An Empirical Analysis of State and Local Public Pension Plan Funded Ratio Change,2001-2009

Abstract

Between 2001 and 2009, all public pension plans suffered losses and saw a drop in their funded ratios. However, some plans saw a much smaller decline than others. In this study, we explore why the ratios fared so differently during this tumultuous period for pension plans. By examining the changes in funded ratio for 84 large public pension plans, we find that the differences can be mostly attributed to variations in annualized investment return and changes in investment return assumption, and to a lesser extent, to the required contributions paid by employers, the contribution rates of employees, and cost-of-living adjustment provisions. The results suggest that pension plans seeking to improve their funded ratios may need to revise their investment strategies, pay a higher percentage of their required contribution, require employees to pay more toward their pension benefit, and limit the use of automatic or consumer price index–linked cost-of-living adjustments.

Keywords

public pension plan funded ratio investment return annual required contribution COLA

Introduction

State and local public pension plans¹ have garnered increasing public attention as the stock market plummeted in 2000. Even though the market recovered to hit new highs after 2000, it crashed again amid the financial crisis of 2008-2009. These two severe drops in investment values within a 10-year period have wreaked havoc with the asset value of public pension plans due to their large asset allocation to stocks. As a direct impact of this drop in asset value, the funded ratio of most public pension plans (or the ratio of pension assets to pension liabilities) has declined. Just before the 2000 crash, the U.S. average funded ratio for all public pension plans was around 100%, meaning that there were just enough assets to cover all pension benefit obligations as of that year. By 2009, the average ratio had dropped to 78%, and the overall funding shortage in 2009 was US$660 billion (Pew Center on the States, 2011). This drop in funded ratio and the resulting large funding gap have placed a significant strain on state and local government budgets, as both governments and employees must contribute additional funds to reduce the gap. Public pension financing is thus a major component of the fiscal problems facing state and local governments, generating heated debates about the viability of public pension plans in the media, within the public sector, and among the general public.

While virtually all public pension plans saw their funded ratios drop between 2001 and 2009, observers often overlook the fact that not all plan ratios dropped by the same amount. Some pension plans experienced a much smaller drop than others, providing a unique opportunity for research on the factors that affect funded ratios. Up to now, most studies on what determines public pension funded ratios have examined ratios at the time of the study. In this study, we take a different approach. Instead of examining funded ratios only at a particular point in time, we examine the changes in funded ratio between 2001 and 2009. By looking at the changes that occurred over this relatively short time period, we can analyze the impact of common factors that affect all pension plans. By identifying these factors, we can contribute to the growing public discourse on public pension reform.

This article contains five sections. The first section provides a literature review on relevant pension research. The second section outlines the research objective and the hypotheses explored in the present study. The third section discusses the methodology and data collection. Empirical results are presented in the fourth section, and the article concludes by discussing policy implications.

Literature Review

Funded ratios of defined benefit (DB) pension plans have been widely studied in corporate contexts. Due to the profit-seeking nature of private-sector entities, however, many of these studies have focused on the relationship between funding level and profitability, taxation or stock market performance (Feldstein & Seligman, 1981; Franzoni & Marin, 2006; Tepper, 1981). There are several significant differences between private and public DB plans, of which the most important one is that private-sector DB plans are regulated by the federal Employee Retirement Income Security Act (ERISA), whereas public-sector DB pension plans are exempt from ERISA (Peng, 2009). As a result, literature on private pension plans is largely inapplicable to public pension plans. While there is a growing body of research on public pension plans, we limit our literature review in this section to some major studies on the determinants of public pension funding level that are directly relevant to this study.

As the funding level of a pension plan is determined by both pension assets and liabilities, it is then natural for researchers to examine factors that affect the value of assets and liabilities. With a few exceptions, these factors fall into three broad categories: factors affecting asset value, factors affecting liability value, and factors affecting both asset and liability value.

In this first category, while pension asset value depends on two major sources of income, contributions from both employer and employee, and investment return on these contributions, almost all the studies in this category focused on the employer contribution. They generally examine the factors affecting the pension contribution level, and the relationship between pension contribution level and pension funding level. Marks, Raman, and Wilson (1988) found that state and local government sponsors under tight financial constraints and political pressure will provide lower cash contributions to pension plans. Mitchell and Smith (1994) examined 42 large pension plans in 1989 to explore the determinants of pension funding in the public sector. The authors found that unionized employers are less likely to fully fund future pension obligations, and that funding is sensitive to fiscal pressures. Mitchell and Hsin (1997) found that pension systems are more likely to underfund their plans during times of fiscal stress. Chaney, Copley, and Stone (2002) examined the extent to which fiscal stresses and constitutional requirements to balance state budgets affect the funding of state public pension plans. The results indicated a negative relationship between pension funding levels and both factors; that is, both state fiscal stresses and the presence of balanced budget requirements (BBRs) were associated with lower levels of pension funding. Munnell, Haverstick, and Aubry (2008) sought to explain why some plans are less well funded than others by exploring four categories of factors: funding discipline, governance, plan characteristics, and the state’s fiscal health. For the funding discipline, one variable the authors examined was whether the employer has paid the full amount of the contribution. By analyzing the funded ratios for 126 plans in 2006, the authors found that paying the full amount increased the funded ratio by 6%.

