Borrowing from Friends of Friends: Indirect Social Networks and Bank Loans *

Transcription

1 Borrowing from Friends of Friends: Indirect Social Networks and Bank Loans * Sterling Huang *, Bo Li, Massimo Massa, Hong Zhang This draft: January 30, 2018 Abstract We study how indirect connections in social networks (i.e., friends of friends) may affect the loan contracts offered by banks. Using board interlocks as a proxy for social networks, we find that, for firms and banks that are socially unconnected, having a third party that is socially connected to both parties significantly reduce loan spreads. Using changes in board composition due to thirdparty director death and retirement to identify disruptions of indirect connections, our differencein-differences results support a causal interpretation. Subsequent firm performance deteriorates after loan initiations, which is reflects in increases in the probability of default. We find that indirect connections fail to incorporate information into the loan pricing and impede the informativeness of debt contracts with respect to default risk. Our results support the favoritism hypothesis that indirect connections fail to facilitate better information flow in social networks and trigger sub-optimal level of monitoring in the lending market. JEL classification code: G12, G10, G21, G30, G32, G34 Keywords: Social Networks, Bank Loans, Loan Spreads, Covenants. * We would like to thank Cheng Jiang, Kose John, Meng Miao, Wanli Zhao, Hao Zhou and conference and seminar participants at Five Star Workshop, PBC School of Finance, and Renmin University for their helpful comments. * Singapore Management University, Singapore; shuang@smu.edu.sg. PBCSF, Tsinghua University, 43 Chengfu Road, Beijing, PR China, ; lib@pbcsf.tsinghua.edu.cn. INSEAD, 1 Ayer Rajah Avenue, Singapore, ; massimo.massa@insead.edu. PBCSF, Tsinghua University, 43 Chengfu Road, Beijing, PR China, ; zhangh@pbcsf.tsingua.edu.cn.

2 Introduction The effectiveness of the banking business is affected by asymmetric information and moral hazard (e.g., Diamond, 1984, James, 1987). As a result, the ability for banks to collect information about the borrowing firms plays a major role in shaping the structure and operation of the lending market (e.g., Berger et al., 2005). But what affect the information ability of the banks in the first place? While traditional answers point to geographical proximity (i.e., originated from the work of Coval and Moskowitz 2001) and customer relationship (e.g., Sharpe, 1990; Rajan, 1992), more recent studies have identified a new channel that can potentially affect information in general and lending activities in particular: social networks (e.g., Cohen, Frazzini, and Malloy 2008; Schmidt 2008; Engelberg, Gao, and Parsons, 2012, Larcker, So, and Wang 2013). Indeed, given the importance of soft information involved in firms and bank loans (Stein 2002; Berger et al., 2005), it is conceptually appealing, if not crucial, to introduce social networks into banking studies and to understand its potential influences. Empirical studies of social networks mainly focus on outcomes of direct connections. Although direct connections are important, an equally fundamental, if not more intriguing, observation is that nodes of networks can be indirect linked via other nodes. Instead of being direct friends, for example, two participants A and B can share a common friend C, thereby being indirectly linked as friends of friends. Not only are such indirect connections prevailing in networks (they overweight direct connections in many networks), they are also considered one of the most fundamental properties of networks (e.g., Jackson and Rogers 2007) which can give rise to interesting social and economic consequences (Jackson, Rogers, and Zenou 2014). If we interpret the common friend C as an intermediary, for instance, this prototype of network structure can help us to model the transactions observed in certain financial markets (e.g., Garmaise and 1

3 Moskowitz 2003). Given the theoretical importance of networks and the transmission mechanisms, however, empirical evidence on the influence of indirect connections is still scares in the bank lending market. Our paper aims to contribute by exploring the potential influence of indirect connections of social networks in the lending market. To see our intuition, consider the case in which firm A borrows some loans from bank B. Assume that the two are not directly connected in social networks (which we will specify shortly). Rather, they are indirectly connected via some common friends in social networks (firm C or similar firms). We want to examine whether their loans can be affected by the density of common friends they have, which we refer to as third-party networks or indirect network interchangeably. Intuitively, third-party networks can affect lending because they reduce the network distance between A and B: the more common friends, the closer they are located in social networks to receive soft information about each other. Hence, network proximity can also facilitate information flows and reduce costs associated with asymmetric information. The closeness through social ties contrasts the geographical proximity as banks can expand their branching networks to alleviate the geographical distance but not necessarily identify the inter-personal linkages between directors. Third-party networks can enhance the informativeness of lenders and lead to lower the cost of loans, which is referred to the network proximity hypothesis. Under the network proximity hypothesis, we should observe a negative relation between the indirect social connections and the loan spreads, which is also associated with better monitoring and less borrower defaults. We also consider alternative hypotheses that third-party networks may play in the lending market. First, to the extent that social networks may contaminate corporate governance of the borrowing firms (e.g., Fracassi and Tate, 2012; Kramarz and Thesmar 2013), which is referred to 2

4 as the agency cost hypothesis. Consequently, banks may want to charge a higher price of loans to compensate for the weakening of board monitoring. Under the agency cost hypothesis, we should observe a positive relation between the indirect social connections and the loan spreads, which is also associated with more borrower defaults. Second, banks may exhibit favoritism to more connected or more central firms (e.g., Guiso and Zingales, 2014; Haselmann, Schoenherr and Vig, 2017). Banks are likely to charge lower fees to firms within a network, but pricing is uninformative as no additional information is being acquired by banks, which is referred to as the favoritism hypothesis. Under the favoritism hypothesis, we should observe a negative relation between the indirect social connections and the loan spreads, but is also associated with more borrower defaults. Finally, firms within social networks are linked to other firms in a random way (Jackson and Rogers 2007). Third-party connections neither affect banks information nor loan pricing, which is referred to as the random effect hypothesis. Under the random effect hypothesis, we should observe no significant relation between the indirect social connections, loan spreads, and subsequent borrower defaults. We test these hypotheses using the sample of loans issued by all public firms in the US over the period from 2000 to Connections in social networks are proxied by shared board directorates i.e., board interlocks, because the literature provides vast evidence not only for the general importance of board but also the special role played by board interlocks in facilitating information, knowledge, and experiences (e.g., Stuart and Yim, 2010; Chiu, Teoh, and Tian, 2012; Larcker, So, and Wang, 2013). We proxy for the intensity of third-party networks between firm A and bank B by the logarithm of one plus the number of their common friends (e.g., firm C, who shares board directorates both with firm A and bank B). To highlight the power of third-party networks, we focus on loans between firms and banks that are neither directly connected via board 3

