Balance Sheet Conservatism and Debt Contracting* Prudence au bilan et emprunt

Transcription

1 Balance Sheet Conservatism and Debt Contracting* JAYANTHI SUNDER, University of Arizona SHYAM V. SUNDER, University of Arizona JINGJING ZHANG, McGill University ABSTRACT We study the role of borrowers balance sheet conservatism (i.e., conservatism in asset values) in debt contract design. We find that borrowing costs are decreasing in the degree of balance sheet conservatism, and this effect is stronger for firms with lower credit quality. This is consistent with balance sheet conservatism reducing lenders uncertainty about the liquidation value of assets, thus facilitating the ex ante screening of borrowers. We predict that better ex ante screening also reduces the need for ex post monitoring, and find that balance sheet conservatism is associated with less restrictive covenant terms. Further, we find that asymmetric timeliness in earnings is associated with lower borrowing costs only when balance sheet conservatism is not high. This result suggests that lenders appear to recognize the constraining effect of high balance sheet conservatism on future conservatism in earnings. Prudence au bilan et emprunt RESUME Les auteurs se penchent sur le r^ole que joue la prudence observee au bilan des emprunteurs (quant a l evaluation des actifs) dans la conception des contrats d emprunt. Ils constatent que les co^uts d emprunt diminuent avec le degre de prudence au bilan, cette tendance etant plus marquee dans le cas des societes dont le niveau de solvabilite est moins eleve. Cette observation confirme l hypothese selon laquelle la prudence au bilan reduit l incertitude des pr^eteurs au sujet de la valeur de liquidation des actifs, ce qui facilite la selection ex ante des emprunteurs. Les auteurs posent qu une meilleure selection ex ante reduit egalement la necessite du contr^ole ex post et constatent que la prudence au bilan est associee a des clauses restrictives moins rigoureuses dans les contrats d emprunt. Ils constatent au surplus que l asymetrie de la prise en compte en resultat est associee a des co^uts d emprunt plus faibles uniquement lorsque le degre de prudence au bilan est modeste. Ces observations laissent croire que les pr^eteurs auraient conscience de l effet contraignant d un degre eleve de prudence au bilan sur la prudence ulterieure des resultats. 1. Introduction The balance sheet has traditionally been the main source of financial information for lenders, and early U.S. bondholders did not even demand earnings or sales data (Benston * Accepted by Sudipta Basu. We thank Sudipta Basu and two anonymous referees for their guidance and constructive comments. We also appreciate comments from Anne Beatty, Ningzhong Li, Thomas Lys, Darren Roulstone, Sugata Roychowdhury, workshop participants at Northwestern University, the Ohio State University, Florida State University, the Financial Accounting and Reporting Section midyear meeting, the University of Minnesota Empirical Conference, and the London Business School Accounting Symposium. The authors acknowledge support from the Accounting Research Center while the authors were at the Kellogg School of Management. Corresponding author. Contemporary Accounting Research Vol. 35 No. 1 (Spring 2018) pp CAAA doi: /

2 Balance Sheet Conservatism and Debt Contracts , 1973). Further, balance sheet asset values have provided estimates of the abandonment option (Holthausen and Watts 2001) in keeping with lenders demand for conservatism. SFAC 2 (FASB 1980, 23) states that understatement for its own sake became widely considered to be desirable, since the greater the understatement of assets the greater the margin of safety the assets provided as security for loans or other debts. We examine how conservatism in asset values reported on the balance sheet, henceforth referred to as balance sheet conservatism, affects debt contracting. Balance sheet conservatism is arguably a stable proxy for accounting conservatism because it reflects the cumulative effects of conservative accounting choices of the firm in prior periods, and past conservative write-offs of assets cannot be easily reversed under U.S. GAAP. We conjecture that balance sheet conservatism affects debt contracting through two mechanisms by providing conservative estimates of asset values and by affecting the firm s ability to write down assets in the future. The first mechanism (Asset Value Hypothesis) focuses on the direct effect of balance sheet conservatism on contractual terms. Balance sheet conservatism provides a lower bound estimate of the collateral value of the assets. In the presence of information asymmetry, conservative asset values reduce lenders uncertainty regarding asset liquidation value and ameliorate shareholder-debtholder conflicts of interest (Ahmed, Billings, Morton, and Stanford- Harris 2002; Watts 2003). Thus, conservatism in asset values improves the screening of borrowers. Further, firms can use conservatism in reported asset values as a bonding mechanism to not exploit lenders (Watts 1977; Basu 1997), which in turn reduces the need for monitoring mechanisms in the debt contract. For this reason, we expect that greater balance sheet conservatism would be associated with lower borrowing costs and a less restrictive covenant structure. Alternatively, if balance sheet conservatism adds noise to reported asset values, it would be contracting neutral, or would even increase borrowing costs (Ball and Shivakumar 2005; Gigler, Kanodia, Sapra, and Venugopalan 2009). The second mechanism (Constraint Hypothesis) focuses on the moderating effect of balance sheet conservatism on the relation between conditional conservatism in earnings and debt contract terms. According to Zhang (2008), firms that have been conditionally conservative in the past get favorable interest spreads over LIBOR because lenders expect these firms to continue their conservative practices in the future. However, past conservatism reduces the carrying value of assets on the balance sheet, which constrains future conditionally conservative write-downs (Basu 2001; Giner and Rees 2001; Beaver and Ryan 2005; Roychowdhury and Watts 2007). 1 As a result, we predict that firms with high balance sheet conservatism will not get favorable loan spreads, even if they have been conditionally conservative in prior periods. 2 The measurement of balance sheet conservatism is challenging. Some prior studies have used book-to-market ratio (hereafter, BTM) as a proxy for conservatism because BTM captures the wedge between market and accounting values of assets. 3 However, BTM would be a noisy measure of balance sheet conservatism, given that both the 1. For a given size of bad news, the magnitude of the write-off would be lower for firms with more conservative asset values. In the limit, if the asset values are conservative enough, then no write-off could be taken, as, for example, with in-house advertising and R&D assets in the United States. However, investments in new assets could potentially create new write-off opportunities and weaken this constraining effect, which biases against finding results in support of this hypothesis. 2. We do not have a formal hypothesis on the relation between conditional conservatism and covenant use because Dyreng, Vashishtha, and Weber (2017) find that firms with income-based covenants are not necessarily more conditionally conservative. 3. Studies also use market-to-book ratio (MTB) as a measure of conservatism. Following Beaver and Ryan (2000, 2005), we focus on BTM, since it has better distributional properties than MTB. Further, while some studies use BTM of equity, we use BTM of assets because lenders likely care about conservatism in overall asset valuations.

