SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

This article assesses the cost and risk faced by public sector, defined benefit plan providers arising from uncertain mortality, including longevity selection, mortality improvements, and unexpected systematic shocks. Using longitudinal microdata on Australian pensioners, we quantify the extent of longevity selection at both aggregate and scheme level. We also show that as the age-membership structure in a pension scheme matures, scheme-specific longevity selection risk and systematic shocks become quantitatively more important and have larger consequences for plan liabilities than aggregate selection risk or the impact of mortality improvements.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

Public sector employees are traditionally covered by government-sponsored pension schemes exclusive to civil servants. In recent times, the funding of public sector pensions has attracted much policy attention around the world, because the unfunded liabilities associated with civil service defined benefit (DB) schemes are both substantial and uncertain.1 A major source of uncertainty is the future life expectancy and mortality outcomes of civil servants, which may be distinguished as a group from the general population. This article explores the exposure of unfunded liabilities to longevity risk in public sector plans, using a unique data set covering the entire population of existing civil service DB pensioners in Australia. In particular, we assess and quantify the impact of increasing longevity and differential pensioner longevity on the legacy costs of the public pensions system.

The primary contribution of this article is to use microdata to directly examine longevity selection risk in public sector pension schemes and evaluate its impact on pension liabilities vis-à-vis other aspects of longevity risk. Specifically, we construct mortality rates based on actual plan experience and employ these in liabilities valuation. This departs from previous studies that commonly rely on a standard pensioner period life table for the empirical analysis of pension liabilities (e.g., Dushi, 2010; Novy-Marx and Rauh, 2011). Standard pensioner life tables such as the U.S. RP-2000 table, while convenient, are not tailored to the particular set of lives being considered. For instance, the RP-2000 combined mortality table employed by many U.S. states for the valuation of public pension liabilities is constructed mainly based on the mortality experience of private sector uninsured pensioners.2

There are a number of reasons why the exposure of pension liabilities to longevity risk comes into sharper focus for the public sector plans than corporate plans. First, substantially larger proportions of full-time civil servants are covered by DB plans as compared to private sector workers. For example, in the United States, DB plan coverage still remains very much the norm in the public sector despite its dramatic decline in the private sector over the past three decades (Brown, 2011). Likewise, it was common practice for Australian state governments to require their employees to join a DB scheme up till the 1980s. Although Australia has enforced a closure of these schemes in recent years, frozen plans will continue to accrue liabilities for the next few decades. Second, public sector plans tend to be characterized by much lower normal retirement ages and more generous pension benefits than corporate plans (see Palacios and Whitehouse, 2006; Beshears et al., 2011; Clark, 2012). This suggests that, all else equal, greater longevity among future cohorts of retirees is likely to hold larger fiscal implications for state employee pension liabilities.

Yet earlier studies examining the impact of inappropriate mortality assumptions on pension liabilities have largely focused on corporate plans rather than public sector plans (see Antolin, 2007; Pitacco et al., 2009; Dushi, 2010; Sithole, 2011). These studies generally conclude that private pension fund providers in the United States and other OECD countries fail to fully account for expected mortality improvements and thus understate reported pension liabilities. For example, Dushi, Friedberg, and Webb (2010) find that projected benefit obligations are understated by about 2.9–4.6 percent for typical male participants in DB plans if the plan provider uses RP-2000 mortality without incorporating population-based mortality declines.3 That study also shows that if mortality should decline at faster rates as predicted by the Lee–Carter model (Lee and Carter, 1992) instead, the resultant understatement in pension obligations will be higher (6.9–8.7 percent).4 In sum, the existing literature largely focuses on the omission of expected mortality improvements in pension liabilities valuation which, albeit important, is but one aspect of longevity risk.

In this article, we consider three additional aspects of longevity risk pertinent to the providers of the public sector pension promise, namely, the government and taxpayers. The first two risk aspects relate to differential pensioner longevity: public sector pensioners are potentially a select group that is longer-lived than the general population (“aggregate longevity selection risk”), and further, certain sectors of pensioners may be more long-lived than average (“scheme-specific longevity selection risk”). The third aspect is the risk of an unexpected, immediate, and permanent decrease in mortality (“longevity shock”). Together with the risk that pension providers may fail to account for expected mortality improvements (“computational risk”) as highlighted in earlier studies, we present—in total—four risk aspects ensuing from the uncertain mortality of public sector pensioners.

We find compelling evidence of longevity selection at both the aggregate level and the scheme level. Our estimates show that a 60-year-old male public sector pensioner, on average, has 46 percent lower mortality than a same-aged male randomly drawn from the general population. Furthermore, mortality is heterogeneous across schemes. Male pensioners in one of the occupational schemes are particularly long-lived; they face 30 percent lower mortality risk than the average pensioner, controlling for age and pension size. In contrast, pensioners in a scheme based in Western Australia have 7–16 percent higher mortality than average. Differential pensioner longevity is quantitatively important when evaluated against the two other risk aspects. Given an entry age of 60, the increase in pension obligation accruing from aggregate and scheme-specific longevity selection risk is 4.6 and 6.4 percent, respectively, which is higher than that arising from computational risk (2.7 percent) or a plausible longevity shock (2.0 percent). Moreover, we find that the resultant impact of scheme-specific selection risk (and also longevity shock) on liabilities increases with age. On the other hand, the impact of computational risk on liabilities decreases with age. This has implications for maturing public sector DB schemes around the world. As the age-membership structures in pension plans mature, more attention needs to be focused on managing risks stemming from scheme-specific selection and systematic permanent increases in longevity that have larger consequences for plan liabilities.

We also quantify plan liabilities by scheme, allowing for heterogeneity in pension received, entry age, and mortality across individuals.5 Our weighted estimates show that, as at valuation year 2006, the remaining pension obligation per public sector pensioner averages about A$0.367 million (male) and A$0.309 million (female) in present value terms. These figures are indicative of the huge fiscal burden on Australian governments with respect to legacy costs of the now-frozen public sector plans. Additionally, we find that older schemes tend to be associated with higher per capita liabilities largely due to more generous pension benefits. Our results should be of interest to policymakers concerned with overall public debt, as well as taxpayers and fund administrators. The findings are also germane to existing public sector DB plan participants whose income security may be threatened should the government decide to cut benefits in order to hold down increases in (unfunded) pension liabilities.6 Munnell et al. (2013) report that seriously underfunded U.S. state-administered pension plans have recently enacted a mix of benefit cuts and revenue increases. Governments in several developed countries have also reduced benefit entitlements in the broader social security system to resolve severe underfunding issues (Cocco and Lopes, 2011).

The article proceeds as follows. In the next section, we provide contextual background and discuss the magnitude of unfunded pension liabilities by state. We then describe the data and methodology used to empirically test for the presence of aggregate and scheme-specific longevity selection. This is followed by a two-part evaluation of the impact of longevity risk on pension liabilities: we first quantify the four aspects of longevity risk using a discounted annuity approach and then calculate the present dollar value of pension obligations for a cross-sectional sample of individual pensioners. The final section concludes.

UNFUNDED LIABILITIES IN STATE PENSION PLANS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

The Australian public sector, thought of as comprising federal (or Commonwealth) employees and the employees of Australia's six state and two territory governments, is an interesting setting to examine the impact of pensioner longevity on unfunded pension liabilities for two reasons. First, unlike some countries where public employee pension system borrowing is kept “off the books,” unfunded net pension liabilities in Australia are transparently calculated and acknowledged in all state and federal government budget documents. Politicians have also taken steps to close the funding gap in a systematic and gradual manner. Second, earlier studies suggest that public sector pensioners in Australia enjoy favorable longevity up to about age 85 (Knox and Nelson, 2007; Sithole, 2011; Mercer, 2011).7 Whether this has important fiscal implications in terms of pension obligations is an empirical question yet to be explored.

The public sector represents about 18 percent of the total Australian workforce (Industry Skills Council, 2012). Until the 1980s, it was common practice for the federal/state governments to require their employees to join a DB pension (or superannuation) scheme.8 As in the United Kingdom and Canada, the Australian public sector pension system is characterized by a large number of individual schemes from various employers in the public sector, rather than a large and harmonized centrally administered scheme. These schemes mostly have an unfunded employer-contribution component, which is financed on a pay-as-you-go (PAYG) basis.9 Retirement benefits are paid out as a lifetime pension, often inflation-indexed. Over the last 30 years, pension reforms in Australia have resulted in a progressive shift from DB to DC schemes. For example, the 2011–2012 NSW budget statement reports that less than 20 percent of NSW public sector workers are presently members of DB schemes (NSW Government, 2011). Nonetheless, the frozen DB schemes still contractually owe pension benefits that have already been promised and accrued to past employees and, furthermore, continue to accrue pension entitlements due to present active employees. These employee superannuation entitlements sit on the government's balance sheet as a financial liability.

To clarify the valuation of public sector employee pension liabilities, it is useful to begin with the information provided in the governments' annual budget documents. The reported “unfunded superannuation liabilities” in the budget documents reflect the difference between the estimated gross liabilities and assets of DB schemes within each jurisdiction. Pension assets are valued on a market value basis. Pension liabilities, on the other hand, are based on the present value of accrued pension entitlements, derived from forecasts of salary growth, CPI increases, retirement rates, and benefit payments. In accordance with the Australian accounting standard AASB 119 (Employee Benefits), the liabilities are discounted using long-term government bond rates.10 Note that this contrasts with the current practice in United States where public pension liabilities are discounted at the expected rate of return on pension assets (Novy-Marx and Rauh, 2011). Consequently, the levels of unfunded liabilities are somewhat comparable across Australian jurisdictions, but less so internationally.

