Managing earnings management Compensation committees' treatment of by mkr13579

VIEWS: 20 PAGES: 46

									    Managing earnings management: Compensation committees’ treatment of earnings
                        components in CEOs’ terminal years†

                                        Mark R. Huson
                                      University of Alberta

                                           Yao Tian
                                      University of Alberta

                                      Christine Wiedman
                                     University of Waterloo

                                         Heather Wier
                                      University of Alberta




                                        Abstract
The incentive to manipulate earnings to enhance earnings-based compensation increases
in managers’ terminal years. We examine this horizon problem by considering the role of
the compensation committee in setting terminal-year compensation. We predict that
compensation committees are aware of the horizon problem and intervene when setting
pay in terminal years. We find that the relation between changes in cash compensation
and discretionary accruals changes is significantly reduced in the terminal years. This
relation is symmetric for both positive and negative accrual changes. We also find weak
evidence that the compensation committee views discretionary spending as less desirable
as the CEO approaches retirement. Finally, we document that the relation between the
level of option pay and the level of discretionary accruals is reduced in the terminal
years. Our findings suggest that compensation committees do not act naively when
setting CEO compensation in the terminal period and provide an explanation for the
limited evidence of accounting manipulation in CEOs’ terminal years.


JEL Classifications: M41, J33
Keywords: horizon problem, compensation committee, discretionary accruals, real
activity management


†
 Huson (mark.huson@ualberta.ca) is from the Department of Finance, and Tian (ytian3@bus.ualberta.ca)
and Wier (heather.wier@ualberta.ca) are from the Department of Accounting and Management
Information Systems at the University of Alberta School of Business, Edmonton, Alberta T6G 2R6,
Canada. Wiedman (cwiedman@watarts.uwaterloo.ca) is from the School of Accounting and Finance at the
University of Waterloo, Waterloo, Ontario, N2L 3G1, Canada.
I. Introduction

           The temptation confronting outgoing chief executive officers (CEOs) to inflate

income in order to enhance earnings-based compensation in their final years, referred to

as the “horizon problem”, has been extensively debated in the accounting literature.

Research examining terminal year accruals provides conflicting results. Pourciau (1993)

examines non-routine executive changes and finds no evidence that outgoing CEOs

manage total accruals and Murphy and Zimmerman (1993) report that accruals are

actually lower in transition years. Kalyta (2009) documents evidence of income-

increasing discretionary accruals in the CEO’s final years only when performance-based

compensation is a determinant of pensionable earnings. The evidence on whether

managers manipulate research and development (R&D) expenditures as they near the end

of their tenure with the firm is mixed (Dechow and Sloan 1991, Butler and Newman

1989, Gibbons and Murphy 1992, Murphy and Zimmerman 1993, and Cheng 2004).

           We reexamine the horizon problem using a sample of voluntary turnovers from

the Forbes compensation survey for the period 1995–2005.1 Unlike research that looks

for evidence of real or accrual-based earnings management, we use the relation between

earnings components and compensation in the terminal period to assess the magnitude of

the horizon problem. Our maintained joint hypotheses are that (i) compensation

committees can detect managed earnings and treat them differently than other earnings

components when setting pay and (ii) compensation committees are aware of the horizon

problem and exercise special judgment when determining compensation in CEOs’


1
    We thank Vincent Intintoli for sharing the sample firms used in Intintoli (2008).




                                                                                        1
terminal years. While the compensation committee may be unable to unravel managerial

attempts at income manipulation completely, they should have a significant advantage

over researchers because they are insiders and have specific knowledge of the firms and

managers that they monitor. Compensation committees’ unraveling of earnings

manipulation mitigates the horizon problem by reducing the expected payoff from

earnings and real activity management. This is a potential explanation for why there is

only limited evidence of real or accrual-based management in CEOs’ terminal years.

        Cheng (2004) reports evidence suggesting that the compensation committee

mitigates opportunistic reductions in R&D expenditures when CEOs are near retirement.

Specifically, he finds a significant increase in the association between changes in R&D

spending and changes in the value of CEO annual option grants when the CEO is at least

63 years of age. Our work compliments and extends that of Cheng in the following ways:

First, our results address both accrual-based earnings management and real activity

management over a number of expenditures. Second, while Cheng starts with the Forbes

compensation list, he restricts his sample to firms in eight R&D intensive industries, and

his results speak to one particular type of real earnings management (R&D expenditures)

where the expenditures are observable. We use a much broader sample, and therefore our

sample and our examination of earnings management are less specialized than those of

Cheng, making our results generalizable beyond R&D-intensive industries. Third, Cheng

focuses on the compensation committee’s use of option-based pay to affect managerial

behavior. Our focus is on cash compensation.2


2
 Cheng does provide an analysis of cash compensation and R&D activities, but this is not his main focus,
and his cash compensation results do not demonstrate statistical significance. Similarly, we provide an




                                                                                                      2
        Understanding how cash (non-option based) pay relates to the horizon problem is

important for at least two reasons: First, not all firms rely strongly on options to provide

incentives, and in fact the compensation mix has recently shifted away from option-based

pay. Cohen, Dey, and Lys (2008) document that, in their sample of 87,217 firm-year

observations, option pay peaked at 45% of total compensation in 2001, but declined back

to the 25% range by 2005. Second, as described later in the paper, recent work

demonstrates mixed results on the relation between the direction of earnings management

and option pay (see for example McAnally, Srivastava, and Weaver 2008, Cohen, Dey,

and Lys 2008, Cheng and Warfield 2005, Coles, Hertzel, and Kalpathy 2006, and

Bergstresser and Philippon 2006). The prospect of new grants provides incentives to

manage earnings down to obtain lower strike prices while the presence of vested and

maturing grants provides incentives to manage upwards to increase stock prices and the

proceeds from exercise.

        Like prior empirical work on compensation committees, we do not examine the

committees directly, but rather infer their decisions from the earnings/compensation

relation. The existing literature supports the contention that compensation committees

function knowledgably and do not simply apply the bonus formula mechanically to

bottom-line income (see, for example, Dechow, Huson, and Sloan 1994, Gaver and

Gaver 1998, Balsam 1998, and Adut, Cready and Lopez 2003). Much of the research has

focused on situations where management bonuses are shielded from the effect of negative

income items. We expect that the compensation committee will exercise special


analysis of option grants and earnings management, but the main focus of our paper is the relation between
earnings management and cash compensation.




                                                                                                        3
discretion in the terminal period not only to shield managers from the adverse effects of

taking needed write-downs but also to constrain upward earnings management.

       Our empirical results are consistent with our expectations and suggest that

compensation committees are aware of changed incentives in the terminal period and

exercise discretion accordingly. We find that the relation between changes in cash

compensation and changes in discretionary accruals is reduced in the terminal period.

This finding holds for both positive and negative changes in discretionary accruals,

suggesting that the compensation committee both shields managers for taking necessary

write-downs and fails to reward them for upward accrual-based earnings management.

This latter finding is important for two reasons. First, it is consistent with our hypothesis

that boards intervene to mitigate the horizon problem. Second, the existing literature

documents board intervention that increases compensation. To our knowledge this is the

first evidence of intervention that reduces compensation.

       We also observe a positive relation between changes in abnormal discretionary

spending and changes in cash compensation in the non-terminal period, but not in the

terminal period. This suggests that compensation committees view discretionary

spending as value-enhancing for early to mid-career CEOs, but not for CEOs nearing

retirement. Finally, using a subsample of firm-years with option pay, we also observe a

negative relation between the level of option pay and the level of discretionary accruals.

       The paper is organized as follows: Section II presents our hypotheses; Section III

describes our sample and data collection procedures; Section IV presents our results; and

Section V concludes the paper.




                                                                                             4
II. Hypothesis Development

Managerial Incentives: theory and evidence

       One problem with compensation policies that make incentive payments based on

accounting income is that they may encourage management to maximize short-term

income, and hence short-term compensation, at the expense of future income. The

incentive for such myopic behavior is particularly strong for a manager whose term with

the firm is ending, and who therefore does not benefit from future profits. Two

mechanisms for boosting income in this context have been examined: increasing

discretionary accruals, such as by inflating sales or understating depreciation expense and

cutting discretionary expenditures, such as research and development. Both actions

would result in improved performance in the current period at the expense of future

profits when the CEO is no longer in tenure.

       A long literature dating back to Jones (1991) indicates that firms manage the

discretionary portion of accruals to achieve accounting objectives. Specific to the horizon

problem, Pourciau (1993) examines the discretionary behavior of departing CEOs by

examining changes in total accruals and write-offs in their last year. Her sample consists

of 73 non-routine executive changes, including both voluntary and non-voluntary

resignations. She argues that because these changes are often unplanned, directors and

stockholders may find it more difficult to structure the turnover to minimize opportunities

and incentives for earnings management. In contrast to her predictions, however, she

finds no evidence of opportunistic behavior in the terminal period. In fact, she finds that

unexpected accruals are low in years t-1 and t, with higher values in year t+1, where year

t represents the year of the CEO’s departure. However, Pourciau calculates unexpected



                                                                                         5
accruals as the first difference in total accruals rather than using the discretionary

accruals models implemented more recently. Further, her measure may also be capturing

declining firm performance (Murphy and Zimmerman 1993). In contrast, Kalyta (2009)

finds evidence of accrual-based earning management in the terminal period by a

subsample of CEOs whose pension depends on performance-based pay (typically cash

and bonus) in their pre-retirement years.