3 496 Contemporary Accounting Research denominator and numerator are affected by factors other than conservatism. Ignoring these factors would lead to misestimating the effect of conservatism on credit risk. We construct our measure of balance sheet conservatism by extracting other sources of variation in BTM such as economic rents, distress, market sentiment, inflation, and mark-tomarket accounting in order to isolate the portion that represents conservatism in asset values. While some studies view BTM as a proxy for unconditional conservatism, we recognize that balance sheet conservatism arises from past conservatism in earnings, both conditional and unconditional, and from initial conservative recording of assets on the books. We show that measures of conservatism used in prior literature are positively correlated with balance sheet conservatism, but not with the other drivers of BTM. We examine loan contracts during the period 1997 through 2007 to avoid the financial crisis years. With respect to our Asset Value Hypothesis, we find that firms with higher balance sheet conservatism, on average, face significantly lower interest spreads on the loans and have less restrictive covenants, measured by the number of covenants and covenant slack. Consistent with conservative asset values better capturing liquidation values, we also find that balance sheet conservatism is important only for firms with a moderateto-high likelihood of default, in which case liquidation values are more relevant to lenders. Further, the likelihood of using balance sheet-based covenants is decreasing in balance sheet conservatism, suggesting that conservative asset values reduce the need to monitor borrowers. Overall, our evidence highlights the screening role of balance sheet conservatism, in that loan spreads and covenant restrictiveness are decreasing in balance sheet conservatism. In exploratory analyses, we find that lenders value the portion of balance sheet conservatism attributable to the conservative recording of assets and conditional conservatism, but not unconditional conservatism in earnings. Our results are robust to alternative subsamples and specifications, as well as different control variables. With respect to the Constraint Hypothesis, we find that lenders reward past conditional conservatism in earnings with lower loan spreads only for firms with low balance sheet conservatism. Lenders appear to recognize that high balance sheet conservatism impairs contracting efficiency because firms are constrained from taking conservative write-offs in the future based on new information. Our study highlights the dual role of balance sheet conservatism in the design of debt contracts. Conservative asset valuations provide lenders with a lower bound estimate of borrowers asset values, which can be used by lenders to screen borrowers. 4 Our results underscore the ex ante screening role of balance sheet conservatism that complements the ex post monitoring role of conditionally conservative earnings documented in prior literature. However, conservatism in asset values can also adversely affect contracting efficiency because understated asset values constrain the borrowers ability to record losses in a timely manner in the future, even if they are willing to be conservative. The rest of the paper is organized as follows. Section 2 outlines the research hypotheses. Section 3 describes the measurement of balance sheet conservatism, and section 4 explains the research design. Section 5 discusses sample construction and presents the summary statistics. Section 6 provides the empirical results, and section 7 concludes the study. 4. Ahmed et al. (2002) take a first step in establishing a role for a balance sheet-based measure of conservatism in debt contracting. Our paper complements Ahmed et al. and extends it in the following ways. First, while Ahmed et al. measure conservatism using the persistent bias component of BTM, which captures both accounting conservatism and market power, we extract the economic rents from BTM to focus on conservatism. Separating conservatism and market power is important for interpreting results, because both components likely affect loan terms. Second, Ahmed et al. study debt ratings while we examine loan spreads, a more direct measure of the cost of debt. In contrast to Ahmed et al., we document a nonmonotonic relation between the cost of debt and balance sheet conservatism. Third, we examine the effect of conservatism on covenant-based monitoring of debt, which was not studied in Ahmed et al.

