Gambling in the Loan Market: Why Banks Prefer Overconfident CEOs *

Transcription

1 Gambling in the Loan Market: Why Banks Prefer Overconfident CEOs * Yehning Chen Department of Finance National Taiwan University Taipei City, Taiwan ynchen@ntu.edu.tw Po-Hsin Ho Department of Business Administration National Taipei University New Taipei City, Taiwan phho@mail.ntpu.edu.tw Chih-Yung Lin ** College of Management & Innovation Center for Big Data and Digital Convergence Yuan Ze University Taoyuan, Taiwan d @ntu.edu.tw Ju-Fang Yen Department of Statistics National Taipei University New Taipei City, Taiwan jfyen@mail.ntpu.edu.tw * We thank Yan-Shing Chen, Iftekhar Hasan, and Tse-Chun Lin for helpful comments and suggestions. ** Corresponding author: Tel.: ; Fax: address: d @ntu.edu.tw (C.-Y. Lin). 1

2 Gambling in the Loan Market: Why Banks Prefer Overconfident CEOs Abstract This paper examines whether banks charge higher or lower interest rates on loans to firms with overconfident CEOs. With a hedge against the downside risk of loan payments, banks favor firms with overconfident CEOs with the result that these borrowers enjoy lower loan rates especially when they have rich firm-specific growth opportunities or during prosperous periods. Evidences show that firms with overconfident CEOs bring more future business opportunities to banks, luring banks to relax lending standards for these firms. The implication is that banks may prefer high-risk borrowers if there is enough future benefit from doing businesses with these borrowers. JEL: G21, G32, G33, G34 Keywords: CEO overconfidence, bank loan contract, growth opportunities, downside risk. 2

3 1. Introduction Literature documents that banks gain from building a lending relationship with borrowers (see, e.g., Berger and Udell, 1995; Burch, Nanda, and Warther, 2005; Petersen and Rajan, 1994). A lending relationship not only helps the lender reduce information asymmetry, but also raises the lender s likelihood of attracting future business from the borrower (Bharath, Dahiya, Saunders, and Srinivasan, 2007; Burch, Nanda, and Warther, 2005; Drucker and Puri, 2005; Yasuda, 2005). 1 It implies that when a borrower s potential need for future business is greater, a high likelihood of winning future business is a more significant benefit to its relationship lender. Therefore, while banks are searching out potential borrowers to establish a lending relationship, borrowers prospects as well as their potential need for future business would be a critical factor. Most important decisions of a firm are made by its CEO. Their personal traits should be closely linked to the firm s prospects and thus matter to banks when making loan decisions. 2 A venture capitalist, Vinod Khosla, co-founder of Sun Microsystems Inc., has seen thousands of would-be revolutionaries and describes unbridled confidence and arrogance as key characteristics for being a visionary business leader. 3 However, the chance of great victory is always accompanied by the chance of big defeat. Facing these borrowers who potentially are a successful business visionary but comes with a certain degree of risk, are banks willing to take a chance with the future business and establish a lending relationship with them? The paper 1 Recent studies show that relationship lenders would be better placed to win future loan business (Bharath, Dahiya, Saunders, and Srinivasan, 2007) and other fee-generating services such as seasoned equity offerings (Drucker and Puri, 2005) and public debt underwriting (Burch, Nanda, and Warther, 2005; Yasuda, 2005) from its relationship borrower. 2 The common determinants of a business loan are the financial conditions of the borrower, which will affect the uncertainty of its cash flows as well as future loan payments. This study focus on an intriguing question is whether banks also care about any personal traits of borrowers managers when they make loan decisions, as emphasized in the traditional credit 5P or 5C rule. 3 Who Will Be the Next Steve Jobs?, Wall Street Journal, October 8, 2011, 3

4 investigates whether banks prefer overconfident borrowers. If it should be, how will that affect banks lending decisions, and how do lenders benefit from the gamble? The literature does not offer a clear answer to these questions. It is well documented that overconfident firms have a strong preference for investing. 4 Malmendier and Tate (2005, 2008) show that overconfident CEOs are convinced of their ability to generate higher returns on their investment projects, often resulting in overinvestment. Existing studies also show that overconfident CEOs have a greater likelihood of success in obtaining higher payoffs by choosing riskier projects when they enjoy rich growth opportunities (Goel and Thakor, 2008; Hirshleifer, Low, and Teoh, 2012; Ho, Huang, Lin, and Yen, 2016). 5 These studies suggest that overconfident CEOs are more effective in exploiting growth opportunities than other firms. Providing financing for these investments and advisory services is a lucrative business for banks. Hence, banks may want to establish lending relationships with overconfident firms, hoping to sell a succession of lucrative services in the future. One way to establish a lending relationship is to offer attractive loan contracts. Specifically, banks may be willing to charge lower rates on loans to overconfident firms than non-overconfident firms. However, the tendency to invest is a two-edged sword for overconfident firms. Overinvestment increases overconfident borrowers default probabilities as well as banks monitoring costs. 6 Research finds that banks 4 We call firms with overconfident CEOs overconfident firms and those whose CEOs are not overconfident non-overconfident firms. Research finds that overconfident firms have higher investment-cash flow sensitivity, pursue more mergers and acquisitions (M&As), and invest more and gain more success in innovation (see, e.g., Malmendier and Tate, 2005, 2008; Hirshleifer, Low, and Teoh, 2012; Ferris, Jayaraman, and Sabherwal, 2013). 5 Goel and Thakor (2008) show that the choice of a riskier project by an overconfident CEO increases his promotion probability because of the higher likelihood of extreme payoffs. Focusing on innovative industries that present risky and challenging projects, Hirshleifer, Low, and Teoh (2012) show that overconfident firms have greater return volatility, invest more in innovative activities, and also achieve greater success in innovations. Recently, using a sample of U.S. banks, Ho, Huang, Lin, and Yen (2016) find that CEO overconfidence leads to higher return on assets in a period of prosperity. 6 We provide a simple model in Appendix showing that compared with non-overconfident firms, overconfident firms have higher default probabilities, which raises banks monitoring costs. Empirical evidence shows that banks consider monitoring costs when they design loan contracts. 4

