Mandatorily Conservative Accounting: Evidence and Implications

Transcription

1 Mandatorily Conservative Accounting: Evidence and Implications Alastair Lawrence Richard Sloan Yuan Sun Haas School of Business University of California at Berkeley 2220 Piedmont Avenue Berkeley, CA October 2012 ABSTRACT: A large body of accounting research concludes that various economic incentives induce cross-sectional variation in managers discretionary application of conservative accounting. We extend existing research by presenting evidence suggesting that mandatory accounting rules are also a significant determinant of conservative accounting. For example, accounting rules mandate asset impairments when fair values drop materially below book values. After attempting to model the determinants of mandatorily conservative accounting, we find that some previous variables representing economic incentives become insignificant. KEYWORDS: Mandatory conservatism, conditional conservatism, book-to-market, asset impairment. JEL CLASSIFICATION: M41, C23, D21, and G32. DATA AVAILABILITY: Data are publicly available from sources identified in the article. We have received valuable comments from an anonymous referee, Patricia Dechow, Yongtae Kim, Yaniv Konchitchki, Miguel Minutti-Meza, Panos Patatoukas, Scott Richardson, Jieying Zhang, Jerry Zimmerman (the editor), and workshop participants at Santa Clara University, University of California at Berkeley, and University of Southern California. In addition, we thank Jefferson Duarte for providing the PIN data.

2 1. Introduction Conservatism has long been an important accounting convention. The Accounting Principles Board (APB) in Statement No. 4 (APB 1970, paragraph 171) summarizes the convention of conservatism as follows: Frequently, assets and liabilities are measured in a context of significant uncertainties. Historically, managers, investors, and accountants have generally preferred that possible errors in measurement be in the direction of understatement rather than overstatement of net income and net assets. While conservatism has been a long-standing attribute of financial reporting [e.g., AICPA 1939; Devine 1963; Financial Accounting Standards Board (FASB) 1980], the FASB s recent Conceptual Framework for Financial Reporting (2010) drops conservatism as a desired attribute of financial reporting. 1 In practice, however, conservatism remains a prevalent feature of Generally Accepted Accounting Principles (GAAP), with a prime example being the requirement that most non-financial assets must be written down in bad times but cannot be written up in good times. The conservatism convention has also been a popular topic in academic research (e.g., Watts and Zimmerman 1986; Basu 1997; Givoly and Hayn 2000; Watts 2003a, 2003b; Roychowdhury and Watts 2007; LaFond and Watts 2008; Khan and Watts 2009). A large body of literature documents that variables designed to capture various economic incentives explain cross-sectional variation in conservatism. 2 For example, Ahmed, Billings, Morton, and Stanford- Harris (2002, 868) argue that firms experiencing more severe bondholder-shareholder dividend policy conflicts adopt more conservative accounting and Nikolaev (2010, 138) argues that 1 This change is due to a concern that conservatism is inconsistent with the board s desire for neutrality defined as: without bias in the selection or presentation of financial information. 2 See Ball, Kothari, and Robin (2000), Holthausen and Watts (2001), Ball, Robin, and Sadka (2008), LaFond and Roychowdhury (2008), LaFond and Watts (2008), and Nikolaev (2010). For a more detailed review of the conservatism literature see Watts (2003a, 2003b) and Ryan (2006). 1

3 when debt contracts rely on accounting based covenants, bondholders are likely to provide higher incentives for timely loss recognition to the firm s management and its auditors. Overall, this research concludes that managers exercise discretion over the degree of conservatism employed in their financial reporting in response to various economic incentives. In this paper, we extend prior research by modeling the impact of mandatory accounting rules on conservative accounting. For example, firms are required to write down assets when their fair values drop materially below their book values. 3 Previous research has recognized that conservative accounting consists of both a discretionary and a nondiscretionary or GAAPmandated component (e.g., Ahmed et al. 2002, ; Beaver and Ryan 2005, 302; Givoly, Hayn, and Natarajan 2007, 78, 102). The GAAP-mandated component refers to the component arising from the application of GAAP. We define mandatorily conservative accounting as conservatism resulting from the neutral application of GAAP. 4 We then define discretionary conservatism as conservatism arising from purposeful intervention in the financial reporting process to adjust the amount and timing of conservative accounting. We emphasize that discretionary conservatism can exist because of the subjectivity embedded in many GAAP. If GAAP contain no subjectivity, there can be no discretion. 5 3 As with all of FASB s provisions, impairment standards need not be applied to immaterial items. FASB s conceptual framework (FASB 2010) states that an item is material if omitting it or misstating it could influence decisions that users make on the basis of the financial information of a specific reporting entity. Consistent with prior literature (e.g., Strong and Meyer 1987; Elliott and Shaw 1988; Elliott and Hanna 1996) we highlight material asset write-downs by using total assets or market capitalization as a relative benchmark. Moreover, we acknowledge that the notion of materiality introduces judgement into the write-down process as immaterial write-downs need not be taken. The interactions between write-downs and materiality are similar to those between disclosure and materiality as highlighted in Heitzman, Wasley, and Zimmerman (2010). Heitzman et al. (2010) show that materiality thresholds vary systematically as a function of the earnings response coefficient (ERC) and the magnitude of the item, and hence, we conduct analyses, in Section 2.3 Research design, where we model material write-downs as a function of ERCs and current earnings before write-downs. 4 This definition follows from Schipper s (1989) widely-accepted definition of nondiscretionary accruals as those arising from the neutral operation of GAAP. 5 For example, a firm having $100 in a bank account must report an asset of $100. Management cannot choose to report this asset at $50. 2