In the second category, a few studies examined the relationship between the level of employee pension benefit (which is employer’s liability) and pension funding level. Johnson (1997) examined 102 state-level pension plans from 1983 and 1988, and found positive relationship among plan underfunded levels, individual pension benefit level, and taxpayer mobility. The author argued that the relative generosity of pensions among government workers is related to their ability to underfund public employee retirement plans. As underfunding can reduce tax burdens for residents expecting to leave the community before retirement benefits are paid, governments can offer employees generous but inadequately funded pensions. Munnell et al. (2008) also examined the effect of pension benefit level on funded ratio. They posited that teachers tend to have better pension benefits than other general government employees due to their long tenure and higher education level. They found that plans including teachers tend to have lower funded ratio.

In the third category, researchers examined another factor that has impact on both pension assets and pension liabilities value. This has to do with the pension fund’s discount rate,² which plays a critical role in determining the present value of future pension liability, employer contribution, and thus pension funding level. It is the most important economic assumption made in pension actuarial valuation. Empirical studies (Chaney et al., 2002; Giertz & Papke, 2007; Hess, 2005; Marks et al., 1988; Mitchell & Smith, 1994; Vermeer, Styles, & Patton, 2010) have found evidence that governments and retirement systems, under certain circumstances, adopt actuarial assumptions for purposes of reducing their annual required contribution (ARC) to a pension plan and/or for the purposes of raising its funded ratio. Both the reduction of ARC and the increase in funding level are politically advantageous, because pension contributions come out of the general fund. Such contributions compete with other government programs for limited resources, but they do not have the same immediacy and urgency. Stalebrink (2014) also provided empirical evidence that the adoption of investment return assumptions by DB plans may be partly explained by political opportunism.

In summary, in exploring the factors that affect the funding level of a pension plan, previous studies have primarily focused on the employer contribution (and the related fiscal stress), pension benefit level, and pension discount rate assumption. While all these factors are certainly important in this respect, one important variable, the actual investment performance measured by annual rate of return or annualized rate of return over a period of time, is missing from these studies. Investment return is a key factor in determining asset accumulation over a long period of time, and thus in determining the funded ratio. Any empirical model exploring determinants of pension funding level is incomplete if it lacks this variable. With the absence of this variable, the magnitude of other variables’ impact on funding level can be improperly identified. One potential reason this important variable has not been included in previous studies is because of the time period of the data used. Most previous empirical works on public pension plans have generally relied on a particular year of data. While there are good reasons for such an approach, results from a single year of data may have limited generalizability, because these models cannot capture the volatile nature of some important variables such as funded ratio, contribution percentage, and annual investment return, all of which vary considerably from year to year.

The present study aims to improve on the existing literature in several areas. First, while incorporating many variables identified in previous studies in determining the funding level, we will also directly incorporate the variable of investment performance into the model. In doing so, we want to not only build a more complete model in exploring the determinants of pension funding level but also understand the magnitude of each determinant’s impact on funding level. This latter objective is important because it can inform us as to what can be done policy wise to have the most impact on improving pension funding level. Second, unlike the approach adopted by most prior studies, we will use the change in funded ratio, rather than the funded ratio at a particular point in time, as our dependent variable. In our view, a dependent variable measuring change over time is a better measure for the purpose of this study, as it will allow us to incorporate some volatile variables, such as investment return, over a longer period of time. Last but not least, this study is based on a large-scale and updated nationwide data set of state government DB pension plans. To date, only a few studies (Lucas & Zeldes, 2009; Munnell et al., 2008; Stalebrink, 2014) have examined the determinants of funded ratio or investment return assumptions using nationwide state data drawn from pension plans’ comprehensive annual financial reports (CAFRs).

Research Design

A public employee retirement system (PERS) is a statutorily established government unit that serves as a funding and administrative instrumentality with one simple and clear goal: to accumulate sufficient assets to pay for current and future pension benefits. Unlike many complex governmental financial problems, the funding status of pension plans is relatively easy to understand, at least in theory. Public pension assets come from three sources: employee pension contributions, employer pension contribution, and investment return on these contributions. If the total of these three sources is greater than the benefit payment in a particular year, pension asset value will increase; otherwise it will decrease. While a growth in pension assets indicates that state and local governments have accumulated more assets to pay for pension liabilities, however, the increase itself does not indicate whether state and local pension plans have enough assets at any given time. What really determines the sufficiency of a pension fund is whether the pension assets accumulated are greater than the projected pension benefits to employees (or pension liabilities to employers) accumulated at a particular time. If pension assets are less than pension liabilities, the pension plan is underfunded. A plan’s adequacy of funding is typically described in terms of its funded ratio, calculated by dividing pension assets by pension liabilities. Peng (2009), who tracked changes in funded ratios from the 1990s to recent years, concluded that the rise in funded ratios in the 1990s was driven by the strong stock market and that the rapid declines in funded ratios since then have been caused by poor performance of financial markets, coupled with imprudent policies such as increasing benefits and reducing pension contributions.

To explore the problem of unfunded pension liability, we need to examine the three key factors that determine the funded ratio: contributions from employees and employers, pension benefits, and investment management. On the contribution side, because government sponsors have discretionary funding authority, they may either manipulate investment return assumptions to reduce the annual required pension contribution (Stalebrink, 2014) or simply delay or reduce their payments into the pension fund due to fiscal stress (Adler & Sacco, 1995; Clark, 1990; National Conference of State Legislatures [NCSL], 2001; Retkwa, 1990). On the benefit side, each pension plan has its own benefit design, which includes such elements as normal service benefits, early retirement benefits, postemployment benefit adjustments, purchase of service credits, and ancillary benefits. As with pension contributions, government sponsors tend to be more generous (i.e., increase benefits) during economic upturns and reduce them during economic downturns.

