The impact of governmental accounting standards on public-sector pension funding

Abstract

The funding policy for defined benefit pension plans covering government employees represents an important decision for governments sponsoring those plans. Many state and local government plans have become severely underfunded (e.g., New Jersey, Illinois, and Detroit), raising concerns about whether governments are contributing enough to their pensions. Governmental Accounting Standards Board Statements 67/68 (GASB 67/68) fundamentally alter the financial reporting of pension liabilities, by (i) requiring pension liabilities to be estimated using a potentially lower discount rate (increasing estimated liabilities and any funding deficits), and (ii) mandating balance sheet recognition of funding deficits/surpluses. Although GASB 67/68 only change financial reporting and acknowledge specifically that funding is outside their scope, we find, for 100 large state-administered plans, that governments increase pension contributions significantly upon applying GASB 67/68. This funding response is stronger from governments likely to face greater political consequences once pension deficits are made prominent by GASB 67/68. Benefit cuts are also more likely post GASB 67/68, but plans that increase funding are less likely to cut benefits—suggesting that these responses substitute for each other and that pension funding is more of a fiscal priority in some states than others. Overall, our findings suggest that purely accounting changes can have “real” effects on governmental pension policy.

1 Introduction

Employer-sponsored pensions are a key pillar of retirement income security for workers in the United States (US). Within state and local governments—which employ 14% of the US workforce—retirement plans have typically been defined-benefit (as opposed to defined-contribution) in nature, and 80% of state and local government employees rely solely on a defined-benefit (DB) plan for retirement income (Munnell and Soto 2007).¹ As of December 2019, state and local DB plans held $4.5 trillion in assets to fund the retirement of 14.7 million active and 11.2 million retired employees of state and local government.²

Yet, despite the tremendous importance of public-sector pensions, many state and local DB plans are in near crisis.³ In 2015, the largest state and local pension systems reported collective unfunded liabilities of $1.4 trillion using valuation techniques from extant accounting standards. Alternative valuation techniques that hew more closely to finance theory value the deficit at closer to $4 trillion (Rauh 2017). While the largest state and local plans were funded to cover 73% of liabilities on average in fiscal years 2016 to 2021, funding levels vary widely, with about one-fifth of all plans less than 60% funded.The fiscal pressure from deteriorating pension funding is large enough to have led to credit rating downgrades in states with critically underfunded pensions, such as Illinois, New Jersey, and Connecticut (S&P State Global Market Intelligence 2017).

We examine whether accounting regulation impacts governments’ funding policies with regard to public-sector pensions. Prior to the accounting changes that we examine, pension commentators point to decades of generous benefit promises combined with persistent underfunding as having led plans to their current depleted state.⁴ Many states have historically contributed less than their annual required contribution (ARC), which represents the employer’s cost of retirement benefits earned by employees in the current year.⁵ Currently, proposals to relieve the funding burden by cutting benefits, eliminating cost-of-living adjustments (COLAs), and requiring employees to work longer or contribute more are being debated in courts across the US (Brainard and Brown 2016). The fiscal pressures imposed by the pandemic have only made these concerns more urgent. With rapidly burgeoning moves to curtail or restructure benefits, the public retirement system in the US has arrived at a crossroads. Decisions to fund pension plans (or not, as the case may be) carry enormous implications both for the beneficiaries who depend upon pension promises and for the taxpayers who may be called upon to fulfill those promises, either through higher taxes or cuts to public spending.

Two GASB pronouncements, issued in June 2012, dramatically alter the accounting and reporting of pension expense and funding for US state and local governments, aiming to increase the transparency and comparability of pension commitments that governments have accrued over time. GASB Statement No. 67 Financial Reporting for Pension Plans (applicable to plan reporting, effective for plan years starting after June 15, 2013) and Statement No. 68 Accounting and Financial Reporting for Pensions (applicable to sponsor reporting, effective for fiscal years starting after June 15, 2014) collectively introduce changes to (1) measurement and (2) recognition (Governmental Accounting Standards Board (GASB) 2012a, b).

With respect to measurement, previously the pension obligation was valued by discounting projected future benefits at the long-term expected rate of return (ERR) on pension assets. Now the ERR can only be used to the extent that plan assets are projected to be sufficient to meet benefit payments, with a high-quality, tax-exempt municipal bond rate applied to any balance of (unfunded) benefit payments. This results in a “blended” discount rate applied by any plan that is only partially funded. As governmental pension plans invest extensively in equities, their ERRs were typically much higher than high-quality municipal bond rates; the requirement to use a blended rate therefore lowers discount rates, resulting in larger estimates of pension liabilities and funding deficits. With respect to recognition, any net pension liability that is determined by subtracting the fair value of pension assets from the pension obligation, as measured above, is to be recognized on the financial statements of state and local governments for the first time. Prior to these changes, GASB standards only required footnote disclosure of the unfunded liability.

The pension liabilities of state and local governments are so substantial that pension funding considerations shape public policy, budgets, and credit quality for many states (Kilroy 2015). By (1) potentially increasing measured pension liabilities and (2) requiring recognition of the underfunded liability on governmental balance sheets, GASB 67/68 have the potential to place economically substantial liabilities on state and local government financial statements.

These changes, in turn, could heighten scrutiny of funding deficits from an array of stakeholders: (i) credit rating agencies, for whom pension deficits are a key consideration in the rating process; (ii) municipal bond investors, whose general obligation bonds are typically junior to pension deficits (Novy-Marx and Rauh 2012); and (iii) taxpayers, given that balanced budget requirements in most states necessitate higher taxes or cutbacks to other government expenditures to close pension deficits (Allen and Petacchi 2018; Costello et al. 2017). More scrutiny from these stakeholders raises the specter of increased financing costs and citizen oversight of pension management. Given these eventualities, it is plausible that governments respond in ways that minimize the negative impact to their reported financials, which begs the questions: Do governments respond to GASB 67/68, and if so, how? To the extent to which GASB 67/68 create negative economic consequences for sponsoring governments, do they attempt to minimize the impact on reported numbers? Answering these questions is the objective of our study.

We examine one readily available outcome for most plans: how much cash is contributed into the plans. Increasing contributions reduces the GASB 67/68 liability that needs to be recognized in multiple ways. For a plan that was underfunded before GASB 67/68, increasing contributions not only increases the fair value of plan assets but also—due to the mechanics of GASB 67/68 valuation—decreases the estimate of the pension liability. That said, GASB 67/68 did not explicitly require governments to take any real actions to improve pension funding; in fact, one of the GASB’s explicit aims was to implement a theoretically motivated approach to pension accounting that was separate from pension funding policy. Therefore, to the extent to which we do observe real actions to improve pension funding (e.g., higher contributions or lower benefits), these would not only indicate unexpected consequences of the standards but also highlight that accounting can indirectly affect cash flows due to its role in municipal financing and elections.

We examine a sample of the 100 largest state plans covered by the Public Pension Database of the Boston College Center for Retirement Research. For this sample, which covers about 80% (90%) of public pension assets (members) nationwide, we document an increase in total contributions of 3.6% of prior-year covered payroll following GASB 67 implementation, after controlling for plan- and state-level determinants of contributions. This increase, of about $108 million in raw (unscaled) contributions, translates to about 18% (40%) of the median (standard deviation) of total contributions, and to almost 1% (0.34%) of states’ median tax revenues (total revenues) pre GASB 67/68.⁶ Importantly, contributions from participating employers and states (eventually borne by taxpayers) increase, rather than contributions from the employees themselves. A very similar increase in contributions persists on adding plan fixed effects, which absorb time-invariant, unobservable factors at the plan (and largely, state) level, moving us closer to causal inference.

In cross-sectional tests aiming to sharpen inferences, we find that the contribution response is stronger from governments subject to greater political scrutiny, for which recognition of liabilities would be more consequential: those with greater public-sector labor union presence and those in states with upcoming gubernatorial elections. As unions are known to exert pressure for securing workers’ retirement benefits and as pension funding is a prominent issue in election platforms, both findings suggest that the responses we document are driven by a desire to minimize political consequences that may arise from pension deficits made prominent by GASB 67/68.

While our baseline tests examine contribution increases, governments could also respond with other real actions that have negative consequences for public employees, such as cutting benefits. We document that benefit cuts along multiple dimensions are generally more likely post GASB 67; importantly, plans that increase contributions post GASB 67 are significantly less likely to cut benefits, suggesting a substitutive relationship between these two kinds of responses.

