Despite a decade-long bull market between the financial crisis and the COVID-19 recession, state defined benefit pension plans had accrued more than $1.37 trillion in unfunded liabilities. However, little work has investigated the actuarial sources of these unfunded liabilities. This paper uses original data hand collected from publicly available financial reports between 2000 and 2020 for 145 state-administered pension plans to determine the sources of unfunded liabilities. The largest unfunded liability contributor is investment experiences (when actual investment returns do not match assumed returns). The second and third largest contributors are changes to actuarial assumptions and expected changes (or interest accruing on existing unfunded liabilities). Benefit experience and legislative changes, demographic experience, and explicit funding shortfalls account for relatively little of the growth in unfunded liabilities. Moreover, the specific sources of unfunded liabilities are heterogeneous over time and across plans.

This paper provides a methodological critique of an influential method for assessing the impact on the Unfunded Accrued Liabilities (UAL) of the gap between assumed and actual investment returns over extended periods, and offers a sound replacement. The method in question simply sums over time the components of the annual actuarial gain/loss report. This implicitly assumes that in the counterfactual exercise, the interest on the additional UAL is covered dollar-for-dollar by amortization. But under actual funding formulas amortization usually varies less than interest. This means there are large intertemporal interactions between the gap in investment returns and subsequent shortfalls between contributions and interest. Using the actual funding formula in the counterfactual can lead to much higher estimates of the UAL impact of the gap in investment returns because it does not assume away these interactions. This method can more accurately inform policy-makers, regarding the importance of cutting the assumed rate of return.