Qiushi Wang Sun Yat-sen University, China Jun Peng University of Arizona
Political Embeddedness of Public Pension Governance: An Event History Analysis of Discount Rate Changes
Abstract: State governments are frequently said to manipulate the discount rate assumption to make pension funding look better, reduce employers’ and employees’ pension contributions, or relieve fiscal stress. Building a model from the political embeddedness perspective and applying an event history analysis to the 81 largest state-administered pension plans in the United States, the authors found that more politically embedded pension boards were actually more likely to reduce their plan’s discount rate. Public union coverage and government political ideology, however, had no significant impact on discount rate changes. These findings reveal the effect of political embeddedness on pension planning decisions and provide useful insights into the intricate process of setting pension discount rates in a new era of more muted investment return expectations. This article points to both political and financial pressures facing pension boards and state governments for many years to come. Evidence for Practice • Facing severe fiscal stress, more public pension plans have chosen to reduce rather than increase their discount rate assumption to reflect the much lower average investment return since 2001. • More politically embedded pension boards are more likely to reduce the discount rate than plans with more elected board members. • Regardless of their composition, public pension boards face both political and financial pressures to take a piecemeal approach to reducing the discount rate. • Because of the gap between the current discount rate and muted investment return expectations in the future, public pension plans will continue to face pressures for the foreseeable future.
D
efined benefit pension plans cover most state and local government employees (U.S. Bureau of Labor Statistics 2017),1 and they are a major variable affecting state and local governments’ fiscal sustainability (Chapman 2008). Since 2001, and especially since the Great Recession, the underfunding of public pension plans and the increasing costs of financing pension benefits for state and local public employees have become important public policy issues (Anzia and Moe 2017), capturing the attention of scholars in public finance and public administration. Some scholars have explored the economic and institutional factors behind the underfunding of public pension plans (Chaney, Copley, and Stone 2002; Chen 2018; Eaton and Nofsinger 2004; Novy-Marx and Rauh 2011; Wang and Peng 2016). Others have focused on the broader implications and serious consequences of deteriorating pension health for the public sector. They suggest that increasing pension costs put pressure on the funding of other major government programs such as education, public safety, and health care (Coggburn and Kearney 2010; Stalebrink 2014), make it harder
Research Article
Qiushi Wang is associate professor in the School of Government and Center for Chinese Public Administration Research at Sun Yat-sen University, China. His research interests include public pension management, municipal bonds and debt policy, government accounting, and nonprofit finance. He holds a PhD in public administration from the University of Nebraska Omaha. E-mail:
[email protected] Jun Peng is associate professor in the School of Government and Public Policy at the University of Arizona. His research focuses on state and local government financial management in the United States, such as budgeting, debt management, and pension management. His book State and Local Pension Fund Management was published by Taylor & Francis in 2008. He holds a PhD in public administration from the State University of New York at Albany. E-mail:
[email protected]
for governments to borrow money or plan their budgets (Brooks 2017), push some governments to make changes to public pension benefit design that may affect public sector employment and turnover in the workforce (Lewis and Stoycheva 2016), and drive some states to enact defined contribution or hybrid plans in order to lower employer contributions and divert investment risk to employees (Thom 2013, 2015). One of the most frequently mentioned culprits for the current pension problems is the assumed rate of return on pension investment, or the discount rate, which is used to calculate the present value of pension assets and liabilities. As we will explain later, the discount rate is the most important assumption determined by a pension board (Brooks 2017), and it plays a key role in pension financing. It has long been noted that the setting of a plan’s discount rate is not simply a result of expert financial and actuarial advice but also involves significant political considerations from the perspectives of various actors and interest groups, such as the state legislature, executive branch (through
Public Administration Review, Vol. 00, Iss. 00, pp. 1–10. © 2018 by The American Society for Public Administration. DOI: 10.1111/puar.12968.
