What Drives Global Syndication of Bank Loans? Effects of Bank Regulations*

Transcription

1 What Drives Global Syndication of Bank Loans? Effects of Bank Regulations* Janet Gao Indiana University Yeejin Jang Purdue University December 12, 2017 Abstract The syndicated lending market has become highly globalized in the past two decades. This study examines how bank regulations affect the decision of lead arrangers and participant banks to form global lending syndicates. Using a sample of loans syndicated by banks from 44 countries and bank regulation indices introduced by Barth, Caprio, and Levine (2004), we find that stringent capital regulations and higher deposit insurance increase banks participation into syndicates led by banks in less regulated countries. Regulations that limit competition in the banking sector have the opposite effect. To confirm banks incentives, we show that loans issued by global syndicates that involve strictly regulated participants have higher spreads and higher credit risk. Our findings are consistent with a regulatory arbitrage incentive of participant banks. Key words: Syndicated Loans, Cross-Border Banking, Banking Regulation, Regulatory Arbitrage. JEL classification: G21, G30 *Both Janet Gao and Yeejin Jang are fellows of the Risk Institute. We are thankful to Daniel Carvalho, Andrew Ellul, Mariassunta Giannetti, Leslie Hodder, Naveen Khanna, Anthony Saunders, and Greg Udell for his comments. We also thank participants in seminars at Indiana University, Purdue University, and Michigan State University for their valuable input.

2 1 Introduction The syndicated loan market has become one of the most important sources of external financing for corporations (Sufi (2009) and Wittenberg-Moerman (2008)). 1 In this market, global syndication plays an important role. During the past two decades, over 59% of syndicated loans are jointly issued by a lead bank and one or more participant banks located outside the country of the lead arranger. The total amount of globally syndicated loans has risen from $822 billion in 1995 to over $1 trillion in the late 2000s. These patterns suggest that it is a common practice to syndicate across borders. Despite the popularity of global syndication, the incentives for lead and participant banks in forming such a lending structure remain understudied, especially the role of banking regulation. Regulatory differences across countries may induce different risk preferences and lending capacity. Through global syndication, lead arrangers can distribute credit risk to banks with higher risk tolerance and participant banks can access investment opportunities that are infeasible in their home country. Recent evidence suggests that regulations can significantly affect syndicated lending. For example, in March 2013, the Office of the Comptroller of the Currency (the OCC) and the Federal Deposit Insurance Company (the FDIC) instituted the Interagency Guidance on Leveraged Lending, which introduces stricter regulations on leveraged loans issued by U.S. syndicated lenders. Following the introduction of the regulation, as shown in Figure 1, the syndicated lending market in the U.S. experienced a decline in the amount of loans issued by U.S. lead arrangers but an increase in loans led by foreign banks. Meanwhile, U.S. banks increased their participation in loans initiated in foreign countries. In this study, we investigate the effects of banking regulations on the extent to which banks originate and participate in global syndicated lending. Among all banking regulations, we focus on three aspects. Since capital requirements directly affect banks risk tolerance and thus their capital to provide credit (Barth, Caprio, and Levine (2004), Laeven and Levine (2009)), our primary focus is on the stringency of capital requirements. 1 According to the statistics in Global Syndicated Loan Reviews by Thomson Reuters (Full year 2015), during the year of 2015, syndicated lending to U.S. corporations amounted to $2.2 trillion. 1

3 Syndication by US Banks around Introduction of Leveraged Lending Guidance % % % % % - Mar2011-Feb2012 Mar2012-Feb2013 Mar2013-Feb2014 Mar2014-Feb % $ US Loans by US Lead (Billion) $ Foreign Loans Funded by US Participant Banks (Billion) % Amount of US Loans by US Lead % Amount of Foreign Loans Funded by US Participant Banks Figure 1. Syndication activities of U.S. banks around the introduction of the Leveraged Lending Guideline in March This figure illustrates the changes in the amount of loans extended inside the U.S. during four years around the introduction. The solid columns represent the total amount of loans led by U.S. banks, and the patterned columns represent the total amount of loans that are initiated by non-u.s. lenders and syndicated with U.S. banks by year starting from March 2011 (in USD billion). The black solid line indicates the loan amount initiated by U.S. lead banks as a proportion of the total amount of loans issued to U.S. borrowers (in percentage), and the red dotted line indicates the percentage of loan amount participated by at least one U.S. bank as the total amount of loans initiated by non-u.s. banks. In addition, we investigate whether and how regulation-driven credit market competition affects global syndication, as competition may motivate banks to extend more credit for profit while suppressing their risk management incentives. Finally, we inspect the deposit insurance coverage ratio. Deposit insurance intensifies potential moral hazard problems, whereby depositors do not have strong incentives to monitor banks activities (Kane and Demirgüç-Kunt (2001)). Therefore, banks in countries with high deposit insurance are more likely to conduct risky lending activities than is desirable to regulators (e.g., Demirgüç-Kunt and Detragiache (2002) and Laeven (2002)). We conjecture that restrictive capital regulations may generate two conflicting impacts on global syndication activities. First, strictly regulated banks may participate more into syndicates initiated by banks in loosely regulated countries (regulatory arbitrage hypothesis). Prior research on cross-border banking activities documents that banks from strictly regulated countries are more likely to direct their capital overseas to fund risky investment in less regulated countries (Barth, Caprio, and Levine (2004), Houston, Lin, and Ma (2012), and Karolyi and Toboada (2015)). Similarly, as banks become more 2

