Using pension expense to manage earnings: implications for FASB standards.

Parker, Paula Diane ; Sale, Martha Lair

ABSTRACT

Recent studies provide convincing evidence firms manage reported earnings to achieve certain capital market reporting objectives. However, there is little empirical evidence on what levers firms use to manage their reported earnings.

This paper presents evidence that pension expense is an active lever used by firms to manage bottom-line, reported earnings. Firms with actual reported earnings in a neighborhood relatively close to their capital market earnings benchmark are selected for testing. Based on a proxy for premanaged earnings, firms hypothetically missing their capital market earnings benchmark are predicted to reduce their actual pension expense to increase actual reported earnings; whereas firms hypothetically beating their capital market earnings benchmark are predicted to increase their actual pension expense to reduce their actual reported earnings.

Both groups of firms are predicted to manipulate reported earnings in the direction that will move them closer to their capital market earnings benchmark than they would have otherwise been. Results suggest both groups of firms use pension expense as a lever to manage actual reported earnings.

INTRODUCTION

This study investigates whether or not firms use pension expense as an earnings management tool to maintain a steady stream of earnings. The capital market benchmark in the current year is equal to the prior year earnings. Prior research studies provide mixed evidence that pension expense is used as an earnings management tool to manage reported earnings. This lack of persuasive empirical evidence is puzzling as survey evidence indicates auditors perceive pension expense is a frequently used earnings management tool (Nelson et al. 2000). Most prior studies are unable to consistently detect pension expense manipulation because they focus primarily on contracting incentives rather than on capital market reporting incentives for explaining earnings management. Another reason may be that most prior studies focus on pension rate manipulation rather than on pension expense manipulation directly.

Interestingly, one research study (Bergstresser et al. 2006) indicates firms are more aggressive with their assumed expected long-term rate of return on pension assets when the firms are near critical earnings thresholds and this rate assumption has greater potential to impact reported earnings. Since changes in either the discount rate assumption or the compensation rate assumption could offset the impact on pension expense caused by the change in the assumed expected long-term rate of return on pension assets, our study extends prior research by focusing directly on pension expense taken as a whole rather than focusing only on the affect of one of the three pension rate assumptions. Therefore, this research takes a more inclusive approach by looking at the cumulative effects of all pension rate assumptions on pension expense.

This study differs from prior studies in that it examines whether or not the prior year earnings benchmark creates an incentive to manipulate pension expense in a rational economic manner. The research findings provide interested parties with relevant information necessary to support the position that pension expense be monitored more carefully in the future by auditors and regulators to prevent its misuse in financial statement reporting.

One obstacle associated with attempting to test for pension expense manipulation is that of determining what a firm's pension expense would be absent the manipulation. The Statement of Financial Accounting Standards No. 87, Employers' Accounting for Pensions (SFAS No. 87), provides a unique measure of what pension expense should be from year to year based on the corridor approach. Accordingly, firms are allowed to spread pension expense over time in order to avoid the immediate recognition of wide swing market fluctuations that affect pension investments. The reason regulation allows firms to spread pension expense is a long-term view whereby market fluctuations are expected to average out over the long-term. Firms are not forced to recognize short-term market fluctuations unless the aggregated unrecognized fluctuations exceed a 10% corridor (10% of the larger of the market related value of pension assets or the projected benefit obligation measured at the beginning of the period). Therefore, we are able to reasonably estimate what a firm's pension expense would be absent the manipulation.

Theoretically, pension expense should be approximately the same from year to year unless there is a change in the number of employees, industry effects, and or time fixed effects. The industry effects and time fixed effects are captured in the model by using dummy variables for each industry and for each year. Therefore, the proxy for pension expense, absent manipulation, is the prior year pension expense.

Managers run "what if analyses" at the end of the year to determine whether or not earnings benchmarks will be achieved. It is common practice to substitute the prior year pension expense as the current year pension expense in these analyses. So, the logic behind common practice supports our proxy for current year pension expense. In addition, actual pension expense is one of the last general ledger accounts that can be adjusted or manipulated at year-end in an attempt to meet the actual current year earnings benchmark.

For a number of reasons, pension expense is an ideal general ledger account for manipulating. The first reason is the lack of precision in the guidelines as set forth in SFAS No. 87 which allows firms great flexibility in choosing their assumed discount rate, compensation rate, and expected rate of return on pension assets. In addition, it is highly probable that firms have access to and authority over superior information regarding their applicable pension plans than is readily available to the public or to other interested parties. Another reason is the lack of timely verification of the rate assumptions and estimates because they cover discounted projections out in the future generally for 20 plus years.

