Public Sector Unions, Partisanship, and Pensions in the U.S. States Carolyn Abott∗ Princeton University January 17, 2015

Abstract State-sponsored public sector pensions, and how well they are (or are not) funded, have become the subject of intense controversy over the last six years. The problem in the funding of liabilities has been a problem for far longer than just since the Great Recession, however. This paper investigates the determination of liabilities and funding levels of these pension funds, emphasizing the role that legislative partisanship plays and does not play in these policy decisions.

∗

Draft prepared for the Southern Political Science Association’s 86th annual meeting. This work would not have been possible without the help of Nolan McCarty and a generous grant from the Russell Sage Foundation. Additional thanks go to Alex Bolton, Chuck Cameron, Ben Fifield, Mary Kroeger, and Lauren Mattioli.

1

1

Introduction

Open any major national newspaper today and chances are you will be able to find some horror story about the fiscal ravages of public sector pension plans. Over the last ten years, so the story goes, the combination of a weak financial market (Baker, 2011) and state governments’ failure of political will and fiscal responsibility has decimated the assets of hundreds of defined benefit plans across the country. Coupled with expanding liabilities for greedy, out-of-touch public sector workers, the state-run public sector pension plans have become an intense financial, political, and fiscal burden on already struggling state budgets. New Jersey’s pension crisis, for example, is a classic example of the backroom political maneuvering meant to short-change the retirement systems by putting off costs to the future. As Lyman & Walsh (2014) relates, New Jersey began diverting pension payments as far back as 1992. Most infamously egregious of these diversions was Governor Christine Whitman’s use of the payments to subsidize a 30 percent income tax decrease in 1994. 1997 saw the chickens beginning to come home to roost; but rather than increase contributions to the pension fund, Whitman issued $2.75 billion in pension obligation bonds to pay off the liabilities (Mysak, 2014). As a result, New Jersey became the first state in the union to be accused of securities fraud by the S.E.C. in 2010 (Walsh, 2010). Governor Chris Christie promised to make sweeping overhauls to the pension system in 2011, but according to Lyman & Walsh (2014), ...while the pension cuts helped lower the cost of the system, the state also created a new, 38-year funding schedule that began with no payment for one year. That was followed by a seven-year interlude, called “the ramp” during which the state would gradually work its way up to proper funding. Under the law, New Jersey does not have to start making the annual contributions that its actuaries say are required until 2018; it will have until 2048 to pay down its unfunded liabilities. 2

But by 2018, the state itself forecasts, its system will have become shakier, with a funded ratio of just 52.3 percent, down from 2010, because its contributions will have trailed far behind the cost of the plan during the seven-year “ramp.” That missing money must be made up, with interest, in subsequent years, meaning the overhaul will have increased the long-term cost of the system. Though New Jersey’s experience is illuminating, the truth of the matter is that we do not systematically know who or what is responsible for decisions to expand or contract liabilities and decisions to expand or contract assets - policies that directly bear on both the short-run and long-run sustainability of pensions funds and state budgets. What I set out to do here is to try to break apart individual policy decisions relating to the status of pension fiscal positions in order to assemble a larger picture of how and why some states have been able to keep liabilities and assets at an even keel while others are circling the fiscal drain. My analysis suggests that while polarized state legislatures under Democratic administrations are associated with increases in benefit generosity and cost of living adjustments (COLAs), Democrats cannot be otherwise held accountable for lack of legislative restraint in expanding liabilities. Further, contrary to conventional wisdom, unionization and high concentrations of state workers are not associated with increases in expansionary legislative pension policy. Work remains to be done on the administrative and executive side of policy decision-making before a final judgment can be made on the efficacy of partisans and unions on fiscal sustainability, however.

2 2.1

Theory The weak predictions

As Anzia & Moe (2013) and Anzia & Moe (2012) discuss, there are no robust theoretical predictions about how interest groups (e.g., unions), governments, or voters should respond 3

to the underfunding of public sector pension plans. On the one hand, unions want to extract as many concessions as they can from their employers and the tax payers by expanding liabilities1 , but they should also care about the stability and fundedness of the plan, and the possibility that the pension may go bankrupt if mismanaged badly enough 2 . In other words, unions care about the financial positions of their pension funds - how many assets are socked away relative to the size of their benefits. In a perfect world, unions would expand assets at a rate comparable with their liabilities. In the real world, however, unions face scarcity and may be forced to make trade offs between expanding their liabilities and maintaining a given net financial position of the pensions. Governments, both Republican and Democrat, may also have conflicting incentives for adequately funding state pension plans. Because long-term liabilities do not enter the general budgets of states, governments can use pensions as types of piggy banks - both for legal reasons (to circumvent balanced budget laws) and for political reasons (to increase spending without increasing taxes, a win-win for everyone). As The Institute The Institute for Truth in Accounting (2009) explains, Because state budgets are calculated using cash-based measures, only the pension contributions paid to the plans are included in state budgets. The budgets only include the pension contributions legislators decide they want to pay...[they] have nothing to do with the amount of retirement benefits earned by the workers during the budget period. Consequently a state budget calculation may not recognize billions of dollars of retirement costs incurred each year, yet the state is deemed balanced even though current revenue is not set aside to adequately fund these promises. (p. 30) 1

Or liabilities per worker. This is another unsettled theoretical point that arises from the union literature. It is not unreasonable to imagine that public unions never believed that they might get the short end of the stick if a state ran into financial trouble - but contemporary events in New Jersey, Rhode Island, Detroit, and elsewhere have made clear that pensioners can be at risk of abrogated contracts and do not necessarily have the first lien on a bankrupt municipality. 2

4

But again, on the other hand, if state pensions were to be so underfunded as to provoke crisis, there is no doubt that their managers - the government - would be held electorally responsible. Furthermore, Republicans might hold an ideological commitment to adequately funding the plans, as not doing so is the moral equivalent of running perpetual deficits3 . Finally, Democrats may have an incentive to keep the plans well funded in order to keep their union-member constituents happy4 . Lastly, it is also unclear as to how voters should react to underfunded pension plans. Those who believe government workers lead cushy lives and are overpaid would want to contract liabilities - potentially increasing the funding rates - but might even want the pension to go bust altogether - drastically decreasing the funding rates. Others may believe that providing adequate compensation to public employees is the best way to guarantee the provision of good public services. Ultimately, however, most voters probably have no sense of the state of public pensions - especially if they are not in public crisis (pre-2008) and especially if governments are systematically lying about how underfunded the plans are.

