The effects of debt contracting on voluntary accounting method changes.
I. INTRODUCTION
Watts and Zimmerman (1986, 216-217) argue that debt contracts that make covenant thresholds a function of financial ratios give borrowers an incentive to change accounting methods to avoid costly covenant violations. However, reviewing the empirical research on accounting choice,
First, previous research that does not use the details of firms' actual debt contracts assumes that contract calculations are based on current accounting methods. However, Mohrman (1996) and Beatty, Ramesh, and Weber (2002) document that private debt contract calculations often prohibit firms from changing accounting methods. For these contracts, borrowers cannot use voluntary accounting method changes to avoid covenant violations. Thus, debt contracts give borrowers an incentive to change accounting methods only if the voluntary accounting method changes affect contract calculations.
Second, existing research that focuses exclusively on debt covenants ignores other accounting-based features of debt contracts. Performance pricing is a relatively new feature in bank debt contracts that explicitly makes the interest rate charged on a bank loan a function of the borrower's current creditworthiness. Asquith et al. (2002) document that performance-pricing features typically measure the borrower's creditworthiness using financial ratios such as debt to earnings before interest, taxes, depreciation, and amortization (EBITDA), leverage, or interest coverage. That is, the interest rate charged in the contract does not remain fixed over the length of the loan, but varies inversely with changes in measures of financial performance. Compared to covenants under which accounting information affects loan rates only when the borrower violates a single threshold, performance pricing creates a more continuous and direct link between accounting information and interest rates. Thus, performance pricing likely gives managers additional incentives to make income-increasing accounting method changes. Previous research does not consider whether this feature of debt contracts influences accounting choice.
Third, previous research has focused on borrowers who were either close to violating or had already violated covenants. These studies may underestimate the effect of debt contracting on accounting choice for cases in which borrowers have effectively used accounting changes to provide slack in financial covenants, and thus never come close to covenant violations.
To provide more conclusive evidence that debt contracting concerns affect borrowers' accounting choices, we address these three limitations of prior research. First, our research design incorporates the fact that debt contracts often prohibit borrowers from using voluntary accounting method changes to affect contract calculations. This cross-sectional variation in bank debt provisions allows us to examine whether managers who have chosen to change their accounting methods are more likely to make income-increasing changes when the debt contract allows these changes to affect contract calculations. Second, we consider whether another debt contracting feature--accounting-based performance pricing-increases borrowers' tendencies to make their voluntary accounting method changes income-increasing rather than income-decreasing. Third, rather than restricting our sample to borrowers approaching covenant violations, we study all borrowers who make voluntary accounting changes, and control for their other incentives to change accounting methods. If borrowers effectively use changes in accounting methods to create slack in financial covenants so that they avoid coming close to violating their covenants, then our tests examining all accounting method changes are more powerful than tests employed in previous research that focused solely on borrowers who are close to or had already violated their covenants.
To provide evidence on how debt contracts affect the sign of voluntary changes that borrowing firms make in their accounting methods, we identify a sample of borrowers with material bank debt who have chosen to make accounting changes. We determine whether their contracts have accounting-based restrictive covenants, dividend restrictions, or performance-pricing requirements. Beatty, Ramesh, and Weber (2002) find that borrowers willingly pay higher interest rates to obtain bank debt contracts that allow voluntary accounting changes to affect the calculation of financial terms in those debt contracts. We therefore expect borrowers to make use of this costly flexibility. Consistent with our hypothesis, we find that borrowers who change their accounting methods are more likely to make income-increasing changes if their debt contracts allow the changes to affect contract calculations. This increase in likelihood of an income-increasing accounting change is attenuated when the cost of a covenant violation is lower because all the firm's bank debt is from a single lender, and occurs only for borrowers whose contracts have performance-pricing provisions and dividend restrictions.
Our debt contracting results hold even after we control for other motives that borrowers have for changing accounting methods. Specifically, our results hold even after we control for the fact that incentives arising from executive compensation contracts and incentives to meet earnings benchmarks increase borrowers' propensity to make income-increasing (rather than income-decreasing) changes. We also control for the fact that a borrower with a new Chief Executive Officer (CEO) who reports a large loss before the effect of the accounting method change is more likely to report an income-decreasing accounting change (consistent with new CEOs of poorly performing firms taking "big baths"). Our debt contracting results also hold after controlling for tax incentives.
Our evidence that debt contracting concerns affect borrowers' accounting choices is more conclusive than previous research because we show that borrowers take advantage of the flexibility to make income-increasing accounting method changes that affect contract calculations. In addition, our evidence suggesting that incentives to lower interest rates through performance pricing and incentives to retain dividend payment flexibility appear to affect borrowers' accounting method choices helps address Fields et al.'s (2001) fundamental questions of whether, under what circumstances, and how accounting choice matters.
Section II develops our hypotheses and Section III explains our research design. Section IV describes our sample selection and provides descriptive statistics. We discuss our empirical results in Section V, provide sensitivity analysis in Section VI, and present our conclusions in Section VII.
II. HYPOTHESIS DEVELOPMENT
Fields et al. (2001) discuss three motives for accounting choice: (1) contracting (including debt and management compensation contracts), (2) asset pricing, and (3) influencing external parties (e.g., the Internal Revenue Service [IRS]). This paper focuses on the relation between accounting method changes and debt contracting while controlling for the other incentives for accounting choice, as explained in Section III. Specifically, we examine the extent to which detailed provisions of debt contracts explain the sign of borrowers' voluntary changes in accounting methods. We focus on accounting method changes because Beatty, Ramesh, and Weber (2002) document that borrowers willingly pay higher interest rates to retain the flexibility to make voluntary accounting method changes that affect bank debt contract calculations. We therefore expect borrowers to make use of this costly flexibility by making voluntary accounting method changes to affect contract calculations.
Does Allowing Accounting Changes to Affect Debt Contract Calculations Influence Accounting Choice?
Previous research examining whether debt covenants affect borrowers' decisions to change accounting methods generally assumes that the calculations stipulated in the covenants will be based on current accounting methods. For example, Healy and Palepu (1990) examine whether borrowers close to violating their dividend covenant restrictions, which are typically defined as a percentage of net income or retained earnings, change their accounting methods to increase those limits. Similarly, Sweeney (1994) investigates whether borrowers use accounting changes to avoid violating financial covenants designed to monitor borrowers' performance. However, Mohrman (1996) and Beatty, Ramesh, and Weber (2002) document that debt contract calculations are often based on the accounting rules that the borrower used when the contract originated. For these contracts, the borrower cannot use changes in accounting methods to avoid covenant violations. The cross-sectional variation in the effects of accounting method changes on contract calculations provides a natural setting that allows us to investigate the effects of debt contracting on borrowers' decisions to make accounting method changes.
Our first hypothesis is that borrowers who change accounting methods will be more likely to make income-increasing (rather than income-decreasing) accounting method changes when such changes affect covenant calculations.
Do the Expected Costs of Covenant Violations Influence Accounting Choice?
Watts and Zimmerman (1986, 215-216) argue that costly covenant violations give borrowers an incentive to make income-increasing accounting method changes to create slack in covenants; however, evidence on the actual magnitude of the cost of covenant violations is mixed. On the one hand, Beneish and Press (1993) provide evidence that covenant violations that lenders have not waived as of the financial-reporting date cost the borrower an average 80 basis point increase in the interest rate the lender charges on the loan after renegotiation. Furthermore, Healy and Palepu (1990) find that firms close to contractually imposed dividend limitations generally cut the dividends they pay rather than violate the dividend-restricting covenants or change their accounting methods to circumvent the covenants. These findings suggest that borrowers perceive violating dividend restriction covenants to be more costly than cutting dividends. On the other hand, Dichev and Skinner (2002) report that lenders may commonly waive technical violations prior to the financial reporting date, and they argue that waived covenant violations are unlikely to be very costly to the borrower. These results suggest that the cost of technical covenant violations will be lower the more likely it is for the lender to waive violations.