        Dechow and Sloan (1991) use a sample of 91 R&D intensive firms to provide

evidence that managers nearing retirement take actions that may inhibit the long-run

profitability of the firm by cutting R&D expenditures. They report that the decrease in

R&D is not related to poor firm performance, and is most severe when the turnover is

planned (i.e., when the manager is 64 or 65). The authors conclude that executives

respond to earnings-based incentives and behave opportunistically in this context. In

contrast, Butler and Newman (1989), Gibbons and Murphy (1992) and Murphy and

Zimmerman (1993) do not find significant changes in R&D in executives’ terminal years.



Compensation Committees

        Compensation committees exist to design incentive plans and to exercise

judgment in their implementation, particularly in non-routine situations.3 Existing

literature in the area is consistent with the idea that compensation committees intervene

when merited by the circumstances to reward components of income differentially.

Balsam (1998) decomposes income into discretionary and nondiscretionary components


3
 See Dechow et al. (1994) for a more complete discussion of the role and composition of the compensation
committee.




                                                                                                      6
and finds that while discretionary accruals are associated with CEO cash compensation,

the relation is significantly lower than both the relation between non-discretionary

accruals and compensation, and the relation between operating cash flows and

compensation. Balsam notes that the association between cash compensation and

discretionary accruals is higher for positive than for negative discretionary accruals, and

demonstrates a strong positive correlation between positive discretionary accruals and

cash compensation when these accruals enable the firm to avoid reporting a loss. The

differential rewards on income components, which are sensitive to the firm’s financial

reporting choices, cause Balsam to conclude that compensation committees reward

managers when their accounting choices help the firm to achieve its reporting goals.

Balsam’s results focus on behavior for which the manager is rewarded, and demonstrate

no evidence of managers’ bonus pay being diminished in response to the compensation

committee’s belief that positive discretionary accruals are overly aggressive.4

        Dechow et al. (1994) provide further evidence of shielding behavior. They

examine the earnings/compensation relation for a sample of 182 restructuring charges

taken by 91 Fortune 500 firms. The majority of sample firms have compensation plans

that are based on earnings-based performance measures. However, none of these

earnings-based plans explicitly provide for the removal of the effect of restructuring

charges. Still, they find that on average, CEO compensation is shielded from

restructuring charges. This shielding is more likely to take place when the restructuring



4
 Balsam’s conclusions about the relative weights of the accrual and cash components of compensation are
driven by the shielding of negative discretionary accruals. This can be observed from Balsam’s estimates
of β3 and β3p in Table 3 of his paper.




                                                                                                      7
charges are nonrecurring and when the CEO has a shorter expected horizon with the firm.

Adut et al. (2003) find similar results for a more recent sample of restructurings.

        Gaver and Gaver (1998) provide evidence that, in general, compensation

committees do not intervene in a symmetric manner for similar types of gains and losses.

Gaver and Gaver demonstrate that cash compensation is significantly and positively

related to above the line earnings if those earnings are positive, but not if they are

negative.5 They find no distinction between the coefficients on above the line earnings

and below the line gains. The coefficient on below the line losses is insignificantly

different from zero. Therefore, gains appear to flow through to compensation while

losses do not.

        Baber et al. (1998) conduct a cross-sectional analysis of cash compensation and

earnings, and find that the sensitivity of compensation to earnings varies directly with

earnings persistence. More persistent components of earnings are weighted more heavily

in the compensation function. The authors do not address the question of the terminal

period directly. However, using age as a proxy for upcoming retirement, they find that

persistent earnings are most significantly rewarded for CEOs over age 60. Baber et al.’s

findings are consistent with compensation committees actively attempting to mitigate the

horizon problem by rewarding outgoing executives based on the long-term consequences

of their decisions even when their own tenure with the firm is limited.

        Similar to our work, Cheng (2004) examines the role of the compensation

committee in mitigating the horizon problem, but focuses only on R&D expenditures.


5
  Gaver and Gaver classify extraordinary items and the results of discontinued operations as below the line
items.




                                                                                                         8
Cheng demonstrates a significant positive relation between changes in option

compensation and changes in R&D expenditures as the executive’s terminal year

approaches, as well as at other times when management might exhibit myopic behavior.

Similar to Baber et al. (1998), Cheng does not confirm the specifics of CEO turnover, but

assumes that the horizon problem exists for CEOs who are at least 63 years old. Cheng’s

results do not hold for changes in cash or total compensation. Cheng hypothesizes that

option pay is the best means by which to motivate the CEO to take actions that increase

the long-term performance of the firm, like positive net present value R&D investments,

because CEOs can hold options after retirement.6

        Collectively, this research suggests that compensation committees do not adhere

strictly to earnings-based formulas when computing compensation, but rather take the

nature of the reported earnings figure and the context into account. In addition, Dechow

et al. (1994), Baber et al. (1998), and Cheng (2004) provide evidence suggesting that

compensation        committees       explicitly     consider     tenure      when      deciding     pay.

Correspondingly, we expect the compensation committee to recognize the horizon

problem and adjust pay to reflect the nature of the reported earnings.

        The existing literature on earnings management, while generally viewing the

phenomenon as evidence of opportunistic management behavior, also acknowledges that

earnings management can be beneficial for shareholders. For example, Subramanyam

(1996) and Bowan, Rajgopal, and Venkatathalam (2008) provide evidence consistent

with the observation that discretionary accruals and accounting discretion can signal

6
  Dahiya and Yermack (2008) document that, while some CEOs do continue to hold options after
retirement, companies have diverse “sunset” policies for modifying option terms on retirement that include




                                                                                                        9
future accounting performance. The signaling explanation is less likely for outgoing

CEOs for two reasons: first, it is not obvious that an exiting CEO can provide a credible

signal about an incoming CEO’s future performance; and second, the incentives to

manipulate earnings up in the terminal period are especially high because the possibility

of ex post settling up no longer exists.

         Since accrual-based manipulations will unwind, they essentially take earnings

from the succeeding CEO and attribute them to the exiting CEO. As such, the

compensation committee should place less weight on the discretionary component of

earnings when determining the outgoing CEO’s pay. The impact of changes in real

discretionary spending on CEO pay as retirement nears is not as clear. Some of the

existing literature views cuts to discretionary spending as detrimental to firm value. For

example, Graham, Harvey and Rajgopal (2005) survey financial executives and report

that these individuals are willing to manage real activities in order to achieve earnings

targets even if these actions are value-destroying in the long run. Alternatively, Chen, Lu,

and Sougiannis (2008) suggest firm value would be enhanced if at least some

discretionary spending were curbed. Chen et al. demonstrate that SG&A costs are

asymmetric, increasing more with sales growth than they decrease with declining sales.

This asymmetry is most evident in situations where agency costs are likely to be more

extreme (for example, when free cash flow is high), and is reduced by the presence of

strong corporate governance. Further evidence of SG&A’s potential for negative

valuation consequences is provided by Armstrong, Davila and Foster (2006), who

demonstrate that, after companies go public, SG&A costs are negatively related to equity


forfeiture or full immediate vesting.



                                                                                         10
value, and Banker, Huang, and Natarajan (2007), who show that the stock market prices

SG&A costs as expenses.

       Under the view that cuts to discretionary spending destroy firm value,

compensation committees should treat cuts to discretionary spending like accrual-based

earnings manipulations: since they increase reported earnings they should be netted out

of the earnings number used to calculate compensation. The alternative view that a

portion of discretionary spending (notably SG&A) is related to empire-building suggests

two things. First, it may not be appropriate to lump SG&A together with other

components of discretionary spending when considering their valuation consequences.

Second, “empire-building” discretionary spending may be treated differently over a

CEO’s career. Shleifer and Vishny (1989) suggest that expenditures that build on

managerial strengths can be value maximizing even when they do have entrenchment

value. While it is logical for the firm to encourage CEOs to develop projects that are

aligned with their talents early in their careers, continued expenditures that are correlated

with CEO specific human capital may not be in the interest of shareholders as retirement

nears. Accordingly, compensation committees may reward cuts to discretionary spending

in the terminal period.

       The above arguments lead to the following hypotheses, which provide a

directional prediction for accrual-based earnings management and a non-directional

prediction for real activity management.

               H1: Compensation committees place less weight on the
               discretionary accrual component of earnings in setting cash
               compensation in a CEO’s terminal years compared to earlier years.




                                                                                          11
                    H2: Compensation committees place a differential weight on the
                    discretionary expense component of earnings in setting cash
                    compensation in the CEO’s terminal years compared to earlier
                    years.



           We recognize that not all positive accruals are the result of manipulation and that

not all manipulation is necessarily detected. Additionally, some negative accruals result

from poor performance rather than managerial foresight. Further, not all discretionary

spending is perquisite consumption, and not all cuts of discretionary spending represent

earnings management. Consequently our tests reflect the compensation committee’s

assessment of the nature of the accruals and expenditures as well as the extent of its

intervention.

           While our hypotheses are about compensation in general, we make cash

compensation the primary focus of our paper for several reasons. First, for our sample of

companies from the Forbes compensation survey, cash salary plus bonus is significant

both in magnitude and as a proportion of the executive’s pay package. As we tabulate

later in the paper (see Table 1, Panel B), CEOs’ average cash compensation of $1.6

million is in excess of 50% of their total pay package in the non-terminal period of our

sample.7 We do include an analysis of option-based pay and earnings management for

completeness.

