關於公司借款成本及零負債經營的兩篇論文 - 政大學術集成

全文

(1)國立政治大學財務管理研究所博士論文. 立. 政治大. ‧ 國. 學. 關於公司借款成本及零負債經營的兩篇論文. ‧. y. sit. io. Zero-debt Policy. n. al. er. Nat. Two Essays on Corporate Financing Costs and. Ch. engchi. i n U. 指導教授：張元晨研究生：石維中. 中華民國. 106. v. 博士撰. 年 7月. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(2) 目錄目錄................................................................................................................................ I 表目錄..........................................................................................................................IV 圖目錄..........................................................................................................................VI Chapter I ....................................................................................................................... 1 Introduction ................................................................................................................ 1. 政治大 Bank Information Monopolies and Hold-Up Effects: International Evidence .......... 3 立. Chapter II ..................................................................................................................... 3. ‧ 國. 學. Abstract ...................................................................................................................... 3 1. Introduction ............................................................................................................ 4. ‧. 2. Source of Data and Sample Analysis ................................................................ 10. y. Nat. er. io. sit. 2.1 Data................................................................................................................. 10 2.2. Trend analysis ................................................................................................ 14. al. n. v i n Ch 2.3. Cross-section analysis.................................................................................... 15 engchi U 3. Hypotheses and Empirical Models ....................................................................... 19 3.1 Cross-country differences in banks’ hold-up effects ...................................... 19 3.2. Business cycles, banking crises, and the 2007–2008 financial crisis ............ 21 4. Empirical Results ................................................................................................. 26 4.1. Banks’ hold-up effects ................................................................................... 26 4.2. Bank regulations: Cross-country differences ................................................ 29 I. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(3) 4.3. Business cycles, banking crises, the 2007–2008 financial crisis ................... 31 5. Robustness Check ................................................................................................ 36 5.1 Endogenous access to public bond markets ................................................... 36 5.2. Alternative specifications .............................................................................. 38 6. Conclusion............................................................................................................ 41 References ................................................................................................................ 43 Chapter III .................................................................................................................. 84. 政治大. 探究美國公司無負債經營的迷思：供給面因素的影響 ...................................... 84. 立. 摘要 .......................................................................................................................... 84. ‧ 國. 學. 1. 緒論 ..................................................................................................................... 85. ‧. 2. 文獻回顧 ............................................................................................................. 88. sit. y. Nat. 2.1 零負債公司 ................................................................................................... 88. io. n. al. er. 2.2 利率與資本結構 ........................................................................................... 89. i n U. v. 3. 變數定義與敘述性統計 ..................................................................................... 92. Ch. engchi. 3.1 資料來源 ....................................................................................................... 92 3.2 零負債的定義 ............................................................................................... 93 3.3 借貸關係衡量指標 ....................................................................................... 94 3.4 樣本分配 ....................................................................................................... 96 3.5 敘述性統計分析 ........................................................................................... 99 3.6 實證模型 ..................................................................................................... 100 4. 實證結果 ........................................................................................................... 102 II. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(4) 4.1 單變量迴歸分析 ......................................................................................... 102 4.2 多變量迴歸分析 ......................................................................................... 104 4.3 銀行危機期間(Banking crisis)與非銀行危機期間..................................... 109 5. 穩健性測試 ....................................................................................................... 111 5.1 主成分分析法 ............................................................................................. 111 5.2 替代的應變數：近似零負債公司(ZLTD) ................................................ 112 5.3 公司信用評等的差異 ................................................................................. 113. 政治大. 5.4 同步性的內生性問題。 ............................................................................. 114. 立. 6. 結論 ................................................................................................................... 115. ‧ 國. 學. 參考文獻 ................................................................................................................ 116. ‧. 附錄 ........................................................................................................................ 123. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. III. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(5) 表目錄 Chapter II Table I Variable Definitions ...................................................................................... 49 Table II Summary Statistics ...................................................................................... 54 Table III Comparing the Hold-Up Cost between the US and Non-US Sample ........ 56 Table IV The Hold-Up Costs by Country ................................................................. 57. 政治大. Table V Loan Spreads for Bank-Dependent and Nondependent Borrowers:. 立. Multivariate Analysis ........................................................................................... 59. ‧ 國. 學. Table VI Comparing bank regulation indices between the US and Non-US Sample. ‧. .............................................................................................................................. 62. Nat. io. sit. y. Table VII Loan Spreads on the Dummy Variables for Bond Market Access and for. er. Regulatory: Univariate Analysis .......................................................................... 63. al. n. v i n CDummy Loan Spreads on the for Bond Market Access and for h e n Variables gchi U. Table VIII. Regulatory: Multivariate Analysis ....................................................................... 66 Table IX. Loan Spreads, Recessions, Banking Crises, and the 2007-2008 Financial. Crisis .................................................................................................................... 72 Table X Results of the 2SLS estimations (test for H1) ............................................. 77 Table XI Results of the 2SLS estimation (test for H2&H3) ..................................... 81 Chapter III IV. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(6) Table 1、敘述統計 .................................................................................................... 131 Table 2、Pearson 相關係數 ...................................................................................... 133 Table 3、單變量迴歸分析 ........................................................................................ 134 Table 4、多變量迴歸分析 ........................................................................................ 138 Table 5、 Banking crisis 期間的影響效果 .............................................................. 148 Table 6、在 Non-Banking Crisis 期間的影響效果.................................................. 159. 政治大. Table 7、穩健性測試：主成分分析 (Principal component analysis) .................... 