5 interlocks nor relationship lending in the past five years in our main tests. We allow and control for direct connections between A and B in robustness checks. We document a strong negative correlation between the third-party networks to loan spreads. The relation is not only statistically significant, but also economically relevant in terms of the magnitude. A one-standard-deviation increase in third-party network from the mean value translates into about 18.7 bps lower spread (or about 9% of the one-standard-deviation value of the loan spread). 1 Subsample analysis shows that the effect attenuates for older firms and firms with outstanding bonds, where there is less information asymmetry. We also observe that the negative relation to be stronger during periods of stress (18.0%) than in normal periods (9.6%), suggesting that third-party networks reduce financing costs even in financial distress periods. Finally, we conduct robustness tests and demonstrate that having a direct social network does not subsume the impact of third party networks, suggesting that direct and indirect networks contain different information to banks in the lending market. Though our pooled OLS regression suggests a negative relation between indirect social connections and bank loan spreads, it is difficult to interpret due to the endogeneity concerns. First, in the main regressions, we focus on the sample of firms that are neither directly connected to banks via board interlocks nor connected with banks through relationship lending in the past five years. Second, although endogeneity issues are major concerns in research on direct connections, we identify exogenous variations in third-part networks to disentangle the selection of firms by 1 Our regression is essentially ln(y) = α + β ln(x) + ε, where y and x are loan spread (in bps) and the number of common friends. We can estimate the one-standard-deviation impact of x either from a linear approximation or from the exact expression of logarithm. The linear approximation states that Δy/y ~β σ x /x, where y and x are the mean values of y and x, respectively, σ x is the one-standard-deviation change in x, and Δy is the corresponding impact. From Table 1 and Table 2, y = 188.6bps, β = 0.1, x = 21.1 and σ x = Hence Δy~ 21.5bps, which, scaled by the standard deviation of y (140 bps), is about 15%. If we use the exact expression of logarithm, then Δy/y = exp (β ln (1 + σ x /x )) 1. In this case, a one-standard-deviation change in x transforms into 13.8 bps lower spread or 9.9% standard deviations of y. 4

6 banks. In particular, we use the death or retirement of directors from third-part networks to identify changes in board composition that are not due to conditions of borrowers. To see the intuition, recall that in our previous example the third-part network essentially consists of two pairs of direct connections (interlocks): that between Firm A and the third-party Firm C, and that between Bank B and the third-party Firm C. If the second interlock gets disrupted by the death a director serving the boards of Firm C (but not Firm A), the fundamentals of firm C will be affected. However, neither the fundamentals of Firm A nor the riskiness of the loans borrowed by Firm A are directly affected, because these retired directors do not work for firm A (exclusion restriction). What gets disrupted is the connection between firms and their third-parties, which affects the informal channel to pass soft information from the Firm A to the Bank B via the intermediation (inclusion restriction). Director death or retirement of the intermediary can serve as a reasonable shock to introduce exogenous variations into third-party networks. We conduct a difference-in-differences regression around third-party director death or retirement events. We find significant increases in loan spreads around the death or retirement of directors from the intermediary but with network ties to the borrowing firms. The increases in the cost of financing is significant compared to the change in loan spreads around the death or retirement of unconnected directors. The difference-indifferences results are more pronounced for more matured firms, and for firms with existing issued bonds. In addition, the increases in loan spreads is stronger during periods of stress if the third-party firm experiences director death or retirement. These event studies suggest that the influence of third-party networks are causal on the price of loan. Having established the causality between indirect social connections and loan contract terms, we further differentiate the network proximity hypothesis from other hypotheses based on distinct 5

7 tests examining whether third-party networks can enhance the informativeness of loan prices. In the first test, we test the informativeness of loan prices by relating the degree of indirect networks to the likelihood of loan failure, an indicator for whether the firm files for bankruptcy before its outstanding loans mature. If the pricing of loans is informative, lower loan spreads should predict a lower probability of default in the future. To empirically examine the information role of thirdparty networks, we regress measures of future default risk, including the likelihood of loan default, on the third-part networks. Our results show that firms with more connected indirect networks are more likely to fail during the existence of a loan contract, which exposes banks to higher loan default risk. These evidence suggests that third-party networks fail to enhance the informativeness of loan spreads, and limited information about the quality of loans is flowed through third-party networks. Banks do not price in borrowers increases in credit risk given the weakening of the monitoring intensity associated with less strict covenants. The increases in loan failures are consistent with our proposed favoritism hypothesis, where a more indirectly network is associated with lower costs of bank financing, and more loan failures, which is captured by lower interest rate spreads, and more corporate defaults. Taken the findings together, we find more evidence supports the favoritism hypothesis, as opposed to the alternative network proximity hypothesis and the random effect hypothesis. We further pin down the mechanism through which the negative relation between indirect connections and loan spread, and positive relation between indirect connections and loan defaults occur by examining the monitoring role of banks. In the second test, we investigate whether the lower spread reported in our main tests simply reflect a tightened control that the bank may want to impose on loans. To test this channel, we examine the different types of covenants that can help banks to exercise control over their loans, including capital expenditure covenant, Debt-to-Ebitda 6

8 covenant, and net worth covenants. We find that third-party networks reduce the number of restrictive covenants, where the intensity of bank monitoring by using the total number of loan covenants, the number of financial covenants, and the number of networth covenants. We show that lenders impose less strict covenants in debt contracts when borrowers are connected through the third-party. Together with the previous evidence on loan spreads and defaults, third-party networks incentivize banks not only to reduce loan spread but also to lessen the control they impose on loan. In the third test, we examine the impact of indirect social connections on length and depth of the relationship of the borrower and the bank. Our results suggest that increases in indirect social connections are associated with enhance of firm-bank relations, captured by increases in the probability of banks become relationship lenders even in absence of previous lending relations, and more lending by banks in the future. We further conduct a series of cross-sectional tests to gauge the effect of third-party network on loan spreads. Our evidence shows that the negative relation between indirect networks and loan spreads is stronger when the intermediary has longer relationship with the bank and bigger lending position between the bank and the third party. These findings are consistent with the literature that a stronger lending relationship allows banks to obtain more accurate information from the borrower (Sharpe, 1990; Rajan 1992), where in our case the soft information is acquired from the third party about the borrowers. Subsample analyses show that the effect attenuates for older firms and firms with outstanding bonds, where there is less information asymmetry. We also observe that the negative relation to be stronger during periods of stress (18.0%) than in normal periods (9.6%), suggesting that third-party networks reduce financing costs even in financial distress periods. 7

