Auditor Quality, Tenure, and Bank Loan Pricing

Transcription

1 Farsiarticles.com Iran-article.ir Iranarticles.com Auditor Quality, Tenure, and Bank Loan Pricing By Jeong-Bon Kim, Byron Y. Song and Judy S. L. Tsui Current Draft March 2007 The first author is at Concordia University and The Hong Kong Polytechnic University. The second and third authors are at The Hong Kong Polytechnic University. We thank Jong-Hag Choi, Annie Qiu, Hanina Shi, Cheong H. Yi, Suk Heun Yoon, Yoonseok Zang, and participants of the 2006 Annual Meeting of AAA, and Ph.D./DBA research seminars at The Hong Kong Polytechnic University, and Seoul National University for their useful comments. The first and last authors acknowledge partial financial support for this research obtained from the Competitive Earmarked Research Grant of The Hong Kong SAR Government and the Area of Strategic Development (ASD) Research Grant, the Faculty of Business, The Hong Kong Polytechnic University. All errors are our own. Correspondence: Judy Tsui, Chair Professor and Dean, the Faculty of Business, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Electronic copy available at:

2 Farsiarticles.com Iran-article.ir Iranarticles.com Auditor Quality, Tenure, and Bank Loan Pricing SUMMARY: Using a large sample of US bank loan data over the 9-year period from 1996 to 2004, we investigate the effect of two auditor characteristics, namely auditor quality and tenure, on the price term of bank loan contracts. Our results show the following: First, we find that banks charge a significantly lower rate for borrowers with Big 4 auditors than for borrowers with non-big 4 auditors. Further analysis shows that banks charge a higher loan rate for borrowers who change their auditors in general, and they charge a substantially higher loan spread for borrowers who downgrade their auditors from Big 4 to non- Big 4 auditors in particular. Second, we find that the loan spread is inversely related to auditor tenure, suggesting that banks view auditor tenure as a credit risk-reducing factor. Third, we find that the relation between loan spread and audit quality is conditioned upon the level of credit risk perceived by credit rating agencies. Our study provides direct evidence that banks take into account audit quality when assessing borrowers credit quality and determining the price term of loan contracts. Keywords: Auditor quality; Auditor tenure; Loan pricing; Loan spread; Private debt. Electronic copy available at:

3 Farsiarticles.com Iran-article.ir Iranarticles.com Auditor Quality, Tenure, and Loan Pricing INTRODUCTION Audited financial statements play a crucial role in facilitating financial contracts in general and loan contracts between lenders and borrowers in particular. However, previous research has paid little attention to the role of audit quality in loan contracting, although audit quality is an important factor determining the credibility and quality of audited financial statements. In particular, no previous research has examined the issue of whether audit quality differentiation between Big 4 (previously 5, 6, or 8) and non-big 4 auditors does matter in the market for private debts such as bank loans, though Big 4 audits have been documented to be of greater value to participants in the equity and bond markets, compared with non-big 4 audits (e.g., Mitton 2002; Mansi et al. 2004; Pittman and Fortin 2004). 1 Given the lack of empirical evidence on the role of audit quality in private debt contracting, this study aims to provide systematic evidence on whether two auditor characteristics, i.e., auditor quality and tenure, influence the price term of loan contracts. To do so, we first investigate whether the loan rate that lenders charge to borrowers are lower for borrowers with Big 4 auditors than for those with non-big 4 auditors after controlling for borrowers credit quality and loan-specific characteristics. Second, we investigate whether and how auditor tenure, measured by the length of the auditor-client relationship, affects loan pricing. 1 To our knowledge, there are three studies that examine the role of audit per se in loan pricing. Johnson et al. (1983) provide experimental evidence suggesting that auditor association is not a significant factor affecting the bank loan rate. Blackwell et al. (1998) investigate economic values of varying levels of audit assurance (i.e., audits, reviews, compilations), using a small sample of 212 private (closely held) firms that have revolving credit arrangements with six banks located within a single state in the US. Kim et al. (2005) examine the effect of voluntary, non-statutory audits on the interest expenses (relative to short-term and long-term debts) using a sample of privately held Korean firms. Both Blackwell et al. (1998) and Kim et al. (2005) report evidence that audit per se leads to a lower loan rate or a lower interest rate. However, none of the above studies examine the issue of audit quality differentiation in loan pricing for publicly held borrowers

4 Farsiarticles.com Iran-article.ir Iranarticles.com A series of recent incidents of audit failures that started with the 2001 Enron debacle and the subsequent Andersen collapse have triggered a world-wide debate over whether the long-term auditor-client relationship potentially leads to the impairment of auditor independence and thus audit quality. Since the enactment of the Sarbanes-Oxley Act of 2002 which called for a study and review of the potential effects of requiring mandatory rotation of audit firms, several researchers have examined the effect of auditor tenure on audit and earnings quality (e,g., Davis et al. 2002; Myers et al. 2003). To our knowledge, however, no previous research has examined whether and how lenders take into account auditor tenure when assessing borrowers credit worthiness and setting the price term of loan contracting. In this paper, we aim to provide direct evidence on the effect of auditor tenure on loan pricing. Finally, as a supplemental analysis, we also examine whether the relation, if any, between the loan rate and two auditor characteristics, i.e., auditor quality and tenure, is conditioned upon information intermediation and monitoring activities by credit rating agencies. Previous research suggests that the information intermediation by credit rating agencies helps outside investors reduce the information asymmetry, which in turn contributes to increasing firm valuation (Lang et al. 2004) and lowering a cost of capital from the bond markets (e.g., Mansi et al. 2004, 2005). It is therefore interesting to examine how the audit quality variables interact with the information intermediation variable in the context of loan pricing. For this purpose, our analysis focuses on whether the loan rate-reducing effect, if any, of auditor quality and tenure differs systematically between borrowers with good credit ratings and those with poor credit ratings. Our regression results reveal the following. First, we find that lenders charge a significantly lower loan rate for borrowers with Big 4 auditors than for borrowers with non-big 4 auditors. Further analysis shows that lenders charge a higher loan rate for borrowers who change - 2 -

