Cognitive dissonance as an explanation of goodwill write-offs. SohyungKim.pdf · Cognitive dissonance is an uncomfortable psychological state that may arise after a decision has been

Cognitive dissonance as an explanation of goodwill write-offs

Sohyung Kim*Department of Accounting

The Goodman School of BusinessBrock University

500 Glenridge Ave.St. Catharines, Ontario, L2S 3A1

Tel.: (905) 688-5550 x5016Email: [email protected]

Darlene BayDepartment of Accounting

The Goodman School of BusinessBrock University

500 Glenridge Ave.St. Catharines, Ontario, L2S 3A1

Tel.: (905) 688-5550 x4524Email: [email protected]

Current Draft: January 2014

* Kim gratefully acknowledges financial support from CGA Ontario Research ExcellenceFund.

1

Cognitive dissonance as an explanation of goodwill write-offs

ABSTRACT

While agency theory has been an important part of accounting academic research, many

researchers have suggested that other theories should also be considered. One such theory is

the theory of cognitive dissonance. In this study, we develop competing hypotheses based on

agency theory and the theory of cognitive dissonance regarding the decision to record an

impairment of goodwill. Using an archival research methodology, we test the hypotheses on a

sample of 2,274 firm-year observations. Our results are consistent with the theory of

cognitive dissonance even after controlling for variables that have been found to be

significant under agency theory. We conclude that there is strong evidence to suggest that

agency theory does not uniquely explain the results of management decisions as seen through

financial reporting data.

Keywords: cognitive dissonance; asset write-offs; SFAS No. 142; goodwill

Data Availability: The data used in this study was taken from public sources identified in thestudy.

2

1. Introduction

Financial statements provide information that reflects the cumulative outcome of a

multitude of management decisions. In general, these decisions result in changes to economic

resources that can be documented and verified. However, some accounting standards require

the application of management discretion in determining the measurement and timing of a

financial statement item. One example is the reporting of fair market values for assets in

which an active market does not exist. In this case, the resultant information may be impacted,

intentionally or unconsciously, by characteristics of the decision maker. Thus, it is important

to understand the decision-making behavior of managers, including the incentives,

motivations, and biases to which they may be subject.

Since the publication of classic papers by Jensen and Meckling (1976) and Watts and

Zimmerman (1986), archival accounting research has looked to agency theory to provide the

context for management decisions. This theory assumes that managers are motivated by

economic incentives and an aversion to effort and will opportunistically choose to report in a

manner that maximizes their own self-interest rather than the interests of the shareholders.

This research paradigm has led to an understanding of how management may be using

accounting information to manage earnings, how management determines what information

to disclose in areas such an environmental performance and governance practices, and, more

recently, how management may use actual transactions in an effort to manage (real) earnings.

Agency theory has been criticized as providing too narrow a perspective on human

behavior (see, for example, Baiman, 1990). In related disciplines such as economics

(Camerer and Loewenstein, 2004) and finance (Shiller, 2003), and in other areas of

accounting research such as experimental research (see for example, Brown et al., 2009), a

broader range of theories is being discussed, recommended, and applied. This has the

potential to increase our understanding of decision making and to increase the types of

3

research questions that can be asked. Even in archival accounting research, there is growing

recognition of the need to consider variables suggested by theories other than agency theory

(Kachelmeier, 2010). In the current paper, we follow this path by applying a theory from the

psychology literature, cognitive dissonance, which may explain management behavior as

recorded in financial accounting information. While the theory has been recommended in at

least two review papers (Birnberg et al., 2007; Koonce and Mercer, 2005) it has only rarely

been used in accounting behavioral research (see Jermais, 2001 for an exception) and has not

yet been used in accounting archival research.

Cognitive dissonance is an uncomfortable psychological state that may arise after a

decision has been made, especially if negative consequences occur. It results from

simultaneously holding two conflicting cognitions (opinions, beliefs or views) related to the

decision (Festinger, 1962). A person who experiences cognitive dissonance, if it is strong

enough, will attempt to adjust the cognition that is most easily changed. This may take the

form of ignoring or misinterpreting information that conflicts with the decision, actively

searching for information that confirms the decision, or, if the information that conflicts with

the decision is overwhelming, changing the decision or the perception of the effectiveness of

the decision. Cognitive dissonance has been extensively studied in the psychology literature

(see Cooper, 2007) and has been applied in the marketing literature (see Cummings and

Venkatesan, 1976 for a review).

In order to test the theory of cognitive dissonance in the financial accounting context

and to compare its predictions about management behavior to agency theory predictions, we

study goodwill write-offs. Gilad et al. (1987) have suggested that cognitive dissonance can be

used to explain prior findings that underperforming segments are often not sold until there is

a change in management. Information that would suggest the unit should be sold would

create cognitive dissonance for the manager that made the original investment decision and

4

might thus be ignored. If cognitive dissonance can explain delays in divesting

underperforming segments, it quite likely will also explain evidence that tenured managers

are less likely to record impairments of goodwill (Hamberg et al., 2011). Both decisions

require a willingness on the part of the decision maker to consider, appropriately interpret,

and act on information that might conflict with a belief that the initial decision was a good

one. The decision to acquire a new business unit meets important boundary conditions of

cognitive dissonance: the manager freely made the decision (i.e. was not forced to acquire the

unit), demonstrated commitment to the decision by investing, and could have foreseen the

potential for negative outcomes.

In addition to satisfying the necessary conditions for cognitive dissonance, goodwill

write-offs provide a situation in which agency theory and cognitive dissonance suggest

different hypotheses. According to agency theory, managers are likely to opportunistically

exploit the discretion and lack of verifiability inherent in goodwill accounting. Researchers

that examine goodwill accounting (e.g., Ramanna, 2008; Ramanna and Watts, 2012; Watts,

2003, 2006) argue that managers delay write-offs of goodwill in order to avoid reducing net

income. In addition, when write-offs are eventually recorded, they argue that managers tend

to write-off inappropriately large amounts in order to ensure that no further write-offs

become necessary (the so-called big bath). By contrast, cognitive dissonance would indicate

that, once the amount of conflicting information becomes so great that it cannot be ignored,

the decision maker will actually seek out more conflicting information and will, if necessary,

revise cognitions about the original decision. Once the critical point is reached, write-offs of

goodwill in subsequent years will be likely, since the bias created by cognitive dissonance

will have been removed. Thus, when goodwill is written off, agency theory would predict

that no write-off would occur in the subsequent period, while the theory of cognitive

dissonance would suggest it is quite likely there would be a write-off in the subsequent period.

5

To test these competing hypotheses, we identify a sample of 2,274 firm-years with

positive balances in goodwill and book-to-market ratios greater than one (a potential indicator

of goodwill impairment). We analyze these observations to determine whether a write-off in

an earlier year is associated with a write-off in the current year, after controlling for variables

that have been found to be important in prior studies. We find a statistically significant

positive association between these two decisions. The evidence is consistent with the theory

of cognitive dissonance as an explanation of management decisions about write-offs of

goodwill.