170. 立. Table 8、替代的應變數：近似零負債公司(ZLTD) ............................................... 175. ‧ 國. 學. Table 9、公司的信用評等的差異 ............................................................................ 180. ‧. Table 10、同步的內生性問題 .................................................................................. 185. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. V. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(7) 圖目錄 Chapter II Figure 1.. Mean spreads of US firms by comparing the pricing of loans for bank-. dependent borrowers with the pricing of loans for borrowers with access to public debt markets. ............................................................................................. 48 Figure 2.. Mean spreads of Non-US firms by comparing the pricing of loans for. 政治大. bank-dependent borrowers with the pricing of loans for borrowers with access to. 立. public debt markets .............................................................................................. 48. ‧. ‧ 國. 學. Chapter III. 零負債及負債公司數目在各年度分配情況 ......................................... 129. Figure 2.. 零負債公司數目、平均帳面負債比率及平均市場負債比率。 ......... 129. n. al. er. io. sit. y. Nat. Figure 1.. i n U. v. Figure 3. 零負債公司比例及利率之趨勢圖。 ....................................................... 130. Ch. engchi. 零負債公司比例是零負債公司數目除以整體上市公司數目。FFR 為聯邦基準利率。T10y 為 10 年期政府公債利率。AAA 為 Moody 的 Aaa 級企業債券殖利率。................................................................................................................ 130 Figure 4. 零負債及負債公司在各產業分配情況 ................................................... 130. VI. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(8) Chapter I Introduction 本論文是由兩篇探討公司融資決策的文章所構成。在第一篇文章中，我們利用全球的資料驗證銀行的強劫(Hold up)效果。所謂的強劫效果，根據 Rajan (1992) 定義是指當公司無法有效率將信用風險訊息揭露給外部銀行，較高的資訊不對稱會降低外部銀行競爭該公司的銀行貸款，使得資訊優勢的內部銀行可以向借款公. 政治大. 司索取較高的利率。Santos and Winton (2008)的實證結果支持 Rajan 的理論，他. 立. 們發現公司無發行公開債券的銀行貸款會比有發行公開債券(揭露信用風險)的. ‧ 國. 學. 銀行貸款被索取更高的利率，這個現象在景氣衰退的期間更為明顯。. ‧. 過去文獻探討此議題只有侷限單一國家，但是銀行經營範圍是更為寬廣的，. y. Nat. al. er. io. sit. 本論文認為銀行的強劫行為可能是一個全球的現象。因此我們利用全球資料進行. n. 實證分析，發現確實許多國家存在顯著的強劫效果。除了過去文獻發現美國及英. Ch. engchi. i n U. v. 國有顯著的結果之外，我們還發現另外有 19 個國家有顯著的效果。若以美國樣本及非美國國家的樣本作比較，美國的 Hold up 效果是非美國國家樣本兩倍(92.4 bps vs. 42.9 bps)。接著本論文探討那些因素造成強劫效果在不同國家之間有顯著差異，我們發現在銀行管制較嚴格的國家，存在較為顯著的強劫效果。另外我們也發現和 Santos and Winton (2008)一致的結果，在景氣衰退期間存在顯著的強劫效果。但是在 2007-2008 金融風暴期間及銀行衰退期間(Banking. 1. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(9) crisis)，我們沒有顯著的強劫效果。在第二篇文章中，我探討為什麼越來越多美國上市公司是以零負債的方式營運。近年來美國上市公司以零負債經營的比率有相當高的成長(Bessler, et al.,2013; Strebulaev and Yang, 2013)，近年來文獻較多是以需求面的因素探討零負債迷思，但是至今還未發現一致性的結論，然而文獻比較欠缺由供給面探討零負債的研究。本論文希望能夠提供一個新的面向探討零負債公司形成的原因。Bessler et al.. 政治大. (2012) 認為透過需求面因素不足以解釋越來越多公司是以零負債經營的現象，. 立. 他們認為供給面可能是造成公司零負債的關鍵因素之一。. ‧ 國. 學. 本論文利用 1987 年至 2012 年的美國上市公司做為樣本，研究市場利率是否. ‧. 會影響公司以零負債經營的可能性。實證結果有以下三個發現，第一、市場利率. Nat. io. sit. y. 與公司零負債經營是呈現顯著的負向關係，亦即當市場利率越低時，則公司以零. er. 負債經營的機率越高；第二、銀行借貸關係程度與公司零負債經營呈現顯著的負. al. n. v i n Ch 向關係，亦即當公司過去與銀行借貸關係程度越高，公司零負債經營的機率會下 engchi U 降；第三、當公司過去與銀行借貸關係程度越高時，市場利率與公司零負債經營的機率是呈現顯著的正向關係。. 本論文支持零負債公司的行為是服從順循環(pro-cyclical)理論，亦即景氣衰退期間，公司舉債會較低(Cenesizoglu and Essid, 2012; Baum et al., 2009; Korajczyk and Levy, 2003; Erel et al., 2012)。本論文也發現零負債公司會使用較多保留盈餘及持有較高的現金。 2. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(10) Chapter II Bank Information Monopolies and Hold-Up Effects: International Evidence. Abstract. 政治大. We examine whether hold-up effects are confined to the U.S. market. Specifically, using. 立. an international sample, we investigate whether firms with access to the public debt. ‧ 國. 學. market pay lower loan spreads. We hypothesize that when firms issue public bonds they reveal private information of their credit risk, and thus the monopoly power of inside. ‧. banks decreases, resulting in lower loan spreads. We find extensive evidence to support. y. Nat. io. sit. this conjecture. We also find that hold-up effects are stronger during recessions and. n. al. er. more significant in countries with stringent bank regulations, but insignificant during banking crises and the 2007–2008 financial crisis.. Ch. engchi. i n U. v. Keywords: hold-up effects, bank-dependent and non-bank dependent borrowers, loan spreads JEL classifications: G21, G32, F34. 3. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(11) 1. Introduction The banks’ informational hold-up effects are one of the key issues in corporate finance. Hold-up costs tighten a firm’s cash flows and cause underinvestment. According to Rajan’s (1992) theoretical model, inside banks have informational advantages and bid on loans to a firm only when they know the firm will succeed. Conversely, outside banks have no access to the firm’s private information and. 政治大. encounter the “winner’s curse” problem: That is, they are more likely to win a loan. 立. when the firm has a high chance of failure. When a borrower’s credit risk increases to. ‧ 國. 學. the point of being labeled a “lemon,” the probability that outside banks bid on the loan. ‧. declines, which allows inside banks to hold-up the borrower. Rajan suggests that bank. Nat. sit. y. debts increase monitoring of borrowers; however, the monitoring process can provide. er. io. inside banks with a monopoly control on information about borrowers’ creditworthiness.. al. n. v i n C htheir private information, If borrowing firms fail to disclose e n g c h i U information asymmetries between borrowing firms and outside banks occur. A high level of information asymmetry results in a low probability that outside banks bid on loans to borrowers. As a result, inside banks can hold up borrowers to extract higher interest. The inference of Rajan’s model leads to an important empirical question: How can borrowing firms limit the monopoly power of the inside banks and thereby mitigate the hold-up problem?. 4. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(12) Earlier studies focus on the decision to switch from a single bank relationship to multiple bank relationships. For example, Houston and James (1996) examine the determinants for the mix of public and private debt for a sample of U.S. firms. Using a sample of Portuguese private firms, Farinha and Santos (2002) investigate the likelihood that firms substitute a single bank relation with multiple bank relationships. Both studies find that bank-dependent firms with more growth opportunities and poor. 政治大. liquidity are more likely to switch from a single bank relationship to multiple bank. 立. relationships to reduce the hold-up costs.. ‧ 國. 學. Many recent empirical studies examine whether revealing a firm’s private. ‧. information to the public can reduce inside banks’ hold-up power. Santos and Winton. y. Nat. al. er. io. sit. (2008) find that U.S. firms that have most recently issued public bonds prior to a loan. n. origination pay lower interest on bank loans than firms that have no access to public. Ch. engchi. i n U. v. bond markets. Hale and Santos (2009) report that firms pay lower interest rates on bank loans after a bond initial public offering (IPO). These findings suggest that when a firm issues public bonds, it reveals its creditworthiness to the public. 1 The disclosure of information reduces information asymmetry between outside banks and borrowers.. 1. Hale and Santos (2009) find that a majority of firms with bond IPOs have the first rating from rating. agencies. Also, Liu and Thakor (1984), Ederington, Yawitz, and Roberts (1987), and Hand, Hothausen and Leftwich (1992) report informational spill-over effects of bond ratings. 5. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(13) Therefore, the probability that an outside bank bids on loans to the firm increases, thus diminishing the hold-up power of inside banks. Issuing public bonds is not the only way for firms to disclose private information. Schenone (2010) investigates the pattern of loan interest rates before and after an equity IPO for U.S. firms and finds that information released through equity IPOs spill overs to outside banks. Thus, competition increases among banks, which mitigates the hold-. 政治大. up effects. Pagano, Panetta, and Zingales (1998) report a similar result. They analyze a. 立. sample of Italian firms and find that these firms’ cost of debt declines after their equity. ‧ 國. 學. IPOs. In addition, Saunders and Steffen (2011) find that the delisting of British firms. ‧. shifts the bargaining power from borrowers to lenders and increases the information. er. io. sit. y. Nat. production cost.. In sum, prior studies suggest that firms’ disclosure of private information to the. n. al. Ch. engchi. i n U. v. public significantly reduce lenders’ ability to hold up borrowers. Santos and Winton (2008) and Hale and Santos (2009) provide evidence of hold-up costs, measured by the difference in loan spreads paid by bank-dependent and nonbank-dependent firms. However, most prior empirical research on hold-up costs relates to the United States or a single country (e.g., Mattes, Sascba, and Wahrenburg, 2013). However, banks in different countries do not operate in the same economic and legal environments. Thus,. 6. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(14) the question remains: Are hold-up costs a global phenomenon with significant cost variation across countries? If so, what factors cause the discrepancies? To address these questions, we apply Santos and Winton’s methodology to an international sample to examine various issues related to hold-up effects. To the best of our knowledge, this study is the first to provide an international comparison of hold-up effects. Using indices of bank regulatory and supervision policies compiled by Barth,. 政治大. Caprio, and Lavine (2004, 2006, 2008, 2013), we hypothesize that bank regulations. 立. help to explain cross-country variation in hold-up effects. We also examine the relation. ‧ 國. 學. between business cycles and hold-up effects. Santos and Winton (2008) find that in the. ‧. U.S., hold-up effects are significant during recessions but not expansions. We examine. y. Nat. al. er. io. sit. if this relation holds for an international sample. In addition, we investigate whether. n. hold-up effects occur during banking and financial crises. Banking crises, characterized. Ch. engchi. i n U. v. by a liquidity crunch and increasing uncertainty, reflect the financial distress of lending banks. Therefore, we investigate whether banking crises affect the hold-up power of inside banks. Santos and Winton use recessions to capture the credit risk and uncertainty of borrowers. One consequence of the 2007–2008 financial crisis was a loss of confidence in credit rating agencies. For example, Bedendo, Cathcart, El-Jahel, and Evans (2013) provide evidence that the reputational damage of credit rating agencies. 7. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(15) in rating structured products spilled over to corporate bond rating services. If the informational value of public debt diminishes, information asymmetry between outside banks and borrowers increases, thus increasing inside banks’ ability to hold up borrowers. We measure the impact of banking crisis and the 2008–2009 crisis by examining the difference in loan spreads between bank-dependent (i.e., firms that do not have public bonds prior to the loan origination) and nonbank-dependent firms (i.e.,. 政治大. firms that issued public bonds prior to the loan origination).. 立. We divide our sample between U.S. firms and non-U.S. firms and find that. ‧ 國. 學. significant hold-up effects exist for both subsamples, with a higher effect for U.S. firms.. ‧. Specifically, the average loan spread of U.S. firms (non-U.S. firms) that recently issued. y. Nat. al. er. io. sit. public bonds is 92.4 (42.9) basis points lower than those without public bonds. In other. n. words, average hold-up costs in the United States are more than twice of those for nonU.S. firms.. Ch. engchi. i n U. v. Our regression results show that bank regulations significantly affect hold-up effects. In countries with more stringent bank regulations, firms without public bonds pay higher interest rates on bank loans than those with public bonds. The results are robust to controlling for firm and loan characteristics as well as the possible endogeneity of access to public bond markets. These results show that inside banks in. 8. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(16) countries with stringent regulations are more likely to exploit their monopoly power to capture higher informational rents and thus transfer regulatory costs to borrowers. Like Santos and Winton (2008), we find significant hold-up effects during recessions. However, we find no significant difference in loan spreads between bankdependent and nonbank-dependent firms during banking crises and the 2007–2008 financial crisis. In fact, inside banks’ hold-up power declines during banking crises. The. 政治大. absence of significant hold-up effects during the 2007–2008 financial crisis can be. 立. attributed to the credit rating crisis, which reduces the informational value of public. ‧ 國. 學. bonds. Moreover, when more firms’ bonds are nonrated or have a below-investment. ‧. er. io. al. sit. Nat. public bond markets no longer leads to lower loan spreads.. y. grade, the informational advantage of public bonds declines. As a result, access to. n. The remainder of this paper is organized as follows. Section 2 describes the data. Ch. engchi. i n U. v. and methodology and presents the summary statistics of all variables. Section 3 discusses the hypotheses and empirical models. Section 4 presents the main empirical results, and Section 5 provides the results of robustness tests. Finally, Section 6 concludes.. 9. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(17) 2.. Source of Data and Sample Analysis. 2.1 Data We use data from Loan Pricing Corporation’s Dealscan database (LPC), Securities Data Corporation’s Domestic New Bond Issuance database (SDC), Compustat, Center for Research on Securities Prices’ stock price database (CRSP), World Bank, Laeven and Valencia’s (2013) systematic banking crisis database and the Barth et al. (2004,. 政治大. 2006, 2008, 2013) bank regulatory data set. The LPC’s Dealscan database provides. 立. detailed information on worldwide syndicated loans (e.g., borrowers and lenders. ‧ 國. 學. identities, contract terms, and contract dates). The data go back to the beginning of the. ‧. 1980s, but the database is less comprehensive in the early years. Therefore, following. y. Nat. n. al. er. io. 2014.. sit. Santos and Winton, we start our sample period in 1987; we end our sample period in. Ch. engchi. i n U. v. We use the SDC’s Domestic New Bond Issuances database to obtain information on firms’ access to public bond markets. This database contains the information on firms’ public bond issuance such as the issuance date and whether the bonds are publicly or privately placed. Following Hale and Santos (2009), we exclude Rule 144A bonds in the U.S. markets as public bonds because the issuers are not required to disclose as much private information and the bonds are traded to a limited number of qualified. 10. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(18) investors. We construct the main variable, MRPBOND, in our analysis to capture a firm’s access to bond markets. As defined by Santos and Winton (2008), MRPBOND equals 1 if the borrowing firm’s most recent bond issue prior to a loan is a public bond, and zero otherwise. Data on bank loans come from the LPC-Compustat link data set constructed by Chava and Roberts (2008). The LPC-Compustat data set merges data from Compustat. 政治大. with loan information from the Dealscan database by matching company names. We. 立. expand the data set by merging the LPC-Compustat with the SDC using the Committee. ‧ 國. 學. on Uniform Security Identification Procedures (CUSIP) code. However, the CUSIP. ‧. code only applies to U.S. firms. For other countries, we conduct the matching process. y. Nat. al. er. io. sit. using company names, the Stock Exchange Daily Official List, and the International. n. Securities Identification Number. We exclude bank loans to financial institutions (SIC. Ch. engchi. i n U. v. codes 6000–6999) and those without a disclosed interest rate. The final sample totals 67,787 loans to 14,030 firms from 57 countries. Borrowers issued public bonds prior to the origination of 10,577 loans. Out of these loans, 7,364 are from the United States and 3,213 from other countries. In unreported results (available on request), we also define a firm’s access to the bond market using a dummy variable for bond issuances, including both private and public bonds. Including private bonds, 13,230 bonds are. 11. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(19) issued prior to bank loans: 10,293 from the United States and 4,526 from other countries. Using this alternative definition of access to bond market yields similar results. We construct the bank regulatory variables using Barth et al.’s (2004, 2006, 2008, 2013) data set.2 This data set builds on four surveys that include over 300 questions on various topics, such as bank activities and financial conglomerate, competition regulation, capital regulation, and government supervisory power. The selected bank. 政治大. regulatory variables are overall activities restriction (ACT_RESTRICT), restriction on. 立. banks owning nonfinancial firms (OWN_FIRM), official supervisory power. ‧ 國. 