Mandatorily Conservative Accounting:

Evidence and Implications

Alastair Lawrence

lawrence@haas.berkeley.edu

Richard Sloan

richard_sloan@haas.berkeley.edu

Yuan Sun

yuan_sun@haas.berkeley.edu

Haas School of Business

University of California at Berkeley

2220 Piedmont Avenue

Berkeley, CA 94720-1900

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.

1

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).

2

“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 GAAP-

mandated component (e.g., Ahmed et al. 2002, 873-874; 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.

3

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 write-

downs 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 write-

downs. 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

4

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 1995. 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 write-

down or by a superior managerial assessment of fair value.

5

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.

6

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;

7

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

8

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 cross-

sectional 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.

9

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 write-

downs 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 indefinite-

lived 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, 2011. 11

For more information concerning the impairment of goodwill and indefinite-life intangibles see ASC 350-20-35

and ASC 350-30-35-15, 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.”

10

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 finite-

lived 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 360-10-35 [pre-

codification 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. 121. 13

The lower of cost or market requirements for inventories in ASC 330-10-20 (FASB 2009c) would also support

similar predictions.

11

than one) or later than predicted (i.e., to the right of the expected impairment line when the book-

to-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 non-

compliance 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-to-

market 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-to-

market 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.

12

provisions of ASC 360 provide a buffer before impairments are required even when book-to-

market 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.

13

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-to-

market 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 write-

downs on book-to-market ratios to confirm the predicted negative relation (negative as write-

down 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.

14

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

15

(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

16

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 cross-

sectional 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.

17

(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)

18

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.

19

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

2009. 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 2004.

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:

http://www.owlnet.rice.edu/~jd10/publications.htm.

20

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 firm-

year 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 0.802 (0.823), indicating that write-

downs 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 -0.005 and a median of 0.051. 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

0.186 (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.

21

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 special-

items (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.

22

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 0.0-0.3 (11,789 observations), 0.3-0.5 (17,401

observations), 0.5-0.7 (23,303 observations), 0.7-0.9 (30,100 observations), 0.9-1.0 (20,862

observations), 1.0-1.2 (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.

23

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

24

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 -0.028 whereas the group mean is -0.017—a difference

of 0.012. However, the fitted value (Pred. Val) of WRITEDOWN for the BTM6 partition is -0.034

whereas the group mean drops dramatically to -0.044 for a difference of -0.010. 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

25

BTM partitions increase from BTM1 to BTM5, write-downs are relatively small and increase

slowly across the five partitions (remember that asset impairments are recorded in negative

numbers). However, as firms’ book values of assets rise above their market values, write-downs

increase significantly. Specifically, the figure shows that the extent of write-downs spikes from

BTM5 to BTM6 and then again from BTM6 to BTM7. Moreover, the figure highlights Table 3’s

findings, that the nonlinearity imposed by mandatorily conservative accounting is particularly

prevalent when the BTM reaches one. In terms of the magnitudes of write-downs, the actual

write-downs for observations with a BTM above one are lower than those implied by the increase

in BTM outlined in Figure 1. For example, moving from BTM5 (below one) to BTM6 (above

one) an average increase in write-downs, as a percentage of market capitalization, of 8.0 percent

is implied by BTM; however, the mean write-down in BTM6 is 4.4 percent and represents 54.3

percent of the that implied by the change in BTM. This spread widens for BTM7 and highlights

that significant discretion is inherent in the write-down process.

We further investigate these relations using a generalized additive model (GAM)

following Hastie and Tibshirani (1990) that models the nonlinear nonparametric regression of

special-items or write-downs on the BTM ratio. Again for brevity, we only show the figure for

write-downs as inferences are very similar for special-items. Figure 2, Panel B plots the

nonlinear regression relation. The X-axis represents the BTM ratio and the Y-axis represents the

predicted value of write-downs, and the dotted lines represent 95 percent confidence intervals.

The inferences from this plot are similar to that from Panel A. The nonlinear regression plot

shows no significant write-downs until the BTM ratio approaches one and then significant write-

downs are predicted for firms with BTM ratios greater than one. Although similar, the nonlinear

relation is again more pronounced for write-downs than for special-item charges.

26

Panel B of Table 3 presents two tests for nonlinearity in the relation between special-

items/write-downs and BTM. We first report a Vuong (1989) test comparing the goodness of fit

between the linear specification in Equation (1) with the piecewise linear specification from

Table 3 that uses the seven BTM partitions in place of the continuous BTM. The Vuong test

generates negative and significant Z-values for both SPI and WRITEDOWN, implying that the

piecewise model provides a better fit between write-downs and the BTM ratio than the simple

linear model. The second test is an F-test comparing the linear specification of Equation (1) with

the GAM used in Figure 2, Panel B. The F-test also rejects the linear model in favor of the

nonlinear counterpart. Taken together, the foregoing analyses provide support for the idea that

the rules governing mandatorily conservative accounting generate a nonlinear relation between

write-downs and BTM, with the nonlinearity being most pronounced around a BTM of one.

We perform an analysis (untabulated) similar to that of Table 3 but instead compare the

fitted values (Pred. Val) from Equation (2) with the mean value of each BTM-performance

partitions (e.g., BTMj_G and BTMj_B). Recall that BTMj_G reflects the seven BTM partitions

with stronger financial performance and BTMj_B reflects the seven BTM partitions with weaker

financial performance.22

Consistent with our expectations, we find that each respective BTMj_G

partition has lower special-items than its corresponding BTMj_B partition. Specifically, the mean

values of SPI for BTMj_G decrease almost monotonically from -0.004 for BTM1_G to -0.012 for

BTM7_G and from -0.010 for BTM1_B to -0.067 for BTM7_B. Interestingly, the change in

coefficients from BTM1_G through to BTM7_G is fairly marginal, and the extent of special-item

charges is much more prevalent for the BTM partitions with weaker financial performance, as

every BTMj_G partition has fewer special-item charges than the BTMj_B partition with the

22

For example, BTM2_G equals one if firms are in the BTM2 partition and have average market-adjusted returns

and return on assets in the previous two years that are both greater than or equal to five percent, and zero otherwise.

27

lowest level of special-item charges. A similar pattern is observed when comparing the mean

values of WRITEDOWN across the fourteen partitions.

Following the analysis in Table 3, we compute the fitted values of WRITEDOWN and SPI

for each partition by using the estimated coefficients from Equation (2) and the mean BTM for

each of the BTMj partitions. Figure 2, Panel C plots these results for write-downs and

demonstrates the clear difference in write-downs between BTM partitions with strong and weak

financial performance. The plot shows a relatively flat trend of impairments across the BTM

partitions for firms with strong financial performance and a clear increasing trend of charges

across the BTM partitions for firms with weak past performance especially as the BTM ratio

approaches and exceeds one. As in Figure 2, Panel A, the magnitude of write-downs in Figure 2,

Panel C is also lower than that implied by the change in BTM. Specifically, moving from BTM5

(below one) to BTM6 (above one) for weak performers, actual write-downs approximate 58.0

percent of those implied by the change in BTM, and this spread again increases for BTM7.

The nonlinear regression plot of Equation (2) presented in Panel D provides similar

inferences. We perform the two untabulated tests of nonlinearity for Equation (2). The Vuong

test again has negative and significant Z-values for both SPI and WRITEDOWN implying that the

piecewise model of Equation (2) has a superior fit to the linear version of Equation (2).

Furthermore, the F-test between the linear version of Equation (2) and the nonparametric GAM

version of Equation (2) also reveals that the nonlinear nonparametric fit outperforms the linear

fit. Taken together, the preceding analyses confirm our predictions of significant nonlinearities

between write-downs and BTM. Moreover, these nonlinearities are more pronounced for firms

with a BTM close to one and for firms with poor recent financial performance. These findings are

consistent with the impairment requirements of mandatorily conservative accounting.

