attached are details

Journal of Public Economics 116 (2014) 73–88

Contents lists available at ScienceDirect

Journal of Public Economics

journal homepage: www.elsevier.com/locate/jpube

Defined benefit pension plan distribution decisions by public sector employees☆

Robert L. Clark a, Melinda Sandler Morrill b,⁎, David Vanderweide c

a Poole College of Management, North Carolina State University, USA b Department of Economics, North Carolina State University, USA c Fiscal Research Division, North Carolina General Assembly, USA

☆ This research was supported in part by a grant fro ministration (SSA) funded as part of the Financial Liter opinions and conclusions expressed herein are solely th represent the opinions or policy of SSA, any agency o any other institution with which the authors are affili paper was presented at the “Retirement Benefits for St signing Pension Plans for the Twenty-First Century” c by the Smith Richardson Foundation. The authors wo participants for useful comments, as well as seminar William and Mary. The authors would also like to than ell, Thayer Morrill, Doug Pearce, James Poterba, and Jo and comments. ⁎ Corresponding author at: Department of Economics

sity, Box 8110, Raleigh, NC 27695-8110, USA. Tel.: +1 9 E-mail address: [email protected] (M.S. Mo

0047-2727/$ – see front matter © 2013 Elsevier B.V. All http://dx.doi.org/10.1016/j.jpubeco.2013.05.005

a b s t r a c t

Article history: Received 16 October 2012 Received in revised form 22 February 2013 Accepted 28 May 2013 Available online 11 June 2013

JEL Classification: H7 J4 J3

Keywords: Public employees Defined benefit pensions Distribution choices

Studies examining pension distribution choices have found that the tendency of private-sector workers is to select lump sum distributions instead of life annuities resulting in leakage of retirement savings. In the public sector, defined benefit pensions usually offer lump sum distributions equal to employee contributions, not the present value of the annuity. Thus, for terminating employees that are younger or have shorter tenures, the lump sum distribution amount may exceed the present value of the annuity. We discuss the factors that may influence the choice to withdraw funds or not in this environment. Using administrative data from the North Carolina state and local government retirement systems, we find that over two-thirds of public sector workers under age 50 separating prior to retirement from public plans in North Carolina left their accounts open and did not request a cash distribution from the pension system within one year of separation. Further- more, the evidence suggests many separating workers, particularly those with short tenure, may be forgoing substantial monetary benefits due to lack of knowledge, understanding, or accessibility of benefits. We find no evidence of a bias toward cash distributions for public employees in North Carolina.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

Each year, millions of American workers leave their jobs prior to retirement either by choice or due to termination by their employers. Many of these job changers participate in defined benefit pension plans. On leaving their employers, these workers are often given a choice of keeping their retirement accounts open, thus maintaining a claim on a future life annuity, or accepting an immediate lump sum distribution (LS) of their pension assets. This decision is distinct from that faced upon retirement, since workers maintaining their ac- count will not receive any cash benefits until reaching retirement age.

m the U.S. Social Security Ad- acy Research Consortium. The ose of the authors and do not f the Federal Government, or ated. An earlier version of the ate and Local Employees: De- onference, which was funded uld like to thank conference participants at the College of k Julie Agnew, Olivia S. Mitch- sh Rauh for useful discussions

, North Carolina State Univer- 19 515 0331. rrill).

rights reserved.

Workers who accept the LS are then given a choice of whether they want to roll the funds over into an IRA or to accept the cash as taxable income and also pay a tax penalty for early withdrawal if under age 59.5. These choices can have significant long run implications for fu- ture retirement income and shed light on the magnitude of leakages from retirement saving. In a report describing sources of leakage of workers' retirement savings from 401(k) plans, the Government Accountability Office (2009) concluded that cashing out benefits at job separation represents the principle form of leakage of retirement savings and has the largest impact on retirement wealth accumula- tion. The problem of leakages is greater among younger workers and males, who have been found to cash out benefits at higher rates (AonHewitt, 2011).

Economic theory argues that to maximize lifetime utility one should consume such that utility levels are smooth over time. One method of achieving utility smoothing is through the purchase of annuities (Yaari, 1965). However, a series of national surveys and economic studies found that individuals rarely purchase annuities in the open market (see, e.g., Mitchell et al., 1999). Further, when given the choice in their pension plans of a life annuity or a LS, workers often chose the LS (see, e.g., Brown, 2001; Engelhardt, 2002; Hurd and Panis, 2006). Thus, many workers tend to reject the opportunity to receive a certain flow of income throughout retire- ment in favor of receiving cash now, which therefore results in indi- viduals assuming the task of managing funds on their own during their retirement years. This conflict between theory and individual

74 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

choices has been called the “Annuity Puzzle.”1 If cashed out benefits are spent on immediate consumption rather than saved for retire- ment, this leakage will result in lower income and income security in retirement.

Under federal pension regulations, defined benefit plans in the private sector must offer an annuity and provide participants with in- formation on their future annuities. The LS's for these plans are re- quired to be calculated as at least equal to the present value of the retirement annuity using approved interest and mortality tables.2

Things are very different in the public sector. Public sector defined benefits plans usually require explicit employee contributions each pay period, and LS's are based on the employee contributions and not the present value of the annuity. Thus, a public sector worker's choice of whether to take a LS reflects both his/her individual prefer- ence for annuitization and potential differences in the net value of the LS and annuity options. Defined benefit plans continue to cover most state and local employees and virtually all of the plans offer workers the option of a LS at job separation or retirement (Clark et al., 2011).

We examine the choices terminated workers younger than age 50 make using data from the North Carolina Teachers' and State Em- ployees' Retirement System (TSERS) and the North Carolina Local Governmental Employees' Retirement System (LGERS) by examining information contained in the administrative records of the two retire- ment plans.3 We restrict our attention to individuals under age 50 as they are not yet eligible to retire and receive immediate annuity ben- efits. As of the end of 2010, these two public pension plans covered 803,636 employees and retired workers. Our unique dataset contains all terminations from state and local government employment in North Carolina between 2007 and 2008 and tracks behavior through the end of 2009, allowing us to observe choices made within one year of separation for all terminated workers. The dataset includes relevant economic and demographic information on all individuals who left state or local employment during this time period.

Public sector defined benefit plan participants face a series of choices concerning their pension accounts when terminating employment prior to retirement, as illustrated in Fig. 1. The first decision a worker must make is whether to maintain his/her pension account or accept an immediate LS.4 From an economic perspective, a worker should compare the value of the LS to the present discounted value of the life annuity (PDVA) which is set to begin at some point in the future. How- ever, as we will see later, there are a number of factors that make this decision more complicated that a simple wealth comparison.

The default option is for the worker to maintain the pension ac- count; a departing worker must file a request with the retirement system in order to receive a LS. Depending on the rules of the pension plan, a worker might also have the opportunity to return to work with the same employer and have prior service credits count toward a future retirement benefit.5 Fig. 1 also shows that workers who re- quest a LS must specify whether they want to receive cash or have

1 See Benartzi et al. (2011) for an excellent overview of the annuity puzzle literature. 2 The Pension Protection Act requires that beginning in 2008 the LS be calculated

using a three-segment interest rate yield curve based on the rates of return on invest- ment grade corporate bonds of varying maturities. Purcell (2007) provides additional information on this process and how it affects workers at various ages at termination.

3 While these plans have separate governing boards, they are administered by the same staff and have similar, but not identical, benefits and contributions requirements. Further details of TSERS and LGERS can be found by visiting the retirement systems' home page at: http://www.nctreasurer.com/dsthome/RetirementSystems.

4 The term “maintain” is used because workers may have the opportunity to request a LS at any time after separating from public employment and prior to starting a retire- ment benefit; thus not accepting the immediate LS leaves open the option of requesting such a distribution at some time in the future instead of waiting until one is eligible to start a retirement annuity from the plan. In our analysis, “immediate” means that the terminated worker requested a LS within one year of termination.

5 Returning to work and being covered by the same retirement plan is probably much more likely in the public sector where a single plan typically covers all state em- ployees and teachers in the state. This allows workers to change jobs and government agencies and to move within the state while remaining in the same retirement system.

the funds rolled over into another approved tax qualified retirement plan such as an IRA. If the worker is sent a check, she could subse- quently deposit the funds into an IRA and avoid current taxes and penalties if she follows the IRS guidelines. It is important to remem- ber that individuals preferring to insure against longevity risk by annuitizing have the option to withdraw funds, roll them over into an IRA, and ultimately purchase an annuity. Thus, an informed worker should decide whether to withdraw funds based on the highest pres- ent value of the distribution options, appropriately measured, taking into account predicted inflation, interest rates, and various types of risk.

We calculate how the decisions made by separating workers are affected by the value of the distributional options available to them. The relative generosity of the two options is estimated using details of the plan characteristics and information provided by the retire- ment system. We find that fewer than one-third of all terminating public employees requested a LS within one year of separation, de- spite the finding that for over 70% of terminations, the LS was larger than the estimated PDVA. These results indicate a low probability of leakage from retirement funds, although many workers are seeming- ly forgoing the possibility of higher retirement income possible from rolling over funds to an IRA.

We offer several potential explanations for why the distributional choice from a public pension plan is more complex than a simple wealth comparison at a point in time. First, separating participants in TSERS qualify for retiree health insurance from the State Health Plan with no premium as long as they are receiving a monthly annu- ity from TSERS. This option is available for virtually all vested state employees (participants in TSERS), but local employees (participants in LGERS) are not covered by the State Health Plan.6 Comparing dis- tributional decisions by state employees in TSERS to those of local employees in LGERS provides some indication of the effect of retiree health insurance on the choice to ultimately receive a retirement an- nuity. Despite the difference in coverage of retiree health insurance in the two systems, we do not see a large difference in the distributional choices between separating workers that will qualify for retiree health insurance and those that will not.

Second, we consider the likelihood that terminated participants may plan to return to public employment. The expectation of returning to public employment might make maintaining the account the optimal choice for these individuals. However, we document that workers who ultimately returned to work by December 2010 were actually more likely to withdraw funds within one year of separation. Third, we discuss the influence of alternative investment options, macroeconomic conditions, and confidence in the retirement system. Maintaining the account still allows for the option of requesting a LS at some future date. Because the account balance accrues interest at a guaranteed rate of 4%, financially savvy individuals may choose to maintain their account balances and accept larger LS's at a future date as part of an investment portfolio. However, we do not find that the 12-month return on the S&P 500 is related to the probability of withdrawing funds, once local macroeconomic conditions are added to the model. There are mixed results when considering the 1-year Treasury bond rate, but, if anything, higher bond rates are as- sociated with a reduced probability of withdrawing funds. Moreover, we do not see a large difference between the disposition choice of non-vested workers (who do not earn interest) and vested workers. This indicates that workers are not responding to incentives of out- side investment options. We do find that when the state unemploy- ment rate rises, individuals are significantly less likely to withdraw

6 With a few minor exceptions, workers and retirees covered by LGERS are not cov- ered by the State Health Plan; however, they may be covered by locally-managed health plans that extend coverage to retirees. We cannot match the local health plans to the LGERS retirement data.

Fig. 1. Choices facing separating employees to take a lump sum distribution (LSD) or to maintain their retirement account.

75R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

funds. This could be due to selection into who is separating employ- ment, or it may be that individuals more heavily rely on defaults in times of economic turmoil.

The final explanations we consider for why public sector workers in North Carolina do not withdraw funds at a higher rate are financial literacy, peer effects, and inertia. The default is to leave funds in the system. The behavior we observe is consistent with many individuals accepting the default option and forgoing potentially more valuable benefits. We find evidence consistent with peer effects, but cannot distinguish between correlation in unobserved characteristics and the influence of peers in this study.