學. (SUP_POWER), prompt corrective power (PRMPT_POWER), independence of. ‧. supervising authority (SUP_IND_BANK), deposit insurer power (DEP_POWER), and. y. Nat. al. er. io. sit. financial statement transparency (BANK_ACCOUNT). These variables are widely. n. used in the literature on bank regulations (e.g., Houston, Lin, and Ma, 2012; Beltratti. Ch. engchi. i n U. v. and Stulz, 2012; Ongena, Popov, and Udell, 2013; De Jonghe and Oztekin, 2015). During a banking crisis, the supply of funds to the banking system decreases and uncertainty in the industry increases. To determine whether the conditions of a banking crisis affect the hold-up power of inside banks, we use a dummy variable,. 2. We collect the data for sample period 1987 to 2000 from Barth et al. (2004), 2001 to 2003 from Barth. et al. (2006), 2004 to 2007 from Barth et al. (2008), and 2008 to 2014 from Barth et al. (2013). 12. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(20) BANK_CRISIS, which equals 1 if the loan originates during a banking crisis in the country of the lead bank, and zero otherwise.3 We use the Laeven and Valencia’s (2013) database to determine the periods of banking crises. Laeven and Valencia capture banking crises using several banking system indicators and identify significant banking policy interventions. Traditionally, the definition of a banking crisis primarily relies on the identification of specific events or subjective criteria. Laeven and Valencia argue. 政治大. that banking crises can be more precisely defined by adding government policy. 立. intervention as an indirect measure of financial distress in a banking system (see related. ‧ 國. 學. discussions in Laeven and Valencia, 2013, p. 228).. ‧. To examine whether hold-up effects are more prominent during a recession, we. y. Nat. al. er. io. sit. follow Santos and Winton (2008) and use a dummy variable, RECESSION, which. n. equals 1 if the loan originates during a recession, and zero otherwise. We identify the. Ch. engchi. i n U. v. recession stage in the business cycle using Braun and Larrain’s (2005) definition of recession.4 We also construct a dummy variable to capture the effect of the 2007–2008. 3. We focus on the lead bank in a syndicated loan because the lead bank plays the major monitoring role. (Sufi, 2007). 4. Braun and Larrain (2005) measure cyclical GDP for each country by taking the difference between. actual GDP and its long-term trend. They first identify the troughs, the years when cyclical GDP is below its trend level by more than one standard deviation, and then go backward to detect a local peak. A recession is the period from the year after a peak to the year of a trough. 13. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(21) financial crisis, 2007–2008_CRISIS, which equals 1 for the period from the third quarter of 2007 to the fourth quarter of 2008, and zero otherwise. The appendix provides definitions of all variables. Table 1 provides the descriptive statistics.. [Insert Table 1 here] 2.2. Trend analysis. 政治大. Figure 1 shows the average loan spread for bank-dependent and nonbank-. 立. dependent U.S. firms from 1987 to 2014. A firm is classified as nonbank dependent if. ‧ 國. 學. MRPBOND equals 1, where the most recent bonds issued by the firm prior to the loan. ‧. were public bonds, and zero otherwise. As expected, the average loan spread for bank-. Nat. io. sit. y. dependent firms is much larger than that of nonbank-dependent firms. This evidence. er. suggests that significant hold-up effects exist in the U.S. market. Loan spreads spike. al. n. v i n Cshortly financial crisis. However, for both types of firms during and the 2007–2008 U h e nafter i h gc the difference narrows substantially, indicating that loan spreads to nonbank-dependent firms increase by a larger percentage than loan spreads to bank-dependent firms. Figure 1 illustrates that no significant difference exists in the spreads between bank-dependent and nonbank-dependent U.S. firms during the recent financial crisis. During the financial crisis, loan spreads are lower than those in some earlier periods in our sample. 14. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(22) as well as those in the post-crisis period.. [Insert Figure 1 about here]. Figure 2 shows a similar pattern for the non-U.S. sample. The average loan spread of bank-dependent firms is still larger than that of nonbank-dependent firms but by a. 政治大. smaller margin. In addition, the gap in the loans spreads between bank-dependent firms. 立. and nonbank-dependent firms decreases significantly during the 2007–2008 crisis.. Nat. io. sit. y. ‧. ‧ 國. 學 [Insert Figure 2 here]. er. 2.3. Cross-section analysis. al. n. v i n C heffects across 44 Ucountries. The mean spread for Table 2 compares the hold-up engchi bank-dependent and nonbank-dependent firms for the full sample are 230.7 basis points (bps) and 144.5 bps, respectively, and the difference (86.2 bps) is significantly different from zero. Hence, hold-up effects are significant for the full sample. The difference between the mean spread of bank-dependent U.S. firms (242.9 bps) and nonbankdependent U.S. firms (150.5 bps) is 92.4 bps (t statistic = 48.53), indicating strong hold-. 15. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(23) up effects in the U.S. market. This result is consistent with Santos and Winton’s (2008) findings. The results for the non-U.S. sample are similar. The difference between the mean spread of bank-dependent non-U.S. firms (173.4 bps) and nonbank-dependent non-U.S. firms (130.5 bps) is 42.9 bps (t statistic = 13.88). Therefore, the hold-up effects are not confined to the U.S. market. When borrowing firms are unable to disclose private information about their creditworthiness, inside banks hold up their. 政治大. borrowers for higher interests. However, the informational rent decreases when firms. 立. reveal private information by issuing public bonds.. ‧. ‧ 國. 學 [Insert Table 2 here]. er. io. sit. y. Nat. al. n. The hold-up costs of the U.S. subsample are more than two times those of the non-. Ch. engchi. i n U. v. U.S. subsample (92.4 bps vs. 42.9 bps). Furthermore, the mean spread for U.S. bankdependent firms is 69.5 bps higher than that for non-U.S. bank-dependent firms.5. 5. This result may explain the “pricing puzzle” reported by Carey and Nini (2007) who find that loan. spreads in the United States are significantly higher than those in European countries. They identify this difference as a pricing puzzle as the loan spread difference cannot be explained by various possible explanatory. Our initial findings suggest that the higher loan spreads in the United States may be attributed to the higher hold-up costs in the U.S. market. 16. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(24) Table 3 reports the average hold-up costs of each country. Although the full sample contains loans from 57 countries, borrowers of 13 countries have no access to public bonds (i.e., MRPBOND=0); these loans account for 187 observations. 6 Because MRPBOND equals zero in these cases, we cannot calculate the loan spreads difference between bank-dependent and nonbank dependent borrowers. Therefore, we omit loans from these 13 countries in Table 3. We use the full sample in the regression analyses.. 政治大. 立[Insert Table 3 about here]. ‧. ‧ 國. 學. Table 3 shows that 21 countries have significant hold-up effects. For 16 countries,. y. Nat. al. er. io. sit. the hold-up effects are significantly greater than zero at the 1% level. 7 Only four. n. countries have higher average hold-up costs (i.e., significantly greater than zero) than. Ch. engchi. i n U. v. the United States: Australia (99.6 bps), India (136.9 bps), Portugal (101.9 bps), and Spain (108.5 bps). Loan spreads for bank dependent borrowers in Greece are significantly lower than that of nonbank-dependent borrowers (difference = –591.2 bps,. 6. These countries are the Bahamas, Croatia, Cyprus, Hungary, Kazakhstan, Lithuania, Oman, Pakistan,. Poland, Romania, Turkey, Venezuela, and Vietnam. 7. These countries are Australia, Brazil, Canada, Germany, Hong Kong, India, Italy, Japan, Netherlands,. Portugal, South Korea, Spain, Sweden, Taiwan, United Kingdom, and the United States. 17. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(25) significant at the 1% level). We conjecture that public bonds in Greece convey negative information about the borrowing firms; therefore, banks with recently issued public bonds pay higher interest rates than those with no public bonds. Norway also shows a negative hold-up cost with a difference of –62.8 bps, which is marginally significant at the 10% level. In general, Tables 3 and 4 show significant differences in the hold-up effects across. 政治大. countries. These findings motivate us to investigate the factors that explain cross-. 立. country differences in the hold-up costs.. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 18. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(26) 3. Hypotheses and Empirical Models 3.1 Cross-country differences in banks’ hold-up effects. Banks are highly regulated around the world because of their important role in the economy. Regulations restrict banking activities and impose a regulatory burden on banks. According to the concept of regulatory dialectic (Kane, 1997), banks will try to circumvent regulations and transfer the regulatory burden to customers. Thus, we. 