5 their auditors in general, and they charge a substantially higher loan rate for borrowers who downgrade their auditors from Big 4 to non-big 4 auditors in particular. Second, we find that the loan rate is inversely related to auditor tenure, suggesting that lenders view auditor tenure as a credit risk-reducing factor. Finally, we find that the loan rate-reducing effect of audit quality is greater for borrowers with low credit ratings than for borrowers with high credit ratings. This suggests that high-quality audits are of greater value to lenders when borrowers are faced with high credit risks. Overall, our empirical evidence is consistent with the view that audit quality plays a more important role in loan pricing, in particular, when borrowers have poor credit ratings. Our study adds to the existing auditing literature in the following ways. To our knowledge, this is the first study that documents direct evidence that banks take into account auditor quality and tenure when assessing borrowers credit quality and setting the loan rate. Our study also contributes to the existing loan contracting literature as well. We provide evidence that the quality of external audits is an additional factor that favorably impacts the price term of loan contracting, and that this positive effect is not subsumed by information intermediation and monitoring activities by credit rating agencies. Our evidence is consistent with the view that lenders view higher-quality audits and longer tenure as credit risk-reducing factors. Given the scarcity of empirical evidence on the issue, our findings provide useful insights into the role of auditor quality and tenure in the market for private debts such as bank loans. The remainder of the paper is structured as follows: In section 2, we develop our research hypotheses. In section 3, we specify an empirical model linking the loan spread with our test variables, namely auditor quality and tenure, and borrower-specific and loan-specific control variables. In section 4, we describe our sample and data sources and present descriptive statistics - 3 -

6 on our variables. Section 5 reports the results of our univariate tests. Section 6 reports the results of multivariate tests. In section 7, we conduct a variety of robustness check for our main regression results. In section 8, we perform further analysis to investigate the impact of auditor changes on loan pricing. We also examine whether the loan rate-audit quality relation is conditioned on credit risk perceived by credit rating agencies. The final section provides summary and concluding remarks. HYPOTHESIS DEVELOPMENT The Effect of Auditor Quality on Loan Pricing Banks are the largest providers of private debts and bank loans are the most important source of external finance for most firms around the world. Bank loan officers typically rely on audited financial statements to assess borrowers credit quality. On one hand, the use of highquality auditors enhances the credibility of audited financial statements, and thus alleviates information asymmetries between lenders and borrowers, which in turn reduces lenders monitoring costs. Therefore banks are likely to charge a lower loan rate for borrowers with Big 4 auditors than for borrowers with non-big 4 auditors. On the other hand, banks themselves are more sophisticated information processors, compared with a representative investor in the stock and/or public debt (bond) markets. Banks have the ability, skill and resources to collect, produce, and analyze relevant information and to assess the credibility of financial reports and the credit worthiness of a borrower. Moreover, banks often have access to private information about the borrower, for example, through direct communications with management. One may thus argue that the value of high-quality auditors may not be as high to banks as it is to investors in the equity and bond markets. In other words, the information-enhancing value of high-quality audits, - 4 -

7 Farsiarticles.com Iran-article.ir Iranarticles.com if any, may be subsumed by the superior ability of banks to acquire, verify and process borrowerspecific information and to assess the credibility and quality of borrowers financial statements. In such a case, there would be no significant difference in loan rates charged to borrowers with Big 4 auditors vis-à-vis those with non-big 4 auditors. Given the two conflicting views on the value of auditor quality in the bank loan market, it is an empirical issue whether or not the use of Big 4 auditors by borrowers has an incremental value to banks, when banks assess borrowers credit quality (before loan decisions are made), and monitor credit quality and/or renegotiate loan contract terms subsequent to changes in credit quality (after loans are granted). To provide empirical evidence on the issue, we test the following hypothesis with no prediction on the directional effect: H1: The loan rate charged by banks differs systematically between borrowers with Big 4 auditors and borrowers with non-big 4 auditors, other things being equal. The Effect of Auditor Tenure on Loan Pricing There are two conflicting views on the relation between auditor tenure and audit quality, which is at the center of current debates over the pros and cons of mandatory auditor rotation. One strand of research argues that as auditor tenure increases, auditor independence erodes, and thus client firms are given more flexibility in financial reporting. In this scenario, mandatory auditor rotation may contribute to improving audit quality by truncating the existing auditorclient relationship. Davis et al. (2002) provide evidence suggesting that discretionary accruals increase with auditor tenure. Choi and Doogar (2005) show that auditors with long tenure are less likely to issue going concern opinions, suggesting that audit quality decreases with the length of the auditor-client relationships