9 Finally, we conduct a list of robustness checks. In the first robustness check, we demonstrate that having a direct social network does not subsume the impact of third party networks, suggesting that direct and indirect networks contain different information to banks in the lending market. In the first robustness test, we examine whether third-party networks affect more the demand side or the supply side of the lending market. More explicitly, when prices drop, an increase (decrease) in quantity of loans implies a positive supply shock (negative demand shock). A potential test of the two effects can help further differentiate different hypotheses: while positive supply shocks can happen when reduced information asymmetry allow banks to issue more loans, the occurrence of negative demand shocks could be due to agency issues such as firms scale back from loans that lenders have more information. Empirically, we find that third-party networks mostly lead to a positive supply shock from banks. This observation rules out demand-side explanations and suggests that third-party networks affect more bank incentives than those of firms. Our paper makes several contributions. First, we document the role of indirect networks in the lending market. The existing literature shows that social networks can have significant influences on many aspects of our financial markets, ranging from a positive role in facilitating information (e.g., Cohen, Frazzini, and Malloy 2008; Schmidt 2008; Engelberg, Gao, and Parsons, 2012, Larcker, So, and Wang 2013) to a dark role of enhancing agency problems or favoritism (e.g., Fracassi and Tate, 2012; Kramarz and Thesmar 2013; Guiso and Zingales, 2014; Haselmann, Schoenherr and Vig, 2017). 2 Especially, Engelberg, Gao, and Parsons (2012) find beneficial influences for social networks to enhance information flows in the lending market. We differ by extending the analysis from direct connections to network proximity based on friends of friends. 2 Among others, social networks can also affect managerial decisions and incentives at workplace (Shue, 2013; Bandiera, Barankay, and Rasul 2009), moral hazard (Jackson, and Schneider, 2010), entrepreneurship (Lerner, and Malmendier, 2011), venture capital investment (Gompers, Mukharlyamov, and Xuan, 2012), trade (Cohen, Gurun, and Malloy, 2016), and economic growth (Burchardi, and Hassan, 2013). 8

10 Since such indirect connections are a fundamental feature in some of the most important social networks, such as World Wide Web and citation networks (Jackson and Rogers 2007), it is crucial to understand the extent to which our financial markets are also exposed to them. To the best of our knowledge, we are among the first to explicitly examine the influence of third-party networks in financial markets. Among the very few studies that are somewhat related to friends of friends, Garmaise and Moskowitz (2003) uses a theoretic model of non-financial intermediary to understand the commercial real estate market, and Zhao (2016) empirically examines the benefit of being an information intermediary. Our paper differs in that we focus on the explicit design of loans contracts between friends of friends. We also contribute to the banking literature (e.g., Diamond, 1984; Berger et al., 2005) and in particular to the one that focuses on proximity between borrower and lender and precision of the signal the bank receives on the borrower (e.g., Petersen and Rajan, 1994, Berger and Udell 1995, Hauswald and Marquez, 2000, and Sufi 2005, Dass and Massa, 2011). We provide evidence that a key source of proximity is the existence of third-party networks, which extends the findings that proximity based on geographical, economic, political, and social connections may affect lending activities (e.g., La Porta, Lopez-de Silanes, and Zamarripa 2003; Khwaja and Mian 2005, Fisman, Paravisini, and Vig, 2012). Our results suggest that third-party networks may have important normative implications for bank regulations by considering borrower credit risk. The remainder paper is articulated as follows. Section 2 describes the data and the main variables. Section 3 documents the causal link between the third-party network and the loan contract terms by using the death and retirement of directors from the third-party to identify changes in board composition. In Section 4, we assess whether third-party network provides accurate information to the lender or the rate concession reflects sweetheart deals. In Section 5, 9

11 we conduct cross-sectional tests on the relation between indirect network and loan spreads, and provide further robustness checks. Section 6 concludes. 2. Data and Main Variables In this section, we describe our data and how we construct the main variables we use in the analysis. A. Data Sources Our data are drawn from different sources. We first identify the Third-Party Network from the biographical information on directors and top five disclosed executive of publicly traded U.S. companies from the BoardEx database. Our sample contains the board interlock information for firms between 2000 and The key independent variable Third-Party Network captures the current employment connections between external directorships in the same firm. Second, we extract bank loan contract information from the LPC-Dealscan database. Our bank loan sample also spans between 2000 and 2012 as the loan level information coverage significantly increased during this period. We treat each loan facility as an independent observation and the associated loan covenants. We also keep both the revolver and term loans since these loans contain detailed information on the pricing and strictness of the covenant attached. We construct our final sample by matching the biographical data from BoardEx with loan facility information extracted from the LPC-Dealscan database. For a firm to be included in our sample, we require the connection measure from BoardEx, the loan pricing and covenant information from Dealscan. The loan level controls include: the logarithm of loan maturity in years (LoanMaturity), the logarithm of loan amounts in millions of dollars (LoanSize), and the type of loan (LoanType) equals one if the loan is a term loan, and zero if the loan is a revolver. We keep firms with no missing firm characteristics from the Compustat, and construct the following firm- 10

12 level control variables: logarithm of firm total assets (Size), profitability(profitability), the fraction of tangible assets to total assets (Tangibility), market-to-book (M/B), financial leverage (Leverage), the S&P rating indictor (Rating dummy), cash ratio (Cash ratio), and the Altman s Zscore (Zscore). In account for time trend and the macroeconomic compounding factors, we also include macroeconomic level controls: the credit spread (CreditSpread), which equals the difference between average AAA-rated corporate bond spreads and the average BBB-rated corporate bond spreads; the term spread (TermSpread), which equals the difference in yield spreads between 10-year Treasury bond and 3-month Treasury bills; the GDP growth (GDPGrowth), which equals the annual average of GDP growth. Our final sample consists of 32,751 loan facilities to 2,883 public firms in the U.S. B. Summary Statistics We now report the summary statistics in Table 1. We tabulate the loan level summary statistics, including the All-in Drawn Spreads, the logarithm of loan spreads, the number of covenants, the covenant indicator dummy; the firm default indicator; the Third-Party Network measure, including the number of third-party connection, and the number of board connections. The detailed variable definitions are shown in the Appendix. 3. Third-Party Network and Contractual Provisions We now report the main results. We start by looking at whether Third-Party Network affects the main contractual provisions. We first consider the link between loan spreads and Third- Party Network and then address the issue of endogeneity. A. Third-Party Network and Loan Spreads 11

13 We start by relating Third-Party Network and Bank Loan Spreads. We estimate: Log(allindrawn spreads)i,f,t = α + β1 Thirdparty networki,t 1 +β2 FirmCharateristicsi,t 1 +β3 MacroVariablest 1 + β3 LoanCharateristicsf + Industryj +Yeart + Bankh,t + εi,f,t (1) where Log(allindrawn spreads) i,f,t equals the logarithm of one plus the all-in-drawn spreads in basis points reported from the Dealscan data for firm i, facility f, in year t. Third-party network i,t 1 is our focus variable and Controls i,t 1 includes a list of lagged control variables. They include: the logarithm of firm total assets (Size), profitability (Profitability), the fraction of tangible assets to total assets (Tangibility), market-to-book (M/B), financial leverage (Leverage), the S&P rating indictor (Rating dummy), cash ratio (Cash ratio), and the Altman s Zscore (Zscore). The macroeconomic level controls include: the credit spread (CreditSpread), which equals the difference between average AAA-rated corporate bond spreads and the average BBB-rated corporate bond spreads; the term spread (TermSpread), which equals the difference in yield spreads between 10-year Treasury bond and 3-month Treasury bills; the GDP growth (GDPGrowth), which equals the annual average of GDP growth. The loan level controls include: the logarithm of loan maturity in years (LoanMaturity), the logarithm of loan amounts in millions of dollars (LoanSize), and the type of loan (LoanType) equals one if the loan is a term loan, and zero if the loan is a revolver. These variables are defined in the Appendix. We control for industry and year fixed effects and cluster the standard errors at the firm level in all regressions. We report the results in Table 2. Our main sample includes firms that have financial third-party connections, which contains only the indirect connection between a firm and a bank. Columns (1)- (5) display the regression results from various specifications by including the firm-level controls, 12

