(PDF) Pension plan accounting estimates and the freezing of defined benefit pension plans

Contents lists available at ScienceDirect

Journal of Accounting and Economics

Journal of Accounting and Economics 51 (2011) 115–133

0165-41

doi:10.1

n Corr

E-m1 U

negotia

lower n

selectio

journal homepage: www.elsevier.com/locate/jae

Pension plan accounting estimates and the freezing of definedbenefit pension plans

Joseph Comprix a, Karl A. Muller IIIb,n

a Whitman School of Management, Syracuse University, Syracuse, NY 13244, USAb Department of Accounting, Smeal College of Business, 384 Business Building, The Pennsylvania State University, University Park, PA 16802, USA

a r t i c l e i n f o

Article history:

Received 8 May 2008

Received in revised form

25 May 2010

Accepted 16 June 2010Available online 25 June 2010

JEL classification:

M41

Keywords:

Defined benefit pension plans

Pension plan freeze

Expected rate of return assumption

Discount rate assumption

Sarbanes-Oxley Act

01/$ - see front matter & 2010 Elsevier B.V. A

016/j.jacceco.2010.06.003

esponding author. Tel.: +1 814 865 0202; fax

ail address: [emailprotected] (K.A. Muller III).

sing a sample of seven major U.S. steel pr

tions, producers facing excess capacity and de

et income. However, the lower reported incom

n of accounting estimates.

a b s t r a c t

This study provides evidence that, when ‘‘hard’’ freezing their defined benefit pension

plans, employers select downward biased accounting assumptions to exaggerate the

economic burden of their benefit plans. Downward biased expected rates of return and

discount rates allow managers to increase reported pension expenses and, for discount

rates, allow managers to increase reported pension liabilities. We find that prior to the

Sarbanes-Oxley Act, both rates are downward biased when firms freeze their plans,

whereas after SOX the bias is lower. This finding is consistent with managers

opportunistically biasing pension estimates to obtain labor concessions during periods

of reduced regulatory scrutiny.

1. Introduction

To obtain concessions from labor, managers in some cases face incentives to report lower earnings by adoptingdownward biased accounting estimates. In the only study to investigate this incentive directly,1 Liberty and Zimmerman(1986) examine routine union labor negotiations and fail to find that managers reduce reported earnings through the useof discretionary accruals. Liberty and Zimmerman (1986) offer several possible explanations for their failure to findsupportive evidence. For instance, it may be due to their tests lacking sufficient power, or it may reflect managers notneeding to artificially depress reported earnings due to poor firm performance or managers not manipulating reportedearnings due to unions’ ability to undo such manipulation.

The purpose of this study is to shed further light on the extent to which managers opportunistically choose downwardbiased accounting estimates in an effort to reduce labor costs. We seek to provide evidence on this question by investigatingwhether employers systematically assume downward biased pension assumptions when ‘‘hard’’ freezing their defined benefit

ll rights reserved.

: +1 814 863 8393.

oducers during the 1980s, DeAngelo and DeAngelo (1991) find that during non-routine union

clining demand used one-time special charges for real restructuring activities to report materially

e in their paper is due largely to discretionary timing of one-time charges, rather than discretionary

www.elsevier.com/locate/jae

dx.doi.org/10.1016/j.jacceco.2010.06.003

mailto:[emailprotected]

dx.doi.org/10.1016/j.jacceco.2010.06.003

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133116

pension plans.2 Specifically, we examine whether firms that hard freeze their pension plans show increasingly downwardbiased expected rate of return and discount rate assumptions in the years leading up to and including the freeze year.

We choose to investigate pension assumptions rather than other common measures of accrual management, such asdiscretionary accruals, because doing so allows us to focus on accrual estimates that are directly observable, largelyunrelated to firms’ operating performance, and directly related to employees’ perceptions of labor costs. We further chooseto investigate the hard freezing of defined benefit pension plans for two reasons. First, employers are increasingly claimingthat their defined benefit pension obligations are burdensome. By hard freezing their pension plans, employers can reducethe growth in their pension liability. Not surprisingly, therefore, the number of hard freezes has been substantial, withapproximately 8% of single-employer plans hard frozen as of 2008 (U.S. General Accountability Office, 2009). Second,reports in the financial press (e.g., Francis and Schultz, 2003) suggest that employers face incentives to use their discretionwithin generally accepted accounting principles to exaggerate the burden of their defined benefit plans when freezing theirplans. Under Accounting Standards Codification 715: Compensation—Retirement Benefits (FASB, 2009; this standard includesformer SFAS No. 87: Employers’ Accounting for Pensions, FASB, 1985), employers have considerable discretion in choosingtheir pensions’ expected rate of return and discount rate. By selecting a relatively low expected rate of return, employerscan increase reported pension expenses. Similarly, by selecting a relatively low discount rate, employers can increase bothreported pension expenses and the reported pension obligation.

To increase our confidence that any evidence of downward biased pension estimates reflects managerial opportunismrather than other factors associated with the freezing of pension plans, we take advantage of a relatively exogenous shock,namely, the passage of the Sarbanes-Oxley Act (SOX) by Congress in July 2002. The primary goal of SOX was to reduceaccounting fraud and earnings management. Cohen et al. (2008) and Jiang et al. (2010) find that, consistent with this aim,accrual-based earnings management declined significantly following the passage of SOX. In our setting, this exogenousregulatory shock provides an opportunity to distinguish opportunistic behavior from other factors associated with firmsfreezing their plans. In particular, if downward biased pension estimates reflect opportunistic behavior, then downwardbiased rates should be less prevalent for firms freezing their plans after the passage of the SOX—consistent with reducedmanagerial discretion to select biased pension estimates. In contrast, if downward biased estimates reflect other factors,then we should not observe a reduction in the prevalence of downward biased rates for firms freezing their plans.

Based upon the discussion above, in our primary analyses we investigate whether firms that hard freeze their pensionplans use increasingly downward biased estimates of their pensions’ expected rate of return and discount rate in the freezeyear and the three years prior to the freeze. Focusing first on the expected rate of return estimates, we find that for the pre-SOX period, pension-freezing employers’ expected rate of return estimates are approximately 45 basis points lower thanthose of non-freezing firms in the freeze year, consistent with incentives to increase reported pension costs (as well asdecrease expectations that returns on pension plan assets will cover possible funding gaps in the reported pensionobligation in the future).3 In addition, pension-freezing employers assume relatively lower rates of return in the two yearsprior to the freeze year, consistent with employers gradually adjusting the assumed rates downward.4 For the post-SOXperiod, we find evidence of a significant reduction in the use of downward biased expected rate of return estimates,consistent with SOX mitigating managers’ opportunistic behavior. Turning to the discount rate estimates, we find that forthe years prior to the passage of SOX, pension-freezing employers assume discount rates that are approximately 18 basispoints lower than those of non-freezing employers in the freeze year, consistent with incentives to increase reportedpension costs and the reported pension obligation. We further find that employers assume relatively lower discount ratesin the three years prior to the freeze year.5 For the years following the passage of SOX, we find evidence of a significantreduction in the use of downward biased discount rate estimates, again consistent with SOX mitigating managers’opportunistic behavior. These findings are obtained after controlling for various other determinants of the expected rate ofreturn and discount rate assumptions, which is important given changing macroeconomic conditions surrounding thepassage of SOX (see Zhang, 2007 for a discussion).

In additional analyses, we find that freeze decisions occur more frequently during periods associated with a highereconomic burden of defined benefit plans, that is, when employers face lower operating cash flows and report losses, but

2 By freezing a defined benefit pension plan, an employer can effectively reduce or eliminate the accrual of new benefits under the plan and thereby

its pension costs and growth in its pension liability. A ‘‘hard’’ freeze eliminates the accrual of new benefits for all employees. Pension plans can also be

‘‘soft’’ frozen (growth in benefits is limited) or ‘‘partially’’ frozen (pension plans are closed to some classes of employees while other classes of employees

continue to enjoy the accrual of new benefits; usually, existing employees continue in the plan while new employees are shut out). In this paper we focus

on hard freezes due to lack of reliable information in financial and pension filings on occurrences of soft or partial freezes.3 In supplemental analysis, we find evidence of an even larger downward bias in the pre-SOX period for firms that receive a relatively larger boost to

net income from freezing their plans.4 Alternatively, downward biased estimates could arise if actual rates of return on the plan assets were relatively lower in the years leading up to and

including the freeze decision, or even in future years. However, examination of the actual rates of return on plan assets in the freeze year and the three

years prior to the freeze year (discussed below) fails to find evidence that actual rates of return were systematically lower for the pension-freezing firms.

In addition, future returns are actually higher in the year following a freeze. Thus, potential managerial pessimism regarding the expected rate of return

does not appear to arise from the behavior of prior, contemporaneous, or future actual returns on plan assets.5 The downward bias observed for the discount rate assumption is smaller in magnitude than that for the expected rate of return assumption. This

may reflect managers having less flexibility in setting discount rate estimates.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 117

are constrained by the expected level of employee resistance, that is, when plans involve relatively more employees or aresubject to collective-bargaining agreements.

Taken together, our findings provide evidence that managers opportunistically select downward biased accountingestimates to influence labor contracts during periods of reduced regulatory scrutiny. This evidence contributes to theliterature in two ways. First, it fills an important void in the literature regarding whether managers use biased accountingestimates when seeking labor cost concessions. Second, it lends support to the view that SOX has effectively reducedmanagerial discretion over specific accounting estimates. Using pension plans’ expected rate of return and discount rateassumptions offers a potential advantage over discretionary accrual estimates because they are directly observable and notlikely to be affected by firm performance.

The remainder of the paper is organized as follows. In Section 2 we provide background on the freezing of definedbenefit pension plans and on financial reporting rules with respect to pension plans’ expected rate of return and discountrate estimates, and we develop our research hypotheses on employers’ incentives to opportunistically select lowerexpected rate of return and discount rate estimates. We discuss the sample selection process and provide descriptivestatistics in Section 3. In Section 4 we present our empirical results. Finally, Section 5 summarizes and concludes the study.

2. Background and hypotheses

In this section, we first provide background on the freezing of defined benefit pension plans, and on financial reportingrules with respect to defined benefit pension plans’ expected rate of return and discount rate estimates. We then developour hypotheses on pension plan sponsors’ incentives to opportunistically select lower expected rate of return and discountrate estimates to influence plan participants when freezing their defined benefit plans.

2.1. The freezing of defined benefit pension plans

Employers generally offer one of two types of pension plans to their employees: defined contribution plans and defined benefitplans. Defined contribution plans require employers to make specified contributions towards employees’ retirement plans.Investment risk then falls upon the employee, with the employer having no further pension obligation after the contribution hasbeen made. In contrast, defined benefit plans represent a formal promise that the employer will pay pre-determined benefitsupon retirement. This type of plan assumes that sufficient assets will be on hand to pay the pre-determined benefits. Employers’contributions to defined benefit plans are uncertain, however, as they depend on the returns of the assets in the pension plan,prior employer contributions, and payouts to retirees. Defined benefit plans can therefore become too costly for an employer tomaintain. In such cases, employers will have an incentive to reduce their obligations under the pension plan.