4 Our empirical results provide evidence of mandatorily conservative accounting. We show that write-downs are most pervasive in firm-years where firms' market values have dropped below their book values as a result of poor firm performance. We also document that asset writedowns are a nonlinear function of beginning of period book-to-market ratios and recent past operating performance. Our results suggest that mandatorily conservative accounting is a significant determinant of cross-sectional variation in conservative accounting. Moreover, we find that several variables used in the prior literature to capture incentives for discretionary conservatism become insignificant or substantially weakened after we model mandatorily conservative accounting. Specifically, we find that results relating to debt-contracting incentives become statistically insignificant and that results relating to information asymmetry incentives are less pronounced after modeling the circumstances that warrant GAAP-mandated writedowns. These findings raise the possibility that conclusions in previous research regarding the determinants of discretionary conservatism, in particular the debt-contracting determinants, may have been premature. In addition, they highlight the potential importance of controlling for mandatorily conservative accounting when testing for determinants of discretionary conservatism. While our evidence is consistent with a significant role for mandatorily conservative accounting, we emphasize two limitations of our findings. First, our model of mandatorily conservative accounting explains only a small proportion of the observed variation in conservative accounting. This result is consistent with limitations of our model to capture all mandatory conservatism and also with the existence of significant amounts of discretionary 3

5 conservatism. 6 We therefore emphasize that our results are consistent with the existence of significant discretionary conservatism. The issue that we raise with respect to previous research is not with the existence of discretionary conservatism, but with specific inferences regarding the determinants of discretionary conservatism. A second limitation of our findings is that our model of mandatorily conservative accounting may model discretionary conservatism. In other words, it could be the case that firms with impaired assets are not taking write-downs in response to GAAP, but in response to correlated discretionary incentives. We have been unable to uncover any such incentives. One possibility that we considered was that firms with debt contracts could face incentives to take discretionary write-downs on impaired assets in order to appease debtholders. Inconsistent with this explanation, we find that our model of mandatorily conservative accounting similarly explains conservatism in a sample of firms without debt. Our paper relates to several streams of existing research. First, it extends the extant literature on the determinants of write-downs (e.g., Strong and Meyer 1987; Elliott and Shaw 1988; Elliott and Hanna 1996; Francis, Hanna, and Vincent 1996; Rees, Gill, and Gore 1996; Riedl 2004). These prior studies generally concluded that material write-downs were discretionary big baths, as there was limited specific authoritative guidance on the accounting for asset impairments prior to Following the release of SFAS No. 121 Accounting for the Impairment of Long-Lived Assets and for Long-Lived Assets to Be Disposed of (FASB 1995), GAAP for asset write-downs have been formalized. We use GAAP s authoritative guidance to 6 The limitations of our model are illustrated by the Securities and Exchange Commission s challenge to Barnes & Noble s assertion that its goodwill was not impaired despite operating losses during those periods and the significant difference between market capitalization and stockholders equity at the evaluation date. (WSJ 2012) Even though a GAAP-mandated write-down appeared to be warranted, Barnes & Noble was able to defend its position by arguing that our total market capitalization does not indicate the long-term fair value of the company. What is not clear is whether Barnes and Noble s position was motivated by managerial incentives to avoid a writedown or by a superior managerial assessment of fair value. 4

6 model the determinants of mandatory write-downs. Our evidence suggests that a significant proportion of material write-downs can be explained by our model of mandatory write-downs. While our study is a comprehensive attempt at modeling mandatory asset write-downs, some prior research incorporates certain features of our research design. Most notably, Ramanna and Watts (2012) document a positive association between book-to-market ratios greater than one and incentives to avoid asset write-downs. They conclude that the managers of such firms exercise discretion to delay write-downs that are otherwise mandated by GAAP. 7 Also, while not explicitly attempting to model mandatory asset write-downs, some prior studies find that asset write-downs are related to operating performance and the book-to-market ratio (e.g., Strong and Meyer 1987; Francis et al. 1996; Riedl 2004). Second, our findings build on previous research linking book-to-market ratios to conditionally conservative accounting (e.g., Giner and Rees 2001; Beaver and Ryan 2005; Pae, Thornton, and Welker 2005; Roychowdhury and Watts 2007). This prior research documents a positive relation between beginning of period book-to-market ratios and conditionally conservative accounting. Mandatorily conservative accounting is offered as one possible explanation for this relation. We further develop the implications of mandatorily conservative accounting for the relation between book-to-market ratios and conditionally conservative accounting. For example, we predict that this relation will exhibit a significant nonlinearity around a book-to-market ratio of one and will be concentrated in firms with poor operating performance. Our evidence is consistent with these predictions, both corroborating mandatorily 7 Ramanna and Watts (2012) focus on how managerial incentives influence goodwill impairments under SFAS No. 142 (FASB 2001a). Our study more generally models and controls for mandatory asset write-downs across a broader set of asset classes and uses the model to re-examine several previous studies. Nevertheless, the reasoning underlying the research designs employed in the two studies is similar. 5