Of the three key factors, investment return is the most unpredictable due to the volatile nature of economic cycles and market conditions. Over time, investment return has become the most important source of pension funding. In 1985, the annual investment return surpassed the combined total of employer and employee contributions for the first time, and the gap between the two sources has continued to grow since then, except between 2001 and 2003. In 2011, total investment income of all U.S. public plans reached US$480 billion, more than three times the US$136 billion in pension contributions that year (U.S. Census Bureau, 2011a). As this trend continues into the future, due to the power of compounding, investment return will become more and more important in determining funded ratio.

The objective of this research is to explore the determinants of variations in the change of pension plans’ funded ratios. Based on the literature review and the explanation presented above, factors that determine changes in funded ratio can be conveniently categorized into four groups: investment performance, pension benefits, pension contributions, and other variables. This fourth, umbrella group includes indebtedness, budget situation, union coverage, and demographics of plan participants.

Investment Variables

The first hypothesis explores the impact of investment return on DB plans’ funding status. As noted above, investment return is by far the most important part of pension plan management, because over time investment income far outweighs pension contributions as a funding source for pension benefits. If the contribution level and the benefit structure of a pension plan are held constant, the funded ratio is almost solely determined by investment return. Existing studies, however, have either overlooked variations in investment return or assumed that investment return is largely determined by the market and should be nearly identical for all public pension plans, given that asset allocation behavior is highly similar across state and local pension plans (Lucas & Zeldes, 2009). In actuality, this supposed similarity of returns does not occur, at least not for the years from 2001 to 2009. In our sample, the annualized rate of return ranges from −0.4% to 6.2%, with the mean at 2.4%. Clearly, pension plans differ dramatically in investment performance, most likely because of differences in their investment policies, the quality of asset management, the presence or absence of an investment council, and other factors. As mentioned earlier, omitting the investment performance variable will cause the regression model to be biased. As all pension plans reviewed in this study saw a decrease in their funded ratio during this period, a higher investment return should lead to a smaller decrease in funded ratio, if all other factors are equal. Given this fact, the following hypothesis is proposed:

Hypothesis 1: The investment return, measured by the annualized rate of return over the event window, is negatively correlated with a decrease in funded ratio.

The second factor in the category of investment returns is the change in assumptions regarding the real rate of return, which is calculated as the difference between assumed rate of return and assumed inflation. Real (or net relative to inflation) rate-of-return assumptions are a better indicator than gross rate-of-return assumptions, because two plans with the same assumed rate of return may have very different inflation assumptions. The one with a higher inflation assumption is the more conservative one. A higher real rate-of-return assumption is more difficult to fulfill and thus results in a larger funding gap, and the present value of liabilities also grows at a slightly quicker pace. Likewise, a change in the rate-of-return assumption will also cause the funded ratio to change, if everything else is held constant. Therefore, we propose the next hypothesis:

Hypothesis 2: A decrease in real rate-of-return assumption is positively correlated with a decrease in funded ratio, and an increase in real rate-of-return assumption is negatively correlated with a decrease in funded ratio.

Pension Benefit Variables

In general, the higher a pension fund’s benefits level, the more pension assets are needed to pay for the benefits and therefore the lower the funded ratio will be, if other factors remain constant. Among the three elements that determine the normal service benefit—final average salary, years of credited service, and benefit multiplier—the last one has the most impact, because a slightly higher multiplier can lead to a significantly higher retirement benefit, as it applies to every year of service. Similarly, a more generous postemployment cost-of-living adjustment (COLA) will also add significant cost to the pension plan. We hypothesize that an automatic COLA or one linked to the consumer price index (CPI) leads to faster growth of liability and thus a lower funded ratio than an ad hoc COLA or other less restrictive approaches. We state two hypotheses related to these factors as follows:

Hypothesis 3: The benefit multiplier is positively correlated with a decrease in funded ratio.

Hypothesis 4: Automatic COLA or CPI-linked COLA is positively correlated with a decrease in funded ratio.

Contribution Variables

Contributions paid into DB pension plans are the most secure and reliable funding source for paying pension liabilities. The higher the percentage of ARC that is paid, the more likely a plan is to have a higher funded ratio. The literature reviewed in the previous section reveals that (a) pension contribution is an important factor in funded ratio; (b) government sponsors tend to manipulate or reduce employer contributions under fiscal stress, thereby resulting in lower funded ratios; and (c) government sponsors tend to reduce employee contributions when pensions are well funded, a move that will also lead to lower funded ratios in the medium and long term. State and local sponsors sometimes defer payment to pension plans and then make a large lump-sum payment to compensate for the accumulated deficit, causing the annual contribution to vary widely from year to year. We can deal with this issue more effectively in this study because we examine the average of 9 years of contribution data instead of a single year. It is expected that those plans that consistently pay a higher percentage of their ARC will have a higher funded ratio. The following hypotheses are therefore developed:

Hypothesis 5: Average percentage of employer ARC paid is negatively correlated with a decrease in funded ratio.

Hypothesis 6: Average employee contribution rate is negatively correlated with a decrease in funded ratio.

Other Control Variables

The literature review also suggests that (a) governments are more likely to underfund their plans in times of fiscal stress (Chaney et al., 2002; Marks et al., 1988); therefore, higher deficit or higher outstanding debt level may lead to lower funded ratio; (b) underfunding or deferral of pension contribution to future periods can be partly attributed to fiscal constraints and political pressures (Marks et al., 1988), implying that pension plans in states with strict budget restrictions such as BBR are less likely to be fully funded due to the low priority of pension funding in a state budget; and finally (c) union and collective bargaining power may also play a significant role in funding decisions regarding pension plans (Mitchell & Smith, 1994). To account for these factors, we operationalize and add the following testable hypotheses to the regression equation:

Hypothesis 7: Per capita general fund deficit is positively correlated with a decrease in funded ratio.