8 The other strand of research argues that audit quality increases with auditor tenure, and provides evidence supporting a positive association between audit quality and auditor tenure. Using several accrual measures as proxies for earnings quality (and thus audit quality), Myers et al. (2003) document a positive relation between audit quality and auditor tenure. Ghosh and Moon (2005) find that the magnitude of earnings response coefficients increases with auditor tenure, suggesting a positive relation between auditor tenure and audit quality. Mansi et al. (2004) document an inverse relation between auditor tenure and the cost of debt financing in the public bond market (measured by bond yield spreads over the benchmark yield). 2 To our knowledge, however, no previous research has examined the effect of auditor tenure on audit quality in the context of bank loan pricing. To provide empirical evidence on whether and how banks take into account auditor tenure when assessing borrowers credit quality and setting the loan rate, we test the following hypothesis with no prediction on the directional effect: H2: The loan rate charged by banks differs systematically between borrowers with long- tenure auditors and borrowers with short-tenure auditors, other things being equal. EMPIRICAL MODEL To investigate the effect of auditor quality and tenure on bank loan pricing, we specify the following regression model: AIS = α + α Big + α Tenure + β Size + β Leverage + β MB + β LogCoverageRatio 0 + β CurrentRatio + β Profitability + β Tangibility + β Beta + β Loss + γ LogMaturity + γ LogLoanSize + γ Syndicate + ( LoanPurposesDummies) ( IndustryDummies) + ( YearDummies) + ErrorTerm (1) 2 Johnson et al. (2002) and Geiger and Raghunandan (2002) also provide evidence suggesting a positive association between auditor tenure and audit quality in the context of reporting quality and the likelihood of a bankrupt firm receiving a going concern audit opinion, respectively

9 In Equation (1), the dependent variable, AIS, is the cost of the bank borrowing which is measured by the drawn all-in spread in basis points. This all-in spread represents the interest rate charged by banks (plus annual fee and the upfront or maturity fee) over the benchmark rate, i.e., LIBOR, and is paid by the borrower on all drawn lines of credit. We measure the cost of loan using a spread over LIBOR because most bank loans are priced in terms of the floating rate. Commercial banks typically assess the risk of a loan based upon the information on the business nature and performance of borrowing firms, and then set a markup over a prevailing benchmark rate such as LIBOR to compensate for the credit risk. The AIS variable thus reflects the banks perceived level of risk on a loan facility provided to a specific borrower. Our test variables, Big and Tenure, represent auditor quality and tenure, respectively. Big is a dummy variable which is equal to 1 if the incumbent auditor of a borrowing firm is one of Big 4 (or previously 5, 6 or 8) auditors which include Arthur Andersen, Arthur Young, Coopers & Lybrand, Ernst & Young, Deloitte & Touche, KPMG Peat Marwick, PricewaterhouseCoopers, Touche & Ross, and merged entities among them, and 0 otherwise. To the extent that Big 4 auditors are better able to help banks overcome the information problem, we expect the coefficient on Big to be negative (i.e., α 1 < 0 ), and its magnitude captures the difference in the loan spreads charged to borrowers with Big 4 auditors vis-à-vis those with non-big 4 auditors. Tenure is measured by the number of years of the auditor-client relationship. For example, if banks view longer (shorter) tenure as being associated with higher (lower) audit quality, one would observe a negative coefficient on Tenure i.e., α 2 < 0 (α 2 > 0). To isolate the loan pricing effect of audit quality from the effect of other borrowers characteristics, we include a set of borrower-specific variables that are deemed to affect borrowers credit quality and thus loan pricing, i.e., Size, Leverage, MB, Current Ratio, Log - 7 -

10 Coverage Ratio, Profitability, Tangibility, Beta, and Loss. The Size variable is measured by the natural log of total assets. We expect a negative coefficient on Size (i.e., β 1 < 0) because large firms are likely to have the better capacity to repay the loan, and thus have higher credit quality. The Leverage variable is measured by the ratio of long-term debt to total assets. Firms with high leverage are likely to have higher default risks, and thus have lower credit quality. To compensate for this potential credit risk, banks are likely to charge a higher loan rate for highleverage firms than for low-leverage firms. We therefore expect a positive coefficient on Leverage (i.e., β 2 > 0). We include the MB variable to control for borrowers growth potentials, and is measured by the market value of equity plus the book value of debt divided by the book value of total assets. To the extent that MB proxies for the growth opportunities, one would expect a negative relation between AIS and MB. However, growing firms are often faced with high risk. In such a case, one would observe a positive AIS-MB relation. We therefore do not predict the sign of the coefficient on MB. The Log Coverage Ratio variable denotes the natural log of one plus the interest coverage ratio which is measured by the ratio of earnings before interest, taxes, depreciation and amortization (EBITDA: Compustat data item 13) to interest expenses. We expect a negative coefficient on this variable because banks are likely to charge a lower interest rate for firms with a better ability to repay the debt (i.e., β 4 < 0). The Current Ratio variable denotes the ratio of current assets to current liabilities. The Profitability variable is measured by EBITDA divided by total assets, while the Tangibility variable by the ratio of plant, property and equipment (PP&E) to total assets. Similar to Bharath et al. (2006), we expect that banks charge lower loan rates for more liquid firms, more profitable firms and firms with more tangible assets (i.e., β 5 < 0, β 6 < 0, and β 7 < 0, respectively) because such firms have lower credit risks. The Beta variable denotes the market-model estimate of security beta. To obtain - 8 -