4 capital constrained, they are more likely to participate in a loan deal instead of being a lead arranger. 2 Moreover, investing in a foreign-led loan creates a greater level of information asymmetry between regulators and the actual risky investment. Under Basel II, banks only need to provide broad categories of asset riskiness and assign internal ratings. Investing in foreign assets, which are more opaque, thus gives banks more discretion in assigning and even manipulating their internal ratings (Plosser and Santos (2014)). Analogously, we also expect banks under regulations that invite more competition in the banking sector and under regulations that insure more of their assets to participate more in foreign-led syndicates. On the other hand, we may observe the opposite effect, i.e., participants from loosely regulated countries join the syndicates originated from strictly regulated countries. Since capital stringency increases banks costs of capital (Kashyap, Stein, and Hanson (2010) and Baker and Wurgler (2015)), lead arrangers have the incentives to attract cheap capital from less regulated countries. Which effect dominates ultimately depends on the information friction between countries, the variation in banks costs of capital, and the bargaining power between lead and participant banks. We set out to test the effects of bank regulations on global syndication structure using data from several sources. First, we collect data on globally syndicated bank loans from the LPC Dealscan database. In doing so, we focus on 44 countries whose banks initiate the majority of loans in the syndicated lending market. 3 We maintain a focus on ultimate parent banks while adjusting the ownership structure based on bank merger information from SDC and National Information Center. 4 To measure the level of banking regulations from the countries in our sample, we use survey-based data compiled by Barth, Caprio, and Levine (2004). The surveys are conducted every three to four years from the late 1990s to early 2010s. Our primary regulation of interest is capital requirements (Capital 2 Lead arrangers often commit a larger proportion of capital in a syndicated loan. 3 To generate meaningful within-country variation, we require each country to have at least 5 banks and the average number of loans issued by year above A controlled foreign subsidiary would be subject to regulations both in the foreign subsidiary country and headquarter country. The existence of foreign subsidiaries should mitigate the incentives in conducting cross-border syndication. We aggregate loans issued through foreign subsidiaries at the ultimate parent level, which should bias against us finding an association between differences in banking regulation and global syndication activities. 3

5 Requirement). We further gauge regulatory restrictions in the entry into the banking sector (Entry Requirement). More requirements for new bank entry suggest a higher barrier in entering the sector, thus less competition faced by incumbent banks. Finally, we examine banks moral hazard due to the level of deposit insurance (Funding Insured). We begin by examining whether differences in bank regulation stringency between a lead country and a participant country affect the syndication activity among banks in those countries. We employ two samples for these tests. The first sample is a countrypair-year panel that estimates the overall net capital flow between a pair of countries in a given year. This panel aggregates information regarding global syndication activities among banks to the country level and provides direct evidence on the association between regulatory stringency and syndication activities. This construction methodology is designed to alleviate concerns arising from noisy estimates of bank-specific lending activities. The second sample is a bank-pair-year panel, which characterizes the likelihood and the amount of syndication conducted by every pair of banks. The bank-level sample allows us to explore more granular insights on the incentives of individual banks regarding the formation of lending syndicates. Further, with this bank-pair-level sample, we can control for the initial matching between lead and participant banks and bank-specific characteristics. We find a positive relationship between syndication and the differences in capital regulation stringency between the country of participants and the country of lead arrangers. The results indicate that banks from strictly regulated countries participate more in loans initiated by banks from less regulated countries. As we investigate the effect of regulation-induced competition pressures, we find that banks from countries with more regulatory restrictions on bank entry are less likely to participate in loans led by foreign banks. Finally, we document that banks from countries with a greater proportion of their funding insured are more likely to participate in loans led by foreign banks, which lends support to the notion that deposit insurance promotes risk-taking incentives. These patterns manifest both at the country level and at the bank level. They are also robust to various empirical specifications, such as controlling for year fixed effects, lead lender 4