After the analysis of the decision to accept a LS rather than maintaining one's account balance, we then examine the decision between cash and a rollover of pension assets by those who opted for a LS. The decision on spending versus saving the LS distribution has received considerable attention by economists; however, only a few studies have been able to observe this choice in administrative records rather than survey data (see Bryant et al., 2011). We find that nearly 90% of separating workers that request a LS elect to re- ceive the funds as cash, rather than rolling over, suggesting a high probability of leakage of retirement funds among these individuals. Of course, individuals who select a cash distribution can still move the funds into a retirement account, pay off debts, or save in non- retirement accounts, rather than spending the money on immediate consumption.

2. Relevance of findings for participants in public and private DB plans

Due to similarity of TSERS and LGERS to other state and local pen- sion plans, the results presented in this analysis should be relevant to understanding worker behavior in most public pension plans. While the findings also will shed light on potential behavior of participants in private defined benefit plans, the difference in pension coverage and plan characteristics between the public and private sectors of the U.S. economy is striking. The Bureau of Labor Statistics (2011a) found that in November 2011 only 22% of full-time private sector workers were participating in defined benefit plans, compared to 87% of full-time public sector workers. With 19.2 million individuals working as state and local employees, this implies that there were approximately 15 million public sector employees participating in defined benefit plans (Bureau of Labor Statistics, 2011b).

Historically, most defined benefit plans provided only annuities to their participants; however, over time increasingly plan sponsors have adopted provisions that allow their retirees to choose between receiving a lump sum distribution (at or prior to retirement) or accepting an annuity at retirement. Under federal pension regula- tions, defined benefit plans in the private sector must offer an annu- ity. Traditional plans typically provide participants with information on the expected monthly payout amount of their future annuity. According to federal guidelines, if a lump sum distribution is offered, it must be at least equal to the present value of the retirement annuity using approved interest and mortality tables. As discussed further below, things are very different in the public sector where plans usually require employee contributions and lump sum distributions are based on the employee contributions and not the present value of the annuity.

The lump sum option is becoming increasingly more common in private sector defined benefit plans. In 1989, only 2% of defined ben- efit plans offered by medium and large firms gave workers the option of taking a lump sum distribution, but by 1997 this number had risen to 23% (Bureau of Labor Statistics, 1990, 1999; also see Moore and Muller, 2002). According to data from the National Compensation Survey, in 2007 52% of all workers were in plans that provided em- ployees the option of selecting a lump sum distribution instead of accepting the life annuity (Bureau of Labor Statistics, 2007; Purcell, 2009). Some of the increase in defined benefit plans allowing lump sum options may be due to the growth of cash balance plans and other hybrid plans that specify the account balance as a lump sum throughout the worker's career. Hybrid plans almost always offer a lump sum option for departing and retiring workers. As more and more pension participants in both the public and private sectors of the economy confront distributional choices in their defined benefit plans, it is increasing important to understand the determinants of this choice and its implications for retirement income.

When making comparisons between choices made in the public and private sector, it is important to note that differences in plan de- sign and pension preferences may reflect, in part, a labor market sorting of workers based on workers' risk preferences. Using survey questions designed to measure risk aversion, previous studies in the United States and Europe indicated that employees in the public sector tend to be more risk averse than those in the private sector (e.g., Bellante and Link, 1981; Bonin et al., 2007; Hartog et al., 2002; Pfeifer, 2010). Similarly, public sector workers tended to choose less risky options in experiments (Buurman et al., 2009), took fewer

76 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

financial risks (Roszkowski and Grable, 2009), and were observed to select jobs that have smaller fluctuations in annual earnings and greater job stability (Bonin et al., 2007). Thus, one should be cautious in extending the findings regarding the determinants of choices made by public sector workers to workers in the private sector with DB plans that allow for lump sum distributions.

3. Previous literature examining lump sum distributions

Economists have long been interested in how workers access and utilize the wealth accumulated in their retirement accounts. For the most part, research studies have focused on (1) the choice between a lump sum distribution (LS) and an annuity at retirement, and (2) the decision by terminating workers who take a LS to accept a cash distribution or roll over the funds over into another tax qualified re- tirement plan. Papers on the first topic are often linked to the ‘annuity puzzle’ and try to explain why economic theory suggests individuals would prefer an annuity but retiring workers largely prefer LS's.7

Most studies on the second point use survey data which relies on in- dividuals' ability to recall whether they spent or saved the LS. There are only a few papers that consider the choice at termination prior to retirement between a LS and maintaining the account balance in defined benefit plans.8

Several previous papers examined the distributional choice at re- tirement in public sector retirement plans and found an important role for defaults and for personal preferences and discount rates. Butler and Teppa (2007) found that retirees in Swiss pension funds were more likely to accept the default option in the plan, leading to what they called an “acquiescence bias.” Warner and Pleeter (2001) estimated the decisions of individuals to leave the US military pen- sion plan during the 1990s in response two separation benefit pack- ages offered to mid-career personnel. They found that many more eligible participants accepted an offer of a lump sum payment than analysts had expected based on conventional interest rates. The au- thors argue that this was due to high personal discount rates relative to the interest rate used to equate the lump sum to the annuity. Chalmers and Reuter (2012) examined the distribution decisions of 32,060 retiring public employees in Oregon between 1990 and 2002. They found that 85% of Oregon state employees choose an an- nuity from the state retirement plan and that retirees respond to changes in annuity pricing, although this effect was relatively small.

Most studies that have examined the choice of a LS have focused on respondents in large national data sets (e.g., SIPP, HRS). Economic theory indicates that workers will compare the cost of purchasing an annuity that is based on population age-specific mortality rates to the present value of the annuity using their personal discount rates and their own life expectancies. Workers who believe that an annuity based on population life expectancy will be less than actuarially fair to them should accept the LS (Hurd and Panis, 2006). Individuals with high personal discount rates will place a lower value on the fu- ture annuity and thus be more likely to accept the LS.

Earlier studies have found that when the value of the cash settle- ment is relatively small, there is a greater likelihood that workers

7 Benartzi et al. (2011) discuss distributions options chosen by workers over age 50 the private sector. They find relatively high rates of annuitization among retiring workers thus providing evidence against the “Annuity Puzzle.” They found that institu- tional arrangements, framing of choices and defaults influence the decision to opt for an annuity or a LS.

8 An expanded model of pension decisions by individuals would include the initial choice of whether to select a DB plan or a DC plan as their primary retirement plan. This choice is a function of investment risk, labor market risk, and a variety of other fac- tors including whether the worker prefers their retirement plan to provide a retire- ment annuity or a LS. In the past decade, a number of states have given newly hired employees this option (Clark et al., 2011). Several recent papers have examined the choice of pension plans in the public sector (e.g., Brown and Weisbenner, 2012; Papke, 2004; Yang, 2005)

will take the LS and not roll these funds over into another retirement account, thus rejecting a future life annuity (Hurd and Panis, 2006; Poterba et al., 1998, 2001; Sabelhaus and Weiner, 1999; Engelhardt, 2002). Hurd and Panis (2006) found that about 20% of terminating workers opted for a LS when available and observed that women were more likely to request cash settlements than men, as were workers with less formal education. However, others find that men are more likely to take LS's (Butler and Teppa, 2007; Purcell, 2009). Older workers were found to be more likely to have rolled over their pension distributions (Burman et al., 1999; Moore and Muller, 2002; Warner and Pleeter, 2001). Many of these previous studies addressed workers' implied discount rates and the effect of high dis- count rates that minimize the value of the future annuity compared to the LS.9 The choice between money now and a lifetime annuity begin- ning years in the future is also influenced by financial literacy. Consid- erable recent evidence indicates that American workers have a relatively low level of financial literacy (Lusardi and Mitchell, 2007). Workers that lack a good understanding of financial mathematics, fi- nancial market risks, and the uncertainty associated with mortality might not be able to determine the optimal choice when faced with important financial decisions such as the choice of a LS versus a future annuity (see, e.g., Clark et al., 2012).

Despite the impressive list of studies that have examined distribu- tional choices and the use of LS's, there remain two important short- comings in these analyses. First, relatively few studies have examined the choices made by public sector employees. This is an important gap because the distributional options in public sector plans quite dif- ferent than those in the private sector, and defined benefit plans are much more prevalent in the public sector. Second, most prior studies used survey data and hence relied on respondents' recall of decisions made years earlier. The reliance on memory of these events undoubt- edly introduces considerable noise. In contrast, our analysis focuses on separations from a public defined benefit plan using administra- tive records showing real-time decisions. Thus, we are able to use ac- tual administrative data on the distributional decision, the timing of separation, and the value of the distributional options.

4. Estimating the value of distribution options in the North Carolina retirement plans

4.1. Background on distributional choices in state-managed defined benefit plans

In public sector defined benefit plans, the default option for departing workers who are vested is to leave their funds in the pen- sion plan and receive an annuity when they have attained the re- quired age for starting benefits.10 All public defined benefit plans that we have examined offer departing workers the option of leaving their funds in the pension system, thus retaining their eligibility to re- ceive a retirement annuity when they reach the specified age and ser- vice requirements of the plan.11 Most public plans allow separated workers to request a lump sum distribution (LS) at any point up until the individual starts the retirement annuity.

9 Other papers that examine the utilization of a LS include Bassett et al. (1998), Chang (1996), Copeland (2009), and Yakoboski (1997). 10 Ultimately, the default is that no benefits are paid. Terminated workers who do not request a lump sum distribution are defaulted into keeping their account open. When workers finally satisfy age and service requirements for a benefit, they still must re- quest that their retirement benefits be paid. No request from the terminated worker means that no benefit is paid. Data from the North Carolina retirement system show that only in a relatively small number of cases did workers leave public employment and never request either a LS or the start of an annuity. 11 Workers who had not yet been employed sufficient years to achieve vesting would have to return to a public job covered by the same pension and work additional years to satisfy the vesting requirements before they would be eligible to receive a future pension benefit.

77R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

As described above in Section 2, private sector defined benefit plans usually do not entail employee contributions.12 In public retire- ment plans, on the other hand, employee contributions are typically required.13 All state retirement plans that require employee contribu- tions for all plan participants offer a LS option to terminated em- ployees. Separating workers in these plans are offered a LS at least equal to their own contributions. In some states, individuals are also awarded interest payments on their contributions. The interest rate varies across states and is often a function of years of service. The av- erage interest rate paid on employee contributions among those states with interest payments is 4% per year with a standard deviation of 2% (Clark and Hanson, 2011). While the empirical analysis of the distribu- tional choices reported in this paper uses administrative records only of North Carolina retirement plans, the choices and plan parameters imbedded in the North Carolina system are similar to those in other retirement plans covering teachers and state employees.14

16 Workers with 30 years of service can receive an unreduced benefit at any age; however, in the sample of separated employees that we examine, there are no individ- uals with 30 years of service. In addition, early retirement benefits are available to em- ployees with 20 years of service at age 50; however, there are substantial reductions for accepting early retirement benefits. The magnitude of the reduction in annual ben- efits for starting benefits prior to reaching the normal retirement age vary with age and years of service. The benefit reductions for TSERS are presented in the employee hand- book page 8, http://www.nctreasurer.com/NR/rdonlyres/223AE566-7BA0-471F-B02C- 0A18ABDB97C0/0/NC_TeaState_070111_Final.pdf. System records indicate that only about 25% of terminated vested workers wait until age 65 to start benefits; however,

4.2. Choices facing separating public sector workers in North Carolina

Employees who leave public employment in North Carolina must decide whether they will accept a LS or leave their pension account open in anticipation of a retirement annuity payable from the re- quired age for retirement benefits. The LS can be known with certain- ty; however, the present value of the annuity involves a more difficult calculation which should reflect personal discount rates, expected in- flation rates, and the likelihood of returning to public employment in North Carolina. In order to understand the choice between accepting a LS and leaving pension assets in the system, one must consider key parameters of the pension plan and how they affect the value of the LS that is available at separation versus the annuity expected in the future. Public sector workers in North Carolina are also covered by Social Security and Medicare.