5 often require riskier borrowers to put up collateral or covenant to mitigate investment distortions and protect the downside risk of the loan payments (Berger and Udell, 1990; Chava and Roberts, 2008; Jimenez, Salas, and Saurina, 2006; Rajan and Winton, 1995). We thus conjecture that with the use of covenants or collateral agreements, banks care less about the high default risk of overconfident firms and would be willing to make sacrifices on loan spreads of overconfident firms for winning their lucrative business in the future. To investigate these issues, we collect CEO overconfidence data from publicly listed U.S. firms over We use a stock options-based proxy for CEO overconfidence and construct our measure using Standard & Poor s ExecuComp database. We use the same criteria as Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2011). If a CEO postpones exercising stock options that are more than 100% in the money at least twice during his or her tenure, we classify the CEO as overconfident the first time the exercise is postponed. The rationale behind the options-based measure is that a manager who chooses to hold deep-in-the-money stock options after the vesting period is likely overconfident about the firm s future prospects. 7 The sample of bank loan contracts come from DealScan; it comprises 12,917 bank loan contracts from 2,025 individual firms over Our empirical results show that over the sample period, overconfident firms on average enjoy a lower loan spread than non-overconfident firms after controlling for firm and loan characteristics and macroeconomic factors. We further find that the 7 The options-based CEO overconfidence measure has become widely used in recent empirical research (see, e.g., Malmendier and Tate, 2005, 2008; Campbell, Gallmeyer, Johnson, Rutherford, and Stanley, 2011; Malmendier, Tate, and Yan, 2011; Hirshleifer, Low, and Teoh, 2012; Ho, Huang, Lin, and Yen, 2016). 8 In our sample, loan characteristics include loan spread, loan maturity, loan size, collateral requirement, and numbers of covenants. We follow Roberts and Sufi (2009) and Roberts (2015) to construct the sample of financial contract renegotiations. 5

6 preferential effects of CEO overconfidence on loan spreads occur only in the subsamples in which loans are monitored through the use of covenants or collateral agreements. For example, overconfident firms with collateral agreements enjoy a reduction of in natural log of loan spread at the 1% level. As the average natural log of loan spread in our sample is , the borrowing cost of an overconfident firm is lower by almost 10 (e e ) basis points than non-overconfident firms. Meanwhile, as average loan size is about $ (e = ) million and the average loan s time to maturity is around 2.82 (e /12 = 2.82) years, overconfidence in a firm leads to a reduction about $0.7 million ( bps 2.82 yrs) in bank loan interest payment. The preferential effect of CEO overconfidence on loan spread is not only statistically significant but also economically important. The reduction level of 10 bps is around 70% of the estimated coefficient for the politically connected firm dummy variable in the study of Houston, Jiang, Lin, and Ma (2014) and is similar to the effects of repeat borrowing from relationship lenders in Bharath, Dahiya, Saunders, and Srinivasan (2011). 9 Moreover, we find that, conditioned on the presence of covenant or collateral agreements, banks favor overconfident firms with high growth opportunities most. For example, with the use of collateral agreements, overconfident firms facing high firm-specific growth opportunities enjoy a larger reduction in natural log of loan spread, which increases to from for the all sample. By contrast, we find no evidence of an association between overconfident firms and loan spread when firms have poor growth opportunities, even though the downside risks to loan payments are hedged. In addition, in the subsamples without loan covenants or 9 To distinguish the effect of OC on loan spread from the effect of relationship loans, we also add control for past relationship between lenders and borrowers in the previous one to five years in our analysis and the results on OC effect remain qualitatively unchanged. 6

7 collateral, there is no evidence showing that banks favor overconfident firms. This indicates that only when downside risks to loan payments are hedged, banks would consider taking a chance to establish lending relationships with overconfident firms. And the sort of overconfidence that banks prefer appears when CEOs face greater growth opportunities. Similarly, we find that banks favor overconfident firms more than other firms in prosperous years, which have plenty of growth opportunities. These results are consistent with our conjecture that banks view future business opportunities as a critical consideration in design of loan contracts. We then investigate whether overconfident firms do bring more future benefits to banks. We find that overconfident firms are more likely to get loans from the same lenders than non-overconfident firms, and this result is more significant for overconfident firms with high growth opportunities. We also find that overconfident firms in our sample grow faster than non-overconfident ones. They have higher asset and equity growth rates, undertake more acquisitions, and invest more in the one to three years after signing the loan contracts in our sample. It indicates that overconfident firms may have more financing needs in the future to support their fast growth, which would translate into new business opportunities for their relationship lending banks. 10 In sum, the evidence suggests that in the presence of great growth opportunities, banks can benefit from establishing lending relationships with overconfident firms through future lending business or other non-lending fee-generating services. Finally, we discuss whether banks soften their lending standards for overconfident firms. Our empirical results show that overconfident firms are more likely to enter into loan renegotiation than non-overconfident firms when banks have 10 Bharath, Dahiya, Saunders, and Srinivasan (2011) show that firms are more likely to borrow from banks with which they have an established lending relationship, because repeated borrowing usually results in better loan terms. 7