28

4.3 Mandatorily conservative accounting and conditional conservatism

Our next set of analyses, presented in Tables 4 and 5, examines whether mandatorily

conservative accounting is an important determinant of cross-sectional variation in the popular

Basu (1997) measure of conditionally conservative accounting. This analysis builds on our

previous analysis, because we predict that the conditionally conservative accounting evidenced

for SPI and WRITEDOWN in Tables 2 and 3, and Figure 2 will also be present in the Basu

measure. We begin by comparing conditional conservatism coefficients on D*RET from

Equations (3) and (5)—simple linear specifications using a continuous BTM interaction—with

those obtained using the piecewise nonlinear specifications in Equations (4) and (6). A priori, we

expect that the nonlinearity between conditional conservatism and BTM will be most pronounced

in the region surrounding a BTM of one and for the subsample of poorly performing firms.

Table 4 presents regressions for Equations (3) and (5). Column (1) presents the results for

the original Basu (1997) specification; Column (2) presents the results for the Basu (1997)

specification interacted with BTM; and Column (3) presents the results for the Basu (1997)

specification interacted with BTM and poor past performance (B). Consistent with Basu (1997),

the coefficient on D*RET is positive and significant, consistent with the timelier recognition of

bad news than good news in earnings. However, we observe a significant and negative

coefficient on RET which is inconsistent with Basu (1997) but consistent with more recent

studies (e.g., Ruddock, Taylor, and Taylor 2006; Nikolaev 2010; Patatoukas and Thomas

2011).23

Column (2) demonstrates that the coefficient on D*RET is 0.018 and insignificant once

BTM*D*RET is included in the model. The coefficient on BTM*D*RET, however is 0.701 and

significant, implying that firms with high BTM values exhibit strong conditional conservatism—

23

The significant and negative coefficient on RET is generally found throughout all remaining analyses.

29

consistent with the notion that mandatorily conservative accounting is a key driver of conditional

conservatism. In Column (3) the coefficient on BTM*D*RET is reduced to 0.328 with the

majority of the effects transferring to B*BTM*D*RET, indicating that conditionally conservative

accounting is concentrated in the subset of high BTM firms with poor recent performance. Also

important to note is that the adjusted R-squared increases from 0.099 in the base Basu model to

0.143 with the inclusion of BTM, and to 0.173 with the inclusion of both BTM and past

performance.

Table 5, Column (1) presents the results of Equation (4) where we allow for nonlinearity

by interacting the BTM partitions with D*RET. Consistent with the predicted nonlinearities, the

summed coefficients (Sum Coef.) on BTMj*D*RET are monotonically increasing in each BTM

partition and significantly jump for the BTM6 and BTM7 partitions. These results support the

hypothesis that mandatorily conservative accounting is a determinant of cross-sectional variation

in conditional conservatism. To highlight the specific regions of potential nonlinearities between

conditional conservatism and BTM, we compare the summed coefficients (Sum Coef.) to the

predicted coefficients (Pred Coef.) for each BTM partition. ‘Pred Coef’. is the predicted

coefficient for each BTM partition using the linear parameters from Equation (3) shown in Table

4, Column (2). The differences between ‘Sum Coef.’ and ‘Pred Coef.’ highlight how the

nonlinear effects of mandatorily conservative accounting are particularly evident as the BTM

ratio goes from less than one to greater than one. Specifically, compared to the piecewise model,

the linear model generally overestimates conditional conservatism levels for BTM partitions with

values less than one and underestimates conditional conservatism levels for BTM partitions with

values greater than one. The Vuong test and F-test at the bottom of Table 5 provide evidence that

the piecewise nonlinear specification of Equation (4) and the nonparametric nonlinear

30

specification of Equation (3) are superior to the continuous parametric specification of Equation

(3). While previous research has documented a positive relation between conditional

conservatism and BTM (e.g., Pae et al. 2005; Roychowdury and Watts 2007; Patotoukas and

Thomas 2011) and has alluded to the predicted nonlinearities between conditional conservatism

and BTM (e.g., Ball, Kothari, and Nikolaev 2010), our findings are the first to highlight the

specific nonlinearities in this relation surrounding a BTM equal to one.

We also examine the relation between conditional conservatism and the BTM partitions

conditioning on past financial performance using Equation (6). In unreported analyses, we find

that the coefficients on all three-way interaction terms (BTMj_G*D*RET and BTMj_B*D*RET)

are significantly positive. Moreover, for both stronger financial performers (BTMj_G) and

weaker financial performers (BTMj_B), there is a positive relation between the Basu coefficients

and the BTM partitions. Furthermore, the magnitude of conditional conservatism is greater for

the weaker performers than for the strong performers, as evidenced by the values of the

combined coefficients (Sum Coef.). The differences between ‘Sum Coef.’ and ‘Pred Coef.’

indicate that nonlinearities are present for both the strong and weak performers. The Vuong test

and F-test provide evidence that both the nonlinear fits of Equation (6) and the nonparametric

nonlinear specification of Equation (5) are superior to the continuous parametric fit of Equation

(5) in Table 4. Panels A and B of Figure 3 visually highlight the above findings. The

nonlinearities around a BTM of one are clearly evident in all plots, though they are somewhat

less pronounced than for the comparable write-down plots in Figure 2.

In summary, the foregoing analyses confirm our predictions that (i) there is a nonlinear

relation between conditional conservatism and BTM; (ii) the nonlinearity is most pronounced

around a BTM of one; and (iii) the nonlinear relation is strongest for firms with poor recent

31

financial performance. Moreover, the results in Tables 5 and Figure 4 support our hypothesis that

mandatorily conservative accounting is an important determinant of cross-sectional variation in

conditionally conservative accounting. We next turn to the related question of how controlling

for mandatorily conservative accounting affects inferences regarding the determinants of

discretionary conservatism.

5. Economic incentives and conservatism

5.1 Debt contracting

Numerous studies argue that debt-contracting incentives induce managers to use

discretionary conservatism (e.g., Ahmed et al. 2002; Pae 2007; Beatty et al. 2008; Zhang 2008;

Lara, Osma, and Penalva 2009; Lee 2010; Nikolaev 2010; Wang, Xie, and Xin 2011). The

underlying rationale is that managers commit to using discretionary conservatism to reduce

interest rates, loan covenants, and other bondholder imposed monitoring mechanisms. We test

whether the significance of the debt-contracting demand for conservatism is reduced once

controlling for mandatorily conservative accounting. Following prior research (e.g., Ahmed et al.

2002; Pae 2007; LaFond and Watts 2008; Lara et al. 2009; Lee 2010; Nikolaev 2010; Wang et

al. 2011) we use debt issuance, leverage, and debt covenants as proxies for debt-contracting

incentives. Tables 6 through 8 present the results using debt issuances, leverage, and debt

covenants, respectively.

For comparative purposes, Table 6, Column (1) begins with the results for the original

Basu (1997) specification, which is also presented in Table 4. Consistent with the debt-

contracting demand for conservatism, we find in Table 6, Column (2) that firms issuing debt

report more conservative earnings as the coefficient on DEBT_ISSUE*D*RET is 0.075 and

32

statistically significant (t = 2.78; p < 0.01). However, after controlling for the determinants of

mandatorily conservative accounting in Columns (3), (4), (5), and (6) respectively, the

coefficient on this three-way interaction term decreases and becomes statistically insignificant.24

We omit the controls for mandatorily conservative accounting for brevity. We do, however, note

that the control variables continue to be statistically significant and in all cases are more

statistically significant than the debt issuance interactions. For instance, the coefficient on

BTM*D*RET in Column (3) is 0.689 with a t-statistic of 9.86 and the coefficients on

BTM6*D*RET and BTM7*D*RET in Column (4) are 0.671 and 1.008 with t-statistics of 8.40

and 11.82, respectively.25

In other words, incorporating both debt issuance and the mandatorily

conservative accounting variables leads to insignificance for the former, but has little impact on

the latter.

Table 7 replicates the analysis in Table 6 using leverage in place of debt issuance as a

proxy for the debt-contracting demand for conservatism. We again find support for the debt-

contracting demand for conservatism using leverage before we control for determinants of

mandatory conservatism, as the coefficient on LEV*D*RET is 0.182 and significant (t = 3.24; p

< 0.01). However, once we control for mandatorily conservative accounting, the coefficient

drops sharply and becomes statistically insignificant in all but one specification. Table 8 further

confirms the findings in Tables 6 and 7 using the number of debt covenants (RESTRICT) as a

proxy for the debt-contracting demand for conservatism. In Column (1), we find the coefficient

on RESTRICT*D*RET is 0.013 and statistically significant (t = 3.95; p < 0.01) before controlling

24

Given the number of interaction terms in the specifications presented in Tables 6 through 10, it is important to

consider whether the inferences are influenced by potential collinearity effects. Hence, we check the variance

inflation factors of all specifications. While a few specifications have variance inflation factors above ten, we find

that our main inferences hold in specifications with acceptable levels of collinearity. 25

The statistical significance of the mandatorily conservative accounting variables is similar for the remaining

economic incentive analyses.