Political Embeddedness of Public Pension Governance: An Event History Analysis of Discount Rate Changes 1
proposal), public unions (through negotiation), and pension boards of trustees (Chen 2018). For instance, public pension sponsors are often said to exercise political opportunism and manipulate the rate of return assumption to make pension funding look better, reduce government’s required contributions, or relieve fiscal stress (Chaney, Copley, and Stone 2002; Stalebrink 2014). To a large extent, the outcome of a decision on the discount rate is determined by the interplay and balancing of different political forces, but literature in this area is limited. So far, except for a few studies that have examined the adoption of discount rates mostly from financial and institutional perspectives (e.g., Brooks 2017; Stalebrink 2014), not a great deal is known about how different political forces can influence the setting of the discount rate, and the political dynamics linking pension governing boards to this pension decision have never been systematically explored (Anzia and Moe 2017, 58). In this article, we focus on the political explanation for rate-setting behavior and propose the concept of political embeddedness to capture the complexity of the government-board relationship in pension management. The concept of “political embeddedness” was first articulated by Polanyi (1944), who defined embeddedness as the degree to which economic activity is constrained by noneconomic institutions. This concept was later extended to refer to bureaucratic, instrumental, or affective ties to the state and its actors in ongoing patterns of social relations, and it has become one of the core ideas of social network theory (Granovetter 1985). The embeddedness perspective views political connections to government as associated with norms, expectations, and responsibilities, and they can create both opportunities and constraints for the organizations involved (Uzzi 1997). In public management research, the social, political, or institutional embeddedness perspective has been applied to explore a wide range of organizational behavior and performance issues. In studying institutional collective action and economic development among metropolitan jurisdictions, Feiock, Steinacker, and Park (2009) found that cities with both strong and weak ties were more likely to enter into additional collaborative relationships with new and existing partners for joint ventures. Similarly, Lee, Feiock, and Lee (2012) focused on the role of network embeddedness and showed that, at the micro level, dyadic conceptual ties of cooperation and competition can affect policy network structures for economic development. In studies of nonprofit organizations, scholars have found that political ties to government can help philanthropic foundations or nongovernmental organizations obtain more government subsidies, donations, and market revenue (Ni and Zhan 2017). Finally, higher levels of government involvement (embeddedness) have also been prescribed to deal with the marketfailure problem by privatization theorists, who suggest that embeddedness will reduce agents’ opportunistic behavior (Milward and Provan 2000) and prove to be important for divestment decisions (Amirkhanyan 2007). In the world of public pensions, governments not only oversee and regulate pension activities but also become directly involved in pension governance through board ties. Doing so literally embeds political actors in an independent economic entity (Michelson 2007). As such, a political embeddedness perspective can help us consider both the advantages and the constraints associated with 2 Public Administration Review • xxxx | xxxx 2018
a pension board’s ties to government. Drawing on insights from previous literature, we focus on the political embeddedness of public pension boards along two dimensions: strong ties to government with more influence from state political ideology, and weak ties to government with more influence from outside interest groups such as public unions. The first dimension emphasizes the extent to which government can directly control the outcome of pension decisions. The advantages of strong ties to government are that a pension board can reduce uncertainties by encouraging the flow of ideas and keeping government in touch with interest groups and voters (Moe 1985) and that certain pension policies (e.g., the policy on contribution rate) can be implemented more smoothly. However, the constraints of strong government ties are that pension decisions tend to be more aligned with government’s fiscal and political goals contingent on state political ideology and less attentive to plan participants’ wellbeing and the long-term fiscal health of the plan (Anzia and Moe 2017), because political appointees tend to retain allegiances to the institutions that place them in their jobs (Moe 1982). The second dimension focuses on the extent to which pension decisions are influenced by outside interest groups when ties to government are weak. The advantages of weak ties to government are that boards are more independent of the political influence of elected officials and that plan members can enjoy more union protection (Anzia and Moe 2014). The disadvantages of weak government ties, however, may include less information from government, advocacy for generous pension benefits, preference for the use of actuarial methods that disguise costs (Brooks 2017, 6), and more influence from other political actors and interest groups that may have conflicting role expectations (Kelley 2014). Overall, an emerging consensus in the literature is that political forces have played an indispensable role in pension management, but it is less clear in what ways and to what extent public pension plans have been influenced by their government ties. Building on the existing literature and using the lens of political embeddedness, this article aims to explain different political forces in a complete ecosystem and further resolve the factors that have led a growing number of pension plans to change their discount rate over the past decade. By incorporating and testing a political embeddedness model with recurrent event history data of large U.S. public pension plans for the period 2001–24, this article offers two theoretical and methodological improvements on the existing research. First, we direct a more critical eye to the role of states’ political forces on pension decisions and adopt an “embeddedness” view to examine these forces. As Uzzi (1996) points out, “embedded” ties are more powerful than “arm’s length” ties for trust, rich transfers of information, and problem-solving capabilities. Second, we use a modeling strategy that recognizes the mechanism causing the changes in the discount rate over time, thus overcoming the limitations of the cross-sectional analyses used in most previous studies. The Central Role and Variations of Pension Discount Rate The discount rate uniquely straddles the two sides of pension financing: assets and liabilities. The assets needed to pay for the benefits come from pension contributions and the investment
return earned on those pension contributions. There is an inverse relationship between these two components. If more investment income is expected in the future, less pension contributions are needed now. On the liability side, all pension benefits promised by a government need to be discounted so that the government knows what the total pension liability is worth in today’s dollars. Thus, a higher/lower discount rate can render both the pension liability and pension contribution smaller/bigger, making today’s pension benefit more affordable/costly. Roughly, a 1 percentage point increase in the discount rate reduces the pension liability by 12.5 percent and the pension contribution by 22 percent (Munnell and Aubry 2016). As government pension contributions are paid out of the current year’s budget, the discount rate is also linked to the government’s budget. A higher return assumption makes the pension contribution less of a budgetary burden. This is particularly valuable in a time of fiscal stress caused by economic recession.
public pension plans. Did some pension plans adopt more optimistic (meaning higher) discount rates in order to reduce government pension contributions and thereby relieve budget pressure? Although this is still an open question, recent events have changed the debate over this second concern. Beginning in the early years of the twenty-first century, as figure 1 shows, many plans started reducing their discount rate. This trend accelerated after the financial market meltdown in 2008. By 2014, the median discount rate nationwide was lower than 7.75 percent. This development happened over a period when state and local government finances were already under severe fiscal stress after the Great Recession. Instead of increasing the discount rate to relieve government fiscal stress, as would be suggested by this second line of criticism, far more pension plans reduced rather than increased their discount rate, although about one-third of the plans chose not to change the discount rate throughout this entire period (PPD 2001–14).
Because of the central role of the discount rate, scholars and practitioners alike frequently debate the policy behind selecting an appropriate rate. In these debates, two related concerns usually come up. The first concern centers on the overall level of the discount rate for all public pension plans. For many years since 2001, the long-term national median rate was around 8 percent (figure 1). According to this first concern, this rate was set too high and severely underestimated the true cost of public pension benefits, although opinions differ as to what the appropriate rate should be. While Novy-Marx and Rauh (2011) argue that since pension benefits are guaranteed, they should be valued with a risk-free interest rate, which is much lower than the long-term investment rate of return; others disagree. According to Statement 68, issued in 2012 by the Governmental Accounting Standards Board (GASB 2012), state and local governments are required to calculate and report a net pension liability based on a single discount rate that combines the long-term rate of return on funded plan assets with a AA-rated 20-year municipal bond index rate on the unfunded portion of the liability. While this blended single discount rate is lower than the long-term investment rate of return, it is much higher than the risk-free rate.
What explains this seemingly counterintuitive policy choice behind the setting of discount rates? Most likely, this is a response to the lower average investment returns since 2001 (NASRA 2018). As table 1 reveals, the average median discount rate assumption for the largest state pension plans from 2001 to 2014 was 7.95 percent, while the average median one-year investment return was only 6.45 percent during the same period. If that is the case, why did some plans reduce their return assumption, while others chose not to? The rest of this article attempts to answer these two related questions.