Our second hypothesis is that borrowers whose bank debt contracts allow voluntary accounting changes to affect contract calculations will be less likely to make income-increasing (rather than income-decreasing) accounting method changes when they expect lower technical default costs.
Does Performance Pricing Influence Accounting Choice?
We expect performance pricing, which is a relatively new debt-contracting feature, to give borrowers an additional incentive to make income-increasing accounting changes beyond the incentive provided by covenants. Accounting-based performance pricing explicitly makes the interest charged on a bank loan a function of the borrower's financial ratios, such as debt to EBITDA, leverage, or interest coverage. That is, the interest rate in the contract is not fixed over the length of the loan but varies inversely with changes in accounting measures of financial performance, which are typically assessed quarterly using the previous four quarters' results. Performance pricing may increase the importance of debt contracts in accounting choice by providing a more continuous and direct link between accounting information and interest rates than covenants provide.
Asquith et al. (2002) provide descriptive evidence on this new feature found primarily in bank debt contracts. They document that at the inception of the contract the interest rate charged on the loan is rarely above the top level or below the bottom level of the performance-pricing grid, that the average grid has five levels, and that the average increase in interest rates charged when the borrower's performance deteriorates one performance-pricing level is 15 basis points. Beatty, Dichev, and Weber (2002) document that a contract with performance pricing typically has a covenant based on the performance pricing ratio that is set at one level of performance below the worst level included in the grid. Because performance-pricing grids have multiple levels, deteriorations in a borrower's performance may result in a new performance-pricing level, and thus a higher interest rate, even if the borrower's performance never deteriorates enough to violate the covenant. In addition, improvements in a borrower's performance may result in a lower interest rate if the borrower reaches a better performance-pricing level. Thus, we expect performance pricing to provide an additional incentive for borrowers to make income-increasing accounting method changes.
Our third hypothesis is that borrowers with debt contracts allowing voluntary accounting method changes to affect contract calculations will be more likely to make income-increasing (rather than income-decreasing) accounting method changes if their contracts include accounting-based performance pricing in addition to traditional covenants.
Do Dividend Restrictions Influence Accounting Choice?
Borrowers' incentives to make income-increasing accounting method changes may also depend on whether their debt contracts contain dividend restrictions based on accounting performance. Debt contracts often have covenants that either explicitly limit dividend payments to be less than a given percentage of net income, or implicitly limit dividend payments by including covenants affected by dividends (e.g., net worth, tangible net worth, or leverage ratio covenants). The incentives to make income-increasing rather than income-decreasing accounting method changes provided by covenants that explicitly or implicitly restrict dividends may differ from those provided by other covenants (e.g., working capital or coverage ratio covenants). Healy and Palepu (1990) suggest that borrowers' incentive to make an income-increasing accounting change may be lower for covenants affected by dividend payments, because borrowers have the option to reduce dividends rather than violating the covenant. On the other hand, Sweeney (1994) documents that covenants affected by dividend payments are more likely than other covenants to be binding, suggesting that they may give borrowers a greater incentive to make an income-increasing accounting change. Thus, we are unable to predict the direction in which dividend restrictions will affect borrowers' accounting choices.
Our fourth hypothesis is therefore a nondirectional one: the likelihood that borrowers with debt contracts allowing voluntary accounting method changes to affect contract calculations make income-increasing (rather than income-decreasing) accounting method changes will depend on whether their debt contracts contain accounting-based dividend restrictions.
III. RESEARCH DESIGN
All of our tests use a sample of borrowers who have bank debt and who have also made voluntary changes in their accounting methods. Focusing on firms that make accounting changes has the advantage of requiring that the firms have enough flexibility to make an accounting change. Firms that want to make income-increasing accounting changes but are unable to do so should make no such changes and therefore not be included in our sample. This increases the power of our tests relative to prior research that examines the debt-contracting hypothesis using samples of borrowers who are close to violating covenants, which suffers from the problem that many of these borrowers may not have the flexibility to make accounting changes.
We focus on the effect of bank debt (rather than public debt) provisions on the borrower's voluntary accounting method changes for three reasons. First, Beatty, Ramesh, and Weber (2002) find that, in contrast to public debt, there is variation in whether bank debt contracts allow voluntary accounting changes to affect contract calculations. Second, El-Gazaar and Pastena (1991) find that bank debt contracts virtually always contain covenants, whereas public debt often does not, and that the covenants in bank debt are typically tighter than those in public debt. (1) Third, Asquith et al. (2002) find that, unlike public debt, many bank debt contracts have performance-pricing provisions, an innovation that likely provides additional incentives for firms to make income-increasing accounting changes. The greater cross-sectional variation in whether accounting changes affect contract calculations, combined with the potentially greater importance of covenants and performance pricing in bank debt compared to public debt, should increase the power of our tests.
Testing Hypotheses 1 and 2
We test our first hypothesis by examining whether a borrower's decision to make an income-increasing (rather than an income-decreasing) accounting method change depends on whether that change affects debt contract calculations. We test our second hypothesis by examining whether this relation depends on the expected costs of technical covenant violations. To control for other incentives for making income-increasing accounting method changes, such as those provided by management compensation contracts, asset pricing, and external parties like the IRS, as described by Fields et al. (2001), we estimate the following logistic regression:
(1) INCREASE = [beta.sub.0] + [beta.sub.1]ACC_Change + [beta.sub.2]One_Lender + [beta.sub.3]COMP + [beta.sub.4]SMLOSS + [beta.sub.5]SMLOSS_NL + [beta.sub.6]LGLOSS + [beta.sub.7]LGLOSS_[DELTA]CEO + [[beta.sub.8]NOL + [beta.sub.9]SIZE + [epsilon],
where:
Dependent Variable:
INCREASE = 1 if the accounting method change increases income and equity, 0 otherwise.
Debt-Contracting Independent Variables:
ACC_Change = 1 if contract allows accounting changes to affect calculations, 0 otherwise; and
One_lender = 1 if accounting method changes affect debt contract calculations and a single lender makes the loan, 0 otherwise.
Control Variables:
COMP = 1 if the firm indicates that the managers' bonuses depend on financial performance and the CEO received a bonus in the year prior to the accounting change, 0 otherwise;
SMLOSS = 1 if the firm's earnings before the effect of the accounting change is between 0 and -1 percent of assets, 0 otherwise;
SMLOSS_NL = 1 if the firm's earnings before the effect of the accounting change is between 0 and -1 percent of assets and the firm is not listed on a major stock exchange, 0 otherwise;
LGLOSS = 1 if the firm's earnings before the effect of the accounting change is less than -1 percent of assets, 0 otherwise;
LGLOSS_[DELTA]CEO = 1 if the firm's earnings, before the effect of the accounting change, is less than -1 percent of assets and the firm's CEO changed in the year prior to the accounting change, 0 otherwise;
NOL = 1 for borrowers who have a net operating loss carryforward and make an accounting change that has a tax effect, 0 otherwise; and
SIZE = the log of the firm's assets before the effect of the accounting change.
Our dependent variable is a simple dichotomous variable that equals 1 if the accounting change increases the current period net income, and 0 if the accounting method change decreases net income.