政治大. expect that differences in bank regulations can explain cross-country variation in hold-. 立. up effects. Banks in heavily regulated countries are likely to exploit their informational. ‧ 國. 學. monopoly power and charge higher interests to transfer the regulatory costs to. ‧. borrowers. Importantly, banks can choose not to exploit their monopoly power. y. Nat. al. er. io. sit. depending on the trade-off between maintaining the bank’s reputation and preserving. n. its financial health (Mattes et al., 2013). However, high regulatory costs provide inside. Ch. engchi. i n U. v. banks with an incentive to extract informational rent. In this case, nonbank-dependent firms that reduce the monopoly power of inside banks by issuing public bonds pay lower loan interests than bank-dependent firms. In addition, differences in bank regulations may influence the flow of banks’ capital across markets. Several recent studies provide evidence of bank regulatory arbitrage (Acharya, Wachtel, and Walter, 2009; Houston et al., 2012; Ongena et al.,. 19. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(27) 2013; Karolyi and Taboada, 2015). Banks from countries that are heavily regulated may engage in cross-border banking activities in countries with fewer regulations. Ongena et al. (2013) conclude that stringent bank regulations at home encourage domestic banks to reduce lending standards and take more risk aboard. Houston et al. (2012) show that the probability of capital inflow from foreign banks decreases due to stricter regulations. This finding implies that markets with more restrictive regulations are less attractive to. 政治大. foreign banks. A decrease in foreign competition due to tight regulations may enhance. 立. the monopoly power of inside banks. Based on this discussion, we develop our first. ‧. ‧ 國. 學. hypothesis:. Hypothesis 1: In markets with more stringent bank regulations, nonbank-. y. Nat. n. al. er. io. firms.. sit. dependent firms pay lower interest rates on bank loans than bank-dependent. Ch. engchi. i n U. v. We use the following model to examine Hypothesis 1:. 𝐿𝑂𝐴𝑁𝑆𝑃𝑅𝐸𝐴𝐷 = 𝑐 + 𝛼 ∙ 𝑀𝑅𝑃𝐵𝑂𝑁𝐷 + 𝛽 ∙ 𝐵𝐴𝑁𝐾 𝑅𝐸𝐺𝑈𝐿𝐴𝑇𝑂𝑅𝑌 + 𝛾 ∙ 𝑀𝑅𝑃𝐵𝑂𝑁𝐷 × 𝐵𝐴𝑁𝐾 𝑅𝐸𝐺𝑈𝐿𝐴𝑇𝑂𝑅𝑌 + ∑ 𝛿 ∙ 𝑋 + ∑ 𝜃 ∙ 𝑌 + 𝜀,. (1). where LOANSPREAD is the spread plus annualized upfront fees above LIBOR. If information revealed from public bonds reduces the hold-up power of inside banks,. 20. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(28) nonbank-dependent firms will pay lower spreads than bank-dependent firms. Therefore, we expect α to be significantly negative. Bank regulatory includes the indices of bank regulatory and supervision policies as previously discussed. We expect β to be significantly positive because banks that are subject to more restrictive regulations will charge higher interests to compensate for the regulatory costs they bear. We use the interactions between MRPBOND and bank regulatory indices to capture the impact of. 政治大. bank regulations on hold-up effects. If Hypothesis 1 holds, stricter bank regulations will. 立. be positively related to higher loan spreads for bank-dependent firms compared to. ‧ 國. 學. nonbank-dependent firms. Therefore, we expect γ to be significantly negative. X and Y. ‧. represent the variables that control for the borrowing firms and loan characteristics,. er. io. sit. y. Nat. respectively.. al. n. v i n C h and the 2007–2008 3.2. Business cycles, banking crises, financial crisis engchi U Rajan (1992) illustrates that when borrowers’ probability of success declines, outside banks are less aggressive to bid on loans, thereby increasing the hold-up power of inside banks. Santos and Winton (2008) find that bank-dependent firms pay significantly higher loan spreads than nonbank-dependent firms during recessions but not expansions. We use Branun and Larrain’s (2005) definition of recessions in the 21. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(29) countries of the borrowing firms to examine whether this relation holds globally. In addition, we investigate the impact on the hold-up effects of banking crises and the 2007–2008 financial crisis. Whereas recessions reflect borrowers’ risk, banking crises reflect the financial distress of and liquidity shock to the banking industry. When inside banks suffer a liquidity shock, their ability to hold up borrowers decreases and the probability of bidding by outside banks increases. As such, the hold-up power of. 政治大. inside banks is weaker when the banking industry suffers a liquidity shock. The impact. 立. of a banking crisis varies across different banks; some are more vulnerable to a shock. ‧ 國. 學. than others. Banks that are more severely affected by the crisis can lose market shares. ‧. to banks that are less affected. As a result, crisis-affected banks lose the ability to exploit. Nat. io. sit. y. their monopoly power. For example, Chodorow-Reich (2014) show that during the. er. 2007–2008 financial crisis, Credit Suisse, which suffered huge losses from its. al. n. v i n investments in mortgage-backedC securities, its lending in the syndicated market h e n greduced chi U by almost 79% compared to the precrisis period. Comparatively, U.S. Bankcorp, which was less affected by the crisis, reduced its lending by only 14%. In this case, the role of banks that are severely affected by the crisis declines, as does their ability to hold up borrowers. The 2007–2008 financial crisis represents a systematic economic and financial shock to the whole economy and affected both the real and financial sectors. In addition 22. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(30) to the liquidity shock to the banking industry, the 2007–2008 financial crisis may further affect hold-up effects for two reasons: First, the credit rating crisis diminished the informational content embedded in public bonds, and, second, ratings of most public bonds deteoriated. During the 2007–2008 financial crisis, credit rating agencies were highly criticized for the inaccuracy of their ratings for structured financial products. In fact, credit ratings. 政治大. agencies received blame for creating the financial crisis (Lahinsky, 2008). An article in. 立. USA Today reports that the three major credit rating agencies were trying to restore. ‧ 國. 學. their reputations six years after the 2007–2008 financial crisis (Krantz, 2013). This. ‧. reputational damage to credit rating agencies may has spill-over effects. Jaballah (2015). Nat. io. sit. y. finds that investors neglected rating changes during the financial crisis, and Bedendo et. er. al. (2013) report a possible spill-over effect of reputational damage to bond rating. al. n. v i n C hof bond ratings isUsubstantially weakened, issuing services. If the information content engchi public bonds may fail to reduce the informational advantage of inside banks.. Consequently, nonbank-dependent firms do not enjoy lower loan spreads than bankdependent firms. Santos and Winton (2008) show that during recessions, rates for bank-dependent firms are higher than those for firms with access to public bonds markets; however, the difference is mainly driven by investment-grade firms. Similarly, Hale and Santos 23. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(31) (2009) show that only firms with a rating of investment grade pay significantly lower rates after a bond IPO. These results suggest that only public bonds that convey favorable information about a firm’s credit worthiness reduce the informational rent captured by inside banks. If a bond is rated below investment rate or not rated, the firm does not pay significantly less interests than bank-dependent firms. If the majority of the public bond ratings deteriorates during the 2007–2008 financial crisis, access to. 政治大. public bond markets may not result in lower interests on bank loans. Figures 1 and 2,. 立. the difference between the loan spreads of bank-dependent and non-bank-dependent. ‧ 國. 學. firms had narrowed during the crisis. This discussion lead to our next two hypotheses:. ‧. Hypothesis 2: During a banking crisis, the interest rate difference between bank-. Nat. io. sit. y. dependent and nonbank dependent firms is insignificant.. er. Hypothesis 3: During the 2007–2008 financial crisis, the interest rate difference. al. n. v i n C h dependent firmsUis insignificant. between bank-dependent and non-bank engchi. We use the following two models to test Hypotheses 2 and 3, respectively: 𝐿𝑂𝐴𝑁𝑆𝑃𝑅𝐸𝐴𝐷 = 𝑐 + 𝛼 ∙ 𝑀𝑅𝑃𝐵𝑂𝑁𝐷 + 𝜇 ∙ 𝐵𝐴𝑁𝐾_𝐶𝑅𝐼𝑆𝐼𝑆 + 𝜌 ∙ 𝑀𝑅𝑃𝐵𝑂𝑁𝐷 × 𝐵𝐴𝑁𝐾_𝐶𝑅𝐼𝑆𝐼𝑆 + ∑ 𝛿 ∙ 𝑋 + ∑ 𝜃 ∙ 𝑌 + 𝜀. (2). 𝐿𝑂𝐴𝑁𝑆𝑃𝑅𝐸𝐴𝐷 = 𝑐 + 𝛼 ∙ 𝑀𝑅𝑃𝐵𝑂𝑁𝐷 + 𝜂 ∙ 2007 − 2008_𝐶𝑅𝐼𝑆𝐼𝑆 + 𝜑 ∙ 𝑀𝑅𝑃𝐵𝑂𝑁𝐷 × 2007 − 2008_𝐶𝑅𝐼𝑆𝐼𝑆 + ∑ 𝛿 ∙ 𝑋 + ∑ 𝜃 ∙ 𝑌 + 𝜀. (3). During a banking crisis, banks may suffer from liquidity shock and thus charge 24. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(32) higher loan spreads. Hence, in line with previous research (Santos, 2011; ChodorowReich, 2014), we expect μ to be significantly positive. To investigate the impact of banking crises on inside banks’ hold-up power, we add the interaction of MRPBOND and BANK_CRISIS to the model. If Hypothesis 2 will not hold, bank-dependent firms will pay significantly lower rates than nonbank-dependent firms during a financial crisis. Therefore, we expect the coefficient ρ is insignificantly different from zero. Similarly,. 政治大. if Hypothesis 3 holds, the coefficient of MRPBOND*2007–2008_CRISIS, φ, will also. 立. be insignificantly different from zero.. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 25. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(33) 4. Empirical Results 4.1. Banks’ hold-up effects We analyze banks’ hold-up effects for the full sample and subsamples of U.S. and non-U.S. firms. Table 4 presents the results of the regression analysis. LOANSPREAD is the dependent variable, and we focus on the coefficient of MRPBOND to measure banks’ hold-up power. To control for the effects of unobservable factors, we include. 政治大. time-, firm-, and country-specific controls in Models 2, 4, and 6. All the regression. 立. models are estimated using the heteroscedasticity-robust standard error.. Nat. io. sit. y. ‧. ‧ 國. 學 [Insert Table 4 about here]. er. Model 1 of Table 4 shows that the coefficient of MRPBOND is negative and. al. n. v i n C h with Santos andUWinton’s (2008) findings, this significant at the 1% level. Consistent engchi result indicates that firms that have recently issued public bonds (i.e., nonbankdependent firms) pay significantly smaller loan spreads (–86.24 bps) than firms with no access to public bond markets (i.e., bank-dependent firms). If borrowing firms are unable to disclose information about their creditworthiness to the public, outside banks are less likely to bid on the loans, which enhances inside banks’ hold-up power. Model 2 of Table 4 controls for firm and loan characteristics and includes time, 26. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(34) industry (measured by one-digit SIC code classification), and country fixed effects. The firm-specific variables, which control their impact on the firm’s riskiness, are firm’s tangible assets, interest coverage ratio, leverage ratio, z-score, and volatility of stock return. After including the control variables and fixed effects, we find a statistically significant but smaller coefficient on MRPBOND. On average, firms with access to public bond markets pay 33.48 bps less than bank-dependent firms. When we include. 政治大. the controls for firm and loan characteristics, the full sample size falls from 67,787. 立. observations to 36,253 observations due to missing data. To provide a better picture of. ‧ 國. 學. the impact of MRPBOND on LOANSPREAD, we hereafter report the results with and. ‧. without controls for firm and loan characteristics.. Nat. io. sit. y. Models 3 and 5 of Table 4 show the results for the U.S. and non-U.S. subsamples,. er. respectively. The coefficient of MRPBOND is for the U.S. (non-U.S.) subsample is –. al. n. v i n Cthe 92.32 bps (–42.88 bps). That is, U U.S. subsample is more than h ecoefficient n g c hfori the. twice that of the non-U.S. subsample. Models 4 and 6 show that, after controlling for firm and loan characteristics and fixed effects, the coefficient of MRPBOND, which is still negatively significant but with a smaller magnitude, is –32.88 bps (–23.36 bps) for U.S. firms (non-U.S. firms). These findings confirm our previous results in Table 2, which show that, on average, hold-up costs in the United States are significantly larger than those in other 27. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(35) countries. Firms with access to bond markets may be less risky and thus pay relatively lower interest. If so, our results are subject to a subtle form of sample selection bias. Although we include numerous firm-specific characteristics to control firms’ riskiness, our results may still be driven by other unobservable risk factors. To account for the possible endogeneity of MRPBOND, we re-estimate the relation between MRPBOND and LOANSPREAD using a two-stage least squares (2SLS) procedure as discussed in Section 5.. 立. 政治大. The results for the control variables are largely expected. Loan spreads increase. ‧ 國. 學. with financial leverage but decrease with tangible assets and z-score (i.e., higher z-score. ‧. = lower probability of default). Consistent with the banking literature, loan spreads are. Nat. io. sit. y. inversely related to loans size, maturity, number of lenders, and lending relationship. er. and positively related to dividend restriction. Similar to the findings of Santos and. al. n. v i n inclusionCofh different types of controls engchi U. Winton (2008), the. and fixed effects. substantially increase R2. For example, the R2 for the full sample increases from 4% to 35% when controls and fixed effects are included. This result highlights the important role of firm’s riskiness and loan characteristics in explaining loan spreads. However, hold-up effects remain significant even when we include controls and fixed effects. In sum, Table 4 shows significant hold-up effects for both U.S. and non-U.S. firms. However, the U.S. subsample has stronger hold-up effects than the non-U.S. subsample. 28. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(36) 4.2. Bank regulations: Cross-country differences Next we examine the role of bank regulations in explaining the cross-country variation in hold-up effects. Table 5 reports the average value of the regulatory indices for the U.S. and non-U.S. subsamples. The average values of the seven regulatory indices, as previously explained in Section 2.1, are significantly higher for U.S. firms than for non-U.S. firms. On average, U.S. firms are subject to more stringent regulations than the other counties as a whole, and as previously reported (see Table 2), hold-up. 政治大 effects are much higher in the United States than in other countries. Thus, the results in 立. ‧ 國. 學. Table 5 are consistent with the hypothesis that hold-up effects increase with the. Nat. n. al. Ch. engchi. er. io. [Insert Table 5 here]. sit. y. ‧. intensity of bank regulations.. i n U. v. Table 6 presents the regression results without firm and loan characteristics and the fixed effects. Models 1 through 7 examine Hypothesis 1 and, with the exception of Models 3 and 7, show significantly negative coefficients on MRPBOND. Bank regulatory variables, which range from 8.47 bps to 48.03 bps, are significantly and positively related to bank loans. The interactions between the bank regulatory variables and MRPBOND are significantly negative in all models. This result strongly supports 29. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(37) Hypothesis 1: Loan spreads increase with the intensity of bank regulations. With more stringent regulations, nonbank-dependent firms pay significantly lower interest rates than bank-dependent firms, implying that inside banks exploit their hold-up power and transfer regulatory costs to bank-dependent borrowers. Also, stringent regulations are likely to reduce foreign competition, which further enhances the monopoly power of inside banks.. 政治大. 立[Insert Table 6 about here]. ‧. ‧ 國. 學. Among the seven regulatory indices, the index of financial statement transparency. Nat. io. sit. y. (BANK_ACCOUNT) has the greatest impact on loan spreads. Similarly, the coefficient. er. of BANK_ACCOUNT*MRPBOND has the largest absolute value among the. al. n. v i n coefficients of the interactions C between indices and MRPBOND. Table 7 h e nregulatory gchi U shows the results including controls for firm and loan characteristics and fixed effects. With the exception of the interaction between SUP_IND_BANK and MRPBOND, which is not significant, the interaction terms between the bank regulatory variables and MRPBOND are negative and significant at the 1% level. These results are in line with those reported in Table 6 and support Hypothesis 1.. 30. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(38) [Insert Table 7 about here]. 4.3. Business cycles, banking crises, the 2007–2008 financial crisis Table 8 reports the results for Hypotheses 2 and 3. Consistent with Santos and Winton (2008), Model 2 shows that, after controlling for firm and loan characteristics, borrowers pay higher loan spreads during recessions. The coefficients of the interaction between MRPBOND and RECESSION are significantly negative in both Models 1 and. 政治大 2. In Model 2, banks charge firms with access to bond markets 14.44 bps less than bank立. ‧ 國. 