Table 1 displays the published estimates of unfunded pension liabilities by jurisdiction. In June 2011, the present value of the total accrued pension entitlements due to public sector workers in DB schemes total A$210.0 billion.11 Of this, 62 percent (or A$129.5 billion) is attributable to three large federal schemes, the Commonwealth Super Scheme (CSS), the Public Sector Super Scheme (PSS), and the Defence Force Retirement and Death Benefits Scheme. The remaining 38 percent is split among seven state governments, with New South Wales and Victoria reporting relatively larger liabilities than other states. Interestingly, the Queensland government reports zero unfunded liabilities because of its long-standing policy of setting aside funds to prefund future employee entitlements. Evidently, as more plan participants start to transit into retirement, unfunded liabilities in these frozen DB plans have increased over time. For example, there is a 37 percent increase (from A$136.4 billion to $186.2 billion) between 2007 and 2009.

Table 1. Net Unfunded Pension Liabilities, 2007–2011 (in A$ Billion)
Jurisdiction20112010200920082007
  1. Note: All figures are in nominal terms and are obtained from the most recent budget release of the respective federal and state governments (as at June 30 of each financial year). Estimated actual/actual figures are reported for June 2011 and actual/revised figures are reported for June 2007–2010.

  2. Source: ACT Government (2011), Australian Government (2011), NSW Government (2011), Queensland Government (2011), South Australian Government (2011), Tasmanian Government (2011), Western Australia Government (2011), Northern Territory Government (2012), and Victorian Government (2012). Gross domestic product (GDP) figures at current prices are obtained from the Australian Bureau of Statistics.

Federal (Commonwealth, incl. defence)129.5122.9108.1102.795.5
New South Wales32.232.729.417.614.4
Victoria22.820.324.412.910.1
South Australia8.79.58.96.55.1
Western Australia7.37.47.25.95.7
Tasmania4.44.53.72.52.5
Queensland0.00.00.00.00.0
Northern Territory3.12.62.42.12.1
Australian Capital Territory2.01.92.01.11.1
Total210.0201.8186.2151.3136.4
% of GDP14%14%14%12%12%

In addition to a common valuation approach, federal/state governments have also engaged external actuaries to undertake a detailed actuarial review of the schemes every 3 years.12 The results of these triennial actuarial investigations have been central in ascertaining the size of (unfunded) liabilities and in determining the long-term financing strategy. Appendix A presents details from the actuarial review reports for two federal DB schemes. Note that the older Federal-CSS scheme (open 1976–1990) has more members who are already receiving pensions than the newer Federal-PSS scheme (open 1990–2005). Despite this difference in scheme demographics, the estimated unfunded liability per pensioner works out to be quite similar: a considerable A$0.338 million for CSS and A$0.334 million for PSS in 2008 dollars.13

With an aging population likely to place significant pressure on the Australian government's finances, concrete steps have been taken through legislation in recent years to finance the liabilities. In 2006, the Commonwealth government established “The Future Fund” as a mechanism to accumulate financial assets to provide for future unfunded pension liabilities associated with federal employees. As of March 2011, the Fund's assets are A$74.6 billion and it is expected to generate at least a benchmark return of the CPI plus 4.5–5.5 percent per annum over the long term (Australian Government, 2011). At the state level, governments have also strived to reduce the liabilities through regular contributions from public sector budgets. For example, the New South Wales government aims to fully fund the pension liabilities in the state's PAYG DB schemes by 2030 under the Fiscal Responsibility Act of 2005 (NSW Government, 2005). Similarly, the Victoria government aims to eliminate its pension liabilities by 2035.

PENSIONER DATA AND LONGEVITY SELECTION RISK

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

Data Overview

The information on public sector pensioner mortality used in this study is obtained from the Mercer Pensioner Database for years 2002–2010. The database is an ongoing effort by Mercer (Australia) to collect information on pensioners in Australian occupational pension funds and covers major public sector schemes from 2002 onward. The administrative data are of high quality and are reconciled from year to year to permit updating of older data particularly to redress late reported deaths or exits. It features a wide range of pensioner types, including retiree, spouse, child, and invalidity pensioners. Plan-level data include name and jurisdiction of the scheme. Unit-record data on existing age pensioners include pensioner ID, sex, annual pension received, birth date, commencement date, and cessation date as well as reason for cessation (if applicable).

Table 2 lists the individual schemes, together with the sample breakdown. Our sample comprises 158,623 retiree pensioners from 13 schemes, of which 12 belong to the general government sector.14 The exception is the NSW-EISS (Energy Industries Super Scheme), which technically belongs to the public trading enterprise sector but is included here because it still falls under the NSW government pension arrangements (NSW Government, 2011). Notably, this sample covers all the closed federal/state DB schemes in Australia that pertain to civilian public servants. The majority of the DB schemes closed in the 1990s, although a few had closed as early as the mid-1980s (e.g., the South Australian Super and Police Scheme and the NSW State Super Scheme). Among the last schemes to close are the Federal-PSS scheme and the Queensland schemes. By and large, all 13 schemes had closed as of the end of 2008.

Table 2. Sample Composition
Federal/StateName of SchemeAcronymSample Breakdown
# PensionersIn %
  1. Notes: NSW is New South Wales. The sample of age pensioners is observed from July 1, 2002 to June 30, 2010.

  2. a

    The first pension scheme for Commonwealth government employees was established in 1922. Members of this 1922 scheme were transferred to the CSS at its commencement in 1976.

  3. b

    The NSW-EISS scheme technically belongs to the public trading enterprise (PTE) sector. PTEs are public sector entities, which provide major economic infrastructure assets such as water, power, and public transport, and typically finance the bulk of their operations from own sources revenues and borrowings.

  4. c

    This scheme is part of the so-called Victoria Emergency Services and State Super.

  5. d

    This scheme is part of the so-called QSuper DB account.

FederalCommonwealth Superannuation Scheme (1922–1976–1990)aFederal-CSS68,20043.4
FederalPublic Sector Superannuation Scheme (1990–2005)Federal-PSS9,5635.5
NSWNew South Wales State Superannuation Scheme (1919–1985)NSW-SSS28,11517.2
NSWNew South Wales State Authorities Superannuation Scheme (1988–1992)NSW-SASS3,5942.2
NSWNew South Wales Police Superannuation Scheme (1907–1988)NSW-PSS4330.3
NSWNew South Wales Energy Industries Superannuation Schemeb (closed in 1985)NSW-EISS2860.2
VictoriaVictorian State Superannuation Fundc (closed in 1994)Vic-SSS28,43218.4
VictoriaVictorian Emergency Services Superannuation Schemeb (closed in 1994)Vic-ESSS1,0390.7
South AustraliaSouth Australian Superannuation and Police Scheme (closed in 1986)SA-SASS5,8313.6
TasmaniaTasmanian Retirement Benefits Fund (closed in 1999)TAS-RBF6,0663.9
Western AustraliaWestern Australian Government Superannuation Scheme (closed in 1996)WA-GESB6,1924.0
QueenslandQueensland State Superannuation Schemed (closed in 2008)Queen-SS7780.5
QueenslandQueensland State Police Superannuation Schemed (closed in 2008)Queen-PS940.1
   158,623100

This present analysis focuses on age pensioners who had retired on or after age 55.15 There are more pensioners in the two federal schemes than in the individual state schemes partly because government employees from the Australian Territories are subsumed under the former.16 The Federal-CSS scheme, which had its beginnings in 1922, accounts for 43.4 percent of the sample as most of its members have reached retirement age. The states of NSW and Victoria each contributes to roughly one-fifth of the sample with much smaller representation from the other four states. About 70 percent of the sample is male, consistent with the public sector workforce composition around the 1930s–1960s in Australia. Three of the occupational schemes, in fact, are predominantly male: the police schemes (NSW Police Super Scheme and Queensland State Police Super Scheme) and the NSW Energy Industries Super Scheme.

Aggregate Longevity Selection Risk

To determine the mortality experience of the public sector pensioners over an 8-year observation period (2002–2010), we first compute the observed age-specific mortality hazards by sex. This analysis treats the 111,257 male and 47,366 female pensioners separately since mortality is known to differ systematically by sex. Formally, assume that the underlying distribution of deaths is from a Poisson distribution with parameter inline image. The force of mortality for each exact age inline image is estimated as follows:

  • display math(1)

where dx is the observed number of deaths for lives aged x, inline image is the total time the pensioners are exposed to the risk of dying, and inline image is an estimate of the true mortality hazard. The in-sample tabulations of dx and inline image by sex are summarized in Appendix B. We find that 17.6 percent of the sample (or 27,937 pensioners) died during the observation period. The bulk of deaths occurred in the 80–84 and 85–89 age bands. About 0.7 percent of the deaths occur in the 100+ age bracket, indicative of the presence of right-tail longevity risk among public sector pensioners. The oldest surviving pensioner as of the 2010 cut-off is 103.8 years old.

Unlike large mortality investigations involving a broad population base (e.g., a census used to build the population life tables), our sample is comparatively small so the crude estimates inline image do not progress smoothly. To smooth the mortality estimates, we employ an actuarial technique known as “graduation by reference to a standard table.” This technique involves selecting an appropriate life table (typically the population life table) and using the shape of its mortality hazard function as the given standard for smoothing, and is suitable here for two reasons. First, the true underlying mortality of the pensioners' lives is by and large related to that of the general Australian population, particularly in the overall progression of mortality rates from age to age. Second, the sex-specific population life tables constructed from millions of lives provide a basis to estimate increases in mortality hazard at very advanced ages where pensioner data are scarce. Graduation by reference to the published population life tables is also convenient since it enables use of the published population mortality improvement rates by extension.