This research design models the behavior of pension expense to identify its discretionary and nondiscretionary components. Therefore by design, any change in pension expense from year to year is considered discretionary and is the primary focus of explanation.

The benchmark test addresses whether or not firms use pension expense in an attempt to continue a steady stream of earnings. Barth et al. (1999) show evidence those firms with consecutive earnings increases experience higher stock prices, and when those firms encounter declines in reported earnings, the premium stock prices fall tremendously. As a result, firms have strong incentives to continue a steady stream of earnings to acquire market approbation and to avoid market devaluation.

The remainder of this paper is organized into four sections. The first section describes the background and the earnings-based benchmark. The second section provides the research design, hypothesis development, sample selection and other statistical considerations. The third section provides the results, interpretations, sensitivity analyses, and limitations. The fourth section provides the summary conclusions.

PRIOR LITERATURE AND EARNINGS-BASED BENCHMARKS

In 1985, the FASB issued SFAS No. 87, Employers' Accounting for Pensions, which is currently the primary standard influencing financial measurement for defined benefit pension plans. In 1998, the FASB issued SFAS No. 132, Employers' Disclosures about Pensions and Other Postretirement Benefits, which is currently the primary standard influencing pension disclosure. In early 2006, FASB proposed changes to the current pension standards by issuing the exposure draft, Employer's Accounting for Defined Benefit Pension and Other Postretirement Plans--an amendment of FASB Statements No. 87, 88, 106, and 132[R].

For the last two decades pension research (Kwon, 1989; Blankley, 1992; Ali and Kumar, 1993; Weishar, 1997; Brown, 2001 and Bergstresser et al., 2006) focuses primarily on the explanation of pension rates and how and why firms' select the particular pension rates disclosed in their financial statements. Improved disclosures required by SFAS No. 132 now provide enough information to recalculate pension expense using the three pension rate assumptions. So that, research in the area of pension accounting may experience a paradigm shift where pension rates are no longer the primary focus of explanation.

Kwon's (1989) research focuses only on the explanation of the discount rate. Blankley's (1992) research focuses individually on the explanation of the discount rate, compensation rate, and expected long-term rate of return on plan assets. Weishar's (1997) research focuses on the explanation of the simultaneous effects of the discount rate, compensation rate, and expected long-term rate of return on plan assets. Brown (2001) not only focuses on explaining the three pension rates but somewhat changes the direction of research by including a market valuation model to examine the value relevance of economic factors and reporting incentive factors.

In prior studies, the only explanatory variable that is consistently significant in explaining pension rate assumptions is the funding ratio variable. Other variables such as leverage, unrestricted retained earnings, cash constraints, manager control, size, unionization, tax loss, and change in CEO are not consistently significant from study to study. Possible explanations for the inconsistent findings may be due to omitted variables, measurement error, lack of power, and or misspecified models. Therefore, these models explaining pension rates may not fully capture the impact of pension expense manipulation as it relates to financial statement reporting.

We contend that a more complete research approach is needed to examine pension expense explicitly in relation to capital market based incentives. Whether managers act in self-interest or in the interest of shareholders, their performance is monitored by directors, auditors, investors, creditors, and regulators, which creates strong incentives to manage earnings. Capital market based incentives are expected to capture more fully financial statement manipulation as it relates to pension expense.

Burgstahler and Dichev (1997) theorize that investors in publicly traded firms use simple low-cost heuristics, such as earnings-based benchmarks, in determining firm value. In addition, prospect theory is informative as another reason for using benchmarks, whereby investors value gains and losses using a reference point rather than by an absolute level of worth. Prospect theory (Kahneman and Tversky 1979) is defined differently than expected utility theory. Utility is generally defined in terms of net wealth and value is defined in terms of gains and losses representing deviations from a reference point. The value function in prospect theory comprises a different shape for gains and losses. The value function is relatively steep and convex for losses whereas it is concave and less steep for gains. This occurs because people are risk seeking when faced with a loss but risk averse when faced with a gain. Using this theory, if zero is a natural reference point for change in earnings, then firms have incentives to manipulate earnings for a positive rather than negative earnings outcome. Burgstahler and Dichev (1997) use frequency distribution as a method for demonstrating the existence of earnings management. Evidence indicates a disproportionally low incidence of firms reporting small decreases in earnings and small losses relative to a high incidence of firms reporting small increases in earnings and small positive earnings.