Our study offers the following contributions. First, we provide some of the first large-scale empirical evidence on the consequences of GASB 67/68, a pair of standards that were fiercely debated over an exposure period of six years, and which continue to be controversial.⁷ On one hand, researchers forecasted substantial increases in reported liabilities ex ante (Munnell et al. 2012; Mortimer and Henderson 2014).⁸ On the other hand, compilations of state CAFR data post GASB 67/68 implementation indicate that discount rates have shifted for only a small minority of plans (Moody’s Investor Service March 2015). Early forecasts of the impact of GASB 67/68, therefore, are at odds with the eventual reality of their implementation, a puzzling turn of events that is the subject of much comment.⁹ By demonstrating evidence of governments’ responses along multiple dimensions (contribution increases and benefit cuts) that appear to be designed to mitigate the impact of GASB 67/68, our study provides a bridge between the seemingly incongruous early forecasts and the eventual reality.

Our second contribution lies in demonstrating that even in the governmental setting, accounting standards that affect only the measurement and recognition of accruals-based assets, liabilities, and income—without direct effects on cash flows—can elicit managerial responses that affect cash flows. In other words, even in the governmental arena, accounting does have “real” effects. It is well understood that accounting standards have real effects for corporations, and a rich literature documents evidence of managers altering the terms of business transactions to achieve an accounting outcome or moderate the impact of an accounting change.¹⁰ This literature concludes that many of these actions are motivated by managers’ desire to preserve firm valuation or to preserve accounting-based inputs to (compensation or debt) contracts.

While the primacy of accounting numbers is well-established in the corporate context, whether governments are similarly motivated to alter real transactions to achieve accounting outcomes is not as clear. Governments have distinct organizational objectives; they report to a diverse array of stakeholders, some of whom have no equivalent in the for-profit world (citizens and elected representatives) and some of whom do (creditors and pension beneficiaries); and their accounting reports are prepared by bureaucrats and overseen by elected or appointed officials whose incentives differ from corporate managers’. Allen and Petacchi (2018) find that states with weaker plans and greater fiscal pressures are more likely to oppose the GASB’s proposals, and that opposing state governments are more likely to initiate benefit cuts as an early response. These ex ante actions portend that governments will not be passive bystanders to the expected impact of GASB 67/68; our evidence expands on these findings and confirms that governments are indeed willing to expend resources to minimize that impact. To the best of our knowledge, Khumawala et al. (2017), who document a reduction in US municipalities’ derivatives usage after an accounting rule change requiring balance sheet recognition of derivatives, provide the only other work demonstrating that accounting recognition spurs real actions in a governmental setting.

Section 2 describes the public pensions landscape, the changes introduced by GASB 67/68, and lays out predictions. Section 3 describes the sample and research design. Section 4 discusses results on contributions. Section 5 explores benefit cuts, and Section 6 concludes.

2 Institutional background and hypothesis development

2.1 State and local government pension plans (“public” or “governmental” plans) in the US

The Census of Governments identifies about 2,670 retirement systems sponsored by a state or local government. Munnell et al. (2008c) report that state- (as opposed to locally) administered retirement systems, which cover general state employees and teachers, account for only 8% of all plans but 88% (82%) of active participants (assets).¹¹ Benefits under DB plans are usually determined by multiplying final average salary by years of service and a multiplier, which is around 2% in the public sector (Brainard 2007; US Department of Labor 2007). Governmental plans typically also provide COLAs (either automatically or on a frequent but ad-hoc basis), which are uncommon in corporate-sponsored plans (Munnell and Soto 2007). Also unlike corporate plans, governmental plans usually stipulate contributions from employees, in addition to contributions from employers and non-employer sources (e.g., the state).

Unlike corporate plans, public plans are not subject to the Employee Retirement Income Security Act (1974), and states vary in how the plans are governed. Still, some commonalities exist: most public plans hold assets in trust and are governed by a board of trustees required to act in the beneficiaries’ interest. Trustees either (i) serve by virtue of their public office (e.g., a state treasurer who automatically serves on the board), (ii) are appointed by an elected official, or (iii) are elected by beneficiaries (Fitzpatrick and Monahan 2012). While almost all plans have annual contribution requirements, enforcement varies, so the required contribution is not always contributed.¹²

2.2 The erstwhile pension accounting framework and the shift brought about by GASB 67/68

From 1994, the accounting for public plans was governed by GASB Statements No. 25 and No. 27.¹³ GASB 25 laid out disclosures required in financial statements: plan assets and liabilities, required contributions, and the ratio of actual to required contributions, among others. GASB 27 focused on defining the employer’s annual pension expense reported in sponsors’ financial statements, equal to the annual required contribution (ARC), defined as normal cost (i.e., service cost) for the year plus any payment required to amortize the unfunded liability over 30 years.

GASB 67 and 68 changed the approach towards pension accounting along two dimensions. First, while GASB 25/27 did not intend to provide sufficient conditions for appropriate funding, the ARC had “become a de facto funding standard” that was “used to evaluate funding decisions of the employer” and had even “standardized the funding approaches of state and local governmental employers” (GASB 68 Basis for Conclusions, para 160). GASB 67/68, in contrast, created a “new hard-line division of accounting from funding” (Senta 2014). Second, these pronouncements shifted the focus of accounting from the income statement to the balance sheet. GASB 27 focused on what employers contributed each year relative to the ARC for that year; only if the cumulative ARCs exceeded the cumulative amount contributed to date did a liability appear on the balance sheet, to that extent.¹⁴ In contrast, GASB 68 assigns a plan’s full unfunded liability (total pension liability minus fair value of plan assets) to the balance sheet of the sponsoring employer, with a “smoothed” annual pension expense (calculated in a manner reminiscent of corporate pension expense per FASB) appearing on the income statement.¹⁵

In addition to requiring recognition of the pension deficit or surplus, GASB 67/68 also change how the pension liability is measured. Whereas the pension liability was previously determined by discounting projected future benefits at a discount rate equal to the expected rate of return (ERR) on plan assets, GASB 67 now requires a blended discount rate to be applied in cases where the plan is projected to become insolvent at some point in the future.

To make this determination, the GASB specifies a multi-step process: (1) project future benefit payments to current employees based on current benefit terms; (2) project plan assets forward, taking into account expected inflows (contributions, asset returns) and outflows (benefit payments, expenses) associated with current members; (3) if the assets so projected are sufficient to cover benefit payments for all periods, then use the ERR to discount those benefit payments; (4) if the assets so projected are insufficient to cover benefit payments for all periods (i.e., if the plan is projected to run out of funds at some point), discount all benefit payments until that “projected depletion date” or “crossover point” using the ERR, then discount all benefit payments after that date using a high-quality, tax-exempt municipal bond rate; and (5) the single equivalent discount rate that, when applied to all cash flows, produces the same total present value as the two-step discounting described above, is the blended discount rate.¹⁶ Appendix 1 provides details on this procedure, while Appendix 2 comprehensively lists the changes in GASB 67/68.¹⁷

2.3 Potential implications of GASB 67/68: managerial responses

The GASB identifies three potential user groups for governmental reporting: the bond market, citizens/voters, and elected representatives. Of these groups, investors in the bond market (and information intermediaries such as credit rating agencies and bond analysts) pay close attention to pension funding. Rating agencies state that pension deficits are a major driver of credit quality for states (Kilroy 2015), and deteriorating funding is often quoted as driving ratings downgrades (e.g., S&P Global Market Intelligence 2017). In academic evidence, pension funding and performance have been shown to affect municipal credit ratings as well as bond spreads (Marks and Raman 1988; Novy-Marx and Rauh 2012; Martell et al. 2013).

Many states also have balanced-budget restrictions, which imply that taxes will have to be raised or public expenditures curtailed to close pension funding gaps (Costello et al. 2017); citizens/voters could be concerned about pension deficits as a result. While there is little direct evidence in this regard, Rich and Zhang (2015) document that pension funding is stronger in municipalities that permit direct citizen participation in the legislative process or where elected officials have recently faced recall attempts. These findings imply, at the minimum, that pension funding is an outcome of interest for citizens.

Due to the strong interest of the bond market and the citizenry in public pension funding, we would expect elected representatives from both the executive and legislative branches of government to also be concerned about pension reporting. Kido et al. (2012) document evidence consistent with public pension funding being inflated ahead of gubernatorial re-election campaigns. Reported pension funding, therefore, is a metric that public officials appear to believe is important. The role of public-sector unions is particularly relevant here, as these unions are politically influential and described as “interest groups of the first magnitude” in American government (Anzia and Moe 2015), often fielding candidates in elections and supporting them through fundraising and Political Action Committees.¹⁸ Public-sector unions, however, represent interests that arise from inside the government (Anzia and Moe 2015), typically working for better wages, benefits, and benefit security for participating government employees. Therefore, we would expect public-sector unions to care about pension funding as well as to have the ability to influence funding outcomes (Bonsall et al. 2019).