Hypothesis 8: Per capita debt outstanding is positively correlated with a decrease in funded ratio.

Hypothesis 9: The presence of a strict BBR is positively correlated with a decrease in funded ratio.

Hypothesis 10: Union coverage of pension participants is positively correlated with a decrease in funded ratio.

Finally, the literature also suggests that plan demographics such as the ratio of active to retired pension participants may be associated with the change of funded ratio (Giertz & Papke, 2007). An active member is an employee who has enrolled in a pension plan and is accruing benefits for current service, whereas retired members are already entitled to monthly pension benefit payments. As the Public Fund Survey (2012) demonstrated, the ratio of active to retired members has been declining for 12 consecutive years from 2001 to 2012, due to both reductions in the number of persons employed by state and local governments and increases in the number of retired members. A low or declining ratio of active members to annuitants³ is not in itself problematic for a public pension plan. When combined with an unfunded liability, however, a low or declining ratio can cause fiscal stress for pension plan sponsors (Public Fund Survey, 2012). Therefore, we develop our last hypothesis as follows:

Hypothesis 11: The ratio of active to retired members is negatively correlated with a decrease in funded ratio.

Data and Method

Data

Data to be analyzed in this study were drawn mainly from the new Public Plans Database (PPD) constructed by the Center for Retirement Research at Boston College, and also from some other well-known sources including the U.S. Census Bureau, the NCSL, and the Current Population Survey (CPS). The PPD contains comprehensive financial, governance, and plan design information for 126 state and local DB plans. Even though there are more than 3,000 pension plans in the United States, the 222 state-level pension plans accounted for 90% of the all members and 84% of all assets in public pension plans in 2011, accounting to the U.S. Census Bureau survey of public pension systems. At the state level, the PPD covers 107 plans, representing more than 90% of all state government pension assets and members. At the local level, the PPD covers 19 plans, representing more than 20% of all local government pension assets and members. Overall, the PPD covers more than 85% of all government pension assets and members. Even though the sample size is not particularly large, it still represents the vast majority of the public pension universe, primarily due to the highly centralized nature of public pension administration at the state level. The PPD is thus by far the best available nationwide data source for public pension plans.

For the purposes of our study, it is necessary to make some adjustments to the data set by deleting several plans for the study period based on their pension cost evaluation method, plan type, and some other issues that could obstruct our understanding of variations in funded ratio change. First, in the PPD, there are four types of pension cost evaluation methods: entry age normal (EAN), projected unit credit (PUC), aggregate cost, and frozen entry age cost method.⁴ EAN is a cost allocation method on an individual basis with separate unfunded liability; PUC is a benefit allocation method on an individual basis with separate unfunded liability; the aggregate cost method uses a cost allocation on a group basis with no separate unfunded liability; and the frozen entry age method uses cost allocation on a group basis with frozen initial unfunded liability. Due to the structure of the aggregate and frozen entry age methods, the funded ratio of plans using these two methods is always kept at or very close to 100%; in other words, the funding levels of these plans do not vary. Therefore, these plans have been deleted from the sample. As shown in Table 1, these two account for about 16% of the 126 plans, so excluding them from the sample should not cause severe loss of information.

Table 1.

Distribution of Pension Cost Evaluation Methods.

Evaluation methods	Frequency	%	Cumulative
Entry age normal	806	71.08	71.08
Projected unit credit	148	13.05	84.13
Frozen entry age normal	93	8.20	92.33
Aggregate cost	87	7.67	100.00
Total	1,134	100.00

Second, the PPD divides public pensions into three types according to plan participants: general plans, teachers’ plans, and police and firefighters’ plans. Police and firefighters’ plans are usually distinct from the other two types in that the retirement age for police and firemen is substantially earlier and the benefit offered to them is usually more generous. For this reason, we have deleted all police and firefighters’ plans, which account for about 7.9% of all observations, as can be seen in Table 2.⁵

Table 2.

Distribution of Plan Types.

Plan type	Frequency	%	Cumulative
General plan	684	60.32	60.32
Teachers plan	360	31.75	92.06
Police and firefighter’s plan	90	7.94	100.00
Total	1,134	100.00

Third, to make funded ratios of different plans comparable with the fullest extent possible, we have carefully examined the PPD data set and decided to eliminate a few other plans for various reasons. We deleted the West Virginia Teachers’ Plan because it merged with a defined contribution plan in 2006, causing its funded ratio to jump from 31% in 2006 to 51% in 2007. We deleted the Nebraska Teachers Retirement Plan because the funded ratio remained at 100% throughout the period in the database, an obvious error when checked against the pension plan’s own financial reports. We also deleted the Vermont State Employees and Teachers System because this plan used the frozen entry age method until 2006. The Indiana Teachers plan uses a pay-as-you-go funding mechanism for people hired prior to 1996, so it was deleted. Finally, we deleted the City of Austin Employee Retirement System because its average annual investment return was −6.44%, making it an extreme outlier in the data set.

After all these adjustments and corrections, the resulting sample contained 92 plans spanning 9 years from 2001 to 2009. Because 8 of these 92 had missing data in at least one of three study variables (benefit multiplier, union coverage, and employee contribution rate), the final sample size was 84 pension plans.

Dependent Variable

The dependent variable of this study is the average annual change of pension funding level from 2001 to 2009. We choose 2001 as the starting year because it was the year when almost all state and local pension funds reached their highest funded ratio; 2009 was selected as the ending year because it immediately followed the 2008 fiscal crisis, when most pension plans were at their lowest funded ratios in several decades. During this 9-year event window, all 84 plans examined in the study experienced declines in their funded ratio.