By law, employers are obligated to pay pension benefits already earned through service. However, benefits attached tofuture service are not protected. Companies can therefore reduce or eliminate growth in benefits by ‘‘freezing’’ theirpension plans. Employers benefit from constraining the growth of pension benefits in several ways. First, becauseadditional service by employees does not increase the firm’s reported pension liability (in a hard freeze), or increases thefirm’s pension liability much more slowly than otherwise (in a soft or partial freeze), freezing a pension plan decreases afirm’s pension costs. In addition, because the reported pension liability is reduced immediately by the amount of expectedfuture benefits previously included in the liability that the firm will no longer be liable for, freezing the accrual of planbenefits immediately improves the funding status of the plan. Such effects can boost a firm’s bottom line substantially. Forexample, Berkshire Hathaway booked a $70 million gain in 2004 upon freezing the pension plan of a subsidiary. Finally, ifplan assets grow at a faster rate than the frozen plan obligation, an underfunding problem may resolve itself over timewithout requiring additional contributions from the firm.6

Frozen defined benefit pension plans can be replaced by a new plan (such as a defined contribution plan or a hybridplan—e.g., cash balance plan), limiting the financial impact on employees. However, employers are not required tosubstitute a new plan for an old plan. Employee losses thus depend on the type of new plan introduced, if any.Unfortunately, employers typically do not disclose estimates of employee losses resulting from the conversion of one typeof pension plan to another. Nevertheless, recent studies by leading consulting firms indicate that the two driving forcesbehind employers’ freeze decisions are long-term cost savings and reductions in contribution volatility (Aon Consulting,2003; Hewitt Associates, 2006; Mercer Human Resources Consulting, 2006).

2.2. Financial accounting rules for defined benefit pension plans

Employers with defined benefit pension plans must follow Accounting Standards Codification (ASC) 715: Compensa-

tion—Retirement Benefits (FASB, 2009; this standard includes former SFAS No. 87: Employers’Accounting for Pensions, FASB,

6 In contrast, if a firm decides to immediately terminate its pension plan, the current value of the benefits owed would have to be paid out to

employees at that time (either as a lump sum or through the purchase of an annuity). This means that if an underfunded plan is terminated, the employer

must immediately make up any funding shortfall (except in a distress termination, which is rare). Employers are thus likely to prefer freezing an

underfunded plan, and possibly terminating the plan later, to immediate termination. Stone (1987), Mittelstaedt (1989), Thomas (1989), and Petersen

(1992), among others, provide evidence on employers’ decisions to terminate their defined benefit pension plans.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133118

1985) in reporting their net pension obligation and net periodic pension costs. The net pension obligation is measured as theprojected benefit obligation less pension plan assets measured at fair value. After excluding deferred gains and losses andunamortized prior service costs (which are often large), the net pension obligation is reported on employers’ balance sheets.7

Net periodic pension costs—the sum of service costs, interest costs, amortized deferred gains and losses, and amortized priorservice costs less the expected return on pension plan assets—are reported on employers’ income statements.

ASC 715 requires that three important assumptions be made to calculate the net pension obligation and net periodicpension costs: the expected rate of return (ERR), the salary inflation rate (SIR), and the discount rate (DR). The projectedbenefit obligation is affected over the course of a year by service costs, interest costs, actuarial gains and losses, priorservice costs from plan amendments, acquisitions, divestitures, curtailments, settlements, contributions, and benefits paid.An increase in the assumed ERR does not affect the projected benefit obligation. An increase in the assumed SIR, however,raises the projected benefit obligation. Further, because the assumed DR is used to measure the projected benefit obligationat its present value, an increase in the assumed DR leads to a decrease in the projected benefit obligation. Turning to netperiodic pension costs, an increase in the assumed ERR lowers net periodic pension costs directly.8 An increase in theassumed SIR raises the interest and service cost components of pension costs due to the higher projected benefit obligationand higher assumed future compensation levels. In addition, differences between the actual return on the pension planassets and the expected return and changes in the assumed SIR are treated as deferred gains or losses, which are recognizedin pension costs using the corridor amortization method. Finally, the assumed DR affects the service and interest costcomponents of pension expense. An increase in the assumed DR can increase or decrease reported pension costs. Theimpact on pension costs is ambiguous as a higher assumed DR: (1) both lowers the projected benefit obligation and raisesthe interest rate used to calculate the interest cost component of net periodic pension costs and (2) lowers the service costcomponent of pension costs due to the benefits earned through current service being discounted at a higher rate. Inpractice, firms generally face lower (higher) reported pension costs if the discount rate assumption is raised (lowered).9

Due to the lack of specific guidance in ASC 715, employers have considerable discretion over the choice of the threeassumptions above. The assumed ERR should reflect historical and expected future returns on the plan assets (formerlyparagraph 45 of SFAS 87). The assumed SIR should reflect expected future compensation levels, which include factors suchas expected future changes in general price levels, productivity, and seniority, as well as expected future promotions(formerly paragraph 46 of SFAS 87). And the assumed DR should reflect the rate at which the pension obligation could beeffectively settled (i.e., rates implicit in current annuity contracts), or the rate of return on high-quality fixed incomeinvestments currently available and expected to be available over the duration of the pension liability (formerly paragraph45 of SFAS 87). However, as reliable benchmarks for evaluating these estimates are not readily available, considerablesubjectivity is involved when employers select their estimates.10

2.3. Hypotheses

Given that employees likely receive reduced pension benefits if their pension plans are frozen, one would expect themto resist pension freezes. The popular press suggests that on average employees will show less resistance to cuts in theirbenefits when pension plans appear to be more burdensome to their employers (Francis and Schultz, 2003, C.1).Accordingly, employers wishing to freeze their defined benefit pension plans have an incentive to make their plans appearas costly as possible. Given the discretion available to managers in making pension plan-related assumptions, managersmay opportunistically assume a relatively low ERR to increase reported pension expenses and make it appear less likelythat the pension’s liability can be met though returns on current assets. Similarly, managers may opportunistically assumea relatively low DR (which increases both reported pension expenses and the reported pension liability). However,managers are not likely to select a relatively high SIR to exaggerate the costs of maintaining their defined benefit plans, as itis costly to artificially signal higher expected future salaries to employees.

In other contexts, prior research shows that managers opportunistically select pension ERRs.11 Specifically, Bergstresseret al. (2006) provide evidence that managers assume higher ERRs to boost income prior to undertaking acquisitions and

7 Under SFAS 158: Employers’ Accounting for Defined Benefit Pension and Other Post-Retirement Plans (FASB, 2006; superseded by Accounting Standards

Codification 715: Compensation—Retirement Benefits, FASB, 2009), effective for fiscal years ending after December 15, 2008, the net pension obligation no

longer excludes deferred gains and losses and unamortized prior service costs. In addition, deferred gains and losses and prior service costs not

recognized in net periodic pension costs are recorded as part of other comprehensive income.8 The year-end entry to record pension expenses generally involves three components: net periodic pension costs, the change in the net pension

obligation, and plan contributions. In this entry, the expected return affects both net periodic pension costs and the change in pension plan assets

(adjusted for deferred gains or losses for the year).9 We surveyed the financial statements of 100 randomly selected firms disclosing their sensitivity of pension costs to a change in the DR (a relatively

large number of firms voluntarily disclose such sensitivity analysis in their Management Discussion and Analysis or in their pension footnote).

Interestingly, we found that all of the firms surveyed reported that reducing the assumed DR would increase reported pension costs.10 Note that employers’ choice of pension estimates for financial reporting purposes does not directly affect employers’ required contributions to

their defined benefit pension plans, as ERISA and IRC funding rules for minimum funding contributions to defined benefit plans are based on different

criteria than financial accounting rules for pensions.11 In a context of post-retirement benefits, Amir and Gordon (1996) find that managers select discount rate and health care cost trend assumptions

that understate post-retirement benefit obligations when leverage is high (i.e., when the probability of violating debt covenants is high), but that evidence

of opportunistic assumption setting when plans are amended or when firms have extreme earnings-price ratios is inconsistent.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 119

CEO option exercises, whereas they assume lower ERRs to reduce income prior to CEO option grants. Comprix and Muller(2006) provide further evidence that managers assume higher ERRs to boost income when pension income is reported, inperiods prior to CEOs selling shares in the open market, and when leverage is high.

This discussion leads us to predict that employers with defined benefit plans select downward biased ERR and DR

estimates in the freeze year, as managers face incentives to make their plans look as costly as possible in the freeze year. Inaddition, we predict that employers select downward biased ERRs and DRs in the years leading up to the freeze year, asauditors closely examine changing accounting estimates and thus managers also face incentives to make their plans appearmore costly in the years just prior to the freeze year.

Ultimately, whether employers opportunistically select pension accounting estimates when they hard freeze theirdefined benefit plans is an open question. Offsetting the advantages of choosing downward biased pension assumptionsare important potential costs. For instance, the resulting increase in reported pension expenses and in the reported pensionobligation can decrease executive compensation based on reported earnings, increase the risk of violating debt covenants,and adversely affect stock prices. In addition, facing the loss of pension benefits, employees are likely to closely evaluatethe claims of their employers. Moreover, a number of large corporate scandals in the early 2000s led to more stringentdisclosure requirements that aim to strengthen the integrity of firms’ financial statements.

The increase in regulatory scrutiny is of particular relevance to this study. On July 20, 2002, Congress passed the Sarbanes-Oxley Act in an effort to reduce the induce of accounting fraud and earnings management. Consistent with SOX achieving thisgoal, Cohen et al. (2008) and Jiang et al. (2010) find that accruals-based earnings management significantly declinedfollowing the passage of SOX, but that harder to detect and more costly real earnings management increased. Similarly, inDecember of 2002 the SEC cautioned companies that it might contest expected rate of return assumptions, especiallyrelatively high assumptions, and in 2003 the FASB passed SFAS 132 (R): Employers’ Disclosures About Pensions and Other Post-

retirement Benefits (FASB, 2003; superseded by Accounting Standards Codification 715: Compensation—Retirement Benefits,FASB, 2009), effective for fiscal years ending after December 15, 2003, which required extensive new disclosures to assistfinancial statement users when evaluating an employer’s plan assets, expected long-term rates of return, pension obligations,and periodic pension costs.12 These exogenous regulatory shocks, occurring in relatively close proximity, provide us anopportunity to distinguish between opportunistic behavior and other factors associated with firms freezing their plans (e.g.,increased pessimism among managers of pension-freezing firms). In particular, if the use of downward biased pensionassumptions reflects opportunism, then to the extent that the passage of SOX (and other related regulatory changes) reducedmanagers’ discretion to select biased pension estimates, we should observe a significant reduction in the prevalence ofdownward biased pension assumptions for firms that freeze their pension plans following the passage of SOX.

3. Sample selection and descriptive statistics

3.1. Sample selection

We begin our sample selection with all firms on Compustat that have defined benefit pensions plans during the years1991–2008.13 Firms that hard freeze all of their defined benefit plans are then identified as follows. First, we identify thosefirms that go from having non-zero service costs (Compustat data item #331) to zero service costs. We then examine thesefirms’ annual reports to determine whether a switch to zero service costs is due to the freezing of all of the firm’s pensionplans, and if so, the year of the freeze. (A number of firms switched to zero service costs because they disposed of a divisionor segment associated with their defined benefit plans, or because they terminated their plans; these firms are not coded asfreezing their plans.)14 In addition, we identify firms that hard freeze their plans by searching Lexis-Nexis using the term‘‘freeze w/12 of benefits or frozen w/12 plan.’’ This alternative approach is necessary as some firms that hard freeze all oftheir defined benefit pension plans have non-zero service costs due to the existence of supplemental executive retirementplans (SERPs). All other firms with defined benefit plans on Compustat not identified as hard freezing their plans are usedas our control sample.

