Many countries have undertaken large reforms to their public pension systems over the past two decades, and more seem likely to follow suit in the near future. A common theme of these reforms has been to induce workers to retire later (e.g. Gruber and Wise 1999, OECD 2019, Barr and Diamond 2009) in order to restore fiscal sustainability in light of ageing populations. That is, in addition to reducing the generosity of public pensions generally, reforms in many countries have introduced or strengthened incentives favouring longer working lives. Such incentives have desirable fiscal effects – workers who retire later pay more tax – but their overall welfare effects are still poorly understood.
In a recent paper (Kolsrud et al. 2021), we propose a framework to analyse the welfare effects of pension reforms that incentivise later retirement. The policy question we study essentially concerns the optimal steepness of the pension benefits profile as a function of the retirement age. How rapidly should pension benefits rise for workers who retire later in life? Our analysis enriches the existing literature, which, in contrast, mainly revolves around questions of the overall generosity of pensions, whether they are funded or pay-as-you-go, and the response of retirement timing to pension incentives.
The cost of a steeper profile
A key quantity to evaluate the cost of a steeper benefit profile is the value that individuals retiring at different points in life attach to the pension benefit as a means of smoothing consumption when retired. If people who retire earlier value pension benefits more than those retiring later do, then incentivising later retirements via the pension system entails welfare costs. Specifically, these reforms redistribute resources away from people who value them relatively more (early retirees) and toward those who value the benefits relatively less (late retirees). In our paper, we show that the size of this welfare cost depends on the marginal utility of consumption of workers retiring at different ages. If, all else equal, early retirees have much higher marginal utility of consumption than late retirees, for instance, this suggests that a steeper profile has a large welfare cost.
But do early retirees actually have higher marginal utilities of consumption than late retirees, and if so, just how much higher? To shed light on this question, we use administrative data from Sweden and registry-based measures of household consumption (Kolsrud et al. 2020). Mapping consumption data to marginal utilities requires some assumptions, so we employ a variety of ways of looking at consumption and supplement consumption data with data on wealth, health, and life expectancy.
According to economic theory, one key determinant of the marginal utility of consumption is the level of consumption itself: the higher the latter, the lower the former. So, we start by simply examining how overall consumption post-retirement varies across workers who retire at different ages. Figure 1 shows how consumption at age 68 varies with retirement ages, comparing each retirement age group to those retiring at 65 (which we consider as a benchmark for a normal retirement age). The overall gradient of consumption with respect to the retirement age is steep, with late retirees enjoying over 20% more consumption than premature retirees. This suggests that rewarding later retirement by giving later retirees more generous pensions redistributes from low-consumption to high-consumption households. Notably, while the overall consumption gradient between retirees at ages 55 to 70 is large and positive, we also document a notable non-monotonicity among those retiring at ages 60 to 65. Individuals retiring between 60 and 63 years have similar or higher consumption on average compared to those retiring near the normal retirement age of 65. We confirm the same overall patterns when studying surveyed consumption in other European countries and the US.
Figure 1 Consumption level by retirement age
Notes: The figure reports estimates of our comparison of consumption levels across all retirement ages. Individuals who retire at 65 are the reference category. In this specification, only year and age fixed effects are included.
The role of socioeconomic background
Naturally, the level of consumption might not be the sole determinant of marginal utility. Other factors may matter, including health and life expectancy which directly affect the redistributive weights one wants to assign to different retirees as well. We analyse rich data on the characteristics of retirees at various ages to help us understand how these factors might change our evaluation of pension reforms. Figure 2 illustrates how individuals who retire early/late differ in their observable characteristics, compared to those retiring around the normal retirement age of 65. The first panel considers socio-demographic characteristics and the second panel considers health and life expectancy. The results broadly reinforce our finding that a steeper pension benefit profile redistributes from those with a high value of pension benefits (earlier retirees) to those with a smaller value (later retirees). In particular, later retirees tend to have more education, more productive careers and more financial resources than those retiring very early. We also document, in Figure 2B, that health and life expectancy are significantly worse for those retiring early. Hence, steeper incentives take away resources from people who are not only less resourceful, but also suffer from worse health.
Figure 2 also sheds light on the drivers underlying the non-monotonicity in the consumption gradient. Those retiring between ages 61 and 63, where the non-monotonicity appears in Figure 1, are more likely to be cohabiting and/or female. They have higher household assets and tend to be in households where another member of the household earns significant income. The non-monotone pattern is also reduced when controlling for household composition. Hence, within this range of retirement ages, from 61 to 65, incentivising later retirement is arguably less costly than at other ages.
Figure 2 Heterogeneity and selection into retirement ages
Notes: The figure documents patterns of heterogeneity across retirement age groups. We define premature retirees by those retiring between 56 and 60, early retirees by those retiring between 61 and 63, normal retirees by those retiring at 64-65 and late retirees by those retiring at 66-69. Panel A displays estimates from a multinomial logit prediction model for retiring in one of the four different age groups. We report for each regressor the estimated marginal effects predicted at the mean on the relative probability to select into each of the groups, using normal retirees as the reference category. Panel B explores selection on health and life expectancy. We use two indices for bad health (i.e. standardised principal components extracted from all health outcomes in the HEK and ULF surveys) and two measures of “life expectancy” (dummies for being dead by age 70 or by age 75).