Like most public retirement plans, both retirement systems in North Carolina (TSERS and LGERS) require employee contributions. The required employee contributions are equal to 6% of total annual salary and are deducted from paychecks every pay period. These con- tributions are deposited in the retirement funds of the two state sys- tems and help finance the benefits for retirees. In both North Carolina systems, vesting occurs when an employee completes five years of service.15 If workers leave public employment prior to being vested, their LS is simply the total of their own contributions to the system; in other words, non-vested separating workers do not receive any interest credited on these contributions and they do not receive any portion of the employer pension contributions. Workers leaving public employment with at least five years of service are offered a LS equal to the total of their own contributions during their employ- ment plus interest credited at 4% per year. As is typical of most state defined benefit plans, even once vested North Carolina public em- ployees who request a LS have no claim on the implied employer con- tributions to the retirement plan.

12 The Bureau of Labor Statistics (2011a) reports that only 4% of workers that were participating in defined benefit plans in the private sector were enrolled in plans that required an employee contribution. 13 According to Clark and Hanson (2011), only a few state retirement plans did not require workers to contribute a portion of their salary in support of the retirement plan. These plans include Arkansas PERS, Connecticut SERS, Florida FRS, Hawaii ERS, Michigan PSERS and SERS, Missouri MSEP, Tennessee CRS, and Utah SRS. In these noncontributory plans, non-vested terminated workers were not eligible to receive a future retirement annuity, nor were they eligible for a LS. The Bureau of Labor Statistics (2011a) reported that 79% of state and local workers that were participating in defined plans were enrolled in plans that require employee contributions with the mean con- tribution rate being 6.5% of earnings. 14 Clark and Hanson (2011) provide a detailed summary of the distribution choices available to teachers and state employees in all 50 states. 15 Legislation in 2011 raised the vesting requirement from five years of service to ten years for all newly hired teachers and state employees; however, in the sample we consider, all workers are covered by the five year vesting provision.

The value of the LS is equal to the total employee contributions (plus interested if vested) and can be calculated directly from the sal- ary history. Appendix A describes the calculation, which is a direct function of starting salary, wage growth, years of service, and the in- terest rate, but is independent of sex or age. Departing employees can learn the value of their LS by checking their account balances on-line or by directly contacting the retirement system. Thus, the value of one's LS can be known with certainty at the time of separation.

The retirement benefit is calculated in a similar fashion to that typically found in the private sector, using a benefit formula based on years of service and final average salary. The systems do not pro- vide an estimate of the present discounted value of the annuity (PDVA) to its members. Using the estimated benefit for each de- parting worker, we derive the present value of a future annuity for vested workers in the North Carolina TSERS or LGERS retirement sys- tem using the same assumptions employed by the retirement plans to calculate their pension liabilities. Details of this calculation are pro- vided in Appendix A. The PDVA is calculated assuming that the indi- vidual will begin receiving an unreduced benefit at age 65.16

To reiterate, the PDVA is not used by the plan to calculate the LS and this value is never provided to terminated workers. Instead, the PDVA variable represents our estimate using the same life tables used by plan actuaries to evaluate the financial status of the plan, along with real and nominal interest rates, described in Appendix A. Workers with higher personal discount rates will place a smaller value on the annuity, as will those that believe that they have lower life expectancies. The value of the annuity is increasing in salary level and the rate of salary growth, as well as number of years of ser- vice, similar to the LS. However, the PDVA is also an increasing func- tion of age at separation. In addition, because women's survival probabilities are higher than men's, women's PDVA will be (slightly) higher than men's with the same annual retirement benefit.17

A potentially important component of the retirement benefits for public workers in North Carolina is that the state will continue to pro- vide health insurance for retirees in the TSERS system at no premium, provided that they are receiving a retirement annuity.18 Benefits are also available for the spouses and dependents of retirees, with only an implicit subsidy.19 For retirees over age 65, Medicare is the prima- ry insurer and the state health plan (SHP) becomes the secondary in- surer. For eligible state employees and teachers, the value of health insurance should be included in the worker's decision whether to

the reduction factors imposed by the system for early retirement mean that this as- sumption does not substantially alter the expected present value of the annuity. In oth- er words, the reduction factors are, on average, approximately actuarially fair. 17 One could instead consider the size and terms of an annuity which could be pur- chased at the time of separation with the funds that are withdrawn. In the private an- nuity market, insurance companies will use similar interest rates to those we use in our calculations, but may also charge some commission and may adjust for adverse selec- tion (see, e.g., Mitchell et al., 1999). Because individuals could chose a variety of annu- ity products, we think framing the discussion in present value dollars using the assumptions adopted by the plan actuaries is a more straightforward comparison. 18 All states have some form of retiree health insurance for their employees; howev- er, the value of these plans differs markedly across the states, see Clark and Morrill (2010). 19 In North Carolina, spouses and dependents must pay the full price of the premium, as calculated based on prior years' expenditures, so we do not consider the value of spousal or dependent benefits. There is an implicit subsidy due to risk pooling. Ac- counting for this would obviously increase the estimated value of the health insurance option.

$5,000

$15,000

$25,000

$35,000

25 30 35 40

Age at Hire

Years of Service: 10

LS

PDVA

$5,000

$15,000

$25,000

$35,000

$45,000

$55,000

$65,000

$75,000

$85,000

$95,000

5 10 15 20

Years of Service

Hired at Age 30

LS

PDVA

Fig. 2. Simulation of relative values of lump sum distribution (LS) versus the present discounted value of the annuity (PDVA). Notes: The values are calculated for a hypo- thetical separating employee earning a starting salary of $30,000 per year with 3% wage growth. The top chart indicates the data points in dollar values calculated for a person with 10 years of service by the age of hire. The bottom chart indicates the data points in dollar values calculated for a person who was hired at age 30 by the number of years of service. The present discounted value of the annuity (PDVA) does not include the value of health insurance. Other simulation results (shown in Table 1) that vary the age of hire and years of service show a clear pattern that the lump sum distribution (LS) is greater relative to the PDVA for those hired at younger ages and with fewer years of service.

Table 1 Value of lump sum (LS) and present discounted value of annuity (PDVA) for a hypo- thetical worker.

Age at Hire

Age at Separation

Years of Service

LS PDVA

25 30 5 $10,741 $4,916 25 35 10 $25,521 $14,858 25 40 15 $45,485 $33,694 25 45 20 $72,074 $67,967 25 50 25 $107,089 $129,212

30 35 5 $10,741 $6,408 30 40 10 $25,521 $19,377 30 45 15 $45,485 $43,972 30 50 20 $72,074 $89,167

35 40 5 $10,741 $8,357 35 45 10 $25,521 $25,287 35 50 15 $45,485 $57,687

40 45 5 $10,741 $10,906 40 50 10 $25,521 $33,174

45 50 5 $10,741 $14,308

Notes: Values are for a hypothetical worker with a starting salary of $30,000 and wage growth of 3%. Calculations of the lump sum distribution value (LS) and present discounted value of the annuity (PDVA) are described in the text and in detail in Appendix A. The shaded boxes highlight the larger value between the LS and PDVA.

78 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

request a lump sum distribution. However, as discussed further below, it is difficult to know how much an individual worker values access to state-provided retiree health insurance. We describe our calculation of the present value of the health insurance option in Appendix A.

The literature on the “annuity puzzle” highlights a potentially im- portant role for health shocks (see, e.g., Brown, 2001). Individuals worried about unanticipated health care expenses may choose to keep assets liquid in order to have funds accessible in the event of a health shock. If a retiree does not have access to employer-provided health insurance, she may worry about the cost of health insurance coverage should an adverse health event occur. Retiree health insur- ance insures against this type of risk, and thus may be even more valuable to workers than the actuarially equivalent value.

4.3. Simulating the relative values of the annuity and lump sum distribution

As the above discussion indicates, the value of the LS depends on years of service but not on age at hire, while the PDVA is higher for those who are closer to the minimum required age of eligibility for retirement benefits. Thus, as employees accumulate years of service, the present value of the annuity will grow relative to the lump sum. The top portion of Fig. 2 illustrates this trend using a hypothetical male worker in the TSERS retirement system with a starting salary of $30,000 who experiences 3% wage growth per year. The values of the LS and PDVA are calculated for various ages of hire for a person with 10 years of service. The age where the two lines cross indicates the youngest age of hire at which the present value of the annuity exceeds the lump sum amount for a person with 10 years of service. The simulations indicate that for a separating employee with 10 years of service, workers hired at age 35 or older (thus age 45 or older at separation) have higher PDVA while workers hired before age 35 have higher LS values.

The bottom portion of Fig. 2 illustrates a similar pattern, this time varying the years of service. For a worker hired at age 30 with a starting salary of $30,000 and wage growth of 3%, the LS value will ex- ceed the PDVA until about 15 years of service, at which point the PDVA becomes relatively more valuable. Fig. 2 illustrates how age at hire and years of service affect the relative values of the LS and the PDVA; separating employees with fewer years of service or who are younger when hired are more likely to be facing LS values that exceed the PDVA values.

To get a more complete picture of how the various combinations of age at hire and years of service yield relative values of the two dis- tribution options, we calculated the LS and PDVA amounts for a hypo- thetical worker with starting salary of $30,000 per year and annual wage growth of 3%. Table 1 reports these simulated values for workers hired at various ages as their years of service increase from 5 to 25 years. The shaded boxes indicate the distributional option with the larger value for the indicated age/service combinations. One should note the years of service that must be completed before the PDVA overtakes the LS amount and how this crossing point varies by age of hire. For workers hired at age 25, the crossover point does not occur until after 20 years of service. At the other extreme, a work- er hired at age 40 should anticipate an annuity with a present discounted value that exceeds the LS by five years of service, when vesting begins.

These simple comparisons illustrate why it would not be surprising if a substantial proportion of terminated employees select the lump sum option. The LS will be greater than the present value of the annu- ity for many departing workers. This is in stark contrast to private sec- tor defined benefit plans which are required by law to price the LS to be at least equal to the present discounted value of the annuity. These relative values suggest that one should observe a considerable proportion of vested workers requesting a LS at termination.

It is important to recognize that if one were to add the estimated present value of health insurance (calculation described in Appendix A) to the discounted value of the annuity for participants in TSERS, the value of the (cash plus health insurance) annuity is larger than the LS for vested terminated workers throughout their career. Thus,

Table 2 Percent taking a lump sum distribution (LS) by vesting status.

Vested (5+ yrs) Not vested (b5 yrs)

N % LS N % LS

Full sample 11,368 32.36 35,545 35.29

Separation year Year: 2007 5619 34.95 17,836 38.34 Year: 2008 5749 29.83 17,709 32.22

Sex Men 3948 34.75 11,822 35.31 Women 6903 26.08 21,924 30.19 Unreported gender 517 98.07 1799 97.33

Total years of service (reported) Yrs service less than 1 11,839 28.46 Yrs service 1–3 16,606 37.84 Yrs service 4 2753 41.30 Yrs service 5–19 11,039 32.60 Yrs service 20–39 280 12.14

Age at separation (calculated) Age 18–24 17 29.41 5183 29.89 Age 25–34 3256 30.62 15,297 33.95 Age 35–49 7055 33.37 13,257 38.63

Retirement system LGERS 3644 40.45 11,023 40.82 TSERS 7724 28.55 24,522 32.80

Job classification (October 2007–December 2008 only)

Education professionals (excludes higher ed) (1) 2798 19.16 8690 23.27 Skilled labor (2) 772 47.28 2982 36.82 Professional, government, admin (3) 2076 35.69 6148 39.75 University, extension, and community college (4) 184 14.67 323 35.60 Public safety (5) 1148 45.21 3486 43.49 Health and social service professionals (6) 622 27.01 2226 33.38

Regression sample (non-missing salary and gender)

Regression sample 10,818 29.30 LS > PDVA 7977 29.55 LS ≤ PDVA 2841 28.62

Notes: The sample includes all separating employees ages 18–49 who are not eligible to start an immediate retirement annuity. The sample size varies across worker and plan characteristics due to missing values for some workers. Note that for the regression sample of vested workers, 550 observations were dropped because the present discounted value of the annuity (PDVA) could not be determined from the data due to missing gender and/or salary. We report the number of workers and the percent taking a lump sum distribution (LS) versus maintaining their account.