The second concern, rather than centering on the overall level of discount rate, focuses on the variations in discount rates among
Return Assumption
9
8
7
6
20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14
5
Figure 1 Box Plot of Investment Return Assumptions, 2001–14
Theoretical Framework While the elected officials of state and local governments set pension benefit levels for public employees, the actual management of public pension benefits is performed by a separate legal entity, typically called a public employee retirement system, that is governed by a board of trustees. This governing board is responsible for setting the discount rate, which is the key variable of this study. The board trustees are selected in three ways. First, they can be appointed by elected officials, such as the governor or the legislature at the Table 1 Median Return Assumption and Investment Return, 2001–14, for 81 Largest State Pension Plans Year
Median Return Assumption (%)
Median 1-Year Investment Return (%)
2001
8.00
–5.80
2002
8.00
–6.40
2003
8.00
4.50
2004
8.00
15.40
2005
8.00
10.10
2006
8.00
11.74
2007
8.00
17.70
2008
8.00
–4.80
2009
8.00
–18.70
2010
8.00
13.50
2011
7.95
21.22
2012
7.90
1.90
2013
7.75
13.05
2014
7.75
16.90
Average
7.95
6.45
Political Embeddedness of Public Pension Governance: An Event History Analysis of Discount Rate Changes 3
state level. Second, they can serve ex officio, in a position such as state treasurer or comptroller. Third, they can be elected by plan members, including both currently working and contributing members and retirees. What makes pension boards interesting is that no two pension boards have the same trustee composition, ranging from entirely politically appointed to entirely elected by pension members. According to a 2008 report by the U.S. Government Accountability Office, on average, 51.7 percent of trustees were appointed by politicians, 19.1 percent of the trustees were ex officio, and the remaining 29.2 percent were elected by plan members. While the pension board is directly involved in setting the discount rate, its behavior can also be influenced by many other parties whose political, financial, or personal interests are at stake. These parties can be elected officials, plan members, unions that represent plan members, or voters/taxpayers. As figure 2 illustrates, the extent to which a pension board can be influenced by these parties is dependent on the composition of the pension board. If the pension board is mostly composed of political appointees and ex officio members, it is more likely to be influenced by state ideology but less likely to be influenced by public unions; conversely, if the board is mostly composed of elected members, it is more likely to be influenced by public unions but less likely to be influenced by state ideology. Thus, for a complete understanding of the political interplay behind the rate setting, we need to theorize the motivation of each party in this process and analyze the dynamics of the political embeddedness of the pension boards. Elected versus Appointed Trustees
Pension board trustees, however they are selected, have a solemn fiduciary duty to protect the interests of the pension plan members, ensuring that sufficient funds are available to pay benefits when they come due. This involves picking a discount rate that is in line with an expected long-term rate of return to ensure that sufficient contributions are made by government employers and employees. Given this fiduciary duty, and in light of a 14-year annualized return since 2001 that was significantly below the discount rate, many pension boards have reduced their discount rate to reflect this reality. In doing so, higher periodic pension contributions were required to be paid to make up for the lower investment returns. The higher contribution amount can have at least two financial consequences. First, it leads to an increase in the cost of financing public pension benefits, putting pressure on the government’s budget. Second, if the government is politically or legally unable to Public Unions
Government Ideology
Appointed and Ex Officio Members
Strength of Ties
Elected Members
Figure 2 Illustration of Political Embeddedness of Pension Boards 4 Public Administration Review • xxxx | xxxx 2018
raise sufficient taxes to cover such a large increase in pension cost, this can lead to political and public demand for pension reform, which usually means “reducing” or “cutting” in some fashion. Even though the benefits for current members of a pension system cannot be reduced because of legal protections, the financial interests of these members can still be negatively affected in other ways as a result of pension reform. For current contributing members, the government can require them to pay more into the pension system, effectively reducing their take-home pay. For current retirees, their future pension benefits can be cut by not granting a cost-of-living adjustment (COLA) if it is not automatic. More importantly, the government will simply reduce the pension benefits for new members to the pension system. These financial consequences can lead to divergent rate-setting behavior among board trustees with different political backgrounds. Board trustees who are elected by plan members will be more aligned with the interests of current members. They want to ensure the long-term safety of pension benefits. However, since a reduction in the discount rate can increase the possibility of a reduction in other financial benefits