7 conservative accounting as an explanation for this relation and allowing us to more effectively model mandatorily conservative accounting. Our research also builds on Roychowdhury and Watts (2007) findings that conditionally conservative accounting is positively related to beginning of period book-to-market ratios but not necessarily end of period book-to-market ratios. The beginning of period ratios reflect the likelihood that mandatory write-downs will be required during the period. The end of period ratios do not, because the book values are measured after any such mandatory write-downs have been taken. Thus, it is essential that beginning of period book-to-market ratios are used to model expected mandatory write-downs. Our study also provides new insights into Roychowdhury s (2010) argument that explicit or implicit mechanisms must exist to facilitate managers commitment to conservative accounting policies. Specifically, our results highlight mandatorily conservative accounting under GAAP as one important commitment mechanism. Finally, our study is related to previous research emphasizing the importance of controlling for mandatory accounting rules when evaluating discretionary accounting choices. In particular, Kaplan (1985) and Dechow, Sloan, and Sweeney (1995) emphasize the importance of controlling for nondiscretionary accruals in accrual-based tests of earnings management and Heitzman, Wasley, and Zimmerman (2010) emphasize the importance of controlling for mandatory accounting disclosures in tests of voluntary disclosure. While our analyses focus on write-downs and in turn conservatism, we also highlight the importance of considering the implications of unbiased applications of GAAP and GAAS in assessing discretionary financial reporting practices. Our findings also speak to an apparent conflict in the conservatism literature. In particular, the debt-contracting research (e.g., Nikolaev 2012) argues that debt covenants induce more conservatism while the impairment research (e.g., Beatty and Weber 2006; 6

8 Ramanna and Watts 2012) finds that debt contracts induce less conservatism. Our analyses support the latter findings, indicating that the former findings may instead reflect mandatorily conservative accounting. Our study has three main implications for subsequent research. First, it highlights the importance of mandatory accounting rules in explaining the properties of accounting numbers and their associated relations with stock returns. Second, our study highlights the importance of controlling for mandatorily conservative accounting when examining the determinants of discretionary conservatism. Mandatorily conservative accounting is correlated with economic characteristics such as financial distress, and so may be a correlated omitted variable in previous research. Third, our study provides a framework for modeling mandatorily conservative accounting, demonstrating that it is a nonlinear function of several underlying economic variables. We encourage future research to further develop our model of mandatorily conservative accounting, thus enhancing our ability to model mandatory conservatism and distinguish it from discretionary conservatism. The remainder of this paper is organized as follows. Section 2 describes the study s motivation and research design, while Section 3 describes the data. Our results are presented in Sections 4 and 5, with Section 4 reporting results on mandatorily conservative accounting and Section 5 reporting results on the determinants of discretionary conservatism. Section 6 summarizes additional analyses and Section 7 concludes. 2. Motivation and research design 2.1 Motivation Conservatism is a long-standing attribute of accounting information and a popular topic 7

9 in academic research. 8 A large body of literature hypothesizes that various economic incentives induce cross-sectional variation in managers application of conservatism. For example, this literature argues that managers use conservative accounting to reduce the agency costs of debt (e.g., Ahmed et al. 2002; Watts 2003a; Beatty, Weber, and Yu 2008; Nikolaev 2010) and equity (e.g., Ahmed and Duellman 2007; LaFond and Roychowdhury 2008; LaFond and Watts 2008; Ramalingegowda and Yu 2012), with conservatism serving as a substitute for external monitoring. The underlying argument is that managers are able to ex ante commit to implement relatively conservative accounting as a substitute for loan covenants, higher interest rates, and other shareholder and debtholder imposed monitoring mechanisms and costs. For example, Nikolaev (2010) argues that bondholders are likely to provide more incentives for conservatism when debt contracts rely on accounting based covenants. This body of research generally finds strong supporting evidence, concluding that economic incentives introduce significant crosssectional variation in managers application of conservatism. 2.2 Mandatorily conservative accounting We define mandatorily conservative accounting as conservative accounting that is required under GAAP. The foregoing research does not explicitly control for mandatorily conservative accounting, raising the possibility that it is a correlated omitted variable and in turn, provides an alternative explanation for prior results. Our focus is on conditionally conservative accounting, and specifically the accounting principles requiring firms to write down assets when their fair values drop materially below their book values. 9 We predict that such write-downs will 8 See Watts and Zimmerman (1986), Feltham and Ohlson (1995), Basu (1997), Dechow, Hutton, and Sloan (1999), Beaver and Ryan (2000), Watts (2003b), Ryan (2006), and Khan and Watts (2009) for evidence of conservative accounting. 9 The other major type of conservative accounting is unconditionally conservative accounting, which includes requirements that expenditures such as R&D and advertising be immediately expensed in most cases. 8