33

for determinants of mandatorily conservative accounting—supporting the findings in prior

research (Nikolaev 2010) that covenants in debt contracts are related to the degree of conditional

conservatism. However, after we control for determinants of mandatorily conservative

accounting in Columns (2) through (5), the RESTRICT*D*RET coefficient drops considerably

and becomes statistically insignificant in all specifications. The results in Tables 6 through 8

offer mandatorily conservative accounting as an alternative potential explanation for the debt

contacting results in prior research. Specifically, highly levered firms that are seeking additional

debt financing tend to be poorly performing firms. Thus, the association with conservative

accounting may be attributable to the fact that these firms are coincidently required to take

mandatory asset write-downs under GAAP. It is possible, however, that the variables we use to

model mandatorily conservative accounting instead happen to model discretionary conservatism

relating to debt-contracting incentives. While acknowledging this possibility, we hasten to add

that it is not clear to us these variables would be better at modeling debt-contracting incentives

than variables chosen by prior research for this specific purpose. The unreported correlations

between BTM and DEBT_ISSUE, LEV, and RESTRICT are 0.130, 0.071, and 0.121, respectively,

and are all statistically significant (p < 0.01).

5.2 Equity issuance

To test shareholders’ demand for discretionary conservatism after controlling for

mandatorily conservative accounting, we examine the relation between equity issuance and

conditional conservatism. Table 9, Column (1) shows that the coefficient on

EQUITY_ISSUE*D*RET is 0.432 and significant (t = 3.42; p < 0.01) consistent with

shareholders demanding discretionary conservatism. After controlling for BTM and financial

performance either with continuous variables or partitions, we continue to find that firms with

34

higher equity issuances report more conservative earnings than do those with lower equity

issuances. Specifically, the coefficient on EQUITY_ISSUE*D*RET increases from 0.432 to

0.464 and is even more statistically significant (t = 4.68; p < 0.01) once controlling for partitions

sorted by both BTM and B. Therefore, in this setting controlling for mandatorily conservative

accounting strengthens the findings regarding discretionary conservatism. It appears that firms

issuing equity face strong incentives to take discretionary asset write-downs. In terms of

economic magnitudes, we find, in an untabulated analysis, that discretionary levels of

conditional conservatism are approximately 40.0 percent lower for firms without equity issuance

than for firms with equity issuance.

5.3 Information asymmetry

Lafond and Watts (2008) argue that information asymmetry between inside and outside

shareholders generates managerial incentives for discretionary conservatism in order to constrain

aggressive accounting. LaFond and Watts (2008) argue that higher levels of information

asymmetry, proxied by a larger PIN score, are associated with greater asymmetric recognition of

losses, as they find a significant and positive coefficient for PIN*D*RET. Table 10, Column (1)

confirms the results for the original Basu (1997) specification within this sample as the

coefficient on D*RET is positive and significant (t = 8.78; p < 0.01). Table 10, Column (2)

provides results consistent with LaFond and Watts (2008) as the coefficient on the three-way

interaction PIN*D*RET is 1.395 and significant (t = 6.04; p < 0.01).

Given our findings in the previous analyses that the BTM ratio and financial performance

are joint determinants of mandatory conservatism, we re-examine the relation between PIN and

discretionary conservatism after controlling for mandatory conservatism. The coefficients on

PIN*D*RET in Columns (3) and (4) decrease to 0.838 (t = 3.44; p < 0.01) and to 0.871 (t = 3.65;

35

p < 0.01) after controlling for BTM and the seven BTM partitions, respectively. The coefficients

on PIN*D*RET in Columns (5) and (6) further decrease to 0.768 (t = 3.31; p < 0.01) and to

0.802 (t = 3.51; p < 0.01) when we control for both BTM and past performance, and the

partitions double sorted based on BTM and past performance, respectively. Although the

coefficient on PIN*D*RET remains positive and statistically significant after controlling for non-

discretionary determinants of conservatism, the significance is less pronounced. In terms of

economic significance, in an unreported analysis we find that discretionary levels of conditional

conservatism are 122 percent lower for firms below the median value of PIN relative to firms

above the median value of PIN and specifically, that the coefficient on D*RET for firms below

the median value of PIN is slightly negative and insignificant. Interesting to highlight from Table

10 is that the adjusted R-squared increases more from controlling for determinants of mandatory

conservatism than for information asymmetry incentives. Also, it is noteworthy that all

inferences are similar in Tables 6 through 10 whether we control for the continuous or

partitioning BTM and performance variables. The consistency of the approaches likely results

from the fact the discretionary determinants are modeled as linear relations.

Overall, the preceding findings show that our controls for mandatorily conservative

accounting significantly impact inferences in previous research regarding the determinants of

discretionary conservatism.

6. Additional analyses

To examine the sensitivity of our main results, we perform additional analyses using

alternative specifications, methods, and variable measurements. First, we separate the full

population into sub-samples in attempt to further isolate mandatory and discretionary

conservatism. To the extent that debt-contracting incentives motivate higher levels of

36

conservatism, asset write-downs are predicted to be timelier among firms with debt versus those

without debt. Hence, we run our main analyses separately for firms with debt (117,461 firm-

years) and for firms without debt (22,142 firm-years). In line with our main findings, we do not

find evidence of more timely write-downs or higher levels of conservatism for firms with debt

versus those without debt. Specifically, consistent with prior research (e.g., Beatty and Weber

2006; Ramanna and Watts 2012) asset impairments are actually more timely for unlevered firms

than for levered firms, and that levered firms appear to be less conservative. Not surprising given

the foregoing impairment practices, we find that unlevered firms have higher levels of

conditional conservatism than levered firms for each BTM partition from BTM2 to BTM7.

Moreover, we confirm that all of the findings presented in Tables 2 through 5 hold for firms

without debt. These results make it unlikely that our model of mandatory conservatism is

mistakenly modeling variation in discretionary conservatism that is motivated by debt-

contracting incentives.

Moreover, as previously mentioned, it is possible that the variables we use to model

mandatory conservatism may also model discretionary conservatism driven by other contracting

incentives. To the extent that firms voluntarily engage in conservatism that is not mandated by

GAAP, it is more likely to be manifested in write-downs where BTM is lower than but

reasonably close to 1. We therefore alter our modeling of mandatorily conservative accounting to

replace BTM with write-downs which are set to zero for all observations where BTM<1 and find

qualitatively similar results to those reported in Tables 6 through 10. Taken together, the

foregoing results mitigate concerns that our results arise because we mistakenly model

discretionary incentives instead of mandatory conservatism.

37

Second, we re-examine our findings by partitioning the BTM ratio into two groups

(BTM1=1 if BTM<1, and BTM2=1 if BTM>=1) instead of the seven BTM partitions that were

used in the main analysis (untabulated). We find that firm-year observations with BTM ratios

greater than or equal to one have significantly more special-item charges and write-offs than

observations with BTM ratios less than one [the BTM2 partition mean values are -0.031 (t = -

6.47; p < 0.01) and -0.112 (t = -12.87; p < 0.01) for SPI and WRITEDOWN, respectively].

Furthermore, we find that the mean values are higher for firms with BTM ratios greater than or

equal to one which also have weak past performance and that these firms have also higher Basu

(1997) coefficients than the other three partitions. Therefore, our main inferences pertaining to

the determinants of discretionary and mandatory conservatism hold when partitioning the BTM

ratio into two groupings rather than into seven different partitions.

Third, we rerun our main analyses using an alternative definition of the BTM ratio.

Specifically, instead of using the BTM of assets, we follow the standard definition using the BTM

of equity. This alternative has no appreciable impact on any results; however, we believe that the

asset BTM ratio is more consistent with identifying conditions warranting special-item and write-

down charges.