Debt-Contracting Variables
To test our first hypothesis, our logit includes a dichotomous independent variable (ACC_Change) that equals 1 if the borrower has any bank debt contracts that allow accounting changes to affect contract calculations, 0 otherwise To classify the effect of accounting changes on a borrower's contract calculations, we obtain copies of all of the bank debt contracts the borrower has entered at the time of the accounting change. We then use an approach similar to Mohrman (1996) and Beatty, Ramesh and Weber (2002) to identify whether accounting changes affect the contract calculations. Specifically, we classify a contract as not allowing voluntary accounting changes to affect calculations if the contract states either (1) that the borrower shall not make any significant change in accounting treatment or reporting practices, except as required by generally accepted accounting principles (GAAP) or (2) that accounting terms will be based on financial information prepared in accordance with GAAP as in effect and applied on the date of the agreement. If any of a borrower's bank debt contracts allow accounting changes to affect calculations, we then set ACC_Change equal to 1. Only when all of the firm's bank debt contracts prohibit accounting changes from affecting calculations do we set ACC_Change equal to zero.
To test our second hypothesis, we include a dichotomous independent variable (One_lender) equal to 1 if accounting method changes affect debt contract calculations and a single lender makes the loan, 0 otherwise. Asquith et al. (2002) argue that renegotiation costs are lower when there is a single lender, because contracts typically require a majority, supermajority, or unanimous consent of all lenders to waive a covenant. Therefore, the expected costs of technical default are lower when there is only one lender. (2)
Control Variables
To isolate the relation between accounting changes and debt contracting costs, we control for firms' other incentives to make accounting changes. In their summary of the empirical and theoretical accounting choice literature, Fields et al. (2001) suggest that management compensation contracts, asset pricing, and external parties such as the IRS provide additional incentives affecting firms' accounting choices beyond those provided by debt contracting.
To control for incentives arising from accounting-based compensation contracts, we include a dichotomous variable (COMP) equal to 1 if a firm indicates that its managers' bonuses depend on accounting-based measures and its CEO received a bonus in the year prior to the accounting change. We expect firms that pay accounting-performance-based bonuses to be more likely than other firms to make income-increasing (rather than income-decreasing) accounting method changes.
To capture borrowers' asset-pricing incentives to manage earnings, we control for borrowers' potential attempts to increase share prices by avoiding reporting a loss. Specifically, we include two variables to control for borrowers' use of accounting changes to achieve the positive earnings threshold Burgstahler and Dichev (1997) document. We set the first variable (SMLOSS) equal to 1 if the firm would have reported a loss of 1 percent of assets or less prior to the effect of the accounting change. These firms have an incentive to make an income-increasing accounting change to avoid reporting a loss. We construct the second variable (SMLOSS_NL) by interacting SMLOSS with a dichotomous variable equal to 1 for firms not listed on a major stock exchange, 0 otherwise. Based on the results reported in Beatty, Ke, and Petroni (2002), we expect firms not listeed on a major exchange to have less incentive to avoid missing earnings benchmarks.
For firms expecting to report a large loss in the absence of an accounting change, management compensation and asset-pricing incentives may prompt managers to increase the magnitude of the loss to create reserves for the future. We include two variables to control for this "big bath" incentive. The first variable (LGLOSS) equals 1 if the firm would have reported a loss greater than 1 percent of assets prior to the effect of the accounting change. We construct the second variable (LGLOSS_[DELTA]CEO) by interacting LGLOSS with a dichotomous variable equal to 1 if the firm changed CEOs in the year prior to the accounting change, 0 otherwise. We expect the big bath incentive to be higher for new CEOs who can use poor past performance to explain the magnitude of the loss.
To control for borrowers' tax incentives to change accounting methods, we include a dichotomous variable (NOL) equal to 1 for borrowers who have net operating loss carry-forwards, and who also make accounting changes that have tax effects; 0 otherwise. We expect firms with NOL carry-forwards to be more likely than others to make income-increasing accounting method changes that affect tax accounting.
Finally, we include the log of a firm's assets to control for firm size (SIZE). We make no prediction about the effect of firm size on the likelihood the manager makes an income-increasing rather than income-decreasing accounting method change.
Testing Hypotheses 3 and 4
To test our third and fourth hypotheses, we re-estimate our model by substituting the specific type of debt contract feature affected by the accounting method changes for our ACC_Change variable. We partition firms whose contracts allow accounting changes to affect contract calculations into four mutually exclusive categories: those with neither performance pricing nor dividend restriction provisions; those with performance-pricing but no dividend restrictions; those with dividend restrictions but no performance-pricing; and those with both performance-pricing and dividend restrictions. Specifically, we estimate the following logistic regression:
(2) INCREASE = [beta.sub.0] + [beta.sub.1a]NoPPorDiv_Res + [beta.sub.1b]PP + [beta.sub.1c]Div_Res + [beta.sub.1d]PP&Div_Res + [beta.sub.2]One_Lender + [beta.sub.3]COMP + [beta.sub.4]SMLOSS + [beta.sub.5]SMLOSS_NL + [beta.sub.6]LGLOSS + [beta.sub.7]LGLOSS_[DELTA]CEO + [beta.sub.8]NOL + [beta.sub.9]SIZE + [epsilon]
where:
NoPPorDiv_Res = 1 if the contract allows accounting changes to affect contract calculations but contains neither an accounting-based performance-pricing provision nor an accounting-based dividend restriction, 0 otherwise;
PP = 1 if the contract allows accounting changes to affect contract calculations and contains an accounting-based performance-pricing provision but not an accounting-based dividend restriction, 0 otherwise;
Div_Res = 1 if the contract allows accounting changes to affect contract calculations and contains an accounting-based dividend restriction but not an accounting-based performance-pricing provision, 0 otherwise; and
PP&Div_Res = 1 if the contract allows accounting changes to affect contract calculations and contains both an accounting-based performance pricing provision and an accounting-based dividend restriction, 0 otherwise.
All other variables are as defined in Model (1).
Since all borrowers in our sample have some type of accounting-based covenant, this research design allows us to examine the incremental effect of performance pricing and to differentiate between covenants that are linked to dividend restrictions vs. those that are not.
IV. SAMPLE
To obtain a sample of firms that have bank debt and also made material voluntary accounting changes, we conduct a keyword search on the Lexis/Nexis database. (3) When a firm makes a nonmandated accounting change, the Securities and Exchange Commission (SEC) requires the firm's independent accountant to issue a "preferability letter" commenting on whether they prefer the new accounting method relative to the prior method. The firm must file these letters with the SEC as exhibit 18 in their annual 10-K or quarterly 10-Q financial statements.
To identify firms that change accounting methods, we search all 10-K and 10-Q reports available on Lexis/Nexis from 1995 through 2000 for exhibit 18 disclosures. Then we read each exhibit 18 report to determine whether the firm made a material accounting change. Table 1 describes the results of this sample selection process. This search yields a sample of 296 firms that made accounting changes from January 1, 1995, through June 30, 2000. From these 296 firms, we exclude 48 that made accounting changes that had no effect on income statement or balance sheet accounts (e.g., changes in the format of the income statement, changes in the definition of a cash equivalent, or immaterial changes in accounting methods) and we also exclude two firms whose preferability letters were actually for mandatory accounting changes. We also excluded five firms that were banks or had banks as significant subsidiaries, because banks' accounting method choices may be affected by regulatory incentives.
We then used the debt footnotes to determine the amount and type of debt outstanding, and used the exhibit list at the end of each filing to determine whether the borrower filed a debt contract and, if so, the exact filing that includes that debt contract. Based on this analysis, we excluded another 65 firms with no bank debt or with immaterial bank debt.(4) These adjustments left a sample of 176 nonfinancial firms with material bank debt that also made voluntary material changes in accounting methods.