6 fixed effect, participant lender fixed effects, and time-varying macroeconomic conditions of both the lead and participant countries. Taken together, our baseline findings are consistent with the regulatory arbitrage hypothesis, i.e., banks allocate their capital towards risky investment by syndicating abroad. We further investigate whether the variation in syndication activity arises from the extensive or intensive margins. As a bank seeks to invest in a foreign-led syndicate, it might resort to previous syndicate partners or start new syndication relationships. Differentiating between the two margins can help shed light on the frictions related to the formation of global syndicates. We regress syndication activities on the interaction of regulatory differences between participant and lead countries and an indicator for whether the bank pair has been in the same syndicate before. The results suggest that the variation in global syndication activities is mostly concentrated on bank pairs with previous syndication experience. In other words, our base effects are mostly driven by the intensive margin. Moreover, there is a strong pattern of lead-participant bank pairs to syndicate together in the future. These patterns suggest that global syndication activities represent repeated interactions between banks. As banks seek new investment opportunities, they resort to trusted partners instead of seeking broadly for new relationships. An alternative explanation to our base findings is that lead lenders prefer to choose strictly regulated banks as syndicate participants, for those banks have stronger balance sheets and more capital reserves to withstand credit shocks. Prior research suggests that negative capital shocks to a syndicate member may trigger ripple effects across all banks in the syndicate (Nirei, Caballero, and Sushko (2016)). Moreover, syndicated lending represents repeated games between lead lenders and participants (Gopalan, Nanda, and Yerramilli (2011) and Cai (2010)). As a result, lead arrangers may prefer to invite strictly regulated banks as syndicate members (risk-sharing hypothesis). We empirically distinguish between the regulatory arbitrage hypothesis and risk-sharing hypothesis through a gravity model, in which we regress global syndication activities on banking regulation indices in both lead and participant countries separately. The gravity model thus helps distinguish whether the relation between global syndication patterns and regula- 5

7 tion differences is driven by regulations from the lead lender s country or the participant s country. We find that such a relation is largely driven by regulations from participant countries. Moreover, evidence from lead lenders regulations seems inconsistent with the risk-sharing hypothesis. For example, the risk-sharing hypothesis suggests that a strong capital requirement should encourage lead lenders to share risks with multiple lenders, while having deposit insurance should decrease such an incentive. In contrast, we find that stringent capital regulation from the lead country suppresses syndication activities, while deposit insurance facilitates global syndication. Taken together, we do not find evidence in support of the risk-sharing hypothesis. We next analyze which banks are more likely to conduct regulatory arbitrage using a cross-sectional test. Our base results suggest that global syndication is affected by differences in the regulatory-stipulated risk level and shareholders desired risk level. We expect that the motive to participate in foreign-led syndicates should be stronger for capital-constrained banks, so we examine whether banks capital adequacy could moderate the relation between bank regulations and global syndication activities. Consistent with our conjecture, banks with a lower level of Tier-1 capital ratio are more likely to participate in foreign-led syndicated deals when they are subject to stringent capital requirements, face more competition from other banks, and have more insured assets. In the last step of our analysis, we seek to confirm the regulatory arbitrage incentive by inspecting the terms and performance of globally syndicated loans. To maintain comparability across borrowers, we focus on a sample of loans extended to U.S. firms. Using this sample, we first document that globally syndicated loans on average have higher spreads and higher rates of default, which indicates that those are riskier loans. More importantly, we find that loans syndicated with strictly regulated participant banks are also riskier, as those loans have higher spreads and higher default likelihood. These results lend further support to the regulatory arbitrage hypothesis, i.e., banks from strictly regulated countries seek profitable, yet riskier, investment opportunities abroad by joining foreign-led lending syndicates. Our study contributes to several strands of literature. First, this study is among 6

8 the first to examine the determinants and formation of global lending syndicates. The extant literature on debt contracting and costs of bank credit largely takes as given the structure of lending syndicates and the sources of funding for syndicate loans (e.g., Lin, Ma, Malatesta, and Xuan (2011), Hertzel and Officer (2012), Valta (2012)). Several recent studies examine factors that could affect bank syndication networks. Cai (2010) documents a reciprocal arrangement for syndicate lead arrangers, i.e., lead arrangers will later participate in loans led by their participant banks. Gopalan et al. (2011) find that past negative credit events diminish the reputation of a lead bank. Cai et al. (2017) document that banks form syndicates based on their lending expertise. We contribute to this literature by showing the importance of global syndicated lending, and that bank regulations in the lead and participant countries have profound impacts on the structure of global syndicates. Second, our study adds to the recent research on banks cross-border activities and regulatory arbitrage (Houston et al. (2012), Ongena et al. (2013), and Karolyi and Taboada (2015)). Our study complements this literature by suggesting an additional channel through which banking regulations affect bank risk taking, i.e., participation in global syndicates. We depart from the existing literature by documenting a complex role of banking regulation, for not all stringent regulations encourage more risk taking. We find that regulations on bank entry reduce the competitive pressure of incumbent banks, thus alleviating their incentives to undertake risky investment. Moreover, we add to existing studies by showing the implications of cross-border lending on corporate borrowers. More broadly, our study is related to the extant literature that examines the effect of banking regulations on banks costs of capital, risk-taking behavior, and the health of the banking sector. Prior studies document that a higher capital adequacy requirement increases banks funding costs (Kashyap et al. (2010), Baker and Wurger (2015)), promotes their resilience to negative shocks (Dewatripont and Tirole (1994) and Berger, Demsetz, and Strahan (1999)), and reduces their willingness to screen borrowers and extend loans (Koehn and Santomero (1980), Thakor (1996), and Gorton and Winton (2003)). Barth, 7