14 the macro-economic controls, the industry and year fixed effects, the firm fixed effects, and bank fixed effects. The results show a strong negative correlation between Third-Party Network and the loan spreads. This negative relation holds across all the specifications and it is not only statistically significant, but also economically relevant. If we focus on the main sample, we see that one standard deviation higher network affiliation translates in between and lower loan spreads. B. Addressing Endogeneity Firms that are part of social networks can have characteristics that make them different from the other firms that are not part of the same network. The OLS regressions are subject to endogeneity concerns that make the relation between indirect social networks and loan spreads difficult to interpret. In the cross-section, board interlocks may vary across firms with different fundamentals (e.g., firms with better performance are more likely to be connected to third parties) and those differences in performance may be responsible for any differences in the cost of bank loans. We therefore need to find an exogenous driver in third part network that allows us to isolate the impact of the third part network from spurious effects. To address the endogeneity of board composition, we isolate a subset of changes in indirect network connectedness that are unrelated to the borrower s performance. Specifically, we consider decreases in interlock connectedness due to director deaths and retirements of the third party instead of the borrower. The intermediary firm s deaths and retirements of directors satisfy both the exclusion restriction and inclusion restriction. The idea is that death or retirement break the established ties, which serves as an exogenous shock to the network connectedness and death is not spuriously linked the borrower s performance. From the previous illustration, if the board interlock gets disrupted by the death a director serving the boards of the intermediary Firm C (but 13

15 not the borrower Firm A), only the fundamentals of the intermediary will be affected. However, neither the fundamentals of Firm A nor the riskiness of the loans borrowed by Firm A are directly affected, because these retired directors do not work for firm A (exclusion restriction). What gets disrupted is the connection between firms and their third-parties, which affects the informal channel to pass soft information from the Firm A to the Bank B via the intermediation (inclusion restriction). Director death or retirement of the intermediary can serve as a reasonable shock to introduce exogenous variations into third-party networks. Our main identification strategy is to compute the difference in loan spreads around the death or retirement events from the third party. An advantage of our research question is that we can utilize the natural sets of treatment and control events from the intermediary. In our sample, there are 1,281 death and retirement events in which an indirect network is served. First, to obtain the control firms where deaths or retirements do not affect the indirect linkages, we track the borrower and the intermediary are not connected through board interlock but the intermediary firm experiences director deaths or retirements. To ensure the treatment firms and the controls firms have similar characteristics ex-ante, we conduct a propensity-score matching based on the firmlevel control variables, and the two-digit industry classification. We also restrict the control sample to have bank loans issued prior to and after the death or retirement events in order to compare the differences in loan spreads before and after the events. To illustrate the impact of deaths or retirement of the third-party directors on loan spreads before and after the events, we plot the mean loan spreads for the 13 year window around the death or retirement of an independent director, where year 0 is the end of the fiscal year in which the death or retirement occurs. We plot separately paths for borrowers directors with and without direct network ties to the third-party directors that dies or retires. Specifically, mean loan spreads 14

16 are measured as the average loan spreads within the event windows [-6, -3], [-3,0], [0, +3], [+3, +6] for the event years respectively. "Connected" means the borrower as an existing board interlock connection with the third-party directors, where third-party network and board connections are greater than or equal to one. Among the connected directors, we observe s noticeable increases in loan spreads over the event windows after the deaths or retirements. Second, we conduct a difference-in-differences regression around third-party director death or retirement events with the following specification: Log(allindrawn spreads) i,f,t = α + β 1 Treat dummy i,jt 1 +β 2 After i,t 1 + β 3 Treat dummy i,j,t 1 After i,t 1 + γ Controls i,t 1 + ε i,t, (2) where Treat dummy i,j,t 1 equals one for the treated firm i if it is connected to the intermediary Firm j that experiences director deaths or retirements at t-1 and After i,t 1 is a dummy variable covering the period following the deaths or retirements. Controls i,t 1 includes a list of lagged control variables as in the previous OLS regressions. We report the difference-in-differences results in Table 4 for various event windows. We control for both the industry and year fixed effects and cluster the standard errors at the firm level in all regressions. Columns (1) displays the regression results within the [-10, +10] window. We estimate an overall increase in loan spreads in a magnitude of 0.08 among firms that lose a connected third-party director, which is statistically significant at the 1% level. By focusing on the interaction term between the treatment dummy and the event indicator after, we see that third-party connections is positively related to the loan spreads. This holds across all the specifications and it is not only statistically significant, but also economically relevant. These results help us assign a 15

17 causal interpretation to the negative link between third-party network and loan spreads, which confirm the benchmark results. The difference-indifferences results indicate significant increases in loan spreads around the death or retirement of directors from the intermediary but with network ties to the borrowing firms. The increase in the cost of financing is significant compared to the change in loan spreads around the death or retirement of unconnected directors. The difference-indifferences results are more pronounced for more matured firms, and for firms with existing issued bonds. In addition, the increases in loan spreads is stronger during periods of stress if the third-party firm experiences director death or retirement. These event studies suggest that the influence of third-party networks are causal on the price of loan. Overall, both the OLS regression and the difference-in-differences tests demonstrate that thirdparty network affects the contractual provisions of the lending contract reducing the interest rate (loan spreads). This is consistent with more intensive monitoring and information flow that allows the lenders to rely on instead of formal provisions. Alternatively, it is also consistent with the favoritism hypothesis that rates concessions reflect sweetheart deals lenders offer. In the next section we will try to distinguish these two hypotheses. 4. Information or Favouritism? In this section, we directly investigate whether Third-Party Network increases the ability of the bank to screen and monitor and therefore reduces the probability of distress of borrowers, or just raises the confidence in the firm without this being justified in terms of better information. A. Third-Party Network and Loan Defaults 16

18 Previous results show that the greater the link between third party and lender, the less the lender requires to be compensated for the riskiness of the borrower. This may be due to better information or just higher confidence. We distinguish between the information hypothesis and the favoritism hypothesis in the following tests. In particular, we investigate whether the third-party network increases the ability of the bank to screen and monitor and therefore reduces the probability of distress of the borrower, or just raises the confidence in the firm regardless of the firm fundamental information. Under the information hypothesis, we expect that the third-party network predicts less default in general and less default per-rate. In the second case, we expect that third-party networks predict more default in general and more default per-rate. We consider the direct link between borrower defaults and indirect social networks. We therefore estimate: Default dummy i,t+1 = α + β Third-party network i,t 1 + θ Controls i,t 1 + ε i,t, (3) where Default dummy I,t+1 equals the probability of bankruptcy by the S&P for firm i in year t. All the other variables are as in the previous table and as defined in the Appendix. We control for industry and year fixed effects and cluster the standard errors at the firm level in all regressions. The regression results on loan outcomes are reported in Table 5. Columns (1)-(4) present the regression estimates with the probability of entering default by the S&P in year t+1 as the dependent variable; Columns (5)-(8) present the regression estimates with the probability of entering default by the S&P in year t+2 as the dependent variable. The results show that in general that indirect social networks lead to increases in the probability of default in general. More specifically, both the third-party network measured by all types of connections or board connections measured by interlock through independent directors are associated with greater loan failures. These results suggest that Third-Party Network assuages lender s concerns and induces them to lend at a lower rate but risky loans. This also induces to sub-optimally screen/monitor so 17

