TO TELL OR NOT TO TELL: The Value of Corporate Disclosure

Transcription

1 TO TELL OR NOT TO TELL: The Value of Corporate Disclosure Sumon C. Mazumdar, Atulya Sarin and Partha Sengupta This version: November 7, 2000 Mazumdar is at LECG, LLC, Emeryville, CA and the Finance Department, Haas School of Business, University of California-Berkeley. Sarin is at the Finance Department, Leavey School of Business, Santa Clara University. Sengupta is at the Accounting Department, R. H. Smith School of Business, University of Maryland, College Park, MD Sengupta is the corresponding author and may be contacted through Phone: (301) ; fax: (301) ; We thank Bipin Ajinkya, Sanjeev Bhojraj, Nick Ordway, Russell Taussig, Jenny Teruya, and workshop participants at the University of Hawaii at Manoa, University of Maryland, University of Missouri at Columbia, Texas Christian University and the University of Texas at Dallas for many helpful comments. Mazumdar gratefully acknowledges an SSHRC grant that provided partial support for this project. The usual disclaimer applies.

2 TO TELL OR NOT TO TELL: The Value of Corporate Disclosure Abstract Corporations have widely varying disclosure strategies. On one extreme, some firms periodically provide detailed guidance to the investment community in an attempt to avoid surprises. On the other hand, other firms are reluctant to disclose information and in extreme cases even face law suits as a result. This decision has been further complicated by the implementation of Regulation FD by the SEC that requires that the company make material market moving information available to all investors at the same time. In this paper we provide empirical evidence which shows that a high quality of disclosure is valuable. Specifically, we find that corporations disclosure quality is incorporated by banks in their loan pricing decision despite their privileged access to corporate information and superior monitoring ability.

3 TO TELL OR NOT TO TELL: The Value of Corporate Disclosure I. Introduction Corporations have widely varying disclosure strategies. On one extreme, some firms periodically provide detailed guidance to the investment community in an attempt to avoid surprises. On the other hand, other firms are reluctant to disclose information and in extreme cases even face law suits as a result. A recent study by Bajaj, Mazumdar and Sarin (2000) shows that between 1991 and 1999 there were 328 lawsuits filed against corporations for "failure to disclose." On 25 th October, 2000, the SEC introduced Regulation FD aimed at ensuring "fair disclosure". Regulation FD is intended to prevent selective disclosure by public companies, and put individual traders on equal informational footing with Wall Street money managers. This has put an additional wrinkle into an already difficult management decision. A number of researchers have argued that firms may withhold private information if there are sufficiently large costs of disclosing such information. Verrecchia (1983) referred to the costs of disclosing information as proprietary costs. Such proprietary costs include direct costs of preparing and disseminating information and other costs associated with the disclosure of proprietary information, such as adverse action by competing firms, entry of new firms into the industry and political costs arising from possible threat of regulation and antitrust investigations. Verrecchia argued that in the absence of costs of disclosures, firms would release all information. The incentives for full disclosures arise from the assumption that investors know when firms are withholding private information. Nondisclosure in this framework causes investors to infer that the firms have bad news and this has a negative impact on stock prices. Firms, being aware of this, would release all information. If there are some costs to disclosing information, however, 2

4 investors would not be able to ascertain whether firms are withholding bad news or proprietary information. Thus an equilibrium with some nondisclosure can exist when there are sufficiently large costs of disclosing information. Extending Verrecchia's idea of "proprietary costs," a number of recent researchers like Darrough and Stoughton (1990), Wagenhofer (1990), and Feltham and Xie (1992) have shown that a partial-disclosure equilibrium is feasible in situations where the disclosing firm has a potential opponent. The existence of competitors in the market indicates that there are potentially large costs of disclosing information in the form of loss of profits and market share. In this study, we attempt to estimate the benefit of disclosure. Specifically, we estimate whether a policy of detailed, timely and informative disclosures can reduce the interest rate a firm pays on its private debt contracts. Such an examination of the potential link between corporate disclosure quality and the cost of private debt is important for at least two reasons. First, private debt financing is the largest source of external financing for most nonfinancial firms in the US. Petersen and Rajan (1994) document that private borrowing from financial institutions comprised about 61 to 76 percent of all debt (depending on firm size) for a sample of 3,404 small firms. Johnson (1997) examines the sources of debt for all non-financial firms on the Compustat primary, secondary and tertiary files in 1989 and finds that the average ratio of private debt to total debt for this sample is 74 percent. 1 Given the reliance on private borrowing by US firms, it is important to examine if corporate disclosures have an impact on the cost of private debt financing. If such a link can be established, it would point to a major benefit of corporate disclosures. Second, modern banking theory postulates that banks have superior credit assessment 1 Other studies, such as Houston and James (1996) found similar results. 3