This paper responds to calls to go beyond agency theory in attempting to understand

management behavior as it is recorded in the financial accounting information. We highlight

the possibility that potentially biased financial information may result from causes other than

a deliberate attempt to mislead—in this case, resulting instead from a psychological bias to

which most human beings are subject. A broadened understanding of the potential for such

biases and their impact is particularly important in areas such as goodwill write-offs where

some degree of subjectively is inherent in the determination of the fair values of business

units and the decisions to record (or not to record) an impairment. Thus, our contribution is

important to standard setters and others in the profession as they consider the trade-offs

between more relevant fair value information and other more reliable measures. A further

contribution of our study is the introduction of a theory, cognitive dissonance, which has been

used in other disciplines and strongly recommended for use in accounting research. We show

that this theory from psychology can be usefully applied in an accounting setting.

The remainder of the paper is organized as follows. In the next section, we briefly

review accounting for goodwill, with emphasis on relatively recent changes in reporting

standards intended to implement fair value accounting in this area. Following, we review

literature and develop hypotheses, first with respect to goodwill write-offs and agency theory

6

and then for cognitive dissonance. Next, we detail the methodology employed in this study

and describe our sample. Finally, we provide the results of our analyses and discuss the

implications.

2. Accounting for Goodwill under SFAS No. 142

Recently, the Financial Accounting Standards Board (FASB), in concert with the

International Accounting Standards Board, has moved toward fully adopting fair value

accounting for financial instruments and certain other items (FASB 2007). Academics (e.g.,

Benston, 2008; Holthausen and Watts, 2001; Watts, 2003) as well as the financial press (e.g.,

Henry and Kopecki, 2004; Rockness et al., 2001) have expressed concerns that unverifiable

fair value accounting will make accounting numbers “soft” and less useful for investment

decisions.1 Other authors recognize the trade-off between the usefulness of fair value

information and the reliability of cost information (see Quagli and Avallone, 2010). An

important aspect of the debate pertaining to fair value accounting is asset write-downs. As

noted in Elliott and Shaw (1988, 92), asset impairment write-offs “differ from most financial

statement information because of greater discretion as to their magnitude and timing.”

Accounting for goodwill and particularly the determination of impairment losses are

excellent examples of the use of fair value accounting and the related decisions of when and

how much to record as impairment losses.

Goodwill can only be recorded as a result of the purchase of an entire firm (or a unit

of the firm if the goodwill is specific to a unit). Goodwill represents the difference between

the price paid for the acquisition and the sum of the fair values of the identifiable net assets.

The initial valuation of goodwill is relatively straightforward (as long as fair values of the net

1 To address these concerns, the FASB released SFAS No. 157, Fair Value Measurements (FASB 2006), whichdefines fair value as the price that would be received to sell an asset or paid to settle a liability in an orderlytransaction between market participants at the measurement date (i.e., exit values). However, SFAS No. 157does not fully resolve the issues related to the measurement of fair values of the asset or liability for which anactive market does not exist (e.g., goodwill). See also Benston (2008) for a discussion of SFAS No. 157.

7

identifiable assets are readily available) but valuation of goodwill subsequent to the

acquisition is inherently subjective because goodwill is not a separable economic asset.

Before the issuance of SFAS No. 142, Goodwill and Other Intangibles (FASB, 2001), this

subjective valuation of goodwill was avoided. Instead, the FASB mandated systematic

amortization over a maximum of 40 years. This requirement created burden for some firms

and a benefit for others. Firms which were successfully managing their acquisitions had to

periodically expense goodwill even though they may have been enhancing its value. Firms

with underperforming or troublesome acquisitions were potentially able to report far less

expense than economic circumstances might dictate.

To remedy the problems associated with amortization of goodwill, the FASB

promulgated SFAS No. 142.2 In place of systematic amortization, SFAS No. 142 requires

that goodwill be tested annually for impairment. On the surface, the new standard appears to

have resolved concerns about overvaluation and at least partially addressed issues of

undervaluation. However, the impairment test required by SFAS No. 142 revived the issue of

subjectivity. In the first step of this test, the fair value of the reporting unit to which goodwill

has been assigned is estimated and compared with the book value of the unit. Fair values are

estimated using quoted market prices if available or using present value or other valuation

technique such as multiple of earnings. Since the fair value of goodwill is determined at the

reporting unit level (usually a level lower than that of the total entity), quoted market prices

are not available unless the reporting unit is a traded subsidiary, thus often requiring the use

of less objectively determinable fair value measurement techniques. If the book value of the

reporting unit exceeds its estimated fair value, goodwill may be impaired and step two of the

test—determination of the amount of the impairment—is triggered. The potential impairment

is the difference between the carrying amount of goodwill and its implied fair value. Similar

2 However, many articles in the business press regard SFAS No. 142 as a political product (for example, Weil2000).

8

to determining the amount of goodwill arising from an acquisition, the implied fair value of

goodwill is the excess of the fair value of the reporting unit over the fair values of the unit’s

identifiable net assets.

Thus, in addition to the subjectivity that is likely to be present in defining reporting

units and allocating newly created goodwill to these units, the impairment test includes two

additional layers of subjectivity by requiring the estimation of the fair value of the reporting

unit(s) as a whole and of the unit(s)’ assets and liabilities excluding goodwill (Hayn and

Hughes, 2006). While the fair values of the unit’s net assets were estimated on acquisition,

the process must be repeated, giving effect to changes in the composition of the net

identifiable assets and the passage of time.

As a result of fair value accounting and the impairment test, goodwill write-offs

represent a particularly informative area for a study of management decision making. The

timing of the decision to record an impairment loss and the measurement of the amount to be

reported both contain significant levels of subjectivity, making it possible that any decision

making biases would impact the outcome. Further, some characteristics of goodwill reporting

make this particular financial statement item especially useful for this type of study. The first

step of the impairment test must be conducted annually rather than only on the occurrence of

certain triggering events, as is the case for other assets write-offs (FASB, 1995). Thus,

goodwill accounting imposes a more exact reevaluation requirement, ensuring that the

decision to consider reporting an impairment must be made frequently and reflected in the

financial statements. In addition, we can limit our sample to firms that are most likely to be

required to consider goodwill write-offs, by selecting only those with a book-to-market ratio

greater than one. Other researchers (e.g., Beatty and Weber, 2006; Ramanna and Watts, 2012)

have argued this indicates that the book value is overstated (assuming that the market price is

efficient) and that at least some of the overstatement is due to an overstatement of goodwill.

9

Given the nature of goodwill, it is quite likely that a significant portion of the overstatement

relates to goodwill.

3. Competing hypotheses

3.1 Agency Theory and Goodwill Write-offs

Agency theory, according to the seminal work by Jensen and Meckling (1976),

concerns the relationship, generally established by an explicit contract, in which one party

(the principal) delegates work to another (the agent), who is compensated by the principal.