19 that the Third-Party Network-related loans are more likely to go sour and the borrowers to lose value. Besides the benefits from loan pricing we document in previous section, a natural question that arises is whether the positive effect of indirect networks on loan contracts persists beyond 1 year. The tests of persistence look at the continuation of bank-firm relationship. We further examine the impact of third-party networks on length and depth of the relationship with banks. The regression results are reported in Table 6. The dependent variable is future bank relationships, which is measured by the probability of continue a lending relationship (Future relation dummy), the total amount of lending extended by the same bank scaled by the current amount of lending (Future relation amt), and the number of bank loans offered by relationship lenders in the future (Future relation num). The positive and significant coefficient on both Third-party network and Board connections in all columns indicate that, ceteris paribus, borrowers with extensive indirect networks increase the likelihood of forming lending relationship by 11% with the indirectly connected banks in the subsequent years. Firms with indirect connections to financial intermediaries experience increase in the loan amount and the number of loan facilities issued by the relationship lender. The results on subsequent relationship lending are consistent with the favoritism hypothesis as lenders offer sweetheart deals. Together, we demonstrate that benefits associated with borrower indirect networks in lower loan spreads and the continuation of firmbank relationships despite the high likelihood of future default. Finally, in an unreported table, we also look at whether Third-Party Network has predicted power with respect to the performance of the firm. We therefore estimate a similar specification but replacing the dependent variable with measures of firm value. We consider two proxies. The first is market-to-book and the second is the amount of cash holding of the firm. The results show 18

20 that Third-Party Network is negatively related to firm value and positively related to cash holdings. These results confirm our working hypothesis suggesting that Third-Party Network proxies more for trust and confidence than actual information. Firms whose borrowing is linked to Third-Party Network are more likely to enter distress and lose value. These firms are also more likely to hoard cash as a precautionary motive if lenders decide to increase loan spreads or tighten covenants in the long-run. B. Third-Party Network and Covenant Protection Next, we ask whether Third-Party Network, by affecting the trust of the lender changes the contractual provisions of the contract. We focus on covenants. We expect that Third-Party Network will reduce the need for the covenants. We therefore estimate: Covenants i,f,t = α + β Third-party network i,t 1 + θ Controls i,t 1 + ε i,t, (4) where Covenants i,f,t refers to the number of covenants. We consider alternative proxies: the number of covenants, the existence of any covenant, the existence of capital expenditure covenant, the existence of Debt-to-Ebitda covenant, and the existence of net worth covenant as reported from the Dealscan data for firm i, facility f, in year t. All the other variables are as in the previous table and as defined in the Appendix. We control for industry and year fixed effects and cluster the standard errors at the firm level in all regressions. We report the results in Table 7. The dependent variable in Columns (1)-(3) is the number of financial covenants. The dependent variable in Columns (4)-(6) is the total number of financial covenants, which includes various performance related covenants. We see that Third-Party Network is always negatively related to the presence of covenants. In particular, Third-Party Network reduces by 11.2% of the mean number of covenants, by 3% of the existence of any 19

21 covenant. These results confirm our working hypothesis that Third-Party Network by providing further information on the lender increases trust and therefore reduces the need for bank monitoring by imposing strict covenants. These results suggest that Third-Party Network induces them to lend at a lower rate and more covenant-lite. These evidence suggests that indirect social networks lead to insufficient monitoring and information flow on bank loans, which supports the favoritism hypothesis. One of the important role of financial intermediary is to monitor managers and exercise control rights if a firm violates covenants. Conversely, indirect networks through board interlock connection may prevent efficient monitoring by banks and produce less information, which can potentially destroy firm value. In this section, we test whether indirect network ties between borrowers and third parties affect monitoring intensity by examining the ex-post creditor control rights. In particular, we focus on the likelihood of covenant renegotiations if firms violate loan covenants and become technical default. Loan covenant renegotiations have important implications to corporate governance and firm value when borrowers are in distress. First, covenant renegotiations provide a direct measure of creditor monitoring intensity upon covenant violations. Second, how creditors resolve borrower distress can have substantial consequences for firm shareholders. The literature documents the economic implications associated with creditor restructuring when firms file for Chapter 11. Third, firms are required to disclose the detail about loan renegotiations when violating any of the loan covenant, which allows us to evaluate the consequence of loan renegotiations. In Table 8, we analyze the effect of indirect network ties on the probability of subsequent loan renegotiations. We estimate both the OLS (Columns 1 and 3) and the probit (Columns 2 and 4) regression in which the outcome variable indicates whether creditors agree to renegotiate the loan 20

22 contract upon covenant violations. In Columns 2 and 4, we find a modest positive effect of indirect network on the likelihood of lender renegotiations. In terms of the economic magnitude, the odd of a loan renegotiation is 15% higher for a company with a one standard deviation higher percentage of connections. In all specifications, we control for industry and year fixed effects. In an unreported table, the regression estimates are also robust after controlling for lender fixed effects. These fixed effects capture time-invariant differences across firms and banks may affect the likelihood of renegotiations. The higher probability of loan renegotiation by creditors is consistent with the previous findings on the loosening of covenant at the loan initiation. Together, these results support the argument that lenders have incentive to reduce monitoring initiatives both ex-ante at the loan initiation and ex-post when borrowers are in distress. Dennis and Wang (2014) document that debt covenants are frequently renegotiated, and these renegotiations primarily relax existing restrictions. These results strengthen the interpretation by providing direct evidence of weaker monitoring by the financial intermediary with more connections to the third party. 5. Characteristics of Indirect Social Networks A. The lending relation between banks and the intermediary We next try to explore whether the negative relation between the third-party network and loan spreads is driven by the size of the link between the lender and the third party. We start by looking at whether the impact of Third-Party Network is related to the size of the link between lender and third party. As we have argued, the information-role of the Third-Party Network should be stronger the bigger the link between the third party and the lender. The bigger the size, the more likely 21