11 Beta for each year, we estimate the market model for each year using daily returns on an individual stock and the equally-weighted market returns. Loss is a dummy variable which is equal to 1 for loss firms and 0 otherwise. We expect a positive coefficient on both Beta and Loss. Previous research on bank loan contracts shows that several loan-specific characteristics are related to the interest rate charged by banks (e.g. Strahan 1999; Dennis et al. 2000; Bharath et al. 2006). We include in Equation (1) a set of loan-level variables to isolate the potential effect of loan characteristics on the loan spread from the effect of our test variables, namely Big and Tenure. The Log Maturity variable is the natural log of the loan maturity period (in months). The Log Loan Size variable is measured by the natural log of the amount of loan facility given to a borrower. Previous research provides evidence that banks charge a higher interest rate for the longer-term loan and for the smaller loan facility, respectively (e.g., Bae and Goyal 2003; Bharath et al. 2006). We therefore expect a positive sign on Log Maturity and a negative sign on Log Loan Size (γ 1 > 0 and γ 2 < 0, respectively). The Syndicate variable is a dummy variable that equals 1 for the syndicate loans and 0 otherwise. We include this variable to capture any difference, if any, in the interest rate charged between the syndicate and non-syndicate loans. In addition, we include Loan Purpose Dummies to control for any difference in loan pricing associated with the different purposes of loan facilities. 3 Finally, we include Industry Dummies and Year Dummies to control for potential differences in the loan spreads across industries and over years. 3 Our sample includes loan facilities with 22 different purposes specified by the LPC Dealscan database. We use only seven Loan Purpose dummies to capture the seven most common purposes, that is, corporate purposes, debt repayment, working capital, CP backup, takeover, acquisition line, LBO/MBO. The number of loan facilities with each of these seven purposes exceeds one percent of our sample, while the number of loan facilities with each of the other purposes is less than one percent of our sample

12 Farsiarticles.com Iran-article.ir Iranarticles.com SAMPLE, DATA SOURCES, AND DESCRIPTIVE STATISTICS The initial list of our sample consists of all publicly traded firms with bank loan data that are included in the LPC Dealscan database during the sample period, The LPC Dealscan database is an online database which contains a variety of historical bank loan data and other financial arrangements that are compiled from the Securities and Exchange Commission (SEC) filings by public firms and self-reporting by banks. 4 The database includes the loan data starting from 1986, and expands its coverage over time, in particular, after Thus we select 1996 as the starting year of our sample period. The loan data in the LPC Dealscan database are compiled for each deal and facility. Each deal, i.e., a loan contract between a borrower and bank(s) at a specific date, may have only one facility or have a package of several facilities with different price and non-price terms. 5 We consider each facility as a separate observation in our sample since many loan characteristics and the loan spreads vary across facilities. 6 Our sample includes term loans, revolvers and 364-day facilities, but excludes bridge loans and non-fund based facilities such as lease and standby letters of credit. We also require that all loan facilities in our sample are senior debts. 7 We then match the loans with borrowers financial statement data in Compustat, using the ticker symbol and name of each borrower. 8 We require that all the relevant annual accounting data of borrowers are available in the fiscal year immediately before 4 Other papers which use the LPC Dealscan database include Strahan (1999), Dichev and Skinner (2002), Beatty and Weber (2003), Asquith et al. (2005), Bharath et al. (2006), etc. 5 For instance, a deal may comprise a line of credit facility and a term loan with longer maturity. 6 As will be further discussed in Section 7, we also estimate our main regressions using only one facility within each deal and each firm year, and find that the results remain qualitatively identical with those using each facility as a separate observation. 7 Our sample selection criteria are similar to those used by Bharath et al. (2006). 8 This procedure leads to a substantial reduction in the number of available loan facilities because many borrowers included in the Dealscan database are subsidiaries of public firms, private firms and government entities rather than publicly traded companies, and some public companies are not covered by Compustat (Strahan 1999; Dichev and Skinner 2002)

13 the loan year. After applying the above procedures, we obtained a sample of 7,656 loaned facilities borrowed by 1,911 firms over the period. Table 1 presents the distribution of loan facilities in our sample by year and loan type. As shown in the table, nearly 57 percent of 7,656 loan facilities in our sample are for revolvers, while about 23 percent and 20 percent are for term loans and 364-day facilities, respectively. The number of loan facilities increases with years, reflecting an increase in the Dealscan coverage. (Insert Table 1 here) Panel A of Table 2 provides descriptive statistics on various characteristics of loan facilities in our sample. As shown in Panel A, the mean and median of the drawn all-in spread over LIBOR (i.e., AIS) are 172 and 150 basis points, respectively, with its standard deviation of about 130 basis points. The large standard deviation of AIS relative to its mean indicates a wide variation in AIS across loan facilities and deals. The mean (median) maturity period is about 41 (36) months with its standard deviation of about 24 months. The mean and median of loan facilities size are $313 and $146 millions (in US dollars) with a large standard deviation of $652 million, suggesting that its distribution is skewed with a wide variation across loan facilities and deals. About 93 percent of the loan facilities are syndicate loans with an average of more than nine different lenders (commercial banks and other financial institutions such as investment banks and insurance companies) in a syndicate group underwriting the loan facilities. The loan characteristics in our sample are, by and large, comparable with those of the Bharath et al. (2006) sample except for the size of the loan facility. The mean and median of loan facility size in our sample is much bigger than those in their sample ($177.5 and $50 millions), reflecting the fact that our sample period includes more recent years and the facility size has increased over time. (Insert Table 2 here)