Fourth, we examine the relation of two additional measures of operating performance

with mandatorily conservative accounting: sales growth over the prior three years and the free-

cash flow to assets ratio from years t-2 to t+2. We employ these measures in attempt to reflect

the notion that certain assets are only impaired if their carrying values exceed the estimated gross

undiscounted cash flows from the use and disposition of the assets. We show that these variables

stand alone as determinants of mandatory conservatism; however, their significance is subsumed

by our earlier measures of operating performance (i.e., past stock returns and past returns on

38

assets).

Fifth, following Duarte and Young (2009) who decompose the standard PIN measure into

an information asymmetry component and a liquidity component, we repeat the PIN analysis

solely using the information asymmetry component of PIN. We find similar inferences to those

documented in the main analysis using this alternative PIN measure.

Sixth, as an alternative approach to our main analyses, we re-compute earnings by

backing out the fitted values for mandatory write-downs and then re-running the original Basu

specification with the revised earnings numbers. We find that the economic significance of the

Basu coefficient decreases by approximately 27 percent, further supporting the notion that

mandatory asset write-downs are an important determinant of conditional conservatism.

Lastly, we anticipate that the extent of mandatorily conservative accounting will vary

with the type of assets that are on firms’ balance sheets given the different impairment standards

for each asset class. Intangibles and inventories are subject to strict fair value tests for

impairment, as so mandatory impairment should occur close to a BTM of one. In contrast, PP&E

is subject to the less stringent recoverability test, and financial assets such as receivables are

subject to the less stringent incurred loss model. To examine this prediction, we repeat the write-

down and the conditional conservatism analyses when partitioning firms based on their asset

structures. We group firms based on whether the majority of their assets are financial assets (i.e.,

cash and receivables), inventories, PP&E, and intangible assets. We find that all four groups

display the predicted characteristics of mandatorily conservative accounting. However, we find

that the write-down and conditional conservatism nonlinearities surrounding a BTM of one are

most pronounced for the intangibles and inventory firm groupings. These findings are consistent

39

with the impairment provisions requiring a strict fair value test for intangibles and inventory,

while PP&E and financial assets reflect less stringent tests.

7. Conclusion

In this study, we point out that the extant accounting rules mandate the use of

conservative accounting under specific economic conditions. We attempt to model these

conditions and show that our model explains significant cross-sectional variation in conditionally

conservative accounting. We also find that after using our model to control for mandatorily

conservative accounting, several inferences from previous research regarding managers’

incentives for discretionary conservatism are significantly altered.

We caveat our analysis by noting that our model is crude and far from complete. We

likely capture but a small proportion of mandatory conservatism and our model of mandatorily

conservative accounting could reflect an alternative theory of discretionary conservatism

yielding the same predictions as those made by mandatorily conservative accounting; however,

we are currently unaware of such a theory. Furthermore, improving our model could further

impact inferences from previous research. Nevertheless, the fact that a crude model of

mandatorily conservative accounting significantly alters the inferences from several previous

studies regarding discretionary conservatism highlights the need for more research in this area.

More generally, this study highlights the potential importance of mandatory accounting

rules in explaining material accounting outcomes, and suggests that researchers should attempt to

control for mandatory accounting practices when testing hypotheses regarding managerial

discretion. Moreover, the study provides a framework for modeling mandatorily conservative

accounting and shows that it is a nonlinear function of underlying economic variables. Lastly, the

study sheds light on the inconsistency between FASB’s recent conceptual framework (FASB

40

2010) which omits conservatism “as an aspect of representation faithfulness because including

(it) would be inconsistent with neutrality”, and the current accounting standards in place which

are shown to result in mandatorily conservative accounting.

While our study suggests that the Basu (1997) measure of conservatism reflects

mandatorily conservative accounting practices, we do not intend to resolve the debate concerning

potential problems with this measure (e.g., Dietrich, Miller, and Riedl 2007; Ball et al. 2010;

Patatoukas and Thomas 2011). We also stress that our study does not claim to reach any

definitive inferences regarding the determinants of discretionary conservatism. Rather, we hope

to encourage future research to further consider the importance of mandatory accounting rules in

influencing the properties of accounting numbers and specifically, to enhance the model of

mandatorily conservative accounting proposed in this study.

41

REFERENCES

Accounting Principles Board (APB). 1970. APB Statement No. 4, Basic Concepts and Accounting

Principles Underlying Financial Statements of Business Enterprises. New York: AICPA.

Ahmed, A., B. Billings, R. Morton, and M. Stanford-Harris. 2002. The role of accounting conservatism in

mitigating bondholder-shareholder conflicts over dividend policy and in reducing debt costs. The

Accounting Review 77 (4): 867-890.

——–, and S. Duellman. 2007. Accounting conservatism and board of director characteristics: An

empirical analysis. Journal of Accounting & Economics 43 (2-3): 411-437.

American Institute of Certified Public Accountants Committee on Accounting Procedures (AICPA).

1939. Accounting Research Bulletin 2. New York: AICPA.

Ball, R., S.P. Kothari, and A. Robin. 2000. The effect of international institutional factors on properties of

accounting earnings. Journal of Accounting & Economics 29 (1): 1-51.

——–, A. Robin, and G. Sadka. 2008. Is financial reporting shaped by equity markets or by debt markets?

An international study of timeliness of conservatism. Review of Accounting Studies 13 (2-3): 168-

205.

——–, S.P. Kothari, and V. Nikolaev. 2010. Econometrics of the Basu asymmetric timeliness coefficient

and accounting conservatism. Working paper, University of Chicago.

Basu, S. 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of

Accounting & Economics 24 (1): 3-37.

Beatty, A., and J. Weber. 2006. Accounting discretion in fair value estimates: An examination of SFAS

142 goodwill impairments. Journal of Accounting Research 44 (2): 257:-288.

——–, ——–, and J. Yu. 2008. Conservatism and debt. Journal of Accounting & Economics 45 (2-3):

154-174.

Beaver, W., and S. Ryan. 2000. Biases and lags in book value and their effects on the ability of the book-

to-market ratio to predict book return on equity. Journal of Accounting Research 38 (1): 127-148.

——–, and ——–. 2005. Conditional and unconditional conservatism: concepts and modeling. Review of

Accounting Studies 10 (2-3): 269-309.

Dechow, P., R. Sloan, and A. Sweeney. 1995. Detecting earnings management. The Accounting Review

70 (2): 193-225.

——–, A. Hutton, and R. Sloan. 1999. An empirical assessment of residual income valuation model.

Journal of Accounting & Economics 26 (1-3): 1-34.

Devine, C.T. 1963. The rule of conservatism reexamined. Journal of Accounting Research 1 (2): 127-138.

Dietrich, D., K. Muller, and E. Riedl. 2007. Asymmetric timeliness tests of accounting conservatism.

Review of Accounting Studies 12 (1): 95-124.

42

Duarte, J., and L. Young. 2009. Why is PIN priced? Journal of Financial Economics 91 (2): 119-138.

Elliott, J., and W. Shaw. 1988. Write-offs as accounting procedures to manage perceptions. Journal of

Accounting Research 26 (Supplement): 91-119.

——–, and D. Hanna. 1996. Repeated accounting write-offs and the information content of earnings.

Journal of Accounting Research 34 (Supplement): 135-155.

Feltham, G., and J. Ohlson. 1995. Valuation and clean surplus accounting for operating and financial

activities. Contemporary Accounting Research 11 (2): 689-731.

Financial Accounting Standards Board (FASB). 1980. Qualitative Characteristics of Accounting

Information. Statement of Financial Accounting Concepts No. 2. Norwalk, CT: FASB.

——–. 1995. Accounting for the Impairment of Long-Lived Assets and for Long-Lived Assets to Be

Disposed Of. Statement of Financial Accounting Standards No. 121. Norwalk, CT: FASB.

——–. 2001a. Goodwill and Other Intangible Assets. Statement of Financial Accounting Standards No.

142. Norwalk, CT: FASB.

——–. 2001b. Accounting for the Impairment or Disposal of Long-Lived Assets. Statement of Financial

Accounting Standards No. 144. Norwalk, CT: FASB.