Given that (i) key stakeholders are concerned about pension funding and (ii) GASB 67/68 are expected to lower reported funding status, it is plausible that sponsoring governments will try to mitigate the fallout from GASB 67/68. Mortimer and Henderson (2014) posit that governments with low funded ratios could be “tempted to manage reported pension values,” either through the opportunistic use of discretion in estimating the crossover point and the GASB 68 pension liability or through economic transactions and activities (i.e., “real” actions). Many real actions are possible, with varying impacts on beneficiaries: on one hand, governments could reduce reported pension liabilities by closing DB plans or laying off employees, leaving beneficiaries worse off. On the other hand, they could contribute more to their plans to reduce the reported deficit, improving benefit security for participants.

In this study, we focus on examining how annual contributions shift due to GASB 67/68. We focus on contributions for two key reasons. First, at a conceptual level, GASB 68’s valuation approach makes contributions a heightened lever with which to reduce reported pension funding deficits. Increasing contributions not only increases the fair value of plan assets but can also reduce the estimated pension liability, both directly (by increasing the funded base that can be discounted at the ERR rather than the municipal bond rate) and indirectly (by helping to justify assuming a higher stream of expected future contributions when projecting the plan’s depletion date). For plans with a formal funding policy that are allowed to apply judgment in projecting historical contributions into the future, higher current contributions should help to justify projections of higher future contributions. For plans lacking a formal funding policy, higher current contributions increase the five-year average of most recent contributions that serves as the maximum projected future contribution. Higher projected contributions, in turn, reduce the likelihood that the plan will be projected to run out of assets in the first place. Second, at a practical level, contributions are readily observable and can be changed in response to the standards. As higher contributions lower reported pension deficits, we expect that contributions to public pensions will increase post GASB 67/68.¹⁹ Our main prediction follows (in alternative form):

• H1: Contributions to public pensions increase upon GASB 67/68 implementation.

Our prediction that GASB 67/68 will elicit real actions is not straightforward, however; it hinges on some assumptions. The first of these assumptions is that financial statement measurement and recognition matter for governments. Note that none of the GASB 67/68 provisions alter the underlying fundamentals of what is owed to pensioners: to quote Munnell et al. (2012), “$1,000 owed to a retired teacher in ten years under current standards will remain $1,000 owed in ten years under the new standards.” Estimates of the pension liability developed by discounting those benefits at the ERR were disclosed in governments’ financial statements since long before GASB 67/68. The new standards only discount those benefits at a potentially lower rate, yielding a higher single number representing what is owed in present value terms, and place that number (net of pension assets) on the government-wide balance sheet. If bond investors/voters were already discounting benefits at lower rates and were also impounding the resulting higher liabilities into their decisions, then financial statement recognition of those higher liabilities need not, by itself, create economic consequences (or elicit responses to minimize those consequences).

Whether financial statement measurement and recognition of pensions indeed matter in the governmental setting is not clear, with extant evidence painting a mixed picture. On one hand, there is evidence that rating agencies were already discounting pension benefits at rates lower than the ERR. For example, Moody’s already used a high-grade corporate bond index to discount governmental pension liabilities (akin to the FASB-mandated discount rate); it announced not only that its methodology would remain unchanged but also that it did not expect ratings to change after GASB 67/68; so did S&P.²⁰ On the other hand, the exposure drafts for GASB 67/68 elicited an overwhelming response, with many arguing that the proposals would prompt governments to close plans (Allen and Petacchi 2018). Vermeer et al. (2012) document that one-third of the local governments they study did not even report the unfunded pension liability on the CAFR under GASB 27, which is prima facie hard to reconcile with the rating agencies’ contention that GASB 67/68 will not change ratings. Hence, ex ante, it is not clear whether GASB 67/68 measurement and recognition will alter key stakeholders’ decision processes.

The second assumption is that these economic consequences incentivize government decision-makers to respond. In the corporate setting, poor reported performance leads to scrutiny from analysts, investors, and boards and lowers valuations with impacts on managers’ compensation and tenure. In the governmental setting, even if borrowing costs should increase, it is not clear whether the bureaucrats and elected officials who set pension policy face consequences. There is indirect evidence that poorer pension funding brings a greater threat of recall (Rich and Zhang 2015) and a lower chance of reelection (Kido et al. 2012), suggesting some consequences for public officials.²¹ However, public policy decisions often do not take effect immediately—policies enacted in one administration may only take effect in the next. Politicians frequently shift blame by claiming to have “inherited” problems from their predecessors (McGregor 2017, Washington Post). Thus, it is not clear whether the economic consequences of GASB 67/68 will motivate decision-makers to respond.

The third assumption is that government officials have the discretion to take actions—real or otherwise—that can minimize the impact of GASB 67/68, and that those actions will be observable in the time period we examine. Funding increases may take time to effect, as budget priorities have to be reset, particularly in states facing balanced-budget mandates. Legislation must be amended in states where contributions are set by statute. It is possible that decision-makers can exercise discretion in “accrual” estimation of the crossover point for funding levels but are unable to undertake “real” actions, in which case we would not find support for our hypothesis. For all these reasons, it is an empirical question whether contributions increase after GASB 67/68.

2.4 Cross-sectional differences in the response to GASB 67/68

Governments’ motivations to minimize the impact from GASB 67/68 could vary; the countervailing forces described above could also vary. Hence, we do not expect uniform responses from all governments. We motivate a series of cross-sectional tests from these differences.

First, the dollar impact from GASB 67/68 varies across plans. For plans that are not reasonably expected to run out of assets, the blended discount rate will continue to be the same as the ERR. Pension liability measurement for these plans need not change; the impact will be that this liability, previously only disclosed in footnotes, is now recognized on the government-wide statement of net position (a “recognition versus disclosure” shift). For plans that are projected to run out of assets, however, the discount rate will shift downwards, resulting in a larger reported pension liability—a measurement effect in addition to a recognition effect. Moreover, some plans will experience a greater upswing in the estimated pension liability than others. Finally, the pension liability forms a more economically significant part of the balance sheet for some governments than for others. To the extent that there is a response to mitigate GASB 67/68’s impact, this response should intuitively be stronger from governments facing a larger impact. Accordingly:

• H2: The increase in contributions after GASB 67/68 is stronger for governments facing a larger expected magnitude of financial impact from the standard.

Second, the financial statement recognition of one dollar of pension liabilities could generate more adverse economic consequences for some governments than for others. To the extent to which bond market perceptions are affected by GASB 67/68, higher reported liabilities may trigger credit rating downgrades, increase the cost of debt financing, or raise the specter of covenant violation and renegotiation. We expect these consequences to be more severe for governments that rely more heavily on debt financing (Allen and Petacchi 2018). Accordingly, we predict that these governments have stronger incentives to mitigate the impact of GASB 67/68.

• H3: The increase in contributions after GASB 67/68 is stronger for governments that rely more heavily on debt financing.

A caveat to this prediction is that governments relying heavily on debt financing may not have the ability to reduce the GASB 67/68 liability by increasing contributions, because they are financially constrained to begin with. These governments may attempt to reduce the GASB 67/68 liability through other actions, such as cutting benefits or using accounting discretion.

Third, the financial statement recognition of a dollar of pension liabilities could have greater political ramifications in some settings. State and local government workforces are marked by particularly high levels of union membership, with public-sector unions being politically influential in many states. States with stronger public-sector unions not only provide more generous pension benefits for their members (Ehrenberg and Schwartz 1983; Anzia and Moe 2015) but also are more likely to increase pension contributions following the financial crisis (Bonsall et al. 2019), consistent with unions’ incentives to exert their collective bargaining power to ensure greater benefit security for their members. To the extent to which GASB 67/68 make pension funding deficits more visible and salient, the resulting funding pressures could be particularly strong in states with strong public-sector unions. Accordingly:

• H4: The increase in contributions after GASB 67/68 is stronger for governments with relatively strong public-sector unions.

Fourth, the political ramifications could be greater for some elected officials. Highly visible pension deficits could be particularly problematic for governors facing re-election, as elections heighten scrutiny on the state’s fiscal health and how it has evolved under the governor’s watch.²² If state pension funding is an important and visible policy issue in the election, there could be greater motivation to increase pension funding ahead of elections. Accordingly:

• H5: The increase in contributions after GASB 67/68 is stronger for governments with upcoming gubernatorial elections.