           Second, the earnings management incentives vis-à-vis option-based pay are not

clear. McAnally, Srivastava, and Weaver (2008) demonstrate that managers with option

pay may seek to miss earnings targets and that option grants following missed earnings


7
    This percentage drops to 43% in the terminal period.




                                                                                           12
targets tend to be larger and more valuable. Cohen, Dey, and Lys (2008) find that new

option grants are negatively associated with earnings management activity (though not

with real activity management).         By contrast, Cheng and Warfield (2005) find that

managers with high equity incentives are more likely to report earnings that meet or just

beat analysts’ forecasts, and Bergstresser and Philippon (2006) find that discretionary

accruals are used to manipulate reported earnings to a greater degree when more of the

CEO’s total compensation is tied to stock and option holdings.8 Baker, Collins and

Reitenga (2009) find evidence that discretionary accruals are lower when option pay is

high and firm performance poor, but only when firms infrequently issue grants following

earnings announcements. They do not observe these results for firms that follow a

predictable grant schedule.

        Third, Core and Guay (1999) argue that the purpose of equity grants is not solely

to reward performance. The authors demonstrate empirically that the firm’s decision

about the size of a new equity grant is in part a function of bringing the CEO to the

appropriate level of equity incentives from an optimal contracting perspective. Thus, the

incentive effect of equity grants muddies any attempt to measure the pay/performance

relation for these instruments.

        Finally, as Dahiya and Yermack (2008) report, corporate stock option plans have

diverse “sunset” provisions that potentially modify the terms of option grants on

retirement. For mid-career resignations, 95.6% of unvested options are forfeited. For later

career retirements, the forfeiture rate falls to 44%, and 33% of options vest immediately.


8
 Cheng and Warfield (2006) and Bergstresser and Philippon (2005) both use aggregate values of equity-
based pay which do not separately identify the values of option grants.




                                                                                                  13
The vesting schedule of only a small percentage of unvested options continues as

originally stipulated. For forfeited options, the value of terminal period grants is zero. For

options that vest immediately, the long-term incentive effect of options is diminished

such that their incentive effects become closer to those of a cash bonus.9



III. Data and Descriptive Statistics

Sample Selection

        We focus on voluntary CEO turnovers. There are several reasons for doing so.

First, previous research suggests that the circumstances surrounding CEO turnover

influence the opportunities to manipulate accruals. As Murphy and Zimmerman (1993)

explain, managers’ discretion over financial variables is constrained by firm

performance. This constraint is likely to be more binding with poorer firm performance,

and forced turnovers are more likely for poorly performing firms. Poor performance,

coupled with the lack of advance notice often associated with forced turnovers, reduces

the opportunity to manipulate earnings upward. Further, compensation policy differs for

voluntary and forced CEO departures. Yermack (2006) compares the severance pay for a

sample of forced and voluntary CEO departures, and finds that the mean severance

package for forced turnovers equals $15.1 million, over six times the average severance

package of $2.3 million that he observes for voluntary turnovers.10 Additionally, we


9
  Sunset provisions on option grants are a particular problem for us because we focus exclusively on CEO
turnovers. Research such as Cheng (2004) that examines the more general topic of managerial myopia is
less affected by the loss of incentives associated with options that are forfeited at retirement because his
sample includes firms that manage earnings to avoid a loss or a small earnings decline in addition to firms
where the CEO is facing retirement.
10
   Severance packages for voluntary turnovers will bias against our findings by introducing noise into the
compensation/earnings relation; however, for forced turnovers, the presence of severance packages likely




                                                                                                        14
expect any shielding of management compensation from negative charges taken as the

manager approaches the end of his tenure to be particularly evident in the sample of

voluntary turnovers.

        Table 1, Panel A outlines our sampling procedure. Our sample of voluntary CEO

turnovers over the period 1995-2005 is obtained from the database used in Intintoli

(2008). Intintoli uses the Forbes executive compensation surveys to identify CEO

changes, identifies announcement and actual succession dates in the Wall Street Journal,

and employs the procedures outlined in Parrino (1997) to identify forced and voluntary

turnovers. We remove from our sample firms for which we cannot identify a valid

GVKEY, observations without fiscal year-end data, firms in the financial services sector

(SIC codes 6000-6999), firms missing either return or financial statement data that is

necessary to our analysis, and CEOs with tenure of less than three valid years.11 We

elaborate below on our criteria for determining what a “valid” year is.

        In our sample a CEO’s tenure begins the first year in which he holds the position

of CEO for at least 9 months. We define the CEO’s terminal year (t) as the last fiscal year

where the CEO maintains control of the firm’s financial statements through the annual

general meeting (AGM) at which they are presented. We assume that, if a CEO leaves

office prior to the end of the third month of year t+1, he was not present at the AGM at

which the year t financials were presented. In this case, we define year t-1 as the terminal


renders any empirical evaluation of the pay/performance relation invalid. Forced turnovers tend to occur in
poorly performing firms, and also tend to result in very large severance packages. Combined, these facts
suggest that an empirical analysis of pay for forced turnovers is likely to generate the counter-intuitive
finding that poorer performance is rewarded with larger compensation in the terminal year.
11
   We remove CEOs who have tenure of less than three years because we require at least one prior period
for comparison with the terminal period. Since our specification is in terms of changes, this necessitates a
minimum of three years.




                                                                                                        15
year. We assume that if a CEO leaves office subsequent to the sixth month of year t+1, he

was present at the AGM at which the year t financials were presented, and define year t

as the terminal year. For all departures between three and six months after the fiscal year-

end, we check the date of the AGM; we then define year t as the terminal year if the CEO

continued in office through the AGM and year t-1 as the terminal year if the CEO

resigned before the AGM. The timeline below illustrates this process. This leaves us with

a final sample of 164 turnovers at 156 firms.



                                                     Months in Year t+1

 CEO left prior to preparing year t                                           CEO left after preparing year t
 financials. Last year is year t-1.                                           financials. Last year is year t.




          1        2          3          4         5          6           7          8            9        10    11   12



                       Check date of AGM.
                       If CEO exits prior to AGM then last year is t-1.
                       If CEO exits after AGM the last year is t.




Measuring earnings components

         Our first research question is concerned with whether the compensation

committee evaluates changes in discretionary accruals differently in the terminal period.

We estimate our measure of discretionary accruals using a modification of the Dechow,

Richardson, and Tuna (2003) forward looking Jones model (FLJM) estimated by year for




                                                                                                                      16
each two-digit SIC code.12 Total accruals (TA) are defined as the difference between net

income before extraordinary items (Compustat item #18) and cash flows from operations

(Compustat item #308).



TA
  i, t                    ⎡ 1 ⎤               ⎡ (1 + k )ΔSales i ,t − ΔARi ,t ⎤        ⎡ PPE i ,t ⎤        ⎡ TAi ,t −1 ⎤
                     = αt ⎢          ⎥ + β 1t ⎢                               ⎥ + β 2t ⎢          ⎥ + β 3t ⎢           ⎥ + ε i ,t (1)
         A                ⎢ Ai ,t −1 ⎥        ⎢                               ⎥        ⎢ Ai ,t −1 ⎥        ⎢ Ai ,t −1 ⎥
          i, t − 1        ⎣          ⎦        ⎣           Ai ,t −1            ⎦        ⎣          ⎦        ⎣           ⎦




Parameter k represents the slope coefficient from a regression of change in sales on

change in receivables, and captures the expected change in receivables for a given change

in sales. ΔSalesi ,t and ΔARi ,t represent the annual change in revenue and in accounts

receivables from operating activities, respectively. PPEi ,t is current year gross property,

plant, and equipment. TAi ,t −1 and Salesi ,t −1 are lagged total accruals and lagged revenue,

respectively.13 All terms are scaled by lagged total assets ( Ai ,t −1 ). εi,t is a zero-mean

random error, and forms our estimate of the discretionary component of accruals.



Measurement of discretionary expenses

We follow Roychowdhury (2006) in our measure of discretionary expenses, which we

estimate using the following regression:




12
   Dechow et al. 2003 show empirically that this model performs better than other versions of the Jones
(1991) model in terms of explanatory power.
13
   We do not include the final term of the FLJM, expected sales growth, typically calculated as the
difference between current and next period sales scaled by current sales, because this would cause terminal




                                                                                                                                 17
DISEXP                         ⎡          ⎤      ⎡ Sales          ⎤
      i, t                         1 ⎥                   i, t − 1 ⎥                               (2)
                       = α0 +α ⎢            +β ⎢                    +ε
             A                t⎢A         ⎥   1t ⎢ A              ⎥ i, t
              i, t − 1         ⎣ i, t − 1 ⎦      ⎣    i, t − 1 ⎦




DISEXPi ,t , discretionary expenses, is the aggregation of selling, general, and

administrative expenses (SG&A), advertising, and research and development (R&D). εi,t

is a zero-mean random error, and forms our estimate of the abnormal component of

discretionary expenses.




                                           [Insert Table 1 about here]

Descriptive Statistics

        For consistency with Pourciau (1993) and Dechow and Sloan (1991), we group

years t–1 and t together to form our estimate of the terminal period. As Pourciau notes,

executives may know well in advance when their tenure with the firm is coming to an

end; thus, examination of year t alone will likely not document the full extent of the

horizon problem.

        Table 1, Panel B presents a comparison of the non-terminal and terminal period

averages of the following variables: cash compensation (CASH_COMP), the Black-

Scholes      values      of    options      granted       (OPTION_COMP),       total   compensation




year accruals to be affected by year t+1 sales, which reflect the performance of the new CEO rather than
the outgoing CEO.