2.2

The strong(er) predictions

As suggested in the previous section, theoretical predictions about funding rates of public sector pensions funds are still murky. Part of the reason for this is the cross-cutting of incentives for multiple actors, but an equally significant part of the reason for this has to do with the fact that funding rates are a consequence of many, many decisions made by both employee and employer. Here I try to break down the funding ratios into smaller, more testable components of financial and political decisions. 3 Certain Republicans might also hold an ideological commitment to running specific sectors of the government out of business, but my work assumes that no actors want to purposefully force a public entity into bankruptcy. 4 I am working on building a formal model that more rigorously considers these ideas.

5

Statutory decisions Most of the decisions made regarding the fiscal and financial health of the state pension funds are done at the legislative level, by statute. For these decisions, the primary negotiating takes place in a traditional lawmaking environment and amongst individual legislators or across parties. Unions, however, have the ability to influence the final outcome of the law much in the same way that any type of interest group can influence policymaking. For these types of statutory decisions, then, we would expect the right to collective bargaining to have little to no direct influence on how favorable the outcome is for the union. The electoral and financial resources of the union should be the main tools through which public sector workers exert their power on legislative decisions. In my data, I can investigate the relationship between the legislature and public sector unions and its impact on the following policy outcomes: • Changes in benefit replacement rates and the Cost of Living Adjustment (COLA), • changes in vesting ages and eligibility requirements, and • size of state payments into pension funds.

Administrative/executive decisions There are, however, also decisions made at the executive and administrative level that have an impact on the size of pension liabilities and overall fundedness of state funds. These policy choices are made via hiring decisions at the administrative level and via collective bargaining at the executive level. Unions, of course, play a role in these negotiations in all states and sectors, but it is reasonable to imagine that they exert greater leverage when the organizations are protected by collective bargaining laws. Electoral and financial resources likely still play a role in union power, but negotiations here may hinge more on collective bargaining strength. The policy decisions I have identified in this arena are:

6

• Changes in the size of the active workforce, • changes in total payroll, • changes in salary per active worker, and • weakened accountability via changes in actuarial assumptions (indirect test). Though this paper will focus only on the statutory side of pension policy-making, these administrative and executive side decisions are important to keep in mind, and to note that legislative decisions are surely not occurring without regard to other branch activity.

3

Data

Data used to investigate the relationship between union power, partisan politics, and pension policy were collected from a number of disparate sources.

3.1

Dependent variables

Statutory changes to benefit replacement rates, COLAs, vesting periods, and eligibility requirements The National Conference of State Legislatures (National Conference of State Legislatures (NCSL), 1999-2011) provides extensive summaries of all passed state legislation relating to pension reform for each year from 1999 to 2011. Data were hardcoded as {−1, 1} to reflect whether a state passed legislation that would result in an eventual contraction or expansion of liabilities, or 0 if they passed no legislation in the individual policy arena.

Size of state payments Pew (The Pew Center on the States, 2010) provided self-reported actuarial data from the 50

7

states for the years 1997-2008. By using data on self-reported Actuarial Required Contribution numbers and actual legislative contribution rates I was able to back out exactly how much the state paid into their pension funds in a given year.

Size of active workforce, total payroll, and average salary per active worker As part of a larger project, Nolan McCarty and I collected the Comprehensive Annual Financial Reports (CAFRs) of 128 state funded pension plans over 17 years. Workforce and payroll data were hardcoded directly from these reports.

Changes in actuarial assumptions I have two variables in my data that allow me to investigate the impact of politics on (almost always detrimental) changes in actuarial assumptions. The first is data collected from the NCSL legislative summaries that were coded as {0, 1} for changes in governance or funding structure. The second comes out of data collected and analyzed from the CAFRs. By collecting a number of key variables from the CAFRs we were able to construct a data set that accurately and reliably reflects the liability and asset status of states’ public pension plans, and does not force us to rely on states’ self-reported, potentially biased figures5 . Taking these newly calculated liability, asset, and funding rate numbers, I contrast them with self-reported data taken from the Center for Retirement Research’s Public Plans Database (Public Plans Database, 2001-2010). The distance between these two measures - the revised and the self-reported - gives us a measure of how much the administration is “lying” to state workers and the public about how well funded the pensions are. 5

See appendix for a fuller description of this data and process.

8

3.2

Independent variables

Public sector unionization rates and state worker percentages Both state public sector unionization rates and state worker percentages were calculated from survey data taken from the Current Population Survey (United States Census Bureau and Bureau of Labor Statistics, 1995-2012). The first measures the percent of state government workers who belong to a union while the second measures the percent of all full-time employed individuals who work for the state government.

Governor partisanship and legislature partisanship Party of the governor and party composition of the state legislatures were taken from NCSL. Party composition is operationalized here by the percent of Democrats across both houses of the legislature.

State legislature polarization Polarization data was provided by Nolan McCarty and Boris Shor (Shor, 2014). The metric I use in my analysis is average polarization across both chambers of the state legislature.

