The information role of audit opinions in debt contracting

Transcription

1 The information role of audit opinions in debt contracting Peter F. Chen School of Business & Management Hong Kong University of Science & Technology Shaohua He Department of Accounting & Finance Lancaster University Zhiming Ma Guanghua School of Management Peking University Derrald Stice* School of Business & Management Hong Kong University of Science & Technology Draft: July, 2014 JEL Classification: G01, M4, M49 Keywords: Debt Contracting, Audit Opinions, Going Concern Opinion, Explanatory Language We thank Valerie Li, Michael Minnis, Tomomi Takada, and Hansang Yi for helpful comments and suggestions. We appreciate comments and suggestions from participants at the 2013 European Accounting Association Conference in Paris, France; 2013 Korean Accounting Association Conference in Gyeongju, South Korea; 2013 AAA Annual Meeting and from workshop participants at Singapore Management University. *Address for correspondence: Department of Accounting, School of Business and Management, Hong Kong University of Science & Technology, Clear Water Bay, Kowloon, Hong Kong. Phone:

2 ABSTRACT This study examines the effect of audit opinions on private debt contracting. We use the auditor s explanatory language to partition modified audit opinions into Inconsistency, caused by an accounting change or restatement, and Inadequacy, arising from material uncertainty or a going concern opinion. Using the loan contracts of firms with modified audit opinions during the period ; we find that, compared with loans initiated after clean opinions, loans initiated after modified opinions are associated with higher interest spreads (17 basis points on average), fewer financial covenants, more general covenants, smaller loan sizes, and a higher likelihood of being secured. We find that the effect on loan spreads (as well as other non-price terms) varies by the type of modified opinion - ranging from no effect for an accounting change to an increase of 107 basis points for going concern opinions. Additional analyses of GC opinions using propensity score matching show that auditor opinions are still associated with price and non-price contract terms. Overall, our empirical results suggest that audit reporting communicates auditors private information to lenders. 1

3 1. Introduction The audit opinion communicates to financial report users an auditor s degree of assurance that the financial statements faithfully reflect a client s underlying economic activities. Under the Security and Exchange Commission (SEC) requirement and the Generally Accepted Auditing Standards (GAAS), the auditor s reporting options are either an unqualified opinion or an unqualified opinion with explanatory language commonly called a modified audit opinion (MAO). 1 This discretionary explanatory language is the only difference between a standard clean opinion and an MAO. 2 In fact, it is the only practical channel for the auditor to communicate private information discovered during the auditing process to outsiders. 3 With the exception of going concern (GC) opinions, there is limited evidence on whether and how audit opinions are informative to financial statement users. In this study, we examine the information role of MAOs in private debt contracting by incorporating the auditor s explanations for modification. To investigate the effect of audit opinions on debt contracting, we manually classify MAOs in to categories relating to accounting changes, restatements, material uncertainty, and going concern opinions. We predict that these four types of MAOs will have different implications for evaluating a borrower s credit risk and for designing appropriate monitoring mechanisms. Generally, audit opinions may be useful to lenders for at least two reasons. First, audit opinions may inform lenders about the reliability of financial reports. Accounting 1 Per Rule 2-02 of Regulation S-X, the SEC will not accept financial statements with an audit opinion that is not unqualified. Throughout the paper we use the term modified to denote unqualified opinions with explanatory language. Under SAS 58 (effective for reports issued after January 1, 1989) certain opinions previously classified as qualified (such as changes from one GAAP method to another and material uncertainty) are now classified as unqualified with explanatory language. Consistent with prior research conducted during our sample period, our MAOs are almost exclusively unqualified with explanatory language (e.g., Butler et al., 2004). 2 The auditor s discretion is based on the professional standard that Certain circumstances, while not affecting the auditor s unqualified opinion, may require that the auditor add an explanatory paragraph (or other explanatory language) to the audit report (AU Section ). 3 One exception is the auditor change 8-K. However, prior research has found that, in addition to being infrequent, these 8-Ks likely underreport disagreements between auditors and clients (Smith and Nichols, 1982; DeFond and Jiambalvo, 1993). 1

4 information is useful in evaluating a borrower s liquidity and asset values and in predicting future cash flow and the likelihood of bankruptcy. Additionally, accounting numbers can be used in contracting with relative ease and low cost. Unreliable financial information increases uncertainty about borrowers and the costs of contracting. Second, through the audit report, an auditor reveals to financial statement users private information about a client s risks and future expenses of interest to investors (ex. future costs related to ongoing litigation). In particular, a GC opinion may signal an auditor s judgment that a client is closer to financial distress or bankruptcy than what is conveyed elsewhere in the financial report. For these reasons, we predict that loans initiated after MAOs will have higher loan spreads, compared to loans initiated after clean opinions, and that this effect on spreads will be higher following GC opinions than after the other three MAO types. The use of covenants to monitor borrowers performance is an essential component of debt contracting (Jensen and Meckling, 1976; Smith and Warner, 1979; Watts and Zimmerman, 1986). A necessary condition for financial covenants to be effective is that these financial statement numbers faithfully reflect a borrower s economic activities. If an MAO signals a borrower s lower financial reporting quality, lenders may prefer to use general covenants that do not use accounting numbers as inputs. Specifically, we predict that lenders will choose to rely less on accounting numbers and more on non-accounting monitoring mechanisms - resulting in a decrease in financial covenants and an increase in general covenants - after MAOs. In addition, we expect the effect on covenants to be stronger for GC opinions than for other MAO types. Audit opinions can also be informative to lenders in negotiating other loan terms such as loan size, collateral requirements, and loan maturity. For example, if a GC opinion signals higher than expected borrower default risk, then lenders will be more likely to require 2