5 skills compared to individual investors. 2 Hence banks serve as delegated monitors on behalf of their depositors. Banks monitoring ability emanates from their information-gathering activities [Diamond, (1984)] or as a result of long-standing business relationships with their borrowers [e.g., Black (1975), Fama (1985)]. The notion that banks have superior information about a firm s prospects is supported by existing empirical evidence. Several studies have identified a significant increase in the borrower s (abnormal) equity return following a private loan announcement, 3 in contrast to the generally negative reaction to a firm s issue of public securities [Scott and Smith (1986)]. Since private lenders are experts at gathering and processing information, they are least likely to value public disclosures by the borrower and factor such disclosures into their loan pricing. Therefore, if firms actually receive bank loans at lower rates due to their timely disclosure policy then such a cost saving might be interpreted as the lower bound on the value of a corporate disclosure policy. We find such evidence in this paper. Specifically, we find that the loan spread is negatively associated with the quality of a firm s overall disclosure quality after controlling for other potential determinants of loan spread, including other sources of information such as crossmonitoring by bond rating agencies that may reduce banks own monitoring costs. Our results further show that a firm pays an average of 0.92 basis points less if its disclosure quality measure is higher by one point. The firm with the highest disclosure score in the sample enjoys an interest rate that is about 50 basis points (0.5%) lower than the firm with the lowest disclosure score in the sample. Since the firm with the highest disclosure score in our sample had a debt issue of 2 In addition to direct monitoring, according to some recent studies banks may rely on the information generated by other monitors such as bond rating agencies [Booth (1992)]. However, evidence in support of such a crossmonitoring hypothesis appears to be mixed since Chen et al (2000) found that the influence of cross-monitoring on bank loan pricing is not significant if banks differential monitoring ability is considered explicitly in the regression model. 3 See Mikkleson and Partch (1986), James (1987), Lummer and McConnell (1989), Slovin et al. (1992), Best and Zhang (1993), Preece and Mullineaux (1994), Billett et al. (1995), and Thakor (1996). 4

6 $400 million, annual interest savings of approximately $2 million could be attributed to improved disclosures, suggesting that the benefits from improved disclosures could be significant. 4 Overall, we conclude that the a firm s disclosure policy is relevant even to private lenders who enjoy monitoring cost efficiencies. Although private lenders can collect important information directly from the borrowers, it does not completely eliminate information asymmetry between the two parties. In fact, the presence of such information asymmetry is the raison d etre of the various covenants imposed by private lenders. Asymmetrically informed lenders would be concerned about two possibilities: (i) the firm is withholding information that could increase the default risk of the loan and (ii) the firm did not truthfully disclose the requested information. An estimate of these two possibilities would enter into the default risk calculations and should thus affect the covenants attached to the loan and the loan spreads. The higher the probability of withholding or false reporting, as assessed by the lender, the higher is the estimated default risk of the firm and the larger the risk premium (loan spread) charged by the lender. In order to estimate the probability of withholding and false reporting, the lender could examine the firm s current and past disclosures. Firms that consistently make timely, detailed, and clear disclosures, and are easily accessible to financial analysts for discussions and clarifications are less likely to withhold unfavorable information or provide false information. Our results show that firms are associated with a lower loan spread. The rest of the paper is organized as follows. Section 2 describes the research method, section 3 describes the sample, section 4 reports the results and the last section concludes the paper. 4 This figure of $2 million represents the portion of such benefits passed on by the private lender to the borrower due to competition in the lending market. Therefore, the total benefits of disclosure could be even larger if the lending 5

7 2. Research Method We examine the hypothesis that the loan spread on private debt is negatively associated with corporate disclosure scores. Our test of this hypothesis involves regressing the loan spread on the disclosure variable, and other variables that have been shown to affect loan spread, using the following model: SPREAD t+1 = f (DISC t, Control Variables) (1) where SPREAD t+1 is the quoted loan spread in basis points over the benchmark rate (LIBOR) on the first new loan of year t+1. DISC t is a measure of disclosure quality over three years ending in year t. These variables and the control variables are discussed below. Loan Spread (SPREAD) Data on private corporate debt contracts were obtained from the November 1994 release of the Dealscan database compiled by the Loan Pricing Corporation. 5 This database provides detailed information on 23,340 loan agreements between U.S. corporations and a variety of financial institutions such as commercial banks, finance companies, insurance companies, investment banks and savings and loans over the period 1987 through early Information is provided on loans made to more than 8,400 firms. The data is primarily collected from SEC filings. Information on loan characteristics such as size, maturity, the loan s purpose and type, and spread are provided. About 70 percent of the loans are floating rate loans with the interest rate given as a basis point spread over one of three indices: LIBOR, PRIME or CD. We selected market was not perfectly competitive. 5 See Carey et. al. (1998) for a description of the Dealscan data. 6