23 information flows and confidence arises. Indeed, we know from the standard banking literature arguing that the more significant a loan is for the borrower, the more influence the lender has over the managers and the better the information (Rajan (1992)). In our case, the significance of the loan with the third party increases the influence and information of the lender over the third party, and therefore the reliance on the information the lender can extract. We proxy this link by resorting to standard banking literature and we consider three proxies. The first proxy is a measure of relationship lending: 3rd party-bank Rel. Following Bharath, Dahiya, Saunders, and Srinivasan (2011), we define relationship borrower dummy variable equals one if the third-party firm is a relationship borrower of the bank from year t-5 to year t, and equals zero otherwise. The intuition is that, if the third-party has had a longer relationship with the bank, the bank can trust its information more. The second proxy is: 3rd party-bank Amount. It equals the amounts of bank loans that the third-party obtained in the past five years from a lender divided by the total amounts of bank loans from all lenders in the past five years. The intuition is similar: if the third-party represents a significant stake of the bank s lending, the bank has already spent resources to learning about it and therefore can trust its information more. The third and fourth proxies are: 3rd party-bank Rel dispersion and 3rd party-num Rel lenders, which equal the standard deviation of the relation dummy with other lenders that the third party obtained loans in the past five years and the number of other relationship lenders that the third party obtained loans in the past five years, respectively. This variable is inversely related to the degree of information/trust of the bank on the third party. Indeed, the more fragmented is the borrowing of the third party, the more each bank will rely on the other banks to monitor and screen it and therefore the lower the information the bank would have on it. We therefore estimate: Log(allindrawn spreads) i,f,t = α + β 1 Third-party network i,t 1 22

24 +β 2 3rd party-bank Relat i,t 1 + γ Third-party network i,t 1 3rd party-bank Relat i,t 1 + θ Controls i,t 1 + ε i,t, (5) where Log(allindrawn spreads) i,f,t equals the logarithm of one plus the all-in-drawn spreads in basis points reported from the Dealscan data for firm i, facility f, in year t. Third-party network i,t 1 is the lagged third-party connection through board interlock. Controls i,t 1 includes a list of lagged control variables as shown in the Appendix. 3rd party-bank Relat i,t 1 represents our proxy for relationship between bank and third party. We control for industry and year fixed effects and cluster the standard errors at the firm level in all regressions. We report the results in Table 9. In column (1), we report the results for the first variable (3rd party-bank Rel), while in Columns (2), (3) and (4) present the regression estimates for an alternative measure of bank connection with the third party (3rd party-bank Amount, 3rd partybank Rel dispersion, and 3rd party-num related lenders) respectively. The results confirm the strong negative correlation between loan spreads and Third-Party Network. Such a relationship is reinforced in the case of a closer relationship between bank and third party. In the case of 3rd party-bank Relation (column (2)), the fact that the bank has a closer relationship with the third party makes Third-Party Network reduce the loans spread by a further 8.9% (16.8 bps). Similarly, in the case of 3rd party-bank Amount (column (4)), higher loan size issued to the third party by the lender in the past makes Third-Party Network reduce the loans spread. In contrast, Third-Party Network is associated with lower loan spreads in the case of higher dispersion in borrowing by the third party (column (6)). There is no direct effect in terms of number of relationship by the third party. 23

25 Intuitively, we expect the impact to be stronger when there is less information (i.e., less mature, younger firms, firms with no bonds outstanding) as well as during periods in which information is worth more (i.e., periods of stress). B. The Credit Market Distress, Third-Party Network, and Loan Spreads In this section, we examine whether the negative relation between indirect social and loan spreads is more pronounced during the normal period or financially distress times. During hard times, it can be beneficial to have close connections with other firms which established strong bank relationship in the past, which facilitates information flow. However, the intermediary can prevent the flow of information to banks if the borrowers default risk increases. The intuition is that better flow of information on borrowers credit risk can potential trigger higher loan spreads to the intermediary especially they are suppliers and customers. Campello and Gao (2017) demonstrate that higher customer concentration increases interest rate spreads and the number of restrictive covenants. We examine the relation between indirect networks and loan spreads during financial distress periods versus normal periods. This allows us to study whether the lenders are able to price in the increases in default risk through the indirect networks formed by executives and directors. The results are shown in Table 10. We separate columns into distress periods versus nondistress periods. The Distress periods are defined as the periods with greater than the median value of commercial paper spreads from the Federal Reserve, while the Non-stress periods are defined as the periods with less than the median value of commercial paper spreads. The results still show a strong negative correlation between Third-Party Network and loan spreads. This holds across all the specifications and it is not only statistically significant, but also economically relevant. The negative relation is more pronounced during the distressed period, we see that one standard 24

26 deviation higher network affiliation translates in between bps and bps lower spreads. Oure findings are consistent with Babina, Garcia, and Tate (2017) which demonstrate that firms with more direct connections on the eve of 1929 financial market crash have higher 10-year survival rates during the Great Depression. First, our results differ from Babina, Garcia, and Tate (2017) as we examine the impact of indirect board interlocks on bank loan pricing. Second, Babina, Garcia, and Tate (2017) show the positive effect associated with networks through increases in survival rates. However, we demonstrate a negative effect--banks offer favorable loan terms to firms with high default risk, which is associated with inadequate monitoring. C. The Direct Network versus the Indirect Network D. Robustness Checks The findings in previous sections indicate that the indirect network between the intermediary and the borrower weakens creditors incentive to conduct monitoring. Still, it is an open research question whether lenders react to changes in borrowers fundamentals given the existence of indirect networks. To provide further evidence on the favoritism hypothesis, we analyze the sensitivity of loan spreads to borrower performance. Since banks often have information advantage over market participants on borrower fundamentals, we use shocks to borrower credit risk from their customers or suppliers to pin down the specific mechanism. Table 11 examine the interaction effect of the third-party network and shocks to borrowers' fundamentals on loan pricing. The regression results illustrate whether lenders react to changes in the stock returns of borrowers' customers or suppliers. The Ret lag 1 quarter, Ret lag 2 quarter, Ret lag 3 quarter, and Ret lag 4 quarter indicate whether a borrower's customers or suppliers experience a negative or positive shock in the lagged one quarter, lagged two quarters, lagged three quarters, and lagged four quarters respectively. We measure the quarterly changes 25

27 in customers or suppliers returns suing the daily abnormal returns as the return to the supplier s stock minus expected returns from a market model. The interaction term captures the marginal effect of the interlock connection for unexpected changes in borrowers' fundamentals on loan spreads. The coefficient estimates on the interaction terms are insignificant in all of the columns. These results suggest that lenders fail to price in changes in borrower credit risk if borrowers and the third-parties are connected through board interlocks. The insignificant results are consistent with the evidence on the weakening of creditor monitoring intensity. Overall, our results mirror the main findings on the value reduction of indirect networks on loan outcomes in the syndicated loan market. We provide a specific mechanism through which indirect networks destroy corporate governance by loosening loan covenants and initiate more renegotiations. Finally, we provide other robustness checks by re-estimating the main specification with an alternative proxy of Third-Party. Network Boardmgt connections i,t 1 measures the number of connections if the manager of one firm serves as the director of another firm. This variable is an even stronger proxy of connections as it requires the direct cross-board affiliation. Our sample includes the sample of firms that have financial third-party connections, which contains the direct connection between a firm and a bank when a board director is also a director of the bank that obtains loans. As in the previous specifications, we control for industry and year fixed effects and cluster the standard errors at the firm level in all regressions. In an unreported table, we demonstrate a strong negative correlation between indirect social networks and loan spreads. In particular, one standard deviation higher number of connections (i.e., compared to an average of 12.1) translates in between 36.6 bps and 42.7 bps lower loans spreads in the whole sample, between 31.2 bps and 32.2 bps lower spreads for mature firms,