9 Caprio, and Levine (2004) provide wide-spread empirical evidence on the actual practices of banking regulations and how these practices relate to bank development and stability. Our paper builds on this literature and provides evidence on how country-level bank regulation affects bank loan syndication activities by altering the structure of syndicates in an international setting. Section 2 introduces the data source, variable definitions, sample construction, and our empirical strategy. Section 3 describes the evolution of global syndication as well as the composition of participants. Section 4 introduces the baseline empirical results. Section 5 examines the implication of global syndication on borrowers. Section 6 concludes. 2 Data 2.1 Sample construction We obtain data on syndicated bank loans during the period of 1995 through 2016 from the LPC Dealscan database. We first restrict the sample of institutions to only those classified as banks in Dealscan. Next, we restrict our sample to loans that are originated by banks from the countries with at least 5 banks that collectively initiate on average 100 loans per year to ensure a comprehensive set of banks in each country in our analysis. Single-lender loans and loans with missing country information of lenders are excluded. 5 These sample selection criteria limit our sample to 44 countries, which represent 96% of the universe of loans in Dealscan in terms of deal number and loan amount. We sample on global syndication partners, i.e., pairs of countries or pairs of banks from 44 countries that have initiated or have participated in the same syndicated loan at least once in our sample period. We classify loans and locations based on the ultimate parent lenders. To identify the ultimate parent banks of each lender, we rely primarily on bank ownership structure information provided by Dealscan and revise that information based on bank mergers. 6 We 5 We additionally exclude Vietnam from the sample because of limited availability of bank regulation information. 6 One limitation of using ultimate parent information of banks from Dealscan is that the information 8

10 focus on ultimate parents, for the capital adequacy of parent banks can affect the lending behaviors of subsidiaries and branches. To the extent that affiliated subsidiary banks located in a foreign country may be subject to both the banking regulation of the home country as well as the regulation of the host country, lending through the subsidiary will not help circumventing regulations. As such, having an affiliated foreign subsidiary may only limit the regulatory arbitrage incentives of parent banks in conducting global syndication through that subsidiary, thus biasing against us finding an association between differences in regulation and global syndication activities. 7 We follow Bharath et al. (2011) and identify lead lenders and participants in a syndicate. Specifically, a lender is defined as a lead lender if the lead arranger credit variable in the Dealscan database is Yes or if its role is identified as Agent, Administrative Agent, Arranger, or Lead Bank. When we pool all syndicated loans across countries and years based on the criteria above, the final sample contains 114,992 loans initiated by 1,108 lead arrangers and participated by 3,487 banks from 44 countries. To examine the determinants of syndicate formation, we organize the sample in three ways for our empirical tests. First, we construct a country-pair-year panel dataset to estimate the syndication activity between two countries. For each lead-participant-country pair in a given year, we aggregate the number of loans that are led by the lead bank country and syndicated with the participant country. We drop pair observations within the same countries and we require both countries in a pair to have available information for bank regulations. This sample includes 40,576 observations with 1,936 country pairs and spans the period of 1995 through Next, we construct a sample of bank-pair-year panel to examine the syndication activity between two banks located in different countries. The unit of observation in this sample is a pair of lead bank i and participant bank j in a given year t. This samis based on the most recent ownership status. For example, Wachovia was an independent bank entity prior to the merger with Wells Fargo in 2008, but it is identified in Dealscan as a subsidiary of Wells Fargo based on the current ownership status. We re-assign loans to ultimate parent banks prior to the merger based on bank acquisition and merger information extracted from SDC database, supplemented by institution history information from National Information Center (NIC). 7 In robustness analyses, we define bank locations as the location of the subsidiary banks, and the results remain unchanged. 9

11 ple provides more granular information on individual banks responses to country-level regulation changes. One advantage of using a bank-level sample is that it allows us to analyze intensive and extensive margins of syndication activity. In particular, we are able to examine whether banks resort to previous syndicate partners to explore investment opportunities or search for new partners. To avoid inflating the sample, we restrict the sample to cross-border bank pairs. This sample contains 865,087 observations of 64,259 bank pairs and spans the years from 1995 through Using this bank-pair-year data, we examine whether the stringency of bank regulations in the country of a bank i affects the likelihood of bank i to initiate a global syndicate and whether the regulation stringency affects the likelihood of bank i to join global syndicates initiated by banks from other countries. Given that the bank-level activities more directly reflect the incentives of individual banks to form loan syndicates, we focus on the results from the bank-level sample for the majority of our analyses. 8 Finally, we examine loan-level data, in which the unit of observation is a syndicated loan package. Prior literature documents that loan contract terms vary significantly across countries (e.g., Carey and Nini (2007) and Qian and Strahan (2007)). To maintain the comparability of loan contract terms and the quality of borrowers, we restrict the loanlevel sample to those issued to U.S. public firms with non-missing borrower information. The resulting sample contains 28,494 loans. 9 Using the loan-level sample, we examine whether the average bank regulation stringency of lead or participant banks is associated with loan spreads and loan performance. 2.2 Global syndication measures To examine the determinants of syndicate formation, we consider two main measures of syndication activities. First, we consider Syndicate, which is a dummy variable that equals one if banks from two countries issue at least one syndicated deal together in a given year, in which one bank acts as the lead arranger and the other bank as a participant, 8 We find similar results when we estimate the regressions at the country-pair level. 9 The sample of U.S. syndicated loans seems representative; it presents 52% of the syndicated loans in our full sample in terms of the number and 67% in terms of the total amount. 10