DeGeorge et al. (1999) use a similar research design as Burgstahler and Dichev (1997) and report earnings are the single most value relevant item provided to investors in financial statement reporting. Earnings are used as performance measures, which in turn, provide the enticement for firms to manipulate earnings. Their research reveals how efforts to exceed thresholds, that is, to sustain recent performance, to report positive earnings, and or to meet analysts' expectations, induce particular patterns of earnings management. Clearly emerging patterns show earnings falling just short of thresholds are managed upward. Additional evidence suggests future performances of firms just achieving thresholds are poorer than performances for control firms that are less suspect of managing earnings (DeGeorge et al. 1999).

Barth et al. (1999) depict firms with longer strings of repeated earnings increases are priced at a premium but when these firms experience declines in earnings, the premiums fall intensely. Moehrle (2002) finds evidence suggesting some firms record restructuring charge reversals to avoid earnings declines, to avoid reporting net losses, and to meet analysts' earnings forecasts.

In aggregate, prior benchmark studies suggest that firms manage earnings to avoid an earnings decline, to avoid reporting losses, and to meet analysts' earnings forecasts. Based on the logic of prior studies, our study tests whether firms use the discretionary portion of pension expense in a rational economic manner to meet their current year earnings benchmark which is established as their prior year earnings.

RESEARCH DESIGN

The primary objectives of investigating the phenomenon of earnings management are to discover how firms manipulate earnings, to determine what motivates firms to manipulate earnings, and to evaluate what costs and benefits are associated with firm manipulation. The aggregate accruals method, the specific accruals method, and the earnings-based distribution method are the three research designs prevalent in the earnings management literature (McNichols 2000). Each research design is operationally equipped with its own advantages, disadvantages, and tradeoffs.

Healy and Wahlen (1999) conjecture future research contributions in the area of earnings management will come primarily from documenting the extent and magnitude of the effects of specific accruals and from identifying factors that limit firm ability to manage earnings. The specific accruals research method is based on a disaggregated approach that examines individual accounting items that are subject to substantial manager judgment and are able to significantly impact reported earnings. The most important advantage of the specific accruals research method is the provision for yielding directional predictions based on researcher knowledge and skill. Whereas, the core disadvantage of the specific accruals research method is its inability to analyze simultaneously aggregated effects of accounting levers used by managers in managing earnings (McNichols 2000, Fields et al. 2000, Francis 2001).

We use a specific accruals research model with an earnings-based benchmark as the explanatory variable. The research design is an amalgamation of prior research fundamentals that provide discovery, understanding, and explanation as to whether pension expense is manipulated in a rational economic manner to achieve the earnings-based benchmark. The distinction from prior research is determining whether or not there is an association between the change in pension expense and the amount that firms hypothetically beat or hypothetically miss their benchmark based on premanaged earnings.

The theoretical concepts are operationalized. Whereby, the hypothesis is formalized and stated below in alternate form.

H1: Pension expense is manipulated in a rational economic manner to achieve the current year earnings benchmark, which is the prior year reported earnings.

The hypothesis tests for benchmark behavior. However, an analysis of smoothing behavior is also included in the research. Benchmark behavior is where a firm decreases actual pension expense to increase actual current year earnings in an attempt to reach their earnings benchmark (i.e., prior year earnings). Smoothing behavior exists when firms store up reserves for meeting their earnings benchmark in future periods.

Lagged assets are used to scale variables in an attempt to control for size variations in firms. This procedure works much like common size financial statements for comparative analyses of small firms with large firms and vice versa. The primary cross sectional regression model used to test the hypothesis is presented below.

ChgPE = [[alpha].sub.0] + [[alpha].sub.1] Miss_Dummy + [[alpha].sub.2] Incent + [[alpha].sub.3] Interact + [[alpha].sub.4] ChgEmp + [t=2001.summation over (t=1996)][alpha]t x yrDt + [i=41.summation over(i=1)] [alpha]i x ind[D.sub.i] + [epsilon]

* ChgPE is the change in pension expense equal to current year pension expnse minus prior year pension expense all scalled by lagged assets.

* Miss_Dummy is a dummy variable that equals 1 if the continuous variable, Incent < 0, and 0 otherwise.

* Incent is a continuous variable equal to pretax income absent manipulation minus the applicable benchmark all scaled by lagged assets.