5 collateral because covenants are unlikely to provide effective lender protection. In the same vein, we further predict that loans initiated after MAOs are likely to be smaller in size and shorter in maturity length. To examine the effects of audit opinions on debt contracting, we collect a sample of firms with MAOs that have at least one loan issued during an MAO period and at least one loan issued during a non-mao period during the 18-year period from 1992 to For the 2,056 modified opinions in the final sample, we read through the explanatory language contained in the 10-K filings in order to classify the modified opinions. We classify MAOs in to two general categories that are further partitioned into four types: Inconsistency caused by an Accounting Change or a Restatement; and Inadequacy related to Material Uncertainty or a GC Opinion. MAOs labeled Inconsistency alert financial statement users about the potential incomparability, perhaps even quality, of data contained in the financial statements arising from restatements and accounting changes. Inadequacy MAOs express auditors more serious concern about the validity of the financial statements as a whole. 4 Material Uncertainty concerns are related to the resolution of future economically relevant unknowns (ex. contingent liabilities, litigation risk, and business uncertainty), and going concern opinions indicate that a key assumption of the accounting model is violated (i.e. that the firm will continue as a going concern for a period of at least one year). Relative to the standard clean opinion, these four types of MAOs, convey differential degrees of negativity about a client s financial reporting quality and risk and increase in severity as follows: Accounting Change, Restatement, Material Uncertainty and GC Opinion. As discussed, we predict the greatest effect on loan terms to come from GC opinions. 4 According to AU Section 508, the auditor may add an explanatory paragraph (or other explanatory language) in certain circumstances including: lack of consistency caused by a change of accounting principles or misstatements, substantial doubt about the entity s ability to continue as a going concern, and emphasizing a matter regarding the adequacy of the financial statements to reflect significant subsequent transactions or events. 3

6 Consistent with our predictions, we find that loans initiated after MAOs have higher loan spreads (on average, 17 basis points higher) compared to loans initiated after clean opinions, controlling for the other determinants of the interest rate. In addition, the effects of MAOs on loan spreads vary significantly depending on the type of modified opinion. The loan spread effect (in basis points) increases from 0 after an Accounting Change MAO to 25, 49, and 107 after Restatement, Material Uncertainty, and GC Opinion MAOs, respectively. We also find that lenders decrease the use of financial covenants and increase the use of general covenants in loans initiated after an MAO. Specifically, lenders decrease the use of financial covenants by 3.8% and increase the use of general covenants by 4.2%, on average, after an MAO. Again, the effect of an MAO depends on the type of modification. In general, we find that Inconsistency MAOs have a smaller, but still significant, effect on the use of financial and general covenants than do Inadequacy MOAs. As expected, GC Opinion has a large effect on the use of covenants - the number of financial covenants decreases by 9.1% and the number of general covenants increases by 12.3% in loans issued after GC opinions. Material Uncertainty opinions are associated with the largest effect on the use of general covenants (an increase of 18.2%), but they are not associated with the use of financial covenants. We perform additional tests on other loan terms beyond covenants - we find that lenders reduce loan sizes and increase the likelihood of requiring collateral following an MAO. The results on loan maturity are mixed and indicate a slight decrease (increase) in loan maturities after GC Opinion (Material Uncertainty). In addition, as predicted, the effect of an MAO on each of the loan terms is largest for GC opinions. Overall, these results are consistent with differential non-price costs for firms receiving MAOs. To further establish the unique value of the audit report, we perform separate analyses on the different types of GC opinions, because these opinions in particular signal more of the 4

7 auditor s private information likely to be relevant to lenders. Following Menon and Williams (2010), we classify GC opinions as being related to firm performance, financing concerns, or other issues and investigate the effects of each on the loan contract terms previously investigated. We find that each of these categories of GC opinions influences the composition of loan terms, reinforcing the value of the private information provided by the auditor in the audit opinion. A key inference drawn in this study is that auditors communicate private information to lenders through the audit opinion. To mitigate concerns that our results are driven by financial stress of the borrower we employ a variety of control variables in each of our tests. However, managers of firms preparing to secure financing are likely to be in contact with potential lenders preceding a loan issuance, and we are unable to directly control for information that managers privately provide to lenders. To alleviate the concern this raises, we construct a new indicator variable Before_MAO that takes a value of 1 if a loan is issued in the 12 months preceding an MAO. If the private information conveyed by the auditor is actually preempted by managers communicating to lenders before an MAO, then we would expect a significant coefficient on this new variable. However, Before_MAO is insignificant in tests of each of the contract terms we investigate. This result alleviates the concern that our results are driven by unobserved information leakage instead of by the private information communicated by auditors. We also match our firms with GC opinions with out-of-sample firms based on the determinants of receiving a GC opinion and bankruptcy using variables from prior studies (Zmijewski, 1984; DeFond et al., 2002). If GC opinions simply capture the probability of financial distress or bankruptcy, as predicted by these mechanical models, they should have no incremental effect on loan contract terms in the matched-sample specifications. However, we find that GC opinions continue to have a significant effect on all of the loans terms 5