The choice of the standard table is important in the graduation process. There are two candidates: one is the Australian Life Table (ALT) published by the Australian Government Actuary and the other is the Australian Bureau of Statistics (ABS) Life Table. The sex-specific mortality rates in both tables are very similar up to about age 90.17 The ALT table, however, is more suited for our purposes since it extends up to age 109 and thus better captures the right-tail longevity among pensioners. The ABS table terminates at age 100.

Denote inline image as the set of graduated estimates we seek, and inline image as the set of hazard rates (forces of mortality) in the 2006 ALT period life tables. Given the assumed similarity in the true underlying mortality of the public sector pensioners' lives and that of the general population, simple functions are specified to link the pensioner mortality experience to the standard table. We first test a linear relationship: inline image (“function F1”), followed by a multiplicative relationship: inline image (“function F2”), where a, b, and c are suitable constants to be estimated. In order to adjust for heteroskedasticity, weighted least squares estimation is used. The weights (wx) are given by the reciprocal of the estimated variance of the crude rates:

  • display math(2)

Table 3 reports the weighted least squares estimates of the fitted parameters and standard errors for functions F1 and F2. Also reported are several test statistics assessing the error variance from the regression, as well as the fit and smoothness of the fitted rates. Overall, it appears that function F2 is superior since it leads to lower error variance (lower root mean square error) and better adherence to the data (higher Pearson correlation coefficient and lower chi-square test statistic). Nevertheless, we detect some trade-off between fit and smoothness. Smoothness is evaluated by calculating the sums of the squares of the third-order differences in the graduated values (see Bayo and Faber, 1983). The smoothness statistic for function F1 works out to be marginally lower than that for function F2 (difference is in five decimal places).

Table 3. Fitted Estimates Using Weighted Least Squares Methods
SexFunctionsF1: inline imageF2: inline image
  1. Notes: N = 111,257 male and N = 47,366 female retiree pensioners. Standard errors are in parentheses. MSE is mean square error. The test statistic that is superior in each row is shown in bold for ease of comparison between functions F1 and F2. A lower MSE value indicates lower error variance. A higher Pearson correlation coefficient (between the raw and graduated mortality rates) indicates better fit. A lower chi-square test statistic (using the observed and expected deaths) indicates better fit. A lower smoothness test statistic indicates greater smoothness in the graduated rates.

MalesFitted parametersâ = −0.00299 (0.00032)inline image = −0.60116 (0.09466)
   ĉ = 0.01949 (0.00119)
 Root MSE0.00230.0014
 Fit: Pearson correlation coefficient0.9140.930
 Fit: Chi-square test statistic (df)540.2 (48)171.3 (47)
 Smoothness test statistic1.67E053.15E−05
FemalesFitted parametersâ = −0.00146 (0.00019)inline image = −0.32260 (0.09716)
   ĉ = 0.01597 (0.00122)
 Root MSE0.00130.0009
 Fit: Pearson correlation coefficient0.9120.906
 Fit: Chi-square test statistic (df)129.1 (48)50.5 (47)
 Smoothness test statistic4.34E068.17E−06

Figure 1 plots the smoothed hazard rates, constructed based on 610,765 years of exposure in respect of male retirees and 260,233 years of exposure in respect of female retirees. The circle markers represent the raw hazard estimates and the lines represent the smoothed hazard function. The age axis extends up to 109, consistent with the standard table we used to perform the graduation. At advanced ages, we see considerable volatility in the raw estimates due to the small number of exposure years. This graphical comparison also confirms that function F2 (solid line) is a much better fit than function F1 (dotted line) for both sexes. In what follows, we use the F2 graduated mortality rates. Our set of mortality estimates technically applies to year 2006, which is the mid-point of the 2002–2010 observation window.

image

Figure 1. Smoothed Hazard Rates for Retiree Pensioners by Age and Sex

Notes: N = 111,257 male and N = 47,366 female retiree pensioners observed over 2002 to 2010.

Download figure to PowerPoint

As a major fund actuary for government pension schemes, Mercer (Australia) earlier developed pensioner mortality tables for use in its actuarial investigations. The “Mercer 0205” and “Mercer 0509” tables are widely recognized as the standard pensioner life table in the Australian pensions business (Sithole, 2011). The schemes that Mercer used to construct its mortality estimates, however, do not correspond directly to the 13 federal/state schemes in this present study (see Table 2). The Mercer 0205 table is based on the combined mortality experience from only five of the 13 schemes,18 whereas the Mercer 0509 table encompasses local government and quasi-government schemes that are beyond our scope. Consequently, we directly construct our own mortality estimates from raw data. Note also that our estimates are based on a longer 8-year observation period as compared to the 3- or 4-year observation periods used by Mercer.19 It is useful, however, to validate our discrete 1-year probabilities of death inline image against those in Mercer 0205. The first two columns of Table 4 show that our mortality estimates are quantitatively similar to Mercer's estimates for the 2006 valuation year; deviations are mostly well below 10 percent (see column 4).

Table 4. Pensioner and Population Death Probabilities (qx) by Sex
Age(1) Mercer 0205 (Updated to 2006)(2) This Study: 2006 Rates(3) 2006 ALT Table(4) = inline image Deviation From Mercer's Rates (%)(5) = (2)/(3) Ratio of Rates
  1. Source: Column (1) figures are derived using the mortality rates and future mortality improvements contained in the “Mercer 0205 updated to 2007” document obtained from private correspondence with Mercer.

Panel A: Males
550.00190.00210.004570.45
600.00370.00390.007260.54
650.00710.00760.012070.63
700.01370.01400.019230.73
750.02640.02690.033120.81
800.05140.05230.057620.91
850.09170.09880.099181.00
900.16300.17750.162991.09
950.26900.27270.231111.18
1000.32630.35530.282191.26
1050.38440.42440.3207101.32
Panel B: Females
550.00170.00150.0027−140.53
600.00290.00260.0044−90.61
650.00500.00470.0068−60.69
700.00850.00840.0112−10.75
750.01610.01640.019820.83
800.03200.03280.036620.89
850.06400.06860.070970.97
900.12770.13680.130971.05
950.23920.23510.2090−21.13
1000.32600.33750.282841.19
1050.40150.42160.336951.25

To investigate whether aggregate longevity selection risk exists, we compare pensioner mortality against population mortality. The ratio of rates is reported in the last column of Table 4; a value less than unity indicates the presence of longevity selection favoring the pensioners. We find that pensioner mortality is considerably lighter than the general population mortality at ages below 85–90, as consistent with Knox and Nelson (2007) and Sithole, Haberman, and Verrall (2011). For instance, a 60-year-old male pensioner has a one-period death probability of 0.0039, which is 46 percent lower than that of a general population male (0.0072). Similarly, an average 60-year-old female pensioner has a 39 percent lower mortality rate than a same-age female drawn from the population.20

Aggregate longevity selection may have ensued from two factors: the pensioners selecting into a public sector career, and subsequently electing to receive retirement benefits in the form of an indexed lifetime pension.21 Inasmuch as the option to switch to DC schemes is offered only in recent years and only in some of the DB schemes (e.g., not offered in the Federal-CSS scheme, which is the largest scheme in our sample), we infer that scheme mobility did not contribute greatly to the selection.22 The positive selection effect detected is thus likely due to Australian civil servants generally being more highly educated and having higher income. Several studies that examined public–private sector wage differentials using Australian data find that public sector employees enjoy higher wages than their private sector counterparts; civil servants also tend to be more educated and hold white-collar jobs (Baron and Cobb-Clark, 2008; Cai and Liu, 2011).23 Moreover, the public sector is typically associated with job stability whereas private-sector workers are more likely to move between jobs (Palacios and Whitehouse, 2006). Less visible to us, however, is whether the sample pensioners are also positively selected in terms of health. We have no data on the pensioners' health statuses and lifestyle choices, but because our sample excludes early retirees (people who may have retired early due to poor health) and invalidity pensioners, it is plausible that the normal retiree pensioners observed might have retired in good health.24

Interestingly, the results also show that pensioner mortality is heavier than population mortality from about age 90. The ratios exceed unity. One way to rationalize this is through the distribution of deaths among retiree pensioners. Because pensioners face lower chances of death below ages 85–90, most of them survive to advanced ages. Consequently, given an age ceiling, pensioner deaths are compressed at advanced ages above 85–90, causing mortality rates at those ages to be higher than those of the general population. In comparison, the death distribution in the general population may not be as skewed. Another reason proposed by earlier studies is that the selection effect among pensioners somehow diminishes or wears off over time (Knox and Nelson, 2007; Mercer, 2011). In sum, we conclude that aggregate longevity selection risk is characterized by the likelihood of large numbers of pensioners living to their 85th or 90th birthday before exiting the system, rather than an extreme long-tail risk.

Scheme-Specific Longevity Selection Risk

In the United States and United Kingdom, pensioner life tables have been constructed from actual annuitant mortality experience and used in place of population life tables in the valuation of retirement products such as payout annuities and lifetime pensions. While it is well established that mortality differentials exist between the pensioner subgroup and the general population, the extent to which longevity is heterogeneous among pensioners and whether that is quantitatively important in terms of valuation of pension liabilities is largely unclear. Most famously, a long-running cohort study of mortality among British male civil servants (the Whitehall Study) finds that workers in the lower employment grades (e.g., messengers, doorkeepers, etc.) had much higher mortality than those in the higher employment grades (e.g., administrators). Similarly, one may anticipate members of a police or emergency services public sector pension scheme to have higher postretirement mortality compared to members in occupational schemes for teachers or administrative workers.

