2.1 Calculating unexpected deficits

To estimate the impact of fiscal stress on pension contributions, I first calculate unexpected deficit shocks following Poterba (Reference Poterba1994). Unexpected deficit shocks measure the estimated gap between forecasted and actual budgets when adjusting for within-fiscal-year changes to spending (ΔSpend) and tax revenue (ΔTax) resulting from legislation enacted after the initial budget or during the fiscal year, respectively.Footnote ⁵ This measure of fiscal stress captures the difference between what legislators think the budget will look like and how it actually turns out. This method continues to serve as a workhorse in the literature (e.g., Clemens and Miron, Reference Clemens and Miron2012).

Let DeficitShock _it be the per capita unexpected deficit for state i in fiscal year t, where positive deficit shocks are deficits and negative shocks are surpluses. DeficitShock is calculated by subtracting RevenueShock from ExpenditureShock. In years of fiscal stress, positive expenditure shocks and negative revenue shocks both contribute to positive deficit shocks.

(1) $$ExpenditureShoc{k_{it}} = ActualExpenditur{e_{it}} - ExpectedExpenditur{e_{it}} - {\it \Delta} Spen{d_{it}}.$$

(2) $$RevenueShoc{k_{it}} = ActualRevenu{e_{it}} - ForecastRevenu{e_{it}} - {\it \Delta} Ta{x_{it}}. $$

(3) $$DeficitShoc{k_{it}} = ExpenditureShoc{k_{it}} - RevenueShoc{k_{it}}.$$

Deficit shocks are then separated into positive values (unexpected deficits) and negative values (unexpected surpluses). That is, if DeficitShock is positive, then UnexpDeficit equals DeficitShock and zero otherwise. Similarly, if DeficitShock is negative, then UnexpSurplus equals DeficitShock and zero otherwise.

2.2 Modeling pension contribution ratios and expected cuts

Pension contributions are measured by contribution ratios, which divide actual state government contributions by ARCs.Footnote ⁶ Undercontributions occur when pension contributions are less than ARCs, that is, the contribution ratio is less than 100%. To determine the effect of unexpected deficits on contribution ratios for state i in fiscal year t, I study linear models of the following form:

(4) $$\eqalign{{ContributionRati}{{o}_{it}} = & {a_0} + {a_1} \bullet {UnexpDefici}{{t}_{it}} + {a_2} \bullet {UnexpSurplu}{{s}_{it}} + {{\bi X}_{it}} \bullet \alpha \cr & + {{\bi A}_i} + {\varepsilon _{it}}.}$$

The coefficient of interest, a ₁, is negative if unexpected deficits cause undercontributions, and a ₂ is close to zero if unexpected surpluses do not cause extra contributions. Covariates ( X _it ) include a number of controls. Lagged total end-of-year balances (which includes ending balances and budget stabilization funds), control for higher expected contribution ratios due to available funds from previous fiscal years. Fractions of workers in a state that are public employees (federal, state, and local) or public union members control for possible pressure to fund pensions with explicit state tax dollars through higher contribution ratios, rather than employee contributions (which are deducted from wages). Lagged funding ratios control for the persistence of contribution ratios.Footnote ⁷ A tax limitations dummy variable, which is one when raising any state taxes requires a legislative supermajority and zero otherwise, controls for an institution that may cause states to lower contribution ratios due to revenue constraints. State fixed effects ( A _i ) control for other persistent state institutions that may affect contribution ratios, such as whether a state constitutionally protects pension benefits.

To estimate expected per capita pension contribution cuts – that is, the counterfactual cut in state contributions if they were cut proportionally with other spending in response to unexpected deficits – I first calculate expected contribution cuts for each year by multiplying a state's per capita budget cuts as a fraction of expenditures by its pension contribution. To estimate a more representative expected value, state contributions (StateContrib _i,−t ) come from years other than t, specifically, an average of contributions from years t − 1 and t + 1.

(5) $${ExpContributionCu}{{t}_{it}} = {BudgetCu}{{t}_{it}}/{ Expenditure}{{s}_{it}} \bullet StateContri{b_{i, - t}}.$$

Expected contribution cuts are then regressed on unexpected deficits and the same controls used in (4), except for lagged funding ratios.

(6) $$\eqalign{{ExpContributionCu}{{t}_{it}} = & {b_0} + {b_1} \bullet {UnexpDefici}{{t}_{it}} + {b_2} \bullet {UnexpSurplu}{{s}_{it}} + {{\bi X}_{it}} \bullet \beta \cr & + {{\bi B}_i} + {\varepsilon _{it}}.}$$

Next, I convert the units of a ₁ to dollars per capita (á ₁). If states cut pension contributions more than other spending in response to unexpected deficits then the ratio of á ₁ to b ₁ will be greater than one.