8 better downside protection and when the borrowing firms have more growth opportunities, which are the situations that overconfident firms enjoy lower loan rates. It means banks will relax their lending standards via approving lower lending rate to attract overconfident firms in exchange for future business. These results imply that while banks expect to obtain future benefits from doing business with overconfident firms, they also have to prepare to face a possible loss in loan renegotiation. Our work is closely related to Acharya and Naqvi (2016), who propose a theoretical model to discuss the investment preferences of managers. They show that the first priority of financial intermediaries with ample liquidity is to invest in risky projects to reap higher potential yields. Both Acharya and Naqvi (2016) and we conclude that banks may prefer riskier projects despite the higher default probabilities. Yet the two papers have different focuses, and the main features of the models are different. To study how borrower overconfidence affects loan contracts, we consider the future business that borrowers with high growth potential may bring to banks, while Acharya and Naqvi (2016) do not. Complementing Acharya and Naqvi (2016), we propose another reason for why banks may prefer riskier borrowers. Our work contributes to the literature in three ways. First, our research complements several recent studies on the determinants of bank loan contracting. These perspectives include reputation (Sufi, 2007); financial restatement (Graham, Li, and Qiu, 2008); accounting quality (Bharath, Sunder, and Sunder, 2008), ownership structure (Lin, Ma, Malatesta, and Xuan, 2011); bank relationship (Bharath, Dahiya, Saunders, and Srinivasan, 2011); political connections (Houston, Jiang, Lin, and Ma, 2014); and tax avoidance (Hasan, Hoi, Wu, and Zhang, 2014). We show that CEO traits of borrowing firms have an impact on loan contracting. Banks, for example, may favor firms with overconfident CEOs and offer them lower loan rates. 8

9 Second, our work relates managerial characteristics to firms financing policies. Few authors have looked at the influence of managerial characteristics on corporate financing policies. Malmendier, Tate, and Yan (2011) find that, compared with non-overconfident firms, overconfident firms prefer debt to equity when they raise external funds. Ho, Huang, Lin, and Yen (2016) show that overconfident banks add to their leverage in non-crisis periods. Our paper complements these studies by exploring how CEO overconfidence affects firms costs of loan financing. Third, our paper is also related to the literature about banks risk taking behavior in lending decisions. Thakor (2005) argues that for reputational reasons, banks tend to honor borrowing requests under commitments to lend in the future, raising the possibility of overlending during economic booms. Dell Ariccia and Marquez (2006) show that as banks obtain private information about borrowers and information asymmetries across banks are reduced, banks loosen lending standards. Acharya and Naqvi (2012) argue that bank loan officers have an incentive to increase lending by lowering lending rates if their compensation is linked to loan volume. By examining why banks give overconfident firms lower loan rates, our paper provides another scenario in which banks favor high-risk borrowers by approving lower loan rates and even loosening lending standards in exchange for future business opportunities. The paper is organized as follows. Section 2 develops the empirical hypotheses. Section 3 explains the overconfidence measure and data. Section 4 presents our main empirical results. Section 5 provides additional supporting evidence and addresses the endogeneity issues. Section 6 concludes. We propose a simple model in Appendix. 2. Hypothesis development This section provides literature review and presents hypothesis development of 9

10 this study. The hypotheses examine whether and when banks prefer overconfident firms. These hypotheses predict that with a hedge against the downside risk of the loan payments, banks charge lower interest rates on loans to overconfident firms, especially when these firms have rich firm-specific growth opportunities or during prosperous periods. Literature on managerial overconfidence show that overconfident CEOs overestimate their ability to generate returns on their investment projects, often resulting in overinvestment (see, e.g., Malmendier and Tate, 2005, 2008; Goel and Thakor, 2008; Campbell, Gallmeyer, Johnson, Rutherford, and Stanley, 2011; Ben-David, Graham, and Harvey, 2013). For example, Malmendier and Tate (2005) propose a model showing that overconfident firms overinvest compared to the first best level of investment in the equilibrium. Malmendier and Tate (2008) further suggest that overconfident CEOs overinvest in acquisitions and overpay for target companies. Ben-David, Graham, and Harvey (2013) present empirical evidence that firms with overconfident CEOs pursue more aggressive corporate policies: investing more and using more debt financing. This studies indicate that overconfidence, as a managerial trait, leads risk-averse CEOs to overinvest, which would incur high potential need of external financing (Malmendier, Tate, and Yan, 2011). 11 Providing financing for these investments and advisory services is a lucrative business for banks. To capture such lucrative business in the future, banks may want to establish lending relationships with overconfident firms. However, overinvestment may increase overconfident borrowers possibility of failure to meet their obligation as well as banks monitoring costs. Psychological 11 Malmendier, Tate, and Yan (2011) find that when overconfident CEOs overestimate the return on an investment but are unable to fully finance the investment with their own capital, they need to borrow external funds and prefer debt to equity. 10