14 Panel B of Table 2 presents descriptive statistics on borrowers characteristics in our sample. As shown in Panel B, 95.1 percent of all firm-years are audited by Big 4 auditors. The mean and median of Tenure (i.e., the length of the auditor-client relationship in years) are years and years, respectively, with its standard deviation of years, suggesting that the Tenure variable is reasonably distributed in our sample. The mean and median of the Size variable are and 6.753, respectively, with its standard deviation of On average, our sample firms have the long-term debt-to-total asset ratio of 26.2 percent, the market-to-book ratio of 1.753, and the current ratio of The Log Coverage Ratio variable, measured by the natural log of one plus the interest coverage ratio, has the mean (median) of (1.969) with a standard deviation of The descriptive statistics on Profitability and Tangibility show that for our sample, 14.5 percent and 34.9 percent of total assets are, respectively, EBITDA and tangible assets (i.e., PP&E). Our sample firms have, on average, security beta close to one, and 18.4 percent of our sample firms have experienced a loss during the sample period. RESULTS OF UNIVARIATE TESTS To assess the effect of auditor quality (Big 4 vs. non-big 4 auditors) on loan pricing, we partition the full sample into two sub-samples: (1) the Big 4 sample of borrowers with Big 4 auditors; and (2) the non-big 4 sample of borrowers with non-big 4 auditors. Panel A of Table 3 reports descriptive statistics on our major research variables separately for the Big 4 sample and for the non-big 4 sample, along with the results of tests for the mean and median differences between the two samples (t-test and Wilcoxon z-test, respectively). As shown in Panel A, the mean and median of the drawn all-in spread (AIS) are 168 and 150 basis points, respectively, for the Big 4 sample, while they are 252 and 250 basis points, respectively, for the non-big 4 sample

15 Both the mean and median differences of 84 and 100 basis points are significant at less than the onepercent level, suggesting that banks charge a significantly lower loan rate for borrowers with Big 4 auditors than for borrowers with non-big 4 auditors. These differences are economically significant as well, considering the mean and median of AIS for the full sample are 172 and 150 basis points, respectively (as reported in Table 2). The mean and median of Tenure are and years, respectively, for the Big 4 sample, while they are and years, respectively, for the non-big 4 sample. These mean and median differences are significant at less than the one percent level, suggesting that, on average, Big 4 auditors have a longer tenure than non-big 4 auditors. With respect to a set of nine variables representing borrowers characteristics (Size to Loss), we observe that the mean and median of Size, Leverage, MB, Current Ratio, Tangibility and Beta are significantly different between the Big 4 and non-big 4 samples at less than the one percent level. On average, borrowers in the Big 4 sample are larger, more leveraged, have a higher market-to-book ratio and a lower current ratio, more tangible assets and a higher beta, compared with borrowers in the non-big 4 sample. However, we observe no significant difference in the mean and median of Log Coverage Ratio, Profitability, and Loss between the two sub-samples. With respect to a set of four variables representing loan characteristics, borrowers in the Big 4 sample, on average, have a larger loan facility, and are more likely to have a syndicate loan, and attract more participant lenders, compared with those in the non-big 4 sample. To assess the effect of auditor tenure on loan pricing, we partition the full sample into two sub-samples on the basis of the median tenure of 8 years: (1) the long-tenure sample of borrowers with their auditor tenure longer than or equal to eight years; and (2) the short-tenure

16 sample of borrowers with their auditor tenure less than eight years. Panel B of Table 3 reports descriptive statistics on our major research variables separately for the long-tenure sample and for the short-tenure sample, along with the results of tests for the mean and median differences between the two sub-samples. As shown in Panel B, the mean and median of AIS are about 145 and 111 basis points, respectively, for the long-tenure sample, while they are about 202 and 200 basis points, respectively, for the short-tenure sample. Both the mean and median differences of 57 and 89 basis points, respectively, are significant at less than the one percent level. These differences are economically significant as well, considering the mean and median of AIS for the full sample are 172 and 150 basis points, respectively (as reported in Table 2). In short, our data reveal that the loan spread decreases significantly with auditor tenure, suggesting that banks take into account auditor tenure when assessing the credibility of financial statements and setting the loan rate percent of borrowers in the long-tenure sample have Big 4 auditors, while 93.1 percent in the short-tenure sample. This difference is significant at less than the one percent level. With respect to a set of nine variables representing borrowers characteristics, there are significant differences in their mean and median values of most variables between the longtenure and short-tenure samples. Compared with borrowers in the short-tenure sample, on average, those in the long tenure sample are larger in size, and have a lower current ratio, more tangible assets, a smaller beta, and a lower likelihood of incurring a loss. With respect to a set of four variables representing loan characteristics, borrowers in the long-tenure sample, on average, have a shorter maturity period and a larger loan facility, are more likely to have a syndicate loan, and attract more participant lenders, compared with those in the short-tenure sample. (Insert Table 3 here)