79R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

for vested workers who are eligible for retiree health insurance when they start receiving their retirement annuity, the simple present value calculation would suggest that virtually all of these terminated em- ployees should maintain their funds with the state retirement plan. The value of retiree health is not very sensitive to age at separation due to the assumption that medical care costs are increasing at about the same rate as inflation, so the discounting effect on future health insurance is relatively small. For this calculation we assumed that workers retire at age 65 and do not claim retiree health insurance benefits until age 65, at which point they are eligible for Medicare as the primary insurer. If workers were to request an annuity and claim retiree health insurance earlier, the value of health insurance would obviously be larger, and we would predict even more individuals to choice the annuity option.

5. Separating public employees in North Carolina: 2007 and 2008

The state retirement system maintains records on current em- ployees, terminated workers, and retirees consistent with the data needed to calculate and pay retirement benefits to plan participants. The data presented here are from the State of North Carolina's retire- ment systems (TSERS and LGERS) and contain information on all workers who left public employment during 2007 and 2008. The data include the employee's date of birth, sex, salary, and the retire- ment plan. In addition, the status of the account is included indicating whether the retiree is currently receiving retirement benefits (and when the benefits were initiated), whether and when she took a lump sum distribution (LS), or whether she left the funds in the plan and the account remained active (the retirement system refers to these as “dormant” accounts).20

20 In 2007, the state adopted a new reporting system, ORBIT, that records much more detailed information about the workers' separations. Data from earlier years do not provide sufficient information to analyze economic factors that influence the distribu- tional choice.

Our sample includes workers who terminated employment in 2007 and 2008 and who did not retire and begin an annuity within one year of leaving the system. We restrict our attention to separating employees younger than 50 years old, in order to more closely ap- proximate a sample that is not eligible to immediately begin a pen- sion, even at a reduced level. Our sample includes 11,368 vested and 35,545 non-vested separating employees. Appendix B provides details on the data construction and how specific variables are de- fined, including vesting status and years of service.

To illustrate the distribution choices made by departing workers, Table 2 reports the number of vested and non-vested workers who left public employment in each year and the percent of each of the groups that accepted the LS within one year of separation. In the first column of Table 2, we see that about one-third of vested workers who left the retirement systems requested a LS within a year of termi- nation. The second column of Table 2 reports a similar breakdown for non-vested workers. These are individuals who, based on service to date, will not be eligible for a retirement annuity and will not receive any interest on funds left with the system. While one might have pre- dicted that nearly all non-vested terminated workers would select a LS, this is clearly not the case. Thus, it is important to attempt to ex- plain why individuals made a choice that, on its face, seems to be fi- nancially costly. The patterns of distributional decisions are reported separately for vested and non-vested workers for subgroups based on economic and demographic characteristics.

Several similarities are observed in the behavior of vested and non-vested workers reported in Table 2. Interestingly, roughly one- third of both groups, vested and non-vested, accepted the LS within one year of termination. Women in both groups were significantly less likely to withdraw funds. Older workers in both groups were more likely to request a LS and those with more years of service in the non-vested group were also more likely to withdraw their pen- sion funds, perhaps due to having a larger payment. All of the non-vested workers had relatively small account balances. For exam- ple, an individual that separated after three years of service whose

80 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

annual salary was $30,000 would have had approximately $5500 in his retirement account. Hence, there may be a threshold level that provides an incentive for workers to make a positive election for a LS, while those with only a few thousand dollars in their account may be less inclined to seek an immediate distribution.21 In contrast, among vested workers, those with the most years of service were the least likely to withdraw, which is not surprising given that the pen- sion system is most generous for longer tenure workers. This is also consistent with findings that private sector workers tend to be less likely to request a LS the larger the account balance. Vested and non-vested participants in the LGERS were more likely to desire a cash distribution than those in the TSERS system. As described above, workers in TSERS are covered by the state health plan (SHP) and are eligible for free retiree health insurance, but only if they are currently receiving a retirement annuity from TSERS.

There are interesting differences in the desire for a LS across em- ployment groups. The administrative records sort individuals by broad job classifications. Teachers and other educational profes- sionals were much less likely to request a LS compared to other groups, with only about 20% having cashed out their pension ac- counts. On the other hand, skilled labor and public safety officers are among the most likely to have withdrawn.

For each of the individuals in our data, the exact value of the LS that they could have received at the time of separation is reported. For those that have non-missing salary and sex information, we use the assumptions described above to calculate an estimated present discounted value of the retirement annuity (PDVA) for vested partic- ipants. The value of the LS for each worker is then compared to our estimate of the PDVA. The bottom rows of Table 2 show that, of the 10,818 vested terminated workers for whom we can calculate the PDVA, the LS exceeds the PDVA for 7977 individuals (73.7% of all vest- ed terminations).22 Surprisingly, only about one third of separating workers in both groups (those where the LS exceeds the PDVA and those where the LS is less than the PDVA) requested a LS within one year of termination. This suggests that the relative value of the two distribution options did not strongly influence the choice made by separating employees.23

6. Distributional decisions of vested workers

Next, we conduct a multivariate regression analysis of the choice to withdraw funds and accept a lump sum distribution (LS) within one year of separation for vested workers. Theoretically, this assumes that the decision to leave is made and then the individual considers what to do with his retirement accounts. Access to a LS option and the size of account balance may influence a worker's decision to leave government employment; however, we do not model this relationship.

Earlier discussion has shown that we should anticipate differences in distributional choices based on certain personal and plan charac- teristics for two reasons. First, as the simulations reported in Section 4.3 demonstrate, the relative values of the distribution op- tions are a function of a worker's age at hire, years of service, sex, and salary. Given differences in the two pension plans, one might

21 See Benartzi et al., 2011, for a discussion of threshold levels from behavioral economics. 22 Remember that this measure of PDVA includes only the value of the cash retire- ment benefit and does not include the estimated value of having access to the subsi- dized retiree health insurance. When this value is included, practically all vested workers have a higher PDVA compared to their account balance. 23 Butler and Teppa (2007) describe a traditional measure of an annuity's value as its Money's Worth Ratio (MWR), which is the ratio of the present discounted value of the annuity payments and the initial cost. The MWR is then determined to be equal to one if the annuity is well-priced, with the difference usually attributed to adverse selection and administrative costs. In our formulation, the PDVA equal to the LS would be theo- retically equivalent to a MWR of one.

also anticipate that distributional decisions will vary by participation in TSERS or LGERS. Second, previous research has shown that the choice to annuitize, holding constant the relative generosity of the distribution options, varies by sex, age, and the size of the pension account.

We seek to determine whether the choice to withdraw funds is due to underlying demographic characteristics affecting desire to annuitize, or to plan parameters affecting the relative generosity of the distribution options. We attempt to measure these two avenues by including both demographic controls (to the extent available in the data), as well as the size of the LS and our approximation of the present discounted value of the annuity (PDVA). Not only are some of the variables closely related, but the size of the LS and PDVA are both functions of salary and years of service, and age is an important determinant of the PDVA through its discounting effect (i.e., the same annual retirement benefit starting at age 65 has a higher present value for those age 45 compare to those age 35). The annual retire- ment benefit is exactly determined by annual salary and years of ser- vice, and the present value of this benefit is based on the assumed interest rate and age of the employee at termination. Thus, one should be concerned about the inclusion of age, years of service, sal- ary, the value of the LS and the PDVA in the same specification, and we should expect that the estimated coefficients will be sensitive to the inclusion of all of these variables in the same specification.

With these caveats in mind, Table 3 presents coefficients estimat- ed from a linear probability model of the decision to withdraw funds within one year of separation (request a LS) among workers who were vested in the retirement plan.24 We estimate the LS decision as a function of the value of the LS (the worker's account balance at separation), the calculated PDVA, and an indicator for whether the value of the LS exceeds the PDVA. We also include a separate inter- cept by year of termination to account for differences in economic conditions between 2007 and 2008. In columns (2) through (5), we incrementally add variables controlling for gender, state or local re- tirement plan (TSERS versus LGERS), a quadratic in final average sal- ary, a quadratic in age at separation, a quadratic in years of service to observe how these variables affect the estimated coefficients on the values of the LS and PDVA.

First, we observe that the value of the LS, which is equal to the em- ployee contributions (6% of salary) plus interest, has a quadratic rela- tionship whereby workers with larger values are less likely to request a LS until the account balances reaches approximately $73,000. Simi- larly, the estimate of the PDVA has a nonlinear effect on the distribu- tional decision with greater values increasing the likelihood of selecting a LS up to about $93,000 and larger values reducing the probability of requesting a LS. The final variable in column (1) is an indicator variable for having an account balance (value of the LS) larger than the approximated PDVA. The estimated coefficient indi- cates that having a LS that is larger than the PDVA is associated with a 5.4 percentage point higher probability of requesting a LS. This is consistent with expectations, although, as discussed further below, the effect disappears when age at separation is added in Col- umn (4). The estimates indicate that workers separating in 2008 were about 2 percentage points more likely to request a LS than those who terminated in 2007, and this difference becomes slightly larger once additional covariates are added to the model.

Columns (2) through (5) of Table 3 sequentially add covariates to the model. First, in Column (2), we see that men were 10 percentage points more likely to request the LS than were women who terminat- ed employment. This is consistent with women being more likely to accept defaults. Terminated workers in TSERS were approximately 10 percentage points less likely to request a LS compared to

24 The coefficients reported in Tables 3, 4, and 5 were estimated using linear probabil- ity models. Marginal effects from probit models are nearly identical and are available from the authors upon request.

Table 3 Withdrawal decisions of vested workers.

Mean/percent (1) (2) (3) (4) (5)

Value of the LS (10 K) $2.02 −0.188** −0.209** −0.193** −0.217** −0.236** (0.018) (0.018) (0.018) (0.019) (0.024)

Value of the LS (10 K)2 0.013** 0.014** 0.012** 0.013** 0.017** (0.002) (0.002) (0.002) (0.002) (0.003)

PDVA (10 K) $1.75 0.093** 0.100** 0.108** 0.136** 0.162** (0.013) (0.013) (0.013) (0.015) (0.016)

PDVA (10 K)2 −0.005** −0.005** −0.005** −0.006** −0.007** (0.001) (0.001) (0.001) (0.001) (0.001)

LS > PDVA 73.7% 0.054** 0.050** 0.061** −0.017 −0.009 (0.015) (0.015) (0.015) (0.018) (0.018)

Male 36.4% 0.101** 0.104** 0.115** 0.119** (0.009) (0.009) (0.009) (0.009)

TSERS 68.3% −0.104** −0.105** −0.099** −0.097** (0.009) (0.009) (0.009) (0.009)

Final average salary (10 K) $3.36 −0.027** −0.045** −0.053** (0.008) (0.009) (0.012)

Final average salary (10 K)2 0.001 0.002* 0.001* (0.001) (0.001) (0.001)

Age at separation 38.47 0.082** 0.075** (0.010) (0.010)

Age at separation2/100 −0.115** −0.107** (0.013) (0.013)

Years of service 9.01 0.024** (0.006)

Years of service2 −0.001** (0.0002)

Separated in 2008 52.9% 0.018* 0.025** 0.026** 0.027** 0.027** (0.009) (0.009) (0.009) (0.009) (0.009)

Constant 0.418** 0.477** 0.511** −0.809** −0.742** (0.017) (0.018) (0.021) (0.171) (0.171)

Notes: The sample is all workers ages 18–49 that terminated employment in 2007 and 2008 and who were vested with valid entries for the above covariates, N = 10,818. For more information on the sample for this table, see Table 2 and Appendix B. The dependent variable is the decision to withdraw the account balance and take a lump sum distribution (LS) within one year of separation versus maintaining the account (leaving it open with the possibility of electing for an annuity once eligible), approximately 29.3% of the sample requests a LS. The (LS > PDVA) is an indicator for whether the value of the LS is greater than the calculated present discounted value of the annuity (PDVA). Coefficients are esti- mated from a linear probability model with standard errors in parentheses. *Significant at 5%; ** significant at 1%.