* Interact is an interacton variable equal to Miss_Dummy times Incent.

* ChangEmp is a control variable equal to the number of employees for the current year minus the number of employees for the prior year all scaled by lagged assets.

* yrDt is a dummy variable for each applicable year 1995-2001 with the 1995 dummy effects captured in the intercept.

* IndDi is a dummy variable representing each applicable industry.

* [[alpha].sub.0] is the intercept for Incent [greater than or equal to] 0 where Miss_Dummy = 0.

* [[alpha].sub.0] + [[alpha].sub.1] is the intercept for Incent < 0 where Miss_Dummy = 1.

ChgPE is our measure of earnings management. The proxy development is accomplished by using the feature of SFAS No. 87 whereby the prior year pension expense provides a logical approximation for the firm's premanaged or premanipulated pension expense. That is assuming the number of employees remains constant from year to year. ChgPE is defined as the current year pension expense minus the prior year pension expense all scaled by lagged assets. Thus, ChgPE is the proxy for the extent of manipulation in pension expense after controlling for any change in the number of employees.

Premanipulated actual earnings relative to the earnings benchmark (i.e., prior year earnings) represents the level of capital market incentive for earnings management. The capital market based incentive measure to manipulate earnings is represented by the continuous scaled variable, Incent. Premanipulated actual earnings are derived by adding current year pension expense back to current year earnings to zero-out the effect of current year pension expense and then subtracting prior year pension expense. In essence, prior year pension expense is simply substituted in place of current year pension expense to calculate earnings absent pension manipulation.

Following Burgstahler and Eames (2002), benchmark earnings, as well as premanipulated actual earnings, are reported on a pretax basis rather than an after tax basis because pension expense is reported in financial statements on a pretax basis. Again, the proxy measure for pension expense absent pension manipulation is the prior year pension expense.

Because both benchmark and smoothing incentives exist, it is important to distinguish firms that hypothetically miss their benchmark from firms that hypothetically beat their benchmark. Therefore, a dummy variable (i.e., Miss_Dummy) for hypothetically missing the benchmark is included in the analysis. Miss_Dummy is coded zero for firms that hypothetically beat their benchmark using premanaged earnings. Whereas, Miss_Dummy is coded one for firms that hypothetically miss their benchmark using premanaged earnings. If [[alpha].sub.1] is significant and positive, firms missing their benchmark have a higher intercept than the other firms. If [[alpha].sub.1] is significant and negative, firms missing their benchmark have a lower intercept than the other firms. If "1 is insignificant, there is no difference between the two groups of firms.

After controlling for the change in the number of employees, industry effects, and time fixed effects, the association between ChgPE and the level of capital market incentives (i.e., Incent) for earnings management constitutes this study's test of interest. ChgPE is expected to be positively correlated with the incentive variable, Incent. The slope coefficient for the group of firms that hypothetically beat their benchmark is represented by [[alpha].sub.2]. The slope coefficient for the group of firms that hypothetically miss their benchmark is represented by [[alpha].sub.2] + [[alpha].sub.3]. Thus, we predict that [[alpha].sub.2] > 0, and that [[alpha].sub.2] + [[alpha].sub.3] is > 0.

If [[alpha].sub.2] + [[alpha].sub.3] is significant and positive, this suggests the primary firms of interest hypothetically missing their benchmark are actually decreasing pension expense to increase reported earnings to avoid missing their benchmark. If [[alpha].sub.2] + [[alpha].sub.3] is significant and negative, this suggests firms hypothetically missing their benchmark are not actually decreasing pension expense.

If [[alpha].sub.2] is significant and positive, this suggests the secondary firms of interest hypothetically beating their benchmark are actually increasing pension expense to decrease earnings to move closer to their benchmark than they would otherwise be. If [[alpha].sub.2] is significant and negative, this suggests firms hypothetically beating their benchmark are not actually increasing pension expense.

The logic behind the predictions for [[alpha].sub.2] and [[alpha].sub.2] + [[alpha].sub.3] is that ChgPE is expected to move in the same direction as Incent. For example, if a firm has premanaged earnings equal to $.25 per share and benchmark earnings (i.e., prior year earnings) equal to $.23 per share, the firm is expected to manipulate actual earnings by increasing pension expense by $.02 in order to offset the $.02 excess in premanaged earnings. In this situation, there is a positive $.02 excess in premanaged earnings and the change in pension expense (i.e., ChgPE) is expected to move $.02 in a positive direction as well. Incent (i.e. [[alpha].sub.2]) captures the positive $.02 excess in premanaged earnings. Therefore, because ChgPE and Incent move together in the same direction, a positive correlation is predicted.