8 the LIBOR rate for the present analysis as this yielded the most observations. If the PRIME (CD) rate is used instead, the sample size drops by about 30 (49) percent. The Disclosure Quality Measure (DISC) Data on corporate disclosure efforts was obtained from the annual volumes of the Report of the Financial Analysts Federation Corporate Information Committee (FAF report) currently published by Financial Analysts Federation (FAF) branch of the Association for Investment Management and Research (AIMR). 6 A detailed description of this data is provided in Lang and Lundholm [(1993), (1996)] so only the main features are summarized here. Briefly, the FAF report includes summary evaluations of overall disclosure efforts of a selected set of firms. The evaluations are performed by industry sub-committees comprised of analysts specializing in that industry. Each analyst within a sub-committee examines annual reports, quarterly reports, proxy statements, other published information such as press releases and fact books, and direct disclosures to the analysts in the form of meetings and responses to analyst inquiries, of all selected firms within that industry. The evaluations of multiple analysts are then averaged and this is typically reported in the FAF report in the form of a score out of 100 possible points. 7 The FAF evaluates between four hundred to five hundred firms each year. The set of firms selected remains fairly stable from year to year. On average, thirteen analysts are on an industry sub-committee. In evaluating a firm s disclosure efforts, analysts emphasize the detail, clarity and timeliness of disclosures. Since all major forms of corporate disclosures are considered, the FAF scores provide an appropriate proxy for a firm s overall disclosure quality. However, the FAF 6 The disclosure scores are currently published by the Association for Investment Management and Research (the umbrella organization for the Financial Analysts Federation and the Institute of Chartered Financial Analysts) under the title: An Annual Review of Corporate Reporting Practices. The old title is used in the paper to be consistent with prior literature (Lang and Lundholm 1993, 1996). 7

9 scores are based on analysts perceptions of corporate disclosure practices. Therefore, biases and errors in analysts judgments could affect these scores. To control for this possibility, the FAF reports only the average scores across industry analysts. The FAF also provides the analysts with detailed guidelines and a comprehensive checklist of criteria to help standardize the ratings process both within and across industries. Recent studies using the disclosure scores generally indicate that the scores are correlated with common measures of information asymmetry in the market. 8 In making default risk calculations, financial institutions should be particularly interested in identifying if the borrowing firm has a consistent strategy of disclosing private information in a timely manner. In this evaluation, the financial institutions should explore all forms of corporate disclosures, rather than look at specific disclosures such as management forecasts. Hence the FAF scores seem to be appropriate for this study. Furthermore, private lenders should examine disclosures over a number of periods rather than just current period disclosures, in order to identify the borrower s overall willingness to disclose information. As a result, following Sengupta (1998), we averaged the total disclosure score of each firm over the years t, t-1, and t-2 to obtain an aggregate measure of a firm s disclosure effort over three years. The Control Variables Control variables included in this study were developed from the recent literature on the determinants of loan spread [e.g., Scott and Smith (1986), Berger and Udell (1990), Booth (1992), Berger and Udell (1995), and Chen et al. (2000, 1996)]. These can be classified into 7 A few industries provide only rankings rather than scores. A majority of the industries also report separate scores for annual reports, quarterly reports, and investor relations. 8 Lang and Lundholm (1993, 1996) document a negative association between the scores and analyst forecast errors, the standard deviation of stock returns, and dispersion of security analysts' earnings per share forecasts, while Welker (1995) documents a negative relation between the disclosures scores and the bid-ask spread set by the market makers. Healy et. al. (1995) showed further that firms with increases in the disclosure scores enjoy greater stock liquidity, analyst following and institutional ownership. 8

10 three categories: loan contract characteristics (such as size, maturity, and special features of the loan), borrower characteristics (capturing the default risk of the firm), and macroeconomic characteristics that affect the time series variations in loan spread. Based on these studies the following control variables were included: Loan contract characteristics: 9 LSIZE = natural log of the loan size. An increase in loan size could suggest greater default risk indicating a positive association between SPREAD and LSIZE. LMATUR = natural log of the loan's original days-to-maturity. Loans with longer maturity could be associated with greater default risk and higher SPREAD. However, Booth (1992) had concluded that the effect of loan maturity on loan spread is uncertain. 10 SECURE = binary variable which equals 1 if the loan is collateralized and 0 otherwise. Ceteris paribus, a secured loan indicates lower default risk. However, prior studies suggest that costly collateralization is normally undertaken only for risky loans so that SPREAD is expected to be positively associated with SECURE. BID = binary variable which equals 1 if the loan contains the option to price in terms of a different index and 0 otherwise. The bid option denotes a cost to the lender of foregone interest if the borrower can switch to a lower index. Therefore, BID should be positively associated with SPREAD. RESTRUC = binary variable which equals 1 if the loan is used for corporate restructuring (leveraged buyouts, takeover acquisitions, recapitalization, or debt consolidation), 9 Data on these variables were collected from Dealscan. 10 Merton (1974) demonstrates that the risk premium on a bond, defined as the difference in the yield to maturity on the risky debt provided that the firm does not default and the riskfree rate, can be either an increasing or a 9