8 examined except financial covenants. Overall, these results suggest that the information value of GC opinions to lenders likely arises from the auditor s private information about the client. Our study makes several contributions to the existing literature. First, this is the first study to document the information role of audit opinions in debt contracting by incorporating the auditor s explanatory language. The empirical results provide evidence that modified audit opinions along with the explanatory language of the audit report are informative to lenders in negotiating loan agreements. These results complement the finding in prior studies that the voluntary use of auditing or the employment of Big 5 auditors is associated with a lower cost of debt (Fortin and Pittman, 2007; Lennox and Pittman, 2011; Minnis, 2011). In addition, our empirical results have implications for the current regulatory initiative to expand the scope of audit opinions to include a discussion of critical audit matters (PCAOB, 2013) and other items to enhance the value of auditing (Carson et al., 2013; DeFond and Zhang, 2013). One implication of our study is that regulators need to weigh the informative role of additional disclosures in audit reports against the potential reluctance (cost) of auditors to reveal private information about clients, given the economic consequences of MAOs documented here. Second, our empirical evidence on the role of private information conveyed by GC opinions contributes to the vast auditing literature on GC opinions. Prior studies report evidence that GC opinions are useful in predicting subsequent bankruptcies and are associated with negative stock market reactions (Hopwood et al., 1989; Raghunandan and Rama, 1995; Chen and Church, 1996; Menon and Williams, 2010; Kaplan and Williams, 2013). Our empirical results on the effect of GC opinions on loan terms shed light on the unique value of auditors private information that is not only unavailable elsewhere in the financial statements but is also inaccessible through other channels of sophisticated lenders (Mutchler, 1985; Menon and Schwartz, 1987). 6

9 Last, our study is related to the literature on the role of financial reporting quality in debt contracting. Bharath et al. (2008) provide evidence that a borrower s access to the private versus public debt market and its loan terms depend on the quality of the borrower s accounting information, and Costello and Wittenberg-Moerman (2011) document that lenders trade-off between different monitoring mechanisms when internal control weakness information is disclosed. However, these studies are silent about the specific role of the audit function within financial reporting. Our study provides empirical evidence on how audit opinions are incrementally informative to lenders in the private debt market. In the next section we develop our hypotheses. We describe the sample selection procedures and variables used in this study in Section 3. Section 4 presents the empirical results of our hypotheses and Section 5 presents the results of additional analyses. A summary and conclusions are provided in Section Background and Hypothesis Development As capital providers, lenders are interested in protecting the timely repayment of the loan and interest that are claims on the borrower s future cash flow and assets. When contemplating a loan facility, banks and other lenders perform a credit analysis of the borrower on several dimensions. They analyze the risk of default, estimate the market value and liquidation values of assets, and evaluate the character and ability of the management (Tirole, 2007). If lenders decide to initiate a loan after the credit analysis, they negotiate the price and non-price terms of the debt contract that compensate them for risk and allow them to monitor the borrower s performance over the life of the loan. Audit opinions are potentially informative to lenders in evaluating the borrower s default risk and negotiating loan terms that lead to more efficient monitoring. 7

10 Financial statements provide important information to lenders in evaluating the borrower s credit worthiness and default risk. Lenders use financial statements to analyze the financial data and to estimate the value of a firm s assets. Financial ratios are useful in predicting the likelihood of borrower default (e.g., Beaver, 1966; Altman, 1968; Ohlson, 1980). Research has shown that accounting measures can predict the losses that will be sustained by lenders in the event of borrower default (Varma and Cantor, 2005; Acharya et al., 2007), and recent work has demonstrated that accounting numbers also possess significant ability to predict future loss given default at the debt issuance date (Amiram, 2012). An independent audit is a critical component of financial reporting (Watts and Zimmerman, 1983). Through the audit opinion, the auditor communicates to users its degree of assurance that the financial statements faithfully reflect the firm s underlying economic activities. Under the current professional standards, the auditor s degree of assurance is communicated through the inclusion of explanatory language within an unqualified report. Lenders are likely to find auditors disclosures useful for at least two reasons. First, auditors disclosures inform lenders about the quality of financial data by revealing an auditor s judgment about qualitative aspects of the financial reports such as potential bias and assumptions that may not faithfully reflect the firm s economic activities. 5 Czerney et al. (2013) provide evidence that explanatory language mentioning accounting changes and restatements is associated with a higher probability of subsequent misstatements. Graham et al. (2008) find that earnings restatements are associated with increases in price and 5 This is consistent with the notion that auditors are responsible for assuring a level of financial reporting quality that is more than a mechanical compliance with accounting standards. ASN No.14 requires that auditors evaluate the qualitative aspects of a company s accounting practices, including potential biases in management s judgment. Fair representation, in accordance with GAAP, requires the use of professional judgment in making estimates and assumptions that reflect the firm s underlying economic activities. 8