We investigate how mortality differs by age, pension size, and occupational scheme in the sample of retiree pensioners. As before, we sample males and females separately. Applying a Cox proportional hazards model, the mortality hazard of a pensioner at a given time is:

  • display math(3)

where inline image is the resultant hazard rate for the jth subject given age t and the subject's vector of covariates inline image is the baseline hazard function, and β is the vector of regression coefficients to be estimated. This equation states that the death hazard that pensioner j faces is multiplicatively proportional to a baseline hazard that a same-aged person would face, modified by his or her personal characteristics expressed as a vector xj.

The semi-parametric Cox model is selected over a parametric estimation because it allows for greater flexibility. Also, we do not need to impose a parametric assumption on the underlying hazard function inline image. Because we had sufficient data to model the human (pensioner) mortality process over the desired age range of 55–110, we are able to leave the baseline hazard unconstrained (and consequently unestimated). In other words, the aggregate sex-specific smoothed mortality rates derived earlier using function F2 serves as the pseudo-baseline hazard.25

Prior studies using Australian data generally find a significant inverse relationship between socioeconomic status (SES) and mortality (Knox and Tomlin, 1997; Philip and Leigh, 2011).26 Here, we use pension size as a proxy for SES. Pension size is modeled as a categorical variable: “low,” “average” (ref.), and “high.” The distribution of annual pensions received (in real dollars) is positively skewed; a small number of pensioners receive payouts much higher than the average. Consequently, we set the 50th percentile (median) as the midpoint of the “average” category, and use the 25th percentile as the marker for the “low” category. The median pension is A$27,400 and A$17,900 for males and females, respectively. The pension size categories are < $16,000 (“low”), $16,000–38,000 (“average”), and ≥ $38,000 (“high”) for males, and < $9,000 (“low”), $18,000–27,000 (“average”), and ≥ $27,000 (“high”) for females.

We also create individual scheme dummy variables. The Federal-CSS scheme is set as the reference category since it is by far the largest scheme in the sample (accounts for 46 percent male and 37 percent female pensioners). Several of the individual scheme variables, however, do not satisfy the proportional hazards assumption based on a test of Schoenfeld residuals (Cleves et al., 2010).27 This is because some small schemes have noisy data, or data that are concentrated only around a short age span (e.g., all members are ages 80–95). One possible reason for the latter is scheme mobility whereby younger employees had taken up options to switched to a DC arrangement and exited the DB scheme. Consequently, we group schemes that do not independently satisfy the proportional hazards assumption into a “grouped schemes” category. This is done primarily to retain sample size, although it is noted upfront that the coefficient on this variable cannot be meaningfully interpreted. Members of these “grouped schemes” are assumed to face a scheme-specific mortality risk identical to that of the reference CSS scheme.28

Because few observables are available in the Mercer data set, pension size and scheme are the only two explanatory variables used. These are collectively entered into the regression. Table 5 reports the relative effects of pension size and scheme on mortality. A hazard ratio larger (smaller) than unity indicates an increased (decreased) risk of death associated with the explanatory variable. Across the board, we see that pension size has a statistically significant effect on mortality (p < 0.01). For any given age t, pensioners with low pension income face 13–23 percent higher risk of death than those receiving average pension. In contrast, pensioners with high annual pension income—and presumably higher SES—face between 23 and 36 percent lower mortality.

Table 5. Effects of Selected Covariates on Mortality
 MalesFemales
Hazard Ratio[95% CI]Hazard Ratio[95% CI]
  • Notes: Reported hazard ratios are the partial effects of the explanatory variables on the odds of mortality; 95% confidence intervals (CI) of the marginal effects are reported in square brackets.

  • NA(a) = not applicable, because this scheme comprises 95% or more males and is excluded from the female sample.

  • NA(b) = not applicable, because this individual scheme dummy does not independently satisfy the proportional hazards assumption, and so not included as a covariate.

  • ***

     Indicates 1% significance level;

  • **

     Indicates 5% significance level;

  • * Indicates 10% significance level.

Pension sizes:
Low1.23***[1.20,1.27]1.13***[1.06,1.21]
Average (ref.)1.001.00
High0.64***[0.61,0.66]0.77***[0.72,0.83]
Specific schemes:
Federal-CSS (ref.)1.001.00
Federal-PSS0.71***[0.60,0.84]0.84[0.67,1.05]
NSW-SSSNA(b) 1.05[0.97,1.13]
NSW-SASSNA(b) 0.90[0.73,1.10]
NSW-PSSNA(b) NA(a) 
NSW-EISSNA(b) NA(a) 
Victoria-SSSNA(b) NA(b) 
Victoria-ESSS1.01[0.91,1.12]0.86[0.54,1.37]
South Aust-SASSNA(b) NA(b) 
Tasmania-RBF1.00[0.94,1.08]1.01[0.86,1.17]
Western Aust-GESB1.07**[1.01,1.12]1.16**[1.01,1.32]
Queensland-SSNA(b) 0.97[0.76,1.22]
Queensland-PS1.38[0.90,2.12]NA(a) 
Grouped schemes1.03**[1.00,1.06]1.13***[1.05,1.22]
# subjects111,257 47,350 
Chi-squared (df)1,162 (8) 102 (10) 
Adjusted R23.2% 1.1% 

Table 5 also indicates that scheme variables have little explanatory power after controlling for age and pension size. The only individual scheme variable that is statistically significant for both sexes is that of the Western Aust-GESB scheme. Specifically, female pensioners in the GESB scheme face about 16 percent higher mortality risk after retirement than those in the reference category (CI = 1.01, 1.32; p < 0.05). GESB male pensioners similarly face significantly higher mortality risk than their male counterparts. Notably, this higher mortality risk detected for the GESB scheme could stem from a mix of factors including prior occupational profiles, geographical characteristics, health characteristics, etc. Nevertheless, we rule out differences in income since we have controlled for pension size.29

To investigate whether the mortality differentials are due to geographical differences, we turn to general population statistics. We find that the life expectancy of the population in Western Australia to be comparable to that of other states' population (see Table 6). We conclude that the scheme-specific mortality selection does not show up at the population level and is likely pertinent only to the schemes themselves. Lacking detailed demographic data on the pensioners, however, we are unable to pursue this further. In the same vein, the significantly lower mortality risk detected for male Federal-PSS pensioners is hard to pin down. In sum, these results provide evidence of scheme-specific longevity selection risk, implying that some sectors of public sector plan providers potentially face greater exposure to longevity risk than others.

Table 6. Life Expectancy by Jurisdiction
State/TerritoryLife Expectancy at BirthLife Expectancy at Age 65
MalesFemalesMalesFemales
  1. Note: The statistics for Western Australia are bolded for ease of comparison with those of other states.

  2. Source: Australian Bureau of Statistics (2006).

New South Wales78.083.317.721.2
Victoria78.583.317.921.2
Queensland77.882.917.821.1
South Australia78.083.117.721.3
Western Australia78.683.318.121.5
Tasmania76.781.816.820.2
Northern Territory72.378.016.419.0
Australian Capital Territory79.783.918.621.5
Average across states78.183.017.821.1

IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

Longevity risk affects the unfunded liabilities of DB pension plans through the expected annuity payments owed to existing pensioners over their remaining lifetimes. This section uses a discounted annuity value approach to examine the impact of longevity risk on pension obligations. Formally, the actuarial present value (APV) of the annuity payments for pensioner i can be expressed as:

  • display math(4)

where $Ai is the annual dollar pension (in real terms) received by the individual, ν is the discount rate, e is the age as at valuation date (or so-called entry age), and inline image is the set of cumulative survival probabilities differentiated by age, sex, pension size, and scheme.

In what follows, we base all calculations on a valuation year of 2006. This is selected to coincide with the mid-point of our observation span. We also assume a real annual interest rate of 3.5 percent, consistent with the historical margins between nominal rates and inflation in Australia for that period.30 A terminal age of 110 per the ALT population tables is used.

We evaluate the impact of longevity risk on pension liabilities in two segments. In the first segment, we quantify the four aspects of longevity risk by applying alternative sets of aggregate mortality assumptions for a representative individual. That is, we apply Equation (4) at an aggregated level (no individual parameters used), normalize $Ai to $1, stipulate a value for e, and only allow the mortality component to vary. In the second segment, we compute discounted annuity values for a cross-sectional sample of pensioners across all 13 DB schemes. Individual parameters are used. Specifically, $Ai and e are given by the individual's pension income and age as at 2006. inline image is derived for each pensioner using aggregate sex-specific pensioner cohort mortality, adjusted for individual pension size and scheme (adjustment factors per the Cox regression).

Four Aspects of Longevity Risk

To examine the impact of the four distinct aspects of longevity risk on pension liabilities, we assemble five sets of aggregate mortality assumptions, including:

  • “general population” mortality per the 2006 ALT tables,
  • “pensioner (period)” mortality per our mortality estimates for year 2006, and
  • “pensioner (cohort)” mortality, which is obtained by incorporating future mortality trends into “pensioner” mortality.

The other two sets of mortality assumptions are essentially derived from “pensioner (cohort)” mortality and described later.

