Regulatory oversight and trade-offs in earnings management: evidence from pension accounting

For example, the total pension assets of S&P 500 firms in 2012 were more than $1250 billion and the total pension expense was $64 billion. If each firm in the S&P 500 used an additional 20 basis points of discretion in setting the expected return assumption, aggregate net income in 2012 would have increased by approximately $2.5 billion (i.e., .002 *$1250).

Each pension assumption is defined in Appendix 1.

On the contrary, the ownership and qualification of all financial reporting pension assumptions is exclusively the domain of the firm and its auditors. The plan actuary is responsible for the assumptions set for ERISA purposes.

The SEC warned that it would begin asking firms to justify expected return assumptions, particularly those over 9%. The SEC warning was followed by an investigation into the accounting estimates used at six firms; see https://​www.​wsj.​com/​articles/​SB10980635466324​7759?​ns=​prod/​accounts-wsj.

The full statement is available at http://​www.​fasb.​org/​summary/​stsum132r.​shtml. This standard mandated additional quantitative disclosures that allow financial statement users to assess the reasonableness of two of the three critical pension assumptions, the discount rate and the expected return on assets. No new disclosures were mandated for the third critical pension assumption, the compensation rate. Each assumption was disclosed both prior to and after SFAS132R; the only difference was the new supporting information for the discount rate and expected return.

An overview of the new SFAS132R disclosure requirements for the discount rate and expected return and how they can be used to assess the reasonableness of the discount rate and expected return assumptions is provided in Appendix 2.

Securities and Exchange Commission v. General Motors Corporation. United States District Court for the District of Columbia. Case: 1:09-cv-00119, January 22, 2009.

In addition, and unlike the expected return, the discount and compensation rates also affect the amount of other comprehensive income through their effect on the determination of the pension liability.

An exception is the theoretical literature that examines a substitution effect between accounting and real earnings management, and some studies of the banking sector that consider the trade-off between earnings management and capital management.

The ERISA reporting requirements are at the pension plan level, whereas the financial reporting requirements are at the firm level. Some of the firms in my sample sponsor more than one pension plan. In these cases, I take the weighted average pension assumption from the ERISA filings, using the Retirement Protection Act of 1994 (“RPA”) current liability for the weights. The RPA current liability, for plans complying with ERISA, is the present value of accrued plan benefits based on the interest and mortality rates prescribed by the Retirement Protection Act of 1994 (RPA). Because the assumptions used to calculate the RPA current liability are prescribed by statute, this liability measure is generally comparable across firms. As an example of how this calculation works, suppose that a firm has two pension plans with $75 and $25 of RPA current liability and asset return assumptions of 8% and 10%. In this case, I would calculate the asset return assumption for the firm to be 8.5% (0.75*8% + 0.25*10%).

This sequence isn’t necessarily the case for all firms, as the interplay between the duration of the pension plan, its funded status, and the pension plan provisions can result in a lower discount rate producing higher pension expense. However, as highlighted in the calculations provided in Table 1, this is extremely unlikely for the pension plans in my sample during the period of my study. Additional discussion of how the duration and the discount rate assumption can affect reported pension expense is provided by Fried, Davis, and Davis-Friday (2014).

The duration concept that is typically applied to bonds is defined similarly for pension plans. In the case of bonds, the duration is a measure of the average time to receipt of cash payments from the bond. For pension plans, the duration is a measure of the average length of time over which benefit payments will be made from the plan.

The prescribed approach is discussed in more detail in EITF Topic No. D-36.

The Citigroup Pension Discount Curve is composed of individual, zero-coupon interest rates for durations up to 30 years that are mathematically derived from observable market yields for AA-rated corporate bonds. It is the only approach that specifically complies with the requirements of FAS87. Other bond indices, such as Moody’s AA corporate Bond Index or Merrill Lynch’s High Quality Bond Index, have not been designed to match the benefit payment stream of a pension plan, so as such only indicate the yields produced by a select group of high-quality bonds at a fixed date. Therefore changes in these indices can result simply from a change in the duration of the bonds that comprise the index, rather than a change in interest rates. As a result, these indices are inappropriate for complying with pension accounting requirements. I downloaded the Citigroup Pension Discount Curve from the Society of Actuaries website (http://​www.​soa.​org).