Union membership (H4) and elections (H5) capture aspects of the state’s political environment; they also capture increased monitoring and scrutiny of public pensions. Governments could respond to this scrutiny with real actions to improve contributions, as we predict, but also with accrual-based actions that reduce estimated pension obligations and present a more favorable picture of pension funding. For example, pension funding is inflated through accrual choices ahead of gubernatorial elections (Kido et al. 2012) and in states with large public-sector unions (Bonsall et al. 2019). To the extent to which accrual manipulation (to increase reported funding) serves as a substitute to actual increases in funding, it could act as a countervailing force to these predictions.

3 Sample and research design

3.1 Sample selection

We obtain plan-level data from the Boston College Public Plans Database (PPD). The PPD contains data for 170 public pension plans: 114 (56) of the largest state-administered (locally administered) plans from all US states, comprising 82% (13%) of public pension assets nationwide. As local plans vary widely in governance structure and political and regulatory environments, we restrict our sample to state plans. Of the 114 state plans on the PPD, our final sample is 100 plans with non-missing values for all our variables of interest. Table 1 describes our variables of interest, many of which we obtain directly from the PPD.

Table 1 Variable Measurement

3.2 Empirical specification

We estimate the impact of GASB 67/68 on contributions with the following model:

$$\begin{aligned}Contributions&=\beta_0+\beta_1\;POST\hbox{-}GASB67+\beta_2\;Funding\;Ratio(Prior\hbox{-}Year)\\&+\beta_3\;\%Required\;Contributions\;Paid+\beta_4\;Plan\;Covers\;Teachers\\&+\beta_5\;Accrued\;Liability/Tax\;Revenue+\beta_6Unfunded\;Accrued\;Liability/Tax\;Revenue\\&+\beta_7\;Exp\_Minus\_Rev+\beta_8\;TTLBAL\;Ratio+\beta_9\;Per\;Capita\;Total\;Debt\;Issued\\&+\beta_{10}\;Per\;Capita\;Growth\;in\;Gross\;State\;Product+\beta_{11}\;BalancedBudget\\&+\beta_{12}\;Union\;Membership+\beta_{13}\;Election+\beta_{14}\;Corruption\\&+\beta_{15}\;Constitution+\beta_{16}\;Gratuity+Year\hbox{-}Quarter\;Fixed\;Effects\end{aligned}$$

(1)

The dependent variable Contributions is cash contributions for the fiscal year, scaled by prior-year covered payroll. We estimate the model separately for contributions from (i) employees, (ii) employers, and (iii) all sources (employees + employers + state). The increase in contributions, if any, could come from any source; notice, however, that if employee contributions increase (presumably because higher contributions are mandated), this would be tantamount to a benefit cut from the employees’ perspective. On the other hand, if employer/state contributions were to increase, that implies a greater ultimate burden on taxpayers.

Our independent variable of interest is the POST-GASB67 indicator, which is set to one for all fiscal years applying GASB 67; i.e., fiscal years beginning after June 15, 2013. While GASB 67 (which introduced the new pension liability measurement) is applicable to plan-years beginning after June 15, 2013, and GASB 68 (which mandated financial statement recognition) is applicable to fiscal years beginning after June 14, 2014, we use the earlier of the two effective dates as our cutoff, as both standards were issued together and the revised pension liability estimates must be produced as of the effective date of GASB 67.²³

Our specification relies on time-series differences in contributions between pre- and post-GASB 67 periods to infer the impact of GASB rule changes in the absence of a feasible control group of plans that remain unaffected by GASB rules.²⁴ As a result, contemporaneous events or macroeconomic shifts that affect contributions for the universe of plans have the potential to confound our results. In order to more confidently attribute any shifts we observe to GASB 67/68, we incorporate (i) a comprehensive set of control variables and (ii) plan-level fixed effects.

As governors, legislatures, and pension boards must work in concert to implement funding increases, we control for financial conditions as well as political environment. We start with fixed effects reflecting the calendar quarter and calendar year in which each observation’s fiscal year ends, to capture macroeconomic effects.²⁵ We control for plan-specific drivers of contribution policy: Funding Ratio (Prior Year), as poorly funded plans may face pressures to contribute more; and % Required Contributions Paid, to control for historical funding policies and governments’ commitment to funding. The indicator Plan Covers Teachers captures the higher benefits teachers accrue by virtue of their longer tenure and higher earnings (Munnell et al. 2008a).

Accrued Liability / Tax Revenue (Unfunded Accrued Liability / Tax Revenue) bridges plan-level and state-level controls by measuring the pension liability (unfunded pension liability) relative to the state’s ability to generate taxes; both are key measures of how economically significant the pension commitment is to that state. While states might contribute more to larger (and hence more visible) plans, state resources are also likely to be constrained when servicing larger plans.

At the state level, we control for the state’s financial condition along many dimensions. We measure state financial constraints with the per capita difference between final expenditures and final revenues in the general fund (Exp_Minus_Rev, Naughton et al. 2015). If Exp_Minus_Rev is positive (negative), the state is running a deficit (surplus). In contrast to this “flow” measure, TTLBAL Ratio (the ratio of general fund balances plus budget stabilization or “rainy day” fund balances to expenditures in the general fund) serves as a “stock” measure of financial constraints. Governments with low TTLBAL Ratio have a smaller cushion of reserves to draw upon when faced with an unexpected demand for resources. Per Capita Total Debt Issued captures the state’s reliance on the debt market. Per Capita Growth in Gross State Product captures the overall performance of the state economy; states contribute more in times of prosperity.

To control for state political environment, we include a Balanced Budget index; higher values flag states with more rigorous balanced budget provisions, which likely have less discretion to increase contributions. Public-sector Union Membership is a key aspect of the state’s political environment, as powerful unions pressure states for more generous benefits and to contribute more (Bonsall et al. 2019); as is Election, set to one if the state has a gubernatorial election in the upcoming year, to control for governors’ incentives to report favorable pension health. Corruption reflects federal corruption convictions of state public officials.

Our final set of controls captures legal protection of pension promises, which varies greatly across states. At one end of the spectrum are states that have explicit language in their constitution prohibiting benefit reductions for current employees (sometimes for future accruals as well), providing the strongest protection. At the other end are states that treat pensions as a “gratuity” from the employer, amounting to virtually no legal protection.²⁶ We incorporate Constitution and Gratuity indicators to flag states with the strongest protection (constitutional) and the weakest protection (gratuity approach) (Munnell and Quinby 2012). Table 1 defines all variables.

We further modify our baseline model (Eq. 1) to incorporate plan-level fixed effects (FEs). Plan-level FEs rule out alternative explanations attributable to time-invariant, unobservable factors specific to each plan—and many factors specific to the state as well, as most states have only one or two plans in our sample—thus moving us closer to causal inference. This plan-FE model (Eq. 1a) only incorporates controls that vary within plans over time, as constant features (Plan Covers Teachers, Balanced Budget, Constitution, and Gratuity) are absorbed by plan FEs.

$$\begin{aligned}{Contributions}&=\beta_0+\beta_1\;POST\hbox{-}GASB67+\beta_2\;Funding\;Ratio\;(Prior\hbox{-}Year)\\&+\beta_3\;\%\;Required\;Contributions\;Paid+\beta_4\;Accrued\;Liability\;/Tax\;Revenue\\&+\beta_5\;Unfunded\;Accrued\;Liability\;/\;Tax\;Revenue+\beta_6\;Exp\_Minus\_Rev\\&+\beta_7\;TTLBAL\;Ratio+\beta_8\;Per\;Capita\;Total\;Debt\;Issued\\&+\beta_9\;Per\;Capita\;Growth\;in\;Gross\;State\;Product\\&+\beta_{10}\;Union\;Membership+\beta_{11}\;Election+\beta_{12}\;Corruption+\;Plan\;Fixed\;Effects\end{aligned}$$

(1a)

4 Empirical results on pension contributions

4.1 Descriptive statistics of key variables

In Table 2, Panel A (B) we describe the variables used in multivariate tests for the two event years prior to GASB 67 (two event years after GASB 67). All variables are winsorized at the 1^st and 99^th percentiles. We define fiscal years using the cutoff for the effective date of GASB 67, which is effective for plan-years beginning after June 15, 2013. Hence, GASB 67 is first applicable to plan-years ending on June 15, 2014, to June 14, 2015, which we denote as year t0. Next, we define fiscal year t−1 as plan-years ending on June 15, 2013, to June 14, 2014, and so on. In our tests, we use two years immediately preceding GASB 67 (i.e., event years t−2 and t−1) and two years immediately following GASB 67 (i.e., event years t0 and t + 1), to allow sufficient time for contributions to evolve in response while also limiting potential confounding effects from unrelated events introduced by a longer time series.