On the other hand, if a firm has premanaged earnings equal to $.23 per share and benchmark earnings (i.e., prior year earnings) equal to $.25 per share, the firm is expected to decrease pension expense by $.02 to offset the $.02 negative premanaged earnings. Incent (i.e., [[alpha].sub.2] + [[alpha].sub.3]) captures the negative $.02 deficiency in premanaged earnings. Here again, because ChgPE and Incent move together in the same direction, a positive correlation is predicted.

So in summary, the prior year earnings (i.e., benchmark) create incentives for firms that are in opposite directions depending on their level of premanaged earnings. Therefore, firms hypothetically missing their benchmark are expected to exhibit benchmark behavior by manipulating pension expense to increase actual earnings in order to reach their benchmark earnings. On the other hand, firms hypothetically beating their prior year earnings (i.e., benchmark) are expected to exhibit smoothing behavior by manipulating pension expense to decrease actual earnings so that their actual earnings are closer to their benchmark earnings than they would otherwise be.

Another research consideration is big bath behavior where firms write off excessively large losses in a one time hit against earnings rather than take these losses over time because capital markets place a high premium on firms with steady growth in earnings. Firms generally take a big bath when they will miss reaching their target earnings by a significant amount. Supposedly, these excessively large one time write offs clean up the balance sheet, which in turn, allows these firms to once again produce a steady stream of earnings out in the future. Big bath behavior is not expected to be dominant in the present study because the research design uses a sample screening process that should filter out most, if not all, of these firms. The screening process eliminates firms whose actual earnings performance is not close to their prior year earnings (i.e., benchmark). Benchmark behavior is where a firm decreases actual current year pension expense to increase actual current year earnings in an attempt to reach their target performance. Here target performance is prior year earnings. Smoothing behavior is where a firm increases pension expense to decrease actual earnings in an attempt to store up reserves and be closer to their target performance than they would otherwise be. The logic is that firms closer to their benchmark earnings are more likely to exhibit sensitivity to earnings management incentives such as benchmark behavior and smoothing behavior, whereas; firms missing their benchmark earnings by a large amount are expected to exhibit big bath behavior.

ChgEmp is a control variable that accounts for any variation in the dependent variable (i.e., ChgPE) caused by the change in the number of employees from year to year. ChgEmp is calculated as the current year number of employees minus the prior year number of employees all scaled by lagged assets. In addition, the inclusion of the control variable, ChgEmp, should mitigate confounding results attributable to changes in organizational structure such as mergers and acquisitions. A positive relationship is expected between the change in pension expense (i.e., ChgPE) and the change in the number of employees from year to year (i.e., ChgEmp). The reasoning is plausible because an increase in the number of employees is expected to result in an increase in pension expense, whereas a decrease in the number of employees is expected to result in a decrease in pension expense. Therefore, a positive slope coefficient is predicted for ChgEmp.

Sensitivity tests are performed to examine the robustness of the regression results from equation one. A detailed discussion of the sensitivity test results are presented later in the paper.

The initial sample consists of 4,203 cross-sectional firm observations with applicable data for the period 1995 through 2001 which are derived from the Compustat database. The data is selected because it is cost effective. At the time the sample is selected, it includes all years for which pension data is available from the data source.

Earnings manipulation is expected to be toward the benchmark earnings for levels of actual earnings within a neighborhood near the benchmark earnings amount. So following the concept in Dhaliwal et al. (2002), the final sample consists of 315 firm observations whose difference between the actual earnings per share and the benchmark earnings per share are within a specified range. These firms are more suspect of managing earnings in response to capital market incentives. The pattern of manipulation is considered indeterminate outside the neighborhood near the benchmark earnings amount because incentives for big-bath behavior arise outside the neighborhood.

Dhaliwal et al. (2002) use a five cent after tax earnings per share screen to analyze whether or not firms manipulate taxes in any predictable manner in managing earnings. After consideration of tax effects on earnings, a twelve cent pretax earnings per share screen for the benchmark is initially selected for the present study. However, since the selection of a twelve cent pretax earnings per share screen may be considered as ad hoc, additional sensitivity tests are conducted using both a ten cent pretax earnings per share screen and a fourteen cent pretax earnings per share screen.