The above sample selection procedures identify 274 firms that hard freeze their defined benefit plans and 4439 controlfirms, resulting in a total of 28,452 firm-year observations.15 Note that 6% of our sample firms are found to hard freeze theirplans. This compares favorably to the 8% observed by the GAO (2009) for 2008. The slightly lower percentage in our samplemay be due in part to the fact that, while our sample period is greater than that observed by the GAO, we only examinelisted firms, which tend to have larger plans that are less likely to be hard frozen. The rate reported by the GAO also

12 For instance, the revised standard required firms to provide the following new disclosure related to the expected rate of return estimate: (1)

asset allocation information for each major category of plan assets, (2) narrative descriptions of investment policies and strategies, including

target allocation percentages or range of percentages, (3) narrative descriptions of the basis used to determine the estimate, and (4) disclosure of other

information regarding the risk of the plan assets.13 We begin our sample period in 1991 because we obtain union status information from firms’ Form 5500 filings, which first reports this information

for 1991 filings.14 Our results (reported below) are insensitive to the deletion of firms terminating their plans.15 Observations subsequent to the freeze year for the freeze firms are excluded from all analyses (with the exception of our sensitivity analysis

examining future actual rates of return on pension plan assets).

Table 1Annual market conditions and distribution of firms hard freezing their defined benefit plans by freeze year and industry.

Year Average actual plan

asset return (%)

S&P 500 return (%) 30-year treasury

rate (%)

Average funding

level (%)

# of Firms with

hard freezes

Panel A: Annual market conditions and number of firms hard freezing their defined benefit plans

1991 14.17 18.86 7.77 104.92 6

1992 7.62 7.34 7.21 101.28 9

1993 9.70 9.76 6.22 96.11 11

1994 1.17 �2.32 7.69 96.72 14

1995 15.26 35.20 6.02 96.68 21

1996 10.79 23.61 6.80 103.01 10

1997 14.14 24.69 5.80 106.85 15

1998 9.56 30.54 5.09 101.46 13

1999 11.04 8.97 6.49 113.80 14

2000 4.15 �2.04 5.54 106.64 17

2001 �5.31 �17.26 5.43 88.77 11

2002 �7.78 �24.29 4.85 73.24 10

2003 12.63 32.19 4.97 76.01 15

2004 8.09 4.52 4.59 78.17 27

2005 7.99 4.44 4.57 78.54 27

2006 8.67 12.64 4.83 84.48 30

2007 6.00 4.10 4.52 89.59 20

2008 �22.39 �36.27 2.96 72.73 3

Total 274

Industry All Hard frozen All (%) Hard frozen (%)

Panel B: Industry membership (of employers with defined benefit plans and those with hard frozen defined benefit plans)

Basic 2771 15 9.74 5.47

Capital 4877 53 17.14 19.34

Construction 734 5 2.58 1.82

Consumer 9223 112 32.42 40.88

Energy 1215 9 4.27 3.28

Finance 3578 51 12.58 18.61

Transportation 1508 9 5.30 3.28

Utilities 4546 20 15.98 7.30

Total 28,452 274 100.00 100.00

Panel A shows the annual market conditions and number of firms hard freezing their pension plans. Hard frozen defined benefit pension plans are plans

where employers have eliminated the accrual of additional pension benefits for all employees. Panel B shows the distribution of pension plans and hard

pension freezes by industry. Industry definitions are from Sharpe (1982).

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133120

includes foreign domiciled firms, which we exclude from our empirical analyses (resulting in the deletion of 4202 firm-year observations) as their pension rate assumptions are based on home country economic conditions.

Data for other financial statement information are obtained from Compustat.16 Data on firms undertaking acquisitions,and on the raising of equity and debt capital, come from the Securities Data Corporation (SDC) database. Data regardingCEO trading activity are obtained from Thomson Financial Insiders Filings. These additional data requirements reduce thenumber of firm-year observations to 17,560.

We obtain information on the union status of firms’ plans by searching firms’ Form 5500 filings. The EmployeeRetirement Income Security Act (ERISA) of 1974 and the Internal Revenue Code (IRC) of 1986, as amended, require privatepension plans with over 100 participants to file detailed information regarding the funding status of their plans with theIRS annually. While firms are required to disclose whether their plans are subject to a collective-bargaining agreement, alarge number of plans do not disclose such information (Petersen, 1992). We classify a firm’s plans as being subject to acollective-bargaining agreement if any of the firm’s plans are subject to such agreements.17

3.2. Descriptive statistics

Table 1 provides information on the distribution of firms that hard freeze their defined benefit plans by year (Panel A)and by industry (Panel B). Panel A shows that the number of plans frozen varies considerably over time, with 1995 and

16 Due to inconsistent collection of actual return on plan assets by Compustat during the years 1998�2003 (because of the different tabular reporting

requirements under SFAS 132), we hand collect actual return data (8519 observations). This issue was apparently resolved with the implementation of

SFAS 132 (R) in 2004.17 Due to Form 5500s not being publicly released for two years following their filing with the IRS, we assume that firms subject to collective

bargaining agreements in 2006 were also subject to such agreements in 2007 and 2008.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 121

2004–2007 displaying the largest number of hard freezes. This result appears to reflect defined benefit plans’ overallhealth. For instance, as Panel A further demonstrates, the years associated with a large number of hard freezes tend tofollow years with a decrease in average funding levels. The level of funding appears to be related in part to equity marketreturns (as indicated by the actual returns to the S&P 500 index) and falling discount rates (which increases pensionliabilities). With declining interest rates during the sample period, defined benefit pension plan obligations becameincreasingly burdensome, due to their being discounted using lower interest rates.

In Panel B of Table 1 we see that the industry membership of firms that hard freeze their plans is generallyrepresentative of the industry membership of firms that have defined benefit pension plans in the first place. There aresome notable exceptions, however. For example, the number of firms in the basic, transportation, and utility industries thathard freeze their defined benefit pension plans is relatively low. This is likely due to the fact that these industriescommonly have collective-bargaining agreements, which could reduce firms’ ability to hard freeze their defined benefitpension plans as employees must agree to such freezes.

Table 2 provides descriptive statistics for both our main explanatory variables and the control variables used in ourempirical analyses. To reduce the influence of extreme observations, all variables are winsorized at the top and bottom 1%.The table first shows that, on average firms assume an ERR of 8.50%. There is considerable variation in the assumed rates,however, with the 25th percentile (8.00%) 100 basis points lower than the 75th percentile (9.00%). Panel A of Fig. 1 allowsus to look at the variation for individual years. We find that within the same sample year, considerable variation inassumed rates exists, with the difference between the 25th and 75th percentiles averaging 115 basis points within eachsample year. Turning attention to assumed DRs, Table 2 shows that firms assume an average DR of 6.91%. We again find

Table 2Descriptive statistics.

Variable Mean Std. dev. P25 P50 P75

Freeze 0.01 0.11 0.00 0.00 0.00

ERR 8.50 0.99 8.00 8.50 9.00

DR 6.91 0.98 6.00 7.00 7.50

Underfunded 0.66 0.47 0.00 1.00 1.00

Funding% 0.91 0.28 0.75 0.90 1.07

FirmSize 7.03 2.02 5.71 7.01 8.34

PlanSize 0.16 0.17 0.04 0.10 0.20

OperatingCashFlow 0.08 0.07 0.04 0.08 0.12

Loss 0.18 0.38 0.00 0.00 0.00

DSales% 0.09 0.23 �0.01 0.06 0.15

UnionPlan 0.12 0.33 0.00 0.00 0.00

UnionKnown 0.41 0.49 0.00 0.00 1.00

PostSOX 0.36 0.48 0.00 0.00 1.00

ARR 5.79 11.76 2.51 8.17 12.52

TaxRate 0.25 0.34 0.21 0.34 0.38

PensionIncome 0.17 0.38 0.00 0.00 0.00

PensionSensitivity 1.77 3.38 0.33 0.78 1.70

Acquisition+1 0.15 0.35 0.00 0.00 0.00

EquityIssue+1 0.01 0.04 0.00 0.00 0.00

DebtIssue+ 1 0.02 0.05 0.00 0.00 0.00

Leverage 0.28 0.23 0.10 0.24 0.42

CEO_NetSelling+1 0.00 0.01 0.00 0.00 0.00

Distributional statistics are presented for the variables used in our empirical analyses. The sample consists of all firms on Compustat having defined

benefit pensions plans during the years 1991–2008 and also having data on the SDC and Thomson Financial Insiders databases (17,560 firm-year

observations). The sample period begins in 1991 because union status information is obtained from firms’ Form 5500 filings, which first reports this

information in 1991. Freeze is an indicator variable equal to 1 if the firm’s defined benefit pension plans are hard frozen (subscript indicates year relative

to freeze decision), and 0 otherwise. ERR is the assumed expected rate of return on the defined benefit pension plan assets. DR is the assumed discount

rate for the projected benefit obligation. Underfunded is an indicator variable equal to 1 if the firm’s defined benefit pension plans are underfunded, and 0

otherwise. Funding% is the pension plan assets divided by the projected benefit obligation. FirmSize is the natural log of total assets measured in millions

of dollars. PlanSize is the projected benefit obligation divided by total assets. OperatingCashFlow is the cash flow from operations divided by total assets.

Loss is an indicator variable equal to 1 if the firm reported a loss for the year, and 0 otherwise. DSales% is the percentage change in sales. UnionPlan is an

indicator variable equal to 1 if any of the firm’s defined benefit pension plans are subject to a collective-bargaining agreement, and 0 otherwise.

UnionKnown is an indicator variable equal to 1 if the status of whether a firm’s defined benefit pension plans are subject to a collective-bargaining

agreement is known, and 0 otherwise. PostSOX is an indicator variable equal to 1 if the fiscal year occurs following the passage of the Sarbanes-Oxley Act

of 2002, and 0 otherwise. ARR is the actual rate of return on the defined benefit pension plan assets. TaxRate is income taxes divided by the absolute value

of pre-tax income. Funding% is the defined benefit pension plan assets divided by the projected benefit obligation. PensionIncome is an indicator variable

equal to 1 if the firm reports pension income (i.e., negative pension cost), and 0 otherwise. PensionSensitivity is the defined benefit pension plan assets

divided by the absolute value of operating income. Acquisition+ 1 is an indicator variable equal to 1 if the firm undertakes an acquisition during year t+1,

and 0 otherwise. EquityIssue+ 1 is the amount of public and private common and preferred equity raised during year t+1 divided by total assets.

DebtIssue+ 1 is the amount of public and private common and preferred debt raised during year t+1 divided by total assets. Leverage is long-term debt

divided by total capital (i.e., market value of equity plus long-term debt). Finally, CEO_NetSelling+ 1 is the number of net shares sold by the CEO of the firm

during year t+1 divided by the number of common shares outstanding. To reduce the influence of extreme observations, all variables have been

winsorized at the top and bottom 1%.

Fig. 1. The figure provides evidence that sizeable differences exist in the expected rate of return (ERR) and discount rate (DR) estimates for defined benefit

pension plans. ERR and DR data comes from Compustat. Panel A: Expected rate of return assumption, pre- and post-SOX. Panel B: Discount rate

assumption, pre- and post-SOX.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133122

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 123

evidence of considerable variation in the assumed rates, with the 25th percentile (6.00%) 150 basis points lower than thatof the 75th percentile (7.50%). Panel B of Fig. 1 indicates that some of this variation in rates is attributable to employers’assumed DRs moving with prevailing interest rates, which were declining during the sample period. However, even withinthe same sample year, considerable variation in assumed rates is found to exist—the difference between the 25th and 75thpercentiles within each sample year averages 39 basis points.