The theory was developed by incorporating the agency problem into the risk-sharing

paradigm (e.g., Arrow, 1971; Wilson, 1968) in the 1960s and early 1970s in the economics

literature (Eisenhardt, 1989). Research based on agency theory addresses two problems that

can occur in an agency relationship: the agency problem and the problem of risk sharing. The

first problem arises when (1) the goals of the principal and agent conflict and (2) it is difficult

or impossible for the principal to verify what the agent is actually doing. The second problem

arises when the principal and the agent have different preferences for actions due to different

attitudes toward risk.

Agency theory has become one of the most important theoretical backgrounds in

accounting research (Lambert, 2001; Watts and Zimmerman, 1986). This is not surprising

given that financial reporting is a primary means for managers (the agents) to communicate

information on the firm’s performance to external capital providers (the principals). To

facilitate this role of financial reporting, accounting standards often permit managers to

exercise judgment so that managers can effectively convey their private information to

outsiders in a timely manner. Ideally, managers would exercise their discretion in financial

reporting to better portray the firm’s financial position and performance but agency theory

suggests they may behave in a less helpful manner. Indeed, studies on earnings management

10

(e.g., Burgstahler and Dichev, 1997; Healy, 1985; Jones,1991;) provide evidence that

managers opportunistically choose reporting methods and estimates for a variety of self-

interested reasons. A recent survey of CFOs by Graham et al. (2005) confirms that managers

sacrifice long-term value to manage earnings in the short term in order to improve the

perceived value of the firm.

Agency theory has played an important role in the debate over the movement toward

fair value accounting. The alleged decline in usefulness as a result of fair value information

cited above is attributed to the fact that managers must exercise their discretion when they

estimate the fair value of an asset. The value to investors of reported fair values depends

critically on how those numbers are measured and the extent to which they are trustworthy.

Goodwill impairments provide an excellent example of this issue. According to agency

theory, it is likely that managers may avoid recording a write-off even when the

circumstances would suggest that one is warranted, in order to avoid negative stock market

reactions. Thus, as with write-offs of other assets, goodwill write-offs may be recorded only

when earnings are particularly high in order to smooth earnings or as part of a so-called big

bath (Zucca and Campbell, 1992), when other negative impacts on earnings cannot be

avoided, and write-offs are recorded in order to make earnings in subsequent years seem high

by comparison and to avoid write-offs in future years.

Criticisms of goodwill accounting due to the lack of verifiability (e.g., Ramanna, 2008;

Ramanna and Watts, 2012; Watts, 2003, 2006) implicitly and explicitly rely on agency theory

to predict managers’ behavior and suggest negative impacts on the reliability of the

information. For example, Ramanna (2008, 255) explicitly states that “from agency theory

we can expect at least some firms to use their discretion opportunistically.” At the heart of the

criticism found in these studies is that managers delay write-offs of goodwill, which is in

11

contrast with the FASB’s assertion that SFAS No. 142 will benefit investors by spurring

companies to provide better information regarding acquired goodwill.

There is strong empirical evidence suggesting that managers do delay recording

impairments of goodwill. Using data from both before (Hayn and Hughes, 2006) and after

(Ramanna, 2008) SFAS No. 142, it can be shown that evidence of the economic impairment

of goodwill surfaces in market and financial statement information prior to the actual booking

of the impairment. Thus, managers seem to be exercising some discretion with respect to the

timing of goodwill write-offs. Agency theory would suggest that the reason for the delay is

to avoid negative stock market reactions. Other studies have investigated managers’ decisions

to take write-offs of several assets at the same time, the big bath, intended to record write-offs

in the current year so that fewer will have to be reported in future years (Riedl, 2004). Along

with other discretionary write-offs such as restructuring charges, goodwill write-offs have

been shown to occur often in conjunction with a big bath (Francis et al., 1996).

Thus, agency theory would suggest that managers do opportunistically determine the

timing of write-offs of goodwill. They may delay the decision until such time as it becomes

possible to report the write-off during a period of strong earnings, thus avoiding negative

reactions or until a big bath is being taken, at which time the write-off of goodwill is only a

part of the negative information that is being released in the current period. In this case,

managers will tend to write-off inappropriately large amounts in order to reduce the

likelihood of further write-offs in subsequent years. This leads to the following hypothesis:

H1: Ceteris paribus, managers are less likely to write off goodwill when there

was a goodwill write-off in the previous year.

3.2 Cognitive Dissonance and Goodwill Write-offs

Cognitive dissonance is defined in the seminal work by Festinger (1962) as an

uncomfortable psychological state caused when a person holds two conflicting cognitions at

12

the same time. Festinger (1962) described cognitions very broadly as knowledge, opinions or

beliefs about oneself (one’s behavior, emotions, character, and wants or needs) and the

environment (the behavior of others, general economic conditions, social norms, etc.).

Cognitive dissonance is most likely to occur after a decision of some type has been made

(Festinger, 1962). For example, a when a person makes a decision that increases her own

payout at the expense of another’s, she may experience cognitive dissonance if she believes

in fairness as an important value (Konow, 2000).

In the decades following Festinger’s work, experiments in psychology have shown

that important boundary conditions exist in order for dissonance to be activated. Decisions

resulting in dissonance must have been freely taken (the decision maker has free will), there

must be commitment to the decision, some negative results of the decision must arise, and the

negative consequences must have been foreseeable (Cooper, 2007). The decision of

importance to the current study would be a decision to acquire a new business unit. The

boundary conditions are unmistakably met. An investment decision is within the scope of

management as a choice among available opportunities. Further, the act of investing

substantial funding in the new venture demonstrates commitment clearly and publicly. The

failure of the unit to produce the expected level of benefits to such a degree that a write-off of

goodwill would need to be considered is certainly a negative outcome. The potential for

negative results is endemic to any business decision and thus would have been, at least to

some degree, foreseeable.

A decision maker that experiences cognitive dissonance is often motivated to take

steps to reduce the discomfort. Unconscious efforts to bring the cognitions into agreement

will be centered around the beliefs or events that are least resistant to change. Since many

decisions are not easily reversed, information that conflicts with the decision may be judged

to be unimportant or inaccurate or may even be forgotten entirely (Festinger, 1962) and

13

efforts may be made to seek out information that is consonant with the decision, or to

increase the importance attached to such information (Wicklund and Brehm, 1976).

Information that is ambiguous is particularly subject to this misinterpretation or devaluation

(Festinger, 1962). For example, in an experimental setting, Jermias (2001) found that

participants who had become committed to a costing system were more likely to ignore

negative feedback and less likely to be willing to change the costing system. Because efforts

to reduce cognitive dissonance involve focusing on information that supports the decision or

adjusting attitudes to conform to the decision, the confidence of the decision maker in the

decision is often increased as a result (Festinger, 1962).

The more important the decision and the conflicting cognitions, the greater will be the

magnitude of the dissonance experienced (Festinger, 1962). In equation form, the magnitude

of dissonance can be represented as (Cooper, 2007, 9)

Sum (all discrepant cognitions their importance)Magnitude of DissonanceSum (all consonant cognitions their importance)

.