For each of these 176 borrowers, we systematically searched their debt footnotes, 10-Ks, 10-Qs, 8-Ks, and all their registration statements filed online to obtain copies of their debt contracts. We identified 24 firms whose footnotes indicate that they have a material bank debt contract, but whose exhibit indexes indicate that they did not file copies of the bank debt with the SEC. More specifically, we examine the exhibit index in the borrower's 10-K filed before and after the accounting change to determine whether the borrower filed the bank debt contract. For each of these 24 firms, no reference to any bank debt contracts appeared in the entire exhibit index. If the exhibit indexes of the 10-Ks filed before and after the accounting change did not mention bank debt, we assume that the borrower never filed it with the SEC. (5) In addition, we could not access debt contracts for another 27 firms, even though their exhibit index indicates that they filed bank debt contracts with the SEC. (6) Because our study focuses on the effects of debt contract provisions such as covenants, dividend restrictions, and performance pricing on borrowers' voluntary accounting changes, and we do not know whether these debt contracts allowed accounting changes to affect contract calculations, we must exclude these 51 borrowers from the sample.
This sample selection process yielded our final sample of 125 firms. (7) Although our sample attrition rate is high, we argue that our inferences are unlikely to be an artifact of sample selection bias for two reasons. First, we examine the types of accounting changes made by the firms we exclude, and the effects of these changes on net income, to ensure that our data requirements are not systematically related to our dependent variable. We find these firms' accounting changes are remarkably similar to those made by the sample firms. Of the 51 firms excluded from the sample, 29 (57 percent) made income-increasing accounting changes and 22 (43 percent) made income-decreasing accounting changes, compared to 54 percent increases and 46 percent decreases in our sample firms. The difference between these two groups is not statistically significant. As for the sample firms, the most common accounting changes for the excluded firms were, in order of frequency, for inventory, depreciation/capitalization, and revenue recognition. Thus, we conclude that our data requirements did not result in an endogenous sample selection bias.
Second, we compare other characteristics of our sample firms and the excluded firms to ensure that these two groups of firms are not systematically different on other dimensions. Specifically, we compare asset sizes, leverage ratios, and the proportion of borrowers reporting small and large losses before the effect of the accounting changes. None of these characteristics was significantly different between our sample and the excluded borrowers. Furthermore, we confirmed that these characteristics did not differ between the sample and excluded borrowers after partitioning on the accounting changes' effect on net income.
Descriptive Statistics
Table 2 provides descriptive evidence on our sample firms' accounting method changes. The sample is split fairly evenly between income-increasing and income-decreasing accounting changes, with 67 increases and 58 decreases. The most common change is in the method of accounting for inventory, which represents 42 percent of our sample. Of the 53 inventory accounting changes, 41 are changes from LIFO to some other inventory accounting method, 6 are changes from FIFO to LIFO, and 6 are other changes in inventory methods, such as changes in overhead allocation. Changes in capitalization of assets, depreciation, and revenue recognition methods each represent roughly 15 percent of the voluntary accounting changes in our sample. Nearly two thirds of the changes in capitalization reduced income, whereas the split between increases and decreases is fairly even for depreciation changes and changes in revenue recognition as well as for the sample as a whole.
V. RESULTS
Univariate Analysis
Table 3 presents univariate analyses of the direction and (signed) magnitude (scaled by assets before the effect of the change) of the income effect of the accounting method change. Panel A of Table 3 presents the actual and expected (assuming independence) number of income-increasing vs. income-decreasing accounting changes for borrowers whose debt contracts allow voluntary accounting changes to affect calculations vs. those that do not. We find that 75 borrowers have at least one debt contract that allows voluntary changes to affect contract calculations, and 50 of our borrowers do not have any bank debt contracts that allow voluntary accounting changes to affect calculations. The accounting method change increases income for 61 percent (46/75) of borrowers whose debt contracts allow the changes to affect calculations vs. only 42 percent (21/50) of borrowers whose contracts do not allow voluntary accounting changes to affect the calculations. The Chisquare statistic for independence of these two characteristics is 4.51, which is significant at the 4 percent level.
Panel B of Table 3 presents the actual and expected (assuming independence) rank sum of the magnitude of the income effect of the accounting method change for borrowers whose debt contracts allow voluntary accounting changes to affect calculations vs. those that do not. We find that when accounting method changes affect the contract calculations, borrowers make accounting changes that more positively affect income. We perform a Wilcoxon rank sum test of the distribution of the income effect of the accounting change for borrowers with debt contract calculations affected by accounting method changes vs. those without, and reject independence of the distribution of the income effect of the accounting change across these two types of borrowers at the 0.008 level.
Table 4 presents the means and standard deviations of our independent variables, partitioned by the direction in which the accounting change affects income, as well as t-statistics for the differences between the means of these two groups. Several of the debt-contracting variables differ significantly between borrowers making income-increasing (vs. income-decreasing) accounting changes. Voluntary changes in accounting methods affect debt contract calculations for 69 percent of the firms that make income-increasing accounting method changes, compared to only 50 percent of those that make income-decreasing changes. Borrowers that make income-increasing accounting method changes are more likely to face dividend restrictions affected by those changes than are borrowers who make income-decreasing accounting changes.
Table 4 also reports that some of our control variables differ significantly between borrowers who make income-increasing vs. income-decreasing changes. Borrowers who pay their executives bonuses based on accounting performance are more likely to make income-increasing accounting method changes. An income-increasing accounting method change is less likely to be made by an unlisted firm that reports a small loss, or by a borrower that reports a large loss, especially if the firm changed its CEO in the year of the accounting change.
Correlations among Independent Variables
Fields et al. (2001) suggest that a limitation of previous accounting choice studies is that they typically examine only one potential motive for making accounting choices, without controlling for other possible motives. The correlations among our independent variables reported in Table 5 provide evidence on the extent of the "correlated omitted variable" problem associated with focusing on a single motive.
In general, we observe no significant correlations between our debt-contracting variables (ACC_Change, NoPPorDiv_Res, PP, Div_Res, and One_Lender) and other factors that might influence the accounting choice decision (COMP, SMLOSS, SMLOSS_NL, LGLOSS, LGLOSS_[DELTA]CEO, NOL, SIZE, NLIST, and [DELTA]CEO). In particular, whether a debt contract allows accounting method changes to affect contract calculations (ACC_Change) is not significantly correlated with any of the control variables. The second column of Table 5 does show, however, that when accounting changes affect contract calculations and no accounting-based performance-pricing or dividend restrictions exist (i.e., NoPPorDiv_Res = 1), borrowers are less likely to pay their executives a bonus based on accounting performance and are more likely to report large losses. These correlations suggest that it may be important to control for management compensation and incentives to meet earnings benchmarks when we test our debt-contracting hypotheses.
Many of our independent variables are correlated with borrower size. Specifically, larger borrowers are more likely to have accounting-based dividend restrictions that are not combined with performance pricing, and are more likely to pay their executives accounting-based performance bonuses. Meanwhile, larger borrowers' debt contracts are less likely to allow accounting changes to affect contract calculations when there are no dividend restrictions or performance pricing. Larger borrowers are also less likely to report a large loss before the effect of the accounting change, or an NOL, and are less likely not to list their shares on a major exchange. Not surprisingly, borrowers with NOLs are more likely to report losses, either large or small; are more likely to experience CEO turnover; and are less likely to pay their executives a bonus based on accounting performance.