                                                                                                     18
(TOTAL_COMP),14 cash compensation as a percentage of total compensation

(CASH_PERCENT), option compensation as a percentage of total compensation

(OPTION_PERCENT), income before extraordinary items (IBEI), the annual change in

IBEI scaled by opening total assets (ΔROA), the annual change in discretionary accruals

(ΔDA), where both opening and closing discretionary accruals are the residuals obtained

from Model (1), and the change in abnormal discretionary expenses (ΔAB_DISEXP),

where both opening and closing abnormal discretionary expenses are the residuals

obtained from Model (2).

        Both cash and total compensation are significantly higher in the terminal period

(p-values < 0.0001 and 0.03), which may indicate the presence of special severance

arrangements. While the percentage of cash compensation to total compensation is high

in both the terminal and non-terminal periods, at 51% and 43% respectively, cash

(option) pay forms a significantly lower (higher) percentage of pay in the terminal versus

the non-terminal period. IBEI is statistically indistinguishable across the terminal and

non-terminal periods. ΔROA is significantly lower in the terminal period than in the non-

terminal period (p-value = 0.01). Neither the changes in discretionary accruals nor the

change in abnormal discretionary expenses differ significantly across the two periods. It

is worth noting that our research hypotheses do not necessarily predict an increase in

earnings management in the terminal period, but rather predict an increased monitoring

of any observed earnings management by the compensation committee. The fact that



14
  We measure total compensation as TDC1 from Execucomp, which is equal to salary plus bonus plus
other annual compensation plus restricted stock grants plus long-term inventive payments plus option
grants.




                                                                                                 19
managers are likely to be aware that monitoring is particularly high in the terminal period

is likely to constrain, though not eliminate earnings management.15

           Table 2 Panel A (Panel B) presents Pearson correlations between the main

variables of interest used in our Table 3 cash compensation regressions (Table 4 option

compensation regressions). Both panels present correlations separately by non-terminal

and terminal periods, with the non-terminal period correlations presented above the

diagonal and the terminal period correlations below the diagonal.

           The annual percentage change in cash compensation ( ΔCASH _ COMP ) is

positively related to the annual change in return on assets ( ΔROA ) in both terminal and

non-terminal periods, although the correlation is only weakly significant in the terminal

period (p-value = 0.093).           ΔCASH _ COMP is positively correlated with excess stock

returns ( EXC _ CRET ) at a p-value < 0.001 in both periods, and is positively (negatively)

correlated with the change in discretionary accruals ( ΔDA ) in the non-terminal (terminal)

period, although the terminal period relation is not statistically significant.

ΔCASH _ COMP is significantly positive related to the change in aggregate abnormal

discretionary spending in the non-terminal period (p-value < 0.0001) suggesting that, in

the normal course of business, the compensation committee encourages discretionary

expenditures. This relation disappears in the terminal period. ΔROA and ΔAB_DISEXP

are also positively related in the non-terminal period (p-value <0.0001) but negatively

related in the terminal period (p-value = 0.08), which is consistent with the idea that non-

terminal period discretionary expenditures enhance profitability while terminal period

15
     Similarly, the existence of photo radar devices is likely to discourage speeding, but clearly does not




                                                                                                        20
expenditures do not. ΔDA and ΔAB_DISCRET are significantly negatively correlated in

both the terminal and the non-terminal periods (p-value < 0.01).

         The level of option compensation (OPTION_COMP) is positively related to

return on assets in the non-terminal period, but the relation is insignificant during the

terminal period. OPTION_COMP is negatively related to discretionary accruals, and

positively related to abnormal discretionary expenditures, in both the terminal and non-

terminal periods, and positively correlated with excess returns in both periods.



IV. Multivariate Results

Cash Compensation

         To test our hypotheses, we regress the change in CEO cash compensation on the

change in the components of annual income in pooled time-series, cross-sectional

regressions. The OLS assumption that all observations are independent can lead to

misspecification due to serial correlation of the error terms, and thus we follow the

guidance provided for panel data regressions by Petersen (2009). We cluster residuals by

CEO for equation (3) and all subsequent regressions.

         Consistent with Baber et al. (1998) and Matsunaga and Park (2001), we examine

changes in compensation rather than levels to allow every firm to function as its own

control, which reduces the need for control variables. The specification of compensation

depending on earnings is consistent with the descriptive evidence of bonus plan design

provided by Murphy (1999) indicating that almost all companies included in his sample



eliminate it.




                                                                                      21
used at least one measure of accounting profits in the determination of annual bonuses of

top managers (Murphy 1999, p. 12).

        We present a number of variants of the following regression model:

ΔCASH _ COMPit = α 0 + β1 * TERMYR i , t + β 2 × ΔROAi , t + β 3 × ΔROAi , t * TERMYR i , t +
                      β 4 × EXC _ CRETi , t + β 5 × EXC _ CRETi , t * TERMYR i , t + β 6 × ΔDAi , t
                                                                                                      (3)
                      + β 7 × ΔDAi , t * TERMYR i , t + β 8 × ΔAB _ DISXEPi , t
                      + β 9 × ΔAB _ DISEXPi , t * TERMYR i , t + ε i , t

where ΔCASH _ COMPi,t is the year-to-year percentage change in the cash compensation

of the CEO, TERMYRi,t is a 0/1 indicator variable taking on a value of 1 in the terminal

period and 0 otherwise, and ΔROAi,t is the annual change in return on assets.16 We also

control for firms’ market performance by including a firm’s excess return relative to a

size-matched portfolio in the regressions ( EXC _ CRETi,t ). Failure to control for

EXC _ CRETi,t could lead us to observe a relation between discretionary accruals and/or

discretionary expenses and compensation if these items affect compensation only to the

extent that they are impounded in stock price. ΔDAi,t is the year-to-year change in

discretionary accruals, where both opening and closing discretionary accruals are the

residuals obtained from Model (1). If the compensation committee adjusts any

accounting-based compensation in the terminal period to downplay the role of

discretionary accruals due to concerns about the horizon problem, the coefficient ΔDAi,t

should decline in the terminal period. Therefore, we expect the estimates of β 7 to be

significantly negative.


16
  More specifically, ΔROA is equal to year t income before extraordinary items scaled by total assets at the
beginning of year t less year t-1 income before extraordinary items scaled by total assets at the beginning
of year t-1.




                                                                                                        22
         ΔAB _ DISEXP is the change in abnormal discretionary expenses, which we

measure as the residual from equation (2) and ΔAB _ DISEXP * TERMYR is the

interaction of ΔAB _ DISEXP and TERMYR . Our second hypothesis predicts a

significantly different relation between changes in abnormal discretionary expenses and

changes in compensation in the terminal period. Since more spending on expenses

reduces income, a positive estimate of β9 implies that the compensation committee

encourages spending on discretionary expenses in the terminal period (consistent with a

desire to curb a management tendency to manage earnings upward by cutting necessary

expenditures), while a negative estimate of β 9 is consistent with an agency cost

explanation in which the compensation committee associates excessive terminal-year

discretionary spending with perquisite consumption or investments in managerial-

specific assets.

        We note that equation (3) does not control for loss years. Given the Gaver and

Gaver (1998) finding that the earnings/compensation relation differs depending on

whether earnings are positive or negative we expect that if earnings are negative in both

the current and the prior period, the bonus component of income should be nil in both

years. Our sample contains only nine firm-year observations with consecutive losses and

replication of our results including main and interactive effects for loss years does not

change the significance of any of our variables of interest. Therefore, for parsimony, we

present results excluding controls for losses.17




17
  As an additional check on the robustness of our results, we re-estimate the regressions trimming the top
and bottom 1% of each regression variables. Our inferences are unchanged.




                                                                                                       23
       The first column of Table 3 presents an abbreviated version of equation (3) that

includes discretionary accrual changes but not changes to abnormal discretionary

expenses (that is, it includes the variables          ΔDA and    ΔDA* TERMYR       but not

ΔAB _ DISXEP and ΔAB _ DISEXP * TERMYR ). The second and third columns of Table 3

present the results of the column (1) specification run separately for positive and negative

discretionary accrual changes, respectively. The fourth column of Table 3 presents an

abbreviated version of equation (3) that includes changes to abnormal discretionary

expenses but not discretionary accrual changes (that is, it includes the variables

ΔAB _ DISXEP and ΔAB _ DISEXP * TERMYR but not ΔDA and ΔDA* TERMYR ). The fifth

column of Table 3 includes the full version of equation (3).

       The explanatory power of our first Table 3 specification, which focuses on

discretionary accrual changes, exhibits an adjusted R2 of 0.121, which is slightly higher

than that of Baber et al. (1998) who use a similar specification and somewhat lower than

Matsunaga and Park (2001) who include a large number of dummy variables in their

regression. As expected, changes in compensation are strongly related to changes in both

accounting earnings and stock returns (p-values < 0.01). The coefficient on ΔDA, the

change in discretionary accruals, measures the incremental explanatory power of

discretionary accruals over other income components in the non-terminal period. This

measure is not significantly different from zero suggesting that discretionary accruals are

treated similarly to other earnings components. However, the incremental effect of

terminal period discretionary accruals ( β 7 ) exhibits a coefficient of -1.042 (p-value =

0.003). The sum of the coefficients on β 6 and β 7 represents the valuation of




                                                                                         24
discretionary accruals in the terminal period compensation function relative to other

income components, and is significantly negative with a p-value = 0.001. In combination,

these results indicate that, consistent with H1, the compensation committee intervenes in

the terminal period to reduce the effect of the discretionary accrual component of income

on compensation. The effect of income on the compensation function is otherwise

unaltered in the terminal period.

        We investigate whether the compensation committee takes a symmetric approach

towards positive and negative changes in discretionary accruals in the terminal period.