The screening process is done on a firm by firm per share pretax basis. To be included in the sample, a firm's pretax actual earnings per share amount must be within $.12 of the pretax benchmark earnings per share amount.

Outlier observations are windsorized so that large and small outlier values are still large and small values within the dataset but are less likely to disrupt the mean, standard estimates, and other statistics that depend upon them. The action taken to address outlier observations should mitigate the possible influence these observations bias the overall statistical outcome.

Multicollinearity diagnostics indicate no problem exists with independent variables being highly collinear. Based on White's joint test for model misspecification and heteroscedasticity, t-statistics are reported using White's corrected t-statistics where applicable, and are otherwise reported using the OLS t-statistics.

RESULTS, INTERPRETATIONS, AND SENSITIVITY ANALYSES

The sample begins with the total number of firms with defined benefit pension plans and no missing data from the Compustat files. After applying the twelve cent earnings per share screening process, there are 315 firms in the final sample. The data is cross-sectional and covers the seven-year period 1995-2001. There are 42 industries in the final sample. Of the sample, 138 firms hypothetically missed their earnings benchmark (i.e., prior year earnings) whereas 177 firms met or exceeded their earnings benchmark. Out of the 138 firms that hypothetically missed their earnings benchmark 95 reduced their actual pension expense as we predicted. So that actual earnings for the 95 firms were closer to their earnings benchmark than they would have been otherwise. Of the 177 firm that hypothetically met or exceeded their earnings benchmark 68 increased their actual pension expense as we predicted. So that actual earnings for these 68 firms were closer to their earnings benchmark than they would have been otherwise.

Table 1 summarizes the sample observations, mean values, standard deviations, minimums, medians, and maximums for selected variables.

In analyzing the information provided in Table 1, a positive mean value for ChgPE implies that on average pension expense increases which decreases earnings; whereas a negative mean value implies that on average pension expense decreases which increases earnings. A positive mean value for Incent indicates that on average firms hypothetically beat their benchmark earnings; whereas a negative value indicates that on average firms hypothetically miss their benchmark earnings. A positive mean value for ChgEmp indicates that on average firms increase in the number of employees in the current year from the prior year; whereas a negative mean value for ChgEmp indicates that on average firms decrease the number of employees in the current year from the prior year. A positive mean value for Miss_Dummy indicates the percentage of firms in the sample that hypothetically miss their benchmark earnings.

The regression results reported in Table 2 use the equation (1) regression model. ChgPE, representing firm manipulation, is expected to be positively correlated with the incentive variable of interest, Incent. The incentive slope is captured in the model for the firms that hypothetically beat their benchmark by [[alpha].sub.2] and for the firms that hypothetically miss their benchmark by [[alpha].sub.2] + [[alpha].sub.3]. The slope on Incent (i.e., [[alpha].sub.2] and [[alpha].sub.2] + [[alpha].sub.3]) represents the estimated average change in pension expense when the applicable incentive variable increases or decreases by one unit. If firms are more concerned with reaching their benchmark than smoothing, we predict that [[alpha].sub.3] > 0.

The slope coefficient (i.e., [[alpha].sub.2] > 0) for the firms that hypothetically beat their benchmark earnings is expected to be statistically significant. The slope coefficient (i.e., [[alpha].sub.2] + [[alpha].sub.3] > 0) for the firms that hypothetically miss their benchmark earnings is expected to be statistically significant as well and is our key variable of interest [alpha].

The foregoing rationale is based on our belief that pension manipulation is a function of the value of the magnitude of hypothetically missing or hypothetically beating the benchmark earnings (i.e., prior year earnings) based on premanaged earnings. The economic substance is captured in the regression model by the main effects of the incentive variable for the two distinct groups of firms. Thus the results on the control variables are not important for interpretation and are not reported.

Table 2 reports the results of the association test using the twelve cent pretax earnings per share screen. The significant F-statistic (i.e., p-value = .0001) for the model indicates strong evidence of linearity. The R2 and Adjusted R2 are .5004 and .4035 respectively, which indicate a high proportion of the change in pension expense is explained by the combination of independent variables.

The slope on Incent captures the average magnitude of the change in pension expense when there is a one unit change in the incentive variable for the two distinct groups of interest. As predicted, the incentive variable for both groups of firms is positive and significant. This pattern of evidence supports the notion that both groups of firms are using pension expense in a predictable rational economic manner based on the magnitude of hypothetically missing or hypothetically beating their benchmark earnings.