11 literature indicates that people tend to be more overconfident about their performance in difficult rather than easy tasks (see, e.g., Griffin and Tversky, 1992), suggesting that overconfident CEOs are especially enthusiastic about risky and challenging projects that has higher chances of failure due to its dependence on various unpredictable conditions. Hirshleifer, Low, and Teoh (2012) show that overconfident firms have greater return volatility. We also provide a simple model in Appendix showing that overconfident firms have higher default probabilities than other firms. Collateral or covenants are used more intensively in loan contracts involving firms in need of monitoring (Rajan and Winton, 1995). Chava and Roberts (2008) discover that the use of covenants helps mitigate investment distortions. Research also finds that riskier borrowers are often required to put up collateral (Berger and Udell, 1990; Jimenez, Salas, and Saurina, 2006). With collateral or loan covenants, banks have better downside protection, so they care less about the higher default risk of overconfident firms and would be willing to charge them lower loan spreads than other firms in exchange for potential future benefits. A testable implication therefore is that when lending banks are hedged against the downside risk of the loans, overconfident firms are more likely to enjoy lower spreads than non-overconfident firms. We formalize this implication in Hypothesis 1: Hypothesis 1. Overconfident firms enjoy lower loan spreads if banks are hedged against the downside risk of the loans. Note that although overconfident firms are more likely to enjoy lower loan spreads than non-overconfident firms when collateral is pledged, this does not imply that overconfident firms have an incentive to pledge collateral to receive lower spreads. As Proposition 1 in Appendix shows, in our model firms pledge collateral only when doing so is necessary for them to get the bank loan. 11

12 As we discuss in the introduction, studies on lending relationship show that bank s gains from building a relationship with successful borrowers. A successful borrower grows fast and thus needs more financial services. Empirical evidence shows that, when borrowers need further financing, they often go to lenders with which they have lending relationships to obtain better loan terms and non-lending fee-generating services, translating into a succession of future business for its relationship lender (Bharath, Dahiya, Saunders, and Srinivasan, 2007; Burch, Nanda, and Warther, 2005; Drucker and Puri, 2005; Yasuda, 2005). Hence, banks may prefer overconfident borrowers more if there is more future benefit from doing businesses with these borrowers. Literature shows that, overconfident CEOs are more effective in exploiting growth opportunities than other CEOs in the presence of great growth opportunities. For example, Hirshleifer, Low, and Teoh (2012) focus on innovative industries, which are often rife with risky and challenging projects, and show that overconfident firms invest more and achieve greater success in innovations. Lending to them gives banks a better chance to reap the benefits from doing business with successful high-growth borrowers in the future. In addition, while Adrian and Shin (2009, 2010) find that firms need external financing when they see strong growth opportunities during a credit boom, we infer that overconfident firms would have greater need for external financing in prosperous years because these firms perceive business opportunities as stronger than they really are. Ho, Huang, Lin, and Yen (2016) find that managerial overconfidence on average leads to an increase in leverage and higher returns on assets in a period of prosperity. These studies suggest that in the presence of rich growth opportunities, overconfident firms can bring more future lending business or other fee-generating service to their relationship lenders than non-overconfident firms. 12

13 Therefore, if banks are willing to take a chance with future benefits and lend to overconfident firms, the preferential effect of CEO overconfidence on loan spread will be more significant when firms have more firm-specific or macro-level growth opportunities. That is, for a lender, the establishment of a lending relationship with overconfident firms is more lucrative than that with non-overconfident firms in the presence of great growth opportunities. We thus predict that overconfident firms are more likely to enjoy lower loan spreads when they have more growth opportunities. We capture this conjecture in Hypothesis 2: Hypothesis 2a. Overconfident firms enjoy lower loan spreads when they have rich firm-specific growth opportunities. Hypothesis 2b. Overconfident firms enjoy lower loan spreads when they have more macro-level growth opportunities. In Appendix, we propose a simple model and solve the equilibrium loan contract. Whether overconfident firms enjoy lower loan payments depends on the trade-off between high riskiness due to the nature of overconfidence and the future benefits from doing business with them. Our model shows that overconfident firms enjoy lower loan spreads than non-overconfident firms (i) if banks are hedged against the downside risk of the loan payments, and (ii) if there are ample growth opportunities. Note that both the loan spread and whether collateral is pledged are endogenously determined. The model implies that, when the downside risk of lending is reduced by collateral or loan covenants, banks will prefer high-risk borrowers because of expanded future business opportunities. This result is consistent with the prediction in Acharya and Naqvi (2016) that bankers prefer to invest in risky projects. 13