——–. 2009a. Intangibles—Goodwill and Other. Accounting Standards Codification 350. Norwalk, CT:

FASB.

——–. 2009b. Property, Plant, and Equipment. Accounting Standards Codification 360. Norwalk, CT:

FASB.

——–. 2009c. Inventory. Accounting Standards Codification 330. Norwalk, CT: FASB.

——–. 2010. Conceptual Framework for Financial Reporting—Chapter 1, The Objective of General

Purpose Financial Reporting, and Chapter 3, Qualitative Characteristics of Useful Financial

Information. Statement of Financial Accounting Concepts No. 8. Norwalk, CT: FASB.

Francis, J., D. Hanna, and L. Vincent. 1996. Causes and effects of discretionary asset write-offs. Journal

of Accounting Research 34 (Supplement): 117-134.

Giner, B., and W.P. Rees. 2001. On the asymmetric recognition of good and bad news in France,

Germany, and the United Kingdom. Journal of Business Finance & Accounting 28 (9-10): 1285-

1331.

Givoly, D., and C. Hayn. 2000. The changing time-series properties of earnings, cash flows, and accruals:

has financial reporting become more conservative? Journal of Accounting & Economics 29 (3):

287-320.

——–, ——–, and A. Natarajan. 2007. Measuring reporting conservatism. The Accounting Review 82 (1):

65-106.

Hastie, T., and R. Tibshirani. 1990. Generalized Additive Models. London: Chapman and Hall.

43

Heitzman, S., C. Wasley, and J. Zimmerman. 2010. The joint effects of materiality thresholds and

voluntary disclosure incentives on firms’ disclosure decisions. Journal of Accounting &

Economics 49 (1-2): 109-132.

Holthausen, R., and R. Watts. 2001. The relevance of value-relevance literature for financial accounting

standard setting. Journal of Accounting & Economics 31 (1-3): 3-75.

Kaplan, B. 1985. Comments on Paul Healy: Evidence on the effect of bonus schemes on accounting

procedure and accrual decisions. Journal of Accounting & Economics 7 (1-3): 109-113.

Khan, M., and R. Watts. 2009. Estimation and empirical properties of a firm-year measure of accounting

conservatism. Journal of Accounting & Economics 48 (2-3): 132-150.

LaFond, R., and S. Roychowdhury. 2008. Managerial ownership and accounting conservatism. Journal of

Accounting Research 46 (1): 101-135.

——–, and R. Watts. 2008. The information role of conservatism. The Accounting Review 83 (2): 447-

478.

Lara, J.M.G., B.G. Osma, and F. Penalva. 2009. The economic determinants of conditional conservatism.

Journal of Business Finance & Accounting 36 (3-4): 336-372.

Lee, H., 2010. Conditional conservatism, agency costs, and the contractual features of debt. Working

paper, University of Arizona.

Nikolaev, V. 2010. Debt covenants and accounting conservatism. Journal of Accounting Research 48 (1):

51-89.

Pae, J., D. Thornton, and M. Welker. 2005. The link between earnings conservatism and the price-to-book

ratio. Contemporary Accounting Research 22 (3): 693-717.

——–. 2007. Unexpected accruals and conditional accounting conservatism. Journal of Business Finance

& Accounting 34 (5-6): 681-704.

Patatoukas, P., and J. Thomas. 2011. More evidence of bias in the differential timeliness measure of

conditional conservatism. The Accounting Review 86 (5): 1765-1793.

Penman, S., and X. Zhang. 2002. Accounting conservatism, the quality of earnings and stock returns. The

Accounting Review 77 (2): 237-264.

Ramalingegowda, S., and Y. Yu. 2012. Institutional ownership and conservatism. Journal of Accounting

& Economics 53 (1-2): 98-114.

Ramanna, K., and R. Watts. 2012. Evidence on the use of unverifiable estimates in required goodwill

impairment. Forthcoming Review of Accounting Studies.

Rees, L., S. Gill, and R. Gore. 1996. An investigation of asset write-downs and concurrent abnormal

accruals. Journal of Accounting Research 34 (Supplement): 157-169.

Riedl, E. 2004. An examination of long-lived asset impairments. The Accounting Review 79 (3): 823-852.

44

Roychowdhury, S., and R. Watts. 2007. Asymmetric timeliness of earnings, market-to-book and

conservatism in financial reporting. Journal of Accounting & Economics 44 (1-2): 2-31.

——–. 2010. Discussion of: “Acquisition profitability and timely loss recognition” by J. Francis and X.

Martin. Journal of Accounting & Economics 49 (1-2): 179-183.

Ruddock, C., S. Taylor, and S. Taylor. 2006. Nonaudit services and earnings conservatism: Is auditor

independence impaired. Contemporary Accounting Research 23 (3): 701-746.

Ryan, S. 2006. Identifying conditional conservatism. European Accounting Review 15 (4): 511-525.

Schipper, K. 1989. Commentary on earnings management. Accounting Horizons 3 (4): 91-102.

Strong, J., and J. Meyer. 1987. Asset writedowns: managerial incentives and security returns. The Journal

of Finance 42 (3): 643-661.

The Wall Street Journal CFO Report (WSJ). 2012. SEC Makes Barnes & Noble Justify Unimpaired

Goodwill. July 17, 2012.

Vuong, Q. 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57

(2): 307-333.

Wang, C., F. Xie, and X. Xin. 2011. Managerial ownership of debt and accounting conservatism.

Working Paper, Chinese University of Hong Kong.

Watts. R., and J. Zimmerman. 1986. Positive Accounting Theory. Englewood Cliffs, NJ: Prentice-Hall.

——–. 2003a. Conservatism in accounting – Part I: explanations and implications. Accounting Horizons

17 (3): 207-221.

——–. 2003b. Conservatism in accounting – Part II: evidence and research opportunities. Accounting

Horizons 17 (4): 287-301.

Zhang, J. 2008. The contracting benefits of accounting conservatism to lenders and borrowers. Journal of

Accounting & Economics 45 (1): 27-54.

45

APPENDIX Variables Definitions

= Firm i’s total assets scaled by the sum of market capitalization plus total assets minus

the book value of common equity as of fiscal t-1’s year-end;

= J ranges between 1 and 7. equals to “1” if firm i’s is in the interval

0-0.3, “0” otherwise. , , , , , and

equal to “1” if firm i’s is in the intervals 0.3-0.5, 0.5-0.7, 0.7-0.9,

0.9-1, 1-1.2, and 1.2 & above, respectively, “0” otherwise;

= An interaction between and firm i’s past performance G. G equals to “1” if

firm i’s and both equal to one, “0” otherwise;

= An interaction between and firm i’s past performance B. B equals to “1” if

firm i’s or equal to zero, “0” otherwise;

= Firm i’s earnings per share excluding extraordinary items for fiscal year t scaled by its

price as of fiscal t-1’s year-end;

= Firm i’s cumulative raw return beginning nine months before fiscal year-end t to three

months after fiscal year-end t;

= “1” if firm i’s is less than zero, “0” otherwise;

= “1” if firm i’s (XRETt-2 + XRETt-1) / 2 is greater than or equal to five percent, “0”

otherwise; is firm i’s value-weighted market-adjusted annual return beginning

nine months before fiscal year-end t to three months after fiscal year-end t;

= “1” if firm i’s (ROAt-2 + ROAt-1) / 2 is greater than or equal to five percent, “0”

otherwise; is firm i’s pretax income before special-items divided by total assets

as of fiscal t’s year-end;

= “1” if firm i’s equals to zero or equals to zero, “0” otherwise;

= Firm i’s special-items for fiscal year t scaled by its market capitalization as of fiscal t-

1’s year-end; Special-items include any significant nonrecurring items, asset write-

downs, impairments of goodwill, and restructuring charges;

= The sum of asset write-downs and goodwill impairments for firm i for fiscal year t

scaled by its market capitalization as of fiscal t-1’s year-end;

SPI%t = “1” if firm i’s is less than zero, “0” otherwise;

WRITEDOWN%t = “1” if firm i’s is less than zero, “0” otherwise;

= Long-term debt for firm i scaled by its total assets as of fiscal t’s year-end;

= Firm i’s long-term debt issuance for fiscal year t scaled by its market capitalization as

of fiscal t-1’s year-end;

= The count of covenants included in a debt contract for firm i for fiscal year t. Following

Nikolaev (2010), the RESTRICT analysis includes 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);

= Firm i’s sale of common and preferred stock for fiscal year t scaled by its market

capitalization as of fiscal t-1’s year-end; and,

= Firm i’s probability of an information based trade for fiscal year t.