3.3

Control variables

Confidence range of unionization rate Because the sample sizes of the cells in the CPS used to calculate the state-year level rates of unionization amongst state government workers could be quite small for certain states (usually for the smallest states in the country), unionization rates were more reliable for some state-years than for others. I try to control for this potential measurement error by including the size of the 95% confidence intervals for unionization rates in my analysis.

9

Unemployment and per capita personal income The unemployment rate for each state-year was obtained from the Bureau of Labor Statistics (United States Bureau of Labor Statistics, 1995-2012); per capital personal income comes from the Bureau of Economic Analysis (United States Bureau of Economic Analaysis, 19952012).

Manufacturing concentration in 1963 and private sector unionization rates in 1964 In order to control for unobserved state specific variables like hostility or friendliness to labor in general, I tried to find variables that would be as plausibly exogenous as possible to the current political and fiscal climate but still say something about long-lasting, semipermanent state-level attitudes towards labor. To this end I obtained the 1963 Census of Manufactures (United States Census Bureau, 1963) and hard coded data on manufacturing industry value-added to the state and 1963 levels of GSP from the Bureau of Economic Analysis6 to construct a measure of manufacturing concentration. I also use private sector unionization rates downloaded from the Union Membership and Coverage Database from the CPS (Hirsch et al. , 2001).

Total debt and budget surplus All state budget data were obtained from the Census of Governments Annual Survey of State Government Finances (United States Census Bureau, 1995-2012).

6

1963 is as far back as the gross state product accounts extend.

10

0.4

Figure 1: Average Pension Reform Involving Benefit Changes Across States, by Year

0.0 -0.4

-0.2

Average Legislation Change

0.2

Employer Contributions Employee Contributions Benefits COLA

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Year Figure 1: Increases (decreases) on the graph correspond with increases (decreases) in benefit allowances, increases (decreases) in COLAs, and decreases (increases) in contribution rates, all of which are expected to expand (contract) liabilities and worsen (improve) the pensions’ fiscal positions.

11

Figure 2: Average Pension Reform Involving Eligibility Requirements Across States, by Year

0.2 0.0 -0.4

-0.2

Average Legislation Change

0.4

Vesting Requirements Eligible Members

1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Year Figure 2: Increases (decreases) correspond with loosening (tightening) of eligibility requirements and decreases (increases) in vesting periods which are expected to expand (contract) liabilities and worsen (improve) the pensions’ fiscal positions.

12

0.030

Average Payment ($ Millions) Ratio of Avg. Payment to Self-Reported Liabs Ratio of Avg. Payment to Revised Liabs

0.010 0.015 0.020 0.025 Ratio of Average Payment to Liabilities

1000 900 800

0.005

700 500

0.000

600

Average Payment ($ Millions)

1100

1200

Figure 3: Average State Legislature Payment into Pension Fund Across States, by Year

1997

1999

2001

2003

2005

2007

Year Figure 3: Average payment from state legislature into pension funds across states, by year. Left-hand axis corresponds with absolute average payment, while the right-hand axis corresponds with the ratio of average payment to different measures of pension liabilities.

13

4

Analysis

The literature on the politics of budgets, much like the literature on public sector unions, has been relatively dormant for the past thirty years, with the notable exception of Anzia and Moe’s work, which has led the charge in reviving the research agenda. Anzia & Moe (2012) claim that public sector unionization directly increases the costs of doing (governmental) business, at least on the local level. Using Novy-Marx & Rauh (2011)’s revised data on 2009 pension liabilities, they also find that unionization increases both liabilities and unfunded liabilities per capita and per GSP. I argue, however, that their analysis falls short in that 1) their data is only cross-sectional, 2) the liabilities include local liabilities, which the state cannot be held politically accountable for, 3) their measure of unionization includes all unionized public sector workers (rather than just unionized state workers), and 4) politics does not enter into the analysis in any real substantive sense at all7 . I add to their analysis by running a series of more realistic models. 7

Anzia & Moe (2013) take a closer look at the politics behind pension legislation, finding that Republicans and Democrats alike vote to expand liabilities pre-2008, and that it was not until the Great Recession that Republicans began to say no to the benefit expansions. While helpful, this analysis does not take into account legislatures’ contributions to the asset side of the pensions or the size of the legislative benefit expansions, essentially making a comparison of outcomes impossible.

14

4.1

Benefits and COLAs

I first look at a fixed effects linear probability model that predicts the likelihood of combined benefit and COLA increases enacted by legislation. This model is specified as

Bst = β1 Ust + β2 DemGovst + β3 SWst + β4 P olarst + β5 U niGovst + β6 DemsLegst +β7 DemGovst ∗ Ust + β8 DemGovst ∗ SWst + β9 DemGovst ∗ P olarst + β9 U niGovst ∗ P olarst +β10 U niGovst + β11 Ust ∗ DemsLegst + β12 SWst ∗ DemsLegst +β12 Ust ∗ SWst + β13 U niGovst ∗ P olarst + γXst + νs,t,r + +st

where Bst is combined benefit and COLA increase in state s and in year t and can take on a value of {−2, −1, 0, 1, 2}; Ust and SWst are rates of unionization and state worker concentration, respectively, in state s and year t; DemGovst and U niGovst each take the value of 1 if Democrats control the governorship or there is unified government, respectively, in state s and year t; and DemsLegst and P olarst are the percentage of democrats in the state legislature and the average magnitude of polarization across the two state legislative chambers, respectively. Xst is a vector of control variables described above and. All standard errors are clustered by state. The model shows promise. One of the most robust findings across the three specifications of this first model, as displayed in Table 1, is the relationship between polarization and Democratic governors. As visualized in Figure 4 and Figure 5, the likelihood of passing expansionary legislation increases with polarization under a Democratic governor and divided government, and it decreases under a non-Democratic governor and divided government. This finding is in line with theoretical expectations about polarization: by paralyzing legislatures, polarization can allow executives to to bend policy more towards their ideal

15

point. Another robust finding across the three models in Table 1 is the overall null effect of the concentration of stateworkers on the likelihood of passing expansionary legislation. This effect is plotted in Figure 6. At low levels of concentrations of Democrats in legislative office, the state workforce has a positive impact on the likelihood of passing expansionary benefit legislation, but quickly decreases as Democratic percentages increase and eventually becomes statistically insignificant over the largest range of the data. This may seem slightly counterintuitive, but I suggest that this may be due to the fact that benefit and COLA decreases never 8 impact current employees. It is possible that union and stateworker electoral forces are in fact agnostic about benefit cuts to the extent that it does not affect their own benefits and potentially improves the long-run fiscal position of their pensions.