14 3. Overconfidence measure and data 3.1. Overconfidence measure We use a stock options-based proxy for CEO overconfidence, constructing the measure using the Standard & Poor s ExecuComp database between 1993 and We adopt the criteria for the CEO overconfidence indicator in Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2011). 12 We categorize three levels of managerial overconfidence: high, moderate, and low overconfidence, using 100% and 30% moneyness as the cutoff points. A CEO is identified as highly overconfident if he or she postpones the exercise of 100% in-the-money options at least twice during the tenure period. We assign such a CEO to the highly overconfident category the first time this behavior is observed. 13 A low-overconfidence CEO is a CEO who exercises stock options that are less than 30% in the money and does not maintain any exercisable options that are more than 30% in the money. As in the high overconfidence measure, we require that CEOs exhibit this conservative options exercise behavior at least twice during their tenures, and assign them to the low overconfidence category when they first exhibit this behavior. CEOs who hold or exercise options with moneyness between 30% and 100% are classified as moderately overconfident. Both Goel and Thakor (2008) and Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2011) have theoretically and empirically found that only highly overconfident CEOs exhibit overinvestment behavior. Therefore, we characterize 12 This is a modified version of the stock options-based overconfidence measure in Malmendier and Tate (2005, 2008). 13 We compute options moneyness as follows. Realizable values per option are estimated from the total realizable value of exercisable options divided by the number of exercisable options. Then, the estimated average exercise prices of the options are computed from the fiscal year-end stock price minus the realizable value per option. Hence, the percentages of average moneyness are obtained from the per-option realizable value divided by the estimated average exercise price. We employ a similar methodology to measure the percentage of moneyness of exercised options. 14

15 high-overconfidence CEOs as the overconfident group, and moderate- and low-overconfidence CEOs as the non-overconfident group Data We start with all U.S. companies whose data can be found in both the Standard & Poor s Compustat and ExecuComp databases. We exclude financial firms with SIC codes between 6000 and Due to data availability of the ExecuComp database, we choose 1993 as the first year of the sample. The bank loan data for our analysis come from Reuters DealScan database. 14 The final sample includes 12,917 bank loan contracts from 2,025 individual firms over We obtain accounting data of the firms from Compustat and stock returns from the Center for Research in Security Prices (CRSP). To prevent outliers from biasing the results, we winsorize all variables at 1% and 99%. We also follow Roberts and Sufi (2009) and Roberts (2015) to construct the sample of the renegotiation of financing contracts. CEO age is obtained from ExecuComp and macroeconomic variables from Datastream. Following Cao, Simin, and Zhao (2008), the growth opportunity variable is Tobin's Q [(price common shares outstanding + preferred stock + current liabilities current assets + long-term debt) / total assets]. A firm has more growth opportunities if its Tobin s Q is higher. The firm characteristic variables include Assets (natural logarithm of total assets of the firm); Leverage (long-term debt plus debt in current liabilities divided by total assets); Profitability (earnings before interest, taxes, depreciation, and amortization (EBITDA) divided by total assets); Tangibility (net property, plant, and equipment 14 DealScan provides loan characteristics including loan spread, loan maturity, amounts of loan size, collateral requirement status, numbers of all covenants, and purpose and type of loan. 15

16 divided by total assets); Cash flow volatility (standard deviation of quarterly cash flows from operations over the four fiscal years prior to the loan initiation year scaled by total debt); Z-score (modified Altman s Z-score, which equals [(1.2 working capital retained earnings EBIT sales) / total assets] ; and Rating (Standard & Poor s senior debt rating, converted into an index from zero to ten as follows: 10 = AAA, 9 = AA, 8 = A, 7 = BBB, 6 = BB, 5 = B, 4 = CCC, 3 = CC, 2 = C, 1 = D, and 0 = no rating). The loan characteristic variables are Spread (natural logarithm of all-in spread drawn from DealScan, which is the difference between the interest rate that the borrower pays and LIBOR or LIBOR equivalent in percentage points); Maturity (natural logarithm of loan maturity in months); Loan size (natural logarithm of amount of loan in US$ million); Collateral (dummy variable that takes a value of 1 if a loan is secured, and 0 otherwise); Syndicate (dummy variable that takes a value of 1 if a loan is a syndicate loan, and 0 otherwise); Performance (dummy variable that equals one if the loan facility uses performance pricing); FinCov (number of financial covenants); GenCov (number of general covenants); TotalCov (number of total covenants); Loan type (dummy variable for loan types, including term loan, revolving loans longer than one year, revolving loans shorter than 1 year, and 364-day facility); Loan purpose (dummy variable for the loan purposes, including e.g., corporate purposes, debt repayment, working capital, takeover, etc); and Renegotiation (dummy variable that takes a value of 1 if a firm renegotiates any of its loans with lenders during the contract period of the loan in our sample). The macroeconomic variables include Credit spread (difference between the AAA and BAA corporate bond yields) and Term spread (difference between the 10-year and 2-year Treasury yields). 16

17 Finally, the proxies for the future benefits for banks from building a relationship with borrowers include Assets, Equity, Acquisitions, and CE. Assets is the annual change in firm assets normalized by the previous year s asset value. Equity is the annual change in outstanding common equity normalized by the previous year s outstanding common stock value. Acquisition is acquisitions fees divided by total assets. CE is capital expenditures divided by total assets. Appendix B lists the definitions of all the variables employed in this paper Descriptive Statistics Table 1 presents the summary statistics of CEO overconfidence, growth opportunity variables, characteristics of borrowing firms and loan contracts, and macroeconomic factors. From Panel A, overconfident firms represent 32.16% of the total sample. In Panel B, the means of growth opportunity variable, Q, is In Panel C, the average Assets, Leverage, Profitability, Tangibility, Cash flow volatility, Z-score, and Rating values are , , , , , , and , respectively. In Panel D, the average bank loan spread is %, with a standard deviation of The average of Maturity, Loan size, Collateral, Syndicate, Performance, FinCov, GenCov, and TotalCov are , , , , , , , and , respectively. In Panel E, the average of Credit spread and Term spread are and , respectively. 15 [Insert Table 1] Table 2 compares the descriptive statistics of overconfident and non-overconfident firms. For the loan spread in Panel A, the mean values of overconfident firms and non-overconfident firms are % and %, and the 15 In Appendix C, we also present the Pearson correlation coefficient matrix of the variables over The correlation coefficients between OC and other variables are less than 0.26, so multicollinearity is seen as less of a concern. 17