for current members, such as an increase in pension contributions and elimination of COLA, at least in the short term, then these board trustees will be more reluctant to reduce the discount rate, especially if they are members of the system themselves. While this may harm the long-term safety of the system, the impact will not be felt for many years. In contrast, appointed or ex officio board trustees will likely be more aligned with the political interests of elected officials (Chen 2018; Moe 1982). For elected officials, the policies they pursue should reflect public opinion and voter preferences (Jacoby and Schneider 2001; Wood and Theobald 2003). As Anzia and Moe (2017) point out, prior to the Great Recession, public interest in public pension benefits was minimal. However, with widespread media coverage of growing unfunded public pension liabilities, the public became far more aware of the growing cost of public pension benefits and the effect it had on public services, leading to increasing calls for pension benefit reform from the public. This public concern is also aligned with elected officials’ concerns about the impact on government finance and the need to rein in public pension benefits. Political appointees thus should embody this concern about financial sustainability and the need for pension benefit reform. Such concern should prompt them to reduce the discount rate. This leads to the first hypothesis: Hypothesis 1: Pension systems whose boards have strong ties to government (more politically embedded) are more likely to reduce the discount rate. Public Unions
Unions are a strong political force in the public sector. Generally, public employees are a unique interest group in that they can affect pension compensation through both the political process and collective bargaining. There is a long list of studies examining the effect of public unions. While the findings of these studies do not always agree, more studies than not have found that the presence of a public union led to an increase in salary or benefits or both (Brooks 2017) and that the positive impact of collective bargaining on pension benefits was even more evident under worsening
economic conditions (Hoang and Goodman 2018). If employees are unionized, then the union can speak with a more concerted voice to board trustees if a reduction in the discount rate could harm the financial interests of its members. The union is concerned about the interests not just of current members but of future members as well, since it will also represent them and does not want to see their benefits reduced. As Kelley (2014) states, special interest groups can also influence the information available both to voters and politicians. If special interest groups are able to present information in a self-interested way, they can affect the policies formulated by the politicians. The discount rate is one such piece of information that can influence public policy on pension benefits. While unionization is strong in the public sector, its strength also varies from state to states. This leads to a set of two related hypotheses: Hypothesis 2a: Pension systems in states with a higher degree of unionization are less likely to reduce the discount rate. Hypothesis 2b: Pension systems whose boards have weak ties to government and in states with a higher degree of unionization are even less likely to reduce the discount rate. Government Ideology
Government political ideology serves as a proxy for the political influences that elected officials and the voters/taxpayers who elect them exert on the setting of the discount rate when the board has strong ties to government. Polarization is a prominent feature of the political system in the United States. The Democratic Party is known for favoring more government spending, including spending on government employee benefits, and the Republican Party is known for favoring tax cuts and smaller government. As Anzia and Moe (2017) note, while polarization on public pension benefits was not particularly prominent before the Great Recession, it has become a more polarized issue since then, with the Republican Party calling for more benefit reductions than the Democratic Party. Therefore, board trustees appointed by the Democratic Party are less inclined to reduce pension benefits (Chen 2018; Thom 2015) and thus less inclined to reduce the discount rate. Here is the last set of hypotheses: Hypothesis 3a: Pension systems in more conservative states are more likely to reduce the discount rate. Hypothesis 3b: Pension systems whose boards have strong ties to government and in more conservative states are even more likely to reduce the discount rate. Control Variables
The first control variable is the ratio of active to retired plan members, an important pension plan characteristic that partly determines the revenue and expenditure structure of the plan. In line with the argument of special interest groups in the discussion of public unions, there may be a difference in interests between current contributing members and retirees. As retirees are already receiving benefits, the long-term safety of pension benefits is of less concern to them than it is to current contributing members. They are thus less inclined to see a reduction in the discount rate because it will have a more immediate negative effect on their pension benefits. However, self-interested younger workers have a larger