In terms of how to measure this dependent variable, the change in funded ratio, there is no universally accepted method in the literature. Generally, there are three options. The first option is to use the change in pension funded ratio between 2001 and 2009. This approach, however, relies too heavily on the choice of starting and ending year of the sample, and is thus too vulnerable to historical idiosyncrasies. If the funded ratio changed dramatically at a certain point of time as a result of legislative mandates or for some reason, choosing even a slightly different cutoff year could lead to a totally different set of results. The second option is to use the year-to-year change in funded ratio and put together a panel data set with 84 plans, each of them containing year-to-year comparisons. This is a tempting approach, because creating such a large data set of individual instances of year-to-year change has technical merit and can typically generate more convincing empirical results. It is, however, not the best choice for this research. An intrinsic problem with pension studies is that many determinants, especially those controlling for contribution, benefit structure, and policy issues, vary by only a small amount if at all. Ignoring this distinctive feature of pension variables will cause such statistical problems as insignificant coefficients and low R².

To overcome the difficulties described above, we choose the third option, which is to use the average of yearly changes in funded ratio as the dependent variable for this study. In constructing the dependent variable, we first calculate the change in pension funded ratio from the immediate past year for all plans and then take an average of the changes. For the specific time period that we have chosen for this study, the resulting average ratios are negative for all plans in the sample, so we add a negative sign (in effect using the absolute value of the average change) so as to facilitate presentation and interpretation of the results. The final dependent variable, as shown in the equation below, is the absolute value of the average funded ratio changes for 84 state and local public pension plans:

avgchg = - {average ({funded ratio}_{t} - {funded ratio}_{t - 1})},

where t = 2002, . . ., 2009.

Figure 1 shows a histogram of the dependent variable. It reveals that the average funded ratio change for the majority of the plans is between 1% and 5%. The distribution of the 84 observations approximately follows a normal curve, indicating that the dependent variable is appropriate for regression analysis.

Figure 1.

Histogram of the dependent variable.

Independent Variables

The independent variables to be included in the empirical model flow out of the theoretical framework explained in the previous section. The main variable of interest, investment return, is measured in terms of annualized rate of return, which means taking the return on an investment over a given period (8 years, in this case) and computing what the annual rate would be if the period were 1 year. This is a more accurate measurement than a simple average of multiple years of annual rate of return, because it smoothes out year-to-year differences and makes investment return more comparable across different plans. The formula to calculate annualized rate of return is given below:

Annualized rate of return = {{[\prod_{i = 1}^{n} (1 + {annual rate of return}_{i})]}^{1 / n}} - 1,

where n is the number of years. In our sample, the minimum and maximum annualized rates of return from 2001 to 2009 were −0.4% and 6.2%, with the mean at 2.4%.

To capture the change in rate-of-return assumption, we construct a set of two dummy variables: real rate of return reduced (“realrorred”) and real rate of return increased (“realrorinc”). “Realrorred” will take on the value of 1 if the assumption was reduced during the period and 0 otherwise; “realrorinc” will take on the value of 1 if the assumption was increased during the period; if both dummy variables are 0, there was no change in return assumption for the plan. Out of the 84 observations in our sample, 49 plans increased their return assumption, 8 reduced their return assumption, and 27 remained unchanged from 2001 to 2009.

Pension benefit change is described by two variables: benefit multiplier (“retfactorcode”) and cost-of-living adjustment (“cola”). The benefit multiplier is an important element in the formula to calculate normal service benefits for retirees. The benefit multiplier data are directly drawn from the PPD. As for “cola,” automatic or CPI-linked COLA types are coded as 1, and other types are coded as 0. In our final sample, the benefit multiplier varied between 0.8% and 3% with a mean at about 2%; about two thirds of the plans had automatic or CPI-linked COLAs whereas one third of the plans had other types of COLAs.

We use two variables to capture the impact of contribution: percentage of ARC paid (“percentarc”) and employee contribution rate (“eecrate”). The first variable measures the average percentage of ARC paid to the pension plan during the 9 years. There was great variability in this percentage: the maximum average contribution rate was 183% and the minimum was only 12%, with the mean at 93%. This wide variance reveals that different plans have very different practices regarding pension contributions. The employee contribution rate, meanwhile, measures the percentage of salary that a plan requires employees to contribute regularly to the pension fund. This amount also varies considerably from plan to plan; as shown in Table 4, the minimum employee contribution rate is zero, the maximum is more than 11%, and the mean is around 5%.

Next, we include average general fund deficit per capita, average debt outstanding per capita, BBR, union membership coverage, and average change in the ratio of active to retired members as control variables to measure financial stress, fiscal constraints, union influence, and member characteristics. The data on the first two variables are standard census data. The information on BBRs comes from a survey conducted by the NCSL. There are different definitions of this requirement (NCSL, 2010), but we adopt the most strict form of the requirement, referred to as “cannot carry over deficit,” for this study. In all, 38 states impose this form of requirement while 12 others implement less strict forms or none at all. The union membership variable measures the percentage of state employees who are union members. These data come from the CPS, which is compiled and updated regularly by Hirsch and Macpherson (2003).⁶

Finally, the average change in active-to-retired-members ratio (“avactchg”) is constructed as the number of active employees divided by the number of retired members for each pension plan included in the sample. The data on retired and active members are drawn from the PPD. In our sample, the average change of this ratio varies from −0.55 to 0.05, with a mean of −0.08. The vast majority of the data points fall between −0.2 and 0, revealing that the ratio has been declining for most pension plans over this period.

Tables 3 and 4 provide a description and summary statistics for all variables.

Table 3.

Variable Specification.