28 bps lower spreads for firms that have issued bonds previously, between 30.2 bps and 48.5 bps lower spreads during distress and non-distress periods respectively and 35.7 bps lower spreads for firms that have financial third-party connections, which contains the direct connection between a firm and a bank when a board director is also a director of the bank that obtains loans. These results confirm the previous ones and further support evidence in favor of the working hypothesis of a direct impact of Third-Party Network on lending. 6. Conclusion In this paper, we study how indirect social networks affect the lending market. We focus on how the connections between the borrowers and a third party affect lenders debt contracts. We consider whether the firms share the same director, or the director of one firm serves as the director of another firm, or the manager of one firm serves as the director of another firm. We use the sample of bank loans issued by public firms in the US over the period We articulate our argument in three parts. First, we document a strong negative correlation between Third-Party Network and the lending spreads. Higher network affiliation translates in lower spreads and lower reliance on covenants. The impact is stronger when there is less information (i.e., less mature, younger firms, firms with no bonds outstanding) as well as during periods in which information is worth more (i.e., periods of stress). Second, using an identifying restriction based on sudden executives deaths from the third-party, we provide evidence that link from Third- Party Network to contractual characteristics is casual. We document that, in line with both hypotheses, the impact of Third-Party Network is related to the size of the link between the lender and the third party. The bigger the size of the borrowing position of the third party vis-à-vis the lender, the closer is its lending relationship and the more 27

29 exclusive it is, the higher the impact of Third-Party Network on the lending conditions. In line with the confidence hypothesis, Third-Party Network is related to more default in general. Furthermore, we ask whether this is more information or just more confidence i.e., we try to distinguish between the two hypotheses. The information hypothesis posits that Third-Party Network helps the lender to know and monitor the borrower better. This allows the lender to lend at a lower rate and with less formal contractual protections: i.e., lower loan spreads and more covenant-lite contracts. The better information/power translates in better screening of loans and less loan failures, which also implies that in the presence of Third-Party Network the borrowing firms are less likely to face distress or turn sour. The alternative favoritism hypothesis posits that Third-Party Network just increases the confidence on the borrower s credit merit without any well-grounded justification for it. This leads to lower loan spreads and more covenant-lite contracts, but also to more loans defaults. Overall, our results suggest that Third-Party Network reduce lender s concerns and induces them to lend at a lower rate and more covenant-lite, which supports the favoritism hypothesis. The sub-optimal monitoring indicates that Third-Party Network-related loans are more likely to go sour and lenders suffer losses. Given that Third-Party Network makes lenders less oblivious of the real conditions of the firm, we expect it to delay the adjustment of loan spread to the deterioration of the firm s conditions. 28

30 References Ahern, Kenneth, Daniele Daminelli, and Cesare Fracassi, Lost in translation? The effect of cultural values on mergers around the world. Journal of Financial Economics 117, Bandiera, Oriana, Iwan Barankay, and Imran Rasul, Social connections and incentives in the workplace: Evidence from personnel data. Econometrica 77, Berger, Allen N., Nathan Miller, Mitchell Petersen, Raghuram Rajan, and Jeremy Stein, Does function follow organizational form? Evidence from the lending practices of large and small banks. Journal of Financial Economics 76, Berger, Allen N., and Gregory F. Udell, Relationship lending and lines of credit in small firm finance. Journal of Business 68, Bharath, S., S. Dahiya, A. Saunders, and A, Srinivasan, So what do I get: A bank s view of lending relationships? Journal of Financial Economics 85, Bharath, Sreedhar T., Sandeep Dahiya, Anthony Saunders, and Anand Srinivasan, Lending relationship and loan contract terms. Review of Financial Studies 24, Boot, A., Relationship banking: what do we know? Journal of Financial Intermediation 9, Burchardi, Konrad B., and Tarek A. Hassan, The economic impact of social ties: Evidence from German reunification. Quarterly Journal of Economics 128, ECB, 2007, EU banking structures, Discussion paper European Central Bank, Frankfurt. Chiu, P.C., Teoh, S.H., Tian, F., Board interlocks and earnings management contagion. Accounting Review 88, Cohen, Lauren, and Andrea Frazzini, Economic links and predictable returns. Journal of Finance 63, Cohen, L., Frazzini, A., Malloy, C., The small world of investing: Board connections and mutual fund returns. Journal of Political Economy 116, Cohen, Lauren, and Dong Lou, Complicated firms. Journal of Financial Economics 104, Cohen Lauren, Umit G. Gurun, and Christopher Malloy, Resident networks and corporate connections: Evidence from World War II Internment Camps. Journal of Finance, forthcoming. Coval, Joshua D. and Tobias Moskowitz, Home bias at home: Local equity preference in domestic portfolios. Journal of Finance, 54, Coval, Joshua D. and Tobias Moskowitz, The geography of investment: Informed trading and asset prices. Journal of Political Economy, 109, Dass, N. and M. Massa, The impact of a strong bank-firm relationship on the borrowing firm. Review of Financial Studies 24, Diamond, D., Financial intermediation and delegated monitoring. Review of Economic Studies 51,

31 Diamond, D., Monitoring and reputation: the choice between bank loans and directly placed debt. Journal of Political Economy 99, Engelberg, Joseph, Pengjie Gao, and Christopher A Parsons, Friends with money. Journal of Financial Economics 103, Fisman, Raymond, Daniel Paravisini, and Vikrant Vig, Cultural proximity and loan outcomes. Working paper, London Business School. Fracassi, C., Corporate finance policies and social networks. Working Paper, University of California, Los Angeles. Fracassi, C. and Tate, G., External networking and internal firm governance, Journal of Finance 67, Garmaise, M.J., Moskowitz, T, Informal financial networks: Theory and evidence. Review of Financial Studies 16, Gompers, Paul, Vladimir Mukharlyamov, and Yuhai Xuan, The cost of friendship. Discussion paper NBER. Granovetter, Mark, The impact of social structure on economic outcomes. Journal of Economic Perspectives 19, Guiso, Luigi, Paolo Sapienza, and Luigi Zingales, The role of social capital in financial development. American Economic Review 94, Jackson, C. Kirabo, and Henry S. Schneider, Do social connections reduce moral hazard? Evidence from the New York City taxi industry. Discussion paper NBER. James, Christopher, Some evidence on the uniqueness of bank loans. Journal of Financial Economics, 19, Haselmann, R., D Schoenherr and V. Vig, Rent-seeking in elite networks, Journal of Political Economy, forthcoming. Jensen, Michael C., and William H. Meckling, Theory of the firm: Managerial behavior, agency costs and ownership structure, Journal of Financial Economics 3, Karlan, Dean, Using experimental economics to measure social capital and predict financial decisions. American Economic Review 95, Markus Mobius, Tanya Rosenblat, and Adam Szeidl, Trust and social collateral. Quarterly Journal of Economics 124, Jackson, Matthew O., and Brian W. Rogers, Meeting strangers and friends of friends: How random are social networks? American Economic Review 97, Jackson, Matthew. O., Brian W. Rogers, and Yves Zenou, Economic consequences of social network structure. Working paper. Khwaja, Asim, and Atif Mian, Do lenders favor politically connected firms? Rent provision in an emerging financial market. Quarterly Journal of Economics 120, Kramarz, Francis, and David Thesmar, Social networks in the boardroom. Journal of the European Economic Association 11,