4 Balance Sheet Conservatism and Debt Contracts Hypotheses development Asset value hypothesis Balance sheet conservatism captures the difference between book value and market value of assets reported on the balance sheet due to the cumulative effects of past conservative accounting. Conservative asset values could be the outcome of (i) recording assets at a conservative value relative to market value; 5 (ii) rapid expensing that occurs when the firm applies conservative accounting methods, also referred to as unconditional conservatism; and (iii) asset write-downs that are conditional on bad news and cannot be reversed if economic circumstances turn around. 6,7 We do not distinguish between the alternative mechanisms of conservatism because all three eventually lead to the understatement of asset values relative to market values, as desired by lenders. This is regardless of whether the reason for the conservatism is contracting, litigation, regulation, or taxation. Also, each source of conservatism affects the ability of the firm to apply the other sources of conservative valuation (Basu 2001), thus justifying a composite construct. Finally, the challenge of isolating the different sources of conservatism in asset valuation makes the use of a composite construct appealing. Watts and Zimmerman (1986) argue that conservative accounting methods are preferred in debt contracting because they result in lower distributions to shareholders and management, and thus leave a bigger pie available for lender claims. In addition, they point out that accounting conservatism helps lenders assess the adequacy of assets to repay loans and mitigate agency problems, thus reducing the credit risk associated with lending. 8 Consistent with the argument that lenders are more concerned with the lowerend distributions of earnings and net assets, DeFond and Zhang (2014) find that bond prices react more quickly to bad news in earnings announcements compared to stock prices, and Johnston, Markov, and Ramnath (2009) find that debt analysts are more likely to cover firms with a greater probability of financial distress. Conservative asset valuation reduces uncertainty regarding the liquidation value and serves as a mechanism to screen borrowers. Moreover, the understatement of net assets limits wealth transfers from debtholders to shareholders (Ahmed et al. 2002). We expect that conservative asset values likely reduce the risk of loans, and lenders would be willing to charge lower interests from borrowers with higher balance sheet conservatism. In addition to affecting the cost of borrowing, balance sheet conservatism can also influence the use of covenants. Covenants in loan contracts are used as a monitoring mechanism to resolve agency problems between shareholders and debtholders. Jensen and Meckling (1976) suggest that a manager can bond himself against exploiting debtholders, which reduces the need for intensive monitoring by the lender. Further, according to Basu (1997), the emergence of the conservatism principle and the preparation of audited financial statements can be ascribed to managerial attempts to bond against exploiting their 5. Examples of more conservative recording of asset values include accounting for acquisitions following the pooling of interest method, and examples of less conservative recording of assets include using fair values under fresh start accounting upon emergence from bankruptcy (Lehavy 2002). 6. Under U.S. GAAP, write-downs are undertaken only if impairments are permanent or other than temporary (Banker, Basu, and Byzalov 2017). 7. A similar classification of sources of conservatism is provided in Mora and Walker (2015). 8. An alternative to conservative reporting by the firm is for lenders to make adjustments to accounting measurement rules in their contracts. While Leftwich (1983) provides evidence that private lenders make adjustments to measurement rules in order to make them systematically more conservative, Beatty, Weber and Yu (2008) conclude that reporting conservatism and modifications in accounting numbers used in the contracts are complements. Thus, contract modification alone does not meet the lenders demand for conservatism, which reiterates the importance of reporting conservatism for lenders.

5 498 Contemporary Accounting Research asymmetrically informed position relative to other claimholders. 9 When firms report asset values at their lower bound, it serves as a bonding mechanism to prevent the firm from exploiting debtholders by paying excessive dividends to shareholders. To the extent that bonding reduces monitoring costs, lenders would impose a less restrictive covenant structure on borrowers, both in terms of the number of covenants included in the contract and the covenant slack. This leads to our first hypothesis: Hypothesis 1a. Loan spreads are decreasing in balance sheet conservatism. Hypothesis 1b. The restrictiveness of the covenant structure is decreasing in balance sheet conservatism. It is important to note that while balance sheet conservatism reflects conditional and unconditional conservatism in earnings, as well as the conservative recording of assets, some prior studies have attributed the understatement of assets entirely to unconditional conservatism (e.g., Pae, Thornton, and Welker 2005). Under both views, unconditional conservatism is an important determinant of balance sheet conservatism. If unconditional conservatism merely introduces a downward bias in asset values, it could reduce contracting efficiency (Ball and Shivakumar 2005; Guay and Verrecchia 2006; Gigler et al. 2009). Taking this argument further, the predicted effect in Hypothesis 1 could be weakened, depending on the relative importance of unconditional conservatism in balance sheet conservatism. Constraint hypothesis Basu (1997, 2005) and Ball and Shivakumar (2005, 2006) stress the importance of conditional conservatism in contracting. Qiang (2007) finds that firms with greater financing costs are more conditionally conservative. Conditional conservatism, through quicker recognition of losses than gains, serves two roles in the contracting process. First, it provides valuable loss information to the contracting parties who have an asymmetric demand for bad news. Second, by triggering financial covenants in a timely manner, it increases the effectiveness of the contract terms. An implication of the above argument is that periodic conditional conservatism in earnings plays a crucial role in the covenant monitoring process and facilitates timely transfer of control to lenders. Since lenders cannot observe future conditional conservatism at loan origination, prior research assumes that lenders use the past level of conditional conservatism to proxy for the borrower s willingness to be conservative in the future. Thus, borrowers that have been conditionally conservative in the past are rewarded with favorable interest terms (Zhang 2008). The argument in favor of the willingness of borrowers to be conservative is based on their reputation concerns (Haw, Lee, and Lee 2014) and constraints, such as litigation or the use of fixed GAAP in computing covenants. However, while these arguments address whether borrowers are willing to apply the same accounting practices after entering into the debt contracts, it is uncertain whether borrowers are able to take the asset write-downs to reflect bad news when assets are conservatively reported. Beaver and Ryan (2005) illustrate how past accounting conservatism reduces accounting slack and preempts future write-offs using simulations. While past conservatism does not 9. Mechanisms used by firms to commit to continued conditional conservatism include hiring a Big N auditor (e.g., Basu, Hwang, and Jan 2001; Krishnan 2005), cross-listing in a litigious regime (e.g., Lang, Raedy, and Yetman 2003; Huijgen and Lubberink 2005), not providing D&O insurance to managers (e.g., Chung and Wynn 2008), putting financial experts on audit committees (e.g., Krishnan and Visvanathan 2008), and increasing the number of asset pools used in impairment testing (e.g., Byzalov and Basu 2016).