18 difference between them is insignificantly negative. Panel B compares loan spreads according to different downside protection arrangements. In the subsamples with downside protection for banks, that is, FinCov > 0, GenCov > 0, TotalCov > 0, and Collateral = 1, the mean difference values between overconfident firms and non-overconfident firms are %, %, %, and %, all significantly negative. In the subsamples without downside protection for banks, FinCov = 0, GenCov = 0, TotalCov = 0, and Collateral = 0, the mean difference values are insignificant. The results suggest that overconfident firms can obtain lower bank loan spreads only when banks can hedge against the downside risk of the loan payments. [Insert Table 2] With respect to growth opportunity variable, overconfident firms are associated with larger Q, suggesting that overconfident firms have better growth opportunities than non-overconfident ones. For firm characteristics, overconfident firms are associated with smaller total assets, lower leverage, tangibility, and credit rating and higher profitability, cash flow volatility, and Z-score. Finally, regarding loan characteristics, loans to overconfident firms tend to have longer maturities and smaller sizes. They are more likely syndicated loans, and an incentive performance scheme is more likely used. In addition, loans to overconfident firms have more downside protection for banks in terms of Collateral, FinCov, GenCov, and TotalCov, which is consistent with other findings that collateral or covenants are used more intensively in loan contracts involving firms in need of monitoring (Rajan and Winton, 1995; Jimenez, Salas, and Saurina, 2006). The differences between overconfident firms and non-overconfident firms suggest it is important to control for these loan structure variables and other firm 18

19 characteristics when we study the relation between CEO overconfidence and bank loan spreads. 4. Empirical results 4.1. CEO overconfidence and bank loan spreads We use linear regressions to investigate the determinants of bank loan spreads. The analysis focuses on the coefficient of the CEO overconfidence dummy variable. A negative coefficient supports our hypothesis that overconfident firms enjoy lower loan spreads because they have greater growth potential that can benefit lending banks in the future. Following Graham, Li, and Qiu (2008) and Hasan, Hoi, Wu, and Zhang (2014), we use an ordinary least squares regression to investigate the effects of CEO overconfidence on loan spread: Spread i,t = α 1 + α 2 OC i,t 1 + β Z i,t 1 + ν i + μ t + ε i,t, (1) where Spread i,t represents the natural logarithm of bank loan spread for firm i in year t. OC i,t 1 is a dummy variable that equals 1 if firm i is an overconfident firm at time t 1 and zero otherwise. 16 Z i,t 1 is a vector of control variables for firm i in year t 1; ν i and μ t capture the firm and year fixed effects, respectively; and ε i,t is the random error. Z includes seven firm characteristics: Assets, Q, Leverage, Profitability, Tangibility, Cash flow volatility, Z-score, and Rating; seven loan characteristics: Maturity, Loan size, Collateral, Syndicate, Performance, FinCov, and GenCov; two macroeconomic factors: Credit spread and Term spread; and the 16 The variable overconfidence measure OC i,t 1 is time-varying in one of two ways. First, Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2011) classify CEOs as overconfident from the first time they exhibit the exercise-postpone behavior. Before this time, CEOs are classified as non-overconfident. Second, because banks may replace their CEOs, the attitude of a new CEO toward project risks may be different from the attitude of the predecessor. 19

20 purpose and type of loan. In all equations, we report t-values based on standard errors adjusted for heteroskedasticity and sample clustering at the firm level (White, 1980, and Petersen, 2009). To save space, we do not report the coefficients of loan purpose, loan type, and firm and year dummies. Table 3 presents the regression results for the effects of CEO overconfidence on loan spread. There are four specifications. The first controls for the loan purpose, loan type, firm and year fixed effects. The second adds the controls for the firm characteristics. The third adds the controls for the loan characteristics, and the fourth adds the controls for macroeconomic factors. Across all specifications, the proxy for CEO overconfidence, OC, is significantly negatively related to loan spread even when we control for other potential factors that may affect loan spread. For example, in Model (4), the coefficient of OC is and is statistically significant at the 1% level, which shows that overconfident firms enjoy lower loan spreads than non-overconfident firms on average. [Insert Table 3] With regard to firm characteristics, loan spread is significantly negatively associated with Assets, Q, and Profitability. The results are consistent with the notion that firms can reduce their financing costs when they are larger, and enjoy better growth opportunities, and higher profitability (Graham, Li, and Qiu, 2008). Higher loan spreads are charged to firms with high leverage and cash flow volatility, consistent with the notion that firms with higher risk are charged higher loan spreads (Chava and Roberts, 2008; Hasan, Hoi, Wu, and Zhang, 2014). The results also indicate that loans with longer maturities, larger amounts, and performance pricing have lower loan spreads. Covenants (GenCov) and collateral are positively associated with loan spread as well, consistent with the notion that firms 20