17 Table 4 reports Pearson correlation coefficients among all the variables included in Equation (1). Consistent with the results of our univariate tests in Table 3, AIS is negatively correlated with Big and Tenure at less than the one percent level with their magnitude of and -0.23, respectively, suggesting that the use of Big 4 auditors and long tenure auditors is inversely associated with a lower loan spread. Consistent with our expectation, AIS is positively correlated with Log Maturity and negatively correlated with Syndicate. This suggests that banks charge a lower (higher) loan rate for short-term (long-term) loans and syndicate (non-syndicate) loans. The negative correlations of AIS with Size, MB, Log Coverage Ratio, Profitability, and Tangibility suggest that banks charge a lower loan rate for borrowers with low credit risks. The positive correlations of AIS with Leverage, Current Ratio, Beta, and Loss support the view that banks charge a higher loan rate for borrowers with high credit risks. With respect to the correlations among explanatory variables in Equation (1), Size is highly correlated with Log Loan Size with the magnitude of This is as expected because banks are highly likely to offer large loan facilities to large firms. The correlations between other explanatory variables in Equation (1) are reasonable with the highest correlation of between Log Coverage Ratio and Leverage. (Insert Table 4 here) In summary, the results of univariate tests suggest that banks charge a lower loan rate for borrowers with Big 4 auditors or long tenure than those with non-big 4 auditors or short tenure, respectively. However, the significant differences in the borrower-specific and loan-specific variables between the Big 4 and non-big 4 samples and between the long-tenure and short-tenure samples suggest that the effect of these variables on loan pricing should be controlled for when assessing the impact of auditor quality and tenure on the loan spread. In the next section, we

18 Farsiarticles.com Iran-article.ir Iranarticles.com therefore conduct multivariate tests to isolate the loan pricing effect of auditor quality and tenure from the effect of borrower-specific and loan-specific characteristics. RESULTS OF MULTIVARIATE TESTS USING THE FULL SAMPLE Table 5 presents the results of the OLS regressions in Equation (1) using the full sample of 7,656 facility-years over the period. As shown in Columns (1) and (2) of the table, the coefficient on Big (Tenure) is significantly negative when AIS is regressed on Big (Tenure) and other control variables, suggesting that banks charge a lower loan rate for borrowers with Big 4 (long-tenure) auditors after controlling for all other borrower-specific and loan-specific variables. As shown in Column (3), when both Big and Tenure are included in the regression, the coefficients on Big and Tenure are both significant with negative signs. The above results are consistent with the view that banks take into account auditor quality and tenure when assessing borrowers credit quality and setting the loan spread. Our results support the view that highquality audits alleviate the information asymmetry between lenders and borrowers and the associated monitoring costs, which in turn contributes to lowering the loan spread charged by banks. Overall, our results are consistent with Mansi et al. (2004) who document that external audits by Big 4 auditors and long-tenure auditors lead to a reduced cost of debt in the public bond market and Pittman and Fortin (2004) who document that Big 4 audits are associated with a lower interest cost of debt in early public years of IPO firms. In Column (3), the coefficient on Big is (t = -2.26), indicating that the difference in loan spread between borrowers with Big 4 and non-big 4 auditors is nearly 14 basis points. As reported in Table 2, the average amount of loan facility is about $313 millions for our sample and the mean maturity is about 41 months or 3.5 years. This means that, on average, borrowers

19 with Big 4 auditors can save the interest cost of about $438,200 per year over the maturity period of 3.5 years, which is economically significant as well. In Column (3), the coefficient on Tenure is (t = -5.52), suggesting that, on average, borrowers can save an interest rate of around 1.3 basis points by retaining their relationship with the incumbent (Big 4 or non-big 4) auditor for one additional year. The associated amount of interest cost saving is about $40,690. (Insert Table 5 here) With respect to the coefficients on control variables, we find all coefficients except for MB and Syndicate are significant at less than the one percent level with their signs consistent with our expectations and the findings of previous research such as Bharath et al. (2006). More specifically, AIS is significantly and positively associated with Leverage, Beta, and Loss, while it is negatively associated with Size, Current Ratio, Log Coverage Ratio, Profitability, and Tangibility. In addition, AIS is positively associated with Log Maturity and negatively associated with Log Loan Size. ROBUSTNESS CHECKS We perform several sensitivity tests to check the robustness of our main results reported in Table 5. Our analyses in Table 5 consider each loan facility as an independent observation although a borrower can obtain several facilities in the same year. As a sensitivity check, we use the following ways to reconstruct our sample and then re-estimate Equation (1): (i) including only one facility of each deal (the largest facility in terms of facility size); (ii) including only one facility for each firm year (the largest facility in the first deal in each year); (iii) conducting Fama-MacBeth regressions on the reduced sample constructed in (ii). Columns (1) to (3) of

