Abstract
We study the employment effects of a large increase in the early retirement age (ERA) of women. Raising the ERA has the potential to extend contribution periods and to reduce the number of pensioners at the same time. However, workers may not be able to work longer or may choose other social support programs as exit routes from employment. Results suggest that the reform increases employment, unemployment, and inactivity rates of women aged 60 and older. However, this is mainly because women remain longer in their respective labor market status, rather than active substitution from employment into unemployment or inactivity.
I. Introduction
Population aging presents enormous challenges for public pension systems (OECD 2015). Most OECD countries have reformed their pension systems to answer the challenges posed by increasing old-age dependency ratios. A central aim of these reforms is to extend working lives to alleviate the decline of the working age population (OECD 2006). A key policy variable is the early retirement age (ERA), the minimal age at which people can claim pension benefits. An increase in the ERA has the potential to extend contribution periods and reduce the number of pension beneficiaries at the same time. However, workers may not be able to work longer or may choose other social support programs as exit routes from employment. As a consequence, large program substitution effects might undermine the potential positive fiscal effects of the pension reform. Moreover, an increase of the ERA might be considered unfair in public opinion because it mainly affects the opportunity sets of workers with unfavorable labor market characteristics (Staubli and Zweimüller 2013). Additionally, an observed increase in unemployment, for example, could result from a higher number of unemployed persons simply remaining in this status for a longer period or from a higher share of persons actively changing from employment to unemployment. Therefore, it is important to empirically assess whether an increase in the ERA induces active or passive substitution into other government programs, such as unemployment or disability benefits.
We provide novel empirical evidence on this important research question by analyzing the labor market effects of a substantial increase in the ERA for women. Germany is characterized by a particularly steep increase in the old-age dependency ratio and low employment rates of older workers. In order to relieve public finances by increasing the employment rates of older workers, in a 1999 pension reform, Germany abolished an important early retirement program for women born after 1951. The reform effectively raised the ERA for women from age 60 to at least 63. The reform provides a clean quasi-experimental setting as it induced a large one-time shift in the ERA. We exploit the sharp discontinuity in the ERA between cohorts to estimate the causal impact on female employment behavior in a regression discontinuity (RD) framework based on high quality administrative data. Macroeconomic time trends and other factors that lead to differences among cohorts are accounted for by the RD research design, which allows us to quantify the causal effects of the reform on female employment, takeup of disability pensions, unemployment, and inactivity rates. Furthermore, we focus not only on the effects on levels, but also on employment outflows into other social security programs as a response to the reform. Following Oguzoglu, Polidano, and Vu (2016), we distinguish between active program substitution from employment into unemployment, disability pension, or inactivity and passive program substitution, which occurs due to continuance of the former labor market status because an exit into early retirement is no longer attainable. Moreover, we examine whether the behavioral reactions are heterogeneous across different groups.
Based on a linear regression discontinuity design, we find that employment rates of women born in 1952 aged 60 and older increase markedly by 13.5 percentage points due to the reform. Employment rates before age 60 are unaffected by the ERA increase, even though the reform was long anticipated and forward-looking women could have adjusted their labor supply to the negative wealth shock. Other reasons why we expect the 1999 reform to affect employment outcomes before age 60 include a delay of the period that can be bridged by unemployment benefits and bunching around the ERA. Although we find evidence for increased program substitution from employment into unemployment and increased inactivity, the increase is mainly driven by the inability of unemployed and inactive women to retire early after the reform. We find only a small increase in active substitution into unemployment, disability pension, and inactivity.
On the basis of these results, we draw several conclusions: first, the reform seems to be an effective tool to extend employment of employed women. Second, unemployed and inactive women remain longer in unemployment and inactivity. Third, takeup of disability pensions increases by a small amount. Fourth, the results indicate that the reform affects certain groups heterogeneously. We find larger effects on unemployment rates in eastern Germany than in western Germany, which is consistent with the fact that unemployment rates are higher, and early retirement is more prevalent in the eastern part of Germany. The main distributional effects of the reform result from the persistence of labor market statuses: unemployed or inactive women remain in their respective status, while employed women continue being employed.
The paper proceeds as follows: Section II provides an overview of the relevant literature. Section III briefly outlines the pension system and the 1999 pension reform, and Section IV describes the administrative data that are used in our analysis. Section V describes our empirical strategy, and Section VI presents the results of the empirical analysis, including a discussion of the heterogeneity of the results across subgroups. Finally, Section VII discusses the validity and limitations of our results, and Section VIII concludes.