6 Balance Sheet Conservatism and Debt Contracts 499 influence the future realization of bad news, it affects the recognition of bad news and the amount recorded. For example, accelerated depreciation results in more conservative asset values early on, compared with straight line depreciation, thus making it harder to use asset write-downs to reflect drops in the market value of assets. Conversely, fresh start accounting is less balance sheet conservative, but leads to future income statement conservatism (Lehavy 1999, 2002). Other papers have also discussed the time-series property of conservatism, in which conservatism in earnings is history-dependent (Basu 2001; Ryan 2006). Consistent with such a conjectured constraining effect, empirical studies document that BTM, a proxy for accounting slack caused by past conservatism, is positively associated with subsequent conditional conservatism in earnings (Giner and Rees 2001; Pae et al. 2005; Ball and Shivakumar 2006; Gassen, Fulbier, and Sellhorn 2006; Roychowdhury and Watts 2007). Following Sridhar and Magee s (1996) theoretical prediction that lenders respond to the inter-temporal dynamics in accounting, we conjecture that contracting parties would recognize the preempting effect of borrowers balance sheet conservatism on future conditional conservatism in earnings. If lenders believe that borrowers past conservatism in earnings will not persist because of the constraining effect of balance sheet conservatism, they will not offer these borrowers lower interest rates. We expect that the negative relation between loan spreads and past conditional conservatism is driven by firms with low balance sheet conservatism that are not constrained in their ability to write down assets to reflect bad news. Formally, our second hypothesis, based on this constraining effect, is stated as: Hypothesis 2. Past conditional conservatism is associated with lower loan spreads only when balance sheet conservatism is low. 3. Balance sheet conservatism We extract our measure of balance sheet conservatism from BTM. While BTM is decreasing in asset value conservatism, it is also related to Tobin s Q or future growth opportunities of the firm. Moreover, firm distress, inflation, and market sentiment can affect the market value (denominator); historical cost accounting and fair value accounting can change the book value (numerator). Some of these factors are also related to a firm s credit risk. Consequently, using an unadjusted measure of BTM creates challenges in interpreting the construct, resulting in potentially erroneous conclusions about the relation between conservatism and debt contracting. To address the concern that the results could be driven by factors other than conservatism, we regress a firm s BTM on a set of variables that proxy for growth and economic rents, distress, market sentiment, unrealized mark-to-market gains, and inflation. To the extent that these variables control for other known drivers of BTM, the residual from the regression captures balance sheet conservatism. We measure conservatism using this approach rather than directly modeling all of the sources of past conservatism because it is easier to proxy for the other known drivers of BTM. We estimate the following equation with year and industry fixed effects: 10 BTM ¼ a þ b 1 LT Growth Forescast þ b 2 Sales Growth þ b 3 Industry Concentration þ b 4 1=Consumer Sentiment Index þ b 5 1=S&P Index þ b 6 Profitability þ b 7 Credit Rating þ b 8 Return Volatility þ b 9 High Inflation þ b 10 AOCI þ e; ð1þ 10. Our industry is defined using ICODE50 in the Fixed Industry Classification (FIC) dataset, which is constructed based on firm pairwise similarity scores from 10-K product descriptions and available at cwis.usc.edu/projects/industrydata/. Hoberg and Phillips (2010, 2016) find that the 10-K text-based industry classification is superior to SIC codes in explaining firm differences.