32 La Porta, Rafael, Florencio Lopez-de Silanes, and Guillermo Zamarripa, Related lending. Quarterly Journal of Economics 118, Larcker, D.F., So, E.C., Wang, C.C.Y., Boardroom centrality and firm performance. Journal of Accounting and Economics 55, Lazarsfeld, Paul. F., and Robert. K. Merton, Friendship as a social process: A substantive and methodological analysis, in Theodore Abel Morroe Berger, and Charles H. Page, ed.: Freedom and Control in Modern Society. pp (Van Nostrand: New York). Lerner, Josh, and Ulrike Malmendier, With a little help from my (random) friends: Success and failure in post-business school entrepreneurship. Discussion paper NBER. Mian, Atif, Distance constraints: The limits of foreign lending in poor economies. Journal of Finance 61, Nini, G., Smith, D., Sufi, A., Creditor control rights and firm investment policy. Journal of Financial Economics 92, Petersen, Mitchell A., and Raghuram G. Rajan, The effect of credit market competition on lending relationships. Quarterly Journal of Economics 110, Petersen, M., Rajan, R., The benefits of lending relationships: evidence from small business data. Journal of Finance 49, Qian, J., Strahan, P., How laws and institutions shape financial contracts: the case of bank Loans. Journal of Finance 62, Rajan, R.G., Insiders and outsiders: the choice between relationship and arm s length debt. Journal of Finance 47, Rauch, James, Networks versus markets in international trade. Journal of Economic Literature 48, Rauch, James, Business and social networks in international trade. Journal of Economic Literature 39, Santikian, Lori, The ties that bind: Bank relationships and small business lending. Working paper USC Marshall School of Business, University of Southern California. Sapienza, Paola, The effects of government ownership on bank lending. Journal of Financial Economics 72, Schmidt, B., Costs and benefits of friendly boards during mergers and acquisitions. Working Paper, University of Southern California. Sharpe, S., Asymmetric information, bank lending and implicit contracts: a stylized model of customer relationship. Journal of Finance 45, Shue, Kelly, Executive networks and firm policies: Evidence from the random assignment of MBA peers. Review of Financial Studies 26, Stein, J.C., Information production and capital allocation: Decentralized vs. hierarchical firms. Journal of Finance 57,

33 Stein, J., Agency, information and corporate investment. In: Con- stantinides, G., Harris,M., Stulz, R. (Eds.), Handbook of the Economics of Finance, vol. 1. North-Holland, Amsterdam, pp Stuart, T.E., Yim, S., Board interlocks and the propensity to be targeted in private equity transactions. Journal of Financial Economics 97, Sufi, A., Information asymmetry and financing arrangements: evidence from syndicated loans. Journal of Finance 62,

34 Appendix: Variable Definitions Definition Source Period Dependent variables Log(allindrawn spreads) The logarithm of the all-indrawn spreads for each loan facility. DEALSCAN Number of covenants The total number of financial covenants contained in a loan facility. DEALSCAN Covenant indicator A dummy variable equals one if a loan facility contains at least one types of financial covenant, and equals zero otherwise. DEALSCAN Default dummy Equals one if a firm receives a default rating, or files for Chapter 11, or get delisted, and equals zero otherwise. COMPUSTAT RATING Network variables Equals the logarithm of one plus the number of connections if Third-party network Boardmgt connections firms share the same director, or the director of one firm serves as the director of another firm, or the manager of one firm serves as the director of another firm. Equals the logarithm of one plus the number of connections if the manager of one firm serves as the director of another firm. BOARDEX BOARDEX Public information variables Issued bonds Mature firm A dummy variable that equals one if a firm has issued any bond prior to year t, and equals zero otherwise. A dummy variable that equals one if a firm s age is greater than or equal to the median asset value in the sample, and equals zero otherwise. DEALSCAN COMPUSTAT rd party-bank Rel A dummy variable that equals one if the third-party firm is a relationship borrower of the bank from year t-5 to year t, and DEALSCAN & BOARDEX equals zero otherwise. 3rd party-bank Amount Equals the amounts of bank loans that the third-party obtained in DEALSCAN & BOARDEX

35 3rd party-bank Rel dispersion 3rd party-num related lenders the past five years from a lender divided by the total amounts of bank loans from all lenders in the past five years. Equals the standard deviation of the relation dummy with other lenders that the third party obtained loans in the past five years. Equals the number of other relationship lenders that the third party obtained loans in the past five years. DEALSCAN & BOARDEX DEALSCAN & BOARDEX Control variables Size The logarithm of total assets. COMPUSTAT Profitability The return on assets is calculated as income before extraordinary items plus interest-related expenses, divided by total assets. COMPUSTAT Tangibility The property, plant and equipment divided by total assets. COMPUSTAT M/B The market-to-book ratio is calculated as the stock price times of number of shares outstanding, plus total assets minus the book COMPUSTAT /CRSP value of equity, divided by total assets. Leverage The sum of long-term debt and current debt divided by total assets. COMPUSTAT Rating dummy A dummy variable equals one if a firm has an S&P rating, and equals zero otherwise. COMPUSTAT Cash ratio The cash and short-term investments divided by current liabilities. COMPUSTAT M/B The market-to-book ratio is calculated as the stock price times of number of shares outstanding, plus total assets minus the book COMPUSTAT /CRSP value of equity, divided by total assets. Profitability The return on assets is calculated as income before extraordinary items plus interest-related expenses, divided by total assets. COMPUSTAT Zscore The Altman's Z score. COMPUSTAT CreditSpread The difference between average AAA-rated corporate bond spreads and the average BBB-rated corporate bond spreads. FEDERAL RESERVE

36 TermSpread The difference in yield spreads between 10-year Treasury bond and 3-month Treasury bills. FEDERAL RESERVE GDPGrowth Equals the annual average of GDP growth. IMF LoanMaturity The logarithm of loan maturity in years. COMPUSTAT LoanSize The logarithm of loan amounts in millions of dollars. DEALSCAN LoanType Equals one if the loan is a term loan, and zero if the loan is a revolver. DEALSCAN

37 Figure 1: Bank Loan Spreads around the Third-Party Director Deaths and Retirements Figure 1: Bank Loan Spreads around the Third-Party Director Deaths and Retirements. Mean Loan Spreads are measured as the average loan spreads within the event windows [-6,-3], [-3,0], [0,+3], [+3,+6] for the event years respectively. "Connected" means the borrower as an existing board interlock connection with the third-party directors, where the third-party network and board connections are greater than or equal to one.