12 and zero otherwise. Specifically, Syndicate is defined as follows: Syndicate i,j,t = max k K i,t 1 i,j,k, in which i indicates a lead arranger, j indicates a participant bank, and k indicates a syndicated loan deal. K i,t represents the collection of syndicated loans initiated by the bank i in year t. In the country-pair-year panel data, Syndicate i,j,t is defined such that i indicates a country of lead banks and j indicates a country of participant banks. In the bank-pair-year panel, i indicates a lead bank and j indicates a participant bank. 1 i,j,k is an indicator function that equals one if a bank (country) i is a lead arranger of a bank loan k and if a bank (country) j is a participant of a bank loan k. K i,t indicates all of the syndicated loans extended by the bank (country) i in a given year t (including both domestically and globally syndicated loans). Syndicate is a coarse measure of the likelihood of collaboration between two countries or two banks because this measure does not take into account the intensity of the syndication activities between two banks or the relative importance of a specific participant to the lead arranger. We next construct a continuous variable that estimates the syndication frequency between two banks (two countries) relative to the total syndication activities of the lead arranger (lead country). Accordingly, for each bank pair, we define %Syndicate as the number of syndicated loans in which a bank i is the lead arranger and a bank j (i j) is the participant as a proportion of the total number of syndicated loans extended by the bank i in a given year t, as follows: %Syndicate i,j,t = k K i,t 1 i,j,k j B k K i,t 1 i,j,k, in which B indicates the collection of potential participant banks (countries). Note that although our country-pair and bank-pair samples exclusively include cross-country pairs, we include the number of loans extended by the lead arranger that are both domestically and globally syndicated in the denominator. In this way, we control for factors that would affect syndication formation within a country as well. 11

13 2.3 Banking regulation To examine the effect of banking regulations on global syndication, we use countrylevel bank regulation measures obtained from Barth, Caprio, and Levine (2013). Among a variety of dimensions of regulation restrictions, we focus primarily on capital reserve requirement, which we call Capital Stringency. Capital Stringency is a composite index measuring the extent to which the capital requirement of a country reflects certain risk elements and market value losses. Previous studies suggest that capital requirements have a profound influence over the risk level of banks investment, assets, and banks financing activities (e.g., Laeven and Levine (2009)). The regulatory arbitrage hypothesis predicts that a more stringent capital regulation will provide stronger incentives for banks to engage in global syndication activities. In particular, banks under stringent capital regulation should participate more in loans originated by banks in less regulated countries. We supplement capital requirement by two other regulation indices that could create risk-taking incentives. First, we examine competitive pressures in the banking sector using Entry Requirement, which is an index that reflects the level of legal procedures that are required for a new entity to obtain a banking license. A higher value suggests a greater hurdle of entry into the banking sector, thus implying less competitive pressure for incumbent banks. As fierce credit competition within a country may fuel more aggressive risk-taking behavior (Ruckes (2004)), we expect that banks from countries that impose more entry restrictions are less motivated to participate in foreign-led syndicated deals. Second, we examine the level of moral hazard in a country using Funding Insured. This variable records the percentage of assets in the commercial banking system s that are funded with insured deposits. A higher value suggests a greater degree of moral hazard, thus a stronger risk-taking incentive of banks in that country. We expect banks whose funding is more insured to be more likely to participate foreign-led syndication. These bank regulation measures are based on the cross-country surveys conducted by the World Bank. There have been four surveys (in the years of 1999, 2002, 2005, and 2011) collected from 107 countries. These surveys collected over time provide timevarying proxies for the bank regulation stringency of each country. Following Karolyi and 12