Table 2 further indicates that 66% of the sample firms are underfunded, with an average funding level of 91%, and thatthe sample firms are relatively large, averaging total assets of $10.816 billion18 and pension obligations that amount to 16%of firms’ total assets. In addition, the sample firms are able to generate large amounts of cash from operating activities,with cash from operating activities averaging 8% of total assets, and are generally profitable, with only 18% of sample firmsreporting losses. Sample firms’ sales growth averages 9%. Looking at union status, we see that 12% of the sample firms havedefined benefit plans subject to collective-bargaining agreements, but we are only able to observe the union status for 41%of our sample firms. As further indicated in the table, 36% of the firm-year observations are in the post-SOX period. Theaverage actual return of plan assets of 5.79% is approximately 271 basis points below the average expected rate of returnassumption, and sample firms record income taxes that average 25% of the absolute value of operating income, reportpension income 17% of the time, and have pension plan assets that average 177% of the absolute value of operating income.Only 15% of the sample firms undertake an acquisition in year t+1. In addition, only a small number of the firms raiseequity or debt capital in year t+1, with an average ratio of new capital raised to total assets of only 3%. The sample firms’long-term debt averages 28% of total capital. Finally, although they are infrequent traders, CEOs’ average net selling activityin the open market averages less than 1% of shares outstanding.

In Fig. 2, we investigate whether, for the years before and after the passage of SOX, pension-freezing firms’ ERRs(Panel A) and DRs (Panel B) are unexpectedly lower in the years leading up to (and including) the freeze. The pre-SOXperiod is defined as fiscal years ending on or before 2002; the post-SOX period is defined as fiscal years ending on or after2003. We begin the post-SOX period with 2003 fiscal year-ends for two reasons. First, as pension estimates are set at thebeginning of each fiscal year, the passage of SOX in July 2002 first affects the selection of pension estimates during the2003 fiscal year. In addition, the warning issued by the SEC in December 2002 concerning greater scrutiny over pensionestimates and the effective date for SFAS 132 (R) first affects pension estimates during the 2003 fiscal year.

The approach we adopt for the analysis in Fig. 2 is similar to that used by Bergstresser et al. (2006), who regression-adjust ERR estimates to investigate the impact of mergers on the assumed rates. The unexpected rates presented inthe figure are derived from a simple regression with firm and year fixed-effects, and with four indicator variablesequal to one for freeze firm in the freeze year and each of the three years prior to the freeze year.19 Firm fixed-effects areincluded to control for possible firm-specific or industry-specific factors (e.g., the systematic use of conservativeaccounting estimates). Year fixed-effects are included to control for prevailing market conditions. To obtain pre- and post-SOX estimates, each of the four freeze year-related indicator variables are allowed to vary for both the pre- and post-SOXperiods.

The patterns presented in Fig. 2 suggest that, in the pre-SOX period, firms freezing their plans assume increasinglydownward biased ERRs and DRs in the years leading up to and including the freeze year. In addition, consistent with thepassage of SOX (and concurrent regulatory events) affecting managers’ ability to engage in accruals manipulation, thepatterns suggest that firms freezing their plans assume estimates that show significantly less downward bias in the post-SOX period.

Panel A of Fig. 2 reports the results for the unexpected ERR estimates. Specifically, Panel A shows that, in the pre-SOXperiod, the unexpected ERR estimates for freeze firms are �9.0, �24.4, �29.2, and �38 basis points in years �3, �2, �1,and 0, where year 0 is the year of the freeze. These unexpected ERR estimates are significantly less than zero for all yearsexcept �3 (t-statistics=�1.61, �4.48, �5.09, and �6.33, respectively).20 The negative unexpected rate in year 0 issizeable, as the inter-quartile range for the ERR assumption within sample years averages 115 basis points (see Fig. 1), andis consistent with magnitudes documented in prior related research—e.g., Bergstresser et al. (2006). In untabulatedanalyses we find that the unexpected ERR estimate for year 0 is significantly more negative than that for year �3(p-value=0.00), and the unexpected ERR estimates for the three years prior to the freeze year are jointly less than zero(p-value=0.00). Taken together, this evidence suggests that managers increasingly select downward biased ERR estimatesin the years leading up to the freeze year. After the passage of SOX, however, the unexpected ERR estimates becomeconsiderably less biased. For instance, while the (absolute) reduction in the unexpected ERR estimates is significant for allyears except year �3 (t-statistics=0.34, 2.66, 3.27, and 4.26), in the post-SOX period the unexpected ERR estimates areinsignificant (t-statistics=�0.48, 0.20, �0.24, and �0.52). In addition, in untabulated analyses, the unexpected ERR

18 The values reported for FirmSize in Table 2 are the natural log of total assets.19 Similar results obtain if only year fixed-effects are included.20 As these analyses, and later multivariate analyses, are essentially repeated for four different time periods—leading to four repeated analyses—the

reader may wish to alternatively consider Bonferroni corrected p-values. That is, an alternative cut-off value for significance of 0.0125 (i.e., 0.05/

4=0.0125), or t-statistics of 2.24 for one-sided tests, could be used. Our primary inferences are unaffected by this alternative cut-off value. We do not

tabulate these family-wise error rates as our primary focus is on year 0, rather than on years �1 through �3, and thus do not consider the tests as

independent multiple comparisons. In addition, note that the Bonferroni correction is very conservative—ignoring both the dependence across tests and

their joint significance (i.e., the probability of observing more than one test out of four being significant at the 5% level).

Expected rate of return assumption, pre- and post-SOX Discount rate assumption, pre- and post-SOX

Year relative to freeze -3 -2 -1 0 Year relative to freeze -3 -2 -1 0

Expected return, pre-SOX -0.090 -0.244 -0.292 -0.380 Discount rate, pre-SOX -0.092 -0.052 -0.258 -0.184

(t-statistic) (-1.61) (-4.48) (-5.09) (-6.33) (t-statistic) (-2.25) (-1.31) (-6.16) (-4.15)

Expected return, post-SOX -0.050 0.016 -0.015 -0.031 Discount rate, post-SOX 0.084 0.092 0.016 0.010

(t-statistic) (-0.48) (0.20) (-0.24) (-0.52) (t-statistic) (1.10) (1.50) (0.33) (0.23)

Difference 0.040 0.260 0.276 0.350 Difference 0.175 0.144 0.274 0.194

(t-statistic) (0.34) (2.66) (3.27) (4.26) (t-statistic) (2.05) (2.00) (4.40) (3.19)

Fig. 2. The figure provides evidence that expected rate of return (ERR) and discount rate (DR) estimates are downward biased at the time that firms hard freeze their defined benefit pension plans (time=0) in

years prior to the Sarbanes-Oxley Act (SOX). In addition, consistent with the passage of SOX (or concurrent regulatory events) affecting managers’ ability to engage in accruals manipulation, the figure provides

evidence that firms’ pension estimates are significantly less downward biased in the post-SOX period. The unexpected ERR and DR estimates in Panels A and B, respectively, use the entire sample of freeze and

non-freeze firms and are from a simple regression with firm and year fixed-effects. Firm fixed-effects are included to control for possible firm-specific or industry-specific factors (e.g., the systematic use of

conservative accounting estimates). Year fixed-effects are included to control for prevailing general market conditions. The unexpected ERR and DR estimates are the coefficients for four indicator variables

equal to one for the freeze firms for the freeze year and the three years prior to the freeze year, and zero otherwise, respectively, which are further interacted with pre- and post-SOX indicator variables. The

post-SOX period is defined as fiscal years ending on or after 2003. The reported t-statistics are from a test of whether the coefficients are significantly different from zero. The pre and post-SOX t-statistics test

whether the ERR and DR are downward biased in the years leading to and including the freeze year. The t-statistics of the difference test whether the ERR and DR are less downward biased following the passage

of SOX, consistent with SOX mitigating opportunistic behavior.

J.C

prix,

K.A

uller

III/

Jou

rna

ccou

ntin

Eco

mics

(20

–1

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 125

estimate for the freeze year is not significantly more negative than that for year �3 (p-value=0.43), and the unexpectedERR estimates for the three years prior to the freeze year are jointly not less than zero (p-value=0.39).

Panel B of Fig. 2 presents the results for the unexpected DR estimates. In the pre-SOX period the unexpected DR estimatesfor freeze firms are �9.2, �5.2, �25.8, and �18.4 basis points in years �3, �2, �1, and 0. These unexpected DR estimatesare significantly less than zero for all years except year �2 (t-statistics=�2.25, �1.31, �6.16, and �4.15). The negativeunexpected rate for year 0 again appears sizeable, as the inter-quartile range for the DR assumption within sample yearsaverages 39 basis points (see Fig. 1). In addition, in untabulated analyses the unexpected DR estimate for the freeze year issignificantly more negative than that for year �3 (p-value=0.04), and the unexpected DR estimates for the three years prior tothe freeze year are jointly less than zero (p-value=0.00). This evidence suggests that managers select increasingly downwardbiased DR estimates in the years leading up to the freeze year. As in Panel A, however, after the passage of SOX the unexpectedDR estimates become considerably less biased. For instance, while the (absolute) reduction in the unexpected DR estimates issignificant for all years (t-statistics=2.05, 2.00, 4.40, and 3.19), in the post-SOX period the unexpected DR estimates areinsignificant (t-statistics=1.10, 1.50, 0.33, and 0.23). Moreover, in untabulated analyses, the unexpected DR estimate for year 0is not significantly more negative than that for year �3 (p-value=0.19), and the unexpected DR estimates for the three yearsprior to the freeze year are jointly not less than zero (p-value=0.93).

Notwithstanding the above evidence, the possibility exists that a downward biased ERR estimate could be driven not bymanagerial opportunism, but instead by actual rates of return (ARR) on plan assets being relatively lower in the yearsleading up to and including the freeze decision (and even after the freeze decision).21 To examine the possible influence ofnegative unexpected ARRs, we follow the same regression-adjusted approach as in Fig. 2, but with the addition of two newindicator variables that allow us to examine the impact of freezes on future unexpected ARRs. These indicators are equal toone for freeze firms in the two years following the freeze year (i.e., +1 and +2), and zero otherwise, respectively. Inuntabulated analysis, we fail to find evidence that unexpected ARRs are negative in the years leading up to and includingthe freeze decision, or after the freeze decision. However, we find some evidence that unexpected ARRs are positive in theyear following the freeze decision. Specifically, the unexpected ARRs are 66.5, �37.5, 21.2, 9.8, 200.8, and 80.5 basis pointsin years �3, �2, �1, 0, +1, and +2, which are not significantly different from zero with the exception of year +1 (t-statistics=1.04, �0.63, 0.38, 0.18, 3.05, and 1.12, respectively). In addition, the unexpected ARR for the freeze year is notsignificantly lower than that for year �3 (p-value=0.21), the unexpected ARRs for the three years prior to the freeze yearare jointly not less than zero (p-value=0.35), and the unexpected ARRs for the two years following the freeze year arejointly greater than zero (p-value=0.01). Thus, it appears that leading, concurrent, and future unexpected ARRs do notexplain freeze firms’ negative unexpected ERRs.