10 occur for firms where (i) the book value of assets is above their corresponding market value; and (ii) operating performance is poor. Given the foregoing conditions under which mandated writedowns occur, we also predict that such mandatorily conservative accounting will be systematically related to firm characteristics, such as financial distress and financial leverage. This latter prediction raises our concern that mandatorily conservative accounting is a correlated omitted variable in prior research. Thus, while discretionary conservatism may be relevant, there is another possible explanation for the results in prior research. FASB s Accounting Standards Codification (ASC) 350, Intangibles Goodwill and Other (FASB 2009a) and ASC 360, Property, Plant, and Equipment (PP&E) (FASB 2009b) are two main standards that mandate conservative accounting through the impairment of assets. ASC 350 requires tests for goodwill and indefinite-lived intangibles when it is more likely than not that their fair values are materially less than their carrying amounts. 10 The amount of the impairment is calculated as the difference between the assets fair values and their carrying values. 11 Given the foregoing impairment rules, goodwill and indefinite-lived intangible impairments should theoretically be linear functions of the underlying assets book-to-market ratios after the ratios exceed one. Figure 1, Panel A plots the theoretical relation, absent discretion, between asset impairments and book-to-market ratios for goodwill and indefinitelived intangibles under ASC 350. Unconditionally conservative accounting also impacts the empirical properties of accounting numbers (e.g., Penman and Zhang 2002), but is not the focus of our study. 10 Prior to December 15, 2011, ASC 350 required at least annual impairment tests for goodwill and indefinite-lived intangibles and hence, more discretion is now inherent in goodwill impairment testing post December 14, For more information concerning the impairment of goodwill and indefinite-life intangibles see ASC and ASC , respectively [pre-codification SFAS No. 142 Goodwill and Other Intangible Assets (FASB 2001a)]. It is important to highlight that the foregoing impairment standards need not be applied to immaterial items. 9

11 The impairment standards for PP&E and finite-lived intangibles are slightly different from those for goodwill and indefinite-lived intangibles. While ASC 360 also requires that PP&E and finite-lived intangibles be tested for impairment when circumstances warrant, these assets are only impaired if they are evaluated and found not to be fully recoverable (i.e., their carrying amount exceeds the estimated gross undiscounted cash flows from their use and disposition). 12 However, as with the impairment of goodwill and indefinite-lived intangibles, the amount of impairment for PP&E and finite-lived intangibles is measured as the excess of the assets carrying amounts over their fair values. Under ASC 360, the carrying values of PP&E and finitelived intangibles can therefore exceed their fair values when the carrying value is recoverable, as determined by expected future cash flows from its use and disposition. Thus, the relation between asset impairments and book-to-market ratios for PP&E and finite-lived intangibles is not a simple linear function of the ratio for values greater than one. Instead, the relation between impairments and book-to-market ratios under ASC 360 will be delayed until ASC 360 s recoverability provisions are breached. Figure 1, Panel B plots the relation, again under the assumption of no managerial discretion, between impairments and book-to-market ratios for PP&E and finite-lived intangibles under ASC 360. These two impairment standards for long-lived assets, inter alia, highlight that conservative accounting is mandatory under certain economic conditions. 13 Managers have some discretion with regard to the exact timing and amount of asset write-downs, and hence, a portion of write-downs may be taken earlier than predicted (i.e., when the book-to-market ratio is less 12 For more information concerning the impairment of PP&E and finite-life intangibles see ASC [precodification SFAS No. 144 Accounting for the Impairment or Disposal of Long-Lived Assets (FASB 2001b)]. As with ASC 350, these impairment standards need not be applied to immaterial items. In addition, it is noteworthy to mention that in 2001 SFAS No. 144 superseded SFAS No The lower of cost or market requirements for inventories in ASC (FASB 2009c) would also support similar predictions. 10

12 than one) or later than predicted (i.e., to the right of the expected impairment line when the bookto-market ratio is greater than one). But absent managerial discretion, we would still expect to observe cross-sectional variation in conservative accounting as a function of the economic conditions facing firms. The rationale underlying these predictions is also articulated in Beaver and Ryan (2005) and Ramanna and Watts (2012), who argue that mandatory asset write-downs are warranted for firms with book-to-market ratios greater than one. For example, Ramanna and Watts (2012, 11) argue that they find evidence on the properties of SFAS No. 142 noncompliance among firms with two consecutive years of BTM > 1. In order to model cross-sectional variation in mandatorily conservative accounting, we need measures of both the book values and the fair values of firms assets. The book values are readily available, but the fair values are more difficult to obtain. We address this issue by focusing on the aggregate book values and fair values of firms assets, and we obtain the fair values by summing the market value of firms common equity and the book values of their liabilities. We focus on the book and market values of assets rather than equity because the former allows us to consistently measure the closeness of the book values and fair values of assets on an unlevered basis. 14 Taken together, ASC 350 and 360 imply that for firms with beginning of period book-tomarket ratios close to or greater than one, there will be a negative relation between the ratios and asset write-downs (i.e., higher book-to-market ratios should result in more negative impairment charges to earnings). Note that there are a number of reasons why this negative relation is not expected to hold starting exactly at a book-to-market equal to one. First, the recoverability 14 For example, consider a firm with assets of $11 at book value and $10 at market value. The unlevered book-tomarket is 1.1, correctly indicating that the firm is close to the asset impairment threshold. However, if the firm also has liabilities with a book value of $9, the levered book-to-market is 2.0, incorrectly indicating that the firm is significantly past the impairment threshold. However, as described in Section 6, our findings are robust with respect to using the equity-level book-to-market. 11