4.1 Cumulative effect of undercontributions on underfunding

How much pension underfunding has actually occurred due to unexpected deficits? The cumulative unexpected deficit for all states except AK and MA was about $170 billion between 1989 and 2008 (in 2008 dollars); meaning $5.20 of undercontributions per $100 of per capita unexpected deficits implies cumulative underfunding of $9 billion. But if the investments had been made in the past then they would have grown with investments. Using each state's average nominal pension returns on investment, Table 3 shows that unexpected deficits explain $16 billion of underfunding, or 4% of mid-2008 actuarial underfunding.Footnote ¹²

Table 3. Estimated cumulative state pension underfunding from positive unexpected deficit undercontributions (millions dollars)

Notes: Annual compounding and assumed $5.20 of undercontributions per $100 of unexpected deficit. US aggregate returns were shown but state specific returns based on US Census data were used in the calculation of cumulative underfunding, where five observations had rates of return top-coded at 30% and three missing values were replaced with average returns. Returns calculated by dividing state investment earnings by previous year assets. Before 2002, US Census reported assets were book values, rather than market, and investment earnings were only realized earnings, suggesting an upward bias to returns. AK and MA excluded.

Sources: PENDAT, Public Fund Survey, US Census, and NASBO.

States may underfund pensions for reasons other than temporary fiscal stress, such as structural issues making it difficult to increase revenues or cut expenditures. Total state undercontributions excluding New Jersey were estimated as in Table 3, where annual undercontributions were estimated with weighted ARCs and state contributions, as reported by the US Census.Footnote ¹³ ARCs for 1993, 1995, 1997, and 1999 were interpolated, and ARCs for 1989–91 were set to the 1992, 1994, and 1996 average. Note that these assumptions mean this is an imprecise estimate; moreover, this is likely an overestimate, as a fraction of these undercontributions represent amortization of previous undercontributions, causing some double-counting. I estimate total cumulative undercontributions between 1989 and mid-2008 of $125 billion, implying that about a quarter of total underfunding was due to state undercontributions for reasons other than fiscal stress [(125–16)/400 = 0.27].

4.2 Alternative measures of fiscal stress

The measure of fiscal stress used thus far, unexpected deficit shocks, may suffer from a number of flaws. Expected revenues and expenditures may be distorted by state budget officers.Footnote ¹⁴ For example, Boylan (Reference Boylan2008) shows that budget forecasts have an upward bias near elections. To control for potential manipulation of forecasts, I estimate how dampened tax revenues affect pension contributions by estimating trend tax shocks as in (2), but replacing the difference between actual and forecasted revenues with quadratic-year actual revenue trend residuals.Footnote ¹⁵ This is similar to Sobel and Holcombe (Reference Sobel and Holcombe1996) and Stansel and Mitchell (Reference Stansel and Mitchell2008). The results of regressions using this alternative measure of fiscal stress suggest similar effects on pension contributions. States cut pension contributions over five times more than other spending in response to positive trend tax shocks (i.e., deficits). Following the same procedure shown in Table 3, over the preceding two decades positive trend tax shocks caused $13 billion of underfunding, similar to the underfunding estimated due to unexpected deficits.

Another issue is possible endogeneity of pension contributions in expenditures. I calculate contribution neutral fiscal stress by subtracting actual pension contributions from actual expenditures and 2-year lagged contributions from expected expenditures. This measure of fiscal stress gives an identical result as the fixed effects model in Table 2, with a significant coefficient of −0.052.

4.3 Institutional differences between states

While all states have laws regarding fiduciary standards for state pensions, institutional differences between states may affect pension contributions. According to the United States General Accounting Office (1996, pp. 3–4):

States with contributions constrained by statute, e.g., a fixed percentage of tax receipts, may undercontribute both because of temporary declines in the tax base or persistently low statutory tax rates.Footnote ¹⁶ For example, the statutorily constrained Oklahoma state teachers’ pension had a contribution ratio of only 56% in 2005. The contribution ratio increased to 86% the next year as dedicated sources increased from 4.0 to 4.5% of state tax revenues and 5.0% of lottery proceeds began going to the pension (Oklahoma Office of State Finance, 2010, FY-2011 Executive Budget, Volume 1).

States with contributions not constrained by statute may use their flexibility to undercontribute more in years of fiscal stress. Similarly, states with annual legislatures may undercontribute more in reaction to unexpected deficits because they have more frequent opportunities to respond. To test these two hypotheses, I add interaction terms for statutory constraints (using state status from Munnell et al. Reference Munnell, Haverstick, Aubry and Golub-Sass2008) and annual state legislatures. For states that were either statutorily constrained or had annual state legislatures there were no significant impacts on contribution ratios.

Balanced budget requirements are another institutional difference between states that may affect how they contribute to pensions. As states with stronger balanced budget requirements have less budget flexibility, I expect a negative effect of balanced budget requirements on pension contributions.