11 0 otherwise. Such restructuring potentially subjects the lender to greater default risk so that a positive relationship is expected between SPREAD and RESTRUC. SENIOR = binary variable which equals 1 if the loan taken is senior to other outstanding debt of the firm, 0 otherwise. Since the seniority feature increases the default risk of the loan, SENIOR is expected to be positively associated with SPREAD. Borrower characteristics: We used two alternative sets of control variables to proxy for borrower characteristics. The first set of variables included an overall proxy for financial strength given by the Altman Z score proposed by Altman and McGough, (1974). The second set of control variables included separate measures of profitability, solvency and growth instead of the Altman Z score. MVAL = market value of the firm s common equity at the end of year t deflated by the aggregate consumer price index. 11 Firms with higher market value of equity are expected to have lower SPREAD. STDRETN = the standard deviation of daily stock returns over years t and t-1. This measures the market risk exposure of the firm and thus is expected to be positively associated with SPREAD. ALTMAN = Altman Z score defined as: 1.2X X X X 4 + X 5 where, X 1 = (current assets current liabilities) / total assets, decreasing function of the loan s maturity. Hence, the relationship between the loan spread (which is a function of the risk premium) and the loan s maturity is ambiguous. 11 Since the sample includes observations from different years, the market value figures are deflated by the seasonally-adjusted monthly consumer price index as of the fiscal year end, to improve the comparability of these numbers across years. Consumer price index information was obtained from the Federal Reserve FRED database. 10

12 X 2 = retained earnings / total assets, X 3 = earnings before interest and taxes / total assets, X 4 = market value of equity / total liabilities, and, X 5 = sales / total assets. All variables are measured at the end of year t. ALTMAN is expected to be negatively correlated with SPREAD. LEVER = long term debt divided by total assets, both at the end of year t. LEVER is expected to be positively associated with default risk and SPREAD. MARGIN = income before extraordinary items of year t divided by net sales of year t. Firms with greater profit margins are expected to have lower loan spreads. MKBK = market value of common equity divided by the book value of common equity at the end of year t. The market-to-book ratio represents growth opportunities of the borrower. If high growth firms are associated with greater risk, we would expect a positive association between LIBOR and MKBK. Macroeconomic conditions: BCYCLE = Average yield on Moody s AAA bonds minus the yield on 30-year U.S. Treasury bill on the day the loan was issued. This variable captures the time series variation in risk premium over the business cycle and is expected to be positively associated with LIBOR. 3. Sample Selection and Description The sample selection process was driven by the availability of loan spread and disclosure data. Table 1 summarizes the process. We obtained data on the total disclosure score for 11

13 companies from the annual volumes of the FAF Reports. All scores were converted to a percentage of total available points for that industry, to make the scores comparable across industries. We then eliminated financial institutions since their financing needs and economic health are affected by factors quite different from industrial firms. This yielded a potential sample of 2,083 firm-year observations covering 585 different firms. We then averaged the scores of each firm over years t, t-1 and t-2 to obtain the disclosure metric (DISC). Firms that did not have three consecutive years information dropped out, resulting in an initial disclosure sample of 989 firm-years (341 different firms). The DISC t metric for the sample selected above was then matched with the first loan of period t+1 collected from the Dealscan database. Firms that did not have a loan issue in period t+1 were deleted, resulting in a sample of 182 firm-year observations and 129 firms. Finally observations with insufficient COMPUSTAT or CRSP data were deleted resulting in a final sample of 141 firm-year observations and 102 firms. Although an analysis could be performed with the full sample of 141 observations, it is clear that multiple observations of the same firm will not be independent of one another, thereby overstating the t-statistics used for regression inferences. The aim of this paper is to explain the cross-sectional variation in the loan spreads so following Sengupta (1998), only one observation per firm (the latest year s observation) is retained. [Insert Table 1] 4. Results Descriptive Statistics and Correlation Analysis Summary statistics of the key variables are provided in panel B of table 2. The table shows that the median disclosure score is There is considerable variation in the disclosure scores 12

14 across the sample with the range of variation of 55.9 and standard deviation of The median size of the firm (total market value of common equity at 1983 s prices) is about $1.9 billion, with large variation in size across the sample. SPREAD also seems to vary quite a bit across firms ranging from 12.5 to 300 basis points. The median loan size is $300 million and the median maturity on the loans is about 3 years. Comparing the characteristics of our private loan sample with Sengupta (1998) s public debt sample, it seems that public debt issues are smaller in size (median size of about $198 million) and longer in maturity (median maturity of 10 years). Firms that issue public debt also tend to be larger in size; mean total assets were about $10 billion in Sengupta (1998) as compared to $5.3 billion in our sample. [Insert Table 2] The data also revealed variation in the disclosure scores across industries. This variation could be attributable to legitimate differences in disclosure practices across industries or differences in the ratings process across industries or both. Lang and Lundholm (1993, 1996) dealt with this issue by industry-adjusting all variables by subtracting out the industry means. However, this method also eliminates any differences in disclosure efforts across industries. Therefore, we follow Sengupta (1998) and present our main analysis using raw (unadjusted) variables. We also performed tests using industry-adjusted variables and these are discussed below. Table 3 presents the simple (Pearson) correlations between variables. The table shows that DISC and SPREAD are negatively correlated as expected, with a correlation coefficient of about 0.3 (statistically significant at the 0.01 level). [Insert Table 3] 13