financial stake in the long-term sustainability of the pension fund because they have a longer time horizon (Kelley 2014). As a result, the higher the ratio of active to retired members, the more likely the plan will be to reduce the discount rate. The second control variable is the percentage of equities in the pension investment portfolio. Eaton and Nofsinger (2001) found that the percentage of the pension fund invested in equity is highly correlated with returns. This variable also measures how much risk a pension plan is willing to take, reflecting the pension board’s investment strategy. If a pension plan adopts a more aggressive strategy, then it expects to earn higher return from the investment and is less likely to reduce the discount rate. The third and fourth control variables concern the actual investment return. Even though the average annualized investment return since 2001 has been substantially below the discount rate, there are significant variations among pension systems, with some higher and others much lower than the average. It is conceivable that for systems with a multiple-year (typically three to five years) average return higher than or close to the discount rate, there is no need to reduce the discount rate. In contrast, for systems with an average return much lower than the discount rate, the need is much greater. Therefore, the greater the investment return gap (or return surplus, if the actual return is higher than the adopted discount rate), the more (or less) likely the governing board will be to reduce the current discount rate. A related variable is a dummy variable indicating whether the discount rate is greater than 8 percent. Given the declining trend in discount rates across the nation (figure 1), it becomes more and more difficult for plans with discount rates above the long-term national median of 8 percent to justify their more optimistic assumption, and these plans are thus more likely to reduce their rate. The last variable is the fiscal condition of the government. Previous studies have found an inverse relationship between government fiscal condition and the discount rate (Chaney, Copley, and Stone 2002; Eaton and Nofsinger 2004). Therefore, we predict that pension plans in states under more severe fiscal stress will be less likely to reduce their discount rate. Data and Methods Based on the previous discussion, we propose the following empirical model: Change in discount rate assumptioni ,t = f (Political embeddednessi ,t −1 , Union coveragei ,t −1 , Government ideology i ,t −1 , Active ratioi ,t −1 , Equity sharei ,t −1 , Return surplus/gapi ,t −1 , Above 8 percenti ,t −1 , Budget deficiti ,t −1 )
(1) Data and Variables
The dependent variable—pension plan rate of return assumptions— and much of the information related to pension plan characteristics, investment strategy, and investment returns were obtained from the Public Plans Database (PPD 2001–14). Another major data source was the U.S. Census Bureau (2016), which provides state finance data and information on union coverage. The political explanatory
Political Embeddedness of Public Pension Governance: An Event History Analysis of Discount Rate Changes 5
.4 .3
Reduced
20 15
20 10
20 00
Event History Model with Recurrent Events and Competing Risks
The main variable of interest in this study, the change in the discount rate, has at least two special features that must be considered in the data analysis. First, the discount rate tends to be stable over time and rarely changes. This is because this rate is
(2)
where a(t) may be any function of time, x1 through xk are constant explanatory variables, and xk + 1 through xp are time-varying explanatory variables. Since the events in our study are both repeatable and competing (meaning that the assumed rate of return might be reduced or increased at any point of time), we modified the standard Cox regression to tackle these problems. Appendix A in
20 05
To test the model specified in equation (1), we operationalized the independent and control variables. Table 3 and table 4 provide brief descriptions, expected signs, and descriptive statistics for all study variables.
log h t = a t + b x +…+ b x + b x () () 11 k k k +1 k +1 ( t ) +…+ b p x p ( t )
.2
For the period from 2001 to 2014, there were a total of 1,134 (81 * 14) plan-years in our sample. However, in constructing the dependent variable “increased” or “reduced,” and in generating lagged variables to account for possible delayed effects, we lost one year of observations. Therefore, the total number of plan-years for the statistical analysis dropped from 1,134 to 1,053. As table 2 indicates, 9.21 percent of all plan-years, or 97 plan-years, saw a reduction in the rate of return assumption, whereas only 1.42 percent, or 15 plan-years, witnessed an increase. Based on these data, figure 3 reports the cumulative incidence curves for reduced and increased return assumptions during 2001–14. It shows the estimated probability of the return assumption being reduced or increased as a function of years. For example, in 2009, the odds of the assumption being reduced are estimated to be about 0.04 percent, and the odds of the assumption being increased are estimated to be a little over 0.01 percent. Furthermore, figure 3 clearly reveals that from 2002 to 2009, the cumulative probability of reducing the return assumption rose only slightly faster than its counterpart, but it soared after 2010 while its counterpart remained at a very low level. Finally, it is worth noting that not one plan in our sample increased its return assumption between 2010 and 2013.