81R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

comparable workers leaving LGERS, which is expected given the po- tential value of retiree health insurance for workers covered by TSERS. Introducing these two variables has virtually no effect on the estimated coefficients in the specification shown in Column (1).

In Column (3), we observe that the final average salary (i.e., the average salary over the past four full years of employment) has a u-shaped relationship with the probability of requesting a LS. The ef- fect is negative up until a salary of over $112,500 (depending on the specification) and then for higher salaries becomes positive. Adding the salary variable also has little effect on the other coefficients. Col- umns (4) and (5) report the estimates when age at separation and years of service are added to the model. Age has an inverted u-shaped relationship with distributional choice, where the youngest and oldest workers are least likely to choose a LS. Depending on the specification, the effect of age on the probability of receiving a LS is positive but declining up until the worker reaches approximately age 35 after which age has a negative effect on accepting a LS. One possible explanation for this pattern is that older workers are closer to being able to start a retirement annuity, so experience less discounting due to having to wait to claim the benefit. It may also re- flect a greater confidence by older workers that retirement benefits will not be reduced for them or greater salience of the need for retire- ment income among workers nearer to retirement. Note that adding a quadratic polynomial in age to the regression reduces the estimated impact of having a larger value of the LS than the PDVA from being positive and significant to being negative and insignificant.

The relative size of the LS and PDVA is highly correlated with age, as illustrated in the simulations reported in Table 1 and shown in Fig. 2. The annuity value is discounted back to the current age, so those who are closer to retirement have less discounting. Meanwhile, the LS is not a function of age, so older individuals are more likely to have higher PDVA than LS relative to younger individuals. We see this

in the data where, for example, among workers age 45 and older at separation only 11% have larger LS's than PDVA's, while almost 90% of workers less than age 45 have a higher value of their LS than PDVA. It is therefore difficult to separately identify the effects of age and the relative value of the LS and PDVA.

In the final column of Table 3, we see that adding a quadratic in years of service leaves the other estimates basically unchanged. Hold- ing all else constant, including the values of LS, PDVA, and age, workers are more likely to withdraw with more years of service up to 12 years of service, at which point longer tenures are associated with a reduction in the probability of accepting a LS. In general, the observed relationships between both demographic characteristics and work history and the probability of withdrawing funds conform to expectations. We next propose several reasons why workers might not be highly responsive to the relative size of the LS and PDVA.

7. Potential explanations for the observed distributional decisions

We postulate four main hypotheses on why the relative size of the lump sum distribution (LS) and the present discounted value of the annuity (PDVA) might not affect the distributional choice in the man- ner initially expected: (1) the potential of subsidized health insurance in retirement if one selects an annuity, (2) the likelihood of returning to work and continuing to build years of service, (3) economic condi- tions and alternative investment opportunities, and (4) inertia, inad- equate financial literacy, peer effects, and the lack of knowledge about the choices and their relative value. We now consider each of these relationships and their potential effect on the distributional choices of terminated vested workers. A discussion of how these fac- tors might influence non-vested workers follows in Section 8.

The calculation of the PDVA makes a series of assumptions includ- ing a personal discount rate and a life expectancy from an actuarial

82 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

life table. In addition, we are assuming that individuals are risk neutral. In this context, risk neutrality is a benign assumption, since technically a worker could withdraw funds and immediately roll them over into an- other tax qualified account. Because workers that request a LS have the option of rolling funds over into another tax qualified account without paying a penalty, or may withdraw funds but still save for retirement using other means, attitudes towards risk cannot explain choices in this particular context. An individual that has a higher personal discount rate or a lower life expectancy would value the annuity less than our calculation and would be more likely to take a LS. Similarly, separating workers that have a bequest motive would be more likely to withdraw funds. These and other personal characteristics have been proposed in the literature to explain the annuity puzzle that people do not purchase annuities, but do little to explain our results.25

7.1. Access to subsidized health insurance in retirement

The present value of the retirement annuity in the regression anal- ysis was based solely on the cash benefit one could expect to receive in retirement. However, as described above, vested employees in the TSERS (teachers and state government workers) system who leave their pension account open and ultimately receive a retirement ben- efit are also eligible to participate in the state health plan (SHP) once they begin receiving the pension. The state will pay 100% of the health insurance premium for these former employees.26 Persons who accepted the lump sum are not eligible for participation in the state health plan in the future.

Appendix A provides a calculation of the present discounted value of health insurance beginning when the retirement annuity starts and ending with death. Health insurance is a relatively valuable benefit, and according to our calculations described in Section 4.3, those who anticipate taking advantage of this benefit would see the PDVA plus the present value of the health insurance exceed the LS amount regard- less of years of service and/or age at hire. To give a sense of the magni- tude of the health insurance benefit relative to the annuity in the data, we estimate that the present discounted value of health insurance for our sample is between $37,225 and $47,854, depending on the individ- uals' age at separation and sex.27 We find that all but ten vested separat- ing workers have smaller LS's when compared to the combined value of the PDVA and health insurance benefit. When considering the choices of workers covered by retiree health insurance, it is much less surprising that only 29% choose to withdraw funds within one year of separation.

Of course, this assessment of the relative value of the distribution- al options assumes that the departing worker believes that the benefit will still be provided by the time she retires and that she intends to claim the health insurance benefit. The cost and liabilities associated with state retiree health plans have been critically examined since new reporting standards were required by the Governmental Accounting Standards Board (Pew Center on the States, 2011; Clark

25 Brown (2001) discusses the annuity puzzle in detail in a similar context for private sector workers. He surveys the empirical work on this issue and develops a concept of annuity equivalent wealth variable to partially explain decisions to annuitize wealth in defined contribution retirement plans. 26 In 2011, the North Carolina General Assembly passed legislation that for the first time required a premium to be paid by active and retired workers for the Standard Plan offered by the state; however, workers and retirees still had access to the Basic Plan without having to pay a premium (Clark and Morrill, 2011). 27 To the extent that some vested workers will start their retirement annuity prior to age 65, the value of retiree health insurance is underestimated, perhaps by a consider- able amount. Individuals who are receiving a retirement annuity are covered by the state health plan regardless of age or whether the person is receiving a normal retire- ment benefit or one reduced due to age and service. For example, a worker who was hired at 25 and left state employment at age 45 could select to begin a reduced retire- ment benefit at age 50, she would also be eligible to participate in the state health plan at no premium. The subsidy for the health insurance does not vary with age at which the individual retires. As noted earlier, the value of the pension benefit is actuarially re- duced to reflect earlier retirement ages; but as we see here, the present value of the health insurance would be much higher for those who accept benefits prior to age 65.

and Morrill, 2010). The unfunded liabilities have been widely reported in the popular press, and many states (including North Carolina) have been modifying the terms of these plans. Thus, it would not be surprising if public employees doubted that the current- ly promised benefits will still be provided in 10 or 20 years. Reflecting this concern, terminated workers may further discount the value of future health insurance in retirement provided by the state. In addi- tion, some workers may have access to health insurance through spousal coverage, so they might not value this benefit as much.

7.2. Potential for returning to work and the distributional choice

If terminated workers anticipate that they may return to public employment in North Carolina, they may wish to keep their accounts open. Terminated workers who return to public employment will re- tain their service based on previous public employment provided they kept their accounts open and do not take a LS. Thus, an employee who temporarily leaves her job due to medical or family reasons with the intent of returning to public employment within a few years may find it convenient and cost-effective to leave her account open. More- over workers who accept the LS can “purchase back” their prior years of service at a price specified by the plan.

To explore this question, we consider all workers who left the re- tirement systems in 2007 and 2008 and observe whether they had returned to public employment by the end of 2009. Surprisingly, a greater proportion of those who returned to public employment had accepted a LS. Among those that returned to employment, 44% of the vested and 48% of the non-vested terminators had selected a LS. In contrast, only about one-third of both vested and non-vested employees who had not returned to public employment chose a LS. It should be emphasized that our time period for return to work is only one to two years, so the impact could be substantially different if we had data over a longer time period.

7.3. Economic conditions and alternative investment options

Accepting a LS gives terminated employees access to retirement funds that can be either rolled over into another tax qualified account, invested in a non-retirement account, or spent on consumption. A ter- minated worker can request a LS at any time after leaving public em- ployment up until she actually starts a retirement annuity. For vested individuals, the account balance continues to increase each year by a plan-specified interest rate of 4%. Thus, a well-informed individual could view leaving her retirement account open as an investment op- tion. In other words, she could request a LS and invest the money her- self or leave the money with the state plans and earn a guaranteed return of 4%.28 Given the uncertainty in the financial markets during this period, a guaranteed return of 4% may have been an attractive in- vestment. Data provided by the retirement system suggests that most LS's occur relatively close to the date of termination.29 A longer time series on when terminated workers requested a LS would provide bet- ter insight into this issue.30

28 The account balance is not adjusted for inflation. 29 In 2010, 12,501 terminated workers accepted a LS. Of these 8473 individuals re- ceived the distribution within one year of termination, another 1234 had been gone for between one and two years, and another 1677 requested a LS between two and five years after leaving the retirement systems. Thus, 91% of all LS's paid in 2010 were to individuals who left public employment within the last five years. 30 An interesting thought experiment is to consider a hypothetical pension plan with all the characteristics of the North Carolina plan except that there is no annuity benefit. This simplifies the distributional choice to taking an immediate LS or leaving the account open and take the LS at some future date with the account balance increasing by 4% per year. If we could observe individuals covered by such a plan, it would be interesting to know the distribution of when workers accepted a LS. In essence, this becomes a portfolio choice with the pension serving as a guaranteed value fund with a 4% return. Depending on per- sonal characteristics and the size of the pension account, we would probably observe some variation in the time since termination that the LS is requested.

4 5

6 7

8

U n e m

p lo

ym e n t R

a te

-4 0

-2 0

0 2 0

S &

P 5

0 0 E

st im

a te

d R

e tu

rn

.2 .2

5 .3

.3 5

.4 .4

5

F ra

ct io

n W

ith d ra

w

2007m1 2007m7 2008m1 2008m7 2009m1

Year and Month of Separation

Fraction Withdrawing S&P 500

Unemployment Rate

Fig. 3. Fraction withdrawing funds, alternative investment options, and local economic conditions.

83R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

To the extent possible given the short time period over which our data span, we can consider the manner in which distributional deci- sions are influenced by outside investment opportunities. In Fig. 3, we plot the fraction of workers that withdraw funds within a year by the month of separation. We see that the probability of withdraw- ing funds drops fairly steadily over the two year period from January 2007 through December 2008. On this same graph, we plot the 12-month return in the S&P 500, which follows a similar pattern.31

However, one must recognize that returns to such investments are correlated with many other macro-level variables, such as the state unemployment rate, which may affect the need for immediate access to funds. Indeed, when adding a plot of the North Carolina state un- employment rate to Fig. 3, we see that the trends mirror those of the S&P 500.32 We expect that job market opportunities, the need for liquidity, and the opportunity costs of investments will all influ- ence an individual's choice of withdrawing funds. Note that over this same time period the 1-year Treasury bond rate reached a high of 5.06% and a low of 0.49%.33 We did not include this plot in Fig. 3 for simplicity, but the decline in the Treasury bond return rate closely tracks the other trends in the figure and is included in the regression analysis below.