46

FIGURE 1 The Predicted Relation between Asset Impairments and under Different Asset Classes

Panel A: The Relation between Goodwill and Indefinite-Lived Intangible

Impairments and the BTM of Assets

Panel B: The Relation between PP&E and Finite-Lived Intangible Impairments and

the BTM of Assets

Figure 1 presents the predicted relations between asset impairments and the asset ratio under two

different asset impairment standards. Panel A plots the predicted relation between goodwill & indefinite-

lived intangible write-downs and the ratio. Panel B plots the predicted relation between PP&E and

finite-lived intangibles and the ratio.

Go

od

will

& I

nd

efi

nit

e-L

ive

d I

nta

ngi

ble

W

rite

do

wn

s/M

kt C

ap

BTM of Asset

Impairment under ASC 350

10

45°

PP

&E

& F

init

e-L

ive

d I

nta

ngi

ble

W

rite

do

wn

s/M

kt C

ap

BTM of Asset

Impairment under ASC 360

10

45°

47

FIGURE 2 Evaluating the Relation between Write-downs ( ) and

Panel A: The Relation between Write-Downs and Groups

Panel B: The Relation between Write-Downs and Using a Nonlinear

Regression

(Continued)

48

Panel C: The Relation between Write-Downs and Groups Conditioning on Firm

Performance

Panel D: The Relation between Write-Downs and Conditioning on Firm

Performance Using a Nonlinear Regression

Figure 2 presents four panels that plot the relation between write-downs ( ) and .

Panel A plots the relation between write-downs and the mean value of seven partitions using the

mean of from Column (7) in Table 3, Panel A. The seven partitions represent the

following intervals: 0-0.3, 0.3-0.5, 0.5-0.7, 0.7-0.9, 0.9-1, 1-1.2, and 1.2 & above, respectively.

49

Panel B plots the relation between write-downs and using a generalized additive model (GAM) in

R. The GAM function allows for nonlinear effects by using a backfitting partial-linear model. The lines

surrounding the regression line represent 95 percent confidence intervals. Panel C plots the relation

between write-downs and the mean value of seven groups conditioning on the firms’ past

performance using the mean of . Panel D plots the relation between write-downs and

conditioning on the firms’ past performance using a GAM in R. The X-axis represents the ratio

and the Y-axis represents the predicted value of .

50

FIGURE 3 Evaluating the Relation between the Basu Coefficient and

Panel A: The Relation between the Basu Coefficient and Groups

Panel B: The Relation between the Basu Coefficient and Groups Conditioning on Firm

Performance

Figure 3 presents two panels that plot the nonlinear relation between the Basu coefficient and .

Panel A plots the relation between the Basu coefficient and the mean value of seven partitions

51

using the combined coefficients (Sum Coef.) from Column (1) in Table 5. The seven partitions

represent the following intervals: 0-0.3, 0.3-0.5, 0.5-0.7, 0.7-0.9, 0.9-1, 1-1.2, and 1.2 & above,

respectively. Panel B plots the relation between the Basu coefficient and the mean value of seven

groups conditioning on the firms’ past performance using the combined coefficients (Sum Coef.) from

Equation (6).

52

TABLE 1 Descriptive Statistics and Univariate Analysis

Panel A - Descriptive statistics

Variable N Mean Std.Dev. Q1 Median Q3

139,603 0.802 0.370 0.553 0.823 1.006

139,603 -0.005 0.285 -0.012 0.051 0.095

139,603 0.186 0.807 -0.206 0.070 0.385

139,603 0.432 0.495 0.000 0.000 1.000

139,603 0.575 0.494 0.000 1.000 1.000

139,603 0.498 0.450 0.000 0.000 1.000

139,603 0.751 0.432 1.000 1.000 1.000

139,603 -0.025 0.117 -0.003 0.000 0.000

47,259 -0.018 0.103 0.000 0.000 0.000

139,603 0.215 0.177 0.069 0.181 0.314

139,603 0.222 0.703 0.000 0.008 0.156

22,483 3.991 4.828 0.000 3.000 6.000

139,603 0.041 0.141 0.000 0.002 0.013

31,992 0.219 0.123 0.140 0.180 0.260

(Continued)

53

Panel B - Mean values of key variables sorted by BTM partitions

Group

Interval N E/P RET D B SPI

WRITE

DOWN SPI%

WRITE

DOWN% LEV

DEBT_

ISSUE

RES

TRICT EQUITY

_ISSUE PIN

1 0-0.3 11,789 0.197 -0.031 0.049 0.587 0.650 -0.008 -0.002 0.326 0.143 0.161 0.029 2.196 0.061 0.168

2 0.3-0.5 17,401 0.405 -0.001 0.105 0.506 0.603 -0.010 -0.004 0.329 0.190 0.180 0.065 3.243 0.044 0.175

3 0.5-0.7 23,303 0.603 0.009 0.143 0.466 0.660 -0.015 -0.009 0.330 0.221 0.220 0.137 4.111 0.042 0.193

4 0.7-0.9 30,100 0.806 0.013 0.164 0.427 0.765 -0.021 -0.014 0.306 0.222 0.243 0.216 4.270 0.041 0.207

5 0.9-1 20,862 0.952 0.020 0.192 0.378 0.865 -0.024 -0.017 0.264 0.159 0.212 0.340 4.547 0.037 0.229

6 1-1.2 21,893 1.081 -0.016 0.270 0.370 0.831 -0.042 -0.044 0.250 0.228 0.235 0.396 4.861 0.040 0.230

7 1.2 & above 14,255 1.454 -0.071 0.379 0.344 0.847 -0.059 -0.072 0.238 0.262 0.180 0.285 4.371 0.030 0.341

This table presents the descriptive statistics and univariate analysis for all the key variables. Panel A provides descriptive statistics for all the key variables and Panel B

presents the means for these variables sorted by the seven partitions. The sample sizes in Panel B refer to the main sample of 139,603 observations. See the Appendix

for variable definitions.

54

TABLE 2 An Analysis of the Relation between Special-Items / Write-Downs and

(1)

(2)

(1) (2) (3) (4)

Exp.

Sign

Dep.Var

Coef.

(t-stat)

Dep.Var

Coef.

(t-stat)

Dep.Var

Coef.

(t-stat)

Dep.Var

Coef.

(t-stat)

Intercept 0.009*** -0.003*** 0.018*** 0.005***

(3.32) (-3.20) (4.59) (4.54)

- -0.043*** -0.006*** -0.048*** -0.017***

(-7.52) (-3.11) (-4.92) (-6.21)

0.012*** 0.011***

(4.71) (4.04)

- -0.042*** -0.035***

(-9.02) (-4.02)

Adj 0.0164 0.0234 0.0247 0.0270

N 139,603 139,603 47,259 47,259

This table presents an analysis of the relation between special-items ( ) or write-downs ( ) and

. *, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-tailed tests. T-

statistics and p-values are calculated using clustered standard errors by firm and year. See the Appendix for variable

definitions.

55

TABLE 3 An Analysis of the Nonlinear Relation between Special-Items / Write-Downs and

Panel A - Test of differences between group means and predicted values

N Mean Pred.Val Diff.

P-val

(2-tailed) N Mean Pred.Val Diff.