4.2

Vesting and eligibility

The model for predicting changes in vesting and eligibility requirements is similarly specified. Table 2 contains the results of the regressions. Unfortunately, these regressions were not as promising. Part of the reason for finding largely null effects may be because there is so little movement in the dependent variable: 73% of the observations were recorded as 0.

8

This has begun to change over the past few years, but these events have occurred outside of the sample period. All recent attempts to cut benefit rates or COLAs for current employees have been subject to numerous lawsuits, still in progress.

16

Figure 4: Marginal Effect of Polarization on Likelihood of Expansionary Legislation

Polarization w/Dem Gov Polarization w.o./Dem Gov. 0.4

Marginal Effect

0.2

0.0

-0.2

-0.4

0.5

1.0

1.5

2.0

2.5

3.0

Polarization

Figure 4: Effects plotted from Model 1 of Table 1. Other variables are held at their mean. Divided government takes the value of the median in the data set, which is 1.

17

Figure 5: Marginal Effect of Democratic Governors on Likelihood of Expansionary Legislation as Polarization Increases

2.0

Marginal Effect

1.5

1.0

0.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

Polarization

Figure 5: Effects plotted from Model 1 of Table 1. Other variables are held at their mean.

18

Figure 6: Marginal Effect of Stateworkers on Likelihood of Expansionary Legislation as Concentration of Legislative Democrats Increases 15

10

Marginal Effect

5

0

-5

-10

0.2

0.4

0.6

0.8

Percentage of Democrats

Figure 6: Effects plotted from Model 1 of Table 1. Other variables are held at their mean.

19

Table 1: Benefit and COLA Increases Model 1 Unionization Rate Dem. Governor Stateworker Percentage Avg. Polarization Unified Govt. Perc. Dems in Legislature Dem. Governor * Union Rate Dem. Governor * Stateworker Perc. Avg. Polarization * Unified Govt. Dem. Governor * Avg. Polarization Unified Dem. Govt. Union Rate * Perc. Dems Stateworker Perc. * Perc. Dems Union Rate * Stateworker Perc. Unified Dem. Govt. * Avg. Polarization Year Year fixed effects Region fixed effects State fixed effects Controls R2 Adj. R2 Num. obs. Years States

Model 2

Model 3

1.843∗∗ 0.812 1.299 (0.727) (1.358) (1.985) ∗∗∗ −0.770 −0.350 −0.557 (0.267) (0.348) (0.407) ∗∗∗ 14.492 3.895 18.514∗∗ (4.655) (9.121) (7.845) ∗∗∗ ∗ −0.279 −0.533 −0.182 (0.058) (0.290) (0.311) −0.033 0.289 0.458 (0.157) (0.299) (0.421) 1.967∗∗∗ 1.415 2.516∗∗ (0.607) (1.219) (1.127) 0.174 −0.340 0.044 (0.244) (0.293) (0.319) 1.113 0.116 −1.461 (2.586) (3.743) (3.708) 0.069 −0.070 −0.176 (0.087) (0.169) (0.236) 0.401∗∗∗ 0.341∗∗ 0.385∗ (0.107) (0.173) (0.210) 0.352 −0.189 0.005 (0.294) (0.435) (0.573) −3.118∗∗∗ −0.071 −2.149 (0.805) (1.814) (1.948) ∗∗∗ −26.128 −26.723 −44.567∗∗∗ (7.617) (18.951) (16.487) 2.488 24.119 22.730 (6.322) (18.990) (22.678) −0.259 −0.047 −0.081 (0.177) (0.260) (0.337) ∗∗∗ −0.048 −0.075 (0.008) (0.046) X X X X

0.270 0.198 422 1999-2011 48

0.267 0.158 502 1999-2011 49

X X 0.312 0.173 422 1999-2011 48

Table 1: *** p < 0.01, ** p < 0.05, * p < 0.01. Estimated intercept not reported. State-clustered standard errors reported in parentheses. See text for description of controls and data sources.

Table 2: Vesting and Eligibility Decreases Model 1 Unionization Rate Dem. Governor Stateworker Percentage Avg. Polarization Unified Govt. Perc. Dems in Legislature Dem. Governor * Union Rate Dem. Governor * Stateworker Perc. Avg. Polarization * Unified Govt. Dem. Governor * Avg. Polarization Unified Dem. Govt. Union Rate * Perc. Dems Stateworker Perc. * Perc. Dems Union Rate * Stateworker Perc. Unified Dem. Govt. * Avg. Polarization Year Year fixed effects Region fixed effects State fixed effects Controls R2 Adj. R2 Num. obs. Years States