In addition to this investment effect, adverse economic conditions may affect employees' confidence in the retirement system and their perception of the state's willingness and ability to honor future retire- ment payments. Over the past few years, the popular press has in- cluded many front page stories about the financial problems facing public pension plans, including the rising cost of providing these ben- efits and low funding ratios. North Carolina has one of the best funded public pension plans in the U.S. Analyzing 2009 pension data from the

31 The S&P 500 rate of return was calculated by comparing the adjusted close in the month prior to separation to that 12 months prior, so that the return for those separat- ing in December of 2007 was equal to the difference in the adjusted close in November 2007 and November 2006, divided by the adjusted close in November 2006. These data were downloaded from the http://finance.yahoo.com/q/hp?s=%5EGSPC+ Historical+Prices, [accessed February 8, 2013]. 32 The North Carolina monthly unemployment rate is from the Bureau of Labor Statis- tics, Local Unemployment Rate Series, LAUST37000003, available at http://data.bls. gov/timeseries/LAUST37000003, [accessed February 6, 2013]. 33 The 1-Year Treasury Constant Maturity Rate (GS1) data were obtained from http:// research.stlouisfed.org/fred2, [accessed February 4, 2013].

Comprehensive Annual Financial Reports of the states, the Pew Center on the States (2011) ranks North Carolina as the fourth best funded state retirement plan, with a funding ratio of 97%.34 Still, many states have been making major changes in their retirement plans. In most states, changes will apply only to new employees. However, some states whose plans are facing substantial funding problems have been altering the benefits for current workers and re- ducing COLAs for current retirees. In this environment, terminated workers in North Carolina might have questioned whether their pen- sion plans would remain unchanged for 10, 20 or 30 years. Doubts about the stability of the plans may have led some separated workers to be more likely to request a LS.

Because these factors tend to move together, it is difficult to disen- tangle the independent effects of each. Still, we explore these rela- tionships further in a regression context in Table 4, Columns 1–5. Using the specification in Column 5 of Table 3, we include the 1-year US Treasury bond rate, the 12-month prior return to the S&P 500, and the monthly unemployment rate.35 We also present a spec- ification where we drop the indicator variable for the year 2008 to allow for the full variation in these trends to be utilized.36 In Table 4, Column 1, we see that when entered individually into the specification, the estimated coefficient for the 1-year Treasury bond rate is negative and insignificant. In contrast, in Column 2, the esti- mated coefficient on the gain in equity values is positive and signifi- cant, indicating that a higher growth in equity values increases the likelihood of accepting a LS. Interestingly, as seen in Column 3, a 1% higher state unemployment rate in the month of termination is asso- ciated with a 3 percentage point decline in the probability of with- drawing funds. This could reflect a difference in who is terminating employment in a given month, or it could indicate that when job

34 While most economists believe that the assumed rates of return in these reports are too high, the relative ranking of the North Carolina retirement system as one of the best funded retirement plans does not change when lower rates of return are employed (Novy-Marx and Rauh, 2011). 35 These parameters are defined in footnotes 31–33. 36 Each of the investment opportunity/macroeconomic indicator variables are mea- sured at the month of termination, although the dependent variable is whether funds were withdrawn within a year of separation. While one could conceive of a model where the cumulative effects of these factors are accounted for, with only two years of data estimating such a model is not practical.

Table 4 Economic conditions, peers, and the withdrawal decisions of vested workers.

Mean (1) (2) (3) (4) (5) (6)

Investment opportunities 1-year treasury bond constant maturity rate 3.1 −0.003 (0.007) −0.017 (0.011) −0.031** (0.001) S&P 500 12 month return −1.3 0.001* (0.0005) −0.0001 (0.001) 0.0004 (0.001)

NC economic conditions State unemployment rate 5.63 −0.030** (0.007) −0.038** (0.014) −0.029* (0.013)

Actions of peers Fraction in 3-digit zip 0.302 0.288** (0.079) Fraction in agency 0.301 0.458** (0.033)

Selected covariates LS > PDVA 0.737 −0.009 (0.018) −0.009 (0.018) −0.010 (0.018) −0.011 (0.018) −0.011 (0.018) −0.010 (0.018) Separated in 2008 0.529 0.019 (0.021) 0.061** (0.016) 0.075** (0.015) 0.040 (0.023) 0.029** (0.008)

Notes: The sample and specification are identical to Table 3, Column (5) with the inclusion of investment opportunities, local economic conditions, and peer behavior, as indicated. Column (5) includes dummy variables for living in a zip code or separating from an agency with fewer than 5 peers in the sample. Coefficients are estimated from a linear probability model with standard errors in parentheses. *Significant at 5%; ** significant at 1%.

84 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

opportunities are fewer then demand for LS fall. This finding does suggest that liquidity constraints are not that important to this deci- sion, since we would expect cash needs to rise when the unemploy- ment rate is higher. In Column 4 of Table 4, we include all three variables in the same specification and find that only the state unem- ployment rate has a significant effect. However, when we do not in- clude the year 2008 dummy variable in Column 5 we see that the estimated coefficient on the 1-year Treasury bond rate becomes larg- er and is statistically significantly negative.37

Although our data only span two years, there is quite a bit of monthly variation in these trends. However, to truly understand the effects of investment options and economic conditions would require a much longer time series and a richer variety of economic and investing environments. Still, the results presented in Column 5 are not consistent with an investment opportunity cost story, where indi- viduals are more likely to take money out when investment opportu- nities are good. Nor are the results consistent with a story about liquidity or the need for cash during an economic downturn. Similar- ly, if confidence in the system is correlated with economic conditions, we again do not see evidence that as “confidence” falls, withdrawals rise. One alternative explanation is that when economic conditions are uncertain, then individuals do not make financial decisions and are more reliant on defaults.

7.4. Financial literacy, plan knowledge, peer effects, and inertia

Considerable survey evidence indicates that workers have a low level of financial literacy, inadequate knowledge about their pension plans, and poor understanding of how choices can affect the value of retirement accounts (Lusardi and Mitchell, 2007; Clark et al., 2012). If workers who leave the North Carolina retirement system possess the same low levels of knowledge, it should not be surprising that they do not always select the option that has the greatest present value. While information about the size of the LS is easily available, separating workers may not be able to determine the value of the re- tirement annuity. Calculating the present value of an annuity 20 or even 30 years in the future is not a simple task, even for individuals who have high levels of financial literacy.

The default option is to maintain the retirement account until the terminated worker either requests a LS or, once eligible, a retirement benefit. There were no changes in the default during our sample peri- od and thus, we are not able to investigate how changes in the default might have altered the distributional choices of departing workers.

37 In a related context, Goda et al. (2012) consider how retirement intentions are influenced by investment returns in the year prior to survey for respondents in the Health and Retirement Study (HRS). They also document that the collinearity of stock returns and labor (and housing) market conditions makes separately identifying the effects difficult.

However, recent research has shown that defaults matter in the selection of various pension options and inertia associated with the acceptance of defaults could also be affecting the distributional choices (e.g., Madrian and Shea, 2001). Brown et al. (2008b) described an important role for framing. If workers accept defaults due to inertia or lack of understanding, or are not appropriately valuing benefits, then public employers could improve separating workers' welfare by providing timely and accessible financial education and information.

The retirement system sends all separated employees who request a LS a personalized form letter (shown in Appendix C), regardless of their vesting status or the relative size of their benefits. The letter contains a warning that if the refund is processed the individual will forfeit a future annuity benefit. A separating worker who is not confi- dent in his/her choice may be swayed by the strong wording in this letter. We do not know how many individuals first applied but then ultimately chose not to withdraw funds after having received this letter.

Workers who are unsure about their distributional decisions may rely on their colleagues or neighbors for advice (Brown et al., 2008a). We investigate the importance of peer effects by adding two variables to the specification in Column 6, Table 3: the fraction of terminating participants who live in the same 3-digit zip code who selected a LS and the fraction of terminating participants who worked in the same agency who chose a LS.38 Both of these variables are positively and significantly related to the probability of selecting a LS. This is consistent with peer effects influence distributional decisions. However, as described in Manski (1993), the estimated coefficients may instead reflect the fact that peers share similar (unobservable) characteristics and so their choices may be correlated. Although we find that individuals behave similarly to their peers, this is not neces- sarily a causal effect.

8. Distributional choices of non-vested workers

Workers leaving public employment with fewer than five years of service are not vested according to the rules of both retirement plans (TSERS and LGERS), so they are not eligible for a retirement annuity based on service at the time of termination. Accordingly, the lump sum distribution (LS) available to them is the sum of their own pension contributions during their employment without any interest. Nevertheless, they are not required to immediately request a LS and the default is to maintain the account with the system. Given the choice of money now or the same amount of money at some future date, we would expect that all non-vested terminated workers would request immediate LS's. Yet, Table 2 shows that only about one-third of these

38 These variables are defined by taking the average across all workers within the cell, excluding the individual.

Table 5 Withdrawal decisions of non-vested workers.

Mean/percent (1) (2) (3) (4)

Value of the LS (1 K) $3.15(K) 0.030** 0.029** 0.028** 0.022** (0.001) (0.001) (0.001) (0.002)

Value of the LS (1 K)2 −0.001** −0.001** −0.001** −0.001** (0.000) (0.000) (0.000) (0.000)

Male 33.3% 0.043** 0.045** 0.047** (0.005) (0.005) (0.005)

TSERS 69.0% −0.062** −0.061** −0.064** (0.005) (0.005) (0.005)

Age at separation 33.3 0.016** 0.017** (0.003) (0.003)

Age at separation2 −0.019** −0.019** (0.004) (0.004)

0–0.99 Years of service 33.3% −0.053** (0.010)

1–1.99 years of service 28.2% 0.009 (0.008)

3–3.99 years of service 12.2% −0.019* (0.009)

4–4.99 years of service 7.8% −0.025* (0.012)

Separated in 2008 49.8% −0.067** −0.003 −0.003 −0.002 (0.005) (0.005) (0.005) (0.005)

Constant 0.311** 0.277** −0.049 −0.023 (0.005) (0.006) (0.050) (0.051)

Notes: The sample is all workers ages 18–49 that terminated employment in 2007 and 2008 and who were not vested (less than five years of membership plus non-contributory service) with valid entries for the above covariates, N = 35,545. For more information on the sample for this table, see Table 2 and Appendix B. The dependent variable is the de- cision to withdraw the account balance and take a lump sum distribution (LS) within one year of separation; approximately 35.29% chose this option. The omitted category is 2– 2.99 years of service. Also included in the specifications in Columns (2)–(4), but not reported, is a dummy variable for gender unknown (5.1% of the sample). Coefficients are es- timated from a linear probability model with standard errors in parentheses. *Significant at 5%; ** significant at 1%.