20 Table 6 report the corresponding results. The magnitude, sign, and significance of the coefficients on Big and Tenure in Table 6 are similar to those in Table 5. To further check whether our inferences on the test variables, namely Big and Tenure, are distorted by the existence of potential endogeneity problems, we re-estimate Equation (1) using one-year lagged values of Big and Tenure. As shown in Column (4) of the table, the use of oneyear lagged values for our test variables does not alter our results reported in Table 5, suggesting that our earlier results are robust to potential endogeneity problems associated with our test variables. In our regression specification in Equation (1), the loan spread is linked to borrowers auditor choice (i.e., Big 4 vs. non-big 4) and many other variables. Suppose that borrowers with high credit quality (and thus having lower AIS) are more likely to choose Big 4 auditors. In such a case, the error term in Equation (1) is likely to be correlated with whether borrowers choose Big 4 auditors or not, and our estimate of the coefficient on Big is likely to be biased. To address a concern over this potential self-selection bias, we estimate the two-stage treatment-effect model (Greene 2000). In the first stage, we estimate a probit auditor-choice model, and then obtain the Inverse Mills ratios. 9 In the second stage, we then estimate Equation (1) after 9 The probit auditor-choice model is specified as follows: Big = α + α Sale + α Liability + α MB α Tangibilit y + α Invrev + α DP + α Turnover 4 + α 8Marg in + α 9Invtrating + α10 HighAnalys t + α11shrinc + ( Industries Dummies ) + ( YearsDummi es) + ε Where Big is an indicator variable which is equal to one for borrowers with Big 4 auditors and zero otherwise; Sale is the natural log of net sales; Liability is total liabilities divided by total assets; MB is the market-to-book ratio; Tangibility is net PP&E divided by total assets; Invrev is the sum of inventory and receivables over total assets; DP is depreciation and amortization over total assets; Turnover is net sales divided by total assets; Margin is income before extraordinary items divided by net sales; Invtrating is a dummy variable which is equal to one for borrowers with S&P investment grade rating (BBB- or above) and zero for firms with non-investment grade rating or without rating value; HighAnalyst is a dummy variable which is equal to 1 for firms followed by more than sevem (the median) analysts and zero for firms followed by less than seven analysts or not covered by IBES; Shrinc is a dummy variable which is equal to one if the number of shares outstanding increases by more than 10 percent during the current fiscal year and zero otherwise. The sample size used for estimating Equation (2) as well as Equation (1) with the Inverse Mills ratio included reduces to 7,559 from 7,656 facility-year observations (2)

21 including the Inverse Mills ratio (obtained in the first stage) as an additional independent variable. Column (5) of Table 6 reports the result of the second-stage regression. We find that the coefficient on Inverse Mills Ratio is significantly positive at less than five percent level, suggesting that self-selection bias may not exist. The coefficients on Big and Tenure are significantly negative at less than one percent level. Overall, the inclusion of the Inverse Mills ratio strengthens our result in the sense that the coefficients on Big and Tenure reported in Column (5) of Table 6 are more significantly negative and larger in magnitude, compared with those reported in Column (3) of Table 5. This suggests that our main regression results reported in Table 5 (without including the Inverse Mills ratio) are robust with respect to potential selfselection biases. Though not tabulated, we also conduct several additional sensitivity checks. First, we estimate Equation (1) after including the Loan Type dummies to distinguish among different types of loan facilities in our sample, i.e. term loans, revolvers greater than one year, revolvers less than one year and 364-day facilities. Not reported is that the inclusion of the loan type dummies does not alter our main results reported in Table 5. Second, we estimate Equation (1) after including an additional dummy variable, Secured, which takes the value of one for secured loans and zero otherwise. Though not reported, we find that the inclusion of this Secured dummy does not alter our main results presented in Table 5. We find that the coefficient on Secured is significantly positive with its magnitude of and its t-value of The result indicates that banks charge a higher loan spread for secured loans by the amount of about 71 basis points because we lose some observations due to missing values required for estimating Equation (2). For brevity, the estimation results of auditor-choice model are not reported here

22 than for unsecured loans. 10 Third, we also estimate Equation (1) after including the Performance Pricing dummy which equals one for loans with performance pricing provisions and zero otherwise. Under a typical performance pricing provision, the loan rate is allowed to decrease directly with the improvement in credit quality. Not reported is that the inclusion of the Performance Pricing dummy in Equation (1) does not alter our main results presented in Table 5. We also find that the coefficient on Performance Pricing is significantly negative with its magnitude of and its t-value of This suggests that banks charge a lower rate for loans with the performance pricing provision by the amount of 31 basis points than loans without it, a finding consistent with Asquith et al. (2005). Finally, in our analyses so far, we measure auditor tenure by the number of years of the auditor-client relationship. We also use a dummy variable which equals one if the tenure for a firm year is longer than the median tenure in our sample (eight years) and zero otherwise, and then re-estimate Equation (1) using this new measure of auditor tenure. Though not reported, we find that the coefficient on this dummy variable is significantly negative. In addition, following the procedure suggested by prior research on auditor tenure (e.g., Myers et al. 2003; Ghosh and Moon 2005; Mansi et al. 2004), we construct a reduced sample of borrowers with at least five years of auditor tenure and re-estimate Equation (1) using this reduced sample. The results using this reduced sample remain qualitatively similar to those reported in Table 5. In short, our main results reported in Table 5 are robust to a variety of sensitivity checks such as alternative treatments of multiple loan facilities of each deal and for each firm year, potential residual cross correlation, potential endogeneity problems associated with auditor 10 This finding is consistent with Dennis et al. (2000) and Berger and Udell (1990) who find that banks are more likely to require collaterals for borrowers with high credit risk and to charge higher rates for secured loans than for unsecured loans