The decision to acquire a new business unit is quite likely viewed as important. Thus, any

information that tends to demonstrate that the decision was not a good one will result in

cognitive dissonance. As more and more information accrues suggesting the unit is

underperforming, the magnitude of the dissonance will increase and the manager will

experience a heightened need to somehow reduce the dissonance. Conflicting information

from a source that the decision maker expected to provide confirming information is likely to

have a stronger impact on the magnitude of dissonance (Wicklund and Brehm, 1976).

Accounting information, including the income of the business unit and its fair value, is

potentially such a source. However, the subjective nature of the fair value information related

to a write-off of goodwill may make it more easily devalued in the mind of the manager.

When dissonance becomes very high as a result of the inability to continue ignoring

conflicting cognitions, the decision maker may actually begin to seek out disconfirming

14

information as a prelude to changing or reducing the commitment to the original decision that

created the dissonance (Festinger, 1962). At this point, it may no longer be possible to avoid

a re-evaluation of the original decision, even if that comes with some cost. Experimental

evidence supports this conclusion (Wicklund and Brehm, 1976). When the evidence becomes

overwhelming that a business unit is not performing to expectations, it may be easier for

management to accept a write-off of goodwill than to continue to hold to the belief that the

decision was effective. This will result in a reduction of cognitive dissonance in the same way

that previous efforts to avoid the conflicting information did.

Once the original decision has been re-evaluated, cognitive dissonance has been

reduced. If no further changes are made, for example, if the original decision has not or

cannot be changed, dissatisfaction results (Oliver, 1997). Some evidence of such an effect

can be seen in a study of post-employment evaluations of the hiring firm by recent graduates.

Initially newly hired students demonstrated the effects of cognitive dissonance by ignoring

negative information and overemphasizing positive information, thus increasing their reports

of satisfaction with their firm above the levels they had stated immediately before the hire.

However, one year later, the average level of satisfaction with their firms had fallen. Three

and one half years later, there were no further changes in the evaluations, i.e. satisfaction was

still lower than immediately after taking the new job (Vroom and Deci, 1971). Thus goodwill

write-offs in years after the initial write-off will be made more objectively in light of the

existing evidence, without the bias caused by cognitive dissonance. Thus, theory of cognitive

dissonance would suggest that a write-off of goodwill is more likely following an initial

write-off which represents the point at which management was forced to re-evaluate the

performance of the business unit. This leads to the hypothesis:

H2: Ceteris paribus, managers are more likely to write off goodwill when

there was a goodwill write-off in the previous year.

15

4. Empirical Test

4.1 Research Design

We test our hypotheses by conducting a binary logistic regression with the probability

of a goodwill write-off in time t as the dependent variable and the existence of a goodwill

write-off in time t–1 as the independent variable. Following prior studies (e.g., Francis et al.,

1996; Riedl, 2004), we include a number of control variables that have been found to explain

write-offs of goodwill. Thus, our statistical model is as follows:

Prob (D_WOt =1) = a + b1BTMt + b2RET12t + b3∆NIt_NIt_∆CFOt + b4D_Bigbatht

+ 5D_∆MGTt + b6WOnon-GWt + b7D_WOt−1 + yearly fixed effects +

et

where

D_WOt = an indicator variable equal to one if the firm reported goodwill write-

offs in year t, and zero otherwise;

BTMt = pre-write-down book-to-market ratio at the end of year t;

RET12t = buy-and-hold returns for 12 months ending three months after the end

of fiscal year t;

∆NIt_NIt_∆CFOt = the first factor from a principal component analysis of the following:

∆NIt (the change in pre-write-off earnings from year t−1 to t, deflated

by lagged total assets), NIt (pre-write-off earnings in year t, deflated

by lagged total assets), and ∆CFOt (the change in cash flows from

operation from year t−1 to t, deflated by lagged total assets);

D_Bigbatht = an indicator variable equal to one if the firm’s pre-write-off earnings

from year t−1 to t, deflated by lagged total assets, is below the median

of nonzero negative values of this variable, and zero otherwise;

16

D_∆MGTt = an indicator variable equal to one if any of the following top

management changed hands from year t−1 to t: chairman of the board,

CEO, and CFO, and zero otherwise;

WOnon-GWt = write-offs of non-goodwill assets (reflected as a positive amount) in

year t, deflated by lagged total assets.

D_WOt−1 = an indicator variable equal to one if the firm reported goodwill write-

offs in year t−1, and zero otherwise;

The first control variable, book-to-market ratio, captures the market’s assessment of

the assets of a company. This variable in various forms (e.g., industry-adjusted or change

rather than level) has been utilized in prior studies (e.g., Beatty and Weber, 2006; Francis et

al. 1996). We include this variable to capture the extent of goodwill impairment. A book-to-

market ratio greater than one implies that some assets are impaired. Clearly, not all

impairment is likely to be related to goodwill but given the nature of goodwill (i.e.,

intangible), it is likely to be responsible for a major portion of total impairment. We construct

this variable such that higher values suggest greater impairment. The coefficient on the book-

to-market ratio is predicted to be positive under both hypotheses. A second market-based

measure, the firm’s stock returns, is included to capture market-wide, industry-specific, and

firm-specific information. This variable has been extensively used in the asset write-off

literature (e.g., Beatty and Weber, 2006; Francis et al., 1996; Riedl, 2004) because it reflects

the firm’s current economic condition as well as market participants’ expectations about

future performance. We predict the coefficient on this variable to be negative under both

hypotheses because the better the stock performance, the smaller the likelihood of write-offs.

The next variable is included to capture the fundamental economic performance of the

firm. The first two components, change in net income and net income, capture accrual-related

performance of a firm and the last component, change in operating cash flows, captures cash-

17

related performance of the firm. These firm-specific performance measures have been used in

prior studies (e.g., Anantharaman, 2007; Riedl, 2004) as determinants of write-offs and have

been found to have a negative association with write-offs. Instead of using three measures

separately, we employ factor analysis as in Anantharaman (2007) to mitigate problems

related to collinearity between these variables.

The next two variables, the proxy for big bath and change in management, are added

to capture managers’ reporting incentives. Among various proxies for big bath, we closely

follow the one used in Riedl (2004).3 To identify change in top management, we compare a

firm’s annual report for a particular year and that for the previous year.4 It has been

documented that (1) managers engage in big bath, i.e., they write off their assets when their

earnings are below normal earnings (e.g., Francis et al., 1996; Riedl, 2004; Zucca and

Campbell, 1992) and that (2) changes in top management tend to concur with asset write-offs

because new management has an incentive to “clear the deck” of impaired assets in order to

obtain more favorable future performance (e.g., Francis et al., 1996; Riedl, 2004; Strong and

Myer, 1987). While these two variables have been used in studies that rely on agency theory,

the behavior they document is also consistent with cognitive dissonance. When earnings are

unexpectedly low, managers can no longer ignore the information that conflicts with the

original decision and may reduce dissonance by changing their cognitions about the decision.