85R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

former public employees requested a LS in the first year following termination.39

At first glance, this is quite puzzling. Unlike the vested workers, non-vested terminated employees cannot look forward to a retire- ment annuity, will not be eligible for state-provided retiree health insurance, and do not earn interest on the money left with the retire- ment system. But workers with short service may have paid very little attention to their pension accounts, and upon termination, may be un- aware of the value of their accounts and the distributional options available to them. Inertia is often given as an explanation of certain types of behavior and could also be a factor in lack of immediate LS's.40

Still, there may be a fully rational reason for a separating employee to leave the monies in the system, even if not vested. Employees who leave public employment but who anticipate returning to a govern- ment job may have an economic incentive to leave their pension ac- counts open, as returning civil servants who have not closed their pension accounts can count prior years of service in the determina- tion of future retirement benefits. If this motive explains the distribu- tional choices of non-vested employees, it would indicate rational decision making based on considerable knowledge of the pension plan. Family circumstances, such as the birth of a child or relocation, can also result in individuals quitting current employment and having temporary periods outside the labor force. The public sector is rela- tively large and all public employees are covered by the same system, offering considerable opportunities for future employment in either TSERS or LGERS. Nevertheless, the analysis of 2007–2008 termina- tions described in Section 7.2 does not support the hypothesis that

39 Using a longer time series, it would be interesting to discover whether these indi- viduals ever return to public employment in North Carolina and ultimately become vested in the retirement plan, whether they ever request a LS, or whether these funds are permanently lost to the individual. 40 Interestingly, inertia may be overcome by certain events. The state retirement sys- tem reports that requests for lump sum distributions from previously terminated workers often spike just before Christmas.

expectations of returning to work is a cause of keeping one's account open since a larger proportion of those that returned to work had ac- cepted a LS than those that had not yet returned to work.

To further examine the distributional choices of non-vested em- ployees, we estimated a linear probability model with the dependent variable being an indicator for whether separating workers chose to withdraw their accounts and accept the LS's from the retirement plan within one year of terminating employment. The results of this regression are presented in Table 5. We first consider how the value of the possible LS (the account balance) influences the choice of with- drawing funds. This relationship is shown to be quadratic, whereby the probability of withdrawing is increasing until the value of the ac- count reaches approximately $15,000, at which point the relationship becomes negative. Since none of these workers is eligible for an annu- ity, it is somewhat surprising that we observe this pattern, although it may be due to the possibility of returning to work in the future.

In Table 5, Column 2, we observe that men were more likely to withdraw money. This could be due to less inertia or more knowledge among men. As was the case for vested employees, members of TSERS were significantly less likely to accept a LS than those in LGERS. Again, consistent with the patterns for vested workers, the youngest and oldest non-vested separating workers were least likely to withdraw funds, with the highest probability of withdrawing around age 42. As discussed earlier, this might be due to lack of knowledge or be- cause of a higher likelihood to return to work among the youngest workers. In the final column of Table 5, we observe that controlling for years of service diminishes the coefficient on the account balance at separation somewhat. Here the omitted category is 2–2.99 years of service and we observe that among non-vested workers those with approximately 1–2.99 years of service are the most likely to with- draw. We observe across the columns of Table 4 that the importance of the size of the LS is only partially explained by observed work his- tory and demographics, and the basic pattern suggests that those with the smallest and largest potential LS values are the least likely to withdraw within one year of separation.

41 In addition, the retirement annuity from the state plan is approximately a real an- nuity with benefits typically being increased each year by the legislature at a rate near the level of inflation. In contrast, the annuity that one could purchase in the market un- der these assumptions would be a nominal annuity.

Table 6 The decision to cash out versus rollover benefits among vested and non-vested workers accepting a lump sum distribution (LS).

Full sample (88.7% cash out) Vested (87.3% cash out) Non-vested (89.1% cash out)

(1) (2) (3)

Mean Coeff. Mean Coeff. Mean Coeff.

Value of the LS (10 K) 0.66 −0.238** 1.75 −0.141** 0.35 −0.487** (0.013) (0.036) (0.029)

Value of the LS (10 K)2 0.020** 0.007* 0.075** (0.002) (0.003) (0.012)

Final average salary (10 K) 3.12 −0.008 (0.018)

Final average salary (10 K)2 0.002 (0.001)

Vested 22.7% 0.074** (0.011)

Male 34.2% 0.031** 43.3% 0.027* 33.3% 0.033** (0.005) (0.012) (0.006)

Unreported gender 13.9% 0.030** 14.0% 0.033** (0.008) (0.009)

TSERS 63.2% −0.017** 60.0% −0.007 64.1% −0.020** (0.005) (0.012) (0.006)

Age at separation 34.9 0.009** 38.4 0.008 34.0 0.010** (0.003) (0.012) (0.003)

Age at separation2/100 −0.010* −0.009 −0.012* (0.004) (0.015) (0.005)

Years of service 3.42 0.024** 8.48 0.014 1.95 0.047** (0.004) (0.011) (0.010)

Years of Service2 −0.0004 0.0001 0.002 (0.0002) (0.0004) (0.002)

Separated in 2008 45.7% 0.024** 54.0% 0.029* 45.5% 0.026** (0.005) (0.012) (0.006)

Constant 0.744** 0.783** 0.734** (0.055) (0.221) (0.060)

Observations 16,213 3,167 12,7537

Notes: The samples include only those separating workers who take a lump sum distribution (LS). For more information on the sample for this table, see Table 2 and Appendix B. Note that in Column (2) only workers whose final average salary could be calculated are included in the sample, as in Table 3. The dependent variable is the decision to cash out the account balance (active termination refund or active termination refund federal tax) instead of directly rolling over into an IRA within one year of separation. Coefficients are estimated from a linear probability model with standard errors in parentheses. *Significant at 5%; ** significant at 1%.

86 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

9. Cash distributions or rollovers: leakages from retirement wealth

The analysis thus far indicates that only about one third of all ter- minating public employees in North Carolina request a LS distribution while two thirds leave their retirement funds in the system. Thus, the extent of retirement saving leakage among these workers is limited. However, among those requesting a LS distribution the risk of leakage is high. Terminated employees who have requested a lump sum dis- tribution (LS) have the option of directly receiving a check or having the funds sent from the retirement system directly to another tax qualified account, often an IRA. This form of distribution from a de- fined benefit plan is typically called a “rollover” of pension funds (see Fig. 1). If the funds are rolled over, the monies are not counted as current income to the taxpayer, nor does the individual incur any tax penalty associated with the early withdrawal of funds from a retirement account. Thus, assets rolled over remain part of the individual's retirement wealth and can be invested through the new tax qualified plan.

The impact of the rollover amount on retirement income will de- pend on the investment choices made and subsequent returns earned compared to the ultimate annuity that a vested worker could receive in the future. In our calculations for vested workers, the future annu- ity is discounted at a nominal rate of 5.8%. Thus, if the vested worker earned returns in excess of 5.8%, she might be able to purchase an an- nuity at retirement with a benefit that exceeded the benefit paid by the pension system. But the retail annuity market for an individual is relatively thin and includes fees, so the worker would likely have

to earn a return greater than 5.8% in order to be able to purchase a comparable benefit in retirement.41

When a terminated worker requests that a check be sent directly to her, the distribution is subject to personal income tax in the year it is received. In addition, because all workers in our sample are under age 50, they will have to pay a 10% tax penalty for an early dis- tribution. Once received, the payment can be spent on current con- sumption, used to pay off outstanding debts, or saved. Of course, if the monies are spent, potential income in retirement is reduced. Our data identify whether the LS was rolled over or paid as a cash dis- tribution; however, we have no information on how the funds were subsequently used by terminated workers.

The possibility that the funds are spent, creating “leakage” from retirement saving, is a concern to many policy analysts. To investigate this pattern, we estimate a regression model exploring the determi- nants of how LS's are sent to terminated employees. The sample in- cludes all individuals who left either of the two North Carolina retirement systems in 2007 and 2008, and who requested the LS. Overwhelmingly, workers who requested the LS opted for a cash dis- tribution; almost 90% of these individuals were sent a check directly

87R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

from the retirement system.42 We estimate the decision separately for vested and non-vested workers. Non-vested workers seeking an annuity do not have the option of maintaining their account and eventually receiving a retirement annuity from the pension plans. We might therefore expect that among workers choosing to with- draw funds, the non-vested group would be more likely to rollover than the vested group, holding all else equal.

Table 6 reports the estimated coefficients from a linear probability model of the decision to cash out (versus rollover) benefits among those electing a LS. The estimates cover three alternative samples: all terminated employees who requested the LS, all vested workers receiving the LS, and all non-vested individuals that received the LS. Estimated coefficients are similar across the three data sets used in the analysis. For all groups, larger account balances significantly in- creased the probability of rolling the funds over into a tax qualified account. This may reflect the desire to avoid the higher tax and penalties associated with the cash distribution. Consistent with ex- pectations given the alternative options for annuitizing, in Column (1) we observe that, holding all else constant, including the value of the LS, vested employees were 7.4 percentage points more likely to request a cash payment.

Across all three samples we see that among those requesting a LS, men are significantly more likely to cash out rather than directly roll- over benefits. In Column (3) of Table 6 we see that non-vested partic- ipants in TSERS were 2.0 percentage points less likely to receive a cash distribution; however, there were no differences between vested members of TSERS and LGERS found in Column (2). The pattern of cashing out has an inverted u-shape by age for non-vested em- ployees, while we do not find an effect for either age or years of ser- vice among vested employees, holding all else equal. For non-vested workers, an additional year of service increased the probability of accepting a check by 4.7 percentage points. Among separating workers requesting a LS, those terminating employment in 2008 were over two percentage points more likely to request cash relative to those separating in 2007.

The most important finding from this analysis is that almost 90% of terminated workers that requested a LS opted for a cash distribution. Given the economic conditions that prevailed during this period, one should not necessary conclude that this magnitude of leakage from retirement saving would prevails under normal situations. Combining these results with those presented earlier, that roughly one-third of all terminated workers request a LS, we estimate that about 30% of all terminated workers close their retirement accounts and receive a cash distribution within one year of separation.

10. Key findings and conclusions

Our analysis shows that public employees face very different choices than do private sectors employees. Whereas private sector plans are constrained by law to offer lump sum distributions (LS's) that are greater than or equal to the present value of the future annu- ity, public sector pension plans base the LS on employee contributions and credited interest. Hence, the LS value is not directly linked to the value of the life annuity.

Leakages of retirement savings when participants in defined ben- efit plans change jobs is a two step decision process. First, if the indi- vidual decides to leave his/her funds in the pension plan, no leakage

42 From October 2007 on, separating workers had the option of taking a partial roll- over and requesting the balance in cash. These are recorded as rollovers in the data, and we cannot distinguish between full or partial rollovers. Therefore, our estimate of “leakage” may be understated, since some of those rolling over funds could also be taking out a portion in cash. We also expect that some separating employees do take cash and deposit it into a tax-qualified account themselves within the allowed 60 day period. We have no way of knowing how many workers exercise this option, either. Thus, the estimate of 90% leakage is slightly low because of partial rollovers and slightly high because of indirect rollovers not being observed.

occurs at this time; however, individuals can request a LS in the fu- ture. Second, having requested a LS, the individual has the option to take the distribution as cash or to directly rollover funds into an IRA. Using administrative records from North Carolina retirement plans, we provide a detailed picture of the distribution decisions of workers ages 18 to 49 that separated prior to retirement from public pension plans in North Carolina between 2007 and 2008. We show that, for younger workers and those with fewer years of service, the LS typically is larger in value than our approximation of the present value of the annuity (PDVA), yet only one-third elected the LS. This fraction does not differ considerably by vesting status or by eligibility for retiree health insurance.

Among participants in the North Carolina retirement system, about one third of terminating workers choose to withdraw funds. However, this is an overestimate of potential leakages of retirement assets. Of those who request a LS, approximately 90% requested cash rather than rolling over directly to another tax qualified retire- ment account. This could reflect a perceived need for cash for current consumption or paying off debts. It might also be the result of poor understanding of the tax consequences of this choice and the need to save these funds for retirement. Of course, workers could still de- posit the cashed-out benefits into an IRA themselves. The high rate of cash-outs suggests a sizeable reduction in retirement wealth accu- mulation and suggests that there is a considerable amount of “leak- age” from the retirement savings of public sector workers in North Carolina.