13 provisions of ASC 360 provide a buffer before impairments are required even when book-tomarket is greater than one. Second, both ASC 350 and 360 state that the impairment rules need not be applied to immaterial items. Hence, if both the firm and auditor deem an asset impairment to be immaterial, the write-down need not be taken. 15 Third, managers are not required to use their firm s own stock price in determining fair values. While we would expect market values and fair values to be highly correlated, management may assess fair values differently from the market. Fourth, all of the above-mentioned reasons involve subjectivity, providing management with the opportunity to exercise discretion in delaying or accelerating asset impairments. Finally, unconditionally conservative accounting for some assets can cause the numerator of firm-level book-to-market ratios to be understated. Under such circumstances, some assets could require impairment even though the aggregate book-to-market is less than one. 16 In light of these reasons, we also predict that the negative relation between beginning of period book-to-market ratios and asset impairments will be stronger when recent past operating performance has been poor. A poor return on operating assets provides prima-facie evidence that a firm s operating assets are impaired and moreover, all else equal, it increases the likelihood that the impairment is material, given that materiality thresholds are generally lower for weak performing than strong performing firms, and that the write-down will be taken. Thus, mandatorily conservative accounting under GAAP leads us to generate the following three predictions concerning book-to-market ratios and asset impairments: P1: There is a negative and nonlinear relation between beginning of period book-to-market ratios and asset impairments. 15 We model material write-downs as a function of firms ERCs and current earnings following Heitzman et al. (2010) who use a similar methodology to identify material advertising expense disclosures. 16 Beaver and Ryan (2005) provide an extended analysis of the interaction between conditional and unconditional conservatism. 12

14 P2: The negative relation between beginning of period book-to-market ratios and asset impairments is stronger for book-to-market ratios greater than one. P3: The negative relation between beginning of period book-to-market ratios and asset impairments is stronger when recent past operating performance has been poor. Our first prediction is consistent with prior research documenting a positive relation between measures of conservatism and beginning of period book-to-market ratios (e.g., Giner and Rees 2001; Pae et al. 2005; Roychowdhury and Watts 2007). Unlike previous research, however, we attribute this relation to mandatorily conservative accounting warranted by GAAP, leading to the additional predictions that this relation is (i) nonlinear; (ii) stronger for book-tomarket ratios exceeding one; and (iii) stronger for firms with poor recent operating performance. 2.3 Research design To provide empirical evidence on the significance of mandatorily conservative accounting, we first examine the relation between asset write-downs taken during a fiscal year and the book-to-market ratios at the beginning of the fiscal year. Specifically, we regress writedowns on book-to-market ratios to confirm the predicted negative relation (negative as writedown charges are recorded as negative values) using the following specification: (1) For the purpose of estimating Equation (1), we scale WRITEDOWN by the market capitalization at the end of period t-1. We also estimate (1) after substituting special-items (SPI) for WRITEDOWN because Compustat does not separately disclose asset write-downs prior to 2000, instead incorporating them into SPI. The beginning of period asset book-to-market ratio (BTM), is defined as total assets scaled by the sum of market capitalization plus total assets minus common equity at the end of fiscal t-1. 13

15 The rules governing mandatorily conservative accounting lead to the prediction that the relation between write-downs and BTM will be nonlinear in BTM. Specifically, the relation is predicted to be weak when BTM is significantly less than one, to gradually increase as BTM approaches one, and to be a constant linear relation as BTM passes one. Given the limitations of our research design and the extent of managerial discretion in applying mandatorily conservative accounting, it is difficult to ex ante identify the precise functional form. Hence, we use coarse BTM partitions to test for evidence of the predicted nonlinearity. The coarse partitions take the form of seven indicator variables based on the rank of BTM. We choose seven partitions (BTMj, j=1 to 7) because this is the minimum number for which we can divide the sample into similar BTM intervals and still have two subsamples with BTM greater than one. BTM6 and BTM7 are the two BTM dummies of most interest, as they reflect firm-years where the BTM ratio is equal to or greater than one. If write-downs are most prevalent after the market value of assets has dropped below their book value, then WRITEDOWN should be highest for the BTM6 and BTM7 partitions. In turn, we predict that relative to the linear relation modeled in Equation (1) asset write-downs will be relatively larger for observations in the BTM6 and BTM7 partitions and relatively smaller for observations in the BTM1 to BTM5 partitions (being those with BTM values less than one). To provide evidence on nonlinearity, we use the estimated parameters of Equation (1) to generate linear predictions of the mean write-down for each BTM partition. We then compare these linear predictions to each partition s actual mean write-down. If our predictions are correct and the nonlinearity is most pronounced in the region surrounding a BTM of one, then the difference between each partition s actual versus predicted mean write-down should be positive 14

