regressions control for log real GDP per capita and log popu- lation. FRACBANKS is the fraction of banks with Bankscope data on board of directors that had a former politician on their boards. SHAREASSETS is the share of the total assets of banks with Bankscope data on board of directors that is represented by connected banks. FRACBANKERS is the fraction of bank directors that had a previous political position. PREVALENCE is the (log) ratio of the actual to the expected number of matches, where the expected number is computed assuming no bias toward politicians in the selection of bankers and assuming that both bankers and politicians are selected from the whole population of a country. MAXSHARE is the largest fraction of a country’s population from which politicians and bankers would have to be selected so that the hypothesis that the selection of bankers is not biased toward politicians could not be rejected at the 5 percent level. Source: Authors’ analysis based on data described in the text and table. Matı 279´as Braun and Claudio Raddatz Downloaded from wber.oxfordjournals.org at International Monetary Fund on September 15, 2010