In a related paper, Clark and Morrill (2012) find that among workers who separated from public employment in 2007–2008 and were eligible for an immediate reduced annuity (age 50 with at least 20 years of service), 90% opted to receive an immediate annuity. Of those who were eligible for an immediate unreduced annuity (30 years of service, age 62 with 25 years of service, or age 65 with 5 years of service), 99% selected some type of immediate annu- ity offered by the plan. Thus, for retiring public employees in North Carolina, almost all select an annuity option and reject the offer of a LS. On the basis of these findings and those in the current paper, it appears that public sector retirees in North Carolina are not part of annuity puzzle and instead overwhelming select annuities when they are offered.

One might expect that all non-vested terminations request a lump sum distribution since they are not entitled to a future retirement benefit and earn no interest on funds left with the system. Yet we ob- serve that two thirds of those leaving the systems in 2007 and 2008 left their accounts open. Given the evidence on financial literacy and inertia in the general population, one might speculate that non-vested terminated workers are unaware of their ability to access these funds, do not understand that they have no claim on a future benefit, do not understand that the funds will not earn any interest, or simply do not take the time to request a LS. Economic and psycho- logical studies indicate that workers often merely accept default op- tions. In this case, the default is to leave the account open, thus we may think that workers leaving the system do not take the time to re- quest a distribution of their pension account. To address this concern, the plan could change the default to be a LS for non-vested workers. Clark and Morrill (2012) find a similar result for non-vested workers over the age of 50 with only 36% of terminated workers aged 50 to 59 selecting a LS. While the proportion of those choosing a LS rises some- what with advancing age (50% of those over age 65 chose the LS), it is hard to explain why these older non-vested terminating workers would leave their funds with the system.

We postulate several factors that might explain why terminating workers for the most part do not appear to respond to the relative size of the PDVA and LS. Given that vested employees receive a 4% an- nual return when funds are left in the system, individuals can consid- er this as an investment option. However, regression analysis does not support the hypothesis that greater returns on investments

88 R.L. Clark et al. / Journal of Public Economics 116 (2014) 73–88

outside the retirement system increase the likelihood that workers will request a LS There is some evidence that as the economy wors- ened, as proxied by state-level unemployment rates, individuals were less likely to withdraw funds. However, we have a limited time series available, so are not able to disentangle the effects eco- nomic conditions, labor market conditions, outside investment op- tions, and confidence in the retirement system. We find that there may be an important role for defaults and workers may not be well informed about the value of their benefits.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.jpubeco.2013.05.005.

References

AonHewitt, 2011. Leakage of Participants' DC Assets: How Loans, Withdrawals, Cashouts are Eroding Retirement Income. http://www.aon.com/attachments/thought-leadership/ survey_asset_leakage.pdf (last accessed February 14, 2013).

Bassett, William, Fleming, Michael, Rodriguez, Anthony, 1998. How workers use 401(k) plans: the participation, contribution, and withdrawal decisions. National Tax Journal 51 (2), 263–289.

Bellante, Don, Link, Albert, 1981. Are public sector workers more risk averse than private sector workers? Industrial & Labor Relations Review 34, 408–412.

Benartzi, Shlomo, Previtero, Alessandro, Thaler, Richard, 2011. Annuitization puzzles. Journal of Economic Perspectives 25 (4), 143–164.

Bonin, Holger, Dohmen, Thomas, Falk, Armin, Huffman, David, Sunde, Uwe, 2007. Cross-sectional earnings risk and occupational sorting: the role of risk attitudes. Labour Economics 14 (6), 926–937.

Brown, Jeffrey R., 2001. Private pensions, mortality risk, and the decision to annuitize. Journal of Public Economics 82, 29–62.

Brown, Jeffrey R., Weisbenner, Scott, 2012. Why do Individuals Choose Defined Contri- bution Plans? Evidence from participants in a large public plan, paper presented at Retirement Benefits for State and Local Employees: Designing Pension Plans for the Twenty-First Century, NBER Conference, Jackson Wyoming.

Brown, Jeffrey R., Ivkovic, Z., Smith, P.A., Weisbenner, Scott, 2008a. Neighbors matter: causal community effects and stock market participation. Journal of Finance 63 (3), 1509–1531.

Brown, Jeffrey R., Kling, Jeffrey R., Mullainathan, Sendhil, Wrobel, Marian V., 2008b. Why don't people insure late-life consumption? A framing explanation of the under-annuitization puzzle. American Economic Review: Papers & Proceedings 98 (2), 304–309.

Bryant, Victoria, Holden, Sarah, Sabelhaus, John, 2011. Qualified Retirement Plans: Analysis of Distribution and Rollover Activity. Pension Research Council Working Paper WP2011-01.

Bureau of Labor Statistics, 1990. Employee Benefits in Medium and Large Firms, 1989. Bulletin 2363.

Bureau of Labor Statistics, 1999. Employee benefits in medium and large establish- ments, 1997. Bulletin 2517.

Bureau of Labor Statistics, 2007. National Compensation Survey: Employee Benefits in Private Industry in the United States 2005. Bulletin 2589.

Bureau of Labor Statistics, 2011a. Employee Benefits Survey. http://bls.gov/ncs/ebs/ benefits/2011/benefits_retirement.htm.

Bureau of Labor Statistics, 2011b. The Employment Situation: November 2011 (Decem- ber 2, 2011).

Burman, Leonard, Coe, Norma, Gale, William, 1999. Lump sum distributions from pen- sion plans: recent evidence issues for policy and research. National Tax Journal 52 (3), 553–562.

Butler, Monika, Teppa, Federica, 2007. The choice between an annuity and a lump sum: results from Swiss pension funds. Journal of Public Economics 91 (10), 1944–1966.

Buurman, Margaretha, Dur, Robert, Van den Bossche, Seth, 2009. Public sector em- ployees: risk averse and altruistic? Journal of Economic Behavior & Organization 83 (3), 279–291.

Chalmers, John, Reuter, Jonathan, 2012. How do retirees value life annuities? Evidence from public employees. Review of Financial Studies 25 (8), 2601–2634.

Chang, Angela, 1996. Tax policy, lump-sum pension distributions, and household saving. National Tax Journal 49 (2), 235–252.

Clark, Robert, Hanson, Emma, 2011. Distribution Options in State Pension Plans. NCSU Working Paper.

Clark, Robert, Morrill, Melinda S., 2010. Retiree Health Plans in the Public Sector: Is There a Funding Crisis? Edward Elgar Publishing, Northhampton, MA.

Clark, Robert, Morrill, Melinda S., 2011. Containing Health Insurance Costs for Active and Retired Public Sector Employees: Lessons from the State of North Carolina. NCSU Working Paper.

Clark, Robert, Morrill, Melinda S., 2012. Retiring Public Sector Employees Prefer Annu- ities to Lump Sum Distribution. North Carolina State University Working Paper.

Clark, Robert, Craig, Lee, Sabelhaus, John, 2011. State and Local Retirement Plans in the United States. Edward Elgar Publishing, Northhampton, MA.

Clark, Robert, Morrill, Melinda S., Allen, Steven, 2012. The role of financial literacy in determining retirement plans. Economic Inquiry 50 (4), 851–866.

Copeland, Craig, 2009. Lump Sum Distributions at Job Change. EBRI Notes. (January, 2-10).

Engelhardt, Gary, 2002. Pre-retirement lump-sum pension distributions and retire- ment income security: evidence from the Health and Retirement Study. National Tax Journal 55 (4), 553–562.

Goda, Gopi Shah, Shoven, John B., Slavov, Sita Nataraj, 2012. Does stock market per- formance influence retirement intentions? Journal of Human Resources 47 (4), 1055–1081.

Government Accountability Office (GAO), 2009. 401(k) plans: policy changes could re- duce the long-term effects of leakage on workers' retirement savings, GAO-09-715 (August).

Hartog, Joop, Ferrer-i-Carbonell, Ada, Jonker, Nicole, 2002. Linking measured risk aversion to individual characteristics. Kyklos 55 (1), 3–26.

Hurd, Michael, Panis, Constantijn, 2006. The choice to cash out pension rights at job change or retirement. Journal of Public Economics 90 (12), 2213–2227.

Lusardi, Annamaria, Mitchell, Olivia S., 2007. Baby Boomer retirement security: the roles of planning, financial literacy, and housing wealth. Journal of Monetary Economics 54 (1), 205–224.

Madrian, Brigitte, Shea, Dennis, 2001. The power of suggestion: inertia in 401(k) participation and savings behavior. Quarterly Journal of Economics 116 (4), 1149–1187.

Manski, C.F., 1993. Identification of endogenous social effects: the reflection problem. Review of Economic Studies 60 (3), 531–542.

Mitchell, Olivia S., Poterba, James M., Warshawsky, Mark J., Brown, Jeffrey R., 1999. New evidence on the money's worth of individual annuities. The American Economic Review 89 (5), 1299–1318.

Moore, James, Muller, Leslie, 2002. An analysis of lump-sum pension distribution recip- ients. Monthly Labor Review 125, 29–46.

Novy-Marx, Robert, Rauh, Joshua, 2011. Public pension liabilities: how big are they and what are they worth? Journal of Finance 66 (4), 1207–1245.

Papke, Leslie, 2004. Pension plan choice in the public sector: the case of Michigan State employees. National Tax Journal 57 (1), 329–340.

Pew Center on the States, 2011. The Widening Gap: The Great Recession's Impact on State Pension and Retiree Health Cost. http://www.pewcenteronthestates.org/ uploadedFiles/Pew_pensions_retiree_benefits.pdf.

Pfeifer, Christian, 2010. Risk aversion and sorting into public sector employment. German Economic Review 12 (1), 85–99.

Poterba, James, Venti, Steven, Wise, David, 1998. Lump-sum distributions from retire- ment saving plans: receipt and utilization. In: Wise, David (Ed.), Inquiries into the Economics of Aging. University of Chicago Press, Chicago, pp. 85–105.

Poterba, James, Venti, Steven, Wise, David, 2001. Preretirement cashouts and foregone retirement saving: implications for 401(k) asset accumulation. In: Wise, David (Ed.), Themes in the Economics of Aging. University of Chicago Press, Chicago, pp. 23–58.

Purcell, Patrick, 2007. Lump-sum distributions under the Pension Protection Act, CRS Report For Congress, Order Code RS22765 (December 3).

Purcell, Patrick, 2009. Pension issues: lump-sum distributions and retirement income security, CRS Report for Congress (January 7).

Roszkowski, Michael, Grable, John, 2009. Evidence of lower risk tolerance among pub- lic sector employees in their personal financial matters. Journal of Occupational and Organizational Psychology 82 (2), 453–463.

Sabelhaus, John, Weiner, David, 1999. Disposition of lump-sum pension distributions: evidence from tax returns. National Tax Journal 52, 593–614.

Warner, John, Pleeter, Saul, 2001. The personal discount rate: evidence from military downsizing programs. American Economic Review 91 (1), 33–53.

Yaari, Menahem E., 1965. Uncertain lifetime, life insurance, and the theory of the consumer. Review of Economic Studies 32 (2), 137–150.

Yakoboski, Paul, 1997. Large plan lump sums: rollovers and cashouts. EBRI Issue Brief, vol. 188. EBRI, Washington, DC.

Yang, Tongxuan, 2005. Understanding the defined benefit versus defined contribution choice, Pension Research Council WP 2005-4.

#143.pdf

• Corporate social responsibility and Eurozone corporate bonds: The moderating role of country sustainability
• 1 Introduction
• 2 Related literature
• 2.1 CSR and credit risk
• 2.2 Environmental performance and credit risk
• 2.3 Employees and credit risk
• 2.4 Corporate governance and credit risk
• 2.5 What moderates the relationship between CSR and CFP?
• 2.6 The impact of geographic location on stakeholders preferences
• 3 Main empirical tests
• 3.1 Methodology
• 3.2 Sample overview and data description
• 3.3 CSP and credit risk: Ratings
• 3.4 CSP and credit risk: z-spreads
• 3.5 Further analysis and robustness checks
• 4 Summary and conclusion
• References