Variable	Description and sources
avgchg	Negative of average funded ratio change (authors’ calculation)
anul_return	Annualized investment return (authors’ calculation)
retfactorcode	Numeric recode for accrual rate for defined benefit formula
percentarc	Average percentage of annual required contribution paid
eecrate	Average employee contribution rate
deficitpc	Average general fund surplus/deficit per capita (census data)
lndebtpc	Average per capita debt outstanding in log (census data)
avactchg	Average change in the ratio of active to retired members (authors’ calculation)
union	Union membership and coverage (CPS)
bbr	Balanced budget requirement (coded as 1 if the state cannot carry over deficit; 0 otherwise; NCSL)
cola	Cost-of-living adjustment (1 if automatic or CPI-linked; 0 if another type of COLA or none)
realrorred	Real rate of return assumption reduced (1 if the assumption was reduced; 0 otherwise)
realrorinc	Real rate of return assumption increased (1 if the assumption was increased; 0 otherwise)

Note. Data on all variables were collected from PPD unless otherwise annotated in parentheses. CPS = Current Population Survey; NCSL = National Conference of State Legislatures; CPI = consumer price index; COLA = cost-of-living adjustment; PPD = Public Plans Database.

Table 4.

Descriptive Statistics.

Variable	Observations	M	SD	Minimum	Maximum
avgchg	84	2.72	1.38	−0.03	8.39
anul_return	84	2.39	1.15	−0.37	6.25
retfactorcode	84	2.01	0.39	0.8	3
percentarc	84	93.01	25.54	11.99	183
eecrate	84	5.20	3.07	0	11.33
deficitpc	84	19.12	318.89	−358.32	1,410.85
lndebtpc	84	7.85	0.49	6.90	9.12
avactchg	84	−0.08	0.07	−0.55	0.05
union	84	39.60	16.72	12.88	66
bbr	84	0.75	0.44	0	1
cola	84	0.67	0.47	0	1
realrorred	84	0.10	0.30	0	1
realrorinc	84	0.58	0.50	0	1

Empirical Results and Discussion

To test the above hypotheses, we developed an ordinary least squares (OLS) model of the change in pension funded ratio as follows,⁷ with predicted signs written above each variable:

avgchg = f (\begin{array}{l} \bar{annual_return}, \overset{+}{retfactorcode}, \bar{percentarc}, \bar{eecrate}, \overset{+}{defictipc}, \\ \overset{+}{lndebtpc}, \bar{avactchg}, \overset{+}{union}, \overset{+}{bbr}, \overset{+}{cola}, \overset{+}{realrorred}, \bar{realrorinc} \end{array}) .

We estimated the OLS model with the data described in the previous section. To compare the relative magnitude of each explanatory variable, we also calculated standardized coefficients. Both results are presented in Table 5.

Table 5.

Regression Results.

	Unstandardized coefficients		Standardized coefficients
	β	SE	β	t	Significance
anul_return	−0.642***	0.123	−0.537***	−5.21	.000
retfactorcode	−0.126	0.380	−0.036	−0.33	.742
percentarc	−0.00952*	0.00561	−0.176*	−1.70	.094
eecrate	−0.108**	0.0514	−0.239**	−2.09	.040
avactchg	−0.563	2.068	−0.027	−0.27	.786
deficitpc	0.000154	0.000485	0.036	0.32	.751
lndebtpc	−0.0254	0.366	−0.009	−0.07	.945
union	−0.00630	0.0102	−0.076	−0.62	.539
bbr	−0.293	0.316	−0.093	−0.93	.356
cola	0.545*	0.308	0.187*	1.77	.081
realrorred	1.387**	0.527	0.297**	2.63	.010
realrorinc	−0.872***	0.295	−0.313***	−2.96	.004
_cons	6.588**	2.983		2.21	.030
n	84
R ²	.405
F	4.028

p < .10. **p < .05. ***p < .01.

We performed several diagnostic tests, which showed that the overall quality of the regression model is quite good. First, the p value of the White test was .42, strongly rejecting the null hypothesis that there is heteroskedasticity in the model. All variation inflation factor (VIF) scores were approximately 2 or less, indicating there is no multicollinearity problem. The p value of Durbin’s alternative test for autocorrelation was .24, failing to reject the null hypothesis that there is first-order autocorrelation at the 5% level. The model also passed the Ramsey Regression Equation Specification Error Test (RESET)(p = .44), suggesting that the model has no omitted variables.

Results from the model specifications are generally consistent with our hypotheses; the signs and magnitudes of the coefficients are not sensitive to changes in the specification, indicating that the empirical results are reliable. The coefficient on the annualized investment return is statistically significant at the 5% level. Its negative sign provides strong empirical support for Hypothesis 1 that a greater annualized investment return reduces the decrease in pension funded ratio. For every 1% increase in investment return, the funded ratio declines by 0.64% less, ceteris paribus. Given that the mean value of the average annual change of pension funding level is about −2.7% and the average annualized return for all 84 plans is 2.4%, the magnitude of the impact of investment return is remarkable. This point is further confirmed by the standardized coefficient on the annualized investment return (−0.54), which is the largest in the model.

As predicted, the coefficient on “real rate-of-return assumption reduced” is positive and statistically significant at the 5% level, indicating that a reduction of the return assumption will cause the funded ratio to drop by about 1.4%, holding other factors constant. Similarly, the coefficient on “real rate-of-return assumption increased” also has the predicted sign and is statistically significant at the 1% level, suggesting that increasing the return assumption will lead to a 0.87% increase in the funded ratio. The standardized coefficients on both variables are also relatively large as compared with those of other variables in the model, suggesting that these two variables carry considerable weight in the determination of the change in the funded ratio. This result is consistent with the finding by Bizley (1950) that a small change in the rate-of-return assumption will alter long-run cost estimates by a large margin.