7 500 Contemporary Accounting Research where the dependent variable BTM is the book value of assets divided by the market value of equity plus the book value of debt. Using the above regression, we decompose BTM into two components. The first component, Balance Sheet Conservatism, is measured as 1 times the residual. The second component, Measurement Error Component, is defined as 1 times the fitted value. We multiply the residual and the fitted value by 1 for ease of interpretation so that Balance Sheet Conservatism is increasing in asset value conservatism and Measurement Error Component is increasing in growth. 11 We use analyst long-term growth forecast and firm sales growth to proxy for expected growth that affects BTM. We also control for industry concentration, given that firms in concentrated industries enjoy higher rents, and thus exhibit lower BTM. Market sentiment may lead to over- or understatement of market capitalization in the denominator of BTM. Therefore, we include 1/Consumer Sentiment Index used by Qiu and Welch (2006) to proxy for sentiment and 1/S&P Index to control for the general level of market prices, since investor sentiment depends on overall market conditions (Rosen 2006). Next, we control for distress using Profitability, Credit Rating, and Return Volatility. Under historical cost accounting for assets, cumulative inflation can decrease BTM, even absent conservatism (Basu 1997). We therefore use High Inflation, an indicator variable constructed from the Consumer Price Index (CPI), to control for inflation. 12 Under fair value accounting, unrealized mark-to-market gains reduce the gap between book and market values. We use AOCI, calculated as accumulated other comprehensive income scaled by total assets, to proxy for the extent of fair value accounting. 13 We expect that Credit Rating, Return Volatility, and AOCI are positively associated with BTM. For the rest of the firmlevel variables, we expect negative coefficients. The measurement of variables for (1) is described in Appendix 1. We estimate balance sheet conservatism using the sample that covers all firm-years (not restricted to our bank loan sample) included in our sample period and report the results in Appendix 1. We find that all of the coefficients on the known drivers of BTM have the expected signs. They are also statistically significant, except for the coefficient on Industry Concentration. BTM is negatively associated with firm growth and inflation, and positively associated with distress factors, fair value accounting, and the inverse of market sentiment. To validate that our measure, Balance Sheet Conservatism, improves upon the raw BTM in representing the cumulative effect of past conservatism in reported asset values, we conduct two separate tests. In the first test, we examine the relation between the two components of BTM and other measures of conservatism that proxy for firms conservative recording of asset values and past unconditional and conditional conservatism in earnings. Specifically, we run the following regressions to contrast the properties of the two components of BTM: 11. Prior studies acknowledge that BTM measures conservatism with noise. For example, Beatty et al. (2008) explicitly control for growth in their specification when they use MTB, the inverse of BTM, as a measure of conservatism. We could similarly control for other drivers of BTM in specifications where the variable of interest is balance sheet conservatism. However, when we examine the moderating role of balance sheet conservatism, using unadjusted BTM requires the inclusion of many interacted variables in the regression and makes it difficult to interpret the results. We therefore take a parsimonious approach of estimating balance sheet conservatism by purging other drivers from BTM. 12. We could alternatively interpret inflation as capturing the effect of historical cost accounting on balance sheet conservatism. In untabulated analyses, we exclude inflation from the first stage, and our inferences are unaffected. 13. AOCI does not capture some effects of fair value accounting, such as unrealized mark-to-market gains and losses on trading securities. However, since finance firms are excluded from the sample, this is not an issue in our study.

8 Balance Sheet Conservatism and Debt Contracts 501 Balance Sheet Conservatism ðmeasurement Error ComponentÞ ¼a þ b 1 LIFO Reserve þ b 2 Accelerated Depreciation þ b 3 Advertising Reserve þ b 4 R&D Reserve þ b 5 Asymmetric Timeliness þ b 6 Nonoperating Accruals þ b 7 Pension þ b 8 Goodwill þ e; ð2þ where the dependent variable is either Balance Sheet Conservatism or Measurement Error Component. We report the results with variable definitions in Appendix 2. We find that in column (1), Balance Sheet Conservatism is positively associated with measures of past conservatism and negatively associated with Goodwill, which reflects recording assets on the books at nonconservative valuations. Except for LIFO Reserve and Pension, all coefficient estimates are statistically significant. The results confirm that balance sheet conservatism reflects the cumulative effect of past conservative accounting choices and practices. When the dependent variable is Measurement Error Component (column (2)), the only coefficient that has the predicted positive sign and is statistically significant is the one on R&D Reserve. We also find that Measurement Error Component is negatively associated with LIFO Reserve and positively associated with Goodwill, providing evidence that this component is unrelated to past conservatism. 14 For completeness, we report the results for raw BTM in column (3). We multiply the dependent variable by 1 to be consistent with the direction of balance sheet conservatism. We note that LIFO Reserve and Goodwill behave differently from column (1), apparently affected by the noise inherent in raw BTM. In the second test, we examine how the two components of BTM relate to past and future conditional conservatism. In groups ranked by the two components of BTM, we run the pooled Basu (1997) regression over a three-year period, either before or after the year when Balance Sheet Conservatism is estimated. In untabulated results, we find that Asymmetric Timeliness, as defined in Basu (1997), increases in groups ranked by Balance Sheet Conservatism in the pre-period, but decreases in the post-period. This reversal pattern is similar to the findings in Roychowdhury and Watts (2007) that past conditional conservatism contributes to end-of-period MTB while beginning-of-period MTB negatively affects future conditional conservatism. Thus, past conditional conservatism increases balance sheet conservatism, which in turn constrains future conditional conservatism. We do not observe such a reversal when groups are ranked by Measurement Error Component. Collectively, the results in the validation tests are reassuring. The contrast in the properties of the two components of BTM confirms that our measure of balance sheet conservatism improves upon the raw BTM by isolating the part that captures the cumulative effect of past conservative accounting. 4. Research design Testing the Asset Value Hypothesis To examine the direct effect of balance sheet conservatism on contract terms, we estimate the following Ordinary Least Square (OLS) model with standard errors clustered at the firm level: 15 Contract Terms ¼ a þ cbalance Sheet Conservatism þ X d i ðcontrol i Þþe; ð3þ 14. The negative association between LIFO Reserve and Measurement Error Component is driven by industry characteristics. Certain industries, such as Chemicals and Consumer Non-durables, have higher LIFO reserve and lower growth. The positive coefficient on Goodwill potentially reflects inorganic growth strategies adopted by some high-growth firms. 15. We also cluster standard errors by firm and by year, and the results are similar. As Angrist and Pischke (2009) point out, clustered standard errors are less reliable when the number of groups is small. Since we only have a limited number of years in our sample, we only report the results using one-way clustering at the firm level.