The results further indicate smoothing behavior is dominant. For every $1 that premanaged earnings are above the earnings benchmark (i.e., prior year pretax earnings), pension expense increases by $.25 to reduce actual earnings. Whereas, for every $1 that premanaged earnings are below the earnings benchmark (i.e., prior year pretax earnings), pension expense decreases by $.06 to increase actual earnings. One plausible explanation is that auditors may be more vigilant in constraining upward earnings manipulation (i.e., benchmark behavior) than downward earnings manipulation (i.e., smoothing behavior).

It is interesting to note the reported results are in agreement with the findings in the Nelson et al. (2000) survey study where evidence suggests income-decreasing earnings management attempts are more likely to occur with respect to imprecise financial standards. SFAS No. 87 can be classified as an imprecise financial standard partly because of the allowed firm flexibility in choosing the discount rate, the compensation rate, and the expected rate of return on plan assets. Assuming the incentive to manipulate earnings upward to meet benchmark earnings is at least equal to the incentive to manipulate earnings downward to meet benchmark earnings, the pattern of evidence from Table 2 suggests auditors are more vigilant in constraining upward earnings management.

To test the robustness of the primary regression analysis reported in Table 2, the regression analysis is repeated using two additional screens. The additional screens are on opposite sides of the ad hoc twelve cent pretax earnings per share screen. Although the results are not reported, the results are similar which indicate a consistent pattern of association between ChgPE and the incentive variable, Incent.

Since some of the models are heteroscedastic, another sensitivity analysis is performed. Based on the studentized residuals, the most extreme observations within each dataset are eliminated with the end results producing desired homoscedastic models. Although these results are not reported, the results indicate a consistent pattern of association between ChgPE and the incentive variable, Incent. The evidence supports both incentive groups of interest are statistically manipulating pension expense in the predicted direction based on their economic conditions and market incentive.

CONCLUSIONS

Managers have incentives to use discretionary accounting levers to manage earnings to continue a steady stream of earnings to reap stock price advantage and to avoid market devaluation. In addition, many contracting incentives are tied directly or indirectly to earnings based measures which also provide incentives for earnings management.

Our study contributes to the literature by helping to resolve the puzzling lack of significance in prior earnings management studies related to pension accounting. Our study provides some evidence that managers are, in fact, using pension expense in a rational economic manner in regard to prior year pretax earnings. The research provides evidence that prior year earnings (i.e., benchmark earnings) create capital markets incentives for firms in opposite directions depending on their economic status as measured by whether or not firms will miss or beat their benchmark earnings based on premanipulated earnings. So by using "what if" analyses, firms that hypothetically miss their benchmark earnings are predicted and shown to manipulate actual pension expense downward to increase actual earnings; whereas, firms that hypothetically beat their benchmark earnings are predicted and shown to manipulate actual pension expense upward to decrease actual earnings. Therefore as we predicted, both groups of interest are successfully manipulating pension expense in the direction that moves their actual earnings closer to their benchmark earnings (i.e., prior year earnings) than they would otherwise be. The results suggest that smoothing behavior is stronger than benchmark behavior. A plausible explanation is that auditors may be more vigilant in constraining efforts to manage earnings upward than in constraining efforts to manage earnings downward.

Rationale is provided that pension expense is likely the earnings management lever of choice as it allows managers to manipulate earnings directionally as needed without easily being detected by interested outside parties. Furthermore, sensitivity analyses support the research findings are robust to controls for industry and time effects, as well as to the change in the number of employees. Sensitivity analyses further support the results are not driven by a few influential outlier observations.

Our research should, therefore, be of interest to a wide audience such as investors, directors, creditors, auditors, and regulators because it provides relevant information about how managers are using pension expense to manipulate the most value relevant amount (i.e., actual earnings) reported to investors.

Since financial statement integrity is vitally important to capital markets and to the accounting profession as a whole, perhaps our research will be a stimulus for the FASB to continue rethinking its current position regarding pension standards on pension measurement and reporting. Public interest in pension accounting is widespread and futile ground for future research.

REFERENCES

Accounting Principles Board (1966). Accounting for the costs of pension plans. Accounting Principles Board Opinion No. 8.

Barth, M., J. Elliott & M. Finn (1999). Market Rewards Associated with Patterns of Increasing Earnings. Journal of Accounting Research 37 (Autumn): 387-413.