I use an index measuring the strength of budget balance rules from Clemens and Miron (Reference Clemens and Miron2012), which considers if a state's governor must submit or the legislature must pass a balanced budget and limitations on carrying over deficits. A year fixed effects model is used for this analysis because balanced budget requirement measures are constant across the study for each state. The significant coefficient on balanced budget strength is −1.1 (Table A1). This suggests that stronger balanced budget requirements are correlated with pension undercontributions. Specifically, these requirements are correlated with a 10% point drop in contribution ratios among the half of states with the most stringent balanced budget requirements. This translates into almost $1 billion in annual undercontributions nationwide.

Year fixed effects results, shown in Table A1, look similar to the cross-sectional analysis of Mitchell and Smith (Reference Mitchell and Smith1994), who also found a persistence of funding behavior and that higher unionization of covered employees led to less state contributions. They suggest that this counterintuitive negative effect of unions may be ‘due to the upward pressure on salaries associated with collective bargaining, to which employers respond by reducing pension contributions.’ (p. 286) Note that the negative effect of unions on pension contribution ratios is more than offset by the positive effect of public employees, which is three times larger on average.

4.4 Robustness checks

A number of alternative models demonstrate the robustness of the regression results. As states pay their full ARC in two-thirds of fiscal years, this bunching of contribution ratios at 100% may lead to non-linear effects. A fractional logit model gives similar results. The average marginal effect for unexpected deficits is −0.038 and significant, while the coefficient for unexpected surpluses is near zero and insignificant (Table 4, column 1).

Table 4. Robustness checks: unexpected deficits and state pension contribution ratios

Notes: The dependent variable is the annual state contribution ratio, weighted by each pension's liabilities. For the fractional logit and annual trend models, the years in sample are 1992–2007, where missing years are 1993, 1995, 1997, 1999. AK and MA dropped. Linear models used for annual trend and 1992–2009. For fractional logit model, contribution ratios are divided by a hundred and topcoded so that they range between zero and one, and average marginal effects times a hundred are reported. Contribution ratio adjustments are described in the text. Errors clustered by state. $pc values are in per capita fiscal year 2010 dollars (CPI-U-RS).

Significance ***p<0.01, **p<0.05, *p<0.10.

Sources: PENDAT, Public Fund Survey, US Census, NASBO, and Union Membership and Coverage Database.

Contribution ratios tend to decline over the sample. To control for this, I add an annual trend that grows from 0 to 15 between 1992 and 2007. The coefficient on unexpected deficits (−0.050) is similar and the negative coefficient on the decade dummy suggests that all else equal, states lowered their contribution ratios about one percentage point each year over the sample. The fractional logit coefficient is unchanged when adding the annual trend.

So far, this study has only considered pension contributions through fiscal year 2007. Extending the data through 2009 attenuates the effect of unexpected deficits, with a coefficient for unexpected deficits of −0.043. The Recovery and Reinvestment Act of 2009 significantly boosted state government inflows, perhaps temporarily breaking the link between estimated unexpected deficits and true fiscal stress. Referring to the end of state fiscal year 2009, Cauchon (Reference Cauchon2009) writes that the ‘flood of federal money lifted total revenues by 7.5%, overcoming an 8% drop in tax collections.’

5.1 Growing liabilities and recent reforms

Liabilities have grown in part because of increasingly generous benefits. New promises often slowly accrue as actuarial liabilities over many years. If these liabilities are in excess of pension assets then this underfunding is amortized, usually over 30 years. This means increases in pension benefits can appear and persist as underfunding long after new promises are made. Johnson (Reference Johnson1997) finds that state and local pensions take advantage of this temptation, as pensions more able to shift costs to future taxpayers through underfunding are more likely to increase the relative generosity of promised benefits.

Increased benefits may be catalyzed by the cyclicality of investment returns. Instead of smoothing out returns over the long-run, temporary pension surpluses may be dispersed to public employees through reduced employee contributions or increased benefits (Peskin, Reference Peskin, Mitchell and Hustead2001; Bader and Gold, Reference Bader and Gold2007). This can lead to a ratchet effect, where during economic expansions states make benefits more generous, but during downturns these benefits persist and eventually lead to underfunded pensions.

Pension liabilities have also increased with changing actuarial assumptions. For example, Rhode Island recently changed its actuarial assumptions to reflect workers retiring earlier and living longer, resulting in an increase of $50 million in liabilities. But Rhode Island is one of many states trying to limit the growth of liabilities with a number of reforms: making workers wait until age 62 to collect benefits, reducing the maximum pension to 75% of average pay near retirement, and limiting annual cost-of-living adjustments (COLAs) for new retirees (Gregg, Reference Gregg2011). Snell (Reference Snell2012) gives many examples of recent reforms tightening eligibility conditions and reducing benefits. However, Novy-Marx and Rauh (Reference Novy-Marx and Rauh2011a ) estimate that even implementing extreme versions of some of these policy changes, such as the complete elimination of COLAs, would only eliminate half of the underfunding they estimate with Treasury discounting. Underfunding can also be addressed by raising employee contributions, which half of states have done since 2008 (U.S. Government Accountability Office, 2012).