Different trends in consumption
Economic theory provides us with several ways to learn about the marginal utility of consumption from data on consumption. Another approach we take in our paper is to examine how the drop in consumption at retirement varies across workers retiring at different ages. This approach relies on weaker assumptions about the relationship between retirement age and marginal utility than the consumption levels approach. It also isolates the ‘social insurance’ value of pensions against work longevity risk, i.e. the risk involving when an individual will need to retire for e.g. health reasons (Diamond and Mirrlees 1978, 1982, 1986), while disregarding redistributive effects across individuals. The drop in consumption at retirement has been widely studied and debated (e.g. Bernheim et al. 2001, Aguiar and Hurst 2005, Battistin et al. 2009, Stephens and Toohey 2018), but without considering how this drop varies by the retirement age, which is key for our question about the welfare effects of incentivising later retirement.
The main result from this analysis is displayed in Figure 3, where we plot the mean household consumption relative to two years before retirement. We estimate that the consumption drop from two years before retirement to two to five years after retirement is much larger for very early retirees compared to later retirees. This fact is consistent with a substantial difference in the marginal value of pension benefits to the former versus the latter groups. In the paper, we use data on the dynamics of health to show that much of this appears to be explained by health shocks incurred by premature retirees. We also examine how marginal propensities to consume out of wealth shocks differ across retirement age groups (Landais and Spinnewijn 2021). These results confirm our overall finding that incentivising later retirement entails a substantial cost because it takes resources away when the marginal utility of consumption is high and provides more resources when the marginal utility of consumption is low.
Figure 3 Differences in consumption drops
Notes: The figure documents consumption dynamics around retirement. It plots average residualised (with respect to cohort fixed effects, age fixed effects, social insurance contributions at age 55 and household structure) consumption as a function of time to retirement, separately for premature, early, normal and late retirees. The group definition is the same as in Figure 2. The graph scales the residual consumption of each group by its level two years prior to retirement (this level is also reported on the graph). Because of the year and cohort coverage of our consumption and retirement pension data, the earliest we can observe consumption among all premature retirees is three years prior to retirement. And the latest we can observe consumption among all the late retirees is three years after retirement. This explains the differential coverage of the residualised consumption series.
Incentivising later retirement is a common theme of pension reform in many countries. Most discussions of these reforms, in academic work or public debate, focus on the fact that strengthening incentives to retire later can help restore fiscal sustainability of the pensions system. However, we find that incentivising later retirement entails potentially pivotable welfare costs as well by reallocating public pension benefits from relatively needy to relatively well-off individuals. The widely employed framework for thinking about optimal provision social insurance (Bailey 1978, Chetty 2006) highlights that introducing incentives that have desirable fiscal properties can often entail these types of costs. By adopting this framework to pensions and showing that the redistributive costs appear to be substantial, we highlight that prior literature on pension reforms has mainly focused on just one side of a difficult trade-off: between providing generous pensions to those most in need, who tend to retire early, and maintaining fiscal sustainability. However, we also find there are some ranges of retirement ages – between ages 60 and 65 – where incentivising later retirement is likely less costly. Here lies an opportunity to provide stronger incentives to work longer and smooth consumption between households.
The overall framework we provide can be used to study other types of pension reforms as well, including changes to the relation between pension benefits and career length, changes to the minimum pension benefit and/or changes to disability insurance for old-age workers. Future work could examine other types of pension reforms that might provide incentives that promote fiscal sustainability without incurring the large welfare costs of broad incentives for later retirement.
Aguiar, M and E Hurst (2005a), “Consumption versus expenditure”, Journal of Political Economy 113(5): 919–948.
Baily, M N (1978), “Some aspects of optimal unemployment insurance”, Journal of Public Economics 10(3): 379–402.
Barr, N and P Diamond (2009), “Reforming pensions: Principles, analytical errors and policy directions”, International Social Security Review 62(2):5–29.
Battistin, E, A Brugiavini, E Rettore and G Weber (2009), “The retirement consumption puzzle: Evidence from a regression discontinuity approach”, The American Economic Review 99(5): 2209–2226.
Bernheim, B D, J Skinner and S Weinberg (2001), “What accounts for the variation in retirement wealth among US households?”, The American Economic Review 91(4):832–857
Chetty, R (2006), “A general formula for the optimal level of social insurance”, Journal of Public Economics 90(10-11): 1879–1901.
Diamond, P and J Mirrlees (1978), “A model of social insurance with variable retirement”, Journal of Public Economics 10(3):295–336.
Diamond, P and J Mirrlees (1982), “Social insurance with variable retirement and private saving”, MIT working papers 296.
Diamond, P and J Mirrlees (1986), “Payroll-tax financed social security with variable retirement”, Scandinavian Journal of Economics 88:25–30.
Gruber, J and D A Wise (1999), “Introduction to ‘Social Security and Retirement around the World’”, In Gruber, J and D A Wise (editors), Social Security and Retirement around the World, International Social Security, University of Chicago Press.
Kolsrud, J, C Landais and J Spinnewijn (2020), “The value of registry data for consumption analysis: An application to health shocks”, Journal of Public Economics 189: 1040–1088.
Kolsrud, J, C Landais, D Reck and J Spinnewijn (2021), “Retirement Consumption and Pension Design”, CEPR Discussion Paper No. 16420.
Landais, C and J Spinnewijn (2021), “The value of unemployment insurance”, Review of Economic Studies 88(6): 3041–3085.
OECD (2019), Pensions at a Glance 2019: OECD and G20 indicators, OECD Publishing, Paris.
Stephens, M and D Toohey (2018), “Changes in nutrient intake at retirement”, NBER Working Paper No. w24621.