11 non-price costs of debt, consistent with financial restatements leading to higher information uncertainty. 6 Second, auditors can reveal private information about the borrower s risks and financial health directly in their audit opinions. For lenders, Inadequacy opinions are particularly relevant because they convey an auditor s judgment about default risks that are not reflected elsewhere in the financial statements these risks may pertain to future events (ex. contingent liabilities). Therefore, the effect of MAOs related to Inadequacy opinions should be stronger than the effect of Inconsistency opinions because they signal greater default risk of the borrower. Also, prior studies find that GC opinions communicate auditors private information about higher default risk beyond information publically available and are useful in predicting bankruptcy (Mutchler, 1985; Menon and Schwartz, 1987; Mutchler et al., 1987; Hopwood et al., 1989; Raghunandan and Rama, 1995). These findings are consistent with the effectiveness of SAS 59 which provides guidance to auditors in evaluating a client s ability to continue as a going concern. We predict that the effect of GC opinions on loan spreads will be greater than the other three opinion types. Given the potential economic consequence of MAOs, auditors are likely to experience pressure from clients to issue a clean opinion. 7 On the other hand, considering auditors legal liability, auditors also have incentives to be more conservative in issuing MAOs as a way of protecting themselves from potential litigation risk. GC opinions may be most affected by this conservative bias because auditors are more likely to be sued when a clean opinion is issued before bankruptcy. If auditors issuance of MAOs is conservative, it should bias 6 The sample of restatements in Graham et al. (2008) is based on restatements announced in the filed financial reports. These restatements are not always mentioned in the explanatory language in the audit report. The restatements in our sample are exclusively restatements mentioned in the explanatory paragraph of the audit report. 7 This is consistent with the use of modified opinions (in particular GC opinions) as a measure of audit quality in prior studies - see DeFond and Zhang (2013) for a review of the literature. 9

12 against our finding an effect of MAOs on loan terms as lenders rationally adjustment for the conservative bias of MAOs. As a result, to what extent audit opinions are informative to lenders in negotiating the interest rates of loan contracts is an empirical question. The discussion above leads to our first empirical hypothesis, stated as follows: H1: Compared with loans issued to firms with clean opinions, loans issued to firms with modified audit opinions have higher loan spreads. In addition, the effect on spread is greater for going concern opinions than for other MAO types. The monitoring of a borrower s behavior through debt covenants to mitigate agency conflicts between shareholders and debtholders is an important part of debt contracting (Jensen and Meckling, 1976; Smith and Warner, 1979). In the case of covenant violation, control rights can be quickly transferred to lenders. Accounting information and financial ratios are widely used to monitor a borrower s performance in debt covenants. A precondition for using financial covenants is the assurance that the accounting information used reflects the actual performance of the borrower. On the other hand, if accounting numbers are unreliable or of low quality, lenders may use general covenants that do not rely on accounting information. General covenants often specify events that will require the borrower to pay down the balance of their loan, whether dividends may be paid, or the allowed uses of borrowed funds. If lenders view an MAO as decreasing the value of including financial covenants, then they may compensate by increasing the number of general covenants. Alternatively, if financial and general covenants are independent in purpose, the optimal number (and type) of included general covenants may already be included and no change will be observed. The implications of an MAO for debt covenants vary by type of modified opinion. If Inconsistency opinions are associated with a high probability of subsequent misstatements, then financial covenants may be less effective as a monitoring tool (Czerney et al., 2013). 10

13 However, the effects of Inadequacy opinions on covenant choices should be stronger than the effects of Inconsistency opinions, because Inadequacy MAOs convey an auditor s information about the adequacy of financial statements to reflect a firm s economic activities. Financial statements with a Material Uncertainty MAO may not reflect the potential risks of corporate events or contingent liabilities. GC Opinion MAOs may indicate that a basic assumption of the accounting model (i.e. that the firm will persist as a going concern) is violated, decreasing the usefulness of the financial statements to lenders as an effective monitoring tool. Lenders may need to identify alternative measures for liquidation values of assets. We expect the effect on the use of debt covenants in loan contracts to be strongest for GC opinions. Stated in the alternative form, our second hypothesis is: H2: Compared with loans issued to firms with clean opinions, loans issued to firms with modified audit opinions are associated with a decrease in the number of financial covenants and/or an increase in the number of general covenants contained in debt contracts. In addition, the effect of GC opinions is stronger than those of other MAO types. To assess the total effect on contract design of an MAO, it is important to consider the many different contract components that lenders can choose from (Gigler et al., 2009). Up until this point we have only considered the use of spread and covenants in contract design. In reality, lenders have other options to consider when designing a firm-specific contract. We consider the effects of an MAO on three additional contracting options available to lenders: loan size, the requirement of collateral, and the duration of the loan contract. We view an MAO as a disclosure event that communicates a negative signal about financial reporting quality and information risk, and we predict that lenders will be more likely to reduce loan size, require collateral and shorten the loan maturity after an MAO. 11