15 Effect of Disclosure Quality on the Bank Loan Pricing Table 4 presents the results of regressions of SPREAD on DISC and other determinants of SPREAD. The results are reported for two models that differ in the choice of control variables. In model 1 (column 3) firm characteristics are summarized by ALTMAN which represents the Altman Z score. In Model 2 (column 4) we use three financial variables: leverage ratio (LEVER), profit margin (MARGIN), and the market-to-book ratio (MKBK) instead of ALTMAN. The Breusch-Pagan χ 2 for both regressions were statistically significant at the 0.01 level indicating that heteroscedasticity could be a problem in these regressions. Hence, the reported t-statistics are based on White s (1980) heteroscedasticity-corrected covariance matrix. The results show that DISC is negatively associated with SPREAD and the coefficient is statistically significant at the 0.01 level under both models. This is consistent with the argument that financial institutions examine corporate disclosure policy in estimating the risk premium to charge a lender. The magnitude of the coefficient for DISC in Model 2 indicates that if a firm s disclosure score was higher by one point then it would, on average, enjoy a 0.92 basis point lower loan spread. This implies that on average, other things remaining constant, the firm with the highest disclosure score in the sample (95.73) enjoys a borrowing rate that is approximately 51 basis points (0.51 percentage point) lower than the firm with the lowest disclosure score (39.83) in the sample. For the high disclosure firm, this translates into annual interest savings of about $2.04 million on its $400 million loan. The adjusted R 2 for both regressions was 0.62 compared to R 2 s of 0.13 to 0.42 reported in prior studies on the determinants of loan spread such as Berger and Udell (1990, 1995), Booth (1992), and Chen et. al. (2000, 1996). Among the control variables, SECURE, BID, ALTMAN and LEVER have their expected signs and are statistically significant at the 0.01 level. 14

16 RESTRUC, MVAL and STDRETN also have their expected signs although the coefficients are not statistically significant. The coefficients for LSIZE, LMATUR, MKBK and BCYCLE turn out to be negative contrary to expectations, although neither coefficient is statistically significant. Similarly, the coefficient for MARGIN is positive, contrary to expectations, although not statistically significant. [Insert Table 4] Additional Tests and Sensitivity Analysis We checked our results for multicollinearity using procedures suggested by Belsley, et. al. (1980). The results suggested that multicollinearity could be a problem in the SPREAD regressions since the condition number for the regressions were in the range of The correlations reported in table 4 showed that the variables LSIZE and MVAL have a correlation of about 0.49, while STDRETN had a correlation of 0.38 with LSIZE, 0.49 with SECURE and with MVAL. To deal with this problem, we ran regressions after dropping some of these variables but this did not affect the qualitative conclusions of the effect of disclosure quality on loan spread. We also ran regressions after adding other proxies for leverage (such as times interest earned ratio) and profit margin and they all resulted in the same conclusions. Regression results reported in table 4 were tested for the presence of influential observations using procedures suggested by Belsey et al. (1980). These procedures identified 10 influential observations. To control for the effects of potential influential observations on the regression results, two procedures were performed: (i) regressions were rerun after dropping all potentially influential observations, and (ii) regressions were rerun using Welsch s (1980) method of bounded influence estimation that runs a weighted least squares regression after 15

17 assigning lower weights to the influential observations. The conclusions of the paper remained qualitatively unaltered under both procedures. 12 We performed additional tests using industry-adjusted variables (as in Lang and Lundholm 1993, 1996). 13 These results (not reported) are consistent with those reported in table 4. The coefficient for DISC was statistically significant at 0.05 level for both models. The adjusted R 2 for both regressions was 0.52 suggesting an overall weaker fit as compared to regressions using unadjusted variables. Do private lenders rely less on disclosures when the borrower has public debt outstanding? We ran additional tests to examine if the relationship between corporate disclosures and loan spread is affected by the presence or absence of public debt. If a borrower has publicly traded debt outstanding, the yields and ratings on its existing debt will provide the private lender with an independent (market) assessment of the borrower s default risk. 14 The lender could rely on this default risk information, and may be less keen on using other publicly available information to generate its own assessment. This would suggest that the importance of corporate disclosure quality would be reduced when the borrower has publicly available debt outstanding. We tested this conjecture in two ways. In the first method, we divided the sample into two groups according to whether the borrower had public debt outstanding or not and ran separate regressions for each. In the second method, we created a dummy variable, PUBDEBT which equals 1 if the firm has outstanding public debt, 0 otherwise. We next constructed a crossdummy, DISC* PUBDEBT, which measures the difference in the effect of disclosures on loan spread between the public-debt and no-public-debt sub-samples. Since the negative 12 The main results are presented including all observations since almost 10 percent of the sample was classified as influential and a careful examination of these observations did not indicate any abnormality in them. 16