The second column of Table 3 examines positive changes to discretionary accruals, and

the third column negative changes to discretionary accruals.18 The results are consistent

with those observed when we use all observations: The coefficients on ΔROA and

EXC_CRET remain significant in the non-terminal period, and the interaction of both

ΔROA and EXC_CRET with TERMYR remain insignificantly different from zero. The

interaction of ΔDA and the terminal period dummy is negative and significant in both

specifications with a p-value of 0.018 (0.041) for positive (negative) changes in

discretionary accruals. These results are consistent with intervention by the compensation

committee in the terminal period to both punish upward accrual-based earnings

management, and to shield the manager from adverse effects of taking needed write-

downs to accruals.19 An additional rationale for the differential terminal period treatment

of the negative discretionary accruals is that in the non-terminal period, managers could


18
   The classification of positive and negative changes is done by firm-year. Firm-years for which the annual
change in discretionary accruals is greater than (less than) zero are presented in Column 2 (Column 3).
19
   The finding that managers are shielded from taking write-downs is consistent with the shielding of
restructuring charges observed by Dechow et al. (1994) and Adut et al. (2003).




                                                                                                        25
use these as a form of earnings management to set up cookie-jar reserves. Since it is not

in the outgoing manager’s interest to set up reserves for his or her successor, downward

accruals in the terminal period are unlikely to represent earnings management.

       Column (4) of Table 3 examines the relation between compensation and changes

in abnormal discretionary expenses in the terminal period. Results of column (4) indicate

that, outside of the terminal period, the compensation committee rewards abnormal

increases in discretionary expenditures: the coefficient on ΔAB _ DISEXP is positive and

significant (p-value = 0.016), providing evidence of the compensation committee’s effort

to discourage managers from real-activity earnings management (i.e., inflating income by

cutting discretionary spending). However, the relation between the change in abnormal

discretionary spending and the change in cash compensation is reduced to insignificance

in the terminal period: the coefficient on ΔAB _ DISEXP * TERMYR is negative (though

insignificant) and equal in magnitude to the coefficient on ΔAB _ DISEXP .

       To analyze these results further, we estimate our column (4) regression

individually by type of discretionary expenditure (untabulated), and find that the main

driver of these results is changes in abnormal SG&A expenses. The compensation

committee pays for abnormal SG&A in the non-terminal period, but this relation is

reversed in the terminal period. The negative coefficient on ΔAB _ SG & Ai,t * TERMYRi,t

is weakly significant at a p-value of 0.10, and is comparable in magnitude to the positive

coefficient on ΔAB _ SG & Ai ,t in the non-terminal period.20



20
  The p-value on the joint significance of the coefficients on ΔAB_SG&A and ΔAB_SG&A_TERMYR is
0.607.




                                                                                           26
       Results of the regressions using changes in abnormal R&D and changes in

abnormal advertising as the dependant fail to attain the consistently significant results

that we observe in the regressions for changes in abnormal SG&A. This may be

attributable both to the smaller sample and lower magnitude of the raw expenses as

compared with SG&A. For example, over a quarter of our firms report no expenditures

on R&D, and over half report no expenditures on advertising. While SG&A is a very

material line item for our sample firms, with a mean (median) value equal to 24% (22%)

of sales, average R&D expenditures represent a mean (median) of only 6% (3%) of sales,

and advertising expenditures a mean (median) of only 5% (3%) of sales.

       Finally, we use the full version of equation (3) to consider the joint effect of

changes in discretionary accruals and changes in abnormal discretionary expenditures on

changes in compensation. Prior literature on the horizon problem has independently

considered accrual-based (Pourciau 1993) and real (Dechow and Sloan 1991, Cheng

2004) earnings management. Additionally a recent and growing body of literature has

examined the simultaneous use of accual-based and real earnings management (Zang

2007, Cohen, Dey, and Lys 2008, Cohen and Zarowin 2008). All of these authors find

evidence consistent with managers trading off real and accrual-based earnings

management. If we assume that earnings management takes place during the terminal

period, it is reasonable to consider the entire portfolio of tools that the manager might use

to manage earnings in one regression model, particularly since the compensation

committee provides an annual bonus that reflects its composite evaluation of all the

manager’s activities during the year.




                                                                                          27
        Table 3, column (5) presents the results of incorporating changes in both

discretionary accruals and changes in abnormal discretional spending on changes in cash

compensation. Consistent with models (1) through (3), changes in discretionary accruals

are not priced differently from other income components in the non-terminal period.

However, the incremental effect of terminal period discretionary accruals relative to other

income components exhibits a coefficient of -1.169 (p-value = 0.003), and the valuation

of discretionary accruals in the terminal period compensation function (the sum of the

coefficients on β 6 and β 7 , which equals -0.936) is significantly negative (p-value =

0.002). Changes in abnormal discretionary expenditures continue to be positively priced

in the non-terminal period compensation function (p-value = 0.011), but the incremental

effect of terminal period discretionary accruals is negative and significant (p-value =

0.042).21 The effect of terminal period changes in discretionary expenditures on

compensation (the sum of β8 and β 9 ) is insignificantly different from zero.

        These results for discretionary expenditures are consistent with the argument that

the compensation committee encourages the CEO to make discretionary expenditures

that are value-increasing to the firm up until the end of her tenure. At this point, the

compensation committee views these expenditures in a different light, consistent with the

contention that discretionary expenditures correlated with the idiosyncratic talents of the

outgoing CEO are no longer valuable in the face of an upcoming CEO change. Consistent

with this explanation, we note that Table 2 shows that the correlation between changes in

21
   The statistical significance of the estimate of the effect of discretionary expenses in the terminal year
increases in Model (5) because ΔAB_ DISEXP is negatively correlated with ΔDA in the terminal year.
Managers who up-accrue also cut discretionary expenses. Failure to control for ΔDA results in a correlated




                                                                                                        28
ROA and changes in abnormal discretionary expenditures is significantly positive in

CEOs’ non-terminal years and significantly negative in their terminal years. A further

explanation consistent with these findings is that if the compensation committee

increased cash compensation as discretionary spending increased, the horizon problem

might encourage managers to spend indiscriminately as they neared the end of their

tenure with the firm.



Option compensation

         Finally, for comparability to Cheng (2004), we investigate the relation between

terminal period earnings management and option pay.22 Unlike Cheng, who examines a

sample of R&D intensive firms, we do not have a sample of firms for which option use is

consistent and pervasive. Examination of changes in option grants on changes in

compensation is not informative for our sample as a measure of the pay/performance

relation because (a) those firms that do grant options do not make concurrent grants every

year and (b) the level of option grants within firm varies significantly by year.

         We therefore focus on the relation between compensation levels and the annual

level of option grants. Use of the levels specification requires the addition of more

control variables, since each firm no longer functions as its own control. We therefore

introduce an indicator variable, LOSS, which takes on a value of 1 if ROA is negative and

0 otherwise. LOSS_TERMYR is the interaction of LOSS and TERMYR, ROA_LOSS is the



omitted variable problem that reduces the precision of coefficient estimate, and in this case also imparts an
upward bias.
22
   Cheng focuses on situations where management myopia is expected to be high, and groups the terminal
period together with periods in which the firm faces a small earnings decline or a small loss.




                                                                                                         29
interaction of ROA and LOSS, and ROA_LOSS_TERMYR is the interaction of ROA,

LOSS, and TERMYR. For consistency with the definition of the right hand side variables,

we scale the left hand side by opening total assets, such that OPTION _ COMPit equals

the Black-Scholes value of option compensation divided by opening total assets. Table 4

presents the results of estimating three versions of the following regression to document

the relation between option compensation and both discretionary accruals and abnormal

discretionary expenditures:


OPTION _ COMPit = α 0 + β1 * TERMYRi ,t + β 2 × ΔROAi ,t + β 3 × ΔROAi ,t * TERMYRi ,t
                    + β 4 × EXC _ CRETi ,t + β 5 × EXC _ CRETi ,t * TERMYRi ,t
                    + β 6 × ΔDAi ,t + β 7 × ΔDAi ,t * TERMYRi ,t +β 8 × ΔAB _ DISEXPi ,t
                                                                                           (4)
                    + β 9 × ΔAB _ DISEXPi ,t * TERMYRi ,t + β10 * LOSS i ,t +
                    + β11 × LOSS i ,t * TERMYRi ,t + β12 × ΔROAi ,t * LOSS i ,t
                    + β13 × ΔROAi ,t * LOSS i ,t * TERMYRi ,t + ε i ,t



       Consistent with the results on cash compensation, we report regressions using

discretionary accruals alone (column 1), abnormal discretionary expenditures alone

(column 2), and both discretionary accruals and abnormal discretionary expenditures in

the same regression (column 3). We measure option grants over the twelve months of the

fiscal year. ROA and EXC_CRET are significantly positively associated with the value of

option grants in all three specifications. ROA_TERMYR is significantly negative,23 and

EXC_CRET_TERMYR significantly positive. Together, these results suggest that the

compensation committee shifts the pay/performance relation for option compensation

away from income and towards returns in the terminal period. Possible explanations for




                                                                                                 30
this include: (a) Returns are less manipulable than accounting earnings and (b) because

returns lead earnings, the committee attempts to reward outgoing executives for actions

they have taken that will benefit future earnings.24 The positive coefficient on TERMYR

is consistent with the idea of options as forward-looking compensation because it

indicates that, after controlling for both accounting and market-based performance, firms

award a higher level of option compensation to CEOs in the terminal period. The

negative relation between ROA_LOSS and the level of option grants indicates that firms

provide more option compensation when performance is poor. This initially

counterintuitive finding is, however, consistent with the results of Yermack (1995),

Dechow, Hutton, and Sloan (1996), and Core and Guay (2001), all of whom find that

cash constrained firms substitute equity compensation for cash compensation. This

relation is reversed in the terminal period, as indicated by the positive and significant

coefficient on ROA_LOSS_TERMYR. The CEO receives fewer options when terminal

period ROA is negative.