To quantify aggregate longevity selection risk, we compute first the APV of the $1/year pension annuity using “general population” mortality, then the APV of the same annuity using “pensioner (period)” mortality, and evaluate the percentage difference between the two. In a similar manner, computational risk is quantified by the percentage difference in APVs arising from differences in “pensioner (period)” and “pensioner (cohort)” mortality. With the benefits of improved health technologies accruing to the elderly, the remaining life expectancy of an 80-year-old in 2036 is likely to be higher than that of an 80-year-old in 2006. Specifically, we use population-based mortality improvement factors obtained from the Australian Government Actuary. These improvement rates are suitable not only because they ensue from a firm historical basis (being carefully calibrated using actual mortality improvements for the Australian population over the preceding 100 years), but also because we seek to be consistent with the valuation approach used by Australian pension fund actuaries.31 The 2006 ALT improvement rates indicate that survival probabilities an older adult around age 60 are projected to increase in a nonuniform (hump-shape) manner over the remaining lifetime.32

Finally, setting the APV with “pensioner (cohort)” mortality as a benchmark, we derive the percentage increase in pension obligations under two scenarios. In one scenario, mortality is 29 percent lower than the benchmark across all ages. This follows from the greater longevity of the male pensioners in the Federal-PSS scheme noted earlier. Accordingly, the percentage increase in annuity value encapsulates scheme-specific longevity selection risk. In another scenario, mortality is 10 percent lower than the benchmark across all ages and times. This longevity shock emulates stress testing scenarios that are typically used to determine the longevity risk capital requirements in life insurance products (Stevens, 2010; Ngai and Sherris, 2011).33 The impact of a longevity shock is immediate, uniform, and persistent, unlike the nonuniform impact of expected mortality improvements.

Results appear in Table 7, where the columns denote various entry ages (e = 60, 70, 80, and 90) as at valuation year 2006. Focusing first on entry age 60, we see that a male pensioner obtains 4.6 percent more in discounted annuity value on average over his remaining lifetime relative to a male from the population-at-large, ignoring any mortality improvements. This is because the sampled public sector pensioners are longer-lived. Taking into account expected mortality improvements, the APV attributable to the pensioner will be another 2.7 percent higher. This 2.7 percent is the resultant understatement in unfunded liabilities if the plan provider used period mortality rather than cohort mortality, which is roughly of the same magnitude as the 3.3 percent estimated by Antolin (2007) for a 65-year-old pensioner.34 An unexpected flat 10 percent longevity shock would increase APV by 2.0 percent. In addition, the Federal-PSS plan provider faces 6.4 percent higher pension liabilities than other public sector employers because male PSS pensioners have significantly higher survival chances. This effect is relatively larger than the other three aspects of longevity risk as the advantageous longevity of these male pensioners is rather substantial; in fact, it is equivalent to a flat 29 percent longevity shock.

Table 7. Percentage Changes in Discounted Annuity Values Under Varied Mortality Assumptions (Male)
Aspects of Longevity RiskMortality Assumptions% Change in APV Values for Different Ages
60 (%)70 (%)80 (%)90 (%)
Aggregate longevity selection riskGeneral population vs. pensioner (period)4.63.4−0.4−8.9
Computational riskPensioner (period) vs. pensioner (cohort)2.72.31.60.7
Scheme-specific longevity selection riskPensioner (cohort): Without vs. with 29% lower mortality per Federal-PSS scheme6.410.316.523.7
Longevity shockPensioner (cohort): Without vs. with 10% lower mortality2.03.35.27.7

Table 7 reveals that the exposure effects are sensitive to the age profile of the pensioner. We see that the impact of computational risk is inversely related to age, as noted earlier by Antolin (2007). This is because the exposure of the pension fund to future mortality improvements is smaller the older the pensioner is today. In contrast, both the impact of scheme-specific selection and longevity shocks increases with age. This is attributable to larger absolute changes in mortality at advanced ages (for a uniform relative increase in survival probabilities across ages), which shows up more prominently for a 80-year-old than for a 60-year-old due to lighter discounting.35 Interestingly, the impact of aggregate longevity selection decreases with age but flips at the advanced ages. This follows from our earlier observation that the pensioners' advantageous longevity is strongest immediately after retirement and does not perpetuate into advanced ages. Consequently, an age 90 male pensioner actually receives 8.9 percent less in discounted annuity value than a same-aged male randomly drawn from the general population. We conclude that scheme-specific selection and unexpected but persistent declines in mortality are the two most quantitatively important longevity risk aspects for schemes with older age-membership structures. Given a maturing public sector workforce in many developed nations, more attention should perhaps be focused on managing risks stemming from these two risk aspects.

Individual Discounted Annuity Values Summed by Schemes

The second segment of the analysis allows for heterogeneity in terms of pension amount, entry age, and mortality at the individual level. We begin by extracting a cross-sectional sample of pensioners as of June 30, 2006 comprising 80,330 male pensioners and 33,692 female pensioners. The female sample excludes the NSW-PSS, NSW-EISS, and Queensland-PS schemes in which there are fewer than 10 female subjects. For each individual, we compute the APV using Equation (4) where e is the age as at 2006, $Ai is pension amount of at June 30, 2006, and inline image is the “scheme-specific pensioner (cohort)” mortality obtained by applying the scheme's mortality adjustment factor from the Cox regression to sex-specific “pensioner (cohort)” mortality.

The valuation results are summed by schemes and reported in Table 8. We rank the schemes in decreasing order of “per capita APV,” which is derived using the scheme's total APV divided by number of pensioners. This measure allows comparability in pension liabilities across schemes. Per capita APV ranges from A$100,855 to A$586,144 for male pensioners; the dispersion for female pensioners is slightly smaller (from A$100,740 to A$526,928). One conclusion from these results is that—on average—an existing male public sector pensioner in year 2006 can look forward to an annuity valued at about A$0.367 million in present value terms over his remaining lifetime (weighted by the schemes' membership size). The corresponding annuity value for female pensioners is A$0.309 million, smaller because of their generally lower pension income.36 Liabilities estimates presented in the actuarial review reports of two large schemes. The liabilities estimates of A$0.29–0.34 million per pensioner derived from the actuarial review reports of two large schemes corroborate these findings (see Appendix A).

Table 8. Annuity Valuation by Schemes (as of June 30, 2006)
Schemes# MalesTotal APV (A$m)Per Capita APV (A$)% by Pension Size% by AgeChange in Total APV (10% Shock)
LowAverageHigh< 65Age 65–80> 80
  1. Note: APV = actuarial present value of the pension annuity; see Equation (4) in text.

  2. The pension size categories correspond to those used in the Cox regression. For males, these are given by < $16,000 (“low”), $16,000–38,000 (“average”), and ≥ $38,000 (“high”). The pension categories for females are < $9,000 (“low”), $18,000–27,000 (“average”), and ≥ $27,000 (“high”). The female cross-sectional sample excludes three schemes, which are predominantly male; no pension annuity valuation is performed for these schemes.

Panel A: Males
NSW-EISS16094586,144112763395742.4%
NSW-SSS11,3695,917520,4661932492449273.2%
NSW-PSS270133493,1325293723704.8%
Federal-CSS38,27315,149395,8211550353541252.7%
Federal-PSS2,267835368,155463123564402.0%
South Aust-SASS2,934985335,7381647371054363.7%
Tasmania-RBF3,3301,114334,5743047242259193.0%
Victoria-SSS15,0713,860256,147375674835172.8%
Western Aust-GESB4,1301,025248,065214831548474.3%
Queensland-PS4511241,2532464111856273.6%
Queensland-SS29470239,4553743191648363.5%
Victoria-ESSS663104157,030265816025755.6%
NSW-SASS1,524154100,85581172825673.8%
 Total = 80,330 Weighted average = $366,627       
Panel B: Females
NSW-SSS7,9124,169526,928924664341162.3%
South Aust-SASS736214291,1631244441058323.1%
Federal-CSS13,3513,475260,2702353243349182.2%
Tasmania-RBF1,198283236,0642954163154142.4%
Federal-PSS2,419566234,075365311663401.7%
Western Aust-GESB934211225,497134146357403.8%
Queensland-SS23552219,6122057231850322.9%
Victoria-SSS6,4561,358210,3183159104836152.3%
NSW-SASS40264159,568395742152272.4%
Victoria-ESSS495100,74024591608925.5%
 Total = 33,692 Weighted average = $308,571       

Another conclusion is that older schemes tend to be associated with higher per capita liabilities largely due to more generous pension benefits. The three top-ranked schemes in the male sample (NSW-EISS, NSW-SSS, and NSW-PSS) were in operation before 1920 and among the earliest public sector DB schemes to close.37 The distribution of pension sizes in these older schemes is heavily skewed: 63–93 percent of members in the NSW-PSS and NSW-EISS receive high pension incomes exceeding A$38,000 annually (compared to 35 percent in the reference CSS scheme). Similarly for the female sample, the two largest per capita APV values are associated with old schemes providing generous benefits (the NSW-SSS and South Aust-SASS). Note also that pension levels are quite dispersed across schemes even though they are all under the ambit of the public sector.

Earlier in the Cox regression, we find that male pensioners in the Federal-PSS have lower than average mortality and those in the Western Aust-GESB have higher than average mortality. Yet we see from Table 8 that Federal-PSS did not rank very highly in terms of per capita APV in the table, ranking even below the CSS. This is partly rationalized by the fact that the bulk of its members receive only low to average-sized pensions. Also, the Federal-PSS scheme has zero pensioners above age 80 at valuation date. Thus the impact of scheme-specific longevity selection risk (which is positively related to age) is currently muted by the youthful age composition of the scheme. We posit that the Federal-PSS scheme's per capita APV may soon outrank that of the other schemes as more of its members reach advanced ages.

The last column of Table 8 illustrates the sensitivity of these APV estimates to the flat 10 percent longevity shock. As expect, we see that the shock has a greater impact on the total APV of schemes with older age-membership structures. For instance, the NSW-PSS and Victoria-ESSS schemes (which comprise of at least 70 percent male pensioners above age 80) experience a 5–6 percent increase in pension obligations, compared to a modest 2.7 percent for the CSS scheme where only a quarter of the pensioners are in advanced ages. Similarly, the increase in total APV ensuing from the longevity shock is largest for the Victoria-ESSS scheme (5.5 percent) in the female sample since more than 90 percent of its members are above age 80. In contrast, the Federal-PSS scheme with an extremely youthful age composition experiences only a modest 1.7 percent increase in pension obligations. The Pearson correlation between the percentage of members above age 80 and the change in total APV from the longevity shock is 0.93 (males) and 0.98 (females).

CONCLUSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

We consider the exposure of unfunded pension liabilities to longevity risk in DB plans using a rich new panel data set of over 150,000 pensioners from a complete set of closed, major civil service retirement schemes in Australia. These DB schemes have been frozen for some time but continue to accrue substantial amounts of unfunded financial liabilities owed to past and present public sector employees. Importantly, our assessment distinguishes between four aspects of longevity risk that is pertinent to public pension providers, including aggregate longevity selection risk, scheme-specific selection risk, computational risk, and unexpected longevity shocks.

We find evidence of longevity selection at the aggregate level over a large range of age values. For example, a 60-year-old male pensioner has a one-period death probability, which is just half of that for the population-at-large. This selection risk, when quantified in present dollar value terms, implies that the age 60 pensioner is owed 4.6 percent more in pension benefits than a same-aged male randomly drawn from the population. Notably, the fiscal impact of aggregate selection decreases with age. At the same time, certain sectors of public pension providers are exposed to scheme-specific longevity selection risk. In our nationally representative sample, we find that male pensioners in one of the occupational schemes (Federal-PSS) are associated with almost 30 percent lower mortality risk. Because of his greater longevity, the Federal-PSS pensioner can expect to receive approximately 6.4 percent more in present dollar value terms (assuming an entry age of 60) over his remaining lifetime as compared to the average public sector pensioner. This translates into roughly A$23,500 more per person in pension obligations that the administrator of the Federal-PSS scheme needs to fund.38

We also quantify the impact of computational risk and an unexpected longevity shock on the level of (unfunded) liabilities. A fund provider who omits future expected mortality improvements will underestimate pension liabilities by 2.7 percent (assuming an entry age of 60). As noted in earlier studies, this exposure is inversely related to age; the resultant underestimation in liabilities is only 1.6 percent if the pensioner is already age 80. Conversely, an immediate and persistent 10 percent decrease in mortality has greater consequences on plan liabilities when the age-membership structure of a scheme is older. This is because a uniform increase in survival probabilities causes larger absolute changes in mortality at older ages than at younger ages. Consequently, our results highlight that more attention needs to be focused on managing risks stemming from scheme-specific selection and systemic longevity shocks as the age composition in pension plans mature.

Unfunded PAYG pension schemes can become unsustainable as populations age, because fewer and fewer workers finance growing numbers of retirees. Aside from the public sector, DB pension promises are also an important legacy issue for several large corporations in Australia (The Australian, ).39 These include mining companies like BHP Billiton and Rio Tinto, Woolworths, Qantas Airlines, and the four major Australian banks. Nonetheless, most of these corporate DB plans are funded. Towers Watson (2009) reports that the unfunded DB pension liability on the books of listed Australian companies is only A$2 billion in mid-2008, which is a tiny fraction of the A$186 billion reported in the same period by the state governments combined. This could be in part attributed to the more generous pension benefits in public sector plans; for example, Palacios and Whitehouse (2006) estimate that a public sector worker in Australia can expect a replacement rate of around 66–88 percent as compared to 52 percent for private sector worker. Our estimates also show that, on average, the remaining pension obligation per public sector pensioner is about A$0.3 million in present value terms (as at valuation year 2006). Such figures are indicative of the huge fiscal burden on the Australian government with respect to accrued unfunded pension liabilities for the next few decades.

A sustained effort has been made by the federal and state governments in Australia to ultimately eliminate these net pension liabilities from public sector balance sheets. An early measure was to initiate a nation-wide changeover to a DC pension system and completely freeze the DB schemes. Most prominently, the introduction of the Superannuation Guarantee mandate in 1992 required that both public and private sectors employers contribute into a DC fund nominated by the employee, with the quantum of employer-contribution determined by the central government (9 percent in 2002–2012 and 12 percent by 2019). Notably, this changeover was not accompanied by an explicit reduction in DB pension benefits although benefits have been made implicitly less generous through rises in the “normal retirement age” over time. We find evidence of this in our broad cross-sectional analysis of unfunded liabilities levels across schemes whereby pensioners in older public sector schemes are observed to have higher per capita annuity values, stemming from their more generous pension benefits. Other government efforts to reduce the preexisting unfunded pension liabilities in Australia include the establishment of specific funds to prefinance future liabilities (such as “The Future Fund” by the Commonwealth government), and legislating regular contributions to target a gradual fall in liabilities.

APPENDIX

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

Extract of Unfunded Liabilities Estimates From Actuarial Review Reports

The table compares the scheme demographics and unfunded liabilities for two large federal DB schemes. The Federal-CSS scheme was open from 1976 to 1990 and replaced by the Federal-PSS scheme (1990–2005). Total members comprise contributors, preserved members, and pensioners, all of whom accrue liabilities.40 Because the PSS scheme is newer, most of its members are still contributors; less than 7 percent are pensioners. By contrast, more than 70 percent of CSS members are already pensioners, of which three-quarters are age (or retiree) and invalidity pensioners. Statistics extracted from the schemes' actuarial review reports show specifically the estimated unfunded liability pertaining to pensioners. From this, we derive a ballpark estimate of unfunded liability per pensioner which works out to be A$0.338 million (CSS) and A$0.334 million (PSS) in 2008 dollars.

 Federal-CSSFederal-PSS
2008200520082005
  1. Note: The unfunded liability represents an estimate of the accrued pension liabilities in respect of service up to June 30 of each valuation year for which no assets are held. The proportion of total unfunded liability accruing solely to pensioners is reported in parentheses.

  2. Source: Australian Government (2006, 2009).

Scheme demographics:
Total number of members149,055157,821252,354252,025
Number of pensioners [line A]115,432113,58816,45211,419
% age and invalidity pensioners75.3%74.9%95.8%95.6%
% dependent pensioners (spouse or child)24.7%25.1%4.2%4.4%
Accrued unfunded liability (A$m):
Estimated total unfunded liability$59,20050,60020,90013,800
Estimated unfunded liability for pensioners [line B]$39,000 (66%)31,900 (63%)5,500 (26%)3,300 (24%)
Unfunded liability per pensioner (=B/A)$0.3380.2810.3340.289

APPENDIX

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References

Total Observed Deaths and Exposure Years by Sex and Age Bands

This table shows the inputs used to derive the graduated mortality estimates inline image. For the given sample of 158,623 retiree pensioners, we tabulate the observed number of deaths (dx) for lives aged x and the total observed waiting time exposed to the risk of dying inline image. In actuarial terminology, inline image is the central exposed to risk. In total, we have almost 880,000 exposure years summed over all 13 schemes for the 8-year period (July 1, 2002–June 30, 2010). In all 27,937 pensioners (or 17.6 percent) died during observation.41 Breaking down the statistics by sex, we see that the number of exposure years for males is higher than that of females. Males also account for a disproportionate percentage of total deaths. Eighty-two percent of the 27,937 deceased are males. For both sexes, however, the majority of deaths are observed in the 80–84 and 85–89 age bands. About 0.7 percent of the deaths occur in the 100+ age bracket; the maximum observed death ages are 106.5 for males and 105.4 for females. Exposure for the last age group (100+) is relatively small. Nonetheless, we decided to include this in the analysis since longevity is a right-tail risk but care is taken in interpreting any results pertaining to this age group.

Age BandTotalMaleFemale
Observed DeathsExposure YearsObserved DeathsExposure YearsObserved DeathsExposure Years
55–59358136,85425687,72210249,132
60–64808178,765596121,12321257,641
65–691,085134,85983690,08424944,775
70–741,756101,7871,40369,35135332,437
75–794,329122,3913,60289,97672732,415
80–847,356117,0986,17990,9581,17726,140
85–896,78457,7605,67645,3541,10812,405
90–944,02817,7693,27113,5827574,187
95–991,2393,4428972,4323421,010
100+19436913224862122
Total27,937871,09522,848610,8315,089260,264
N158,623111,25747,366