In the next section, we more systematically examine firms’ choice to freeze their plans and whether managers use theirdiscretion in selecting pension assumptions to make their firms’ plans appear more burdensome.

4. Empirical analyses and results

In this section we begin by providing more extensive descriptive evidence on the determinants of employers’ hardfreeze decisions. The analysis is intended to provide contextual information regarding the environment in which thepension assumptions are set. In Sections 4.2 and 4.3, we examine for the pre- and post-SOX periods whether employerswith defined benefit plans select relatively lower ERR and DR estimates in the year they freeze their plans while controllingfor other factors shown in prior research to impact the selection of pension assumptions.

4.1. The determinants of hard freezes of defined benefit pension plans

In this sub-section we investigate the determinants of firms’ decisions to hard freeze their defined benefit pensionplans. We employ the following probit regression model:

Freeze0 ¼X2008

i ¼ 1991

d0,tþd1 Underfundedþd2 Funding%þd3 FirmSizeþd4 PlanSizeþd5 OperatingCashFlow

þd6Lossþd7 DSales%þd8 UnionPlanþd9 UnionKnownþB ð1Þ

where Freezej is the indicator variable equal to 1 if the firm’s defined benefit pension plans are hard frozen (subscript j

indicates year relative to freeze decision), and 0 otherwise; Underfunded the indicator variable equal to 1 if the firm’sdefined benefit pension plans are underfunded, and 0 otherwise; Funding% the pension plan assets divided by the projectedbenefit obligation; FirmSize the natural log of total assets measured in millions of dollars; PlanSize the projected benefit

21 Note, however, that while ARRs are important in determining the expected rate of return assumption, they generally do not closely correspond to

the ERR estimate. As discussed earlier, ASC 715 requires only that ‘‘appropriate consideration should be given to the returns being earned by the plan

assets in the fund and the rates of return expected to be available for reinvestment’’ (formerly paragraph 45 of SFAS 87). Over what time horizon historical

and expected future returns are to be evaluated is not specified. Consistent with this limited connection, prior research finds little association between

ERR and ARR, even over relatively long time periods. For instance, Amir and Benartzi (1998) find no evidence of an association between ERR and future

ARRs over one- to five-year horizons. In addition, Comprix and Muller (2006) find correlations of 0.04, 0.03, 0.03, 0.05, and 0.08, respectively, when ERR

and ARR are averaged over periods of one to five years.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133126

obligation divided by total assets; OperatingCashFlow the cash flow from operations divided by total assets; Loss theindicator variable equal to 1 if the firm reported a loss for the year, and 0 otherwise; DSales% the percentage change insales; UnionPlan the indicator variable equal to 1 if any of the firm’s defined benefit pension plans are subject to acollective-bargaining agreement, and 0 otherwise; and UnionKnown the indicator variable equal to 1 if the status ofwhether a firm’s defined benefit pension plans are subject to a collective-bargaining agreement is known, and 0 otherwise.

Year fixed-effects, d0,t , are included in Eq. (1) to capture changing market conditions—e.g., falling interest rates—in theyears leading up to employers’ decisions to freeze their defined benefit plans. Variables are also included to capture thehealth of employers’ defined benefit pension plans, the size of employers’ plans, employers’ financial condition, andwhether employers’ plans are subject to collective-bargaining agreements.

We posit that employers are more likely to freeze plans that are more costly or ‘‘sick,’’ as in this case employees are lesslikely to protest any potential loss in benefits, and thus we expect a positive association between Freeze0 and Underfunded

and a negative association between Freeze0 and Funding% (we include Underfunded in addition to the level of pensionfunding (Funding%), as employers may face relatively lower employee resistance to freezing plans classified as‘‘underfunded’’). Relatively large employers and those with relatively large defined benefit plans likely face greaterconstraints in freezing their plans given the larger number of affected employees who may resist the freeze, and givengreater potential negative publicity of freezing a relatively large plan, and hence we expect a negative association betweenFreeze0 and the variables FirmSize and PlanSize. Similarly, employers with relatively less ability to generate resourcesinternally may face constraints when funding the accrual of future pension benefits, which leads to a negative predictionon the association between Freeze0 and OperatingCashFlow.22 Employers facing losses and declining sales are likely betterable to argue that they are no longer able to maintain their defined benefit plans. We therefore expect a positiveassociation between Freeze0 and Loss, and a negative association between Freeze0 and DSales%. For defined benefit pensionplans subject to a collective-bargaining agreement, all affected employee groups must agree to the freeze. This leads us topredict a negative association between Freeze0 and UnionPlan. Finally, as we discuss earlier, information regarding whetheremployers’ defined benefit plans are subject to collective-bargaining agreements is not always available from Form 5500filings. To control for the unavailability of data we include the indicator variable UnionKnown, but we do not make aprediction regarding its association with Freeze0.23

Table 3 presents the results of estimating Eq. (1) using standard errors clustered by firm (Petersen, 2009). To easeinterpretation, all coefficients have been multiplied by 100. The coefficients for Underfunded and Funding% are insignificant(t-statistics=1.38 and 0.44, respectively). We thus fail to find evidence that employers increasingly freeze plans classifiedas underfunded or plans with lower funding levels. The coefficients for FirmSize and PlanSize are significantly negative (t-statistics=�9.33 and �4.64, respectively), consistent with large employers and employers with large defined benefit plansbeing less able to freeze their plans. The coefficient for OperatingCashFlow is significantly negative (t-statistic=�2.78),consistent with defined benefit plans being frozen when employers are less able to generate resources internally, and thecoefficient for Loss is significantly positive (t-statistic=3.87), which implies that employers facing losses are more likely tofreeze their plans. The coefficient for DSales% is insignificant (t-statistic=1.26). The coefficient for UnionPlan is significantlynegative (t-statistic=�2.14), in line with employers being less able to freeze plans subject to collective-bargainingagreements, and the coefficient for UnionKnown is insignificant (t-statistic=0.33).

In summary, we find that the number of employees affected by the freeze decision and the presence of collective-bargaining agreements are important constraints on freeze decisions, whereas employers facing greater financial distressare more likely to freeze their plans. In the next sub-sections, we investigate whether managers exaggerate the apparenteconomic burden of their pension plans by selecting downward biased ERR and DR assumptions.

4.2. The association between assumed expected rates of return and hard freezes of pension plans

We use the following year and firm fixed-effects regression to investigate, for both the pre- and post-SOX periods,whether employers assume lower ERRs in the freeze year and the three years prior to the freeze year:

ERR¼X2008

t ¼ 1991

l0,tþX2,616

i ¼ 2

l0,iþ½l1 Freeze0þl2 Freeze�1þl3 Freeze�2þl4 Freeze�3�*½PreSOXþPostSOX�

þl5 ARRþl6 TaxRateþl7 FirmSizeþl8 Lossþl9DSales%þl10 Funding%þl11 Funding%2þl12 PensionIncome

þl13 PensionSensitivityþl14 Acquisitionþ1þl15 EquityIssueþ1þl16 DebtIssueþ1þl17 Leverage

þl18 CEO_NetSellingþ1þc ð2Þ

22 Note, however, that the inclusion of OperatingCashFlow may inappropriately reduce the importance of Underfunded and Funding% in Eq. (1) as firms’

ability to generate resources internally likely affects the funding status of the their defined benefit pension plans. In untabulated tests, we find that our

inferences are unchanged if we exclude OperatingCashFlow from Eq. (1).23 In untabulated analysis, we alternatively estimate Eq. (1) using only those observations not missing information regarding whether employers’

defined benefit pension plans are subject to collective-bargaining agreements. Our inferences are similar with the exception of Funding% and

OperatingCashFlow, which become significant and insignificant, respectively. We do not use this sub-sample for our primary analysis because we would

then retain only 41% of our firm-year observations.

Table 3Probit regression of employers’ decisions to hard freeze their defined benefit pension plans (Freeze).

Freeze0 ¼X2008

i ¼ 1991

d0,tþd1 Underfundedþd2 Funding%þd3 FirmSizeþd4 PlanSizeþd5 OperatingCashFlowþd6 Loss

þd7 DSales%þd8 UnionPlanþd9 UnionKnownþB ð1Þ

Explanatory variables Coefficient(Exp. sign) (t-statistic)

Underfunded (+) 8.77(1.38)

Funding% (�) 4.59(0.44)

FirmSize (�) �14.86nnn

(�9.33)PlanSize (�) �117.81nnn

(�4.64)OperatingCashFlow (�) �105.26nnn

(�2.78)Loss (+) 26.38nnn

(3.87)DSales% (�) 13.96

(1.26)UnionPlan (�) �25.00nn

(�2.14)UnionKnown (?) 2.34

(0.33)Pseudo-R2 10.33%N 17,560

The table presents the results of a probit model that regresses the freeze decision on its potential determinants. Freeze is an indicator variable equal to 1 if the

firm’s defined benefit pension plans are hard frozen, and 0 otherwise. Underfunded is an indicator variable equal to 1 if the firm’s defined benefit pension plans

are underfunded, and 0 otherwise. Funding% is pension plan assets divided by the projected benefit obligation. FirmSize is the natural log of total assets

measured in millions of dollars. PlanSize is the projected benefit obligation divided by total assets. OperatingCashFlow is the cash flow from operations divided

by total assets. Loss is an indicator variable equal to 1 if the firm reported a loss for the year, and 0 otherwise. DSales% is the percentage change in sales.

UnionPlan is an indicator variable equal to 1 if any of the firm’s defined benefit pension plans are subject to a collective-bargaining agreement, and 0 otherwise.

Finally, UnionKnown is an indicator variable equal to 1 if the status of whether a firm’s defined benefit pension plans are subject to a collective-bargaining

agreement is known, and 0 otherwise. To ease interpretation, all coefficients have been multiplied by 100. Year fixed-effects estimates are not reported.n, nn, nnn denote significance at the 0.05, 0.01, and 0.001 (one-tail) levels, respectively. Standard errors clustered by firm are computed using the approach

discussed by Petersen (2009).

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 127

where ERR is the assumed expected rate of return on the defined benefit pension plan assets; PreSOX the indicator variableequal to 1 if the fiscal year occurs prior to the passage of the Sarbanes-Oxley Act of 2002, and 0 otherwise; PostSOX the indicatorvariable equal to 1 if the fiscal year occurs following the passage of the Sarbanes-Oxley Act of 2002, and 0 otherwise; ARR theactual rate of return on the defined benefit pension plan assets; TaxRate the income taxes divided by the absolute value of pre-tax income; PensionIncome the indicator variable equal to 1 if the firm reports pension income (i.e., negative pension costs), and0 otherwise; PensionSensitivity the defined benefit pension plan assets divided by the absolute value of operating income;24

Acquisition+1 the indicator variable equal to 1 if the firm undertakes an acquisition during year t+1, and 0 otherwise;EquityIssue+1 the amount of public and private common and preferred equity raised during year t+1 divided by total assets;DebtIssue+1 the amount of public and private common and preferred debt raised during year t+1 divided by total assets;Leverage the long-term debt divided by total capital (i.e., market value of equity plus long-term debt); and CEO_NetSelling+1 thenumber of net shares sold by the CEO of the firm during year t+1 divided by the number of common shares outstanding.

All other variables are as previously defined. As before, because employers’ choice of ERRs should reflect the historicaland expected future returns on pension plan assets, we control for prevailing market conditions by including yearfixed-effects, l0,t . We further control for the possibility that unspecified firm-specific or industry-specific factors (e.g., thesystematic use of conservative accounting estimates) could influence our analyses by including firm fixed-effects, l0,i.