14 Toboada (2011), we apply the regulation variables from the 1999 survey for observations from 1995 through 2001, the values from the 2002 survey for observations from 2002 through 2004, the values from the 2005 survey for observations from 2005 through 2010, and the values from 2011 survey for observations from 2011 through Controls Syndication activities can be affected by the economic conditions of each syndicate member s country. We thus consider country-level macroeconomic conditions that might affect the demand and supply of bank credits. To control for the size and growth of the economy, we use the log of GDP per capita (GDP per Capita) and the growth in real GDP (GDP Growth). To control for the information asymmetry and cultural differences between a pair of countries, we use the log of geographical distance (Ln(Distance)) and an indicator variable denoting whether the country pair shares the same language (Common Language). Distance is defined as the circle distance between the capital cities of two countries. We obtain these country-level macroeconomic variables from the World Bank. In a subset of regressions, we additionally control for bank characteristics using banks financial information from the Bankscope database. We first manually match lenders in Dealscan to banks in Bankscope based on bank names, locations, and entity type. We then calculate bank size as the log of total assets, bank ROA as return on average total assets, and Tier-1 capital as the ratio of Tier-1 capital to risk-weighted assets. In the analysis of loan terms and loan performance, we control for the borrower characteristics, which include firm size, tangibility, and profitability. We use the matching table provided by Chava and Roberts (2008) to combine the syndicated loan information with financial data from Compustat for the U.S. borrowers. Definitions and sources of all variables are provided in Appendix A. 2.5 Empirical Methodology Our baseline approach examines the effects of bank regulations on banks syndication activities in both a country-pair panel and a bank-pair panel. To do so, we investigate 13

15 whether, and how, the difference in bank regulation stringency between country i and county j affects the likelihood that banks in country i will initiate a syndicated loan with banks in country j as syndicate participants. Specifically, we estimate the following model: SyndicateActivity i,j,t = α + β 1 Regulation i,j,t + β 2 Controls + φ i + η j + µ t + ɛ i,j,t, (1) in which i indicates the lead country, j indicates the participant country, and t indicates the year of observation. SyndicateActivity i,j,t {Syndicate i,j,t, %Syndicate i,j,t }. Regulation {Capital Stringency, Entry Requirement, Funding Insured}. 10 Regulation i,j,t measures the differences in banking regulation between the participant country j and lead country i (i.e., participant lead). Thus, higher values of Regulation i,j,t indicate that the participant bank country imposes more stringent regulations than the lead arranger country. In the country-pair panel, we control for time-invariate features of a country that could drive the financial sector activities in that country. We thus include fixed effects for the lead arranger country, φ i, and the participant country, η j. In the bank-pair panel, we include lead-bank fixed effects and participant-bank fixed effects to control for timeinvariant characteristics of those banks. In both samples, we include year fixed effects, µ t, to control for the correlation between syndication activities and bank regulation driven by common time-series trends. We further control for macroeconomic conditions, GDP per Capita and GDP Growth as well as country-pair variables, Ln(Distance) and Common Language. Since our main independent variables of interests are the differences in bank regulations between two countries, standard errors are clustered at the country-pair level. If banks are motivated by regulatory arbitrage in forming global lending syndicates, we should observe β 1 > 0, i.e., banks from strictly regulated countries are more likely to participate in syndicated loans initiated by banks from loosely regulated countries. Alternatively, if capital flows from weakly regulated countries to stringently regulated 10 We estimate the equation using a linear probability model, even for the indicator variable (Syndicate), instead of probit or logit specifications, because nonlinear models can be inconsistent with fixed effects included. 14

16 3,000 2,500 2,000 1,500 1, Global Syndication Over Time: % 95% 90% 85% 80% 75% 70% 65% 60% 55% 50% Total Amount of Globally Syndicated Loans ($ Billion) % Amount of Globally Syndicated Loans Figure 2. Global syndication among banks in 44 countries. This figure presents the time series trend in global syndication activities. The solid line indicates the percentage of loan amount that is syndicated globally, i.e., a syndicate loan occurs when at least one participant lender is located in a different country of the lead arranger, relative to the total amount of syndicated loans extended every year. The blue columns indicate the total amount of syndicated loans with at least one foreign lender relative to the total number of syndicated loans extended every year. The left vertical axis indicates the amount of globally syndicated loans per year (in $billions), the right vertical axis indicate the percentage of globally syndicated loans, and the horizontal axis indicates time. countries, we should observe that β 1 < 0. The regulatory arbitrage hypothesis predicts that β 1 > 0 for capital stringency and funding insurance but β 1 < 0 for entry requirements in the banking sector. 3 Univariate Analyses Figure 2 illustrates the trend in global loan syndication by year. We define a loan as globally syndicated if at least one lender in the syndicate is located outside of the country of the lead arranger. Over our sample period, there is an increasing level of global syndication. In 1995, globally syndicated loans, which are jointly issued by a lead arranger and contains at least one foreign participant bank, had a total face value of around $800 billion. The amount of globally syndicated loans increased to over $2.7 trillion in The participation of foreign lenders in bank loan syndicates experienced a significant decline following the 2008 financial crises. This phenomenon seems consistent with the flight-to-home effect documented in the previous studies (Giannetti and Laeven (2012)). After the crises, global syndication continues to rise, reaching over $2.3 trillion in