References

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. UNFUNDED LIABILITIES IN STATE PENSION PLANS
  5. PENSIONER DATA AND LONGEVITY SELECTION RISK
  6. IMPACT OF LONGEVITY RISK ON PENSION LIABILITIES
  7. CONCLUSION
  8. APPENDIX
  9. APPENDIX
  10. References
  • ACT Government, 2011, Budget Paper No. 3, Budget Overview.
  • Antolin, P., 2007, Longevity Risk and Private Pensions, OECD Working Papers on Insurance and Private Pensions No. 3.
  • The Australian, 2009, Australian Companies Face Pension Fund Shortfalls, April 16. World Wide Web: www.theaustralian.com.au/business/pension-fundshortfalls/story-e6frg8zx-1225699572953 (accessed December 10, 2012).
  • Australian Bureau of Statistics, 2006, Australian Social Trends. World Wide Web: www.abs.gov.au/AUSSTATS/abs@.nsf/DetailsPage/4102.02006?OpenDocument (accessed December 10, 2012).
  • Australian Government, 2009, PSS and CSS Long Term Cost Report 2008: A Report on the Long Term Costs of the Public Sector Superannuation Scheme and the Commonwealth Superannuation Scheme, Prepared by Mercer (Australia) for the Department of Finance and Deregulation.
  • Australian Government, 2011, Budget 2011–12, Budget Paper No. 1, Budget Strategy and Outlook, May.
  • Australian Government, 2006, PSS and CSS Long Term Cost Report 2005, Prepared by Mercer (Australia) for the Department of Finance and Administration.
  • Baron, J., and D. Cobb-Clark, 2008, Occupational Segregation and the Gender Wage Gap in Private- and Public-Sector Employment: A Distributional Analysis, Institute for the Study of Labour Discussion Paper No. 3562, June.
  • Bateman, H., and J. Piggott, 2012, Civil Services and Military Retirement Income Provision in Australia, in: N. Takayama, ed., Reforming Pensions for Civil and Military Servants (Tokyo: Maruzen Publishing), pp. 29-54.
  • Bayo, F. R., and J. F. Faber, 1983, Mortality Experience Around Age 100, Transactions of Society of Actuaries, 35: 37-64.
  • Beshears, J., J. Choi, D. Laibson, and B. Madrian, 2011, Behavior Economics Perspective on Public Sector Pension Plans, Journal of Pension Economics and Finance, 10(2): 315-336.
  • Bikker, J., D. Broeders, D. Hollanders, and E. Ponds, 2012, Pension Funds' Asset Allocation and Participant Age: A Test of the Life-Cycle Model, Journal of Risk and Insurance, 79(3): 595-618.
  • Booth, H., J. Maindonald, and L. Smith, 2002, Applying Lee–Carter Under Conditions of Variable Mortality Decline, Population Studies: A Journal of Demography, 56(3): 325-336.
  • Borland, J., and R. Gregory, 1999, Recent Development in Public Sector Labour Market, in: O. Ashenfelter and R. Layard, eds., Handbook of Labour Economics (Amsterdam: North Holland), pp. 3573-3630.
  • Brown, J. R., R. Clark, and J. Rauh, 2011, The Economics of State and Local Pensions, Journal of Pension Economics and Finance, 10: 161-172.
  • Cai, L., and A. Y. Liu, 2011, Public-Private Wage Gap in Australia: Variation Along the Distribution, British Journal of Industrial Relations, 49: 362-390.
  • Chang, S., and H. Cheng, 2002, Pension Valuation Under Uncertainties: Implementation of a Stochastic and Dynamic Monitoring System, Journal of Risk and Insurance, 69(2): 171-192.
  • Clark, R., 2012, Evolution of Public Sector Retirement Plans: Crisis, Challenges, and Change, ABA Journal of Labor and Employment Law, 27(2): 257-273.
  • Cleves M., R. Gutierrez, W. Gould, and Y. Marchenko, 2010, An Introduction to Survival Analysis Using Stata, 3rd edition (College Station, TX: Stata Press).
  • Cocco, J., and P. Lopes, 2011, Defined Benefit or Defined Contribution? A Study of Pension Choices, Journal of Risk and Insurance, 78(4): 931-960.
  • Commonwealth Superannuation Corporation, 2008, PSS: Explore Your Expanded Options Booklet, July. World Wide Web: www.pss.gov.au/forms-and-publications/publications/ (accessed December 10, 2012).
  • Dushi, I., L. Friedberg, and T. Webb, 2010, The Impact of Aggregate Mortality Risk on Defined Benefit Pension Plans, Journal of Pension Economics and Finance, 9(4): 481-503.
  • HM Treasury, 2011, Independent Public Service Pensions Commission: Final Report (by Lord Hutton of Furness), March.
  • Industry Skills Council, 2012, Public Sector: Industry Overview. World Wide Web: www.governmentskills.com.au/ (accessed December 10, 2012).
  • Knox, D. M., and M. Nelson, 2007, The Mortality Experience of Australian Superannuation Pensioners: Some Surprising Results, Working Paper presented at the International Actuarial Association 2nd PBSS Section Colloquium.
  • Knox, D. M., and A. Tomlin, 1997, An Analysis of Pensioner Mortality by Pre-retirement Income, Insurance: Mathematics and Economics, 20(3): 255.
  • Lee, R. D., and L. R. Carter, 1992, Modeling and Forecasting U.S. Mortality, Journal of the American Statistical Association, 87: 659-671.
  • Mercer, 2011, Pensioner Mortality Investigation: 2005–2009, Australian Prudential Regulation Authority Research Report, Mercer Australia, February 14.
  • Milevsky, M., and K. Song, 2010, Do Markets Like Frozen Defined Benefit Pensions? An Event Study, Journal of Risk and Insurance, 77(4): 893-909.
  • Munnell, A., J. Aubry, A. Belbase, and J. Hurwitz, 2013, State and Local Pension Costs: Pre-Crisis, Post-Crisis, and Post-Reform, Brief No. 30, Center for Retirement Research at Boston College, March.
  • New South Wales Government, 2005, Fiscal Responsibility Act 2005 (Act No. 41), June. World Wide Web: www.legislation.nsw.gov.au/sessionalview/sessional/act/2005-41.pdf (accessed December 10, 2012).
  • New South Wales Government, 2011, NSW State Budget 2011–12, Budget Paper No. 2, Budget Statement, Chapter 7: Liability Management.
  • Ngai, A., and M. Sherris, 2011, Longevity Risk Management for Life and Variable Annuities: The Effectiveness of Static Hedging Using Longevity Bonds and Derivatives, Insurance: Mathematics and Economics, 49: 100-114.
  • Northern Territory Government, 2012, NT Government 2012–13 Budget, Budget Paper No. 2, Fiscal and Economic Outlook, Chapter 8: Uniform Presentation Framework.
  • Novy-Marx, R., and J. Rauh, 2011, Public Pension Promises: How Big Are They and What Are They Worth? Journal of Finance, 66(4): 1211-1249.
  • Palacios, R., and E. Whitehouse, 2006, Civil-Service Pension Schemes Around the World, World Bank, Social Protection Discussion Paper No. 0602, May.
  • Philip, C., and A. Leigh, 2011, Death, Dollars and Degrees: Socio-Economic Status and Longevity in Australia, Economic Papers, 20(3): 345-355.
  • Pitacco, E., M. Denuit, S. Haberman, and A. Olivieri, 2009, Modelling Longevity Dynamics for Pensions and Annuity Business (Oxford: Oxford University Press).
  • Queensland Government, 2011, State Budget 2011–12, Budget Paper No. 2, Budget Strategy and Outlook, Statement 1: Fiscal Strategy, Performance and Outlook.
  • Sithole T. Z., S. Haberman, and R. J. Verrall, 2011, Second International Comparative Study of Mortality Tables for Pension Fund Retirees, The Actuarial Professional Discussion Paper presented to the Institute and Faculty of Actuaries, June.
  • Society of Actuaries, 2000, The RP-2000 Mortality Tables. World Wide Web: http://www.soa.org/ (accessed December 10, 2012).
  • South Australian Government, 2011, Budget Paper No. 3, Chapter 4: Managing the State's Assets and Liabilities.
  • Stevens R., A. De Waegenaere, and B. Melenberg, 2010, Calculating Capital Requirements for Longevity Risk in Life Insurance Products. Using an Internal Model in Line With Solvency II, Working Paper, Netspar.
  • Tasmanian Government, 2011, Budget Paper No. 1, Statement 6: Assets and Liabilities.
  • Towers Watson, 2009, Watson Wyatt Estimates ASX-Listed Companies' Unfunded Defined Benefit Super Liability at $25—A Significant Increase in Second Half of 2008, Press Releases, April. World Wide Web: www.watsonwyatt.com/render.asp?id=20962&catid=1 (accessed December 10, 2012).
  • Victorian Government, 2012, Victorian State Budget 2012–13, Budget Paper No. 2, Strategy and Outlook, Chapter 4: Budget Position and Outlook, May.
  • Western Australian Government, 2011, 2011–12 Budget, Budget Paper No. 3, Economic and Fiscal Outlook, Appendix 1, May.
Notes
  1. 1

    In high income OECD countries—including United States, United Kingdom, Australia, Switzerland, and others—it is reported that spending on civil service pensions makes up one-quarter of total pension spending (Palacios and Whitehouse, 2006). Unfunded liabilities in closed public sector DB schemes amount to some 15 percent of GDP in Australia (Bateman and Piggott, 2012). Concerns in the United States focus on the implications of pension liabilities on state-level budget balance requirements and local pensions (Brown, 2011; Clark, 2012). In emerging economies such as Brazil, China, and India, civil service pensions are fast becoming a major fiscal burden given low income and limited tax bases.

  2. 2

    See Society of Actuaries (2000) for a distribution of exposed life-years by industry.

  3. 3

    Dushi, Friedberg, and Webb (2010) assume an interest rate of 6.17 percent and that employees retire exactly at age 60. Population-based mortality forecasts are based on the Social Security Administration's (SSA) intermediate mortality forecast for the U.S. population per the SSA Trustees' Report.

  4. 4

    This is based on the central (50th percentile) forecast from the Lee–Carter model. Additionally, Dushi, Friedberg, and Webb (2010) evaluate the consequences for plan liabilities if aggregate mortality declines unexpectedly faster than is predicted by the central forecast by taking the upper bound of the 95 and 99 percent confidence intervals of life expectancy projections calculated from the Lee–Carter model.

  5. 5

    Lacking micro data, Antolin (2007) assumes hypothetical DB funds with different age-membership structures. Dushi, Friedberg, and Webb (2010) use cross-sectional statistics on mean tenure, salary, and pensions by broad age groups. Nonetheless, focusing on the mean (or median) pensioner can conceal wide differences in pension sizes and age structures across the pensioner population.

  6. 6

    For instance in the United Kingdom, there have been recent recommendations to peg the pension increments for civil servants to the (lower) CPI rather than the (higher) RPI index, and to calculate pensions based on career average salary rather than final salary (HM Treasury, 2011).

  7. 7

    Past ages 85–95, however, these studies generally find that pensioner mortality is heavier than the general population.