24 For parsimony, we do not include interactions between PensionSensitivity and our control variables, as is done by Bergstresser et al. (2006). In untabulated

analyses of the ERR and DR assumptions, our inferences remain unchanged when such interaction terms are included, which are generally insignificant.25 As the results in Eq. (1) suggest, the firms that freeze their plans are different from those that do not freeze their plans. We control for such firm-

specific differences in Eqs. (2) and (3) using firm fixed-effects. In addition, as discussed earlier, we distinguish between whether the selection of

downward biased pension estimates reflects opportunistic behavior or increased managerial pessimism (and other possible types of endogeneity) using

the relatively exogenous shock of the passage of SOX. It is also possible that freeze decisions could be driven by managers’ pessimism regarding long-term

ERR, in which case we could treat ERR as an (endogenous) explanatory variable in Eq. (1). We do not adopt this approach, however, as the reduced forms

would not be uniquely defined—i.e., the ERR differential across freeze and non-freeze firms would not capture managers’ benefits from freezing their

plans. Accordingly, following the suggestions of Maddala (1991) for such settings, we do not to include ERR in Eq. (1), but instead note that the

simultaneity can be accounted for indirectly through the specification of the choice model.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133128

Our primary prediction is that employers with defined benefit plans select relatively more downward biased ERR

estimates in the year of the decision to freeze their plans, as managers face incentives to make their plans look as costly aspossible in the freeze year. In addition, given that changes in accounting estimates are closely examined by auditors, andthus managers face similar incentives to make their plans look relatively more costly in the years prior to the freeze year,we expect employers to begin assuming lower ERRs in the years leading up to the freeze year. Accordingly, similar to ouranalysis in Fig. 2, we also investigate the three years prior to the freeze year. We expect negative associations between ERR

and Freezej, consistent with managers biasing their ERR estimates downward. To obtain pre- and post-SOX estimates, eachof the four indicator variables representing the years relative to the freeze year are allowed to vary for the pre- and post-SOX periods.26 To the extent that the passage of SOX, or concurrent regulatory events, reduced managers’ ability to engagein accruals-based earnings management, we expect firms’ pension estimates to show significantly less downward bias inthe post-SOX period.

Eq. (2) includes a number of control variables to capture the economic determinants of the ERR assumption. Thevariable ARR is included to capture the possibility that actual rates determine the assumed expected rate.27 Strongeraverage performance by pension plan assets should lead to a higher assumed ERR. We therefore expect a positiveassociation between ERR and ARR. We include the variable TaxRate to capture employers’ tax rates. Because the actualreturns on pension plan assets go untaxed, employers with relatively higher tax rates should face greater incentives toinvest in bonds—the least tax-favored securities during our sample period (Black, 1980; Tepper, 1981). This incentivecould lead to employers with relatively higher tax rates investing in assets with generally lower expected rates of return.Amir et al. (2009) find evidence consistent with this prediction. Accordingly, we expect a negative association between ERR

and TaxRate. We include the variables FirmSize, Loss, and DSales% to capture firm risk. Smaller firms, unprofitable firms, andfirms facing declining sales typically are more risky. Because employers facing higher corporate risk tend to make less riskyasset allocation decisions (Friedman, 1983; Bodie et al., 1984; Amir et al., 2009), investing a greater proportion of pensionplan assets in bonds, such a strategy generally results in lower returns. We therefore expect a positive association betweenERR and FirmSize, and a negative association between ERR and Loss and DSales%. We include the variables Funding andFunding2 to capture the relationship between funding levels and asset allocations.28 Consistent with the predictions ofBader (1991), Amir and Benartzi (1998) and Amir et al. (2009) find evidence that employers of extremely overfunded andextremely underfunded plans invest relatively less in equities, with only moderately funded plans investing more inequities. We therefore expect a positive association between ERR and Funding and a negative association between ERR andFunding2, consistent with an inverted-U relation.29

We also include a number of control variables to capture managers’ incentives to opportunistically select ERR

assumptions. We include the variable PensionIncome to capture managers’ incentives to assume relatively higher ERR

estimates when firms report pension income. Comprix and Muller (2006) provide evidence that compensation committeesasymmetrically treat pension income and pension expenses in CEO cash compensation contracts––weighting pensionincome more heavily and providing managers with an incentive to opportunistically assume higher ERRs when pensionincome is reported. We therefore expect a positive association between ERR and PensionIncome. We include the variablesPensionSensitivity, Acqusition+1, EquityIssue+1, DebtIssue+ 1, and Leverage to capture managers’ incentives to assumerelatively higher ERR estimates when firms’ income is more sensitive to pension amounts, when firms plan to undertakeacquisitions and raise equity and debt, and when leverage is high. Bergstresser et al. (2006) provide evidence thatmanagers opportunistically assume higher ERRs––which increase reported income––when the sensitivity of income topension cost is high and prior to acquisitions or CEO option exercises. They also find mixed evidence of similar behaviorprior to raising equity capital. In contrast, Comprix and Muller (2006) find no evidence of managers assuming higher ratesprior to raising debt or equity capital, but do find evidence of managers assuming higher rates when leverage is high orprior to the firm’s CEO selling shares in the open market. Overall, we expect a positive association between ERR and thevariables PensionSensitivity, Acqusition+1, EquityIssue+ 1, DebtIssue+ 1, Leverage, and CEO_NetSelling+ 1.30

26 The indicator variable PostSOX is not included in Eq. (2) as a separate explanatory variable due to the inclusion of year fixed-effects in the model,

which fully accounts for shifts in means across sample years.27 Alternatively, the effect of ARRs on the assumed ERR could be modeled using the difference between the ERR and ARR as the dependent variable, as

in Comprix and Muller (2006). In this context, due to the one-time nature of hard freeze decisions and their relative infrequency, deviations between

expected and actual rates would be extremely noisy given that most of the variation would arise from changing ARRs. Thus, unlike the pension income

versus pension expense setting examined by Comprix and Muller (2006), this alternative specification would make it difficult to capture systematic

differences between firms’ long-term ERR and the long-term ARR. Our choice of examining the determinants of the expected rate using the actual rate as

an explanatory variable is consistent with prior related research (e.g., Amir and Benartzi, 1998).28 Prior research by Amir and Benartzi (1998) and Amir et al. (2009) defines funding status using the accumulated benefit obligation (ABO) rather

than the projected benefit obligation (PBO). We use PBO to be consistent with funding status in Eq. (1). We also use PBO as ABO is commonly missing on

Compustat—i.e., the use of ABO would result in the loss of nearly 4000 observations. In any event, our inferences remain unchanged when ABO is used

instead.29 Amir et al. (2009) also find evidence that firms allocate more of their pension plan assets to equities when the horizon of the pension obligation is

longer. We do not include an additional variable for retirement horizon due to the horizon of employees being relatively fixed at the firm level.30 Bergstresser et al. (2006) find that managers opportunistically assume higher (lower) ERRs prior to the firm’s CEO exercising (being granted)

options. We do not include variables for these incentives due to the restrictive data requirements (i.e., option data from ExecuComp are only available for

S&P 1500 firms).

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 129

Table 4 presents the results of estimating Eq. (2). The results are generally in line with the univariate results presentedin Fig. 2 above. Although not tabulated, the year and firm fixed-effects explain a significant portion of employers’ assumedERRs (with only the year and firm fixed-effects, the adjusted R2 is 23.05%). Turning to our results for the pre-SOX period, wefind that the coefficient for Freeze0 is significantly negative at �44.62 (t-statistic=�6.83), indicating that the unexpectedERR estimate for freeze firms is �45 basis points in the freeze year.31 In addition, the coefficients for Freeze�1 and Freeze�2

are significantly negative at �30.26 and �23.34 (t-statistics=�4.84 and �4.01, respectively), indicating that the negativeunexpected ERR estimates begin in the two years prior to the freezing of the pension plans. Further analyses (nottabulated) indicate that the coefficient for the unexpected ERR estimate during the freeze year is more negative than thosefor the years �3, �2, and �1 (p-values=0.00, 0.00, and 0.02), and that the coefficient for the unexpected ERR estimates forthe three years prior to year 0 are jointly less than zero (p-value=0.00). Taken together, this evidence suggests that, inthe pre-SOX period, managers select increasingly downward biased ERRs in the years leading up to and including thefreeze year.

After the passage of SOX, however, the unexpected ERR estimates become considerably less biased. The (absolute)reduction in the coefficients for the Freeze variables is significant for all years except year �3 (t-statistics=0.79, 2.44, 3.60,and 4.79). Moreover, the coefficients for the Freeze variables are insignificant in the post-SOX period (t-statistics=0.11,0.23, 0.32, and �0.33). Further analyses (not tabulated) indicate that the coefficient for the unexpected ERR estimate is notsignificantly more negative during year 0 than those for years �3, �2, and �1 (p-values=0.39, 0.34, and 0.29), and thecoefficient for the unexpected ERR estimates for the three years prior to year 0 are not jointly less than zero (p-value=0.38).

The variables capturing the economic determinants of the ERR assumption are generally significant. The coefficient forFirmSize is significantly positive (t-statistic=3.83), consistent with firms facing greater risk assuming relatively lower ERR

(by allocating pension plan assets to less risky assets). The coefficients for Funding and Funding2 are significantly positiveand negative, respectively (t-statistics=2.20 and �3.49), consistent with asset allocations to equities, and thus assumedERR estimates, being relatively lower for extremely underfunded or overfunded plans but being higher for moderatelyfunded plans. The coefficients for the variables ARR, TaxRate, Loss, and DSales%, however, are insignificant.

A number of the variables capturing managers’ incentives to opportunistically select ERR assumptions are alsosignificant. The coefficient for PensionIncome is significantly positive (t-statistic=8.13), consistent with managers assumingrelatively higher ERRs when CEO cash compensation formulas are relatively more sensitive to reported pension costs.Likewise, the coefficient for PensionSensitivity is significantly positive (t-statistic=2.01), consistent with managersassuming relatively higher ERRs when income is relatively more sensitive to pension amounts, and the coefficient forDebtIssue+ 1 is significantly positive (t-statistic=1.95), consistent with firms that are planning to raise debt assumingrelatively higher ERRs. In contrast, the coefficients for the variables Acqusition+ 1, EquityIssue+ 1, Leverage, andCEO_NetSelling+ 1 are insignificant.

In summary, we find evidence of employers assuming unexpectedly lower ERRs in the year of a pension freeze and inthe two years prior to the freeze. In the next sub-section, we investigate whether managers also similarly assumeunexpectedly lower DRs.