學. dependent firms during a recession, supporting our hypothesis that informational rent. Nat. n. al. Ch. engchi. er. io. [Insert Table 8 about here]. sit. y. ‧. increases when firm risk increases.. i n U. v. Models 3 and 4 of Table 8 examine the impact of banking crises on hold-up effects. The coefficients of BANK_CRISIS are 26.77 and 43.67 in Models 3 and 4, respectively. Banks raise loan spreads to borrowers in the midst of a liquidity crunch and increasing uncertainty in the banking industry. However, when we control for firm and loan characteristics and other fixed effects, we find no significant difference in loan spreads paid by bank-dependent firms and nonbank dependent firms during banking crises. The 31. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(39) interaction between MRPBOND and BANK_CRISIS is positive but not significant at the conventional level. When we do not control for firm and loan characteristics and fixed effects, nonbank dependent firms even pay higher interest than bank-dependent firms; the coefficient in Model 3 for the interaction between bank crisis and MRPBOND is 11.49, significant at the 5% level. These results provide evidence of insignificant informational rents during. 政治大. banking crises and thus support Hypothesis 2. The liquidity shock and increasing. 立. uncertainty in the banking industry reduce the ability of inside banks to hold up. ‧ 國. 學. borrowers. While both bank-dependent and nonbank dependent firms pay higher. ‧. interest during banking crises, loan spreads of nonbank-dependent firms increase by a. Nat. io. sit. y. larger magnitude than those of bank-dependent firms. In addition, a comparison of. n. al. er. Models 2 and 4 in Table 8 shows that recessions and banking crises have different impacts on the hold-up effects.. Ch. engchi. i n U. v. The coefficients of the 2007–2008 financial crisis are significantly negative in Models 5 and 6 of Table 8. These results contradict to the findings by Ferreira and Matos (2012). Using the same definition as we do for the 2007–2008 financial crisis, Ferreira and Matos report a significantly positive coefficient for the financial crisis dummy in the loan spreads equation. We believe the difference in sample periods explains these contradictory results. Ferreira and Matos’s study covers the period from 32. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(40) 2003 to 2008. As a result, their financial crisis dummy captures the difference between the financial crisis and the precrisis credit boom period (2003–2006). Not surprising, loan spreads during the financial crisis period are higher than those in the credit boom period. Our sample period covers 1987 to 2014; therefore, our financial crisis dummy variable captures the difference in loan spreads between the 2007–2008 financial crisis and other periods, including the postcrisis period. Due to the government quantitative. 政治大. easing policy to stimulate economy activities, loan spreads during the financial crisis. 立. are lower than those during some of the precrisis periods and the postcrisis period (see. ‧ 國. 學. Figure 1). As a result, we find a significantly negative coefficient for the 2007–2008. ‧. crisis for our sample. In unreported results, we follow Ferreira and Matos and re-. Nat. io. sit. y. estimate Models 5 and 6 of Table 8 using only 2003–2008 data. The coefficient of the. er. 2007–2008 crisis turns positive and is significant in Model 6 (7.445, significant at 5%. al. n. v i n level). Similar to the findings forC banking bank dependent borrowers did not pay h e ncrises, gchi U a significantly lower rate than nonbank-dependent borrowers during the 2007–2008 financial crisis. In Table 8, the coefficient of MRPBOND*2007–2008_Crisis is negative and insignificant (positive and significant) when we consider (do not consider) all firm and loan controls and fixed effects. Therefore, in line with Hypothesis 3, we find no evidence of significant hold-up effects during the recent financial crisis. Our results are 33. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(41) consistent with Kwan’s (2010) finding in that bank-dependent firms, proxied by loan size, do not suffer more from bank tightening than do nonbank-dependent firms during the financial crisis. When we estimate Models 5 and 6 for a reduced sample using only data from 2003 to 2008, the coefficient of MRPBOND*2007–2008_CRISIS remain insignificant. Santos (2011) analyzes loan pricing following the 2007–2008 financial crisis and. 政治大. finds that banks with larger charge-offs during the subprime loan crisis charge. 立. significantly higher interest rates on bank-dependent borrowers but not on nonbank-. ‧ 國. 學. dependent borrowers. However, our results are not directly comparable with those of. ‧. Santos due to differences in research design. Unlike Santos, who compares the loan. Nat. io. sit. y. spreads of firms borrowing from the same bank before and during the subprime loan. n. al. er. crisis, we compare the loan spreads of bank-dependent borrowers and nonbank. Ch. engchi. dependent borrowers during the financial crisis.. i n U. v. We conjecture that during the 2007–2008 financial crisis public bonds failed to convey positive information about firms, which explains the narrowed gap in loan spreads between bank-dependent and nonbank-dependent firms. To address this possibility, we calculate the percentage of public bonds that are not rated or are rated below investment grade during different time periods. For the full sample period (200– 2008 financial crisis period), among the public bonds issued, below-investment grade 34. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(42) and unrated bonds account for 46.86% (46.9%) and 26.77% (40.28%), respectively. These findings show that the percentage of below-investment grade bonds is stable, but a much higher percentage of publicly issued bonds are not rated. Therefore, firms do not benefit from disclosing information through public bonds and fail to reduce holdup costs incurred.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 35. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(43) 5. Robustness Check 5.1 Endogenous access to public bond markets If access to public bond markets is endogenous, our results are subject to sample selection bias. Also, unobservable risk factors may affect both access to bond markets and loan spread. To cope with the potential endogeneity problem, we re-estimate the regressions using the 2SLS method. First, we regress MRPBOND on the firm controls and instrumental variables that are highly correlated with MRPBOND but not. 政治大 LOANSPREAD and obtain the predicted value. In the second stage, we substitute the 立. ‧ 國. 學. predicted value of MRPBOND into the loan spreads regression. In all cases, we include. ‧. the firm and industry fixed effects.. sit. y. Nat. Following Santos and Winton (2008), we include two instrumental variables,. n. al. er. io. INDACESS and OLDERFIRM, in the first stage estimation of MRPBOND.. Ch. i n U. v. INDACESS is the percentage of firms in a given industry (as defined by the two-digit. engchi. SIC code) and country that have access to bond markets. If a firm operates in an industry in which the bond market is highly competitive, investors likely have experience from past trading and can more easily evaluate bonds issued by this firm. Therefore, INDACESS should be strongly and positively correlated with MRPBOND. However, INDACESS does not affect LOANSPREAD. OLDERFIRM equals 1 for older firms, defined as firms older than the median firm age of the sample (23 years), and zero 36. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(44) otherwise. Firms with longer history are more likely to have access to the public bond markets; therefore, we expect a positive relation between older firm and LOANSPREAD. Table 9 shows the result of the 2SLS analysis. The instrumental variables, INDACESS and OLDERFIRM, are significantly and positively related to MRPBOND, consistent with our expectation. Next, we use the predicted value of MRPBOND in the. 政治大. first stage of regression as an independent variable in the LOANSPREAD equation. To. 立. examine Hypothesis 1, Panel A include interactions between bank regulatory variables. ‧ 國. 學. and the predicted value of MRPBOND. Similar to the results in Table 6, the results in. ‧. Table 9 show that the coefficients of all seven indices of regulatory stringency and bank. Nat. io. sit. y. supervisions are significantly positive. The positive relation between bank regulations. n. al. regulatory variables and the. er. and loan spreads is unchanged. Most important, the interactions between bank. v i n C h value of MRPBOND predicted are all negative and engchi U. significant at the 1% level. Again, the results support Hypothesis 1.. [Insert Table 9 about here]. Panel B of Table 9 reports the second-stage of the 2SLS estimations, which examine Hypotheses 2 and 3. The coefficient for the interaction between RECESSION 37. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(45) and the predicted value of MRPBOND is significantly negative. That is, bankdependent firms pay higher loan interests than nonbank-dependent firms during recessions. The interaction between BANK_CRISIS and the predicted value of MRPBOND and between 2007–2008_CRISIS and the predicted value of MRPBOND are significantly positive. These results show that, after considering the possible endogeneity of access to public bond markets, the difference in loan spread between. 政治大. bank- and nonbank-dependent borrowers is insignificant. This finding confirms our. 立. previous results that inside banks extract no significant informational rents during. ‧ 國. 學. banking crises and the 2007–2008 financial crisis.. ‧ sit. y. Nat. 5.2. Alternative specifications. n. al. er. io. In the previous regression analyses, we include all 57 countries in our sample. As. Ch. i n U. v. previously mentioned, sample firms from 13 countries do not have access to the public. engchi. bonds market. Therefore, we exclude these countries (187 observations) and re-estimate the regressions. The results are qualitatively similar. We also use an alternative definition of the 2007–2008 financial crisis: We set 2007–2008_CRISIS to equal 1 for the fourth quarter of 2007 to the fourth quarter of 2008. Comparable to the results in Table 7, the coefficients of 2007–2008_CRISIS are significantly negative, and the interaction between MRPBOND and 200738. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(46) 2008_CRISIS become significantly positive. Specifically, the coefficient of the new 2007–2008_CRISIS in the loan spread equation is –16.21, significant at the 1% level, and the coefficient of the new 2007–2008_CRISIS*MRPBOND is 15.18, significant at the 5% level. These untabulated findings do not alter our conclusion regarding the impact of the recent financial crisis on hold-up effects; that is, bank-dependent firms do not pay higher interest rates than nonbank-dependent firms during the 2007–2008 financial crisis.. 立. 政治大. Given that a banking crisis and a recession can occur simultaneously, we construct. ‧ 國. 學. an alternative dummy variable for banking crisis, which takes a value of 1 if the loan. ‧. originated during a banking crisis but not during a recession, and zero otherwise. We. Nat. io. sit. y. find no material changes in our main results using the new banking crises variable.. er. Specifically, when we include firm and loan controls and fixed effects, the coefficient. al. n. v i n C hvariable in the LOANSPREAD of the newly defined banking crisis equation is 46.76, engchi U significant at the 1% level (compared to 43.67, significant at the 1% level, for the original variable), and the coefficient of the interaction terms between the newly defined banking crises variable and MRPBOND is 4.37, not significant at the conventional level (compared to 6.52, not significant, for the original banking crisis variable). Thus, as before, the newly defined banking crisis variable is significantly and positively related to loan spreads and the coefficient of the interaction between the new 39. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(47) variable and MRPBOND is insignificant.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 40. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(48) 6. Conclusion The inability of borrowing firms to disclose private information to the public reduces the probability that outside banks bid on loans, which allows inside banks to hold up borrowers (Rajan, 1992). This study provides international evidence of these hold-up effects. First, we find that hold-up effects are not confined to the U.S. markets. However, hold-up effects, measured by the difference in loan spreads between bank-. 政治大. dependent and nonbank-dependent borrowers, are more than twice as high for the. 立. subsample of U.S. firms, compared to the hold-up effects for the subsample of non-U.S.. ‧ 國. 學. firms. Second, differences in bank regulations can explain the cross-country variation. ‧. in hold-up effects. Banks in countries with more stringent regulations are more likely. Nat. io. sit. y. to exploit their monopoly power over borrowers and transfer the regulatory burden to. er. borrowers. Accordingly, we find a positive relation between the stringency of bank. n. al. regulations and hold-up. i n C effects.hAfter controlling e n g c h i Uthe. v. firm- and loan-specific. characteristics and taking into account any problems related to endogeneity, our results remain robust. Similar to Santos and Winton (2008), we find significant hold-up effects during the 2007–2008 recession. This finding is in line with Rajan’s (1992) argument that inside banks have higher hold-up power when the probability of borrowers to success is lower and outside banks are less likely to bid on loans. However, we find no 41. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(49) significant difference in loan spreads between bank-dependent firms and nonbankdependent firms during banking crises and the 2007–2008 financial crisis. A liquidity shock to the banking industry reduces the hold-up power of inside banks. We suggest two possible explanations for the insignificant hold-up effects during the recent financial crisis. First, Santos and Winton show that hold-up effect exists only when the information embedded in public debt regarding the creditworthiness of borrowers is. 政治大. favorable. During the financial crisis, more publicly issued bonds are not rated or have. 立. a below-investment grade. Second, the value of information conveyed by issuing public. ‧ 國. 學. bonds may diminish during the financial crisis due to the loss of confidence in the rating. ‧. by credit rating agencies. As the value of private information decreases, firms with. Nat. io. sit. y. publicly issued bonds do not have an informational advantage and pay similar loan. n. al. er. spreads as bank-dependent firms do.. Ch. engchi. i n U. v. 42. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(50) References Acharya, V.V., Wachtel, P., Walter, I., 2009. International alignment of financial sector regulation. In: V.V. Acharya and M. Richardson (Eds.), Restoring Financial Stability: How to Repair a Failed System. Wiley, Hoboken, NJ. Barth, J.R., Caprio, G., Levine, R., 2004. Bank regulation and supervision: What works best? Journal of Financial Intermediation 13, 205–248.. 政治大. Barth, J.R., Caprio, G., Levine, R., 2006. Rethinking Bank Regulation: Till Angels. 立. Govern. Cambridge University Press, Cambridge, UK.. ‧ 國. 學. Barth, J.R., Caprio, G., Levine, R., 2008. Bank regulations are changing: For better or. ‧. worse? World Bank Policy Research Working Paper 4646.. Nat. io. sit. y. Barth, J.R., Caprio, G., Levine, R., 2013. Bank regulation and supervision in 180. er. countries from 1999 to 2011. NBER working paper no. 18733.. al. n. v i n C hL., Evans, L., 2013.UThe credit rating crisis and the Bedendo, M., Cathcart, L., El-Jahel, engchi informational content of corproate credit ratings. Working paper, Universita’ Bocconi. Beltratti, A., Stulz, R., 2012. The Credit crisis around the globe: Why did some banks perform better? Journal of Financial Economics 105, 1–17. Braun, M., Larrain, B., 2005. Finance and the business cycle: International, interindustry evidence. Journal of Finance 60, 1097–1128. 43. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.

(51) Carey, M., Nini, G., 2007. Is the corporate loan market globally integrated? A pricing puzzle. Journal of Finance 62, 2969–3007. Chava, S., Roberts, M.R., 2008. How does financing impact investment? The role of debt covenants. Journal of Finance 63, 2085–2121. Chava, S., Purnanandam, A., 2011. The effect of banking crisis on bank-dependent borrowers. Journal of Financial Economics 99, 116–135.. 政治大. Chodorow-Reich, G., 2014. The employment effects of credit market disruptions: Firm-. 立. level evidence from the 2008–9 financial crisis. Quarterly Journal of. ‧ 國. 學. Economics 129, 1–59.. ‧. De Jonghe, O., Ozekin, O., 2015. Bank capital management: International evidence.. Nat. io. sit. y. Journal of Financial Intermediation 24, 154–177.. er. Ederington, L.H., Yawitz, J.B., Roberts, B.E., 1987. The informational content of bond. n. al. i n C ratings. Journal of Financial hResearch i U e n g c10,h211–226.. v. Farinha, L.A., Santos, J.A C., 2002. Switching from single to multiple bank lending relationships:. Determinants. and. implications.. Journal. of. Financial. Intermediation 11, 124–151. Ferreira, M.A., Matos, P., 2012. Universal banks and corproate control: Evidence from the global syndicated loan market. Review of Financial Studies 25, 2703–2744. Hale, G., Santos, J.A.C., 2009. Do banks price their informational monopoly? Journal 44. DOI:10.6814/DIS.NCCU.Finance.019.2018.F07.