21 with higher risk will be asked to accept loan covenants or to provide collateral (Chava and Roberts, 2008; Graham, Li, and Qiu, 2008). In terms of macroeconomics, the results indicate that firms face higher financing costs when credit and term spreads are tightened. To sum up, after controlling for other potential factors that may affect loan spreads, managerial overconfidence among U.S. firms is generally associated with a lower cost of bank loan financing CEO overconfidence and downside protection But do banks always favor overconfident firms, firms that tend to invest in risky projects and are subject to more return volatility? Literature shows that banks can monitor, and even intervene in management, through the use of covenants (Rajan and Winton, 1995; Chava and Roberts, 2008). Chava and Roberts (2008) discover that the use of covenants helps mitigate investment distortions. Research also finds that riskier borrowers are often required to put up collateral (Berger and Udell, 1990; Jimenez, Salas, and Saurina, 2006). Given the literature and our model prediction, we expect that the preferential effect of CEO overconfidence on loan spread will be more effective when banks have downside protection on loan payments. Table 4 presents the regression results of CEO overconfidence on bank loan spread under different downside protection arrangements. In Models (2), (4), (6), and (8), the overconfidence proxy OC is significantly negatively related to loan spread at the 1% level. Specifically, compared to non-overconfident CEOs, CEO overconfidence translates into a to reduction in natural log of loan spread when there is collateral or covenants. For example, in Model (2), the coefficient of OC is , which means that overconfident firms enjoy a reduction of in natural log of loan spread over non-overconfident firms in the 21

22 subsample with the use of collateral, which is 1.50 times the coefficient of OC ( ) for the whole sample in Table 3. As the average natural logarithm of loan spread in our sample is about basis points, an overconfident firm reduces its bank loan interest cost about 10 (e e ) basis points. Meanwhile, as average loan size is about $ (e = ) million and the average loan s time to maturity is around 2.82 (e /12 = 2.82) years, overconfidence in a firm leads to a reduction about $0.7 million ( bps 2.82 yrs.) in bank loan interest payment. Besides, the converse is that in Models (1), (3), (5), and (7), OC is insignificantly related to loan spread, indicating that overconfident firms obtain lower bank loan spreads only in the subsamples of Collateral = 1, FinCov 1, GenCov 1, and TotalCov 1. That is, the preferential effects of CEO overconfidence on bank loan spreads are concentrated in the subsamples where loan repayment is monitored through the use of covenants or collateral agreements. The results support Hypothesis 1 that overconfident firms enjoy lower loan spreads if banks are hedged against the loans downside risk. [Insert Table 4] 4.3. CEO overconfidence and firm-specific growth opportunities We now turn to whether firm-specific growth opportunities influence the effects of CEO overconfidence on loan spread. Following Graham, Li, and Qiu (2008), Hirshleifer, Low, and Teoh (2012), and Hasan, Hoi, Wu, and Zhang (2014), we use the ordinary least squares regression: Spread i,t = α 1 + α 3 OC i,t 1 Q Hi,t 1 + α 4 OC i,t 1 Q Li,t 1 +β Z i,t 1 + ν i + μ t + ε i,t, (2) 22

23 where Spread i,t represents the natural logarithm of bank loan spread for firm i in year t. D Hi,t 1 and D Li,t 1 are dummy variables that equal 1 if firm i is a firm with high or low growth opportunities at time t 1 and zero otherwise; All the other independent variables are defined as before. Table 5 presents the regression results of CEO overconfidence on bank loan spread for firms under different downside protection arrangements from the viewpoint of firm-specific growth opportunities. For all sample in Model (1), OC Q H are significantly negatively related to loan spread at the 1% level, indicating that overconfident firms on average enjoy lower loan rates when they have rich growth opportunities. [Insert Table 5] In the subsamples of Collateral = 1, FinCov 1, GenCov 1, and TotalCov 1, the coefficients of OC Q H are significantly negatively related to loan spread at the 1% level. For example, conditioned on covenant or collateral agreements, overconfident firms with high growth opportunities enjoy a to reduction in loan spread over other firms, which are much greater than the estimated coefficients of OC in Table 4. We find no evidence, however, showing that banks favor overconfident firms in the subsamples of no loan collateral pledge or covenants even though growth opportunities are high. This indicates that only when downside risks to loan payments are hedged though the use of collateral or covenants, banks are willing to lower loan spread to overconfident firms with great growth opportunities. The result sheds light on the role of collateral and covenant in designing loan contracts to risky firms or firms in need of monitoring, which supports findings in Berger and Udell (1990), Rajan and Winton (1995), and Chava and Roberts (2008). 23

24 Besides, the coefficients of OC Q L in all Models are statistically insignificant by comparison. That is, there is no evidence showing an association between overconfident firms with poor growth opportunities and loan spread. This result implies that banks favor overconfident firms and offer lower loan spreads only when they have high growth opportunities. That is, banks prefer certain traits of overconfident CEOs and this trait appears much more when the CEOs face greater growth opportunities and could be restricted if a firm has poorer growth opportunities. Overall, the results support our Hypothesis 2a that overconfident firms enjoy lower loan spreads when they have rich firm-specific growth opportunities CEO overconfidence and macroeconomic conditions If banks prefer certain traits of overconfident CEOs and these traits appear only when CEOs face rich firm-specific growth opportunities, banks should favor overconfident firms more than other firms in prosperous times when there are plenty of growth opportunities. Ho, Huang, Lin, and Yen (2016) find that managerial overconfidence on average can lead to higher return on assets only during a period of prosperity, and that such firms are subject to poorer financial performance and greater expected default frequency during crisis years. This implies that the relation between managerial overconfidence and firm performance depends on market conditions. Thus, we investigate whether macroeconomic conditions affect the impact of CEO overconfidence on loan spread. Following Ivashina and Scharfstein (2010), Acharya and Naqvi (2012), and Fahlenbrach, Prilmeier, and Stulz (2012), we define the crisis 17 We find consistent results when we run the regression based on the three subsamples: firm with high, medium, and low growth opportunities. In addition, we also find consistent results when we use two alternative measures of firm growth opportunities: Market-to-book (Total Assets Book value of Equity + Price Common Shares Outstanding) / Total Assets) and DTE ((Debt in Current Liabilities + Total Long-Term Debt + Preferred Stock) / (Common Shares Outstanding price)) to do the robustness check. We do not report results here to save space but they are available upon request. 24