14 However, the effects should vary for different types of MAOs. Economic theory on credit rationing (see e.g., Jaffee and Russell, 1976; Stiglitz and Weiss, 1981) shows that loan size is not a linear function of the default risk signaled in a borrower s going concern opinion because credit risk may increase if the loan size can only finance small projects that are more risky. As a result, rather than reducing the size of loan, lenders may simply deny the loan application after an MAO, especially after a GC opinion. Moreover, Diamond s (1991) theory shows that debt maturity is a non-monotonic function of a firm s risk; with low and high risk firms obtaining short-term debt. Finally, we expect GC opinions to have a stronger effect on the likelihood of a loan being secured than the other three types of MAO. Stated in the alternative form, we predict that: H3: Compared with loans issued to firms with clean opinions, loans issued to firms with MAOs are associated with decreases in the size of loans granted, increases in the likelihood that lenders will require collateral, and decreases in the average length of maturity in debt contracts. In addition, the effect of GC opinions is greater than those of other MAO types. 3. Research Design and Sample Selection 3.1 Research Design To examine the information role of audit opinions in debt contracting, we perform empirical tests of the effect of an MAO on the contract terms of loans issued to firms with MAOs by estimating the following model: Loan Term = α + β 1 MAO + β 2 After_MAO + β i (Control i ), (1) where Loan Term is a variable representing the specific contracting terms of the loan agreement that we investigate in each of our tests including: interest spread, the number of financial covenants, the number of general covenants, loan size, whether or not a loan is secured, and the maturity length of a loan. MAO is an indicator variable equal to one if the loan is initiated after the borrower receives an MAO, and zero otherwise. To examine 12

15 whether there is any lingering effect on contract terms for loans initiated beyond the first year after an MAO, we include After_MAO to capture any continuing effect of an MAO after a borrower receives a clean opinion. After_MAO is an indicator variable equal to one if a firm currently has a clean opinion but had an MAO equal to 1 in the previous three years, and zero otherwise. We expect MAO to have a positive (β 1 ) effect on loan spread, use of general covenants, and likelihood of the loan being secured, but a negative (β 1 ) effect on the use of financial covenants, loan size, and maturity length. To investigate the differential effect of different types of MAOs on debt contracts, we replace the generic MAO with our classified MAOs. For the 2,056 modified opinions in the final sample, we read through the explanatory language contained in the 10-K filings in order to classify the modified opinions. Following Butler et al. (2004), we find that most of the modified opinions are related to Inconsistency issues (1,881, or 91.5%) with the rest of the opinions related to a client s Inadequacy (175, 8.5%). Out of these Inconsistency issues, most modifications (1,680) are related to an accounting change (Accounting Change) and the remaining opinions (201) are related to restatements (Restatement) mentioned by the auditor in the audit reports. For Inadequacy MAOs, the majority is related to going concern opinions (GC Opinion), 131 or 74.9% - the rest are related to material uncertainty (Material Uncertainty), 44 or 25.1%. 8 We first examine the incremental effect of MAOs on loan spreads after controlling for the determinants of loan spreads. We control for firm size because small firms have greater information asymmetry and a higher default risk (Forth and Pittman 2004; Bharath et al. 2007). We control for loan size because larger loans are priced at lower interest rates (Booth, 1992; Beatty et al., 2002). We include a number of controls related to financial distress found 8 For those opinions that contain multiple reasons for modification, we classify the modified opinion based on the most severe concern expressed by the auditor in the client s financial report. Our results are robust to allowing overlap across MAO observations and to deleting observations modified for more than one reason. 13

16 in the prior literature: Z-score, market-to-book, leverage, cash flow volatility, tangibility, and credit and term spreads (Graham et al., 2008). We include a measure of abnormal accruals which have been shown to affect debt contracting (Bharath et al., 2008). We include a control for revolvers because these loans have lower loan spread than term loans (Zhang, 2008). Institutional loans, relative to bank loans, have higher loan spread because of the higher information symmetry with the borrower. We control for the existence of performance pricing provisions because a performance pricing provision signals higher adverse selection and moral hazard costs of the borrower (Asquith et al., 2005). We also control for the number of lenders in the loan. A larger number of participants in the loan syndicate is an indication of a higher quality borrower with less information symmetry. Last, we control for other contracting devices available to lenders: loan maturity, collateral, and the number of financial covenants. 9 We include a similar set of control variables in testing the effects of MAOs on the use of financial and general covenants. We replace the number of financial covenants with loan interest rate as a control variable since agency theory on debt covenants predicts a negative relation between loan spread and the use of covenants (Jensen and Meckling, 1976; Smith and Warner, 1979). We select control variables similar to those in prior studies on the determinants of covenants in debt contracting (Beatty et al., 2002; Sufi, 2007; Graham et al., 2008; Costello and Wittenberg-Moerman, 2011). To test hypothesis H3, we examine the effects of MAOs on loan size, likelihood of requiring collateral, and loan maturity. If MAOs signal the borrower s financial reporting quality, lenders may reduce loan size, require collateral, and reduce loan maturity as compensation for the increased cost and reduced efficiency of monitoring after loan initiation. If some MAOs, such as GC opinions, signal an increased default risk of the borrower, the 9 All variables are defined in Appendix A. 14