Table 2 Descriptive Statistics

Panels A-B show that public pension plans are economically significant, with mean (median) assets of $21.8 ($11.2) billion in the pre-period and $23.6 ($12.9) billion in the post-period. Pension liabilities are predictably larger, with mean (median) liabilities of $30.2 ($16.8) billion in the pre-period and $32.3 ($18.1) billion in the post-period.The means being close to double the medians indicates some very large plans in our sample. Mean and median (lagged) funding ratios start and remain at 73% of the liability; we have no prior on how funding ratios might evolve, as real actions could counteract GASB 67/68’s increase to liabilities. In both periods, though, funding ratios vary substantially, with an interquartile range of about 60%-83%.

Moving to state finances, the overall fiscal condition of states appears to have improved slightly over the pre-period to post-period, with slightly higher surpluses, reserves, and GSP growth. 11% of public-sector employees are represented by unions. 16% (10%) of the sample is from states offering constitutional protection for benefits (treating benefits as a gratuity).

Employee contributions remain steady, at around a mean (median) of 6% (7%) of covered payroll; employer contributions increase from 14% (12%) to 15% (13%); contributions from employers and states increase from 15% (13%) to 17% (14%). Overall, mean (median) total contributions increase by about 3% of covered payroll, from 22% (20%) to 25% (23%), with much of the increase seemingly from non-employee sources, i.e., participating employers and state governments. Covered payroll remains steady across periods, helping to rule out the alternative explanation that our scaled contribution variables increase due to reductions in the scalar. Graphing mean total contributions from year t = −6 to + 2 (where year 0 reflects the first year of GASB 67 implementation and the first year of our post-period, Fig. 1) to examine any secular trends, we note that total contributions (i) are relatively flat in early years, (ii) start rising in our pre-period, and (iii) exhibit a sharp increase around GASB 67 implementation, consistent with GASB 67/68 triggering a shift in funding behavior. We turn to multivariate tests next.

Fig. 1

Means of Total Contribution Scaled by Prior-Year Payroll for Each Year Relative to GASB 67. This figure presents the mean values of total contributions scaled by prior-year payroll, for each year relative to GASB 67. Year 0 is the first year that GASB 67 is effective. In our regressions, observations in years −2 and −1 comprise the pre-period, and observations in years 0 and 1 comprise the post-period. Years −6 to −3 and Year 2 are not used in the regressions

4.2 How do contributions increase after GASB 67? Multivariate evidence from Eq. (1)

Table 3, Panel A presents results of estimating Eq. (1) with contributions from the key sources—employee, employer, and total—as dependent variables in Columns 1, 2, and 3 respectively; Panel B presents Eq. (1a) with plan-FE. While Panel A describes contribution behavior, we rely on Panel B to draw tighter inferences on the effect of GASB 67/68.

Table 3 Regressions of Contributions around GASB 67

While no plan- or state-level factors are significant for employee contributions, many are significant in the employer and total contributions models in Panel A, indicating that employer and state contributions are more responsive to fiscal conditions (with the models explaining a healthy 39.5% and 57.4% of variation, respectively). This is intuitive, given that mandated increases in employee contributions are tantamount to a benefit cut and take time and effort to negotiate. Plans with weaker prior-year funding ratios receive more contributions from employers and in total, consistent with stronger pressures to fund. % Required Contributions Paid is positive and strongly significant for employer and total contributions, indicating persistence in funding practices. When the deficit is larger relative to state tax revenues (Unfunded Accrued Liability / Tax Revenue), contributions are lower, possibly indicating a lower ability of the state to fund.

With state-level controls, TTLBAL Ratio shows an interesting pattern across sources: stronger state reserves associate with lower employer contributions but (weakly) higher total contributions, consistent with state governments contributing more and requiring less from participating employers when the state has a stronger fiscal cushion. Governments with lower debt issuances also contribute more, suggesting that high-issuance states may be financially constrained. More rigorous Balanced Budget provisions associate with lower total contributions, possibly due to the reduced discretion that states have in increasing contributions in those contexts.

We find striking effects on the legal protection indicators. For states offering the weakest protection (gratuity), employer (total) contributions are lower by 7% (5%) of covered payroll. In contrast, states offering constitutional protection for benefits have higher total contributions to the tune of 4.1% of covered payroll, although this is marginally significant.

Our chief variable, POST-GASB67, is insignificant in Column 1 (employee contributions) but positive and strongly significant in Column 2 (employer contributions) and Column 3 (total contributions), broadly consistent with univariate patterns. There is no evidence, hence, of increased employee contributions (tantamount to a benefit cut from employees’ perspective); rather, the increased funding burden is placed on participating employers and the state.

Coefficient estimates in Column 2 (3) indicate an increase in employer (total) contributions of 3% (3.6%) of covered payroll in the two years following GASB 67 implementation relative to the two years prior, after controlling for plan- and state-level determinants of contributions. The shift in employer (total) contributions is equivalent to an increase of 25% (18%) relative to the median and 33% (40%) relative to the standard deviation of pre-period values of those contributions. These shifts hence appear economically significant.

4.3 Do contributions increase after GASB 67? Multivariate evidence from plan-FE models

Moving to Table 3, Panel B incorporating plan-level FEs, fewer explanatory variables are significant in Columns 2 and 3 of Panel B compared to Panel A—unsurprising, as Panel B only uses within-plan variation for a relatively short panel of four years. However, the overall explanatory power of these models jumps to a striking 88.8% and 92.8%, respectively. This allows us to draw causal inference on the impact of GASB 67/68 more confidently, as the controls along with plan FEs capture most of the variation, lowering concerns about omitted variables. In Columns 2 and 3, we continue to observe positive coefficients on POST-GASB67 that are lower in magnitude than in Panel A, but statistically significant. Employer contributions increase by 1.3% in Column 2 (p-value = 6.1%), and total contributions increase by 2.5% in Column 3 (p < 5%).²⁷ Therefore, we conclude that GASB 67 triggers a significant increase in total contributions.

Among control variables, total contributions increase as the pension liability becomes more economically significant to the state, and Election has an interesting coefficient: total contributions are lower to the tune of 1.1% of covered payroll ahead of gubernatorial elections. While running counter to the intuition that governors wish to convey a better picture of pension funding ahead of elections, this pattern could first signal the competing and complex fiscal pressures to balance budgets, reduce borrowings, and improve spending on public welfare and services ahead of elections, all of which could lead to kicking pension contributions farther down the road. Second, if accrual manipulation is pervasively used to convey an inflated picture of reported funding status (e.g., Kido et al. 2012), that could in turn necessitate lower actual contributions.

4.4 Cross-sectional variation in the post-GASB 67 shift in contributions

We explore cross-sectional partitions to more sharply identify GASB 67/68’s impact. We expect any response to be stronger from governments for which the financial statement impact of GASB 67/68 is larger (H2) and for which the recognition of pension liabilities engenders more adverse economic consequences (H3) and more adverse political consequences (H4-H5).

We first explore graphical trends, partitioning Fig. 1 along proxies for each of these constructs. First, partitioning by funding ratio to capture the financial statement impact (Fig. 2, H2) yields a clear separation: a noticeable spike in contributions for the poorly funded plans, compared to an almost flat line for well-funded plans.²⁸ The spike around GASB 67 is also more pronounced for governments with high debt issuances (Fig. 3, H3) and strong unions (Fig. 4, H4). For Figs. 2 and 3, the two subsamples run very close for the entire pre-period graphed, with sharp movements only around GASB 67 implementation. For H5 on elections, we graph in Fig. 5 the change in total contributions using elections to define event time (as gubernatorial elections occur every four years for each state), with the two lines representing the pre- and post-GASB 67 periods respectively. In Fig. 5, we observe a sharp spike in contributions around elections, but only for the post-GASB 67 period. Armed with these descriptive trends that are broadly consistent with our expectations, in Tables 4 and 5 we present the results of estimating Eq. (1a) on total contributions, augmented with interactions of POST-GASB 67 with proxies for each of these constructs.