17 Syndicate Compositions Over Time: % 70% 60% 50% 40% 30% 20% 10% 0% #Foreign Participants #Total Participants Percentage of Foreign Participants Figure 3. Globalization of syndicated lending. This figure presents the time series trend in global syndication activities. The solid line indicates the average percentage of number of foreign participants in a syndicate by year. A participant lender is defined as a foreign lender if it is located in a different country than the lead arranger. The blue columns indicate the average number of foreign participant in a syndicate, and the patterned column indicates the average number of participants in a syndicate. and In comparison, the percentage of globally syndicated loans has remained stable at around 90% over time. Figure 3 plots the participation by foreign banks over time. Although the number of syndicate participants together with the number of foreign participants are declining during our sample period, the level of participation by foreign banks remains relatively stable over time. Similar to the overall level of global syndication, the percentage of foreign participants also experienced a temporary decline during the financial crisis period of 2008 to Table 1 summarizes the extent of cross-border syndication activities by banks in each country of our sample. Columns (1) through (4) report the overall syndication activity for banks in each country. We calculate the total number of loans issued, total loan amount, the number of lead arrangers, and the number of participant banks that are involved in loans syndicated by banks for each country. These summary statistics show that with 424 banks assuming the role of lead arrangers and 1,142 banks being the participant lenders at least once in the sample period, the U.S. is the largest country in issuing syndicated loans, followed by Japan, Spain, and Germany. Columns (5) through (7) report the extent of global syndication activities for each country. Specifically, for each country, we calculate 16

18 the average percentage of globally syndicated loans over the total number of loans led by banks in a given country over the sample period. We also calculate the average percentage of loan amount contributed by foreign banks and the percentage of foreign participant banks over the total number of participants. There is a substantial variation in terms of syndication structure across different countries. Japanese banks, for example, maintain a relatively isolated syndication style in that only about 23% of syndicate loans are funded by foreign banks. In a typical Japanese bank-led syndicate, foreign banks account for less than 20% of all syndicate members. European banks, on the other hand, seem to syndicate in a more open style. For example, 99% of the loans led by Swiss banks include at least one foreign lender and around 36% of the loan amount is contributed by foreign banks. U.S. banks exhibit an intermediate degree of global syndication activities, for about 67% of the loans initiated by the U.S. lead arrangers include foreign banks in their syndicates and attract 33% of loan amount from these foreign banks. Table 1 About Here Table 2 provides the summary statistics and the availability of bank regulation variables. Panel A provides summary statistics of our key syndication measures. Capital Stringency and Entry Requirement have mean levels of around 4 and 7, respectively. These two variables have standard deviations of 1.7 and 1.1, which suggests that those regulations remain relatively stable across the sample and over time. On average, banks in our sample countries have 40% of their assets funded by insured deposits, with a 30% standard deviation. In untabulated results, we find that the stringency of bank regulation has been increasing over time. For example, the median value of Capital Stringency is 4 in the 1990s and early 2000s, but increases to 5 in the most recent survey. This suggests that it is important to control for the time trend of bank regulation in our regression analyses. Panel B lists the number of survey values available for each country in our sample. The majority of countries have available values from all four surveys for either capi- 17

19 tal regulation or entry requirement. 11 However, the index for funding insurance is only available for 39 countries after When we compare the stringency of capital requirement across countries, we find that Australia, Singapore, and Spain are the most heavily regulated countries, and Canada, Malaysia, and Sweden are among the least regulated countries. European countries (e.g., UK, Italy, and Spain) tend to have the strictest requirement for entry into the banking sector. In terms of deposit insurance, banks in UK, Italy, Singapore, and Greece have the highest percentage of their assets being funded by insured deposits (above 80%). Table 2 About Here 4 Bank Regulations and Global Syndication 4.1 Country-level analyses We first estimate Equation 1 to examine the relationship between bank regulation and syndicate structure at the country-pair level. For each country pair, we test whether and to what extent banks from these two countries engage in the same syndicate with one of them being the lead arranger and the other being the participant bank. As shown in Table 1, some countries have only a few banks that initiate loans. To alleviate the issue that such small countries might over-represent the country-level pair sample, we drop country-pair observations if a lead country i does not issue any loans. Table 3 About Here We report results from country-level regressions in Table 3. Columns (1) through (3) report estimation results for Syndicate. Column (1) shows a positive, significant coefficient for Capital Stringency, which suggests that a global lending syndicate is more likely to involve participants facing stricter capital requirements than the lead arranger. The estimate suggests that a one-standard-deviation increase in the differences in capital 11 Only three survey values are available for Czech Republic, Japan, Saudi Arabia, and Sweden. 18