18 association between loan spread and disclosure is expected to be weaker if the firm has outstanding public debt, we expect the coefficient for DISC* PUBDEBT to be positive. The results of the tests are summarized in Table 5. Model 1 (column 3) shows the results with the dummy variable DISC* PUBDEBT. The variable is positive and statistically significant at the 0.05 level indicating disclosures have a stronger impact on loan spread for the firms with no public debt outstanding. Models 2 and 3 (columns 4 and 5) show the results of separate regressions for the two sub-samples. In all models the coefficient for DISC is negative, and statistically significant at conventional levels, indicating that corporate disclosures are important irrespective of whether the firms have public debt outstanding or not. Furthermore, in Model 2 (sample of firms with no public debt) the coefficient for DISC is while it is in Model 3 (sample of firms with public debt) suggesting that disclosures are more important in the absence of public debt. This is confirmed by the statistical significance of the coefficient for DISC* PUBDEBT in model Conclusion This study provides evidence of a statistically significant negative association between financial analysts measure of a firm s overall disclosure quality and the interest rate on private debt contracts. This result is especially interesting when considered in light of modern banking theory, which postulates that banks possess superior monitoring skills and thus act as delegated monitors on behalf of their depositors. We demonstrate that despite such monitoring advantage banks nevertheless rely on direct disclosures by the borrower itself in pricing its loan. This suggests that a bank s monitoring does not eradicate all the information asymmetry that may 13 This procedure involved subtracting out the mean for a particular variable for a particular industry from each variable, except for the binary variables. 17

19 exist between itself and its borrower. Therefore, the bank evaluates a firm s disclosure quality and incorporate this into its default risk estimate and loan spread. Firms that are rated favorably on the basis of their disclosure quality enjoy a lower interest on their private loans. This result complements earlier empirical research that finds banks rely on cross-monitoring by others such as ratings agencies to assess a borrower s credit-worthiness. We find that a bank s reliance on a borrower s disclosure quality is strongest in the absence of other sources of public information, such as bond ratings. Finally, this paper contributes to the literature on the value of disclosure. Earlier studies had found a negative relationship between disclosure quality and the cost of public external capital (public debt and equity). We find that such a negative relationship also prevails in the case of private loans, the largest source of corporate financing. Further work needs to be done to evaluate the cost and benefits of disclosure to enable managers to arrive at the optimal disclosure policy. 14 See Booth (1992). 18

20 References Altman, E. and T. McGough Evaluation of a company as a going concern. Journal of Accountancy 138 (6): Bajaj, M, S. Mazumdar and A.Sarin, Settlements in Securities Class Actions: An Empirical Analysis, Working Paper, Santa Clara University. Belsley, D., E. Kuh and R. Welsch Regression Diagnostics: Identifying Influential Data and Sources of Collinearity. New York, NY: Wiley. Berger, A. and G. Udell Collateral, loan quality, and bank risk. Journal of Monetary Economics 25 (January): Relationship lending and lines of credit in small firm finance. Journal of Business 68 (July): Best, R. and H. Zhang, Alternative information sources and the information content of bank loans. Journal of Finance 48, Billett, M., M. Flannery and J. Garfinkel, The effect of lender identity on a borrowing firm's equity return. Journal of Finance 50(2), Black, F Bank funds management in an efficient market. Journal of Financial Economics 2, Booth, J Contract costs, bank loans, and the cross-monitoring hypothesis. Journal of Financial Economics 31 (month): Botosan, C Disclosure level on the cost of equity capital. The Accounting Review 72 (July):

21 Carey, M., M. Post and S. Sharpe Does corporate lending by banks and finance companies differ? Evidence on specialization in private debt contracting. Journal of Finance 53 (June): Chen, A., S. Mazumdar and Y. Yan Monitoring and bank loan pricing. Pacific Basin Finance Journal 8, Chen, A., S. Mazumdar and M. W. Hung Regulations, lender identity and bank loan pricing. Pacific Basin Finance Journal 4: Darrough, M. N. and N. M. Stoughton Financial disclosure policy in an entry game. Journal of Accounting and Economics 12: Diamond, D Financial intermediation and delegated monitoring. Review of Economic Studies 51: Fama, E What s different about banks? Journal of Monetary Economics 15: Feltham, G. A., and J.Z. Xie Voluntary financial disclosure in an entry game with continua of types. Contemporary Accounting Research 9 (Fall): Financial Analysts Federation Report of the Financial Analysts Federation Corporate Information Committee. New York, NY: FAF. Healy, P, K. Palepu, and A. Hutton Do firms benefit from expanded voluntary disclosure? Harvard University Working Paper. Houston, J. and C. James Bank information monopolies and the mix of private and public debt claims. Journal of Finance 51 (December): James, C Some evidence of the uniqueness of bank loans. Journal of Financial Economics 19:

22 Johnson, S An empirical analysis of the determinants of corporate debt ownership structure. Journal of Financial and Quantitative Analysis 32 (March): Lang, M. and R. Lundholm Cross-sectional determinants of analyst ratings of corporate disclosure. Journal of Accounting Research 31 (Autumn): Corporate disclosure policy and analyst behavior. The Accounting Review 71 (October): Leftwich, R Accounting information in private markets: Evidence from private lending agreements. The Accounting Review 58 (January): Lummer, S., and J. McConnell, 1989, Further Evidence on the Bank Lending Process and the Capital Market Response to Bank Loan Agreements, Journal of Financial Economics 25, Merton, R. C., On the pricing of corporate debt: the risk structure of interest rates. Journal of Finance 29, Mikkelson, W., and M. Partch, Valuation effects of securities offerings and the issuance process. Journal of Financial Economics 15, Petersen, M. and R. Rajan The benefits of lending relationships: Evidence from small business data. Journal of Finance 49 (March): Preece, D. and D. Mullineaux, Monitoring by financial intermediaries: banks vs. nonbanks. Journal of Financial Services Research 4, Scott, J. and T. Smith The effect of the bankruptcy reform act of 1978 on small business loan pricing. Journal of Monetary Economics 16 (month): Sengupta, P Corporate disclosure quality and the cost of debt. The Accounting Review 73 (October):

23 Slovin, M., S. Johnson and J. Glascock, Firm size and the information content of bank loan announcements. Journal of Banking and Finance 16, Thakor, A., Capital requirements, monetary policy, and aggregate bank lending: theory and empirical evidence, Journal of Finance 51, Verrecchia, R. E Discretionary disclosure. Journal of Accounting and Economics 5: Wagenhofer, A Voluntary disclosure with a strategic opponent. Journal of Accounting and Economics 12 (March): Welsch, R Regression sensitivity analysis and bounded influence estimation. In Evaluation of Econometric Models, edited by J. Kmenta and J. Ramsey, Academic Press: New York. White, H A heteroscedasticity-consistent covariance matrix estimator and a direct test for heteroscedasticity. Econometrica. 48:

24 TABLE 1 Summary of Sample Selection Filters Sample Selection Filters: Number of firms Firms with total scores Less: Financial Institutions (72) Firms which did not have three consecutive years (244) scores Initial Disclosure sample 341 Less: Firms lacking loan pricing information (212) Firms lacking COMPUSTAT or CRSP information (27) Final sample Total disclosure scores were taken from the Report of the Financial Analysts Federation Corporate Information Committee for the years

25 TABLE 2 Variable Definitions and Descriptive Statistics Panel A: Variable Definitions SPREAD quoted spread in basis points to the LIBOR rate on the first loan of year t+1. DISC = average of total disclosure score for the years t, t-1 and t-2. LSIZE = natural log of the loan size (in $ millions). LMATUR = natural log of the loan's original days-to-maturity. SECURE = binary variable which equals 1 if the loan is collateralized and 0 otherwise. BID = binary variable which equals 1 if the loan contains the option to price in terms of a different index and 0 otherwise. RESTRUC = binary variable which equals 1 if the loan is for corporate restructuring (i.e. leveraged buyout, takeover acquisition, recapitalization, debt consolidation), 0 otherwise. SENIOR = binary variable which equals 1 if the borrowing firm has public debt which is senior to the bank loan, 0 otherwise. MVAL = market value of common equity (in $ millions) deflated by the aggregate consumer price index, at the end of year t. STDRETN = the standard deviation of daily stock returns over years t and t-1. ALTMAN = 1.2(X1) + 1.4(X2) + 3.3(X3) + 0.6(X4) + X5, calculated using end of year t information, where, X1 = (current assets current liabilities) / total assets, X2 = retained earnings / total assets, X3 = earnings before interest and taxes / total assets, X4 = market value of equity / book value of total liabilities, and, X5 = sales / total assets. MVAL = the market value of common equity deflated by the aggregate consumer price index, at the end of year t. MKBK = market value of equity divided by the book value of equity both at the end of year t. MARGIN = income before extraordinary items divided by the total assets, both at the end of year t. BCYCLE = average yield on Moody s AAA bonds for the month of issue minus the average yield on 30 year treasury bills for the month of issue. DISC*PUBDEBT = DISC if the firm has public debt outstanding, 0 otherwise. Panel B: Summary Statistics of Continuous Variables Mean Standard median minimum 25% 75% maximum Variables deviation SPREAD DISC SIZE MATUR MVAL STDRETN ALTMAN MKBK LEVER MARGIN BCYCLE Notes: Sample statistics are based on a sample of 102 observations. 1 SIZE and MATUR information is provided only for sample characteristics. In the regressions, log of these values (LSIZE and LMATUR, respectively) are used. 24