25 time in our sample as the most recent financial crisis, , and the normal time as and We use the ordinary least squares regression: Spread i,t = α 1 + α 5 OC i,t 1 D Normalt + α 6 OC i,t 1 D Crisist +β Z i,t 1 + ν i + μ t + ε i,t, (3) where D Normalt and D Crisist are dummy variables for the normal and crisis periods at time t, respectively. All the other variables are defined as before. Table 6 presents the regression results for the impact of CEO overconfidence on bank loan spread under different macroeconomic conditions. The coefficient of OC D Normal in Model (1) is significantly negatively related to loan spread at the 1% level, showing that overconfident firms enjoy lower loan spreads in good times on average over all our sample. In the subsamples of Collateral = 1, FinCov 1, GenCov 1, and TotalCov 1, the coefficients of OC D Normal are all significantly negatively related to loan spread at the 1% level, while all the coefficients of OC D Crisis are insignificant. To be specific, conditional on collateral or covenant agreements, overconfident firms enjoy a to reduction in loan spread over other firms in good times, greater than the estimated coefficients of OC in Table 4. These results support Hypothesis 2b that overconfident firms obtain lower bank loan spreads only in good times. 5. Additional Supporting Evidence [Insert Table 6] 5.1. CEO overconfidence and the future benefits of banks So far, the results show that as long as there are collateral or covenant 18 Didier, Hevia, and Schmukler (2012) indicate that the recovery phase started in

26 agreements, overconfident firms are charged lower loan spreads, especially when firms have rich firm-specific growth opportunities or when markets are booming. In other words, banks view future business opportunities as a critical consideration in design of loan contracts to overconfident firms. In this section, we investigate whether overconfident firms do bring more possible future business opportunities to banks. Due to the tendency to invest, overconfident firms have more need for external financing and the relative business. Bharath, Dahiya, Saunders, and Srinivasan (2007, 2011) show that, when borrowers need further financing, they often go to lenders with which they have lending relationships to obtain better loan terms. Thus, we first examine whether overconfident firms get more loan from the same banks after the first loan contract. To test this issue, we construct a variable regarding future relation in loans between banks and firms (Future Relation). We estimate the logistic regressions: P(Future Relation i,t+5 = 1 OC i,t 1, Z i,t 1 ) = L(α 1 + α 2 OC i,t 1 + β Z i,t 1 + ν j + μ t ). (4) P(Future Relation i,t+5 = 1 OC i,t 1, Z i,t 1 ) = L(α 1 + α 3 OC i,t 1 Q Hi,t 1 + α 4 OC i,t 1 Q Li,t 1 + β Z i,t 1 + ν j + μ t ), (5) where Future Relation is a dummy variable which equals one if the firm borrows at least one loan from the same lender in the five year after signing the loan contract in our sample, and zero otherwise. ν j is industry fixed effect. 19 All the other variables are defined as before. Table 7 presents regression results for the impact of CEO overconfidence on 19 In the specification, we run cross-sectional regressions, so we control the industry fixed effect rather than the firm fixed effect. 26

27 future benefits of banks in bank loan contracts. 20 In Models (1) (2), the coefficients of OC are significantly positive at the 1% level, meaning that overconfident firms are more likely to borrow from the banks with lending relationships in the future. Consistent with previous findings, this result is only significant for high-growth overconfident firms. In Models (3) (4), the coefficients of OC Q H are significantly positive at 1% level, whereas the coefficients of OC Q L are all insignificant. These results suggest that only overconfident firms with better growth opportunities are more likely to borrow from the banks with lending relationships. One possible reason (equilibrium) of the result is that banks only favor overconfident firms when the firms have better growth opportunities. [Insert Table 7] We also use growth in assets and equity, acquisitions, and capital expenditures as proxies for possible future benefits to banks. Borrowers with stronger growth, more acquisitions, and higher capital expenditures can bring more future business to banks. We use the ordinary least squares regression: Future Benefits = α 1 + α 3 OC i,t 1 + β Z i,t 1 + ν j + μ t + ε i,t, (6) where Future Benefits is one of the four proxies: Assets, Equity, Acquisitions, and CE. Assets is the annual change in firm assets normalized by the previous year s asset value. Equity is the annual change in outstanding common stock normalized by the previous year s outstanding common stock value. Acquisition is acquisition fees divided by total assets, and CE is capital expenditures divided by total assets. All the other variables are defined as before. Table 8 presents regression results for the impact of CEO overconfidence on 20 After we include the future relation in loans between banks and firms, the sample size reduces to 10,