4.3. The association between assumed discount rates and hard freezes of pension plan

Paralleling our investigation of assumed ERRs, we use the following year and firm fixed-effects regression to investigate,for the pre- and post-SOX periods, whether employers assume unexpectedly lower DRs in the freeze year and the yearsprior to the freeze year:

DR¼X2008

t ¼ 1991

b0,tþX2,616

i ¼ 2

b0,iþ½b1 Freeze0þb2 Freeze�1þb3 Freeze�2þb4 Freeze�3�*½PreSOXþPostSOX�

þb5 PensionIncomeþb6 PensionSensitivityþb7 Acquisitionþ1þb8 EquityIssueþ1þb9 DebtIssueþ1þb10 Leverage

þb11 CEO_NetSellingþ1þb12 Underfundedþb13ARRþB ð3Þ

where DR is the assumed discount rate for the projected benefit plan.All other variables are as previously defined. As we discuss earlier, employers’ choice of DRs should reflect the rate at

which the pension obligation could be settled or the rate of return on high-quality fixed income securities that are similarin duration to the pension liability. As the assumed DR should be affected largely by prevailing interest rates, we againcontrol for annual changes in interest rates by including year fixed-effects, b0,t . We also continue to control for the

31 Managers may face greater incentives to bias ERR estimates downwards when the boost to net income from freezing plans is larger. As the boost to

net income comes from future service costs becoming zero in future years, in sensitivity analysis we investigate this possibility by interacting our Freeze

variables with the indicator variable LargeServiceCost, which is equal to one when the lagged ratio of service costs to the absolute value of net income is

above the median value, and zero otherwise. We use the lagged ratio as service costs in the freeze year are typically affected by the freeze decision. For

the ERR analysis, consistent with a greater incentive effect when service costs are high, we find evidence in the pre-SOX period of greater downward bias

in the freeze year and the year before the freeze year (Freeze0 and Freeze�1 coefficients=�61.85 and �83.77; t-statistics=�2.35 and �2.41, respectively).

We fail to find similar evidence for the post-SOX period. For the DR analysis (discussed below), we fail to find evidence of a similar increase in the

downward bias in either the pre- or post-SOX periods.

Table 4Year and firm fixed-effects regression of employers’ choice of expected rate of return (ERR) estimate during the pre- and post-SOX time periods.

ERR¼X2008

t ¼ 1991

l0,tþX2,616

i ¼ 2

l0,iþ½l1 Freeze0þl2 Freeze�1þl3 Freeze�2þl4 Freeze�3�*½PreSOXþPostSOX�

þl5 ARRþl6 TaxRateþl7 FirmSizeþl8 Lossþl9 DSales%þl10 Funding%þl11 Funding%2þl12 PensionIncome

þl13 PensionSensitivityþl14 Acquisitionþ1þl15 EquityIssueþ1þl16 DebtIssueþ1

þl17 Leverageþl18 CEO_NetSellingþ1þc ð2Þ

(Exp. sign) Pre-SOX Post-SOX Difference

Coefficient Coefficient

(t-statistic) (t-statistic) (t-statistic)

Variables of interest

Freeze0 (�) �44.62nnn�2.09 42.53nnn

(�6.83) (�0.33) (4.79)

Freeze�1 (�) �30.26nnn 2.17 32.43nnn

(�4.84) (0.32) (3.60)

Freeze�2 (�) �23.34nnn 2.01 25.35nnn

(�4.01) (0.23) (2.44)

Freeze-3 (�) �8.68 1.16 9.84

(�1.46) (0.11) (0.79)

Control variables

ARR (+) 0.04

(0.66)

TaxRate (�) 0.71

(0.55)

FirmSize (+) 3.99nnn

(3.83)

Loss (�) 0.41

(0.30)

DSales% (�) �0.94

(�0.49)

Funding (+) 22.84nn

(2.20)

Funding2 (�) �15.70nnn

(�3.49)

PensionIncome (+) 12.24nnn

(8.13)

PensionSensitivity (+) 0.33nn

(2.01)

Acquisition+ 1 (+) �5.25

(�3.62)

EquityIssue+ 1 (+) �9.83

(�0.97)

DebtIssue+1 (+) 16.90n

(1.95)

Leverage (+) 3.92

(1.26)

CEO_NetSelling+ 1 (+) �13.29

(�0.22)

Adjusted R2 23.17%

N 17,560

The table presents the results of a year and firm fixed-effects regression that investigates whether employers opportunistically assume downward biased

expected rates of return in the years leading up to the hard freezing of their pension plans, and whether managers’ ability to opportunistically bias

assumed rates is mitigated following the passage of the Sarbanes-Oxley Act. Hard frozen defined benefit pension plans are plans where employers have

eliminated the accrual of additional pension benefits for all employees. ERR is the assumed expected rate of return on the defined benefit pension plan

assets. Freeze is an indicator variable equal to 1 if the firm’s defined benefit pension plans are hard frozen (subscripts indicate year relative to freeze

decision), and 0 otherwise. PreSOX is an indicator variable equal to 1 if the fiscal year occurs prior to the passage of the Sarbanes-Oxley Act of 2002, and 0

otherwise. PostSOX is an indicator variable equal to 1 if the fiscal year occurs following the passage of the Sarbanes-Oxley Act of 2002, and 0 otherwise.

ARR is the actual rate of return on the defined benefit pension plan assets. TaxRate is income taxes divided by the absolute value of pre-tax income.

FirmSize is the natural log of total assets measured in millions of dollars. Loss is an indicator variable equal to 1 if the firm reported a loss for the year, and

0 otherwise. DSales% is the percentage change in sales. Funding is defined benefit pension plan assets divided by the projected benefit obligation.

PensionIncome is an indicator variable equal to 1 if the firm reports pension income (i.e., negative pension cost), and 0 otherwise. PensionSensitivity is

defined benefit pension plan assets divided by the absolute value of operating income. Acquisition+ 1 is an indicator variable equal to 1 if the firm

undertakes an acquisition during year t+1, and 0 otherwise. EquityIssue+ 1is the amount of public and private common and preferred equity raised during

year t+1 divided by total assets. DebtIssue+ 1is the amount of public and private common and preferred debt raised during year t+1 divided by total assets.

Leverage is long-term debt divided by total capital (i.e., market value of equity plus long-term debt). Finally CEO_NetSelling+ 1 is the number of net shares

sold by the CEO of the firm during year t+1 divided by the number of common shares outstanding. All coefficients have been multiplied by 100 for ease of

interpretation. Year and firm fixed-effect estimates are not reported.n, nn, nnn denote significance at the 0.05, 0.01, and 0.001 (one-tail) levels, respectively.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133130

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 131

possibility that unspecified firm-specific or industry-specific factors could influence our analyses by including firm fixed-effects, b0,i.

In comparison to Eqs. (2) and (3) excludes the variables TaxRate, FirmSize, Loss, DSales%, Funding, and Funding2, as thesevariables capture economic determinants of the ERR assumption. However, we continue to include the control variablesfrom Eq. (2) that capture managers’ incentives to act opportunistically. Our predictions for managers’ opportunisticselection of DRs are the same as our predictions for the opportunistic selection of ERRs. We also include an additionalincentive variable, Underfunded, to capture managers’ incentive to mask the extent to which plans are underfunded byassuming higher discount rates. Finally, we include the variable ARR in Eq. (3), but for a different reason than in Eq. (2).Here, ARR captures managers’ incentive to assume higher DRs to mask declines in their reported funding level arising fromdeclining asset returns. We therefore expect positive and negative associations between DR and the variables Underfunded

and ARR, respectively.Table 5 presents the results of estimating Eq. (3). The results are again similar to the univariate results presented in

Fig. 2 above. As before, to ease interpretation all coefficients have been multiplied by 100. Also as before, although nottabulated, the year and firm fixed-effects explain a significant portion of employers’ assumed DRs (with only the year andfirm fixed-effects, the adjusted R2 is equal to 79.19%). Turning to our results on DR, we find that in the pre-SOX period, thecoefficient for Freeze0 of �17.81 is significantly negative (t-statistic=�3.73), indicating that the unexpected DR estimatefor freeze firms is -18 basis points in the freeze year. In addition, the coefficients for Freeze�1, Freeze�2, and Freeze�3 of�32.60, �9.63, and �11.80 are significantly negative (t-statistics=�7.13, �2.26, and �2.71, respectively), indicating thatnegative unexpected DRs begin in the three years prior to the freeze year. Further analyses (not tabulated) indicate that thecoefficient for the unexpected DR estimate during the freeze year is not significantly more negative than those for the years�3, �2, and �1 (p-values=0.13, 0.06, and 0.99), and that the coefficient for the unexpected DR estimates for the threeyears prior to the freeze year are jointly less than zero (p-value=0.00). Taken together, this evidence suggests that, in thepre-SOX period, managers select increasingly downward biased DRs in the years leading up to the freeze year.

After the passage of SOX, however, the unexpected DR estimates become considerably less biased. The (absolute)reduction in the coefficients for the Freeze variables is significant for all years (t-statistics=2.79, 2.50, 5.04, and 2.84). Thecoefficients for the Freeze variables are insignificant in the post-SOX period (t-statistics=1.69, 1.46, 0.13, and 0.14). Furtheranalyses (not tabulated) indicate that the coefficient for the unexpected DR estimate during year 0 is not significantly morenegative than those for years �3, �2, and �1 (p-values=0.07, 0.11, and 0.50), and the coefficient for the unexpected DR

estimates for the three years prior to year 0 are not jointly less than zero (p-value=0.96).Interestingly, the combined effect of assuming both relatively lower ERRs and relatively lower DRs has a sizeable impact

on reported pension costs. Using the negative unexpected ERR and DR estimates documented for our freeze firms, wecalculate that net periodic pension costs would increase on average by 40% for a typical firm.32 The effect of assumingrelatively lower DRs also has a sizable impact on reported funding levels. Using the negative unexpected DR estimate, wecalculate that reported funding levels would decrease on average by 22% for a typical firm.33

In Table 5 the other incentive variables are generally found to be insignificant. However, the coefficients forPensionIncome and CEO_NetSelling+ 1 are significantly positive (t-statistics=2.21 and 2.26, respectively), and the coefficientfor ARR is significantly negative (t-statistic=�3.06). This evidence is consistent with managers choosing higher DRs whenreporting pension income, prior to CEOs selling their shares, and when the pension plan assets are performing poorly.

5. Conclusion

In this study we examine whether employers assume downward biased expected rates of return and discount rates tomake their plans appear more costly in the year they decide to ‘‘hard’’ freeze their defined benefit pension plans. Weconduct our investigation separately for the years before and after the passage of the Sarbanes-Oxley Act (SOX)—whichincreased regulatory scrutiny and penalties for earnings management—to distinguish between opportunistic behavior andother factors associated with firms freezing their pension plans. Specifically, if the behavior is opportunistic, the post-SOXperiod should be characterized by less biased pension estimates. In addition, we examine the determinants of andconstraints on employers’ decisions to hard freeze their defined benefit plans. Employers hard freeze their defined benefitplans by eliminating the accrual of additional future service benefits for all employees. Our examination of the pensionassumptions during the hard freezing of pension plans is motivated by critics’ claims that employers face incentives tomake their defined benefit plans look overly burdensome so as to obtain concessions from labor.

32 To determine the impact of the average change in net periodic pension costs, we conducted the following analysis. For the ERR assumption, we

directly calculated the effect on pension costs (in absolute value) in the year prior to the freeze (due to it being unaffected by the freeze decision) by

multiplying the beginning pension plan assets by 45 basis points. For the DR assumption, the relative increase in reported pension costs in the year prior

to the freeze was approximated using firms’ sensitivity analysis disclosures. We randomly chose 100 firms that voluntarily disclosed the impact of

changing the DR assumption on their pension costs. We then averaged the percentage change on reported pension costs for the 100 firms for an 18 basis

point decline in the DR.33 The lower reported funding levels attributable to the relatively lower assumed DR was computed in a similar manner as the impact of the lower DR

on reported pension costs. We averaged the percentage change in the PBO for the sampled firms for an 18 basis point decline in the DR. We then

determined the impact of the percentage change in the PBO on reported funding levels (in absolute value) for the freeze firms. This calculation is based on

a sub-sample of the 100 firms, due to only 32 of the 100 sampled firms reporting the impact of a change in the DR on their reported PBO.