16 (smaller actual write-downs) for those observations in the BTM4 and BTM5 partitions, and negative (larger actual write-downs) for those observations in the BTM6 and BTM7 partitions. Consistent with prior findings (e.g., Strong and Meyer 1987; Francis et al. 1996; Riedl 2004) we also expect that asset impairments will be greater when the firm has experienced a relatively poor recent operating performance. Managers and auditors assessments of assets fair values are likely to be increasing in the underlying assets profitability. Under such circumstances, a BTM greater than one may simply reflect an unrealistically negative assessment on the part of investors or the anticipation of future negative events (e.g., increased competition, new regulations, costly lawsuits, etc.). In either case, managers are unlikely to book a mandatory write-down until the occurrence of these events and the associated deterioration in firm performance. Moreover, as previously mentioned, weak past performance increases the likelihood that the impairment is material, as well as the likelihood of not meeting GAAP s recoverability conditions, and hence, it increases the likelihood that the write-down will have to be taken. Hence, we modify Equation (1) to model write-downs as a function of the BTM and past performance using the following regression specification: (2) We define past performance to be weak if either the average market-adjusted stock returns or the average pre-tax return on total assets are less than an annualized five percent over the last two years. 17 We create an indicator variable, B, that takes the value of one in the presence of weak past performance and zero otherwise. 18 The interaction term, BTM*B, examines the prediction 17 We use a 5% threshold, as this is a reasonable lower bound for the cost of equity capital over our sample period implying that a return of less than 5% indicates that a firm is earning less than its cost of capital. 18 As it is possible that the average pre-tax return on total assets is more likely to proxy for the effects of materiality thresholds on write-downs and the market-adjusted stock returns is more likely to proxy for the effects of the 15

17 that write-downs will be more prevalent in firms with high BTM ratios and a recent history of poor performance. If this prediction is correct, then the coefficient on the BTM*B interaction term will be significantly negative, and the coefficient on the main BTM term will revert toward zero. To examine the nonlinearities between BTM and WRITEDOWN while conditioning on firm performance, we separate observations in each of the seven BTMj (j=1 to 7) partitions into two sets of sub-partitions. For observations in each BTMj partition with B= 0, we form a separate BTMj_G partition reflecting observations with stronger financial performance and for observations in each BTMj partition with B= 1, we form a separate BTMj_B partition reflecting observations with weaker financial performance. As with our analysis of Equation (1), we compare the mean write-down of each partition with its predicted value from Equation (2) to assess the extent of any nonlinearities between BTM and WRITEDOWN. We predict that each respective BTMj_G partition should have less write-downs than its corresponding BTMj_B partition, and hence that nonlinearities should be more prevalent across the BTMj_B partitions. To demonstrate that mandatorily conservative accounting is a determinant of crosssectional variation in conditionally conservative accounting, we modify the Basu (1997) conditional conservatism specification to include BTM and the seven BTM partitions using the following continuous and piecewise BTM interactions, respectively: recoverability provisions, we rerun our main analyses by defining weak past performance using only the average market-adjusted stock returns or only the average pre-tax return on total assets. We find that all the study s main inferences only hold when defining weak past performance using the average market-adjusted stock returns and not the average per-tax return on assets, suggesting that the write-downs appear to be driven by not meeting the recoverability provisions rather than the lower materiality thresholds. We find similar inferences when we model materiality effects using current period pre-tax return on assets before write-downs and ERCs following Heitzman et al. (2010), who propose that materiality thresholds vary systematically as a function of the ERC and the magnitude of the item. We actually find that firms with lower ERCs have higher levels of mandated write-downs than those firms with higher ERCs. 16

18 (3) (4) where E/P is the ratio of earnings per share for period t to price at the beginning of period t and RET is the cumulative annual raw return beginning nine months before the end of fiscal year t. The main coefficients of interest in Equations (3) and (4) are the 7 coefficient on BTM*D*RET in Equation (3), which is predicted to be positive, and the 7,j coefficients on the BTMj*D*RET interaction terms in Equation (4), which are predicted to be increasing in j. Ex-ante, we expect better measures of fit for Equation (4), with the nonlinearity between conditional conservatism and BTM being most pronounced in the region surrounding a BTM of one. As in Equation (2), we also modify Equations (3) and (4) to include past performance as a conditioning determinant of mandatorily conservative accounting using the following regression specifications: (5) (6) 17

19 The coefficients on B*BTM*D*RET in Equation (5), and on BTMj_G*D*RET and BTMj_B*D*RET in Equation (6) reflect the degree of conditional conservatism as a function of BTM and firm operating performance. We predict that conditional conservatism will be the strongest when BTM is greater than one and operating performance is poor. Thus, for Equation (6) we expect the largest coefficients on BTM6_B*D*RET and BTM7_B*D*RET. More generally, mandatorily conservative accounting predicts that the coefficients on each BTMj_B*D*RET dummy will exceed those on the corresponding BTMj_G*D*RET dummy, implying that firms with weaker prior financial performance have higher levels of conditional conservatism than those with stronger financial performance. To examine whether mandatorily conservative accounting is an important correlated omitted variable in prior research examining managerial incentives to engage in discretionary conditional conservatism, we use the following general regression model: (7) Equation (7) modifies the Basu (1997) specification for interactions with the specific incentive in question (INCENTIVE), while controlling for determinants of mandatorily conservative accounting (CONTROLS). The control variables employed correspond to those incorporated in Equations (3) through (6). Section 5 outlines the specific incentives investigated. A significant coefficient on is consistent with the respective incentive inducing managers use of discretionary conservatism. We first estimate Equation (7) without the controls to confirm that we can replicate the findings of previous research. We then include the controls to determine whether mandatorily conservative accounting is an important correlated omitted variable in previous research. 18