23 Farsiarticles.com Iran-article.ir Iranarticles.com quality and tenure, and the inclusion of various indicator variables representing Loan Type, Secured, and Performance Pricing. (Insert Table 6 here) FURTHER ANALYSES The Impact of Auditor Changes on Bank Loan Pricing To alleviate a concern that our levels results so far are possibly driven by correlated omitted variables and to examine the effect of auditor switches on the change in the loan spread, we examine the relation between changes in auditors and changes in loan spreads. In so doing, we measure the change in the drawn all-in spread, i.e., AIS, by the change in the facility-sizeweighted average of AIS on all loan facilities for a firm from year t - 1 to year t. Similarly, we measure the change in loan maturity, i.e., Log Maturity, by the change in the natural log of facility-size-weighted average of maturity periods (in months) for all loan facilities for a firm from year t - 1 to year t. We measure the change in loan facility size, i.e., Log Loan Size, by the change in the natural log of average dollar amount of all loan facilities for a firm from year t - 1 to year t. We do not include the change in Syndicate for our changes regressions because it is difficult to identify the Syndicate status for the yearly facility-size-weighted average loan. We use five different auditor change dummies, i.e., Change, Upgrade, Downgrade, Big and NonBig to capture any type of auditor change, a change from a non-big 4 auditor to a Big 4 auditor, a change from a Big 4 auditor to a non-big 4 auditor, a change within Big 4 auditors, and a change within non-big 4 auditors, respectively. After applying the above definitions of the change variables, we obtain a total of 2,974 observations available to this change analysis. Out of 2,974, there are 388 observations of all

24 types of auditor changes which include seven observations of upgrade changes, 14 observations of downgrade changes, 353 changes within Big 4 auditors, and 14 changes within non-big 4 auditors. Table 7 presents the results of change regressions where all variables are measured in terms of their changes from year t - 1 to year t. In Columns (1) and (2), we include Change to capture the effect of (any type of) auditor changes on the loan spread change. In Columns (3) and (4), we include the dummy variables indicating four types of auditor changes, i.e., Upgrade, Downgrade, Big and ΔNonBig, instead of Change. As a sensitivity check, two loan-specific control variables (i.e., ΔLog Maturity and ΔLog Loan Size) are excluded in Columns (1) and (3), but they are included in Columns (2) and (4). As reported in Columns (1) and (2) of the table, the coefficient on Change is and , respectively, which is significant at less than the one percent and five percent levels, respectively. This suggests that, on average, banks perceive auditor changes as an event that deteriorates the quality and/or credibility of accounting information, and thus they charge a higher loan spread for borrowers with auditor changes. As shown in Columns (3) and (4), the coefficients on Upgrade and Big are insignificantly positive. However, the coefficients on Downgrade are (t= 2.33) and (t = 2.32) as reported in Columns (3) and (4), respectively. In other words, banks charge a higher loan spread for borrowers who switch their auditors from Big 4 to non-big 4 auditors by the amount of nearly 80 basis points, which is economically significant as well. Also the coefficients on NonBig are (t = 2.27) and (t = 2.32) in Columns (3) and (4), respectively. This suggests that similar to auditor downgrading, banks perceive auditor switches within non-big 4 auditors to be a credit qualitydeteriorating event. Consistent with our expectation, banks charge a significantly higher rate for clients with auditor downgrading than for those with auditors switches within non-big 4 auditors

25 by the amount of about 20 basis points. The results here are in sharp contrast with those reported by Mansi et al. (2004) that only the auditor upgrade leads to a significant decrease in bond yield spreads. However, our results are consistent with the finding of Fried and Schiff (1981) that there is a negative stock price reaction to auditor switches including the switch from a small to a large auditor. (Insert Table 7 here) Effect of Credit Rating Previous research provides evidence that the information uncertainty is greater for highrisk firms than for low-risk firms (e.g., Beneish 1997; Christensen et al. 1999). Moreover, Mansi et al. (2004) find that the favorable effect of auditor quality on the bond yield spread is significant for the non-investment-grade sample, but is insignificant for the investment-grade sample. They also find that the favorable effect of auditor tenure on the bond yield spread is more significant for the non-investment-grade sample than for the investment-grade sample. Their study suggests that the value of high-quality audit in the public bond market is more pronounced for high-risk firms than for low-risk firms. We investigate whether the effect of audit quality on lowering the loan spread is greater for high-risk firms than for low-risk firms. To address this question, we partition the full sample using S&P Issuer Bond Rating data (Compustat item 280). 11 We then partition the full sample into two sub-samples, namely: (1) the investment-grade sample of borrowers with their S&P Issuer Bond Rating of BBB- or above (N =2,275); and (2) the non-investment-grade sample of 11 The Issuer Credit Rating (ICR) is a current opinion of an issuer s overall creditworthiness apart from its ability to repay individual obligation and focuses on the obligor s capacity and willingness to meet its long-term financial commitments. Prior to September 1, 1998, this item is named as S&P Senior Debt Rating