        The relation between discretionary accruals and option grants is negative and

significant in the non-terminal period in both columns (1) and (3). This relation becomes

significantly more negative in the terminal period in both columns (1) and (3), indicating

that, consistent with the results of the cash regressions, the compensation committee

reduces the weight on the discretionary accruals component of income in deciding



23
   In fact, the large negative coefficient on ROA_TERMYR relative to the smaller positive coefficient on
ROA indicates that the CEO’s option pay is unrelated to earnings performance in the terminal period.
24
   This finding is consistent with the results of Dikolli, Kulp and Sedatole (2009), whose results suggest
that compensation committees counter incentives to meet short-term earnings targets by increasing
(decreasing) the sensitivity of cash bonuses to returns (earnings).




                                                                                                       31
terminal-period option pay. The coefficient on terminal year discretionary accruals (the

sum of DA and DA* TERMYR ) is significantly negative in both of columns (1) and (3).

        The relation between abnormal discretionary expenses and option compensation

is positive in both the terminal and non-terminal periods in our sample. However, results

fail to attain statistical significance in either sub-period.

        For consistency with the prior literature, we also examine the relation between

changes in option compensation and changes in performance (untabulated). Of our 763

observations in the options regressions, we lose 186 firms due to lack of grants in

concurrent years, and estimate regression equation (4) in changes form for the remaining

577 observations. When we use the percentage change in options as the left-hand side

variable, we find no relation between changes in discretionary accruals and changes in

option compensation in either of our measurement windows. We expect that this is

largely due to the variability in the dependant variable, which exhibits a mean year-to-

year change in excess of 100%. We therefore replicate the changes regression using the

annual change in option grants scaled by total assets as the dependent variable. These

regressions results exhibit a coefficient on ΔDA_TERMYR that is significantly negative

as predicted (p-value < 0.05); however, results of the regression are otherwise not

consistently intuitive. For example, there is no relation between changes in option grants

and changes in return on assets. We therefore do not place high weight on these results as

a description of the relation between earnings components and option compensation. Our

results of a significant relation between earnings components and changes in cash

compensation but mixed results for the relation between earnings components and




                                                                                       32
changes in option compensation are consistent with Baber, Kang and Kumar (1998).

While they find a significant relation between earnings innovations and changes in cash

compensation, they find no relation between earnings innovations and changes in stock-

based compensation.


VI. Conclusion

       In this paper, we examine the horizon problem by considering the role of the

compensation committee. We use the relation between income and compensation to

evaluate the compensation committee’s assessment of both accrual-based and real

activity earnings management in the terminal period.

       We examine the relation between changes in cash compensation and changes in

ROA and abnormal discretionary accruals over a CEO’s career in a pooled regression

analysis.   We find no significant difference between the treatment of abnormal

discretionary accruals and other components of earnings in CEOs’ non-terminal years. In

CEOs’ terminal years however we find that the compensation committee places

significantly less weight on abnormal discretionary accruals than it does on other

components of earnings when setting cash compensation.         Further, we find that this

relation is symmetric across positive and negative discretionary accruals, suggesting that

in this situation the compensation committee not only shields compensation from

negative accruals, but also intervenes to reduce pay when managers up accrue.

       We also find that the relation between changes in abnormal discretionary

spending and changes in cash compensation is altered in the terminal period. Our results

suggest that the compensation committee values abnormal discretionary spending in the




                                                                                       33
non-terminal period, but this relation is reversed in the terminal period such that changes

in abnormal discretionary spending become a neutral in the pay/performance relation.

Our findings on discretionary spending are driven by the relation between compensation

and abnormal SG&A. The relations between abnormal accruals, abnormal discretionary

spending, and cash compensation are upheld when we include both types of activities in

the same regression model.

       The results on discretionary spending are consistent with the interpretation that

the compensation committee does not view upward earnings management through real

activities as a particular concern in the terminal period, but is more concerned with

encouraging the CEO to abstain from perquisite consumption or managerial-specific

investments at this time. When we combine these findings with those observed in the

abnormal accrual regressions, our results indicate that the compensation committee sees

aggressive accruals as the primary indicator of terminal period earnings management.

       We also observe a reduction in the relation between the level of option

compensation and the level of discretionary accruals in the terminal period. Observation

of this finding for a changes analysis using option compensation is sensitive to the

specification used. We do not observe a differential relation between compensation and

abnormal discretionary spending across the terminal and non-terminal periods.

       A considerable literature has emerged in recent years that examines whether

governance structure can mitigate earnings management (see for example Cornett,

Marcus and Tehranian 2008, Bowen, Rajgopal and Ventakchalam 2008, Larcker and

Richardson 2004, and Klein 2002). Additionally, the finding that the presence of equity

incentives actually encourages earnings management is prevalent in the literature, though



                                                                                        34
whether the direction is up or down is dependent on the circumstance (see for example

Cornett, Marcus and Tehranian 2008, McAnally, Srivastava, and Weaver 2008,

Bergstresser and Philippon 2006, and Cheng and Warfield 2005). Our findings indicate

that the behavior of the compensation committee in the CEO’s terminal period provides

an example of a well-functioning governance mechanism that identifies opportunistic

behavior and prices it in compensation.

       The results in this paper contribute to an understanding of why the existence of a

horizon problem has been hard to document in the literature (Murphy and Zimmerman

1993). A CEO’s terminal period appears to be a time when the compensation committee

exercises special vigilance. The committee’s willingness to intervene and reduce

compensation in the terminal period when they suspect opportunistic behavior may well

lower the incentive to manage earnings at this time.




                                                                                      35
References

Adut, D., W. H. Cready, and T. J. Lopez. 2003. Restructuring charges and CEO cash
       compensation: A reexamination. The Accounting Review 78 (1): 169-192.

Armstrong, C., A. Davila and G. Foster. 2006. Venture-backed private equity valuation
      and financial statement information. Review of Accounting Studies 17:119-64.

Baber, W. R., S. Kang, and K. R. Kumar. 1998. Accounting Earnings and Executive
       Compensation: The Role of Earnings Persistence. Journal of Accounting and
       Economics 25 (2): 169-193.

Baker, T., D. Collins and A. Reitenga. 2009. Incentives and opportunities to manage
       earnings around options grants. Contemporary Accounting Research 26 (3):
       Forthcoming.

Balsam, S. 1998. Discretionary Accounting Choices and CEO Compensation.
      Contemporary Accounting Research 15(3): 229-252.

Banker, R., R. Huang, and R. Natarajan. 2007. Does SG&A expenditure create a long-
      lived asset? Working paper, Temple University, Baruch College, and University
      of Texas at Dallas.

Bergstresser, D., and T. Philippon. 2006. CEO incentives and earnings management.
       Journal of Financial Economics 80 (3): 511-529.

Bowen, R., S. Rajgopal, and M. Venkatchalam. 2008. Accounting Discretion, Corporate
      Governance, and Firm Performance. Contemporary Accounting Research 25(2):
      351-405.

Butler, S., and H. Newman. 1989. Agency Control Mechanisms, Effectiveness and
        Decision Making in an Executive’s Final Year with the Firm. Journal of
        Institutional and Theoretical Economics 145: 451-464.

Chen, C., H. Lu and T. Sougiannis. 2008. Managerial Empire Building, Corporate
       Governance, and the Asymmetrical Behavior of Selling, General, and
       Administrative Costs. Working paper, University of Illinois at Urban-Champlain
       and University of Toronto.

Cheng, S. 2004. R&D Expenditures and CEO Compensation. The Accounting Review
       79(2): 305-328.

Cheng, Q. and T. D. Warfield. 2005. Equity Incentives and Earnings Management. The
       Accounting Review 80 (2): 441-476.




                                                                                        36
Cohen, D. A., A. Dey and T. Z. Lys. 2008. Real and Accrual-Based Earnings
       Management in the Pre- and Post-Sarbanes-Oxley Periods. The Accounting
       Review 83 (3): 757-787.

Cohen, D. and P. Zarowin. 2008. Accrual-based and Real Eearnings Management
       Activities around Seasoned Equity Offerings. Working paper, New York
       University.

Coles, J., M. Hertzel, and S. Kalpathy, 2006, Earnings management around employee
        stock option reissues, Journal of Accounting and Economics 41: 173-200.

Core, J. and W. Guay. 1999. The use of equity grants to manage optimal incentive levels.
        Journal of Accounting and Economics 28(2): 151-184.

Core. J. and W. Guay. 2001. Stock options for non-executive employees. Journal of
        Financial Economics 61(2): 253-287.

Cornett, M., A. Marcus, and H. Tehranian, 2008. Corporate governance and pay-for-
       performace: The impact of earnings management. Journal of Financial
       Economics 87(2): 357-373.

Dahiya, S. and D. Yermack. 2008. You can’t take it with you: Sunset provisions for
      equity compensation when mangers retire, resign, or die. Journal of Corporate
      Finance 14(5): 499-511.

Dechow, P. M., M. R. Huson, and R. G. Sloan. 1994. The Effect of Restructuring
      Charges on CEO Cash Compensation. The Accounting Review 69 (1): 138-156.

Dechow, P., S. Richarson, and I. Tuna. 2003. Why are Earnings Kinky? An Examination
      of the Earnings Management Explanation. Review of Accounting Studies 8:355-
      384.