Fig. 2

Means of Total Contribution Scaled by Prior-Year Payroll for Each Year Relative to GASB 67 Partitioned by Low Versus High Funding Ratio. This figure presents the mean values of total contributions scaled by prior-year payroll, for each year relative to GASB 67. Year 0 is the first year that GASB 67 is effective. In our regressions, observations in years −2 and −1 comprise the pre-period, and observations in years 0 and 1 comprise the post-period. Years −6 to −3 and Year 2 are not used in the regressions. The red (blue) line represents observations with below-median (above-median) values of the prior-year funding ratio. The process of separating observations by the median is described in Table 4

Fig. 3

Means of Total Contribution Scaled by Prior-Year Payroll for Each Year Relative to GASB 67. Partitioned by Low Versus High Debt Issued. This figure presents the mean values of total contributions scaled by prior-year payroll, for each year relative to GASB 67. Year 0 is the first year that GASB 67 is effective. In our regressions, observations in years −2 and −1 comprise the pre-period, and observations in years 0 and 1 comprise the post-period. Years −6 to −3 and Year 2 are not used in the regressions. The red (blue) line represents observations with above-median (below-median) values of total (i.e., short-term and long-term) debt issued in the upcoming year. The process of separating observations by the median is described in Table 4

Fig. 4

Means of Total Contribution Scaled by Prior-Year Payroll for Each Year Relative to GASB 67. Partitioned by Low Versus High Union Membership. This figure presents the mean values of total contributions scaled by prior-year payroll, for each year relative to GASB 67. Year 0 is the first year that GASB 67 is effective. In our regressions, observations in years −2 and −1 comprise the pre-period, and observations in years 0 and 1 comprise the post-period. Years −6 to −3 and Year 2 are not used in the regressions. The red (blue) line represents observations with above-median (below-median) values of union membership in the state. The process of separating observations by the median is described in Table 5

Table 4 Regressions of Total Contributions around GASB 67: Interactions of POST-GASB 67 and the Expected Magnitude of Effect on Financial Statements, and Economic Consequences Thereof

Table 5 Regressions of Total Contributions around GASB 67: Interactions of POST-GASB 67 and Political Consequences

Fig. 5

Means of Change in Total Contribution Scaled by Prior-Year Payroll for Each Year Relative to a Gubernatorial Election. Partitioned by Pre-GASB 67 Versus Post-GASB 67. This figure presents the mean values of the change in total contributions scaled by prior-year payroll, for each year relative to a gubernatorial election. The x-axis includes only four years, because elections occur once every four years for every state. The red (blue) line represents observations in the post-GASB 67 (pre-GASB 67) period. Our regressions only include years −2, −1, 0, and 1. However, to be consistent with the range of years presented in Figs. 1, 2, 3 and 4, this figure includes all observations in the years −6 to −1 to compute the means presented in the blue line for the pre-GASB 67 period, and includes observations in the years 0 to 2 to compute the means presented in the red line for the post-GASB 67 period

Table 4, Column 1 includes POST-GASB 67*Low Funding Ratio; poorer funded plans experience a larger impact from GASB 67’s provisions (H2). While the interaction term is positive (as expected), it is not statistically significant.²⁹ Column 2 includes the interaction POST-GASB 67* High Debt Issued (reflecting above-median per-capita total debt issued). We expect governments that are reliant on the debt market to experience greater economic consequences from GASB 67/68’s balance sheet shock, and to respond accordingly. This interaction, however, is negative and significant: governments with high upcoming debt issuances increase contributions less following GASB 67/68, compared to those without. While this runs counter to our expectations as well as to the univariate trend in Fig. 3—that governments motivated to maintain credit ratings would have stronger incentives to shore up pension funding—we acknowledge the countervailing forces that could yield this finding in a multivariate test: first, that governments making large debt issuances are likely to be facing fiscal pressure and hence have lower ability to shore up pension funding; and second, that when faced with the threat of ratings downgrades, governments might respond in other ways, e.g., delaying pension contributions (either actively or by manipulating actuarial assumptions to reduce required contributions) or cutting benefits.

In Table 5, we present interactions on the political environment: public-sector union strength (H4, Column 1) and upcoming gubernatorial elections (H5, Columns 2 and 3). Consistent with expectation, the contribution response is higher from states with powerful unions (which would serve as monitors of pensions). Also consistent with our expectations, the contribution response to GASB 67 is higher when elections are upcoming (Column 2), with a larger and more significant effect in elections where the incumbent governor is running for re-election (Column 3).

Two points on the election findings are noteworthy: first, the progression from Column 2 to Column 3 is intuitive, as we would expect incumbent governors to have stronger incentives and ability to shift pension funding to influence election outcomes. Second, the base coefficient on Election remains negative and significant, as in Table 3—so, on average, pension contributions are lower ahead of elections, either due to competing fiscal priorities or because accrual manipulation can be used to lower the transparency of reported funding. Importantly, however, GASB 67/68 reverse this relation to some extent, with a positive and significant coefficient on the POST-GASB 67*Election term; we infer from this that GASB 67/68 made pensions a more important fiscal priority than they were previously, made pension funding more visible and hence harder to manipulate, or both. Both forces are consistent with pension funding becoming more salient to stakeholders post GASB 67/68, with funding pressures increasing as a result (Fig. 5).³⁰

While the results in Table 4 are mixed, the results in Table 5 convey a clearer picture of political consequences as a channel through which GASB 67/68 create real effects. Both factors that we examine in Table 5—union strength and upcoming elections—capture public monitoring and scrutiny of pensions. Labor unions have a direct interest in benefit security, and powerful unions have been shown to influence funding policy; elections represent a time when government performance, in general, is being closely scrutinized (by the electorate generally, and by public-sector unions, an important constituency). That we observe stronger contribution responses in these contexts suggests that those responses are motivated by the desire to minimize political consequences from large pension deficits being made prominent by GASB 67/68.³¹

5 Other real actions in response to GASB 67/68: an exploration of benefit cuts

While boosting contributions is a key “real” action that can mitigate the impact of GASB 67/68, governments could also respond with a very different action, one that could leave participants worse off: reduce benefits. Moreover, these two types of responses (increasing funding versus cutting benefits) could interact. In one scenario, these actions could function essentially as substitutes—some states may be more inclined to reduce benefits, while others may be more inclined to improve funding. In another scenario, these actions could coexist—governments that are serious about addressing pension deficits renegotiate benefits while also increasing contributions from various sources (employees, employers, and state) in an attempt to share the burden across stakeholders. We now attempt to provide some evidence on these questions.

Benefit cuts are inherently more difficult to model than contributions. While contributions are made every year and have well-understood determinants, benefit cuts are more sporadic events. Therefore, we lack an understanding of why (and when) some governments find themselves at a tipping point when current benefit levels are deemed unsustainable. Benefits can also be reduced in myriad ways (e.g., requiring employees to contribute more or work longer for the same benefit, applying a more conservative accrual formula, reducing COLAs) and for many classes of employees (e.g., new hires, active employees, retirees). We lack the disclosures necessary to estimate the actual economic significance of each change on expected levels of benefits or funding. We make many research design choices as a result, so our analysis is exploratory and descriptive.

We model benefit cuts for the 400 plan-years in our sample, for the 2012–2015 period, with the following logistic regression model, presented in Table 6:

$$\begin{aligned}{Benefit\;Cut\;Indicator}&=\beta_0+\beta_1\;POST\hbox{-}GASB67+\beta_2\;Funding\;Ratio\;(Prior\hbox{-}Year)\\&+\beta_3\;\%Required\;Contributions\;Paid+\beta_4\;Plan\;Covers\;Teachers\\&+\beta_{5\;}Accrued\;Liability/Tax\;Revenue+\beta_6\;Unfunded\;Accrued\;Liability/Tax\;Revenue\\&+\beta_7\;Exp\_Minus\_Rev+\beta_8\;TTLBALRatio+\beta_9\;Per\;Capita\;Total\;Debt\;Issued\\&+\beta_{10}\;Per\;Capita\;Growth\;in\;Gross\;State\;Product+\beta_{11}\;Balanced\;Budget\\&+\beta_{12}\;Union\;Membership+\beta_{13}\;Election+\beta_{14}\;Corruption\\&+\beta_{15}\;Constitution+\beta_{16}\;Gratuity+\beta_{17}\;High\;Contribution\\&+\beta_{18}\;POST\text{-}GASB67\ast High\;Contribution+Year\hbox{-}Quarter\;Fixed\;Effects\end{aligned}$$

(2)

Table 6 Determinants of Benefit Cuts around GASB 67

We compile plan-year-level data on benefit cuts from briefs issued by the National Association of State Retirement Administrators (Brainard and Brown 2016, 2018). Our dependent variables are indicators set to one once each type of benefit cut is enacted and affecting current plan members (Columns 1–7), and a combined indicator set to one once any type of benefit cut is enacted (Column 8). We are especially interested in how benefit cuts interact with contribution increases. Accordingly, High Contribution indicates plans with an above-median change in total contributions upon GASB 67 implementation. This is intended to flag plans that were particularly responsive to GASB 67/68 with increased contributions. POST-GASB67*High Contribution gauges whether the contribution increases complement or substitute for benefit cuts.³² Appendix 3 details the benefit cuts.