  8. 8

    See Knox and Nelson (2007). The term “superannuation” is the Australian vernacular for “employer-based pensions.” The majority of pension schemes in Australia are coined “super scheme” with “super” being the abbreviation for “superannuation.”

  9. 9

    An exception is the closed Queensland QSuper DB scheme, which features fully funded employer contribution. DB schemes under local governments or quasi-government organizations are not included in this present article.

  10. 10

    In practice, schemes use either a floating discount rate per the 10-year Australian Government Bond rates (∼6.6 percent as at June 2008 and 5.2 percent as at Jun 2010), or an actuarially-determined discount rate that is relatively close to the long-term bond rate.

  11. 11

    As at end May 2012, the Australian dollar traded at almost parity with the U.S. dollar.

  12. 12

    Chang and Cheng (2002) discuss the importance of such regular valuations by pension actuaries and formulate a stochastic and dynamic monitoring system for use in defined benefit pension plans.

  13. 13

    These figures are derived using the actuary's estimate of “unfunded liability for pensioners” divided by the given number of pensioners in that year. Pensioners include all age, invalidity, and dependent pensioners. Actuaries' estimates of unfunded liability incorporate population-based mortality improvements.

  14. 14

    These 12 schemes are identical to those listed in Bateman and Piggott (2012), and cover the entire population of closed public sector DB schemes. The only federal DB schemes left open are the Defence Force Retirement and Death Benefits scheme and a small scheme for judges (Bateman and Piggott, 2012). We focus on the accumulated benefit obligation in the frozen schemes; future pension liabilities for current contributors are not included in our analyses.

  15. 15

    Normal retirement age in the Australian public sector varies from 55 to 65. Our sample excludes invalidity pensioners and dependent pensioners (child and spouse). We also did not include early retirees in the analysis since the mortality experience of early retirees differs from that of normal retirees. The Mercer database has a small sample (∼25,000) of early retirees. A two-tail paired t-test confirms that the raw mortality rates of male early retirees are significantly different from that of male retirees at the 5 percent level.

  16. 16

    Government employees in the Northern Territory and the Australian Capital Territory were enrolled in the federal DB pension schemes for Commonwealth government employees prior to 1988 and 2005, respectively.

  17. 17

    The ALT life table is constructed using deaths and estimates of population over a period of 3 years centered on a Census. The ABS life table is based on the deaths and population data of Australian residents who are physically present in Australia over a 3-year period. We use the 2005–2007 (or year 2006) ALT and ABS tables.

  18. 18

    The five schemes are NSW State Superannuation Scheme, Victorian State Superannuation Fund, Western Australian Government Superannuation Scheme, and the Commonwealth PSS and CSS schemes (Mercer, 2011). Also, as the names of the tables suggest, Mercer 0205 is based on mortality observed over July 1, 2002 to June 30, 2005 (3 years), and Mercer 0509 is based on mortality observed over July 1, 2005 to June 30, 2009 (4 years).

  19. 19

    In addition, we graduated the mortality rates by reference to the population life table but Mercer employed a combination of manual and parametric smoothing (Sithole, 2011).

  20. 20

    Note that our sample excludes early retiree pensioners.

  21. 21

    In lieu of a lifetime pension, the plan participant could have elected to receive retirement benefits in the form of a lump-sum benefit, or switch to a DC scheme prior to retirement. Approximately 60 percent of members in the federal DB schemes are reported to take out benefits in the form of pensions (Australian Government, 2009). Note that idiosyncratic longevity risk (that any one particular pensioner may live longer than anticipated) is eliminated through pooling since these public pension schemes typically encompass tens of thousands of employees.

  22. 22

    The Federal-PSS scheme, for example, provides the option to switch starting only in July 2008 (Commonwealth Superannuation Corporation, 2008). Due to data constraints, however, we are unable to investigate further if scheme mobility contributed to the selection.

  23. 23

    Average hourly wage for male civil servants is about A$23.8 compared to $21.6 for private sector workers; the corresponding figures are $21.2 and $17.8, respectively, for females (Cai and Liu, 2011). Some international studies also find a wage premium of between 3 and 11 percent for public sector employees relative to the private sector (Borland and Gregory, 1999).

  24. 24

    Early retiree pensioners retire before the normal retirement age of 55–65. In the Mercer Pensioner database, some early retiree pensioners commence their pensions as early as age 24. Workers who retire due to invalidity also tend to commence their pensions earlier than age 55 (on average around age 49).

  25. 25

    To ensure validity, we compare the smoothed mortality rates to the estimated Cox baseline hazard built from estimates of baseline hazard contributions (and smoothed via a kernel-smoothing function). We find the two sets of sex-specific hazard rates to be relatively close at ages of 60, 70, 80, 90, and 95. Details are available upon request; the percentage differences between the rates at the specified ages are less than 10 percent on average (range from 2 to 16 percent). This is not surprising since the CSS scheme accounts for more than two-fifths of the total sample. Thus, the mortality experience of CSS pensioners contributes heavily to the aggregate mortality experience.

  26. 26

    Using a smaller sample of male Federal-CSS pensioners observed over 1991–1994, Knox and Tomlin (1997) find a strong inverse relationship between preretirement final salary and postretirement mortality.

  27. 27

    Under the null hypothesis of proportional hazards, the scaled Schoenfeld residuals should have the sample path of a random walk and the slope in a regression of the scaled residuals on functions of time should be zero. A nonzero slope is an indication of a violation of the proportional hazard assumption. Graphically, the estimated survival curves for a particular scheme will be nonparallel to that of the reference Federal-CSS scheme (curves may intersect).

  28. 28

    For the male sample, 7 of the 13 schemes fall under “grouped schemes.” For the female sample, three schemes that are predominantly male are dropped (16 subjects dropped). Two of the remaining 10 schemes fall under “grouped schemes.” STATA 12.0 is used for estimation.

  29. 29

    In most Australian DB plans, pension size is a proxy for preretirement final salary. For example, total pension benefit in the Federal-PSS scheme is calculated by multiplying the final average salary (defined as the average superannuation salary on the three birthdays before leaving the scheme) by a benefit multiple, subject to a maximum benefit limit. In comparison, Bikker et al. (2012) report that most Dutch pension plans have switched from using final pay to average pay for computing benefit entitlements as a means to improve their solvency risk management.

  30. 30

    Between 1995 and 2006, the average annualized nominal rate of return on Australian government 10-year bonds is about 6.5 percent and average inflation (measured by the CPI) is about 2.6 percent; the average margin approximates 3.8 percent. This discount rate is also consistent with that chosen by the fund actuary (Mercer Australia) in its actuarial valuation of pension liabilities for the closed Federal-PSS and CSS schemes (Australian Government, 2006, 2009).

  31. 31

    In relation, previous research (e.g., Booth, 2002) documents that the use of the original Lee–Carter method with Australian data results in poor fit. This is largely due to significant departures from the model assumptions, which include linearity in the time component and a constant age pattern of mortality decline.

  32. 32

    Given an entry age of 60, the most notable increases in survival probabilities occur between ages 75 and 90 (about 15–16 percent) with smaller increases at the tails, thus creating the hump-shape. Likewise for an entry age of 80, the most notable increases in survival probabilities occur between ages 92 and 97 (about 6 percent). This process of incorporating future improvements in life expectancies into period mortality rates is known as “cohortization,” and it effectively shifts the hazard curve downward in a nonuniform manner.

  33. 33

    The E.U. Solvency II proposal recommends using a 25 percent shock scenario but several studies (e.g., Stevens, 2010) find that a 5–10 percent shock scenario is more plausible when assessed against historical data. Accordingly, we assume a 10 percent shock in the mortality probabilities on top of expected mortality improvements.

  34. 34

    Antolin (2007) compares the APVs using period mortality (as at 2005) and cohort mortality (which assumes that life expectancy increases at the rate of 1.2 years per decade as forecast by the OECD). Dushi, Friedberg, and Webb (2010) do not directly report the resultant understatement in liabilities for frozen plans but their results for projected benefit obligations of continuing plans indicate that the understatement is about 4.8–9.1 percent (depending on cohort table used for comparison) for a male pensioner of entry age 60.

  35. 35

    It is plausible that a longevity shock may not affect all ages uniformly. For example, a medical breakthrough could be most impactful in increasing survival possibilities for younger-old ages. In an alternative scenario (results not reported here in detail), we assume that the mortality shock is 10 percent up to age 90, and 0 percent thereafter. We find that the percentage changes in APV are 1.8 percent for the 60-year-old and 4.2 percent for the 80-year-old. The gap works out to be smaller (2.4 percent compared to 3.2 percent in the uniform shock), suggesting that the fiscal impact of this modified longevity shock is now less sensitive to the scheme's age-membership structure.

  36. 36

    In the cross-sectional sample, the median pension is approximately A$27,800 (males) and A$18,900 (females). Also, these per capita APV values represent lower bound values since survivor pensions are excluded in our analyses.

  37. 37

    The NSW-PSS and NSW-SSS started in 1907 and 1919, respectively. Although the NSW-EISS officially started in 1997, it consists of members who transferred over from the NSW-SSS scheme. These are employees of certain designated Energy Industries employers who were formerly under the NSW-SSS.

  38. 38

    This is derived based on our weighted estimates for valuation year 2006. An existing male public sector pensioner on average is expected to receive A$0.367 million in present value terms.

  39. 39

    Using U.S. data, Milevsky and Song (2010) note a positive announcement effect when a publicly traded company discloses that it has (either partially or fully) frozen its existing DB pension plan.

  40. 40

    “Contributors” are existing employees who currently contribute to the scheme, and “preserved members” are past employees who have preserved their benefits upon termination.

  41. 41

    Separately, a small handful of pensioners (0.5 percent) terminated their pensions for other reasons.