Model 2

Model 3

0.317 0.873 0.646 (0.759) (1.461) (2.170) −0.347 0.091 0.132 (0.415) (0.675) (0.716) 5.236 6.956 8.992 (4.861) (12.965) (14.114) −0.186∗ −0.040 0.139 (0.095) (0.329) (0.393) −0.407 0.135 −0.165 (0.262) (0.425) (0.447) −0.286 0.804 0.968 (0.665) (1.564) (1.862) −0.437 −0.245 −0.145 (0.313) (0.343) (0.370) 0.698 4.907 4.601 (3.074) (3.538) (3.948) 0.184 −0.104 0.058 (0.165) (0.273) (0.305) 0.183 −0.193 −0.284 (0.180) (0.366) (0.388) ∗∗∗ 1.091 −0.409 −0.024 (0.376) (0.871) (0.889) 0.235 −1.232 −1.804 (0.835) (2.394) (2.984) −8.584 −16.442 −22.960 (9.304) (25.937) (29.648) −1.096 −6.018 0.968 (8.579) (24.698) (28.693) −0.455∗ 0.316 0.190 (0.255) (0.564) (0.586) ∗∗∗ −0.038 −0.050 (0.011) (0.059) X X X X

0.207 0.131 422 1999-2011 48

0.165 0.043 502 1999-2011 49

X X 0.251 0.102 422 1999-2011 48

Table 2: *** p < 0.01, ** p < 0.05, * p < 0.01. Estimated intercept not reported. State-clustered standard errors reported in parentheses. See text for description of controls and data sources.

4.3

State payments

Finally we come to modeling state payment response. Tables 3 and 4 present OLS regressions of logged payment per liability response for self-reported and revised liability measures, respectively. These models also do not provide too much explanatory power from our independent variables (more of the variation is explained by control variables), though it appears as though unionization has a somewhat robust negative impact on the size of state payments to pension funds, relative to their liabilities. This may be because of cost-shifting by legislatures to well-funded unions; though not described in the reported regressions in Tables 3 and 4, increases in statutory employee contribution rates are indeed associated with decreased payments from state legislatures.

5

Conclusion

Public sector unions are interest groups - but they are not just interest groups. Public sector unions enjoy the unique advantage of being able to both lobby for and determine fiscal outcomes. This is most obviously true in the case of public worker pensions. This paper hopes to have made a first pass at exploring data on fiscal and budget outcomes at the state level and the role that unions and their partisan cohorts might play in their realization.

22

Table 3: Log of State Payment $ per Self-Reported Liability $

Unionization Rate Dem. Governor Stateworker Percentage Avg. Polarization Unified Govt. Perc. Dems in Legislature Dem. Governor * Union Rate Dem. Governor * Stateworker Perc. Avg. Polarization * Unified Govt. Dem. Governor * Avg. Polarization Unified Dem. Govt. Union Rate * Perc. Dems Stateworker Perc. * Perc. Dems Union Rate * Stateworker Perc. Unified Dem. Govt. * Avg. Polarization

Model 1

Model 2

Model 3

−2.427∗ (1.288) −0.386 (0.693) −14.442 (11.684) 0.160 (0.147) −0.247 (0.476) −1.136 (1.319) −0.302 (0.471) 11.755∗∗ (5.292) 0.092 (0.293) −0.250 (0.285) 0.181 (0.751) 2.993∗ (1.639) 10.596 (18.588) 11.725 (12.557) −0.044 (0.431)

−1.818 (1.662) 0.506 (0.492) −3.926 (11.449) 0.181 (0.332) 0.768∗∗ (0.359) −0.325 (1.389) −0.287 (0.326) −0.750 (3.009) −0.485∗∗ (0.219) −0.182 (0.289) −0.862 (0.739) 3.556 (2.451) 12.929 (21.074) −5.261 (19.955) 0.458 (0.443) 0.005 (0.014)

−1.803∗ (1.037) 0.434 (0.364) −11.805 (9.694) 0.236 (0.427) 0.124 (0.312) −2.650∗ (1.377) 0.473∗ (0.250) −1.366 (2.809) −0.196 (0.184) −0.390∗ (0.226) −0.658 (0.499) 3.106 (2.042) 37.934∗ (21.877) −6.061 (23.825) 0.521 (0.331) −0.104 (0.065)

Year Year fixed effects Region fixed effects State fixed effects Controls R2 Adj. R2 Num. obs. Years States

X X X X 0.502 0.443 342 2001-2008 47

0.652 0.596 461 1997-2008 48

X X 0.783 0.727 342 2001-2008 47

Table 3: *** p < 0.01, ** p < 0.05, * p < 0.01. Estimated intercept not reported. State-clustered standard errors reported in parentheses. See text for description of controls and data sources.

Table 4: Log of State Payment $ per Revised Liability $

Unionization Rate Dem. Governor Stateworker Percentage Avg. Polarization Unified Govt. Perc. Dems in Legislature Dem. Governor * Union Rate Dem. Governor * Stateworker Perc. Avg. Polarization * Unified Govt. Dem. Governor * Avg. Polarization Unified Dem. Govt. Union Rate * Perc. Dems Stateworker Perc. * Perc. Dems Union Rate * Stateworker Perc. Unified Dem. Govt. * Avg. Polarization

Model 1

Model 2

Model 3

−1.859 (1.689) −0.098 (0.825) −15.574 (14.591) 0.301 (0.226) −0.212 (0.567) −0.190 (1.779) −0.288 (0.580) 11.180∗ (6.190) 0.065 (0.332) −0.483 (0.370) 0.077 (0.918) 2.531 (2.367) 5.748 (23.169) 14.152 (17.545) 0.056 (0.528)

−3.056 (1.895) 0.965∗ (0.524) −5.606 (12.514) 0.235 (0.319) 0.923∗∗ (0.443) −0.852 (1.454) −0.348 (0.302) −4.500 (3.740) −0.567∗∗ (0.268) −0.336 (0.308) −1.142 (0.869) 3.450 (2.739) 11.812 (23.382) 11.228 (22.550) 0.626 (0.510) −0.021 (0.016)