This results in willingness to record an impairment of goodwill. Likewise, with respect to

change in management, new managers have no reason to experience cognitive dissonance

with respect to a decision that was not theirs. Thus, they are more likely to accept that assets

3 As a sensitivity test, we use an incidence of loss (before write-down, if any) as an alternative proxy for bigbath. The results are qualitatively and quantitatively consistent with those reported here.4 We follow Francis et al. (1996) in defining top management. However, Graham et al. (2005) suggest that theCFO is mainly in charge of financial reporting choices. Thus, we repeat the analysis using only change in CFOas the measure of change in top management. The tenor of the results does not change.

18

in their firm are impaired. Therefore, under both hypotheses, coefficients on these two

variables are expected to be positive.

Our next control variable is the write-off of assets other than goodwill. Our sampling

procedure, to be explained below in more detail, limits our sample to firms with a book-to-

market ratio greater than one. Our implicit assumption is that a ratio greater than one

indicates that some assets are impaired. A write-off of assets other than goodwill implies that

at least part of impairment is unrelated to goodwill, suggesting that managers are less likely

to write off goodwill. On the other hand, the need to write off assets of any type might

indicate a need to write off goodwill as well, since goodwill tends to deteriorate when other

assets are impaired. Based on these two possible scenarios, we do not predict a sign for the

coefficient on write-off of non-goodwill assets.

Finally, we include our variable of interest, write-off of goodwill in the previous year,

in the model. For all of the control variables, as stated above, our two theories predict the

same signs on the coefficients. For this variable, the two theories suggest differing

predictions. Under agency theory, managers are less likely to write off goodwill in the current

year when they have already written off goodwill in the previous year, leading us to predict a

negative coefficient. On the contrary, under cognitive dissonance, managers are more likely

to write off goodwill in the current year when they have written off goodwill in the previous

year, suggesting a positive coefficient on this variable.

4.2 Sample

Table 1 summarizes the sample selection procedure. In total, 22,092 firm-year

observations with a positive beginning goodwill balance under the post-SFAS No. 142

regime were retrieved from the 2011 edition of the merged Compustat/CRSP annual

fundamental file. Using the beginning (rather than the ending) goodwill balance ensures that

19

we include firms that write off the entire balance of goodwill during a particular year.5 From

this initial sample, we remove firms in the utilities (SIC code: 4900-4999) and financial (SIC

code: 6000-6999) industries. We further remove firms that do not have the following

Compustat data: total equity, fiscal year-end share price, number of shares outstanding, net

income, lagged net income, cash flows from operating activities, and lagged total assets. We

also remove firms with non-positive sales and non-positive total assets. This results in 21,631

firm-year observations representing 4,114 firms.

INSERT TABLE 1 HERE

Next, to increase the statistical power of our estimation, we limit our sample to firms

that are most likely to be required to report goodwill impairment losses. Following prior

studies (e.g., Beatty and Weber, 2006; Ramanna and Watts, 2012), we exclude firms whose

book-to-market ratio before impairment is less than or equal to one. Next, we exclude firms

with a fiscal year that ends before December 2004 to ensure that year t−1 does not belong to

the year in which SFAS No. 142 was adopted. We do so because reporting requirements (and

economic incentives) for goodwill impairment losses in a transition year differ from those in

subsequent years (Beatty and Weber 2006). We then require 12-month returns data over the

period that ends three months after the fiscal-year end from CRSP. Finally, we require firms

to have filed an annual report in EDGAR for the current year and the immediately preceding

year in order to identify changes in top management. This results in a final sample of 2,274

firm-year observations representing 1,306 firms.

4.3 Descriptive Statistics

Descriptive statistics are presented in Table 2. For the pooled observations reported in

Panel A, about 43 percent of the sample actually wrote off goodwill. Not surprisingly, the

5 Our final sample includes 305 observations with an ending goodwill balance of zero.

20

mean and median 12-month returns are negative, which explains the high book-to-market

ratio of these firms. Mean and median of key performance measures are negative, with the

exception of change in the cash flows from operations. Approximately 37 percent of firms

experienced change in their top management and about 20 percent of our sample had write-

offs of goodwill in the previous year. Panel B presents descriptive statistics for write-off

observations and for non-write-off observations. Comparison of the statistics clearly suggests

that these two groups are different. For example, the higher BTM averages indicate that the

market believes assets of write-off observations to be more impaired than those of non-write-

off observations and 12-month returns for the former are more negative than those for the

latter. In fact, all the indicators of firm performance suggest that write-off observations

performed poorly relative to non-write-off observations.

INSERT TABLE 2 HERE

Panel A of Table 3 reports industry classifications of sample firms using two-digit

SIC codes. We do not have an ex ante reason to believe that managers in a particular industry

are more likely to write off their goodwill than those in other industries and the results

reported in Panel A are consistent with our conjecture. The proportion of write-off

observations in a particular industry does not deviate much from the proportion of write-offs

for the entire sample, 43.4 percent. It ranges from 36.6 percent in the chemical, biotech, and

drug industries to 52.5 percent in the communications industry. Panel B reports sample

composition by year and clearly suggests that the recent economic downturn has had an

impact on the sample composition. Approximately 40 percent of our sample firms (925 out of

2,274) are from a single year--2008, the year the U. S. economy was hard hit. The years

immediately before and after 2008, i.e., years 2007 and 2009, also contribute more to our

sample than other years. 2008 is also the year with the highest proportion of write-off

observations: 56.3 percent of sample firms in that year reported a write-off of goodwill.

21

Taken together, the results reported in Panel B justify the inclusion of yearly fixed effects in

our model.

INSERT TABLE 3 HERE

4.4 Results

Table 4 contains pairwise correlations for the variables of interest. As can be seen,

nearly all of the control variables are significantly related to the decision to write off goodwill

in the current year. In addition, the signs on the correlations for these variables are as

predicted. Only write-offs of assets other than goodwill is not strongly correlated with the

write-off decision. It is important to note that the positive, significant correlation for change

in management, while consistent with agency theory, is a result that is explicitly predicted by

the theory of cognitive dissonance. Finally, as predicted by the theory of cognitive dissonance,

there is a significant, positive correlation of write-off in the prior year with the decision to

write-off goodwill in the current year.

INSERT TABLE 4 HERE

Table 5 contains a univariate test of the two hypotheses. As stated, agency theory

predicts that the probability of a write-off is less in the current year given that a write-off

occurred in the prior year. The theory of cognitive dissonance predicts the opposite: given a

write-off in the prior year, the probability of a write-off in the current year is greater. In our

sample, the probability of a write-off is 43.4%. Given a write-off in the prior year, that

probability increases to 52.1%. A chi-square test confirms that this difference is statistically

significant at p < .01. Thus, the univariate analysis provides support for the theory of

cognitive dissonance.

INSERT TABLE 5 HERE

22

Table 6 contains the results of the multivariate analysis. We conduct a logistical

regression with write-off in the current year as the dependent variable and write-off in the

prior year as the independent variable. The control variables as specified above are included

as well as indicator variables for each year included in the sample to control for

macroecnomic shooks. The results of the analysis are consistent with the univariate analysis:

the coefficient on write-off in the prior year is positive and highly significant, indicating that,

as suggested by the theory of cognitive dissonance, write-offs in the prior year indicate that

management has recognized the failure of the investment to perform as expected and no

longer suffers from the bias created by cognitive dissonance. Thus, the decision to record a

write-off in the current year was more easily taken. Further, the coefficient on change in

management, which is positive and significant, provides support for the theory of cognitive

dissonance. New managers are not likely to suffer from cognitive dissonance related to a

decision made by prior managers and are thus willing to record a write-off if the appropriate

tests indicate that goodwill has been impaired.