Regression Analysis
Table 6 reports the results of our logistic regressions. Model (1), which we use to test Hypotheses 1 and 2, has explanatory power of 24.3 percent. This model correctly classifies 81.6 percent of the accounting method changes as increasing or decreasing income, compared to a naive prediction accuracy rate of 55.4 percent if we assume that all accounting changes increase income. The second model, which we use to test Hypotheses 3 and 4, has slightly higher explanatory power at 27.8 percent and correctly classifies 84 percent of the accounting changes as either income-increasing or income-decreasing. (8)
The significantly positive coefficient on ACC_Change in Model (1) indicates that borrowers that change accounting methods are more likely to make changes that increase income when the changes affect debt contract calculations. Following Amemiya (1981, 1488), we convert the estimated coefficient to a change in probability by multiplying the estimated logit coefficient by 0.25. Borrowers are 39 percent (1.57 x 0.25) more likely to make income-increasing (rather than income-decreasing) accounting changes when their debt contracts allow the accounting changes to affect contract calculations. (9) This supports our first hypothesis, that debt contracting is an important consideration in borrowers' accounting method change decisions.
The significant negative coefficient on One_Lender is consistent with our second hypothesis that borrowers that change accounting methods are less likely to make income-increasing accounting changes when the expected costs of technical default are lower, as should be the case when all of their bank debt contracts are from a single lender.
Model (2) partitions ACC_Change into four mutually exclusive subgroups to provide evidence on our third hypothesis (that performance pricing provides additional incentives beyond those provided by traditional covenants) and our fourth hypothesis (that the incentives provided by covenants restricting dividend payments, either directly or indirectly, may differ from those provided by other covenants). The significantly positive coefficient on PP indicates that borrowers that make accounting changes are 59 percent (2.34 x 0.25) more likely to make income-increasing accounting method changes when the changes not only affect their covenants, but also affect their interest rates through performance-pricing provisions. This supports our third hypothesis. The coefficient on Div_Res indicates that borrowers are 46 percent (1.84 x 0.25) more likely to make income-increasing (rather than income-decreasing) accounting method changes when the changes affect their dividend restrictions. This is consistent with our fourth hypothesis. The coefficient on PP&Div_Res indicates that borrowers with both performance-pricing and dividend restrictions are 50 percent (2.01 x 0.25) more likely to report income-increasing accounting method changes than are borrowers with neither performance-pricing nor dividend restrictions. However, the coefficient on this variable is not significantly different from the coefficients on PP or on Div_Res. The insignificant coefficient on NoPPorDiv_Res indicates that we find no evidence that borrowers without performance-pricing or dividend restrictions were more likely to report income-increasing accounting method changes.
We also find some evidence that management compensation incentives influence the choice between income-increasing and income-decreasing accounting method changes. Specifically, when managers receive accounting-based performance bonuses (COMP) they are more likely to make income-increasing changes. In addition, consistent with a "big bath" story, borrowers with new CEOs who report large losses before the effects of the accounting method changes (LGLOSS_[DELTA]CEO) are more likely to report income-decreasing accounting changes than income-increasing changes.
We also find some evidence that noncontracting motives to manage earnings affect borrowers' decisions to make income-increasing vs. income-decreasing accounting method changes. Specifically, we find borrowers with shares listed on a major exchange expecting to report a small loss prior to the accounting method change (SMLOSS) are more likely to report an income-increasing change, while those not listed on a major exchange are more likely to report an income-decreasing change (SMLOSS_NL). This is consistent with the Beatty, Ke, and Petroni (2002) conclusion that exchange-listed firms are more likely to manage earnings to beat earnings benchmarks than are firms that are not listed on a major exchange.
Finally, we find some evidence that external parties affect accounting choices. Specifically, borrowers with NOL carryforwards who make accounting changes affecting taxes are more likely to make income-increasing accounting changes than borrowers without such carryforwards.
VI. SENSITIVITY ANALYSIS
Signed Magnitude of Income Effect
The tests of our hypotheses reported in Table 6 examine the sign of the income effect of the accounting method change but not the magnitude of the effect. In Table 7, we report the results of rank regressions of the signed magnitude of the income effect of the accounting change, scaled by assets before the effect of the change. The results are largely consistent with those from our logistic analyses. As our first hypothesis predicts, borrowers make accounting changes that have more positive effects on income when the accounting changes affect their debt contract calculations. Consistent with our second hypothesis, borrowers make accounting changes that have a less positive effect on income when the accounting changes affect their debt contract calculation, but the costs of covenant violation are lower because all of their bank debt contracts are with a single lender. With respect to our third and fourth hypotheses, borrowers whose bank debt has performance-pricing and dividend restrictions make accounting changes that have a more positive effect on income and, in fact, these appear to be the two most significant debt-contracting factors explaining the magnitude of the income effect. The results on our control variables largely conform to the results from our previous analyses. However, the performance-based executive bonus compensation variable (COMP) is insignificant in our rank regressions.
We examine the sensitivity of our rank regression results to two alternative specifications. First, we re-estimate the regressions without ranking the data. After winsorizing the dependent variable for three outliers that fall more than 5 standard deviations from the mean, we obtain results similar to those in Table 7. We also obtain similar results when we re-estimate the rank regressions using the unscaled income effect of the accounting method change as the dependent variable.
Debt-Contracting Variables
We consider all bank debt contracts when we construct our debt-contracting variables. However, these variables do not incorporate covenants on non-bank debt. Since covenants on non-bank debt typically allow accounting changes to affect contract calculations (see Smith and Warner 1979), we include a dichotomous variable (PUB_COV) equal to 1 for firms with covenants on non-bank debt. When we include PUB_COV as another debt-contracting variable its coefficient is insignificant and our other results remain similar. We also include another variable equal to 1 if the borrower has public debt and accounting changes do not affect bank-debt calculations, 0 otherwise (PUB_COV x [1 - ACC_COV]). Again, the coefficient on this variable is insignificant, indicating that covenants on public debt do not appear to affect whether borrowers' accounting changes are income-increasing vs. income-decreasing, regardless of whether these accounting changes affect bank debt calculations. Our other results remain similar.
Our debt-contracting variables capture the average effect of accounting-based covenants on the decision to record an income-increasing vs. an income-decreasing accounting change. However, this relation is likely to differ cross-sectionally, reflecting differences in expected cost of future violations. Since our One_Lender variable measures these differences only indirectly, we considered an alternative (indirect) proxy as an additional debt-contracting variable in our analyses. Following El-Gazaar and Pastena (1991), we consider the number of covenants as a measure of tightness of covenant restrictions. The number of covenants is not significant in our models. This may indicate either that the initial covenant tightness is not an important determinant of the subsequent accounting choice decision, or that the number of covenants is not a good proxy for covenant tightness.
VII. CONCLUSIONS
Fields et al. (2001, 301) argue that "research on accounting choice addresses the fundamental question of whether accounting matters." However, they also state that "the evidence on whether accounting choices are motivated by debt covenant concerns is inconclusive." By analyzing contract-specific details, we provide sharper evidence on the importance of debt contracts in borrowers' accounting choices. Specifically, our tests capitalize on the facts that (1) many recent debt contracts do not allow borrowers to use voluntary accounting method changes to affect contract calculations, (2) debt contracts often include performance pricing features as well as debt covenants, and (3) debt contracting decisions may affect accounting choices of borrowers who are not approaching covenant violations.
We find that borrowers that make accounting method changes are more likely to make income-increasing changes when their debt contracts allow these changes to affect contract calculations. This increase in likelihood of an income-increasing accounting change is lower when the borrower expects costs of technical violation to be lower because all of the borrower's bank debt is from a single lender. These results, which hold even after controlling for other motives for changing accounting methods (such as executive compensation incentives, incentives to meet earnings benchmarks, and incentives to reduce taxes), suggest that debt contracts influence borrowers' accounting choices. We also find that borrowers that voluntarily change their accounting methods are more likely to make income-increasing changes if their debt contracts include accounting-based performance-pricing or dividend restrictions. These results suggest that incentives to lower interest rates through performance pricing or to retain dividend payment flexibility influence borrowers' accounting method choices, thereby providing indirect evidence addressing the Fields et al. (2001) fundamental questions of whether, under what circumstances, and how accounting choice matters.