9 502 Contemporary Accounting Research where the contract terms are either loan spreads or covenant-related variables. Balance Sheet Conservatism is defined in section 3. The coefficient of interest is c. First, we examine the effect of Balance Sheet Conservatism on Spread, which is the allin-drawn interest spread on the bank loan over LIBOR, expressed as a percentage. We expect a negative c. We include a set of control variables to proxy for firm- and loan-specific risks that are likely to affect loan spreads. Following Bharath, Sunder, and Sunder (2008), Costello and Wittenberg-Moerman (2011), and Ivashina and Sun (2011), we control for Log(Assets), Profitability, Loss, Leverage, Return Volatility, Credit Rating and Tangibility, which are known proxies for credit risk, in the regression. Growth can have two opposing effects on debt contracting. Greater growth opportunities suggest higher profitability in the future, which lowers credit risk for the firm. However, there is higher information asymmetry and greater debtholder shareholder conflict of interest in highgrowth firms (Myers 1977; Smith and Watts 1992). Thus, the predicted direction for the effect of LT Growth Forecast on loan spreads is ambiguous. For loan characteristics, we include Loan Amount, Maturity, an indicator variable for collateral (Secured), and the number of covenants (Total Covenants) in the regression (Booth 1992; Dennis, Nandy, and Sharpe 2000; Bradley and Roberts 2015). Because prior studies suggest that loan terms are jointly determined and lenders tend to impose more stringent terms when firm risk is high (Bharath et al. 2008; Costello and Wittenberg-Moerman 2011; Callen, Chen, Dou, and Xin 2016), we expect complementarities among loan terms. We control for the presence of a performance pricing provision, which has been shown to reduce loan spreads (Asquith, Beatty, and Weber 2005). We add Default Spread and Term Spread to control for market-wide time trends because these factors increase loan spreads (Ivashina and Sun 2011). Finally, we have indicator variables for loan type, loan purpose, industry, and year. Second, we investigate how balance sheet conservatism affects covenant intensity and covenant tightness. Covenant intensity, Total Covenants, refers to the total number of covenants contained in a debt contract. Because any covenant violation by borrowers can lead to a technical default, more covenants are generally assumed to result in a more restrictive debt contract. This measure has been commonly used in prior studies (Nikolaev 2010; Christensen and Nikolaev 2012; Ball, Li, and Shivakumar 2015). While covenant intensity captures the overall use of covenants, we also expect that the likelihood of including balance sheet-based covenants in the contract is lower, given that balance sheet conservatism reduces the need for monitoring changes in asset values. We follow Frankel and Litov (2007) to identify balance sheet-based covenants, and create an indicator variable, B-Covenant, to indicate whether the loan contains at least one balance sheet-based covenant. We measure covenant tightness using covenant slack, computed as the difference between the level of the covenant variable at loan origination and the covenant threshold specified in loan agreements. Once a covenant is included in a debt contract, tighter covenant slack imposes higher restrictions on borrowers. Because the different definitions of covenant variables between loan agreements and GAAP complicate the measurement process of covenant slack (Li 2010), we focus on (tangible) net worth covenants, following Drucker and Puri (2009). These covenants are widely used and are defined in a more standardized way in debt agreements (Dichev and Skinner 2002). Net Worth Slack is defined as the slack in a net worth covenant or a tangible net worth covenant, as applicable (only one of them appears in a given debt contract). The firm controls are the same as in the regression examining loan spreads. We expect firms with lower credit risk to have less restrictive covenants. Thus, we expect covenants to decrease in Log(Assets) and Tangibility and increase in Leverage and Credit Rating. At the same time, firms with low profitability and greater volatility likely trigger covenants

10 Balance Sheet Conservatism and Debt Contracts 503 excessively, making covenants less valuable for monitoring. Similarly, it is difficult to write income-based covenants for loss-making firms. Thus, we do not make a directional prediction for Profitability, Loss, and Return Volatility. For loan characteristics such as amount, maturity, and spread, we compute the average value of the variables at the deal level. We do so because all of the loans in a deal have the same covenant structure. Both covenants and collateral are contractual devices to monitor borrowers (Rajan and Winton 1995). We therefore control for collateral if at least one loan in a deal is secured. When we estimate the likelihood of using a balance sheet-based covenant, we include Volatility Ratio proposed by Demerjian (2011) to measure the effect of the shift in accounting standards. When we examine covenant tightness, we control for the level of (tangible) net worth in the quarter before loan origination, because Demiroglu and James (2010) argue that covenant slack is correlated with the level of the covenant variable. We also control for the presence of build-up provisions, which can tighten slack (Beatty et al. 2008). Last, we include loan type, loan purpose, year and industry fixed effects in the regressions. Testing the Constraint Hypothesis To test the moderating effect of balance sheet conservatism on the relation between conditional conservatism and loan spreads, we divide the observations into three groups ranked by Balance Sheet Conservatism and estimate the following regression separately for the low and high groups with standard errors clustered at the firm level: Spread ¼ a þ casymmetric Timeliness þ X d i ðcontrol i Þþe; ð4þ where Asymmetric Timeliness, our measure of conditional conservatism, is estimated from Basu s (1997) market-based model: NI it = b 0 + b 1 R it + b 2 DR it + b 3 R it DR it + ɛ it. Asymmetric Timeliness is captured by b 3. The definitions of the variables used in the model are as follows. NI it is net income before extraordinary items for firm i in fiscal year t deflated by the market value of equity at the beginning of the year and adjusted by the average EP ratio for sample firms in fiscal year t, R it is the 12-month fiscal-year return for firm i less the corresponding size decile return, and DR it is an indicator variable equal to one if the firm s adjusted return R it is negative, and zero otherwise. 