Table 5Year and firm fixed-effects regression of employers’ choice of discount rate (DR) estimate during the pre- and post-SOX time periods.

DR¼X2008

t ¼ 1991

b0,tþX2,616

i ¼ 2

b0,iþ½b1 Freeze0þb2 Freeze�1þb3 Freeze�2þb4 Freeze�3�*½PreSOXþPostSOX�

þb5 PensionIncomeþb6 PensionSensitivityþb7 Acquisitionþ1þb8 EquityIssueþ1þb9 DebtIssueþ1þb10 Leverage

þb11 CEO_NetSellingþ1þb12 Underfundedþb13 ARRþB ð3Þ

(Exp. sign) Pre-SOX Post-SOX Difference

Coefficient Coefficient

(t-statistic) (t-statistic) (t-statistic)

Variables of interest

Freeze0 (�) �17.81nnn 0.66 18.47nnn

(�3.73) (0.14) (2.84)

Freeze�1 (�) �32.60nnn 0.65 33.25nnn

(�7.13) (0.13) (5.04)

Freeze-2 (�) �9.63nn 9.38 19.01nnn

(�2.26) (1.46) (2.50)

Freeze�3 (�) �11.80 13.73 25.53nnn

(�2.71)nnn (1.69) (2.79)

Control variables

PensionIncome (+) 2.30

(2.21)nn

PensionSensitivity (+) 0.07

(0.58)

Acquisition+ 1 (+) 0.17

(0.16)

EquityIssue+ 1 (+) �4.30

(�0.58)

DebtIssue+1 (+) 4.66

(0.74)

Leverage (+) �0.44

(�0.20)

CEO_NetSelling+ 1 (+) 102.24

(2.26)nn

Underfunded (+) �0.56

(�0.90)

ARR (�) �0.14nnn

(�3.06)

Adjusted R2 79.24%

N 17,560

The table presents the results of a year and firm fixed-effects regression that investigates whether employers opportunistically assume downward biased

discount rates in the years leading up to the hard freezing of their pension plans, and whether managers’ ability to opportunistically bias assumed rates is

mitigated following the passage of the Sarbanes-Oxley Act. Hard frozen defined benefit pension plans are plans where employers have eliminated the

accrual of additional pension benefits for all employees. DR is the assumed discount rate for the projected benefit plan. Freeze is an indicator variable

equal to 1 if the firm’s defined benefit pension plans are hard frozen (subscripts indicate year relative to freeze decision), and 0 otherwise. PreSOX is an

indicator variable equal to 1 if the fiscal year occurs prior to the passage of the Sarbanes-Oxley Act of 2002, and 0 otherwise. PostSOX is an indicator

variable equal to 1 if the fiscal year occurs following the passage of the Sarbanes-Oxley Act of 2002, and 0 otherwise. PensionIncome is an indicator

variable equal to 1 if the firm reports pension income (i.e., negative pension cost), and 0 otherwise. PensionSensitivity is defined benefit pension plan

assets divided by the absolute value of operating income. Acquisition+ 1 is an indicator variable equal to 1 if the firm undertakes an acquisition during year

t+1, and 0 otherwise. EquityIssue+1is the amount of public and private common and preferred equity raised during year t+1 divided by total assets.

DebtIssue+ 1is the amount of public and private common and preferred debt raised during year t+1 divided by total assets. Leverage is long-term debt

divided by total capital (i.e., market value of equity plus long-term debt). CEO_NetSelling+ 1 is the number of net shares sold by the CEO of the firm during

year t+1 divided by the number of common shares outstanding. Underfunded is an indicator variable equal to 1 if the firm’s defined benefit pension plans

are underfunded, and 0 otherwise. Finally, ARR is the actual rate of return on the defined benefit pension plan assets. All coefficients have been multiplied

by 100 for ease of interpretation. Year and firm fixed-effect estimates are not reported.n, nn, nnn denote significance at the 0.05, 0.01, and 0.001 (one-tail) levels, respectively.

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133132

We find that, for the pre-SOX period, employers assume downward biased expected rates of return (which increasesreported pension expenses) and lower discount rates (which increases both reported pension expenses and the reportedpension obligation) in the year of the pension freeze. In addition, we find that employers assume downward biasedexpected rates of return and discount rates in, respectively, the two and three years prior to the freeze year. This evidenceis consistent with employers exaggerating the burden of their defined benefit plans in the years leading up to and including

J. Comprix, K.A. Muller III / Journal of Accounting and Economics 51 (2011) 115–133 133

the year they decide to hard freeze their plans. For the post-SOX period, we find that the use of downward biased pensionestimates becomes less prevalent, consistent with SOX reducing opportunities for strategic accrual manipulation.

In further analyses, we find that managers freeze their pension plans when the economic burden of their plans is higher.Specifically, we find evidence that employers are more likely to hard freeze plans when the costs of supporting the planincrease—i.e., when employers face declining cash flows and report losses. In addition, we find that employers faceconstraints when freezing plans—i.e., employers are less able to freeze plans when the firm is relatively large (and thus thenumber of employees is relatively large), when the plans themselves are relatively large, or when the plans are subject tocollective-bargaining agreements.

Taken together, the findings above extend prior research by providing evidence on managers’ opportunistic use ofdownward biased accounting estimates to influence labor contracts during periods of reduced regulatory scrutiny. Inaddition, these findings extend prior research by providing evidence on the effectiveness of SOX in reducing managers’discretion over a specific accounting estimate.

Acknowledgments

We greatly appreciate useful comments from and discussions with Charles Christian, Gail Goodman (Deloitte &Touche), Susan Hamlen, Jim Irving, Denise Jones, S.P. Kothari (the editor), Brad Lindsey, Mike Mikhail, Siva Nathan, MonicaNeamtiu, Jim Ohlson, Kathy Petroni (the referee), Vanessa Radick, Vern Richardson, and workshop participants at the 2008American Accounting Association Annual Meeting and the following universities: Arkansas, Buffalo, California-Riverside,Florida State, North Carolina-Charlotte, Oregon, Singapore Management, Syracuse, Virginia, and William & Mary. We alsogreatly appreciate the excellent research assistance of Zahn Bozanic, Ryan Casey and Hal White.

References

Amir, E., Benartzi, S., 1998. The expected rate of return on pension funds and asset allocation as predictors of portfolio performance. The AccountingReview 73, 335–352.

Amir, E., Gordon, E.A., 1996. Firms’ choice of estimation parameters: empirical evidence from SFAS No. 106. Journal of Accounting, Auditing and Finance11, 427–448.

Amir, E., Guan, Y., Oswald, D., 2009. The effect of pension accounting on corporate pension asset allocation. Review of Accounting Studies, forthcoming.Aon Consulting, 2003. Aon study: pension plan freezes moving to the forefront; more possible without changes to funding rules. Press release, October 29, 2003.Bader, L.N., 1991. The Financial Executive’s Guide to Pension Plans. Salomon Brothers Inc, New York.Bergstresser, D., Desai, M., Rauh, J., 2006. Earnings manipulations, pension assumptions and managerial investment decisions. The Quarterly Journal of

Economics 121, 157–194.Black, F., 1980. The tax consequences of long-run pension policy. Financial Analysts Journal 36, 25–31.Bodie, Z., Light, J.O., Morck, R., Taggart, R.A., 1984. Funding and asset allocation in corporate pension plans: an empirical investigation. NBER Working

Paper No. 1315 (March). National Bureau of Economic Research, Cambridge, MA.Cohen, D., Dey, A., Lys, T., 2008. Real and accrual-based earnings management in the pre- and post-Sarbanes-Oxley periods. The Accounting Review,

757–787.Comprix, J., Muller III, K.A., 2006. Asymmetric treatment of reported pension expense and income amounts in CEO cash compensation calculations.

Journal of Accounting and Economics 42, 385–416.DeAngelo, H., DeAngelo, L., 1991. Union negotiations and corporate policy: a study of labor concessions in the domestic steel industry during the 1980s.

Journal of Financial Economics 30, 3–43.Financial Accounting Standards Board, 1985. Statement of Financial Accounting Standards no. 87: Employers’ Accounting for Pensions. FASB,

Stamford, CT.Financial Accounting Standards Board, 2003. Statement of Financial Accounting Standards no. 132 (R): Employers’ Disclosures About Pensions and Other

Postretirement Benefits. FASB, Stamford, CT.Financial Accounting Standards Board, 2006. Statement of Financial Accounting Standards no. 158: Employers’ Accounting for Defined Benefit Pension

and Other Post-Retirement Plans. FASB, Stamford, CT.Financial Accounting Standards Board, 2009. Accounting Standards Codification 715: Compensation—Retirement Benefits. FASB, Stamford, CT.Francis, T., Schultz, E.E., 2003. Is pension crisis a scapegoat? The Wall Street Journal, November 18, p. C1.Friedman, B.M., 1983. Pension funding, pension asset allocation, and corporate finance: evidence from individual company data. In: Bodie, Z., Shoven, J.B.

(Eds.), Financial Aspects of the United States Pension System. University of Chicago Press, Chicago, pp. 107–152.General Accountability Office, 2009. Defined Benefit Pensions: Survey Results of the Nation’s Largest Private Defined Benefit Plan Sponsors. United States

Government Accountability Office, Washington, DC.Hewitt Associates, 2006. Survey Findings: Hot Topics in Retirement, 2006. Hewitt Associates, Lincolnshire, IL (January).Jiang, J.X., Petroni, K.R., Wang, I.Y., 2010. CFOs and CEOs: Who have the most influence on earnings management? Journal of Financial Economics 96,

513–526Liberty, S.E., Zimmerman, J.L., 1986. Labor union contract negotiations and accounting choices. The Accounting Review 61, 692–712.Maddala, G.S., 1991. A perspective on the use of limited-dependent and qualitative variables models in accounting research. The Accounting Review 66,

788–807.Mercer Human Resources Consulting, 2006. In: A Closer Look at Recent High Profile Pension Plan Freezes. Update. Mercer Human Resources Consulting,

New York, NY (February).Mittelstaedt, H.F., 1989. An empirical analysis of the factors underlying the decision to remove excess assets from overfunded pension plans. Journal of

Accounting and Economics 11, 399–418.Petersen, M.A., 1992. Pension reversions and worker–stockholder wealth transfers. The Quarterly Journal of Economics 107, 1033–1056.Petersen, M.A., 2009. Estimating standard errors in finance panel data sets: comparing approaches. The Review of Financial Studies 22, 435–480.Sharpe, W.F., 1982. Factors in New York Stock Exchange security returns. 1933–1979. Journal of Portfolio Management 8, 5–19.Stone, M., 1987. A financing explanation for overfunded pension plan terminations. Journal of Accounting Research 25, 317–326.Tepper, I., 1981. Taxation and corporate pension policy. Journal of Finance 36, 1–14.Thomas, J.K., 1989. Why do firms terminate their overfunded pension plans? Journal of Accounting and Economics 11, 361–398Zhang, I.X., 2007. Economic consequences of the Sarbanes-Oxley Act of 2002. Journal of Accounting and Economics 44, 74–115.

(PDF) Pension plan accounting estimates and the freezing of defined benefit pension plans - DOKUMEN.TIPS (2024)

References