The coefficients on the variables for the years clearly show that write-offs were much

more common in 2008 and 2009. As discussed above, this is likely the result of the difficult

economic situation in the United States around that time. The coefficients on the control

variables are significant and the signs are as expected. Consistent with prior studies, the

existence of a high book-to-market ratio, low returns, or financial statement indicators of

economic stress can be expected to increase the likelihood of write-offs of goodwill.

Evidence that a big bath is occurring is strongly related to goodwill write-offs. Finally, if

other assets are being written off, goodwill is less likely to be written off.

INSERT TABLE 6 HERE

5. Conclusions

23

This study investigates competing hypotheses: one based on agency theory and the

other based on the theory of cognitive dissonance. We provide empirical evidence that is

consistent with the latter and inconsistent with the former. Thus, it appears that cognitive

dissonance explains management behavior with respect to the decision to record an

impairment of goodwill at least as well as, if not better than, agency theory. We make no

claim that the theory of cognitive dissonance uniquely explains management behavior.

Perhaps the most important result of this study is to emphasize that managers, being human

beings, are subjects to a broad range of influences, motivations, and biases. Thus, no one

theory can be expected to explain the behavior of all managers for all decisions. This finding

has implications for financial statement analysis, regulators, and accounting researchers.

Clearly, neither this evidence nor any other similar type of study can prove or

disprove either agency theory or the theory of cognitive dissonance. In order to do that, one

would have to have access to the cognitions of managers as they make a decision about

reporting a write-off of goodwill. Agency theory assumes that managers may deliberately

provide misleading information in order to increase the firm’s market value and, therefore,

their personal wealth. The theory of cognitive dissonance assumes that managers have an

unconscious bias against accepting evidence that contradicts a decision made at an earlier

time. In either case, the prediction for the current year is the same—managers will write-off

goodwill only under a specific set of circumstances. Our analysis confirms that the situation

suggested by cognitive dissonance, previous write-offs which indicate that management has

been forced by overwhelming evidence to accept the underperformance of the investment, is

a highly significant predictor of write-offs even after controlling for other variables that have

been found in prior studies relying on agency theory to predict write-offs.

While the difference may seem to be minor, it is important in terms of understanding

financial statements. The prevailing view has often seemed to be that any misstatements or

24

biases in the financial statements represent deliberate attempts by management to mislead

investors. Without a doubt, this may sometimes be the case. However, it may also be the case

that managers unintentionally incorporate errors in judgment into the firm’s external reports.

For those who analyze financial statements, the presumption that information will be

misleading to the maximum degree possible and in a specified direction may result in

assuming distortions and untruths that do not exist, leading to “corrections” that leave the

resultant information more distorted than ever. A more nuanced understanding, that some

managers may mislead, others may be reporting subject to unknown biases, and others may

be reporting quite accurate information may, in the long run, be more effective and lead to a

better understanding of financial accounting information.

For regulators, a failure to better understand the broad range of impacts on

management behavior may be similarly detrimental. For example, the trend toward use of fair

values instead of historical costs has been criticized based on the presumption that managers

will take advantage of the subjective nature of fair value to accounting to mislead investors.

This argument may result in the more relevant fair value information being eschewed in favor

of cost information that is suggested to be more reliable. However, if, as this paper

demonstrates, agency theory is not the only impact on management decision making, the

unreliable nature of fair value accounting may have been exaggerated. The cost benefit

calculations with respect to this decision might be reconsidered.

This study is subject to limitations as is any similar study. We employ a model that

assumes that the reasons for managers’ decisions about financial reporting can be uncovered

through analysis of archival information. Because there are many other impacts on this data

(in this study, for example, the year variables demonstrate this point) other than managers’

cognitions, there is a bit of noise in the data. Because other studies are also subject to this

limitation, our results are comparable. We believe that we provide evidence that is

25

informative about the behavior of some managers, on average, but can make no claim about

the reasons for the behavior of each individual manager.

We employ proxies for several constructs such a big bath and financial difficulties.

These proxies are the same as those employed in prior studies. However, they remain proxies

and are therefore estimations, not facts. To the extent that these proxies may have biased the

results, our findings are weakened. However, the highly significant coefficient on the

variable of interest in this study provides some confidence that the impact is not critical.

26

REFERENCES

Anantharman, D. (2007) Has SFAS 142 diminished the reliability of goodwill write-off?Working paper, Columbia University.

Arrow, K. J. (1971) Essays in the theory of risk bearing (Chicago, IL: Markham PublishingCo).

Baiman, S. (1990) Agency research in managerial accounting: A second look, Accounting,Organizations and Society, 15 (4), pp. 341−371.

Beatty, A. and Weber, J. (2006) Accounting discretion in fair value estimates: Anexamination of SFAS 142 goodwill impairments, Journal of Accounting Research, 44(2), pp. 257−288.

Benston, G.J. (2008) The shortcomings of fair-value accounting described in SFAS 157,Journal of Accounting and Public Policy, 27 (2), pp. 101−114.

Birnberg, J. G., Luft, J. and Shields, M. D. (2007) Psychology Theory in ManagementAccounting Research. In Handbook of Management Accounting Research, Volume 1,edited by C. S. Chapman, A. G. Hopwood, and M. D. Shields, pp. 113−136(Amsterdam: Elsevier Ltd.).

Brown, J. L., Evans, III J. E. and Moser, D. V. (2009) Agency theory and participativebudgetingexperiments, Journal of Management Accounting Research, 21, pp. 317−345.

Burgstahler, D. and Dichev, I. (1997) Earnings management to avoid earnings decreases andlosses, Journal of Accounting and Economics, 24 (1), pp. 99−126.

Camerer, C. F. and Loewenstein, G. (2004) Behavioral Economics: Past, Present, Future. InAdvances in Behavioral Economics, edited by C. F. Camerer, G. Loewenstein, andMatthew Rabin, pp. 3 – 51 (New York, NY: Princeton University Press).

Cooper, J. (2007) Cognitive Dissonance: Fifty Years of a Classic Theory (Los Angeles, CA:Sage Publications).

Cummings, W. H. and Venkatesan, M. (1976) Cognitive dissonance and consumer behavior:A Review of the evidence, Journal of Marketing Research, 13, pp. 303−308.

Eisenhardt, K. M. (1989) Agency theory: An assessment and review, The Academy ofManagement Review, 14 (1), pp. 57−74.

Elliott, J. A. and Shaw, W. H. (1988) Write-offs as accounting procedures to manageperceptions, Journal of Accounting Research, 26 (Supplement), pp. 91−119.

Festinger, L. (1962) A theory of cognitive dissonance (Stanford, CA: Stanford UniversityPress).