17 effect of the default risk on the individual loan terms may vary. If GC opinions communicate higher default risk, it is more likely that the loan will be secured to insure the recoverability of the principal. In addition, GC opinions will have larger effects on the decrease in loan size and in loan maturity length than other MAOs. 3.2 Data Sources and Sample Selection We obtain data on MAOs in audit reports from COMPUSTAT (variable name AUOP) for the period from 1992 to We match our MAO sample with public firms in the Dealscan database that contains contractual terms such as interest rate, size, and covenants of loans issued by public firms in the United States. 11 To be included in our sample we require borrowers to obtain a loan during the window just after an MAO (in either the year of MAO or within the three years following the first clean opinion) and outside of this MAO window (either before the MAO or more than three years after the first clean opinion following a MAO). This requirement is to ensure that our results are not driven by a change in the sample composition over time and also to avoid comparing MAO to non-mao borrowers. After eliminating observations with missing data needed in our analyses, our final sample includes 8,473 loans issued to 5,377 borrowers during the period To classify MAOs based on the stated reasons for modification, we read the explanatory paragraph section of the audit report for each MAO identified above from the 10 Audit Analytics provides going concern opinions after we use these data to check our sample of MAOs from COMPUSTAT. We found three cases in which Audit Analytics classified an observation as a going concern but we did not find a going concern statement in the audit report. We also identified 11 cases in which Audit Analytics classified an observation as non-going concern, but we observed a going concern statement in the audit report. Including or excluding these observations does not affect our results. 11 Dealscan is provided by the Loan Pricing Corporation (LPC). Sufi (2007) reports that approximately 90% of the 500 largest nonfinancial firms in COMPUSTAT obtained a loan through private channels during his sample period of 1994 to 2002 and that the market for these loans reached $1 trillion during this period. The value of private loans grew to $1.7 trillion in 2007 (Kim et al., 2011). 12 Syndicated loans often bundle multiple facilities in to one transaction. These different facilities have different contract terms but are syndicated as a single deal. Consistent with other work using private debt contracts, we conduct our tests at the individual facility level. 15

18 borrower s 10-K on EDGAR or LexisNexis. Following Butler et al. (2004), we classify the reason for each MAO into two broad categories: Inconsistency and Inadequacy. Inconsistency MAOS refer to a lack of consistency according to GAAP accounting principles (AU 508 section) - we breakdown Inconsistency into Accounting Change and Restatement depending on whether the auditor mentions accounting changes or restatements in the explanatory paragraph. 13 Likewise, we decompose Inadequacy MAOs into two types: Material Uncertainty and GC Opinion. We classify an MAO as Material Uncertainty if the auditor mentions a business uncertainty, litigation risk, or contingent liability in the audit report we classify an MAO as GC Opinion if the auditor mentions going concern, bankruptcy, financing difficulty or distress. Table 1 reports the annual distribution of loans and annual frequency of each type of modification over our sample period. As shown in the table, the number of MAOs increases from 3 in 1992 to 345 in 2004 and decreases afterward. A notable jump in the number of MAOs takes place starting in 2002 when the Sarbanes-Oxley Act (SOX) was implemented. This is consistent with SOX increasing the pressure for auditors to issue MAOs. Consistent with Butler et al. (2004), most MAOs are related to Inconsistency with less than 10% related to Inadequacy. Our sample of MAOs contains a higher proportion of Inconsistency modifications than does the Butler et al. sample this is likely because firms in our sample are relatively large and were able to obtain financing. Within Inconsistency MAOs, most are Accounting Change with about 10% being Restatement MAOs. Within Inadequacy modifications, GC Opinion MAOs are about 3 times the number of Material Uncertainty modifications. In addition, GC opinions have a wider variation over the years with some 13 There are two major differences between our partition of MAOs and that of Butler et al. (2004). First, we read explanatory paragraphs in the audit reports of all loan observations identified on COMPUSTAT as having an MAO, rather than searching audit opinions by keyword as in their study. Second, for MAOs with multiple reasons for modification in the explanatory language, we classify the observation as having the most severe MAO type, rather than allowing multiple reasons for MAO as in their study. 16