Our focus on contract-specific details for borrowers who need not be approaching covenant violations creates two important data limitations. First, our tests do not include borrowers that do not file their bank debt contracts with the SEC because contract-specific information is unavailable for these borrowers. This could affect the generalizability of our results if the relation between debt contracting and accounting choice differs for these borrowers. To mitigate this concern we provide evidence that these borrowers made voluntary accounting changes that are similar to those made by our sample borrowers. In addition, we document that the excluded borrowers and sample borrowers are similar along a number of dimensions including firm size, income before the accounting change, and leverage.
Second, we cannot directly test whether an accounting change (1) reduced the interest rate charged on the loan based on the performance pricing grid, (2) increased the borrower's flexibility to pay dividends, or (3) reduced the likelihood the borrower would violate a covenant. Since our sample includes borrowers that may be making accounting changes not only to affect the current debt contracting calculations, but also future calculations, we would need to estimate the effect of their accounting change on the interest rate, dividends paid, and covenant violations both in the quarter of the accounting change, and in all future quarters that could be affected by the accounting change as well. Since most borrowers have equity-based covenants and dividend restrictions, this means an accounting change would affect all future calculations until the debt matures. In addition, it is difficult to develop such estimates using only publicly available current period data. For example, when we use balance sheet data available in the financial statements, the ratios we calculate are not equal to the ratios used in the debt contracts. For all these reasons, we leave it to future research to quantify the effects of borrowers' voluntary changes in accounting methods on their interest rates, dividend payments, and probability of covenant violations.
TABLE 1
Sample Selection Process
Borrowers Making Material Voluntary Accounting Method Changes during
1995-2000
Firms that filed "preferability letters" with the SEC (a) 296
Less:
Firms that made accounting changes that had no effect
on balance sheet or income statement accounts (b) 48
Financial institutions 5
Firms that made mandatory accounting changes improperly
classified as voluntary changes 2
Firms with no bank debt or with immaterial bank debt 65
Nonfinancial firms with material bank debt that made
material voluntary accounting changes 176
Less:
Firms with bank debt that exceeds the SEC's materiality
threshold of 10 percent of assets, but do not report filing
the debt contract with the SEC in their 10-K exhibit list 24
Firms that indicated in their 10-K exhibit list that bank debt
was filed with the SEC, but we could not find the contract
on EDGAR (c) 27
Firms in our sample 125
(a) When a firm makes a nonmandated accounting change, the SEC requires
the firm's independent accountant to issue a "preferability letter"
commenting on the preferability of the new accounting method relative
to the prior method.
(b) Examples include changes in the format of the income statement,
changes in the definition of a cash equivalent, or immaterial changes
in accounting methods.
(c) Although most of the indicated filings are available on global
access, we cannot confirm whether the debt contracts were filed or
examine filed contracts because our subscription does not allow us to
view most of these filings.
TABLE 2
Accounting Method Changes Partitioned by Effect on Income for Borrowers
Making Voluntary Accounting Method Changes during 1995-2000
Type of Accounting Change Total Income-
Increasing
Capitalizing to expensing costs or expensing 19 7
to capitalizing
Depreciation policies 18 10
Gains on pension assets 4 4
Goodwill 5 0
Inventory accounting--from LIFO 41 23
Inventory accounting--from FIFO 6 4
Inventory accounting--Other 6 6
Revenue recognition 14 6
Other 12 7
Total 125 67
Type of Accounting Change Income-Decreasing
Capitalizing to expensing costs or expensing 12
to capitalizing
Depreciation policies 8
Gains on pension assets 0
Goodwill 5
Inventory accounting--from LIFO 18
Inventory accounting--from FIFO 2
Inventory accounting--Other 0
Revenue recognition 8
Other 5
Total 58
TABLE 3
Tests of Independence between Whether Accounting Changes Affect
Contract Calculations and the Sign and Magnitude of the Effect of the
Accounting Change on Income
Panel A: Contingency Table of Actual and (Expected) Number of Firms
Making Accounting Changes, Testing for Independence between Whether
Accounting Changes Affect Contract Calculations and Whether Accounting
Change is Income-Increasing vs. Income-Decreasing
Accounting Method
Changes Affect
Contract Calculations
Income-increasing 46
accounting change (40.2)
Income-decreasing 29
accounting change (34.8)
Total 75
Test of independence Chi-square 4.51
Accounting Method
Changes Do Not Affect
Contract Calculations Total
Income-increasing 21
accounting change (26.8) 67
Income-decreasing 29 58
accounting change (23.2)
Total 50 125
Test of independence p-value 0.04
Panel B: Wilcoxon Rank Sum Test of Independence of Distribution of
Income Effect of Accounting Method Change for Borrowers with Debt
Contract Calculations Affected or Not Affected by Accounting Method
Changes
Accounting Method
Changes Affect
Contract Calculations
Actual rank sum 69.42
Expected rank sum (63)
Test of independence Z-statistic 2.42
Accounting Method
Changes Do Not Affect
Contract Calculations
Actual rank sum 53.36
Expected rank sum (63)
Test of independence p-value 0.008
TABLE 4
Mean and (Standard Deviation) of Independent Variables Partitioned by
the Effect of the Accounting Method Change on Income
Income-
Predicted Increasing
Sign of Mean
Variable Difference (std. dev.)
Debt Contracting Variables
+ 0.69
Acc_Change (0.46)
+ 0.09
NoPPorDiv_Res (0.29)
+ 0.13
PP (0.34)
+ 0.25
Div_Res (0.44)
+ 0.21
PP&Div_Res (0.41)
[+ or -] 0.09
One_Lender (0.29)
Control Variables
+ 0.57
COMP (0.50)
+ 0.18
SMLOSS (0.39)
[+ or -] 0.03
SMLOSS_NL (0.39)
- 0.19
LGLOSS (0.40)
- 0.02
LGLOSS_[DELTA]CEO (0.40)
+ 0.12
NOL (0.33)
[+ or -] 13.51
SIZE (1.89)
- 0.15
NLIST (0.36)
- 0.08
[DELTA]CEO (0.27)
Number of Firms (a) 67
Income-
Decreasing Difference
Mean Mean
Variable (std. dev.) (t-statistic)
Debt Contracting Variables
0.50 0.19
Acc_Change (0.50) (2.15) **
0.19 -0.10
NoPPorDiv_Res (0.40) (-1.63)
0.07 0.06
PP (0.26) (1.22)
0.14 0.11
Div_Res (0.44) (1.65) *
0.10 0.11
PP&Div_Res (0.31) (1.64) *
0.17 -0.08
One_Lender (0.38) (-1.38)
Control Variables
0.34 0.23
COMP (0.48) (2.53) ***
0.16 0.02
SMLOSS (0.37) (0.35)
0.12 -0.09
SMLOSS_NL (0.37) (-1.89) *
0.31 -0.12
LGLOSS (0.47) -(1.63) *
0.12 -0.10
LGLOSS_[DELTA]CEO (0.47) (-2.41) ***
0.16 -0.04
NOL (0.37) (-0.58)
13.18 0.33
SIZE (1.83) (1.00)
0.36 -0.21
NLIST (0.48) (- 2.81) ***
0.18 - 0.10
[DELTA]CEO (0.38) (-1.59) *
Number of Firms (a) 58
***, **, and * Indicate significance at the 1 percent, 5 percent, and
10 percent levels, respectively, for one- or two-tailed tests as
appropriate.