26 TABLE 3 Simple Pearson Correlations Among Variables of the Spread Regression SPREAD DISC LSIZE LMATUR SECURE BID RESTRUC SENIOR ALTMAN MVAL STDRETN DISC (0.003) LSIZE (0.001) (0.595) LMATUR (0.111) (0.239) (0.433) SECURE (0.001) (0.587) (0.001) (0.107) BIDD (0.005) (0.814) (0.006) (0.331) (0.058) RESTRUC (0.001) (0.408) (0.522) (0.099) (0.001) (0.052) SENIOR (0.284) (0.198) (0.916) (0.096) (0.270) (0.069) (0.027) ALTMAN (0.013) (0.003) (0.282) (0.273) (0.655) (0.742) (0.230) (0.891) MVAL (0.001) (0.007) (0.001) (0.124) (0.217) (0.768) (0.008) (0.063) (0.005) STDRETN (0.001) (0.106) (0.001) (0.105) (0.001) (0.712) (0.004) (0.603) (0.755) (0.001) MKBK (0.076) (0.395) (0.546) (0.164) (0.015) (0.495) (0.156) (0.759) (0.777) (0.844) (0.173) Notes: Variables are defined in panel A of table 2. Correlations are based on a sample of 102 observations. p-values for two-tailed tests are given in parentheses. 25

27 TABLE 4 Regression results of the effects of corporate disclosures on loan spread Variable Prediction Coefficient (t-statistic) Model 1: Results based on a composite measure of default risk (ALTMAN Z score) Model 2: Results based on separate variables to proxy for default risk INTERCEPT (α 0 )? (1.170) (0.802) Disclosure Index DISC (α 1 ) (-2.980)** (-3.686)** Loan characteristics LSIZE (α 2 ) (-0.484) (-0.160) LMATUR (α 3 ) (-0.053) (-0.248) SECURE (α 4 ) (3.446)** (3.395)** BID (α 5 ) (3.279)** (3.644)** RESTRUC (α 6 ) (1.521) (1.349) SENIOR (α 7 ) (0.140) (-0.129) Firm characteristics MVAL (α 8 ) (-1.205) (-1.606) STDRETN (α 9 ) (1.542) (1.334) ALTMAN (α 10 ) (-2.478)** LEVER (α 11 ) (2.831)** MARGIN (α 12 ) (1.599) MKBK (α 13 ) (-1.236) Market Conditions BCYCLE ((α 14 ) (-0.854) (-0.748) Adjusted R Breusch Pagan Chi-square Notes: Variables are defined in panel B of table 2. Regression results are based on 102 observations. The t-values given in parentheses below each coefficient estimate are calculated using White s (1980) heteroscedasticity consistent covariance matrix. * significant at 0.05 level based on a one tailed test. ** significant at 0.01 level based on a one tailed test. 26

28 TABLE 5 Presence of Public Debt and the Effects of Disclosure Quality on Loan spread. Variable Prediction Model 1 (Full sample with public debt versus no public debt dummy) Coefficient (t-statistic) Model 2: Firms with no Public debt outstanding Model 3: Firms with public debt outstanding INTERCEPT (α 0 )? (1.158) (0.339) (1.394) Disclosure index DISC (α 1 ) (-2.965)** (-1.981)* (-2.135)* Public debt Vs. No public debt dummy DISC*PUBDEBT (1.791)* Loan Characteristics LSIZE (α 2 ) (-0.611) (0.019) (-1.025) LMATUR (α 3 ) (0.236) (0.053) (-0.265) SECURE (α 4 ) (3.399)** (4.111)** (2.870)** BID (α 5 ) (3.499)** (1.453) (2.752)** RESTRUC (α 6 ) (1.752)* (1.560) (1.495) SENIOR (α 7 ) (-0.088) (-1.622) (1.077) Firm Characteristics ALTMAN (α 8 ) (-1.846)* (-0.164) (-1.936)* MVAL (α 9 ) (-1.380) (-2.158)* -001 (-0.849) STDRETN (α 10 ) (1.742) (0.300) (1.343) Market Conditions BCYCLE (-0.149) (0.764) (-0.947) Number of observations Adjusted R Breusch Pagan χ Notes: Variables are defined in panel B of table 2. The t-values given in parentheses below each coefficient estimate are calculated using White s (1980) heteroscedasticity consistent covariance matrix. * significant at 0.05 level based on a one tailed test. ** significant at 0.01 level based on a one tailed test. 27

29 28