19 concentrations in years around the implementation of SOX in The percentage of GC opinions in our sample (6.4%) is less than those reported in prior studies (Butler et al., 2002; DeFond at al., 2002) - this is likely a result of our requirement that firms must have obtained debt financing in the post-mao period to be included in our sample. If a firm was denied a loan or filed for bankruptcy because of a GC opinion, then the economic consequence of a GC opinion on debt contracting will not be captured by our empirical analysis. The implication is that our empirical analysis has a bias of underestimating the cost of MAOs, especially for GC opinions. 3.3 Descriptive Statistics Table 2 Panel A reports the descriptive statistics of the MAO and non-mao loans in our sample. We define our variables in Appendix A. The MAO loans in our sample have a mean spread above LIBOR of basis points. This is higher than the non-mao loan average of basis points and the average spread of all loans included in the DealScan database. The average loan size is $379.12M for firms with an MAO and $258.72M for firms with a clean audit opinion. Loans mature in an average of and months for firms with and without MAOs, respectively. The debt contracts of firms with (without) an MAO include an average of 2.34 (2.63) financial covenants and 5.67 (5.22) general covenants. Most loans include a performance pricing provision, require collateral, and are a revolver, regardless of whether a firm has an MAO or not. Panel B of Table 2 reports the descriptive statistics of firm characteristics for our sample of loan observations. The mean firm size for firms with (without) an MAO is $3,562.07M ($1,796.54). The means of profitability, leverage and tangibility are 0.12 (0.13), 0.26 (0.25) and 0.34 (0.32), respectively, for firms with (without) an MAO. The MAO period accounts for 38% of observations (2,056 / 2, ,321), 22% of observations fall into the after-mao category, and the remaining 40% of observations are non-mao observations. 17

20 Table 2 Panel C provides a correlation matrix. Many of the contracting terms are significantly correlated. As expected, spread is positively associated with the number of debt covenants and with requiring collateral in the univariate; and it is negatively correlated with loan size, profitability, firm size, and Z-score. These univariate results are consistent with lenders having many different mechanisms through which to design contracts (Melnik and Plaut, 1986), not just through interest spreads. We expect lenders to incorporate the information provided by audit opinions in the negotiation of debt agreements. 4. Empirical Results 4.1 Audit Opinions and Loan Spreads Table 3 presents the effects of modified audit opinions on loan spreads. We regress loan spread on MAO, After_MAO, and a set of control variables. Our first hypothesis predicts that if lenders view the auditor s MAO as a negative signal of financial reporting quality and/or of the borrower s risk, then loans issued to firms with an MAO will have a higher interest rate than loans issued to firms with a clean audit opinion. Additionally, if there is a lingering effect of an MAO even after it has been cured, then debt issued during the after- MAO period will have a higher spread as well. In Column 1, the coefficient on MAO is positive and statistically significant; loans initiated during the MAO period have a spread over LIBOR that is basis points higher than loans issued during the non-mao period. This represents an increase in the interest spread of 8.6%. 14 In Column 2 we report a significant incremental effect on loan spreads for Inadequacy modifications, basis points, but no effect for Inconsistency modifications. However, Column 3 reports that firms with GC Opinion, Material Uncertainty, and Restatement MAOs all experience significant 14 Throughout the paper, economic magnitudes are calculated by comparing the coefficient of the variable of interest to the mean of that variable when there is a clean audit opinion as reported in Table 2. For example, the overall increase in the interest spreads for firms with an MAO is: / = 8.6%. 18

21 increases in their loan spreads of , 49.37, and basis points, respectively. The result that MAOs issued because of an accounting change have no effect on the borrower s interest spread is consistent with these modifications being relative mechanical in nature and providing little incremental spread-related information about the borrower to lenders. In contrast, the significant coefficients of the other three types of MAOs suggest that they increase the cost of loans to the borrower increasing with the severity of the MAO. The incremental effect on spread of GC Opinion is significantly larger than any of the other three MAOs, and it represents a 53.4% increase relative to the non-mao period. 15 These results support H1. Many of the included control variables are statistically significant. Spreads are negatively associated with borrower profitability, firm size, market-to-book, Z-score, loan size, the inclusion of a performance pricing provision, loan maturity, and whether or not the loan is a revolver. Spreads are positively associated with leverage, cash flow volatility, credit spread, abnormal accruals, whether or not the loan is secured, and whether or not the loan is institutional. After_MAO is positive but not statistically significant across all specifications, providing no evidence that lenders continue to charge higher interest rates to borrowers who recently received MAOs. The variables of interest and the control variables capture much of the variation in the dependent variable - the R-squared is over 51% across all specifications. 4.2 Audit Opinions and the Use of Financial and General Covenants Table 4 presents the effects of MAOs on the use of financial and general covenants. Hypothesis 2 predicts that lenders will be less willing to rely on financial covenants in debt contracts after the reliability of the financial statements is brought in to question by a modified opinion from an auditor. The first column in Table 4 provides evidence that the 15 The tests of the difference between the coefficient on GC Opinion and those of each of the other opinion types are significant. 19