The two contribution variables bear the predicted signs. The coefficient on employees’ contribution (“eecrate”) is statistically significant at the 5% confidence level; the coefficient on employers’ contribution (“percentarc”) is significant at the 10% level. The negative coefficient on “percent of ARC paid” indicates that plans paying a higher percent of their ARC are likely to have a smaller decrease in their funded ratio. Other things being equal, a 1% increase in ARC contribution will cause the funded ratio to increase by 0.01%. Considering that the mean ARC contribution percentages of DB plans in the sample vary from 12% to 183%, and that the standardized coefficient is 0.18, the effect of ARC contribution is by no means negligible.

The “employee contribution rate” variable pushes the change of funded ratio in the same direction as the ARC contribution percentage does. Holding other variables constant, a 1% rise in the employee contribution rate will cause the funded ratio to increase by 0.1%. As the mean employee contribution rate is 5% and it usually does not vary widely, the effect of this variable comparable with or even slightly greater than that of the ARC contribution percentage as indicated by the standardized betas.

The sign of “COLA” is consistent with our prediction, and it is statistically significant at the 10% level. The positive coefficient indicates that automatically adjusting benefits according to the inflation rate and paying higher benefits to retirees will place a heavier burden on pension plans, causing the funded ratio to decrease by a larger amount. The other benefit variable included in the model, benefit multiplier (“retfactorcode”), is not statistically significant at any acceptable confidence level.

The relative impact of the independent variables included in the model can be compared based on standard coefficients. Annualized investment return is by far the most important factor in explaining the variation in the change of funded ratio. This echoes the fact that total investment income of all U.S. public plans has accounted for an increasingly larger portion of total pension revenue since 1985. The second most important variable is the change in real rate of return assumption. Either an increase or a decrease in this assumption will result in about the same degree of impact on funding level, in the opposite direction. This shows that when analyzing a pension plan’s change in funding level, we also need to take account of the change in underlying assumptions. Employee contribution rate is the third important contributor to the variation in the funded ratio, while the percentage of ARC paid and COLA provision are the forth and their effects are similar. This shows that employee contribution is just as important if not more so than employer contribution in determining the funding level. These results are consistent with the literature and theory presented in previous sections.

Last, contrary to our expectations and to the findings of some previous studies, other variables related to financial stress, fiscal constraint, indebtedness, demographics of plan participants, and union influence are generally not statistically significant. We believe that the reason why the first three of these variables (all related to states’ or localities’ financial status) that measure fiscal stress are not significant is that the effect of fiscal stress on the funded ratio change has already been captured by the contribution variable. As fiscal stress should have no direct impact on the level of pension benefit and investment performance, its primary impact on the funded ratio is through pension contribution, in terms of the percentage of ARC paid. After controlling for ARC, the fiscal stress variables are no longer significant. Similarly, the ratio of active to retired members has only an indirect relationship with the funded ratio through its impact on contributions. Once the contribution variable is included, this ratio adds no further explanatory power to the model.

It is worth mentioning that the overall quality of our model has improved notably from previous empirical studies using the same data set. Compared with, for example, the research of Munnell et al. (2008) and Stalebrink (2014), our study incorporates 9 years of data (instead of one particular year), uses a smaller sample but a more relevant one after excluding plans that are obviously distinct, and uses about the same number of variables, yet achieves a notably higher R² result. F score and diagnostic test results all suggest that the model is well specified and quite reliable. More importantly, we believe that examining the change in funded ratio over a chosen event window instead of modeling the absolute level of funded ratio at one particular point of time is logically more correct and empirically more convincing.

Conclusion and Policy Implications

State and local government pension funds represent about 27% of the total retirement assets⁸ in the United States and cover approximately 89% of state and local government employees as of 2012.⁹ The declines in pension plans’ funded ratios since 2000 and the increasing gap between pension assets and liabilities leave many people wondering if the benefits promised to them will actually be there by the time they retire. The last few years have seen considerable discussion of how to reform public pension plans and keep the system sustainable. This research contributes to this discussion by examining why some pension plans performed better than others over a very tumultuous period in the first decade of the 21st century. This study has contributed to the literature by proposing a model that examines the effect of investments, contributions, and benefits on the long-term change in funded ratios. It has generated empirical evidence to suggest that, in the order of magnitude, deterioration in investment performance, drops in employer and employee contribution, and generous provision of COLAs have been associated with a greater decline in pension plan funded ratios after 2000, while other benefits offered by pension plans and external factors such as deficits, fiscal constraints, and union coverage have played little role in the process after controlling for the above factors. In addition, this article also confirms that changes in investment return assumption also affect funded ratio to a great extent.

The results of this study have several implications for policy makers. Making sufficient contributions to pension funds on a regular basis is of vital importance to maintaining a healthy pension system. Even though this point has been stressed previously, the present study provides further evidence of its importance. One promising reform to address this threat would be to pass laws making it impossible or at least very difficult, to delay or skip actuarially required pension contributions. This study shows that the employee contribution rate is also important in maintaining a healthy funded ratio. A higher employee contribution rate, perhaps equal to the employer contribution rate, could also be a useful part of pension reform packages. This provision will help to ensure healthy funded ratios and lessen the burden on government and taxpayers.

This study provides compelling evidence that investment return has a significant impact on funded ratios. According to PPD, state and local governments differ greatly in investment strategies—and in investment outcomes. Although further research on why some plans perform better in the financial market than others is needed, there is no doubt that many plans could improve their investment performance by adjusting their portfolio, enhancing the use of investment counsel, and managing risk more effectively.