20 3. Data 3.1 Sample selection Our empirical tests employ data from four sources. We obtain financial-statement data from the Compustat annual database, stock-return data from the CRSP monthly stock returns database, debt-covenant data from the Mergent Fixed Income Securities Database, and the probability of information-based trading (PIN) data from Duarte and Young (2009). 19 Our sample period covers all firm-years with available data on Compustat and CRSP from 1974 to We start the sample in 1974 because special-items are not widely available prior to this point. We also use shorter sub-periods for some analyses due to data restrictions. In particular, write-down data is only separately available in Compustat beginning in 2000 and the PIN data is only available from 1983 to We require firm-year observations to have stock price, total assets, market capitalization, and book value of common equity data as of the prior year s fiscal year-end. We also require earnings-per-share before extraordinary items, special-items, and annual return data for the current fiscal year. In addition, we require companies to have return on assets and annual stock return data for the previous two fiscal years. We replace positive special-items and missing or positive write-downs with zero values in attempt to isolate conservative accounting practices. For the analyses examining incentive variables (which include leverage, debt issuance, debt covenants, and equity issuance), we replace negative debt or equity issuance and missing issuance values with zero. Moreover, for the debt covenant analysis following Nikolaev (2010) we include firm-years within a 10-year window starting five years prior to and ending five years after debt issuance (the year of the issue is excluded). For all our analyses, we exclude 19 We thank Jefferson Duarte for the PIN data which is available on his website: 19

21 observations in the top and bottom one percent of the financial variables in order to reduce the effects of outliers. 20 After imposing the aforementioned data restrictions we obtain 139,603 firmyear observations for the special-item and the main conditional conservatism sample, 47,259 firm-year observations for the write-down sample, 22,483 firm-year observations for the debt covenant sample, and 31,992 firm-year observations for the PIN score sample. 4. Evidence of mandatorily conservative accounting 4.1 Descriptive statistics Table 1, Panel A presents the descriptive statistics for the key variables in our analyses. The first key variable is BTM, the beginning of period asset book-to-market ratio. Recall that as BTM approaches and exceeds one, the likelihood that a firm will be subject to mandatorily conservative accounting in the form of asset write-downs increases. We therefore measure BTM at the beginning of the year (t-1) and look for evidence of conservative accounting over the subsequent year (year t). The mean (median) BTM ratio is (0.823), indicating that writedowns are not expected for the majority of firms in our sample. The upper quartile, however, is 1.006, indicating write-downs are expected, absent managerial discretion, for about 25 percent of our sample firms. The next key variable is the earnings-to-price ratio (E/P), which is the dependent variable in the Basu (1997) conservatism tests. It is noteworthy that the distribution of E/P is left skewed, with a mean of and a median of This is consistent with the application of conservative accounting to immediately recognize bad news, but not good news, in earnings. Following E/P is the annual stock return (RET), which is a dependent variable in the Basu (1997) conservatism tests. In contrast to E/P, RET is right skewed, with a mean (median) of (0.070). Thus, the asymmetrically strong bad news we observe for E/P is not reflected in 20 All inferences are robust to winsorizing the financial variables at the top and bottom one-percent levels. 20

22 RET, consistent with conservative accounting. We next report our lagged performance dummy variables, L2_RET and L2_ROA. The mean value of L2_RET is 0.575, indicating that 57.5 percent of the sample observations have lagged annualized abnormal stock returns greater than or equal to five percent. The mean value of L2_ROA is 0.498, indicating that 49.8 percent of the sample observations have lagged annualized return on assets greater than or equal to five percent. Recall that we classify a firm as weakly performing, B, if either L2_RET or L2_ROA equals zero which is the case in 75.1 percent of the sample observations, with the remaining 24.9 percent classified as strongly performing. 21 Next, we report descriptive statistics for specialitems (SPI) and write-downs (WRITEDOWN). The mean value of SPI is negative (-0.025), while the median value is zero. This is consistent with SPI reflecting a relatively small number of relatively large write-downs. Similarly, we see that the write-down variable, WRITEDOWN, mean value is negative (-0.018) and is highly left-skewed, as it is characterized by a small number of relatively large write-downs. In unreported tests, we find that the correlation between SPI and WRITEDOWN in the post-2000 period is 81.7 percent, confirming that much of the variation in SPI is attributable to write-downs. Finally, the leverage (LEV), debt issue (DEBT_ISSUE), number of debt covenants (RESTRICT), equity issue (EQUITY_ISSUE), private information (PIN), and negative returns (D) variables have similar distributions to those reported in prior research that uses these variables to explain variation in conservative accounting. Panel B of Table 1 reports the means of the key variables sorted by the BTM partitions. We classify all firm-year observations into seven BTM groups. In choosing the number of groups, we trade off a number of considerations. First, we want to have a sufficient number of groups either side of a BTM of one to observe any nonlinearities around one. Second, we want 21 All inferences are robust to using three-percent thresholds rather than five-percent thresholds and to defining B as one if both L2_RET and L2_ROA equal zero. 21