16 Observations with deflated earnings or returns falling either in the top or bottom 1 percent are truncated. We estimate Basu s model at the industry level, defined by three-digit SIC codes for each year of the sample period using the prior 10 years of data. 17,18 Industry-years with fewer than 10 observations are excluded to ensure a reliable estimate of Asymmetric Timeliness. The corresponding ranked industry-year measure is assigned to each sample firm. If lenders recognize the constraining effect of balance sheet conservatism on future conditional conservatism and adjust loan spreads, c should differ in subsamples based on low and high values of Balance Sheet Conservatism. We estimate equation (4) for the two subsamples separately using seemingly unrelated regression. This technique allows us to compare the coefficients across groups. We expect that the negative relation between 16. Following Callen et al. (2016), we use net income before extraordinary items. Our results are robust to measuring asymmetric timeliness with net income. 17. We define industry here using SIC rather than FIC because our sample period starts from 1997, and we require the prior 10-year history to estimate Asymmetric Timeliness, while the industry classification constructed by Hoberg and Phillips (2010, 2016) only starts from We also follow Hui, Klasa, and Yeung (2012) to estimate Asymmetric Timeliness by running firm-specific time-series Basu (1997) regressions in a rolling window of at least seven observations for each firm-year. We obtain similar results. Due to the more restrictive data requirements, the sample size shrinks by more than one-third. In addition, Givoly, Hayn and Natarajan (2007) point out that firm-specific time-series regressions with very few observations for each firm are likely to result in a metric with considerable measurement error. We thus report the results using the industry-level estimation of Asymmetric Timeliness.

11 504 Contemporary Accounting Research conditional conservatism and loan spreads documented in Zhang (2008) is driven by firms with low Balance Sheet Conservatism. We include the same control variables as in equation (3), except for industry fixed effects because our variable of interest, Asymmetric Timeliness, is estimated at the industry level. 5. Sample selection and descriptive statistics Our sample includes loan initiations from the DealScan database for the years We stop our sample in 2007 so as to avoid the financial crisis period when the types of firms raising debt and the design of loan contracts depart markedly from other periods (Ivashina and Scharfstein 2010; Christensen and Nikolaev 2012). Consistent with prior studies, we focus on dollar-denominated loans borrowed by U.S. firms, and we exclude borrowers in the regulated financial and utility industries. We retain revolvers and term loans with a maturity greater than one year. Further, we drop any loans without spread, maturity, covenant, or loan amount information. When a deal contains multiple loans, we retain the loan with the highest loan amount. We match borrowers in the loan data to firms in the COMPUSTAT universe using the DealScan-COMPUSTAT Linking Database (Chava and Roberts 2008). Continuous variables are truncated at the top and/or bottom 1 percent, as appropriate. After requiring the availability of data to compute balance sheet conservatism and borrower-specific control variables, the final sample contains 5,158 loans. Table 1 tabulates the distribution of the loans and firms in our sample from 1997 through The distribution of loans/firms across years reflects the business cycle effects on loan markets. We include year fixed effects in our analyses to absorb any aggregate time effect. We also report the distribution of firms in the COMPUSTAT universe in the last column for comparison. Our sample is broadly consistent with the COMPUSTAT sample with the exception of the year We conjecture that the disparity between the COMPUSTAT sample and loan sample in 2000 is because there are a large number of newly public firms in COMPUSTAT following the hot IPO market of the late 1990s, and also because firms prefer to raise equity rather than raise debt when share prices are high (Baker and Wurgler 2002). Table 2 reports descriptive statistics of the major variables used in this study. 19 There is significant variation in firm size, with the mean value of total assets being over $3 billion, while the median is less than $1 billion. The average firm is profitable and the median credit rating is 14, which corresponds to a credit rating of BB. The median loan spread is 150 basis points and the median maturity is five years. On average, the loans are subject to 4.66 total covenants. Most loan variables are comparable with prior studies (e.g., Bradley and Roberts 2015). The median BTM is 0.67, which is comparable to Callen et al. (2016). Balance Sheet Conservatism in the final sample has a mean value of Since the distributions of Assets, Net Worth, Loan Amount, Maturity, Total Covenants, and Net Worth Slack are right-skewed, we transform these variables to their log forms in the regression analyses. 6. Empirical results Testing the Asset Value Hypothesis In Table 3, panel A, we examine the direct effect of balance sheet conservatism on loan spreads. Consistent with the Asset Value Hypothesis, the first column shows that firms with higher balance sheet conservatism experience lower loan spreads. The coefficient on Balance Sheet Conservatism is negative and statistically significant at the 1 percent level. 19. We observe that the firms in the subsample requiring the presence of a (tangible) net worth covenant are quite similar to the firms in the full sample, except that they are smaller (the mean value of total assets is $520 million and the median value is $319 million).