Last, state and local pension sponsors should exercise caution in providing COLA for retirees. As we have seen, either automatic or CPI-linked COLAs will significantly increase pension liabilities in a steady and compounding manner.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Qiushi Wang is an Assistant Professor of Public Affairs and Administration. Prior to joining the faculty of Rutgers, Wang taught at the University of Texas at Arlington, Texas. He also served as a visiting instructor and visiting scholar at the City University of Hong Kong and Sun Yat-Sen University, China. His research interests include pension management, municipal bonds and debt policy, government accounting, and public budgeting.

Jun Peng is an Associate Professor in public administration. He has been teaching at the University of Arizona since earning his Ph.D from the State University of New York at Albany. His research interest primarily focuses on state and local government financial management in the U.S., such as budgeting, debt management and pension management. His recent book “State and Local Pension Fund Management” was published by Taylor and Francis in 2008. He currently is also involved in research on budgetary transparency in China.

References

Adler

Sacco

(1995, May 1). Strapped municipalities defer pension funds. Wall Street Journal, p. A9.

Bizley

M. T. L.

(1950). The effect on pension fund contributions of a change in the rate of interest. Journal of the Institute of Actuaries Students’ Society, 10, 47-51.

Chaney

B. A.

Copley

P. A.

Stone

M. S.

(2002). The effect of fiscal stress and balanced budget requirements on the funding and measurement of state pension obligations. Journal of Accounting and Public Policy, 21, 287-313.

Clark

(1990). Has politics spoiled the Virginia fund? Institutional Investor, 24, 111-116.

Feldstein

Seligman

(1981). Pension funding, share prices, and national savings. The Journal of Finance, 36(4), 801-824.

Franzoni

Marin

J. M.

(2006). Pension plan funding and stock market efficiency. Journal of Finance, 61, 921-956.

Giertz

J. F.

Papke

L. E.

(2007). Public pension plans: Myths and realities for state budgets. National Tax Journal, 60, 305-323.

Hess

(2005). Protecting and politicizing public pension fund assets: Empirical evidence on the effects of governance structures and practices. University of California Davis Law Review, 39, 187-224.

Hirsch

B. T.

Macpherson

D. A.

(2003, January). Union membership and coverage database from the Current Population Survey: Note. Industrial and Labor Relations Review, 56, 349-354.

10.

Johnson

(1997). Pension underfunding and liberal retirement benefits among state and local government workers. National Tax Journal, 50(1), 113-142.

11.

Lucas

D. J.

Zeldes

S. P.

(2009). How should public pension plans invest? American Economic Review, 99, 527-532.

12.

Marks

B. R.

Raman

K. K.

Wilson

E. R.

(1988). Toward understanding the determinants of pension underfunding in the public sector. Journal of Accounting and Public Policy, 7, 157-183.

13.

Mitchell

O. S.

Hsin

P.-L.

(1997). Public sector pension governance and performance. In Valdes

(Ed.), The economics of pensions: Principles, policies, and international experience (pp. 92-126). Cambridge, UK: Cambridge University Press.

14.

Mitchell

O. S.

Smith

R. S.

(1994). Pension funding in the public sector. Review of Economics and Statistics, 76, 278-290.

15.

Munnell

A. H.

Haverstick

Aubry

(2008). Why does funding status vary among state and local plans? (State and Local Pension Plans, No. 6). Chestnut Hill, MA: Center for Retirement Research, Boston College.

16.

National Conference of State Legislatures. (2001). State fiscal outlook for 2001: February update. Retrieved from http://www.ncsl.org/Portals/1/documents/fiscal/sfo2000.pdf

17.

National Conference of State Legislatures. (2010). NCSL fiscal brief: State balanced budget provisions. Retrieved from http://www.ncsl.org/documents/fiscal/StateBalancedBudgetProvisions2010.pdf

18.

National Conference of State Legislatures. (2012). State governments’ public safety retirement plans. Retrieved from http://www.ncsl.org/documents/employ/PENSION-PUBLIC-SAFETY-TABLE-8-6-12.pdf

19.

Peng

(2009). State and local pension plan management. Boca Raton, FL: Taylor & Francis.

20.

Pew Center on the States. (2011). The widening gap: The Great Recession’s impact on state pension and retiree health care costs. Retrieved from http://www.pewtrusts.org/uploadedFiles/wwwpewtrustsorg/Reports/State_policy/State_Pensions_Health_Care_Retiree_Benefits.pdf

21.

Public Fund Survey. (2012). Summary of findings for FY 2012. Retrieved from http://www.publicfundsurvey.org/publicfundsurvey/summaryoffindings.html

22.

Retkwa

(1990, September 1). New York state plan alters funding method. Pension World, 29(9), p. 22.

23.

Stalebrink

O. J.

(2014). Public pension funds and assumed rates of return: An empirical examination of public sector defined benefit pension plans. American Review of Public Administration, 44, 92-111.

24.

Tepper

(1981). Taxation and corporate pension policy. The Journal of Finance, 36, 1-13.

25.

United States Census Bureau. (2011a). Annual survey of public pensions: State and local data. Retrieved from http://www.census.gov/govs/retire/

26.

United States Census Bureau. (2011b). Government employment and payroll. Retrieved from http://www.census.gov/govs/apes

27.

Vermeer

T. E.

Styles

A. K.

Patton

T. K.

(2010). Are local governments adopting optimistic actuarial methods and assumptions for defined benefit plans? Journal of Public Budgeting, Accounting & Financial Management, 23, 511-542.