The specific pension liability measure I use is the RPA current liability. The breakdown of the RPA current liability is provided in the Schedule B attachment to the required ERISA Form 5500 filing and therefore is available for all U.S. qualified plans.

A frozen pension plan is one in which active participants are no longer accruing any additional benefit payments.

Consider the example of Teradyne, Inc. From its 2003 Form 5500 Schedule B item 2(b), the components of the RPA current liability are $34,953,692, $32,695,502, $34,953,692, and $52,413,244, for active vested, active nonvested, terminated vested and retired participants, respectively. The sum of these four components equals total RPA current liability, which is $155,016,130. From Form 5500 Schedule B item 1(d)(2), the RPA normal cost and expected disbursements are $3,510,321, and $5,000,000, respectively. Next, each of these six inputs is scaled by the total RPA current liability, and then each scaled component is multiplied by the coefficients in Table 2 (i.e., 0.2255*16.86 + 0.2109*20.59 + 0.2255*23.69 + 0.3381*10.57 + 0.0226*26.72 + 0.0323*-50.56). This produces a duration for the plan of 16.034 years.

The assumption that I select from the ERISA filings is for the same period (e.g., for calendar year 2004). In some cases, there is not perfect overlap between the plan year and the fiscal year. In these cases, I choose the ERISA assumption for the valuation date following the fiscal year-end. Dropping observations when there is not perfect overlap between the plan year and fiscal year does not change my results.

More details are provided in Actuarial Standard of Practice No. 27. Even though the financial reporting and ERISA assumptions are equivalent, the filing dates are not. The deadline for ERISA filings is 20.5 months after the actuarial valuation date. Therefore a plan year that begins on January 1, 2004, would have an ERISA filing deadline of September 15, 2005. As a result, the assumptions selected by the plan actuary are not available at the time the auditor certifies the assumptions selected by management for financial reporting purposes.

Discretion in ERISA assumptions may be employed to manage the firm’s cash contribution requirements. For most plans, the binding calculation that influences mandatory contributions is set by statute (e.g., Rauh, 2006), and so there is limited incentive to add discretion to certain ERISA assumptions. In addition, if there was discretion in the ERISA assumptions, it would bias against my findings to the extent that the discretion is in the same direction (i.e., that managers want lower contributions and lower pension expense).

The reasonable range for the actuarial assumptions was based on a report, “Accounting for Pensions and Other Postretirement Benefits,” completed by Watson Wyatt Worldwide and available on the firm website. For the asset allocations, I checked all allocations that did not total 100% against the annual report.

The pension assets for the average firm in my sample at the time of the SEC warning were $707 million. The increase in reported pension expense, due to an ERA assumption that is 24 basis points lower, is $1.7 million (i.e., $707 million × 24 basis points). The illustration of pension expense in Table 1 shows that the change in reported income from a 1 basis point change in the DR is approximately equal to a 2 basis point change in the CR and a 3 basis point change in the ERA for the median firm in my sample. This implies that the change in pension expense for a 19 basis point change in the DR is approximately equal to the change in pension expense from a 57 basis point change in the ERA. Therefore the decrease in pension expense associated with a 19 basis point reduction in the DR is $4.0 million (i.e., $707 million × 57 basis points). Similarly, since a 1 basis point change in the CR is approximately equal to a 2 basis point change in the ERA, the change in pension expense for a 7 basis point change in the CR is approximately equal to the change in pension expense from a 14 basis point change in the ERA. Therefore the decrease in pension expense associated with a 7 basis point increase in the CR is $1.0 million (i.e., $707 million × 14 basis points).