P-val

(Two-tailed)

11,789 -0.008 0.001 -0.009 0.000 5,110 -0.002 0.008 -0.010 0.000

17,401 -0.010 -0.008 -0.002 0.000 7,152 -0.004 -0.002 -0.002 0.000

23,303 -0.015 -0.017 0.002 0.000 8,594 -0.009 -0.011 0.003 0.000

30,100 -0.021 -0.026 0.005 0.000 10,036 -0.014 -0.021 0.007 0.000

20,862 -0.024 -0.032 0.007 0.000 7,862 -0.017 -0.028 0.012 0.000

21,893 -0.042 -0.038 -0.005 0.000 5,613 -0.044 -0.034 -0.010 0.000

14,255 -0.059 -0.054 -0.005 0.000 2,892 -0.072 -0.052 -0.019 0.000

Panel B - Nonlinear tests

Vuong test between Equation (1) and a piecewise nonlinear model of Equation (1)

Dep.Var

Dep.Var

Vuong Z-value -2.207 -5.747

P-value 0.027 0.000

Goodness of fit between Equation (1) and a GAM model of Equation (1)

F-value 43.496 46.595

P-value 0.000 0.000

This table presents the means and the predicted means of special-items ( ) and write-downs ( ) partitioned by seven

partitions, and three analyses to examine the nonlinear relation between special-items or write-downs and . In Panel A, predicted

means of special-items or write-downs are calculated as the fitted values of Equation (1) by applying coefficients from Columns (1) and (3) in Table

2, respectively, to the mean value of each partition. T-tests (two-tailed) are performed by comparing whether the means of special-items or

write-downs are significantly different from the predicted values for each partition. Panel B presents two statistical tests comparing the

goodness of fits of the linear and nonlinear specifications. The Vuong test compares the fit of the linear model in Equation (1) to a piecewise version

of Equation (1), which regresses special-items or write-downs on seven dummies variables of partitions. A negative Z-value indicates that the

56

linear model of Equation (1) has a significantly smaller explanatory power than the piecewise version of Equation (1) and hence, a weaker fit. The F-

test compares the fit of the linear model of Equation (1) to the GAM (presented in Figure 2, Panel B for WRITEDOWN). A significant F-value

indicates that the GAM has a better fit than the linear model. See the Appendix for variable definitions.

57

TABLE 4 An Analysis of the Relation between Conservatism and

(3)

(5)

(1) (2) (3)

Exp.

Sign

Coef.

(t-stat)

Coef.

(t-stat)

Coef.

(t-stat)

Intercept 0.061*** 0.032*** -0.010**

(8.02) (2.72) (-2.03)

0.004 0.041*** 0.022***

(0.49) (2.81) (3.11)

+ -0.029*** -0.005 0.002

(-3.52) (-0.60) (0.47)

* + 0.475*** 0.018 -0.050**

(14.81) (0.58) (-2.52)

0.034 0.152***

(1.54) (11.61)

* -0.028 -0.027

(-1.20) (-1.85)*

* -0.027* 0.015

(-1.68) (1.05)

* * + 0.701*** 0.328***

(9.72) (8.32)

0.039***

(3.90)

* 0.024

(1.48)

* -0.014

(-1.26)

* * + 0.122***

(3.66)

* -0.138***

(-9.06)

* * -0.006

(-0.26)

58

* * -0.039*

(-1.84)

* * * + 0.325***

(4.40)

Adj 0.0993 0.1431 0.1732

N 139,603 139,603 139,603

This table presents an analysis of the relation between conservatism, as measured per Basu (1997), and

firms’ past performance. *, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-

tailed tests. T-statistics and p-values are calculated using clustered standard errors by firm and year. See the

Appendix for variable definitions.

59

TABLE 5 An Analysis of the Nonlinear Relation between Conservatism and

∑

∑ ∑

∑

(4)

(1)

N

Coef.

(t-stat)

Sum

Coef.

Pred

Coef. Diff.

P-val

(two-

tailed)

Intercept 0.007

(1.63)

0.008

(1.17)

+ -0.023***

(-9.77)

* + 11,789 0.178*** 0.178 0.142 0.036 0.013

(12.20)

* * + 17,401 0.093*** 0.270 0.291 -0.021 0.389

(3.32)

* * + 23,303 0.164*** 0.341 0.434 -0.092 0.000

(5.68)

* * + 30,100 0.312*** 0.490 0.579 -0.089 0.003

(8.46)

* * + 20,862 0.464*** 0.642 0.684 -0.043 0.542

(5.94)

* * + 21,893 0.688*** 0.865 0.777 0.088 0.278

(8.07)

* * + 14,255 1.016*** 1.193 1.045 0.148 0.088

(11.82)

Adj 0.1527

N 139,603

Vuong test between Equation (3) in Column (2) of Table 4 and Equation (4)

Z-value P-value

-10.071 0.000

Goodness of fit between a parametric specification of Equation (3) in Column (2) of Table 4 and a GAM

nonparametric specification of Equation (3)

F-value P-value

153.270 0.000

60

This table presents an analysis of the relation between conservatism, as measured per Basu (1997), and seven

partitions. In panel A, the combined conservatism coefficients (Sum Coef.) of each partition are

calculated from the partitions’ respective * * regression coefficient plus the * coefficient.

Results for the stand-alone , the two-way interactions: , and , are not

reported for brevity. *, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-tailed

tests. T-statistics and p-values are calculated using clustered standard errors by firm and year. The predicted

conservatism coefficients (Pred Coef.) of each partition are calculated as the fitted value of Equation (3) by

applying the coefficients of * and * * from Column (2) in Table 4, to the mean value of each

partition. F-tests are performed by testing whether the coefficient of * (for , or the sum

of the coefficients of * and (for to ) are significantly different

from their predicted value for each partition respectively. Two-tailed p-values are reported. The Vuong

test compares the goodness of fit of Equation (3) to Equation (4). A negative Z-value indicates that Equation (3) has

a significantly smaller explanatory power than Equation (4) and hence, a weaker fit. The F-test compares the

goodness of fit of Equation (3) to the GAM version of that specification. A significant F-value indicates that the

GAM has a better fit than Equation (3). See the Appendix for variable definitions.

61

TABLE 6 Conservatism and Debt Issuance

(7)

Exp.

Sign

(1)

Coef.

(t-stat)

(2)

Coef.

(t-stat)

(3)

Coef.

(t-stat)

(4)

Coef.

(t-stat)

(5)

Coef.

(t-stat)

(6)

Coef.

(t-stat)

Intercept 0.061*** 0.065*** 0.032** 0.007* -0.010** 0.044***

(8.02) (8.78) (2.77) (1.79) (-2.04) (23.27)

0.004 0.003 0.041*** 0.008 0.021*** -0.002

(0.49) (0.42) (2.82) (1.16) (3.10) (-0.80)

+ -0.029*** -0.028*** -0.005 -0.023*** 0.002 -0.003*

(-3.52) (-3.48) (-0.55) (-9.88) (0.46) (-1.90)

* + 0.475*** 0.454*** 0.020 0.177*** -0.047*** 0.051***

(14.81) (15.34) (0.64) (12.16) (-2.37) (9.15)

-0.020*** -0.022*** -0.024*** -0.019*** -0.022***

(-4.78) (-5.74) (-6.21) (-5.19) (-5.71)

* -0.001 -0.001 -0.002 -0.001 -0.002

(-0.07) (-0.07) (-0.28) (-0.10) (-0.34)

* 0.001 0.001 0.001 0.002 0.002

(0.32) (1.08) (0.67) (1.33) (0.98)

* * + 0.075*** 0.008 0.006 0.007 0.005

(2.78) (0.36) (0.28) (0.33) (0.27)

CONTROLS

No No

,

Adj 0.0993 0.1068 0.1461 0.1565 0.1755 0.1863

N 139,603 139,603 139,603 139,603 139,603 139,603

This table presents an analysis of the relation between conservatism and debt issuance ( . Results for

are not reported for brevity. controls in Column (3) include , and its interactions with

, , and , respectively. controls in Column (4) include six indicator variables:

to , and their interactions with , , and , respectively. controls in Column

(5) include , , and their interactions with , , and , respectively.

and controls in Column (6) include the following thirteen indicator variables: six

and seven , and their interactions with , , and , respectively.

*, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-tailed tests. T-statistics and

p-values are calculated using clustered standard errors by firm and year. See the Appendix for variable definitions.

62

TABLE 7 Conservatism and Leverage

(7)

Exp. Sign

(1)

Coef.

(t-stat)

(2)

Coef.

(t-stat)

(3)

Coef.

(t-stat)

(4)

Coef.

(t-stat)

(5)

Coef.