This study adopts a better modeling strategy that views the change in the return assumption, either a decrease or increase, as an “event.” An event refers to something that does not occur very frequently and consists of a relatively sharp disjunction between the state that precedes the event and the state that follows it. Because these events are defined in terms of change over time, it is increasingly recognized that the best way to study events and their causes is to collect event history data and conduct an event history analysis (EHA) (Allison 2014). In this study, we estimate event history data using a standard Cox regression, which can be written as follows:
.1
One important independent variable, the political embeddedness of the pension board, was available from the PPD only for the period from 2001 to 2009. To make the data set complete, we supplemented the PPD data with a more updated and detailed survey data collected from the National Association of State Retirement Administrators (NASRA). A careful comparison of the PPD data for 2001–09 and the NASRA data reveals that the two data sets are highly consistent with each other. Since NASRA includes pension board information for only 88 large plans, and we removed a few locally administered plans and plans with many missing values, our final sample contains 81 large state-administered plans for the period 2001–14. Put together, our sample represents all U.S. states except Washington,2 and it accounts for 72.6 percent of the total assets of U.S. public pensions in 2013.
designed to reflect a pension plan’s long-run average return, not short-term fluctuations. Second, even when the discount rate does change under certain circumstances, it often changes by a relatively small amount and typically by a notch of 25 basis points. Given these two features, it might be inappropriate to treat the return assumption as a true continuous variable that varies freely, as in previous studies (e.g., Brooks 2017; Stalebrink 2014). Traditional linear regression models (ordinary least squares or panel data) are useful to explain why the rate of return assumption varied among different pension plans, but they cannot tell us anything about what caused the assumptions to change from one level to another.
0
variable—the state ideology score—was retrieved from Berry et al. (1998).
Increased
Figure 3 Cumulative Incidence Functions for Reduced and Increased Return Assumptions
Table 2 Changes in Investment Return Assumptions by Year, 2002–14 Change/Year
02
03
04
05
06
07
08
09
10
11
12
13
14
Total
Percent
No Change
78
73
72
78
76
79
78
77
67
61
68
72
62
941
89.36
2
5
8
3
4
1
2
2
14
20
13
9
14
97
9.21
Reduced Increased Total
1
3
1
0
1
1
1
2
0
0
0
0
5
15
1.42
81
81
81
81
81
81
81
81
81
81
81
81
81
1,053
100.00
6 Public Administration Review • xxxx | xxxx 2018
Table 3 Description of Independent Variables and Expected Signs Variable
Expected Sign
Description
Source
Embeddedness
+
Appointed and ex officio members as a percentage of total board (%)
NASRA & PPD
Union coverage
–
Union coverage of total plan members (%)
U.S. Census
Government ideology
–
Continuous measure of state government’s political ideology: 0 implies the government is extremely conservative, and 100 implies the government is extremely liberal
Berry et al. (1998)
Active ratio
+
Ratio of active to retired members
PPD
Equity share
–
Equities as a percentage of total pension assets (%)
PPD
Return surplus/gap
–
Three-year annualized investment return minus investment return assumption (%)
PPD
Above 8 percent
+
1 = investment assumption higher than 8%, 0 = otherwise
PPD
Budget deficit
–
State budget surplus/deficit per capita in thousands of 2009 dollars
U.S. Census
Table 4 Descriptive Statistics Variable
Obs.
Mean
SD
Min
Dependent variable
Max
Reduced
1,053
.092
.289
0
1
Increased
1,053
.014
.119
0
1
Embeddednesst – 1
1,053
67.132
30.268
0
100
Union coveraget – 1
1,053
39.220
17.586
6.2
75.3
Gov’t ideologyt – 1
1,053
49.049
25.301
0
92.451
Active ratiot – 1
1,047
1.952
.582
Equity sharet – 1
1,051
55.172
10.584
0
Return surplus/gapt – 1
1,022
–2.464
6.342
–18.01
Above 8 percentt – 1
1,053
.263
.441
891
.181
1.386
Independent variable
Budget deficitt – 1
.290
0 –5.263
4.569 76.2 10.600 1 8.667