−2.294∗ (1.306) 0.848∗ (0.381) −16.519∗ (9.541) 0.402 (0.433) 0.616 (0.378) −2.796∗ (1.423) 0.663∗∗∗ (0.183) −3.236 (2.550) −0.491∗∗ (0.247) −0.627∗∗∗ (0.230) −1.535∗∗ (0.622) 3.377 (2.143) 46.270∗ (23.695) −3.389 (24.892) 1.027∗∗ (0.417) −0.073 (0.067)

Year Year fixed effects Region fixed effects State fixed effects Controls R2 Adj. R2 Num. obs. Years States

X X X X 0.454 0.390 342 2001-2008 47

0.708 0.661 461 1997-2008 48

X X 0.822 0.776 342 2001-2008 47

Table 4: *** p < 0.01, ** p < 0.05, * p < 0.01. Estimated intercept not reported. State-clustered standard errors reported in parentheses. See text for description of controls and data sources.

References Anzia, Sarah F, & Moe, Terry M. 2012. Public sector unions and the costs of government. Unpublished draft (May). Available at http://papers. ssrn. com/sol3/papers. cfm. Anzia, Sarah F, & Moe, Terry M. 2013. The Politics of Pensions. In: APSA 2013 Annual Meeting Paper. Baker, Dean. 2011. Origins and Severity of the Public Pension Crisis. Center for Economic and Policy Research. Elliott, Douglas J. 2010. State and local pension funding deficits: A primer. The Brookings Institution. Hirsch, Barry T., Macpherson, David A., & Vroman, Wayne G. 2001. Estimates of Union Density by State. Monthly Labor Review, 124(7). Lyman, Rick, & Walsh, Mary Williams. 2014. Public Pension Tabs Multiply as States Defer Costs and Hard Choices. New York Times, February 25, A12. Mysak, Joe. 2014. Don’t Buy Into Pension Obligation Bond Plans: Joe Mysak - Bloomberg. Bloomberg, January 23. National

Conference

of

State

Legislatures

(NCSL).

1999-2011.

http://hdl.handle.net/1902.1/12650 V2 [Version]. Novy-Marx, Robert, & Rauh, Joshua. 2011. Public pension promises: how big are they and what are they worth? The Journal of Finance, 66(4), 1211–1249. Public Plans Database. 2001-2010. Center for Retirement Research at Boston College and Center for State and Local Government Excellence.

25

Shor, ping

Boris. 2014. of

American

July 2014 Update:

Aggregate Data for Ideological Maphttp://dx.doi.org/10.7910/DVN/26799

Legislatures.

UNF:5:OVOSsr9rJl25lUk4Ck51TA== Harvard Dataverse Network [Distributor] V1 [Version]. The Institute for Truth in Accounting. 2009. The Truth About Balanced Budgets: A Fifty State Study. The Pew Center on the States. 2010. The trillion dollar gap: Underfunded state retirement systems and the roads to reform. United States Bureau of Economic Analaysis. 1995-2012. Regional Economic Accounts. http://www.bea.gov/regional/. United States Bureau of Labor Statistics. 1995-2012. BLS Statistics on Unemployment. http://www.bls.gov/bls/unemployment.htm. United States Census Bureau. 1963.

Annual Survey of Manufactures (ASM).

http://www.census.gov/manufacturing/asm/. United States Census Bureau. 1995-2012. Census of Governments Annual Survey of State Government Finances. https://www.census.gov/govs/state/. United States Census Bureau and Bureau of Labor Statistics. 1995-2012. Current Population Survey (CPS). http://www.census.gov/cps/. Walsh, Mary Williams. 2010. Pension Fraud by New Jersey Is Cited by SEC - The New York Times. New York Times, August 19, A1.

26

A

Appendix

Appendix A One of the greatest challenges of this larger project was to construct a data set that accurately reflected the liability and asset status of states’ public pension plans. With the help of a grant from the Russell Sage Foundation, Nolan McCarty and I set out to collect the Comprehensive Annual Financial Reports (CAFRs) of 128 state funded9 pension plans over 17 years. Following a procedure outlined by Novy-Marx & Rauh (2011) that they used to revise 2009 liabilities, we used these CAFRs to hard-code data on benefit levels, cost of living adjustments, salary structure, employee age and experience distributions, and current annuitant numbers and allowances. Using these (relatively) few data we were then able to construct a standardized measure of liabilities across pension membership tiers, plans, states, and years. Again following Novy-Marx & Rauh we used the Accumulated Benefit Obligation (ABO) as a measure of plan liability, the most conservative of the standard actuarial formulations. This obligation is essentially a measure of what the plan would owe if all of the employees stopped working today. An easy analogy would be if the employer (i.e., the school system in the case of a teachers pension plan) went belly-up today, but was still responsible for paying off the pension benefits that their employees had accrued. This construction of the liability is convenient because it makes the minimum number of assumptions necessary to get a picture of the plan’s obligations - we do not need to guess about termination rates or wage increases, for instance. Current annuitants would continue receiving what they receive today, while active employees would receive only what they have 9

Plans that were at least partially funded by contributions from the state legislature were included in the collection. Municipal plans that receive no revenue from the state government were not included, even if they were administered by the state. Many large and prominent plans were excluded using this standard (including Teachers’ Retirement System of New York City and Detroit’s municipal pension funds), but the idea is to look at funds that are the direct and clear responsibility of the state legislature, even if there are scenarios in which the state bails out local funds.

27

already earned up to today, beginning when they reach normal retirement age

10

.