22 number of financial covenants included in a debt contract is lower in the MAO and after- MAO periods. The coefficient on MAO is and statistically significant, representing a decrease in the use of financial covenants by 3.8%. Column 2 shows a decrease in the use of financial covenants after MAOs related to both Inadequacy (-0.17) and Inconsistency (-0.10). Column 3 provides the results for each type of MAO and shows that the decreases in the use of financial covenants are driven by GC Opinion and Accounting Change. In contrast, the coefficients on Restatement and Material Uncertainty are not statistically significant. The coefficient on After_MAO is negative and significant across all three specifications and indicates that lenders are reluctant to include financial covenants for up to three years after a clean opinion is restored to the borrower. This suggests some lingering effect of MAOs on the use of financial covenants. H2 also predicts that lenders will be more likely to include general covenants. Column 4 provides evidence consistent with this hypothesis. The coefficient on MAO is 0.22 and statistically significant, and it indicates an average increase in the use general covenants of 4.2%. This finding provides evidence consistent with our prediction that when lenders are less willing to use accounting numbers in debt contracts, they will increase their use of the non-accounting contracting mechanisms that they have at their disposal. In contrast to the use of financial covenants, the coefficient on After_MAO is insignificant. Column 5 shows that the general covenants increase for after both Inadequacy and Inconsistency modifications, and Column 6 provides evidence that the use of general covenants increases after all MAOs except those related to restatements. The largest effect (0.95) relates to material uncertainty MAOs, those modifications in which the auditor mentions business uncertainty, litigation risk, or contingent liability issues. This result provides evidence that one way that lenders respond to an uncertain environment is through the increased use of general covenants. As in tests of H1, many of the control variables are statistically significant. Overall, our empirical 20

23 results on the effects of MAOs on the use of financial and general covenants are consistent with H Audit Opinions and the Use of Additional Loan Terms Hypothesis 3 predicts that lenders will include more stringent non-accounting loan terms after an MAO. We investigate three non-accounting mechanisms that lenders can employ in debt contracts: loan size, requirement of collateral, and loan maturity length and report the results in Table 5. Columns 1, 2, and 3 of Table 5 provide evidence that lenders decrease the loan sizes offered to borrowers with any MAO except those related to material uncertainties. Additionally, After_MAO is significant across each loan size specification. These results are consistent with lenders reacting to an auditor s concerns about the client s risk and financial statement quality by reducing loan sizes. The difference between the coefficient of GC Opinion and any of the other three is not significant. This suggests that the loan size component of the loan contract is insensitive to the differential signal of MAOs on the borrower s default risk or financial reporting quality. However, this finding is consistent with the credit rationing literature (see e.g., Jaffee and Russell, 1976; Stiglitz and Weiss, 1981) in that lenders reaction to a signal of default risk in a non-linear fashion. Our empirical results are biased against finding the differential effect of GC opinion from those of other MAOs since loan applications that were denied because of GC opinions are not represented. 16 We also find that the likelihood of requiring collateral increases significantly after Inadequacy MAOs. Column 6 indicates that the probability of requiring collateral increases after GC opinion and material uncertainty MAOs and that this increased probability lingers for up to three years after a clean opinion is issued. The probability of requiring collateral 16 As a result, our estimate of the effect of an MAO on debt terms underestimates the true cost of an MAO because borrowers are likely to incur higher costs of financing if they are denied a loan, especially because of a GC opinion, from lenders in the private debt market. 21

24 increases 15.9% and 14.4% when a firm receives an MAO related to a GC opinion and material uncertainty, respectively - these two coefficients are not statistically different from each other. It is not surprising that lenders are eager to require collateral when firms receive an inadequacy modification because the other monitoring mechanisms, such as covenants, are unlikely to be effective when the ability of the firm to continue as a going concern is in question or when there is serious uncertainty regarding a borrower s prospects. Our last test of H3 investigates loan maturity choices that lenders make at the contract initiation - we predict that lenders will decrease loan lengths after an MAO. Columns 7 and 8 indicate there is no effect on loan maturity after an MAO, even when broken down in to Inadequacy and Inconsistency modifications. However, Column 9 indicates a decrease in maturity length for firms with GC opinions but an increase in maturity length for those with material uncertainty modifications. While it is not surprising that lenders prefer to decrease loan lengths after going concern opinions, it is less clear why an MAO related to a material uncertainty would lead to an increase in loan maturity. One explanation is that Material Uncertainty MAOs do not communicate auditors information about client risk as it pertains to loan maturities. An alternative explanation is that these borrowers have greater bargaining power relative to lenders in negotiating the maturity terms of the loan because of managers information advantage about future cash flows. Overall, the loan maturity results are weaker than those of the other contract devices. 5. Additional Analyses 5.1 The Effect of Different Types of Going Concern Opinions on Loan Terms A large body of auditing literature has examined the information role of GC opinions. This is understandable because the consequence of GC opinions is most severe and GC opinions reveal auditors private information a client s financial health. In the main tests of 22