23 the difference in BTM between groups to be similar, so that we can readily observe any nonlinearities. Third, we want the number of observations within each group to be similar, so that we estimate within-group statistics with similar efficiency. In trading off these considerations, we select BTM intervals of (11,789 observations), (17,401 observations), (23,303 observations), (30,100 observations), (20,862 observations), (21,893 observations), and >1.2 (14,255 observations). We see that as the BTM ratio increases, special-items (SPI) and write-downs (WRITEDOWN) decrease monotonically, consistent with the mandatory application of GAAP s impairment standards. We also present the percentage of firms with special-items (SPI%) and write-downs (WRITEDOWN%) by each BTM partition and highlight that while the percentage of firms with write-downs generally increases monotonically across the BTM partitions (except for BTM5), the percentage of firms with special-items is mixed across the seven BTM partitions. Given the monotonically decreasing results for SPI and the mixed results for SPI%, it appears that several firms take immaterial special-item charges irrespective of their BTM. Moreover, given that only 22.8 and 26.2 percent of firms take write-downs in BTM6 and BTM7, respectively, highlights that GAAP-mandated impairments involve managerial discretion and subjectivity in terms of the timing and the amounts of the write-downs. We also observe a positive association between BTM and LEV, DEBT_ISSUE, RESTRICT, and PIN, and a negative association between BTM and EQUITY_ISSUE. These correlations raise the possibility that mandatorily conservative accounting is a potential correlated omitted variable in previous research examining discretionary determinants of conservative accounting. 22

24 4.2 Documenting mandatorily conservative accounting We begin by examining the relation between asset impairments and BTM. In accordance with mandatorily conservative accounting, we predict that firms with higher beginning of period BTM ratios will have more special-item charges and write-downs. We start in Table 2 by regressing SPI and WRITEDOWN on BTM to check for evidence of the predicted negative relation (negative because charges are recorded as negative numbers). As indicated by the regressions in Columns (1) and (3) of Table 2, the relations are negative and significant (p < 0.01) for both SPI and WRITEDOWN. We further predict that the negative relation will be stronger for poorly performing firms. In order to test this prediction, we use our dummy variable, B, which takes on the value of one whenever either L2_RET or L2_ROA are equal to zero. We estimate the same set of regressions, but add B as an additional main effect and also as an interaction with BTM. For both the SPI and WRITEDOWN regressions, in Columns (2) and (4) respectively, the inclusion of B leads to a large drop in the magnitude of the coefficient on BTM and a significantly negative coefficient on the BTM*B interaction. This result is consistent with our prediction that mandatorily conservative accounting is more prevalent in firms with high BTM ratios and a recent history of poor performance. Under such conditions, it will be difficult for managers to persuade auditors that impairment is not warranted. The results in Table 2 are in line with our prediction that asset impairments will be greater for firms with higher BTM values. But these regressions model asset impairments as a linear relation of BTM, whereas, we predict that the rules governing mandatorily conservative accounting will lead to a nonlinear relation. Specifically, the relation is predicted to be weak when BTM is significantly less than one, to gradually increase as BTM approaches one, and be linear once BTM significantly exceeds one. Given the latitude for managerial discretion in the 23

25 application of mandatorily conservative accounting and the limitations of our research design, it is difficult to specify the exact functional form on a priori grounds. We therefore use our coarse BTM partitions to test for evidence of the predicted nonlinearity. Recall that BTM6 and BTM7 are the partitions for which BTM is greater than one, while BTM4 and BTM5 are the partitions for which BTM is marginally less than or equal to one. We therefore predict that, relative to the linear relations modeled in Table 2, asset impairments will be relatively larger for observations in BTM6 and BTM7 and relatively smaller for observations in BTM4 and BTM5. We do this by first computing the fitted values of WRITEDOWN or SPI by using the estimated coefficients from Equation (1) and the mean BTM for each of the BTM partitions. We then compare the fitted value (Pred. Val) with the mean value of each BTM partitions as shown in Table 3. Table 3, Panel A provides evidence consistent with our expectations that the nonlinearities are pronounced in the region surrounding BTM values of one, as the fitted values of asset impairments are relatively smaller than the mean values for observations in the BTM4 and BTM5 partitions (below one), and are relatively larger than the mean values for observations in BTM6 and BTM7 partitions (above one). For example, the fitted value (Pred. Val) of WRITEDOWN for the BTM5 partition is whereas the group mean is a difference of However, the fitted value (Pred. Val) of WRITEDOWN for the BTM6 partition is whereas the group mean drops dramatically to for a difference of Inferences are similar for the SPI partitions. Figure 2, Panel A plots the foregoing relations for WRITEDOWN (the figure for SPI is similar and omitted for brevity). The X-axis represents BTM values and the Y-axis represents the values of WRITEDOWN. The plot shows both the fitted linear relation from Table 2 and the BTM partition means with the indicated nonlinear relation from Table 3. The plot shows that as the 24