Dechow, P. and R. G. Sloan. 1991. Executive Incentives and the Horizon Problem.
      Journal of Accounting and Economics 14(5): 51-89.

Dechow, P., A. Hutton, and R. Sloan. 1996. Economic Consequences of Accounting for
      Stock-Based Compensation. Journal of Accounting Research, 1996, 1-20.

Dikolli, S. S., S. L. Kulp, and K. L. Sedatole. 2009. Transient Institutional Ownership
        and CEO Contracting. The Accounting Review 84(3): 737-770.

Gaver, J. J. and K. M. Gaver. 1998. The Relation Between Nonrecurring Accounting
       Transactions and CEO Cash Compensation. The Accounting Review 73(2):235-
       253.




                                                                                          37
Gibbons, R., and K. J. Murphy. 1992. Does Executive Compensation Affect Investment.
      Journal of Applied Corporate Finance 5(2): 99-109.

Graham, J., C. Harvey, and S. Rajgopal. 2005, The Economic Implications of Corporate
      Financial Reporting. Journal of Accounting and Economics 40(1-3): 3-73.

Intintoli, V. 2008. The Importance of Strong Internal Governance when extending the
        Succession Process: Evidence from Marathon Successions, Working Paper,
        University of Southern Illinois at Carbondale.

Jones, J. 1991. Earnings Management During Import Relief Investigations. Journal of
        Accounting Research 29(2): 193-228.

Kalyta, P. 2009. Accounting Discretion, Horizon Problem, and CEO Retirement Benefits.
       The Accounting Review, forthcoming.

Klein, A. 2002. Audit Committee, Board of Director Characteristics, and Earnings
       Management. Journal of Accounting and Economics 33(3): 375-400.

Larcker, D. and S. Richardson. 2004. Fees Paid to Audit Firms, Accrual Choices, and
       Corporate Governance. Journal of Accounting Research 42(3): 625-658.

McAnally, M. L., A. Srivastava and C. D. Weaver. 2007. Executive Stock Options,
     Missed Earnings Targets and Earnings Management. The Accounting Review
     83(1): 185-216.

Matsunaga, S. R. and C. W. Park. 2001. The Effect of Missing a Quarterly Earnings
      Benchmark on the CEO’s Annual Bonus. The Accounting Review 76(3): 313-332.

Murphy, K. J. 1999. Executive Compensation’, in O. Ashenfelter and D. Card (eds.),
      Handbook of Labor Economics 3, (North Holland, New York, NY).

Murphy, K. J., and J. Zimmerman. 1993. Financial Performance Surrounding CEO
      Turnover. Journal of Accounting and Economics 16(1,2&3): 273-315.

Parrino, R., 1997, CEO turnover and outside succession: a cross-sectional analysis,
       Journal of Financial Economics 46: 165-197.

Petersen, M. 2009. Estimating standard errors in finance panel data sets: comparing
   approaches. Review of Finance Studies 22(1): 435-80.

Pourciau, S. 1993. Earnings Management and Non-routine Executive Changes. Journal
       of Accounting and Economics 16(1,2&3): 317-336.

Roychowdhury, S. 2006. Earnings Management through Real Activities Management.
      Journal of Accounting and Economics 42 (3): 335-370.


                                                                                      38
Shleifer, A. and R. W. Vishny. 1989. Management Entrenchment: The Case of Manager-
       Specific Investments. Journal of Financial Economics 25 (1): 123-139.

Standard & Poor’s. 1983-1996. Register of Corporations, Directors, and Executives,
       Standard & Poor’s Corporation.

Subramanyam, K. 1996. The Pricing of Discretionary Accruals. Journal of Accounting
      and Economics 22(103): 249-281.

Yermack, D. 1995. Do corporations award CEO stock options effectively? Journal of
      Financial Economics 39(2,3): 237-269.

Yermack, D. 2006. Golden Handshakes: Separation Pay for Retired ad Dismissed CEOs.
      Journal of Accounting and Economics 41(3): 237-256.

Zang, A. 2007. Evidence on the Tradeoff between Real Manipulation and Accrual
       Manipulation. Working paper, University of Rochester.




                                                                                     39
Table 1, Panel A: Sample Construction

This table reports information about sample construction. CEO turnover data from 1995–2005 are obtained
from the database compiled by Intintoli (2008). Tenure of at least three years is required for measuring
changes in compensation. Following Dechow et al. (2003), financial firms (SIC Codes 60-69) are removed.
Compustat data required are: net income before extraordinary items, lagged total assets, current and lagged
accounts receivable and total accruals, current gross property, plant, and equipment, current sales, as well
as lead and lagged sales, and at least one of our discretionary expense categories (SG&A, advertising, and
R&D).


                                                                                       # of             # of
                                                                                       firms        turnover
                                                                                                           s
Voluntary turnovers from Intintoli (2008)                                                337             383
Remove:
Observations without valid GVKEY                                                          (1)            (1)
Observations without fiscal year-end data                                                (11)           (13)
Observations in the financial services sector                                            (52)           (54)
CEO tenure < 3 years                                                                     (35)           (59)
Observations without sufficient Compustat or return data in both the terminal and        (82)           (92)
non-terminal years
Total turnovers                                                                          156            164




                                                                                                        40
  Table 1, Panel B: Descriptive Statistics

  Descriptive Statistics by terminal and non-terminal periods

                       N        Non-terminal
    Variable                       period             Terminal period      Difference     t Value     Pr > |t|
CASH_COMP             821               1,617.330             1,987.940        370.610       4.080     <.0001
OPTION_COMP           821               2,536.780             3,189.250        652.470       1.190      0.234
TOTAL_COMP            819               5,043.700             8,840.650      3,796.950       2.180      0.029
CASH_PERCENT          819                  51.3%                 43.1%           -8.2%      -4.700    <0.0001
OPTION_PERCEN         819
T                                            36.3%                38.5%           2.1%       0.750       0.453

IBEI                  821                 563.172               675.195        112.023       1.300       0.193
ΔRΟΑ                  821                   0.004                -0.012         -0.016       -2.65       0.008
ΔDA                   821                  -0.002                -0.005         -0.003      -0.380       0.703
ΔΑΒ_DISEXP            821                  -0.015                -0.021         -0.006      -0.700       0.483


  CASH_COMP is annual salary and bonus compensation, OPTION_COMP is the annual Black-Scholes
  values of options granted, and TOTAL_COMP (TDC1 from Execucomp) is equal to salary plus bonus plus
  other annual compensation plus restricted stock grants plus LTIP payouts plus the value of option grants.
  CASH_COMP, OPTION_COMP, and TOTAL_COMP are in thousands of dollars. CASH_PERCENT is the
  percentage of cash compensation to total compensation and OPTION_PERCENT is the percentage of
  option compensation to total compensation. IBEI is income before extraordinary items. ΔROA is the
  change in earnings before extraordinary items divided by lagged total assets. ΔDA is the change in
  discretionary accruals where both opening and closing discretionary accruals are the residuals obtained
  from Model (1). ΔAB_DISEXP is the change in abnormal discretionary expenses, defined as the sum of the
  abnormal changes in selling, general, and administrative expenses, advertising, and research and
  development. Both opening and closing abnormal discretionary expenses are residuals obtained from
  Model (2).

  Sample sizes are lower for TOTAL_COMP, CASH_PERCENT and OPTION_PERCENT because of
  missing observations for total compensation.




                                                                                                        41
 Table 2
 Pearson correlation coefficients by terminal and non-terminal periods

 This table presents Pearson correlation coefficients for the non-terminal period above the diagonal, and the
 terminal period below the diagonal. Panel A presents correlations for the Table 3 regressions, and Panel B
 presents the correlations for the Table 4 regressions.

 Panel A: Correlations for main variables in Table 3 ( ΔCASH _ COMP ) regressions

                      ΔCASH _ COMP               ΔROA         EXC _ CRET        ΔDA          ΔAB _ DISEX
ΔCASH _ COMP         Non-terminal period →           0.276             0.336        0.085              0.179
                     Terminal period ↓           (<0.0001)         (<0.0001)      (0.060)          (<0.0001)
ΔROA                                   0.093                           0.302        0.373              0.187
                                     (0.093)                       (<0.0001)    (<0.0001)          (<0.0001)
EXC _ CRET                             0.188        -0.454                         -0.059              0.106
                                     (0.001)      (0.0001)                        (0.191)            (0.019)
ΔDA                                   -0.075         0.759            -0.459                          -0.128
                                     (0.178)     (<0.0001)         (<0.0001)                         (0.005)
ΔAB _ DISEXP                           0.027        -0.098             0.081       -0.231
                                     (0.625)       (0.076)           (0.144)    (<0.0001)


 Panel B: Correlations for main variables in Table 4 ( OPTION _ COMP ) regressions

                      OPTION _ COMP              ROA          EXC _ CRET       DA            AB _ DISEXP

OPTION _ COMP Non-terminal period →                  0.303            0.373        -0.239              0.221
                     Terminal period ↓           (<0.0001)        (<0.0001)     (<0.0001)           (0.0001)
ROA                                     0.011                          0.139       -0.044              0.243
                                      (0.847)                        (0.002)      (0.332)          (<0.0001)
EXC _ CRET                              0.630       -0.298                                             0.020
                                    (<0.0001)    (<0.0001)                                         (<0.0001)
DA                                     -0.316        0.671           -0.609                           -0.180
                                    (<0.0001)    (<0.0001)        (<0.0001)                        (<0.0001)
AB _ DISEXP                             0.138       -0.008            0.099      -0.154
                                      (0.019)      (0.882)          (0.096)      (0.009)


 ΔCASH_COMP is the annual percentage change in salary and bonus. ΔROA is the change in earnings
 before extraordinary items, scaled by lagged total assets. EXC_CRET is the firm’s annual size-adjusted
 stock return. ΔROA is earnings before extraordinary items, scaled by lagged total assets. ΔDA is the annual
 change in discretionary accruals, where both opening and closing discretionary accruals are the residuals
 obtained from Model (1). ΔAB_DISEXP is the change in abnormal discretionary expenses, defined as the
 sum of the abnormal changes in selling, general, and administrative expenses, advertising, and research and
 development, where both opening and closing abnormal discretionary expenses are the residuals obtained
 from Model (2). OPTION_COMP is the Black-Scholes value of options granted to the CEO related to the
 fiscal year. ROA is earnings before extraordinary items, scaled by lagged total assets. DA is discretionary
 accruals (the residuals obtained from Model (1). AB_DISEXP is abnormal discretionary expenses, defined
 as the sum of the abnormal selling, general, and administrative expenses, advertising, and research and
 development (the residuals obtained from Model (2).