(a) We could not obtain proxy statements for four borrowers, so there
are only 121 observations for the COMP, LGLOSS_[DELTA]CEO, and
[DELTA]CEO variables.
Variable Definitions
Debt Contracting Variables:
ACC_Change = 1 if contract allows accounting changes to affect
calculations, 0 otherwise;
NoPPorDiv_Res = 1 if the contract allows accounting changes to
affect calculations but contains neither an
accounting-based performance pricing provision nor
an accounting-based dividend restriction, 0
otherwise;
PP = 1 if the contract allows accounting changes to
affect calculations and contains an accounting-
based performance pricing provision but not an
accounting-based dividend restriction, 0 otherwise;
Div_Res = 1 if the contract allows accounting changes to
affect calculations and contains an accounting-
based dividend but not an accounting-based
performance pricing provision, 0 otherwise;
PP&Div_Res = 1 if the contract allows accounting changes to
affect calculations and contains both an
accounting-based performance pricing provision and
an accounting-based dividend restriction, 0
otherwise; and
One_lender = 1 if accounting method changes affect debt contract
calculations and a single lender makes the loan, 0
otherwise.
Control Variables:
COMP = 1 if the firm indicates that the managers' bonuses
depend on financial performance and the CEO
received a bonus in the year prior to the
accounting change, 0 otherwise;
SMLOSS = 1 if the firm's earnings, before the effect of the
accounting change, is between 0 and -1 percent of
assets, 0 otherwise;
SMLOSS_NL = 1 if the firm's earnings, before the effect of the
accounting change, is between 0 and -1 percent of
assets and the firm is not listed on a major stock
exchange, 0 otherwise;
LGLOSS = 1 if the firm's earnings, before the effect of the
accounting change, is less than -1 percent of
assets, 0 otherwise;
LGLOSS_[DELTA]CEO = 1 if the firm's earnings, before the effect of the
accounting change, is less than -1 percent of
assets and the firm's CEO changed in the year prior
to the accounting change, 0 otherwise;
NOL = 1 for borrowers that have a net operating loss
carryforward and make an accounting change that has
a tax effect, 0 otherwise;
SIZE = the log of the firms assets before the effect of
the accounting change;
NLIST = 1 for firms not listed on a major stock exchange, 0
otherwise; and
[DELTA]CEO = 1 if the firm's CEO changed in the year prior to
the accounting change, 0 otherwise.
TABLE 5
Pearson Correlations among Independent Variables
(significance levels in parentheses)
ACC_ NoPPor Div_ PP&Div_
Change Div_Res PP Res Res
ACC_Change 1.00
NoPPorDiv_Res 0.33 1.00
(0.00)
PP 0.28 -0.14 1.00
(0.00) (0.13)
Div_Res 0.41 -0.19 -0.17 1.00
(0.00) (0.02) (0.06)
PP&Div_Res 0.36 -0.17 -0.15 -0.22 1.00
(0.00) (0.05) (0.10) (0.01)
One_Lender 0.31 0.27 -0.13 0.23 0.03
(0.00) (0.00) (0.15) (0.00) (0.76)
COMP 0.01 -0.23 -0.00 0.12 0.12
(0.94) (0.01) (0.99) (0.19) (0.19)
SMLOSS 0.02 0.01 -0.01 -0.06 0.10
(0.80) (0.92) (0.89) (0.48) (0.29)
SMLOSS_NL 0.04 -0.02 0.11 -0.06 0.05
(0.67) (0.82) (0.23) (0.49) (0.60)
LGLOSS 0.09 0.20 0.11 -0.15 0.00
(0.31) (0.02) (0.23) (0.10) (0.98)
LGLOSS_[DELTA]CEO 0.09 0.19 0.12 -0.13 -0.02
(0.34) (0.04) (0.18) (0.15) (0.85)
NOL 0.04 0.05 0.09 -0.08 0.02
(0.67) (0.60) (0.30) (0.37) (0.84)
SIZE -0.04 -0.24 -0.07 0.24 -0.04
(0.67) (0.01) (0.43) (0.01) (0.70)
NLIST -0.02 0.04 -0.01 -0.10 0.05
(0.80) (0.64) (0.88) (0.26) (0.56)
[DELTA]CEO 0.06 0.08 0.11 0.00 -0.09
(0.51) (0.41) (0.22) (0.98) (0.34)
One_ SM SM LG
Lender COMP LOSS LOSS_NL LOSS
ACC_Change
NoPPorDiv_Res
PP
Div_Res
PP&Div_Res
One_Lender 1.00
COMP 0.03 1.00
(0.76)
SMLOSS -0.04 0.05 1.00
(0.63) (0.55)
SMLOSS_NL -0.01 -0.01 0.62 1.00
(0.88) (0.90) (0.00)
LGLOSS 0.06 -0.09 -0.26 -0.16 1.00
(0.53) (0.33) (0.01) (0.07)
LGLOSS_[DELTA]CEO 0.10 -0.19 -0.12 -0.07 0.46
(0.27) (0.04) (0.20) (0.44) (0.00)
NOL -0.01 -0.18 0.00 0.16 0.31
(0.89) (0.04) (0.92) (0.07) (0.00)
SIZE -0.39 0.21 0.10 -0.10 -0.22
(0.00) (0.02) (0.29) (0.27) (0.01)
NLIST 0.06 -0.16 0.19 0.49 0.14
(0.53) (0.07) (0.04) (0.00) (0.11)
[DELTA]CEO 0.16 -0.11 -0.03 0.10 0.25
(0.07) (0.72) (0.72) (0.27) (0.01)
LGLOSS_
[DELTA]CEO NOL SIZE NLIST
ACC_Change
NoPPorDiv_Res
PP
Div_Res
PP&Div_Res
One_Lender
COMP
SMLOSS
SMLOSS_NL
LGLOSS
LGLOSS_[DELTA]CEO 1.00
NOL 0.27 1.00
(0.00)
SIZE -0.20 -0.16 1.00
(0.02) (0.08)
NLIST 0.16 0.10 -0.34 1.00
(0.08) (0.28) (0.00)
[DELTA]CEO 0.70 0.28 -0.17 0.20
(0.00) (0.00) (0.35) (0.03)
Variables are defined in Table 4.
TABLE 6
Coefficients and (t-statistics) from Logit Regressions of the Sign of
the Income Effect of Voluntary Accounting Changes on Debt Contracting
and Control Variables Sample of 64 Income-Increasing and 57
Income-Decreasing Voluntary Accounting Method Changes, 1995-2000
Model (1) Model (2)
Predicted Coefficient Coefficient
Variable Sign (t-statistic) (t-statistic)
2.36 2.92
Intercept + (1.28) (1.49)
Debt Contracting Variables
1.57
Acc_Change + (3.19) ***
0.20
NoPPorDiv_Res + (0.26)
2.34
PP + (2.42) ***
1.84
Div_Res + (2.72) ***
2.01
PP&Div_Res + (2.52) ***
-2.36 -2.41
One_lender - (-2.99) *** (-2.72) ***
Control Variables
1.31 1.15
COMP + (2.71) *** (2.32) ***
1.69 1.96
SMLOSS + (1.87) ** (2.14) **
-5.13 -5.91
SMLOSS_NL - (-3.23) *** (-3.42) ***
-1.01 -0.96
LGLOSS - (-1.69) ** (-1.55) *
-2.04 -2.59
LGLOSS_[DELTA]CEO - (-1.62) * (-1.78) **
1.14 1.29
NOL + (1.49) * (1.62) *
0.24 0.28
SIZE [+ or -] (1.74) * (1.89) *
Pseudo [R.sup.2] 24.3% 27.8%
Percent Correctly
Predicted 81.6 84.0
***, **, and * Indicate significance at the 1 percent, 5 percent, and
10 percent levels, respectively, for one- or two-tailed tests as
appropriate.