Because the variance captures the shape of the distribution, it is not directly affected by a change in the mean. To illustrate this point, consider what happens when a constant is subtracted from the mean of a distribution. This reduces the mean of that distribution, but it has no effect on its variance.

In robustness tests, I examine whether my results are sensitive to the selection of the pre and post periods through two tests. First, I change the post period to include firms whose fiscal years end between December 15, 2004, and December 15, 2005. Second, I exclude firms that do not have fiscal years that end in December. Four hundred and fifty out of the 500 firms in my sample have fiscal years ending in December. For these firms, the pre period is fiscal years ending on December 31, 2002, and the post period is fiscal years ending on December 31, 2004. These adjustments do not affect my results.

Even though there isn’t an overlap between the FAS132R and SEC warning post periods, part of the disciplining effect of the SEC warning may be captured in later periods. For example, if a firm believed it needed to reduce its ERA by 50 basis points, it may have decided to reduce the assumption by 25 basis points in 2002 and the remaining 25 basis points in 2003. At least part of the reduction in discretion in the ERA in 2003 may be because of this type of choice.

The pension assets for the average firm in my sample after the implementation of SFAS132R are $948 million. The change in reported pension expense, due to an ERA assumption that is 19 basis points lower, is $1.8 million (i.e., $948 million × 19 basis points). As with the prior calculation, I exploit the fact that the increase in reported income from a 1 basis point change in the DR is approximately equal to a 2 basis point change in the CR and a 3 basis point change in the ERA for the median firm in my sample. Therefore the increase in pension expense associated with a 30 basis point reduction in the DR is $8.5 million (i.e., $948 million × 90 basis points), and the decrease in pension expense associated with a 13 basis point increase in the CR is $2.5 million (i.e., $948 million × 26 basis points).

As previously noted, a CR assumption is absent because the pension plan is either frozen (i.e., participants are no longer accruing additional benefits) or has a benefit formula that is not based on pay (e.g., $20 per month for each year of service).

Variations on these cutoffs don’t affect results.

In December 2008, the FASB issued Staff Position FAS 132(R)-1 amending SFAS 132(R), “Employers’ Disclosures about Pensions and Other Postretirement Benefits,” which, among other things, expands the required disclosures regarding assets in an employer’s pension and postretirement benefit plans. This Staff Position has no effect on the principal methods used in this paper, but the expanded disclosures could be used for an additional robustness test similar to equation (5).

For example, a plan with 60% equity, 30% bonds, and 10% real estate would have an estimated ERA of 6.9%.

Hewitt Associates started preparing these reports in 2003 in response to updated accounting guidelines for corporate plans. These reports are not publicly available. However, a similar report, prepared by PCA for 2012, is available online at http://​www.​pensionconsultin​g.​com/​pdfdocs/​2012_​PCA_​Asset_​Class_​Assumptions_​Report.​pdf

When I informally surveyed three other actuarial firms and two auditing firms, I found that the expected investment returns by asset class were similar at each of these organizations and that these return assumptions were generally unchanged between the pre- and post-SFAS132R periods.

For firms that need to report unfunded obligations, as a result of an additional minimum liability calculation, the DR and CR also offer additional benefits through a reduction in the pension liability. The ERA assumption, on the other hand, affects only the determination of expense. For firms that are frozen, overfunded, and don’t have a loss amortization, the impact of the DR and CR, relative to the ERA, would be reduced. However, each of these attributes occurs relatively infrequently. In my sample of 624 firms, 601 have a loss amortization base, and 426 report a service cost that is at least 50% of the expected return on assets. Moreover, only 87 firms have a funded ratio in excess of 100%. In fact, more firms that are likely to have a stronger expense impact from the DR and CR assumptions, relative to the ERA. Ninety-one firms reported a service cost that was higher than the expected return on assets, and 177 have a total loss amortization that is more than double the level used in this example.