                                                                                                          42
       Table 3
       Regression results: Changes in cash compensation on changes in discretionary
       accruals and discretionary expenses

          ΔCASH _ COMPit = α 0 + β1 * TERMYRi , t + β 2 × ΔROAi , t + β 3 × ΔROAi , t * TERMYRi , t +
                                 β 4 × EXC _ CRETi , t + β 5 × EXC _ CRETi , t * TERMYRi , t + β 6 × ΔDAi , t
                                 + β 7 × ΔDAi , t * TERMYRi , t + β8 × ΔAB _ DISEXPi , t + β 9 × ΔAB _ DISEXPi , t * TERMYRi , t + ε i , t

                                                                    Positive ΔDA       Negative ΔDA
                                                             (1)                (2)                (3)            (4)           (5)
INTERCEPT                                                 0.150              0.140              0.164          0.156         0.157
                                                      (<0.0001)           (0.0001)         (<0.0001)       (<0.0001)     (<0.0001)
TERMYR                                                    0.004              0.073             -0.066         -0.006        -0.004
                                                        (0.883)            (0.135)            (0.118)        (0.818)       (0.865)
ΔROA                                                      1.002              1.190              0.697          0.974         0.803
                                                        (0.009)            (0.013)            (0.087)        (0.009)       (0.037)
ΔROA*TERMYR                                               0.585              1.097              1.087         -0.110         0.795
                                                        (0.271)            (0.193)            (0.149)        (0.800)       (0.136)
EXC_CRET                                                  0.190              0.164              0.216          0.181         0.189
                                                      (<0.0001)            (0.002)           (0.0001)       (0.0001)     (<0.0001)
EXC_CRET*TERMYR                                           0.022             -0.076              0.030          0.064         0.023
                                                        (0.752)            (0.489)            (0.766)        (0.352)       (0.740)
ΔDA                                                       0.127              0.184              0.339                        0.233
                                                        (0.476)            (0.597)            (0.168)                      (0.194)
ΔDA*TERMYR                                               -1.042             -1.518             -1.540                       -1.169
                                                        (0.003)            (0.018)            (0.041)                      (0.003)
ΔAB_DISEXP                                                                                                      0.336        0.373
                                                                                                              (0.016)      (0.011)
ΔAB_DISEXP*TERMYR                                                                                              -0.256       -0.426
                                                                                                              (0.184)      (0.042)

Adj. R2                                                    0.121             0.117               0.134          0.111         0.1263
N                                                            821               414                 407            821            821
Tests of significance of terminal year coefficients
ΔDA + ΔDA*_TERMYR β + β      (
                             ˆ
                                  6
                                   ˆ
                                       7
                                           )              -0.915             -1.334              -1.201                      -0.936
                                                         (0.001)            (0.015)             (0.092)                     (0.002)

                               ˆ     ˆ
ΔAB_DISEXP + ΔAB_DISEXP*TERMYR β 8 + β 9          (        )                                              0.080        -0.053
                                                                                                        (0.601)       (0.750)
       ΔCASH_COMP is the annual percentage change in salary and bonus. TERMYR is a 0/1 dummy variable,
       with 1 representing the CEO’s final two years in office. We define the final year in office as the last year in
       which the CEO had control of the financial statements through the annual general meeting in the
       subsequent year. ΔROA is the change in earnings before extraordinary items, scaled by lagged total assets.
       EXC_CRET is the firm’s annual size-adjusted stock return. ΔDA is the annual change in discretionary
       accruals, where both opening and closing discretionary accruals are the residuals obtained from Model (1).
       ΔAB_DISEXP is the change in abnormal discretionary expenses, defined as the sum of the abnormal
       changes in selling, general, and administrative expenses, advertising, and research and development, where
       both opening and closing abnormal discretionary expenses are the residuals obtained from Model (2).
       ΔROA_TERMYR, EXC_CRET_TERMYR, ΔDA_TERMYR, and ΔAB_DISEXP_TERMYR are the
       interactions of ΔROA, EXC_CRET, ΔDA, and ΔAB_DISEXP respectively with TERM_YR.



                                                                                                                         43
     Table 4
     Regression results: Levels of option compensation on levels of discretionary accruals

     OPTION _ COMPit = α 0 + β 1 * TERMYRi ,t + β 2 × ROAi ,t + β 3 × ROAi ,t * TERMYRi ,t
                           + β 4 × EXC _ CRETi ,t + β 5 × EXC _ CRETi ,t * TERMYRi ,t + β 6 × DAi ,t
                           + β 7 × DAi ,t * TERMYRi ,t + β 8 × AB _ DISEXPi ,t + β 9 × AB _ DISEXPi ,t * TERMYRi ,t
                           + β 10 * LOSS i ,t + β 11 × LOSS i ,t * TERMYRi ,t + β12 × ROAi ,t * LOSS i ,t
                           +β 13 × ROAi ,t * LOSS i ,t * TERMYRi ,t +ε i ,t

                                                                (1)                     (2)                       (3)
INTERCEPT                                                   -1.001                  -0.714                    -0.726
                                                           (0.197)                 (0.208)                   (0.187)
TERMYR                                                       3.871                   5.268                     4.644
                                                           (0.033)                 (0.048)                   (0.041)
ROA                                                         22.051                  20.264                    20.208
                                                           (0.020)                 (0.010)                   (0.009)
ROA_TERMYR                                                 -41.780                 -49.756                   -44.827
                                                           (0.043)                 (0.041)                   (0.039)
EXC_CRET                                                     1.822                   2.111                     1.910
                                                           (0.032)                 (0.012)                   (0.020)
EXC_CRET_TERMYR                                             17.912                  20.121                    17.978
                                                           (0.042)                 (0.054)                   (0.040)
DA                                                          -6.064                                            -4.936
                                                           (0.011)                                           (0.087)
DA_TERMYR                                                  -42.588                                           -41.242
                                                           (0.108)                                           (0.088)
AB_DISEXP                                                                             2.222                    1.964
                                                                                    (0.277)                  (0.350)
AB_DISEXP_TERMYR                                                                      6.306                    4.948
                                                                                    (0.322)                  (0.341)
LOSS                                                         -0.393                  -0.339                   -0.452
                                                            (0.753)                 (0.777)                  (0.708)
LOSS_TERMYR                                                   5.644                   3.976                    5.350
                                                            (0.193)                 (0.265)                  (0.192)
ROA_LOSS                                                   -91.509                 -89.256                  -883052
                                                            (0.032)                 (0.025)                  (0.033)
ROA_LOSS_TERMYR                                            228.718                 193.470                  229.877
                                                            (0.046)                 (0.054)                  (0.045)

Adj R2                                                       0.511                   0.486                    0.519
N                                                              763                     763                      763

Tests of significance of terminal year coefficients
ΔDA + ΔDA*_TERMYR β + β     (ˆ
                                6
                                   ˆ
                                    7
                                        )                   -48.652                                         -46.178
                                                           (0.0060)                                         (0.049)
                               ˆ     ˆ
ΔAB_DISEXP + ΔAB_DISEXP*TERMYR β 8 + β 9      (        )                              8.528                    6.912
                                                                                    (0.161)                  (0.152)



                                                                                                             44
OPTION_COMP represents the Black-Scholes value of options granted to the CEO related to the fiscal
year. TERMYR is a 0/1 dummy variable, with 1 representing the CEO’s final two years in office. We
define the final year in office as the last year in which the CEO had control of the financial statements
through the annual general meeting in the subsequent year. ROA is earnings before extraordinary items,
scaled by lagged total assets. EXC_CRET is the firm’s annual size-adjusted stock return. DA is
discretionary accruals (the residuals obtained from Model (1)). AB_DISEXP is abnormal discretionary
expenses, defined as the sum of abnormal selling, general, and administrative expenses, advertising, and
research and development (the residuals obtained from Model (2)). LOSS is a 0/1 dummy variable taking
on a value of 1 if ROA is negative, and 0 otherwise. ROA_TERMYR, EXC_CRET_TERMYR,
DA_TERMYR, and AB_DISEXP_TERMYR are the interactions of ROA, EXC_CRET, DA, and
AB_DISEXP respectively with TERM_YR. LOSS_TERMYR and ROA_LOSS are the interactions of
TERM_YR and ROA respectively with LOSS. ROA_LOSS_TERMYR is the interaction of ROA, LOSS, and
TERM_YR.




                                                                                                      45

								
To top