Bergstresser, D., Desai, M. & Rauh, J. (2006). Earnings Manipulation, Pension Assumptions and Managerial Investment Decisions. Quarterly Journal of Economic (Feb 2006) 121, no.1: 157-195.

Blankley, A. (1992). Incentives in Pension Accounting: An Empirical Investigation of Reported Rate Estimates. Dissertation, Texas A&M University, College Station, Texas.

Brown, S. (2001). The impact of pension assumptions on firm valuation. Dissertation, Northwestern University, Evanston, Il.

Burgstahler, D. & I. Dichev (1997). Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics 24 (1): 99-126.

Dhaliwal, D., C. Gleason & L. Mills (2002). Last Chance Earnings Management: Using the Tax Expense to Achieve Earnings Targets. University of Arizona, Working Paper.

Degeorge, F., J. Patel & R. Zeckhauser (1999). Earnings management to exceed thresholds. Journal of Business 72 (1): 1-33.

Fields, T., T. Lys & L. Vincent(2001). Empirical research on accounting choice. Journal of Accounting & Economics 31: 255-307.

Financial Accounting Standards Board (1980). Accounting and Reporting by Defined Benefit Pension Plans. Statement of Financial Accounting Standards No. 35.

Financial Accounting Standards Board (1985). Employers' accounting for pensions. Statement of Financial Accounting Standards No. 87.

Financial Accounting Standards Board (1998). Employers' Disclosures about Pension and Other Postretirement Benefits. Statement of Financial Accounting Standards No. 132.

Financial Accounting Standards Board (2006). Exposure Draft: Employers' Accounting for Defined Benefit Pension and Other Postretirement Plans-an amendment of FASB Statements No. 87, 88, 106, and 132(R). Statement of Financial Accounting Standards Board Project: Pensions, Exposure Draft Reference Number 1025-300.

Francis, J. (2001). Discussion of empirical research on accounting choice. Journal of Accounting and Economics 31, 309-319.

Healy, P. & J. Wahlen (1999). A review of the earnings management literature and its implications for standard setting. Accounting Horizons 13-4: 365-38.

Kwon, S. (1989). Economic determinants of the assumed interest rate in pension accounting. Dissertation, Univ. of Oklahoma,.

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

McNichols, M. (2000). Research design issues in earnings management studies. Journal of Accounting and Public Policy 19 313-345.

Moehrle, S. (2002). Do firms use restructuring charge reversals to meet earnings targets? The Accounting Review 77-2:397-413.

Nelson, M., J. Elliott & R. Tarpley (2000). Where do Companies Attempt Earnings Management, and When Do Auditors Prevent It? Cornell University and Washington University, Working Paper (October Draft Copy). PBGC (2000). About PBGC--Mission and Background. [email protected] (Version update 7-19-2000).

Powell, G., C. Wasley & G. Waymire (1993). The Stock Price Effects of Alternative Types of Management Earnings Forecasts. The Accounting Review 68-4:896-912.

Weishar, J. (1997). A Cross-Sectional Evaluation of the Ability of Authoritative Standards to Influence the Use of Defined Benefit Pension Plan Actuarial Assumptions as an Income Smoothing Technique. Dissertation, University of Arkansas, Fayetteville, Arkansas.

Paula Diane Parker, University of South Alabama Martha Lair Sale, Sam Houston State University

Table 1: Descriptive Statistics From Regression Analysis

 Standard
Variable n Mean Deviation

ChgPE 315 -0.000124 0.006041
incent 315 -0.000178 0.019516
ChgEmp 315 0.000001 0.000003
miss_dummy 315 0.438095 0.496942

Variable Median Minimum Maximum

ChgPE -0.000028 -0.054763 0.078231
incent 0.000456 -0.248927 0.078475
ChgEmp 0.000000 -0.000018 0.000031
miss_dummy 0.000000 0.000000 1.000000

Table 2: CrossSectional pooled Effects Estimation Using $.12 Screen
with Time and Industry Fixed Effects

 One Tail
Variable Prediction Coefficient p-value

intercept + -0.0058 .9994
miss_dummy - -0.0002 .3937
incent + 0.2450 .0001
interact + / - -0.1854 .0001
[[alpha].sub.0] + [[alpha].sub.1] - -0.0060 .0003
[[alpha].sub.2] + [[alpha].sub.3] + 0.0596 .0006
F-statistic as p-value .0001
[R.sup.2] .5004
Adjusted [R.sup.2] .4035