Financial Accounting Standards Board (FASB) (1995) Statement of Financial AccountingStandards No. 121: Accounting for the Impairment of Long-Lived Assets and forLong-Lived Assets to Be Disposed Of (Norwalk, CT: FASB).

Financial Accounting Standards Board (FASB) (2001) Statement of Financial AccountingStandards No. 142: Goodwill and Other Intangibles (Norwalk, CT: FASB).

Financial Accounting Standards Board (FASB) (2006) Statement of Financial AccountingStandards No. 157: Fair Value Measurements (Norwalk, CT: FASB).

27

Financial Accounting Standards Board (FASB) (2007) Statement of Financial AccountingStandards No. 159: The Fair Value Option for Financial Assets and FinancialLiabilities (Norwalk, CT: FASB).

Francis, J., Hanna, J. D. and Vincent, L. (1996) Causes and effects of discretionary assetwrite-offs, Journal of Accounting Research, 34 (Supplement), pp. 117−134.

Gilad, B., Kaish, S and Loeb, P. D. (1987) Cognitive dissonance and utility maximization,Journal of Economic Behavior and Organization, 8, pp. 61−73.

Graham, J. R., Harvey, C. R. and Rajgopal, S. (2005) The economic implications of corporatefinancial reporting, Journal of Accounting and Economics, 40 (1−3), pp. 3−73.

Hamberg, M., Paananen, M. and Novak, J. (2011) The adoption of IRFS 3: The effects ofmanagerial discretion and stock market reactions, European Accounting Review,20(2), pp. 263-288.

Hayn, C. and Hughes, P. (2006) Leading indicator of goodwill impairment, Journal ofAccounting, Auditing and Finance, 21 (3), pp. 223−265.

Healy, P. (1985) The effect of bonus schemes on accounting decisions, Journal of Accountingand Economics, 7 (1−3), pp. 85−107.

Henry, D. and Kopecki, D. (2004) The snag in accounting-made-simple; Companies may beexploiting new standards for more accurately valuing assets, Business Week (June 04),p. 82.

Holthausen, R. and Watts, R. L. (2001) The relevance of the value-relevance literature forfinancial accounting standard setting, Journal of Accounting and Economics, 31 (1−3),pp. 3−75.

Jensen, M. C. and Meckling, W. H. (1976) Theory of the firm: Managerial behavior, agencycosts and ownership structure, Journal of Financial Economics, 3 (4), pp. 305−360.

Jermias, J. (2001) Cognitive dissonance and resistance to change: The influence ofcommitment confirmation and feedback on judgment usefulness of accounting system,Accounting, Organizations and Society, 26, pp. 141−160.

Jones, J. J. (1991) Earnings management during import relief investigations, Journal ofAccounting Research, 29 (2), pp. 193−228.

Kachelmeier, S. J. (2010) Introduction to a forum on individual differences in accountingbehavior, The Accounting Review, 85 (4), pp. 1127−1128.

Konow, J. (2000) Fair shares: Accountability and cognitive dissonance in allocation decisions,The American Economic Review, 90 (4), pp. 1072−1091.

Koonce, L. and Mercer, M. (2005) Using psychology theories in archival financialaccounting research, Journal of Accounting Research, 24, pp. 175−214.

Lambert, R. A. (2001) Contracting theory and accounting, Journal of Accounting andEconomics, 32 (1−3), pp. 3−87.

Oliver, R. L. (1997) Satisfaction: A Behavioral Perspective on the consumer (New York, NY:McGraw-Hill).

Quagli, A. and Avallone, F. (2010) Fair value or cost model? Drivers of choice for IAS 40 inthe real estate industry, European Accounting Review, 19(3), pp. 461-493.

28

Ramanna, K. (2008) The implications of unverifiable fair-value accounting: Evidence fromthe political economy of goodwill accounting, Journal of Accounting and Economics,45 (2−3), pp. 253−281.

Ramanna, K. and Watts, R. L. (2012) Evidence on the use of unverifiable estimates inrequired goodwill impairment, Review of Accounting Studies, 17 (4), pp. 749−780.

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

Rockness, J. W., Rockness, H. O. and Ivancevich, S. H. (2001) The M&A game changes,Financial Executives, 17 (October), pp. 22−25.

Shiller, R. J. (2003) Form efficient markets theory to behavioral finance, Journal ofEconomic Perspectives, 17 (1), pp. 83−104.

Strong, J. S. and Meyer, J. R. (1987) Asset writedowns: Managerial incentives and securityreturns, The Journal of Finance, 42 (3), pp. 643−661.

Vroom, V. H. and Deci, E. L. (1971) The stability of post-decision dissonance: A follow-upstudy of the job attitudes of business school graduates, Organizational Behavior andHuman Performance, 6, pp. 26−49.

Watts, R. L. (2003) Conservatism in accounting part I: Explanations and implications,Accounting Horizons, 17 (3), pp. 207−221.

Watts, R. L. (2006) What has the invisible hand achieved? Accounting and BusinessResearch, 36 (Special Issue), pp. 51−61.

Watts, R. L. and Zimmerman, J. L. (1986) Positive accounting theory (Upper Saddle River,NJ: Prentice Hall).

Weil, J. (2000) FASB backs down on goodwill accounting rules, The Wall Street Journal,(December 7), A2.

Wicklund, R. A. and Brehm, J. W. (1976) Perspectives on Cognitive Dissoance (Hillsdale,NJ: Lawrence Erlbaum Associates).

Wilson, R. (1968) On the theory of syndicates, Econometrica, 36 (1), pp. 119−132.

Zucca, L. J. and Campbell, D. R. (1992) A closer look at discretionary writedowns ofimpaired assets, Accounting Horizons 6 (3), pp. 31−41.

29

Table 1: Sample selection

Description Number offirms

Number offirm-years

Firms listed with positive beginning goodwill balance onCompustat from 2003 to 2011

5,169 22,092

After excluding utilities and financial institutions (SIC codes:4900-4999 and 6000-6999)

4,148 21,829

After excluding firms with missing Compustat dataa 4,114 21,631After excluding firms with pre-write-off book-to-market less thanor equal to one

1,566 2,948

After excluding firms with fiscal year that ends before December2004

1,436 2,643

After excluding firms with missing returns data from CRSP 1,322 2,302After excluding firms with missing annual report in EDGAR 1,306 2,274

a We require total equity, fiscal year-end share price, number of shares outstanding, net income, lagged netincome, cash flows from operating activities, and lagged total assets. We also exclude firms with non-positivesales and non-positive total assets.