12 Balance Sheet Conservatism and Debt Contracts 505 TABLE 1 Sample distribution by year Loans Loan sample Firms COMPUSTAT Firms Year Number Percent Number Percent Percent Total 5, , Notes: This table provides the breakdown of the number of loans and firms in the sample by year. The distribution of the firms in COMPUSTAT is also provided in the last column for comparison. The sample contains loans originated from 1997 through 2007 with available loan data and firmrelated variables. Controlling for other factors that affect Spread, we note that when Balance Sheet Conservatism increases from the 25th to the 75th percentile, Spread drops by 12 basis points, which represents an 8 percent decrease, compared to the median spread over the sample period. 20 This economic magnitude is similar to the economic effects of other firm characteristics. For instance, an increase (decrease) in Profitability (Leverage) from the 25th to the 75th percentile is associated with a 13 basis-point decrease (a 14 basis-point increase) in Spread. The coefficients on the control variables generally behave as expected. Firms with higher profitability, lower leverage, better credit ratings, and less volatile returns tend to incur lower borrowing costs. The coefficients on loan characteristics, such as Loan Amount and Maturity, are negative and significant, while the coefficient on the number of total covenants is positive and significant. These results confirm the prior findings that loan terms can be complementary (e.g., Bharath et al. 2008; Callen et al. 2016). To alleviate the concern that we have not fully extracted growth prospects while constructing our measure of Balance Sheet Conservatism, we perform regression analyses in two subsamples that exclude high-growth firms. In column (2), we remove high-tech firms, namely, firms in computers and pharmaceuticals industries. In column (3), we drop young firms, that is, the firms in the lowest quintile ranked by firm age. We continue to find statistically significant negative coefficients on Balance Sheet Conservatism, with coefficient magnitudes comparable to those found in the full sample. The findings in the subsamples 20. An alternative way to evaluate the incremental importance of balance sheet conservatism in explaining loan spreads is to compare the adjusted R 2 of the regressions, including and excluding Balance Sheet Conservatism. Using the Vuong test to compare R 2 s, we find a statistically significant improvement in the fit of the regression at the 1 percent level when we include Balance Sheet Conservatism in the specification. The magnitude of the incremental R 2 is similar to that of other important firm characteristics such as leverage and profitability.

13 506 Contemporary Accounting Research TABLE 2 Descriptive statistics Variable N Mean Q1 Median Q3 SD Firm characteristics Assets (millions) 5,158 3, ,077 14,581 Profitability 5, Loss 5, Leverage 5, Credit Rating 5, Return Volatility 5, LT Growth Forecast 5, Tangibility 5, Net Worth (millions) 1, Loan characteristics Spread (basis points) 5, Loan Amount (millions) 5, Maturity (months) 5, Secured 5, Performance Pricing 5, Total Covenants 5, B-Covenant 5, Net Worth Slack (millions) 1, Build-up 1, Conservatism measures BTM 5, Balance Sheet Conservatism 5, Asymmetric Timeliness (rank) 5, Notes: This panel provides descriptive statistics of the major variables used in this study. The full sample contains 5,158 loans. For the variables that we only use in the test for covenant tightness, we report descriptive statistics of the subsample with (tangible) net worth covenants. Variable definitions are provided in Appendix 3. suggest that the negative relation between Balance Sheet Conservatism and Spread is unlikely to be driven by the effect of growth. In additional analyses, we examine whether balance sheet conservatism is more valuable for borrowers with a higher probability of default. This test is based on the intuition that the balance sheet captures liquidation value (Watts and Zimmerman 1986; Hayn 1995; Burgstahler and Dichev 1997) and that conservative asset values provide lenders with greater confidence in the liquidation value. We reexamine the results in Table 3, panel B in three subsamples based on credit ratings. The high credit quality firms are those with credit ratings of A and higher; the medium credit quality group comprises borrowers with ratings of BBB through BBB+; and the low credit quality captures ratings BB+ and below. 21 We find that the coefficient on Balance Sheet Conservatism is insignificant for firms that are investment grade (column (1)). As firms approach junk status (column 2) or already have a junk rating (column (3)), we find that the coefficients on Balance Sheet Conservatism are negative and statistically significant. This result is consistent with our intuition that lenders value balance sheet conservatism when the probability 21. This classification is consistent with the National Association of Insurance Commissioners (NAIC) classification of investment quality.