(t-stat)

Intercept 0.071*** 0.041*** 0.013*** -0.004 0.049***

(8.59) (3.70) (2.75) (-0.82) (24.57)

0.001 0.040*** 0.008 0.021*** -0.001

(0.18) (2.95) (1.21) (2.89) (-0.62)

+ -0.026*** -0.002 -0.020*** 0.003 -0.002

(-3.47) (-0.18) (-8.46) (0.51) (-0.19)

* + 0.440*** 0.015 0.167*** -0.047** 0.047***

(13.43) (0.46) (12.71) (-2.19) (5.92)

-0.056*** -0.058*** -0.078*** -0.055*** -0.071***

(-3.7) (-3.52) (-4.99) (-3.62) (-4.66)

* 0.006 0.007 0.004 0.006 0.003

(0.37) (0.43) (0.24) (0.37) (0.20)

* -0.025* -0.026* -0.030** -0.019 -0.023*

(-1.90) (-1.90) (-2.19) (-1.56) (-1.74)

* * + 0.182*** 0.030 0.083* 0.014 0.064

(3.24) (0.69) (1.80) (0.33) (1.39)

CONTROLS

No

,

Adj 0.1028 0.1449 0.1563 0.1746 0.1859

N 139,603 139,603 139,603 139,603 139,603

This table presents an analysis of the relation between conservatism and leverage ( . Results for

are not reported for brevity. controls in Column (2) include , and its interactions with

, , and , respectively. controls in Column (3) include six indicator variables:

to , and their interactions with , , and , respectively.

controls in Column (4) include , , and their interactions with , , and

, respectively. and controls in Column (5) include the following thirteen indicator

variables: six and seven , and their interactions with , , and

, respectively. *, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-tailed

tests. T-statistics and p-values are calculated using clustered standard errors by firm and year. See the Appendix for

variable definitions.

63

TABLE 8 Conservatism and the Use of Covenants

(7)

Exp.

Sign

(1)

Coef.

(t-stat)

(2)

Coef.

(t-stat)

(3)

Coef.

(t-stat)

(4)

Coef.

(t-stat)

(5)

Coef.

(t-stat)

Intercept 0.052*** 0.065*** 0.024*** 0.011** 0.038***

(9.49) (5.77) (5.40) (2.29) (13.78)

0.012 -0.003 0.016 0.006 -0.002

(1.54) (-0.18) (1.25) (0.77) (-0.36)

+ -0.027*** -0.017 -0.027*** -0.003 -0.005***

(-2.64) (-1.55) (-5.24) (-0.73) (-2.70)

* + 0.352*** -0.088 0.183*** -0.017 0.037**

(11.92) (-1.04) (3.43) (-0.49) (2.25)

-0.001 -0.001 -0.001* -0.001 -0.001*

(-1.25) (-1.12) (-1.86) (-1.41) (-1.82)

* 0.001 0.001 0.001 0.001 0.001

(0.76) (0.45) (0.61) (0.60) (0.70)

* -0.002*** -0.002* -0.002* -0.001 -0.001

(-2.50) (-1.73) (-1.67) (-0.66) (-0.41)

* * + 0.013*** 0.003 0.005 0.002 0.003

(3.95) (0.92) (1.31) (0.45) (0.61)

CONTROLS

No ,

Adj 0.1016 0.146 0.175 0.1737 0.202

N 22,483 22,483 22,483 20,460 20,460

This table presents an analysis of the relation between conservatism and the use of covenants ( following

Nikolaev (2010). Results for are not reported for brevity. controls in Column (2) include ,

and its interactions with , , and , respectively. controls in Column (3) include six

indicator variables: to , and their interactions with , , and , respectively.

controls in Column (4) include , , and their interactions with , ,

and , respectively. and controls in Column (5) include the following thirteen

indicator variables: six and seven , and their interactions with , , and

, respectively. *, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-

tailed tests. T-statistics and p-values are calculated using clustered standard errors by firm and year. Following Nikolaev

(2010), this analysis includes 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). See the Appendix for variable definitions.

64

TABLE 9 Conservatism and Equity Issuance

(7)

Exp.

Sign

(1)

Coef.

(t-stat)

(2)

Coef.

(t-stat)

(3)

Coef.

(t-stat)

(4)

Coef.

(t-stat)

(5)

Coef.

(t-stat)

Intercept 0.066*** 0.039*** 0.014*** -0.007 0.047***

(9.38) (3.18) (3.84) (-1.44) (20.53)

0.001 0.038*** 0.005 0.021*** -0.002

(0.12) (2.67) (0.88) (2.96) (-0.74)

+ -0.022** 0.006 -0.012*** 0.008 0.002

(-2.20) (0.82) (-4.16) (1.63) (0.70)

* + 0.435*** -0.035 0.125*** -0.067*** 0.032***

(13.12) (-1.25) (7.06) (-3.44) (5.85)

-0.214*** -0.211*** -0.209*** -0.202*** -0.197***

(-5.57) (-5.49) (-5.60) (-5.57) (-5.65)

* 0.034 0.029 0.031 0.021 0.026

(0.74) (0.66) (0.72) (0.49) (0.62)

* -0.001 -0.004 -0.005 -0.005 -0.004

(-0.04) (-0.45) (-0.50) (-0.46) (-0.43)

* * + 0.432*** 0.491*** 0.491*** 0.458*** 0.464***

(3.42) (4.86) (4.85) (4.60) (4.68)

CONTROLS

No

,

Adj 0.1172 0.1627 0.1719 0.1912 0.2004

N 139,603 139,603 139,603 139,603 139,603

This table presents an analysis of the relation between conservatism and equity issuance ( . Results

for are not reported for brevity. controls in Column (2) include , and its interactions

with , , and , respectively. controls in Column (3) include six indicator variables:

to , and their interactions with , , and , respectively. controls

in Column (4) include , , and their interactions with , , and ,

respectively. and controls in Column (5) include the following thirteen indicator variables: six

and seven , and their interactions with , , and , respectively.

*, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-tailed tests. T-statistics and

p-values are calculated using clustered standard errors by firm and year. See the Appendix for variable definitions.

65

TABLE 10 Conservatism and Information Asymmetry ( )

(7)

Exp.

Sign

(1)

Coef.

(t-stat)

(2)

Coef.

(t-stat)

(3)

Coef.

(t-stat)

(4)

Coef.

(t-stat)

(5)

Coef.

(t-stat)

(6)

Coef.

(t-stat)

Intercept 0.060*** 0.067*** 0.073*** 0.023*** 0.015*** 0.039***

(13.22) (7.56) (7.53) (2.87) (2.86) (6.03)

0.015** 0.013 0.022* -0.004 0.008 -0.003

(2.33) (0.95) (1.68) (-0.29) (0.82) (-0.25)

+ -0.026* -0.013 -0.003 -0.005 0.004 0.012

(-1.91) (-0.52) (-0.17) (-0.27) (0.32) (0.97)

* + 0.516*** 0.177*** -0.235*** -0.029 -0.135*** -0.106***

(8.78) (2.70) (-4.78) (-0.58) (-3.77) (-2.99)

-0.035 -0.026 0.001 -0.025 0.000

(-1.19) (-0.85) (0.05) (-0.81) (-0.01)

* -0.001 0.008 0.008 -0.002 -0.001

(-0.01) (0.13) (0.13) (-0.03) (-0.01)

* -0.050 -0.043 -0.061 -0.049 -0.064

(-0.79) (-0.64) (-0.85) (-0.71) (-0.90)

* * + 1.395*** 0.838*** 0.871*** 0.768*** 0.802***

(6.04) (3.44) (3.65) (3.31) (3.51)

CONTROLS

No No

Adj 0.1112 0.1366 0.1752 0.1915 0.1954 0.2087

N 31,992 31,992 31,992 31,992 31,992 31,992

This table presents an analysis of the relation between conservatism and information asymmetry ( ). Results for

are not reported for brevity. controls in Column (3) include , and its interactions with

, , and , respectively. controls in Column (4) include six indicator variables:

to , and their interactions with , , and , respectively.

controls in Column (5) include , , and their interactions with , , and

, respectively. and controls in Column (6) include the following thirteen indicator

variables: six and seven , and their interactions with , , and

, respectively. *, **, *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively, using two-tailed

tests. T-statistics and p-values are calculated using clustered standard errors by firm and year. See the Appendix for

variable definitions.