20 regulation stringency is associated with a 0.5% increase in the likelihood of syndication between two countries, which is equivalent to a 2% increase at the mean. Column (3) shows a positive and statistically significant coefficient for Funding Insured, suggesting that as banks face a stronger safety net in the home country, they are more likely to participate in foreign-led syndicates. Columns (4) through (6) show the results for %Syndicate. To mitigate the concern that small countries that have issued only handful of loans can bias the magnitude of allocating loans across countries, we further restrict our sample to country-pair observations where lead country i issues at least 10 loans in a given year. The difference in capital regulation stringency loses significance in Column (4), likely due to lack of variation in country-level syndication activities. Column (5) reports a negative, significant coefficient for Entry Requirement. The estimate suggests that a one-standard-deviation increase in the level of entry requirement is associated with about 0.2% decrease in the level of participation in foreign-led syndicates. Funding Insured bears a positive, significant coefficient, supporting the hypothesis that deposit insurance motivates banks to conduct more global syndicated lending. Taken together, results from country-level analyses are largely consistent with the regulatory arbitrage hypothesis, i.e., as stricter regulations in the home country create incentives for banks to extend risky loans, banks are more likely to participate in global syndicates initiated by foreign, less regulated banks. 4.2 Bank-level analyses Examining syndication activities at the bank level can better shed light on the incentives that individual banks face in forming lending syndicates. We start by estimating Equation 1, in which we regress the syndication activities between a lead bank and a participant bank on the difference in bank regulations of the home country of each bank. Table 4 reports the results. Similar to country-level analyses, we examine the effect of bank regulations on the likelihood of syndication between two banks, Syndicate, in Panel A. In Panel B, we consider the extent of syndication activity, %Syndicate, measured by 19

21 the proportion of syndicated loans issued between two banks over the total number of syndicated loans extended by the lead bank. In each panel, we first report regression results for all three measures of bank regulations with country-level fixed effects and then we do so with bank-level fixed effects. Table 4 About Here We find that the differences in capital regulation stringency between the participant and lead countries are positively associated with syndication activities between the bank pair. The economic magnitude of the effect of the differences in bank regulations from the bank-level analyses is stronger than that from the country-level analyses; a one-standarddeviation increase in Capital Stringency of participant bank country (1.72) is associated with a nearly 1% increase in the likelihood that banks from two countries collaborate in a syndicated lending deal. In Panel B, the analyses of the extent of syndication activity, however, yield a smaller magnitude than that from the country-level test. The estimates in Column (2) indicate that when a participant is from a country with stronger bank regulations (by one standard deviation), a lead arranger is likely to syndicate 0.09% more loans with that participant bank. Given that the median and mean values of %Syndicate are 2.18% and 0%, respectively, the economic magnitude of the effect is equivalent to a 4% increase (at the mean) in the syndication allocation. Columns (3) and (4) of each panel examine the effect of entry requirement. Keeping the lead arranger country s regulation constant, a one-standard-deviation increase in the participant country s entry requirement is associated with a 0.4% decrease in global syndication activities, and about a 0.15% decrease in the percentage of loans that the lead lender will syndicate with that participant bank. This is consistent with the hypothesis that competition within the banking industry intensifies the risk-taking incentives of incumbent banks, leading them to seek more syndication deals from foreign countries. Finally, Columns (5) and (6) report the results from funding insurance. We find a positive association between the funding insurance in a participant country and the likelihood that banks in that country will join syndicates arranged by a less-insured bank. 20

22 This result is in support of the notion that deposit insurance may breed moral hazard, i.e., banks are less concerned about downside risks, thus taking a higher level of risk than is desirable to regulators. A one-standard-deviation increase in the difference of funding insurance is associated with 0.3% increase in the likelihood that a bank will participate in the global syndication deal, which is equivalent to a 4% increase in the likelihood of crossborder syndication. In Column (6), the difference in funding insured is positively related to the intensity of global syndication between two banks. The economic magnitude of the effect of a one-standard-deviation increase in a participant bank s funding insurance is equal to a 8% increase in the percentage of global syndication. Taken together, the results from the bank-level analyses confirm the equilibrium outcome that we find from the country-level analysis; banks facing more stringent capital regulations are more likely to join lending syndicates initiated by lead banks under more lenient regulations. We find that both competition in domestic banking and moral hazard induced by deposit insurance also contribute to the global syndication activities. Our findings so far highlight a positive correlation between the relative regulatory stringency faced by participant banks and their involvement in global syndicates initiated by less regulated lead banks. Such a correlation could originate from either extensive margin or intensive margin. The extensive margin refers to cross-bank-pair comparisons. As such, characteristics within a bank pair may correlate with both the differences in regulations between their countries and their syndication activities. The intensive margin, on the other hand, refers to the changes in syndication tendencies between a given pair of banks. Although both margins are important in their own right, the effect from the extensive margin is likely to be confounded by unobservable characteristics that are specific to the new participant banks. The intensive margin, on the other hand, tracks the recurrence of the syndication activities while keeping the lead-participant bank relationship constant. To separate the effect of the intensive margin from that of the extensive margin, we repeat our bank-level tests (Equation 1) while controlling for lead bank-participant bank-pair fixed effects. More importantly, we examine the interactive effect of regulation 21