Variables are defined in Table 4.
TABLE 7
Coefficients and (t-statistics) from Rank Regressions of the Effect of
Voluntary Accounting Changes on the Income Effect of the Accounting
Change (Scaled by Assets Before the Effect of the Change) on
Contracting and Control Variables Sample of 64 Income-Increasing and 57
Income-Decreasing Voluntary Accounting Method Changes, 1995-2000
Model (1) Model (2)
Predicted Coefficient Coefficient
Variable Sign (t-statistic) (t-statistic)
50.30 53.14
Intercept + (5.79) *** (5.99) ***
Debt Contracting Variable
ACC_Chang 20.98
+ (3.21) ** --
4.57
NoPPorDiv_Res + -- (0.44)
34.25
PP + -- (3.27) ***
26.25
Div_Res + -- (2.86) ***
15.82
PP&Div_Res + -- (1.70) **
-32.27 -29.57
One_Lender - (-3.04) ** (-2.63) ***
Control Variables
8.58 6.65
COMP + (1.34) * (1.02)
27.56 32.02
SMLOSS + (2.62) *** (3.04) ***
SMLOSS_NL -41.56 -47.60
- (-2.59) ** (-2.97) ***
-0.95 0.64
LGLOSS - (0.11) (0.08)
-20.53 -20.14
LGLOSS_[DELTA]CEO - (-1.45) * (-1.43) *
17.16 16.42
NOL + (1.76) ** (1.71) **
-0.05 -0.09
SIZE [+ or -] (-0.53) (-0.87)
Adjusted [R.sup.2] 15.25% 17.8%
***, **, and * Indicate significance at the 1 percent, 5 percent, and
10 percent levels, respectively, for one- or two-tailed tests as
appropriate.
Variable are defined in Table 4.
We thank two anonymous reviewers, Wayne Guay, Peter Joos, S. P. Kothari, Greg Miller, and seminar participant at MIT for helpful comments. We also thank Jennifer Altamuro, Shelley Herbein, and Hyunna Song for research assistance and Dick Dietrich for his insights about data availability at the SEC. Professor Beatty gratefully acknowledges financial support from PricewaterhouseCoopers and the Sloan Foundation.
Submitted February 2002
Accepted August 2002
(1) Our examination of public debt and private placement agreements that are available online revealed that few of these contracts contained covenants or accounting-based performance pricing.
(2) This hypothesis assumes that the covenants in contracts with single lenders are similar to the covenants in contracts with multiple lenders. We investigate the validity of this assumption, and find that the average number of covenants is virtually the same.
(3) We searched the 10-Ks and 10-Qs filed on the Lexis/Nexis database using the phrase "exhibit 18" and "Letter re Change in Accounting Principles."
(4) Rule 601 of regulation SK allows borrowers some latitude in filing their debt contracts. Specifically, the SEC requires borrowers to file a copy of a debt contract if it is material, where the SEC generally defines materiality as 10 percent of the firm's assets. See Press and Weintrop (1990) for more discussion of the SEC's filing requirements for private debt contracts. We classified firms as having immaterial bank debt if no bank debt filings were listed in their exhibit indexes and the amount of the bank debt disclosed in their footnotes was less than 10 percent of their assets.
(5) We tested the validity of this assumption by conducting exhaustive online searches for debt contracts for each of these firms, and we were unable to locate copies of any of the bank debt outstanding at the time the firm made the accounting change. The inability to find any contracts online supports the assumption that the 24 firms never filed these debt contracts with the SEC.
(6) Although most of the indicated filings are available on the Global Access database, we can neither confirm whether these 27 firms filed nor can we examine any contracts these 27 firms might have filed because our subscription to Thomson Financial does not allow us to view most of these filings.
(7) Our logit regressions require additional data, reducing the sample size to 121 in that analysis.
(8) The sign and magnitude of the effect of the accounting change are the same on both net income and net worth when the firm accounts for changes in accounting methods using either the cumulative effect or prospective method. However, 22 of our sample borrowers use the retroactive approach, where changes in accounting methods have different income statement and net worth effects. Excluding these firms from our analyses does not alter our conclusions.
(9) The conversion factor is obtained by taking the derivative of the predicted probability with respect to a particular independent variable. For a linear probability model that derivative equals the estimated coefficient. For a logit model that derivative equals the estimated coefficient multiplied by the exponential of the predicted probability divided by the square of 1 plus the exponential of the predicted probability.
REFERENCES
Amemiya, T. 1981. Qualitative response model: A survey. Journal of Economic Literature 19 (December): 1483-1586.
Asquith, P., A. Beatty, and J. Weber. 2002. Performance pricing in private debt contracts. Working paper, Massachusetts Institute of Technology, Cambridge, MA.
Beatty, A., I. Dicbev, and J. Weber. 2002. The role and characteristics of accounting-based performance pricing in private debt contracts. Working paper, University of Michigan, Ann Arbor, MI.
--, B. Ke, and K. Petroni. 2002. Earnings management to avoid earnings declines across publicly and privately held banks. The Accounting Review 77 (July): 547-570.
--, K. Ramesh, and J. Weber. 2002. The importance of accounting changes in debt contracts: The cost of flexibility in covenant calculations. Journal of Accounting and Economics 33 (June): 205-227.
Beneish, D., and E. Press. 1993. Costs of technical violation of accounting-based debt covenants. The Accounting Review 68 (April): 233-257.
Burgstahler, D., and I. Dichev. 1997. Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics (December) 24: 99-126.
Dichev, I., and D. Skinner. 2002. Large-sample evidence on the debt covenant hypothesis. Journal of Accounting Research 40 (September).
El-Gazaar, S., and V. Pastena. 1991. Factors affecting the scope and initial tightness of covenant restrictions in private lending agreements. Contemporary Accounting Research 8 (Fall): 132-151.
Fields, T., T. Lys, and L. Vincent. 2001. Empirical research on accounting choice. Journal of Accounting and Economics 31 (September): 255-307.
Healy, P., and K. Palepu. 1990. Effectiveness of accounting-based dividend covenants. Journal of Accounting and Economics 12 (January): 97-124.
Mohrman, M. 1996. The use of fixed GAAP provisions in debt contracts. Accounting Horizons 10 (September): 78-91.
Press, E., and J. Weintrop. 1990. Accounting-based constraints in public and private debt agreements: Their association with leverage and impact on accounting choice. Journal of Accounting and Economics 12 (January): 65-95.
Smith, C., and J. Warner. 1979. On financial contracting: An analysis of bond covenants. Journal of Financial Economics 7 (June): 117-161.
Sweeney, A. 1994. Debt-covenant violations and managers' accounting responses. Journal of Accounting and Economics 17 (May): 281-308.
Watts, R., and J. Zimmerman. 1986. Positive Accounting Theory. Englewood Cliffs, NJ: Prentice Hall.
Anne Beatty Pennsylvania State University Joseph Weber Massachusetts Institute of Technology
In addition, make sure to read these articles:
- Tax Considerations for New Businesses
- Accounting Periods and Methods for Businesses
- Why should you care?
- Notice 98-31 and guidance on involuntary method changes.
- Money Digest - August 5, 2004
- Choice of accounting method for manufacturers.
- Adherence to utility regulations is an accounting method.
- OID deferral opportunities.
- Accounting changes.
- IRS-initiated accounting method changes.
- Metromedia Fiber Network, Inc. Reports Third Quarter1999 Results.
Email
Digg It
del.icio.us