We interpret the determinants that are consistently significant across multiple types of benefit cuts. Plans contributing more of the ARC are less likely to cut benefits, presumably having a lesser need to. Benefit cuts are more likely when the plan deficit is economically substantial relative to the state’s tax revenues, and less likely in states with a larger cushion of reserves—indicating that state fiscal position plays an important role. The political environment also matters: plans with stronger unions are, curiously, more likely to cut benefits; this could reflect that benefits tend to be more generous in the first place when unions are powerful. Reductions in COLAs, a particularly controversial move in many states, are less likely ahead of elections.

Legal protection measures also show interesting coefficients. First, Constitution states are less likely to change benefit formulae or increase employee contributions, but more likely to reduce COLAs—perhaps because constitutional provisions are more likely to protect benefits already accrued, while COLAs pertain to the future and may not be considered guaranteed. On the other side, while we might expect Gratuity states to cut benefits, they are, surprisingly, less likely to do so. Correspondingly, when we assess how legal protection interacts with contribution responses (untabulated), we find that Gratuity states also increase contributions significantly less. Collectively, gratuity states are less likely to enact any type of change, be it cutting benefits or increasing funding; pension reform or funding may simply not be a priority for these governments.

POST-GASB 67 is positive and significant in four of seven columns and in the combined analysis of all benefit cuts—in general, plans are more likely to enact benefit cuts following GASB 67/68 implementation. GASB 67/68, therefore, appear to trigger both funding increases and benefit cuts.³³ Interestingly, mandates for increased employee contributions become significantly more likely (Column 6), although these mandates may be phased in such that there is no discernible increase in Table 3, Column 1. Even more important, POST-GASB 67*High Contribution is negative and significant in three of seven columns and for the combined analysis—plans increasing contributions post GASB 67/68 are significantly less likely to also cut benefits. Some plans, therefore, respond by increasing contributions, while others respond by cutting benefits, suggesting a substitution relationship between these two responses.³⁴

6 Discussion and concluding remarks

Two GASB pronouncements (Statement No. 67 and No. 68) issued in June 2012 dramatically alter the reporting landscape for DB pensions sponsored by US state and local governments. Together, they require pension liabilities to be estimated using potentially lower discount rates (hence increasing the estimated liability and funding deficits) and require pension surpluses (deficits) to be recognized as assets (liabilities) on government balance sheets.

With a sample of the largest state-administered DB plans, we find that contributions into plans from employer and state sources increase significantly in the two years following initial implementation of these rules. As higher contributions reduce the pension liability that is required to be recognized on state governments’ balance sheets, this response is consistent with governments acting to mitigate the adverse financial statement impact from GASB 67/68. Furthermore, the contribution increase is driven by governments facing stronger political monitoring and scrutiny, through labor unions and ahead of gubernatorial elections.

GASB 67/68 stand in contrast to other US pension regulation that was enacted specifically to alter contribution behavior, such as the Moving Ahead for Progress for the 21^st Century Act (MAP-21). As MAP-21’s pension provisions were designed to alter contributions, it is not surprising that funding practices shifted (Dambra 2018). In contrast, the GASB was notably agnostic about attempting to alter pension funding behavior when issuing GASB 67/68; it went so far as to state that its goal was to separate pension accounting from pension funding. Still, we document systematic increases in pension contributions post GASB 67/68, and our cross-sectional results suggest that this increase is related to those entities’ motivations to soften adverse consequences of reporting large deficits. These “real effects” highlight the importance of considering incentives that may lead to unintended consequences of accounting regulation.

Increasing contributions reduces the funding deficit directly as well as indirectly—directly by increasing the fair value of pension assets, and indirectly by reducing the estimated pension liability under GASB 68. Increasing current contributions allows plans to justify a larger estimate of future contributions, hence delaying any crossover point and reducing the likelihood of having to use a blended rate. To what extent the latter motivation drives the behavior we document is still unclear. Allowing plans to take credit for expected future contributions was highly controversial during the exposure draft process and remains so post implementation. As the AICPA observed in 2016, “Many significantly underfunded pension plans (i.e., 50 to 60% funded) are using the long‐term rate of return for discounting the entire liability based on the adoption of a law or policy that requires employer governments to increase contributions in future periods so that eventually they will be contributing the full actuarially determined contribution amount. There is some question whether the participating employers will make such contributions in the future.”³⁵ Our findings are relevant in two ways here: first, that we document an increase in contributions post GASB 67/68 adds reassurance that not all governments respond with mandates for future funding increases that may or may not come to fruition; at least some back that up with higher current contributions. Second, whether these increases persist into future years or not—as time will tell—may help to disentangle the motivations behind the response we document. Further research comparing actual contributions made under GASB 67/68 over time to projected contributions used in the estimation can also assist with understanding how much discretion is available to sponsors in practice.

A broader issue is the extent to which governments respond to GASB 67/68 by altering actuarial assumptions used to estimate pension obligations—a process now made potentially more discretionary by the complex procedure necessary to estimate crossover points. Plans adopting a “formal funding policy” have greater discretion in estimating future contributions, which begs the question: Do more plans adopt a formal funding policy? Is “accrual” manipulation more pervasive now that funding deficits (or surpluses) are on balance sheets? How do accrual and real actions interact? How will that mix change with fiscal pressure from the COVID-19 pandemic, and will accrual actions become more pervasive?³⁶ These are interesting questions that arise from our study.

Finally, our findings have implications for the GASB’s ongoing deliberations. While GASB 67/68 now require the full funding deficit (or surplus) to be recognized on governmentwide, accrual-basis financial statements, the governmental fund financial statements—including general fund statements (which most stakeholders focus on)—still follow a modified accrual basis, where pension expenses are reported only to the extent payable this period. In GASB’s Financial Reporting Model Comprehensive Project, currently in the post-comment period, the most frequently mentioned accounting topic across nearly 300 comment letters was retirement benefits, with respondents arguing that fund statements conveyed an incomplete and often misleading picture of retirement liabilities and hence of the state’s financial health.³⁷ This amply highlights how important pension accounting is to the broader discussion on improving governmental reporting.

Moreover, government decision-making happens largely at the fund level, where spending priorities are reflected through the budgeting process. Our finding that a change to government-wide financial statements elicits real responses is even more striking, when considered in view of the belief that fund statements play the driving role in government decision-making. Furthermore, requiring a more complete, accrual-based recognition of retirement obligations in fund statements could have even more far-reaching real effects on the provision and funding of these benefits.

Appendices

Appendix 1 – Estimation of the discount rate under GASB 67/68

• Step 1: Project future benefit payments based on the benefit terms as of the fiscal year end.

  • Benefits should be projected for all current plan members, whether active/inactive, currently receiving benefits/not

  • Projected benefits should include all automatic/substantively automatic COLAs, future increases in salary, and future service

  • Projection to be done on “closed group” rather than “open group” basis, i.e., benefits expected to be paid to future employees should be excluded from the benefit projections

• Step 2: Project plan assets

  • Assets are to be projected taking into account expected inflows (contributions) and outflows (benefit payments, expenses) associated with current members

  • Unlike benefit payment projections, expected contributions for future members can be included to the extent to which these contributions exceed the expected service cost associated with these new members

  • Contributions can come from employers, employees, or non-employer funding source

  • If the plan’s contribution rate is set by statute, or there is a formal written funding policy in place (such that it is reasonable to assume that the contribution will continue to be made), then professional judgment can be used in projecting the most recent five years of contribution history into the future

  • If no statutory contribution basis / formal written policy exists, then the average contribution over the most recent five-year period is the maximum projected future contribution

• Step 3: Determine the single equivalent discount rate

  • If Step 1 and Step 2 demonstrate that assets are projected to be sufficient to cover benefit payments for all periods, then the long-term ERR on plan assets may be used to discount all benefit payments

  • If Step 1 and Step 2 indicate that a “projected depletion date” exists, the actuary must then calculate a blended discount rate, by (i) discounting all benefit payments expected to be paid up until the projected depletion date using the ERR, and (ii) discounting all benefits payments after that date with a municipal bond rate.