30

Table 2: Descriptive statisticsPanel A: Pooled observations (n = 2,274)

Variablea Mean SD Q1 Median Q3

D_WOt 0.434 0.496 0.000 0.000 1.000

BTMt 1.871 1.659 1.140 1.386 1.930

Ret12t −0.203 0.734 −0.624 −0.366 −0.023

∆NIt −0.024 0.324 −0.072 −0.017 0.014

NIt −0.033 0.311 −0.070 −0.001 0.029

∆CFOt 0.021 0.543 −0.049 −0.006 0.032

∆NIt_NIt_∆CFOt 0.000 1.000 −0.123 0.032 0.123

D_Bigbatht 0.324 0.468 0.000 0.000 1.000

D_∆MGTt 0.367 0.482 0.000 0.000 1.000

WOnon-gwt 0.008 0.063 0.000 0.000 0.001

D_WO t−1 0.198 0.399 0.000 0.000 0.000

Panel B: Writeoff observations (n = 988) vs. Non-writeoff observations (n = 1,286)

Writeoff observations Non-writeoff observations

Variablea Mean Median SD Mean Median SD

BTMt 2.312*** 1.628*** 2.256 1.533 1.254 0.833

Ret12t −0.333*** −0.511*** 0.775 −0.103 −0.258 0.685

∆NIt −0.064*** −0.043*** 0.260 0.006 −0.004 0.363

NIt −0.072*** −0.037*** 0.269 −0.003 0.012 0.336

∆CFOt −0.006** −0.009** 0.206 0.042 −0.004 0.699

∆NIt_NIt_∆CFOt −0.119*** −0.048*** 0.747 0.092 0.075 1.149

D_Bigbatht 0.465*** 0.000*** 0.499 0.215 0.000 0.411

D_∆MGTt 0.410*** 0.000*** 0.492 0.334 0.000 0.472

WOnon-gwt 0.010 0.000*** 0.050 0.007 0.000 0.072

D_WO t−1 0.238*** 0.000*** 0.426 0.168 0.000 0.374

a Variable definitions: D_WOt is an indicator variable equal to one if the firm reported goodwill write-offs inyear t, and zero otherwise. BTMt is pre-write-down book-to-market ratio at the end of year t. RET12t is buy-and-hold returns for 12 months ending three months after the end of fiscal year t. ∆NIt is the change in pre-write-off earnings from year t−1 to t, deflated by lagged total assets. Nit is pre-write-off earnings in year t,deflated by lagged total assets. ∆CFOt is the change in cash flows from operation from year t−1 to t, deflated bylagged total assets. ∆NIt_NIt_∆CFOt is the first factor from a principal component analysis of the above-mentioned three performance measures. D_Bigbatht is an indicator variable equal to one if the firm’s pre-write-off earnings from year t−1 to t, deflated by lagged total assets, is below the median of nonzero negative valuesof this variable, and zero otherwise. D_∆MGTt is an indicator variable equal to one if any of the following topmanagement changed hands from year t−1 to t: chairman of the board, CEO, and CFO, and zero otherwise.WOnon-GW

t is write-offs of non-goodwill assets (reflected as a positive amount) in year t, deflated by lagged totalassets. D_WOt-1 is an indicator variable equal to one if the firm reported goodwill write-offs in year t−1, andzero otherwise.

31

*, **, *** denote two-tailed significance levels at the 10%, 5%, and 1% level, respectively for two-sample t-tests(for mean) and two-sample Wilcoxon rank-sums tests (for median).

32

Table 3: Sample compositionPanel A: Sample composition by industry

Writeoff observations Non-writeoffobservations

Industry (two-digit SIC code) n % n % Total

Oil & gas extraction (13) 33 47.8% 36 52.2% 69

Chemical, biotech, & drug (28) 34 36.6% 59 63.4% 93

Computer software (35) 49 38.6% 78 61.4% 127

Electronic equipment (36) 147 48.8% 154 51.2% 301

Transportation equipment (37) 28 41.8% 39 58.2% 67

Medical & scientific instruments (38) 52 42.3% 71 57.7% 123

Communications (48) 73 52.5% 66 47.5% 139

Whole sale-durable goods (50) 29 41.4% 41 58.6% 70

Miscellaneous retail (59) 27 39.1% 42 60.9% 69

Business services (70) 140 47.5% 155 52.5% 295

All other industries 376 40.8% 545 59.2% 921

Panel B: Sample composition by year

Writeoff observations Non-writeoffobservations

Year n % n % Total

2004 31 26.5% 86 73.5% 117

2005 61 36.1% 108 63.9% 169

2006 49 34.0% 95 66.0% 144

2007 100 33.9% 195 66.1% 295

2008 521 56.3% 404 43.7% 925

2009 157 43.6% 203 56.4% 360

2010 59 26.1% 167 73.9% 226

2011a 10 26.3% 28 73.7% 38

a We require 12 month returns data over the period that ends three months after the fiscal year-end. Currently,returns data are only available up to December 2011, which effectively eliminates firms with fiscal year thatends later than September, 2011 from our sample.

33

Table 4: Pearson correlations coefficientsVariablea D_WOt BTMt Ret12t ∆NIt_NIt_∆CFOt D_Bigbatht D_∆MGTt WOnon-gw

t

BTMt 0.233***

Ret12t −0.155*** −0.195***

∆NIt_NIt_∆CFOt −0.105*** −0.067*** 0.078***

D_Bigbatht 0.264*** 0.126*** −0.179*** −0.249***

D_∆MGTt 0.079*** 0.013 −0.067*** −0.024 0.127***

WOnon-gwt 0.025 0.080*** −0.052** 0.032 0.101*** 0.038*

D_WOt−1 0.087*** 0.066*** 0.177*** 0.042** −0.064*** 0.022 0.041*

a Refer to Table 2 for the definition of variables.*, **, *** denote two-tailed statistical significance levels at the 10%, 5%, and 1% level, respectively

34

Table 5: Analysis of probability of write-offs given prior write-offsPrediction under agency theory

Pr(D_WOt = 1) > Pr(D_WOt = 1 | D_WOt−1 = 1)

Prediction under cognitive dissonance

Pr(D_WOt = 1) < Pr(D_WOt = 1 | D_WOt−1 = 1)

Pr(D_WOt = 1) 0.434

Pr(D_WOt = 1 | D_WOt−1 = 1) 0.521

Difference in probability 0.087***

χ2 51.34

*, **, *** denote two-tailed statistical significance levels at the 10%, 5%, and 1% level, respectively

35

Table 6: Multivariate logistic regression analysis of write-offsPredicted sign

Variablea Agencytheory

Cognitivedissonance Coefficient Z-statistic

Intercept none none −1.992*** 71.99

BTMt + + 0.361*** 46.63

Ret12t − − −0.292*** 11.01

∆NIt_NIt_∆CFOt − − −0.169* 3.36

D_Bigbatht + + 0.928*** 75.89

D_∆MGTt + + 0.225** 5.38

WOnon-gwt none none −2.282* 2.91

D_WOt-1 − + 0.687*** 32.07

Y2005 none none 0.293 1.13

Y2006 none none 0.276 0.93

Y2007 none none 0.125 0.24

Y2008 none none 0.821*** 12.32

Y2009 none none 0.762*** 8.80

Y2010 none none −0.102 0.14

Y2011 none none −0.284 0.42

Wald χ2 295.36***

Pseudo R2 0.218

N 2,274

a Refer to Table 2 for the definition of variables.*, **, *** denote two-tailed statistical significance levels at the 10%, 5%, and 1% level, respectively