Once the data is collected, the construction of the ABO happens in four main stages: 1. Calculate averages of active workers and annuitants by age and years of service per year across regions (west, midwest, south, northeast) and/or across plan types (judges, general/mixed, teachers, police and firemen, elected officials) 2. Construct distributions (weights) 3. Calculate individual ABOs • Match plan-year to weights by region and year and/or plan types and year • ABO for active members = plan-year benefit factor * years of service * weight for age-service salary cell * plan-year average salary • ABO for annuitants = weight for age benefit cell * plan-year average benefit 4. Calculate discounted plan ABOs • Obtain life expectancy data (mortality tables) • Calculate expected years of remaining annuity payments • Pick discount rate11 and COLA12 • Calculate individual discounted ABOs 10

Assumed here to be 60. I plan to rerun the simulations assuming that police and firemen retire at 55, teachers at 65, and everyone else at 60. 11 We used the January-of-next-year average of the 10-year Treasury rate, but this is a hotly contested issue. Most plans use the expected return on their assets as a discount rate on their liabilities (which has no basis in economic theory or actuarial practice, but has been encoded in the GASB’s list of recommendations), often at a much-too high rate of 7 or 8 percent. The GASB recently revised their recommendations to suggest that state pension plans use a discount rate (and expected rate of return) that at least partially reflects a high-quality tax-exempt municipal bond rate. 12 We used annual CPI inflation.

28

• Plan discounted ABO = (discounted indiviudal ABO * weight for age-service employee cell * plan-year total active workforce size) + (discounted annuitant ABO * weight for age annuitant cell * plan-year total annuitants) Depending on the level of analysis, the plan ABO can be left as is or again aggregated up to the state level. We also re-adjusted pension assets such that their actuarial value reflected a more reasonable expected rate of return. We collected the market value of net plan assets for each pension-year and then calculated their adjusted actuarial value such that

A1 = A0 + rA0 + s(M1 − (A0 + rA0 ))

where At is the actuarial value in period t, Mt is the market value in period t, r is the expected rate of return on the assets, and s is the smoothing factor (the percent of the difference between market value and expected actuarial value that is incorporated into the current actuarial value). For the most recent calculations, I used a 6.5% rate of return and the standard actuarial assumption of a 20% smoothing rate. All of the assumptions made to revise the actuarial value of pension liabilities and assets are essentially subjective. Some assumptions are more reasonable (i.e., lowering the discount rate) than others (i.e., raising the discount rate), but I cannot claim I am more justified in picking the 10 year Treasury rate over a high grade tax-exempt municipal bond rate, especially when the conversation is still going on (Elliott, 2010). As such, this project does not seek to establish an authoritative measure of state pension liabilities, assets, or overall funding levels; I do not make any claims that plan x needs to raise y dollars in revenue in order to avert financial collapse. What this project does seek to establish, however, is a pecking order of plans’ financial health. While I cannot say with certainty that Texas’s state pensions were only funded at a woeful 37.8% in 2011 (versus the more reasonable 83.0% 29

the state claimed), I can say with certainty that it was 1) not funded as well as the state claimed, 2) better funded than Kentucky (the worst offender for that year, with a revised 20.6% funding level (versus a 50.6% state-reported funding level), and 3) not as well funded as Hawaii (52.2% revised funding level, 59.4% self-reported). By standardizing assumptions and calculations across plans, states, and years and preserving an ordinal relationship between liabilities, assets, and funding levels, I may not be able to say with absolute certainty how bad a state’s financial position is, but I hope to be able to understand why some states are in one worse situations than others.

Appendix B Below are the full year/region fixed effects models, including the controls. Other full models can be provided upon request.

30

Table A1: Likelihood of Passing Expansionary Benefit/COLA Legislation, with Controls and Year and Region Fixed Effects and Clustered Standard Errors

Estimate Intercept 0.85 union.rate 1.84 DemGov −0.77 stateworker.rate 14.49 conf.range −1.12 avg.polar −0.28 not.divided −0.03 avg.dems 1.97 unemp −6.29 va.gsp 0.97 union_64 −0.86 log.lm −0.20 log.debt 0.03 employee.cr 0.13 pc.pinc 0.00 percgensurplus −0.24 ppd.ratio 0.41 union.rate * DemGov 0.17 DemGov * stateworker.rate 1.11 avg.polar * not.divided 0.07 DemGov * avg.polar 0.40 DemGov * not.divided 0.35 union.rate * avg.dems −3.12 stateworker.rate * avg.dems −26.13 union.rate * stateworker.rate 2.49 DemGov * avg.polar * not.divided −0.26 R2 0.270 2 Adj. R 0.198 Num. obs. 422

31

Standard Error 1.08 0.73 0.27 4.66 1.09 0.06 0.16 0.61 2.89 0.26 0.34 0.09 0.04 0.11 0.00 0.28 0.18 0.24 2.59 0.09 0.11 0.29 0.81 7.62 6.32 0.18

t value 0.79 2.54 −2.88 3.11 −1.03 −4.81 −0.21 3.24 −2.17 3.70 −2.50 −2.27 0.77 1.25 1.64 −0.88 2.28 0.72 0.43 0.79 3.73 1.19 −3.87 −3.43 0.39 −1.46

Pr(>|t|) 0.43 0.01 ∗ ∗ 0.00∗∗∗ 0.00∗∗∗ 0.30 0.00∗∗∗ 0.83 0.00∗∗∗ 0.03 ∗ ∗ 0.00∗∗∗ 0.01 ∗ ∗ 0.02 ∗ ∗ 0.44 0.21 0.10 0.38 0.02 ∗ ∗ 0.47 0.67 0.43 0.00∗∗∗ 0.23 0.00∗∗∗ 0.00∗∗∗ 0.69 0.14

Table A2: Likelihood of Passing Expansionary Vesting/Eligibility Legislation, with Controls and Year and Region Fixed Effects and Clustered Standard Errors

Intercept union.rate DemGov stateworker.rate conf.range avg.polar not.divided avg.dems unemp va.gsp union_64 log.lm log.debt pc.pinc percgensurplus ppd.ratio union.rate * DemGov DemGov * stateworker.rate avg.polar * not.divided DemGov * avg.polar DemGov * not.divided union.rate * avg.dems stateworker.rate * avg.dems union.rate * stateworker.rate DemGov * avg.polar * not.divided R2 Adj. R2 Num. obs.