the Supporting Information online provides more details about an EHA model with recurrent events and competing risks. Estimation Results Table 5 presents the main empirical results of our event history analysis. Panel I displays the results of fitting a standard Cox regression estimated on the gap times using the Stata program. To account for the possible dependence among the gap times from the same pension plan, we clustered all plans in the same state, and the results reported in panel I are robust z-statistics. Panel II of table 5 reports the estimation results from a Cox regression with a “shared frailty” component using a profile likelihood method. For both panels I and II, the Efron method was used to handle ties because the number of events (reduction or increase) relative to the number at risk exceeded 15 percent in 2010, 2011, 2012, and 2014 for our sample. The last two panels of table 5 show the results of the proportional “subdistribution” hazards model for competing risks. As indicated by the Wald statistics, each of the first three models reported in table 5 was jointly significant at a very small confidence level, implying that we gained significantly more knowledge about the assumption events from these models than from a null model. Overall, the first three models of table 5 generated similar results. The last model, while explaining the event of an assumption increase, also produced results that are consistent with the first three models, although the Wald statistics were statistically insignificant most likely because of the very limited occurrence of assumption
increase events. In the following discussion, we mainly rely on the hazard ratios (panel I and panel II) and the subdistribution hazard ratios (panel III) for interpretation. The EHA estimation results provide strong empirical support for hypothesis 1: holding other things constant, more politically embedded pension boards had a higher hazard ratio to reduce the return assumption. This outcome suggests that political appointees and ex officio trustees were more conservative in managing public pension funds and more willing to reduce a plan’s return assumption when the return expectation was not met. In contrast, the result in panel IV of table 5 indicates that politically embedded trustees may also have been less inclined to increase the discount rate, which is consistent with previous results in panels I–III. But, as stated earlier, this last result was not statistically significant. Results for other two independent variables in our study were mixed and mostly not significant. Union coverage did not seem to play a significant role in the decision to change the discount rate. Additionally, a plan in a more liberal state seemed to be more likely to increase its discount rate. But a plan in a more conservative state was not more likely to reduce its discount rate. To account for a possible quadratic effect of the political embeddedness of a pension board and its interaction effects with other independent variables, we also estimated one model with a squared term of political embeddedness and two interaction models; the results are reported in table 6. In the first model, the squared term of political embeddedness was not significant, suggesting that the political embeddedness of the pension board was not quadratically associated with the possibility of reducing discount rate assumption. In the second and third models, the interaction terms between embeddedness and union coverage and between embeddedness and government ideology were both positive and barely significant. However, the magnitudes of both interaction terms were very small, and the level terms of embeddedness turned insignificant after the inclusion of interaction terms. Thus, the conclusion is that the data did not provide strong support for hypothesis 2b and hypothesis 3b. As for the control variables, most were significant and consistent with our predictions. When the ratio of active members to retirees was higher—meaning that the plan could rely on more pension contributions coming in the future—the odds for reducing the discount rate were also significantly higher. Pension plans were 12 percent to 13 percent less likely to reduce their discount rate for every one-percentage point increase in the
Political Embeddedness of Public Pension Governance: An Event History Analysis of Discount Rate Changes 7
Table 5 Event History Analysis Results, 2001–14 (I) Cox Regression: Gap Times
(II) Cox Regression: Shared Frailty
Robust z
Hazard Ratio
Robust z
Hazard Ratio
Embeddednesst – 1
1.010**
Union coveraget – 1
1.005
(.67)
1.007
(.83)
Gov’t ideologyt – 1
1.002
(.34)
1.002
(.46)
Active ratiot – 1
2.465***
(3.38)
2.686***
(3.39)
Equity sharet – 1
.995
(–.75)
.999
Return surplus/gapt – 1
***
.875
4.923*** .793***
Above 8 percentt – 1 Budget deficitt – 1
(2.50)
1.011**
(2.37)
(III) Comp. Risk: Reduced = 1
(IV) Comp. Risks: Increased = 1
Robust z
SHR 1.008**
Robust z
SHR
(2.08)
.998
(–.14)
1.006
(.93)
1.004
(.25)
1.000
(–.06)
1.038
(1.99)
1.865**
(2.26)
3.820***
(2.13)
(–.13)
.994
(–.90)
.998
(–.11)
(–4.67)
.876
***
(–5.05)
***
.887
(–4.93)
1.039
(.96)
(5.89)
6.070***
(6.10)
4.338***
(6.02)
.696
(–.46)
(–2.87)
.800**
(–2.30)
.822**
(–2.55)
1.029
(.26)
N
877
877
877
Wald Chi-square
127.89***
103.83***
174.28***
**
877 9.45
Notes: Robust z-statistics in parentheses were obtained by clustering plan IDs. The Efron approximation method was used to handle ties in the Cox regression models. *p