  • Solve for the single equivalent discount rate that, when applied to all the cash flows, produces the same total present value as the dual discount rate streams described above; this single equivalent discount rate ( “blended rate”) is used to calculate the total liability per GASB 67/68.

Alternative evaluations of sufficiency

• Instead of the above approach, which can be complex, both GASB 67 (paragraph 43) and GASB 68 (paragraph 29) allow for alternative evaluations of projected solvency if they can be reliably made.

• The standards prescribe no particular alternative method; it is left to professional judgment and so the determination of whether an alternative approach is warranted is up to actuary and auditor.

• For example: “From a practical standpoint, a good candidate for an alternative approach would be a plan that is relatively well-funded, where contributions are based on a conservative, actuarially-based funding policy. E.g., the plan actuary may be able to demonstrate to the auditor’s satisfaction that a plan that is 80% funded, with a solid track record of adhering to a funding policy based on contributing the normal cost + 20-year closed amortization of unfunded liabilities, is mathematically certain to remain solvent if certain assumptions are met” (Winningham 2014).

Appendix 2

Measurement and recognition changes required by GASB 67 and GASB 68

	GASB Statements No: 25 and 27	GASB Statements No: 67 and 68	Expected impact
Measurement of the Total Pension Liability (TPL)	1. Discount rate: Projected future benefit payments discounted at the long-term ERR on pension plan assets	1. Discount rate: If plan assets are projected to be insufficient to cover projected benefit payments for all periods, then a “projected depletion date” or “crossover point” exists. (i) all benefit payments expected to be paid up until the projected depletion date are to be discounted using the ERR on plan assets, and (ii) all benefits payments after that date are to be discounted using a high-quality municipal bond index rate The single equivalent discount rate that produces the same total present value as the two discounted streams from (i) and (ii) is the plan’s overall, “blended” discount rate.	Typically, ERRs are higher than municipal bond index rates as public pensions invest extensively in equities and alternative assets. Therefore, “blended” discount rates are expected to be lower than the ERR, leading to higher estimates of the TPL.
	2. Substantively automatic benefit changes are not required to be included in benefit projections.	2. When projecting future benefit payments, any “substantively automatic” changes in future benefit payments are also to be included. Substantively automatic” benefit changes are defined as those “ad hoc” changes that the employer determines based on its past practice and future expectations of granting the change that the item will be granted in the future. A common example is ad-hoc cost-of-living adjustments (COLAs). Some factors to be considered in determining whether an ad-hoc COLA should be deemed “substantively automatic”: (i) The plan’s historical pattern of granting (or denying) ad-hoc COLAs (ii) Consistency in the amount of the COLAs, or the amounts relative to a pre-determined inflation index (iii) Whether there is evidence indicating that the ad-hoc COLAs will not be granted in future years	The inclusion of ad-hoc COLAs that are deemed substantively automatic will lead to higher estimates of the TPL.
	3. Any one of six actuarial cost methods are permitted, of which the most common method is Entry Age Normal (EAN), followed by the Projected Unit Credit (PUC) method.	3. Only the EAN method is permitted.	Up until the point of retirement, the EAN method recognizes a larger accumulated pension obligation for active employees than the PUC method.
Measurement of plan assets, i.e., “Fiduciary Net Position” (FNP)	Plan assets may be measured with a market-related value, where changes in market value may be smoothed over a three-to-five year period.	Plan assets must be measured at fair value as of the measurement date.
Balance sheet recognition	Only the net pension obligation (NPO) is recognized as a liability, where the NPO = cumulative excess of required contributions over actual contributions	The Net Pension Liability (NPL) must be recognized as a liability, where the NPL = excess of the Total Pension Liability (TPL) over the Fiduciary Net Position (FNP)
Additional liabilities	Not applicable	Deferred inflows and outflows: (i) Liability gains/losses (ii) Assumption changes (iii) Asset gains/losses
Income statement recognition	Generally, pension expense is equal to the annual required contribution (ARC). The ARC = normal cost + payment to amortize the unfunded actuarial accrued liability in 30 years. The exact computation of pension expense depends on whether a net pension obligation (NPO) was recorded. The NPO, which was the balance sheet liability under GASB 25/27, was calculated as the cumulative difference between the annual pension cost and the employer’s contribution to the plan. If there is no NPO, then pension expense is equal to the ARC (i.e., normal cost). If there is an NPO, then pension expense = ARC + one year of interest on NPO + adjustment to NPO to remove amounts in ARC already in the NPO.	Pension expense = normal cost  + interest on NPL - Expected return on plan assets  ± Liability gain/loss (amortized)  ± Asset gain/loss (amortized over 5 yrs)  ± Plan changes (immediate recognition)  ± Assumption changes (amortized)
Funding policy	Many governments’ funding policies were defined based on the ARC, as defined by GASB 25/27.	No reference to funding policies; GASB 67/68 represent an explicit effort by the GASB to delink accounting and funding for governmental plans.

Appendix 3: Benefit cuts measures

We rely on issue briefs created by the National Association of State Retirement Administrators (NASRA) to create measures of pension benefit cuts (or “reforms,” as the issue briefs refer to them). The issue briefs are available at https://www.nasra.org/issuebriefs. We primarily use the briefs titled “Significant Reforms to State Retirement Systems,” issued in June 2016 and December 2018. We supplement these with briefs titled “State Hybrid Retirement Plans,” “Cost-of-Living Adjustments,” and “Employee Contributions to Public Pension Plans.”

We compile data on the following different types of benefit reforms applicable to current plan members (i.e., currently active employees, terminated vested employees, or retirees):

• (1) Has the annual retirement benefit been capped? e.g., to a certain percentage of final average salary, such as 80%. If so, then “Annual Benefit Capped” is set to one, and else to zero.

• (2) Has the annual benefit multiplier been reduced? i.e., the rate at which employees accrue benefits for every additional year of service, which is typically around 2% per annum. If so, then “Benefit Multiplier Reduced” is set to one, and else to zero.

• (3) Has the calculation of “final average salary” been changed such that the final average salary is likely to be lower? We observe this done in two ways: one, the period over which final average salary is determined is increased, most commonly from three to five years; two, a more restrictive definition of “salary” is introduced, which disallows, for example, overtime pay or bonuses or unusually large year-on-year increases during the determination period from being included, or explicitly caps pensionable compensation to a certain dollar amount. The latter change—i.e., restricting what can be included in salary during the determination period—is often introduced to mitigate concerns of pension spiking. If yes, then “Final Salary Reduced” is set to one, and else to zero.

• (4) Have the age-service requirements for retirement been made more onerous? This is usually done by increasing the minimum age or service requirements for normal or early retirement, by eliminating the option to take normal (unreduced) retirement at any age with a certain number of years of service, or by increasing the rate at which benefits are reduced for taking early retirement. If yes, then “Retirement Age Increased” is set to one, and else to zero.

• (5) Have cost-of-living adjustments (COLA) been reduced? We observe this being done in many ways: reducing the guaranteed COLA rate; delaying COLA onset for retirees; or, most commonly, making the COLAs a function of inflation and the plan’s asset returns and funding status (i.e., “risk-based” COLAs). If yes, then “COLA Reduced” is set to one, and else to zero.

• (6) Have mandated employee contribution rates been increased? This is done either by increasing the mandated rate, which ranges from 5 to 10% of covered payroll, or by instituting a shared-risk system, where employee contribution rates are a function of the plan’s asset returns and funded status. If yes, then “Employee Contributions Increased” is set to one, and else to zero.

• (7) Have other changes been made that reduce the expected pension benefit? For example, the benefit vesting period is increased, the interest rate applicable to the purchase of service credits is increased, or the interest rate paid to refunded member contributions is reduced. If yes, then “Other Benefit Reductions” is set to one, and else to zero.

• (8) If any of the above benefit cuts have been passed, then a joint indicator “Any Benefit Reductions” is set to one, and else to zero.

We tabulate the frequency of benefit cuts here:

	t−2	t−1	t0	t + 1
Annual Benefit Capped	2	0	2	0
Benefit Multiplier Reduced	8	1	1	2
Final Salary Reduced	9	5	0	4
Retirement Age Increased	14	6	2	5
COLA Reduced	5	12	2	2
Employee Contributions Increased	5	10	4	4
Other Benefit Reductions	2	4	1	2
Any Benefit Reductions	15	17	10	7