Estimate

Standard Error

−2.08 0.32 −0.35 5.24 −0.69 −0.19 −0.41 −0.29 −2.25 −0.06 −0.15 0.10 0.05 0.00 −1.03 0.31 −0.44 0.70 0.18 0.18 1.09 0.23 −8.58 −1.10 −0.46 0.207 0.131 422

1.39 0.76 0.41 4.86 1.19 0.10 0.26 0.67 2.76 0.35 0.42 0.12 0.05 0.00 0.48 0.18 0.31 3.07 0.17 0.18 0.38 0.83 9.30 8.58 0.26

32

t value −1.50 0.42 −0.84 1.08 −0.58 −1.96 −1.55 −0.43 −0.81 −0.16 −0.36 0.83 1.07 1.48 −2.15 1.74 −1.40 0.23 1.11 1.02 2.90 0.28 −0.92 −0.13 −1.78

Pr(>|t|) 0.13 0.68 0.40 0.28 0.56 0.05∗ 0.12 0.67 0.42 0.87 0.72 0.40 0.28 0.14 0.03 ∗ ∗ 0.08∗ 0.16 0.82 0.27 0.31 0.00∗∗∗ 0.78 0.36 0.90 0.07∗

Table A3: Log of Payments per Self-Reported Liability, with Controls and Year and Region Fixed Effects and Clustered Standard Errors

Estimate Intercept 2.32 union.rate −2.43 DemGov −0.39 stateworker.rate −14.44 conf.range 2.97 avg.polar 0.16 not.divided −0.25 avg.dems −1.14 unemp −6.14 va.gsp 0.76 union_64 −0.49 log.lm −0.05 log.debt −0.08 pc.pinc −0.00 percgensurplus −0.29 ppd.ratio −1.72 no.def −0.39 employee.cr −0.23 employer.cr 0.02 union.rate * DemGov −0.30 DemGov * stateworker.rate 11.76 avg.polar * not.divided 0.09 DemGov * avg.polar −0.25 DemGov * not.divided 0.18 union.rate * avg.dems 2.99 stateworker.rate * avg.dems 10.60 union.rate * stateworker.rate 11.72 DemGov * avg.polar * not.divided −0.04 R2 0.502 2 Adj. R 0.442 Num. obs. 342

33

Standard Error 2.45 1.29 0.69 11.68 1.72 0.15 0.48 1.32 4.68 0.69 0.99 0.25 0.09 0.00 0.57 0.40 0.13 0.09 0.11 0.47 5.29 0.29 0.28 0.75 1.64 18.59 12.56 0.43

t value 0.95 −1.89 −0.56 −1.24 1.73 1.09 −0.52 −0.86 −1.31 1.10 −0.50 −0.20 −0.99 −2.52 −0.51 −4.29 −3.08 −2.55 0.22 −0.64 2.22 0.32 −0.88 0.24 1.83 0.57 0.93 −0.10

Pr(>|t|) 0.34 0.06∗ 0.58 0.22 0.09∗ 0.28 0.60 0.39 0.19 0.27 0.62 0.84 0.32 0.01 ∗ ∗ 0.61 0.00∗∗∗ 0.00∗∗∗ 0.01 ∗ ∗ 0.82 0.52 0.03 ∗ ∗ 0.75 0.38 0.81 0.07∗ 0.57 0.35 0.92

Table A4: Log of Payments per Revised Liability, with Controls and Year and Region Fixed Effects and Clustered Standard Errors

Estimate Intercept 8.43 union.rate −1.86 DemGov −0.10 stateworker.rate −15.57 conf.range 5.71 avg.polar 0.30 not.divided −0.21 avg.dems −0.19 unemp −7.21 va.gsp 0.69 union_64 −1.09 log.lm −0.53 log.debt −0.13 pc.pinc −0.00 percgensurplus 0.12 ppd.ratio −1.86 no.def −0.35 employee.cr −0.21 employer.cr 0.03 union.rate * DemGov −0.29 DemGov * stateworker.rate 11.18 avg.polar * not.divided 0.07 DemGov * avg.polar −0.48 DemGov * not.divided 0.08 union.rate * avg.dems 2.53 stateworker.rate * avg.dems 5.75 union.rate * stateworker.rate 14.15 DemGov * avg.polar * not.divided 0.06 2 R 0.470 2 Adj. R 0.406 Num. obs. 342

34

Standard Error 4.21 1.69 0.82 14.59 2.12 0.23 0.57 1.78 6.11 0.92 1.31 0.38 0.10 0.00 0.63 0.53 0.19 0.10 0.11 0.58 6.19 0.33 0.37 0.92 2.37 23.17 17.55 0.53

t value 2.00 −1.10 −0.12 −1.07 2.69 1.33 −0.38 −0.11 −1.18 0.75 −0.83 −1.38 −1.31 −2.09 0.20 −3.52 −1.86 −2.15 0.29 −0.50 1.81 0.20 −1.31 0.08 1.07 0.25 0.81 0.11

Pr(>|t|) 0.05∗ 0.27 0.91 0.29 0.01 ∗ ∗ 0.18 0.71 0.92 0.24 0.45 0.41 0.17 0.19 0.04 ∗ ∗ 0.84 0.00∗∗∗ 0.06∗ 0.03 ∗ ∗ 0.77 0.62 0.07∗ 0.84 0.19 0.93 0.29 0.80 0.42 0.92