II. Previous Literature
We contribute to a large empirical literature studying the effects of retirement incentives on labor market behavior (for overviews, see Lumsdaine and Mitchell 1999; Blundell, French, and Tetlow 2016). The literature generally shows that the design of retirement benefits can have substantial effects on the labor market behavior of older workers (for example, Krueger and Pischke 1992; Berkel and Börsch-Supan 2004; Chan and Stevens 2004; Gruber and Wise 2004; Mastrobuoni 2009; Behaghel and Blau 2012; Manoli and Weber 2016a). Most of the earlier studies analyzing labor supply reactions to changes in the ERA or the normal retirement age (NRA)1 rely on out-of-sample simulations (for example, Fields and Mitchell 1984b; Gustman and Steinmeier 1985; Rust and Phelan, 1997; Panis et al. 2002; Berkel and Börsch-Supan 2004; French 2005; Blundell and Emmerson 2007; Haan and Prowse 2014). Those studies typically find some effects on retirement ages. Blundell, French, and Tetlow (2016) note that ex post evaluations generally find larger effects of changes in ERA and NRA—possibly because those studies pick up more relevant factors that are not part of standard retirement models. This corresponds to the fact that the concentration of job exits around specific statutory age thresholds cannot be explained by financial incentives alone (Lumsdaine, Stock, and Wise 1996; Gruber and Wise 2004; Seibold 2017).
The literature discusses different channels through which a change in the ERA can influence the labor market behavior at retirement. Theoretical models of retirement decisions focus mainly on different approaches to specify the financial incentive to retire. Early studies model retirement decisions simply as a function of social security wealth and pension levels (see Fields and Mitchell 1984a, for a review). Subsequent literature took the dynamic structure into account by estimating structural life-cycle models of retirement decisions (Gustman and Steinmeier 1986; Rust and Phelan 1997). Stock and Wise (1990) show that it is not simply the level of retirement wealth or the accrual with one additional year of work that matters, but the entire evolution of future wealth with future work. Because by working, the worker is also buying the option to attain additional income that results from work in future years. The option value, developed by Stock and Wise (1990) and adapted, for example, by Börsch-Supan (2000) and Berkel and Börsch-Supan (2004) in the German context, consolidates the dynamic optimization problem by focusing on the value of retaining the option of postponing retirement.
Standard lifetime budget constraint models predict bunching at the ERA; Manoli and Weber (2016a) show this for Austria, which is comparable to the German system. In the German context, pension benefits are a function of lifetime income, and deductions for early retirement are not actuarially fair, inducing strong incentives for early retirement at the ERA. An increase of the ERA lowers lifetime income due to a shorter period of pension benefits receipt. This induces bunching at the new ERA and generally leads to increasing labor supply and later retirement of affected cohorts. However, if affected women behave forward-looking, the labor supply response may be observed long before the first cohort reaches the new ERA (Cribb, Emmerson, and Tetlow 2016).
While standard lifetime budget constraint models solely based on financial incentives can explain some bunching at the ERA, they fail to explain large peaks in retirement entries at the ERA. Gustman and Steinmeier (2005) come up with a structural life cycle model that replicates the observed retirement peaks at the ERA by introducing heterogeneous time preferences and liquidity constraints. Individuals with high time preference and liquidity constraints will retire at the earliest possible age at which benefits can be claimed, despite forgone future benefit increases. If people do not have enough financial resources to finance an earlier exit, they may have to continue working until reaching the new ERA.
Overall, theoretical models of retirement decisions have difficulties depicting the complexity of different financial incentives stemming from sophisticated pension systems (Blundell, French, and Tetlow 2016). Furthermore, reduced form evidence strongly suggests that many other mechanisms affect retirement. For example, ERA and NRA may be seen as a focal point for decision-making or constitute a social norm about the “normal” retirement reference age (Behaghel and Blau 2012; Seibold 2017).2 If such norms take time to manifest, we expect the short-run effects to be smaller than the long-run effects.
Our study is closely related to a number of recent empirical studies that take advantage of pension reforms inducing quasi-experimental variation in the statutory retirement age (for example, Krueger and Pischke 1992; Mastrobuoni 2009; Behaghel and Blau 2012; Hanel and Riphahn 2012; Vestad 2013; Engels, Geyer, and Haan 2017). Most studies focus on changes in the NRA, while only a few analyze the effect of changes in the ERA. Studies that analyze a change in the ERA include Staubli and Zweimüller (2013) and Manoli and Weber (2016b) for Austria; Atalay and Barrett (2015) and Oguzoglu, Polidano, and Vu (2016) for Australia; Cribb, Emmerson, and Tetlow (2016) for the UK; and Ardito and d’Errico (2018) for Italy. All studies find a substantial increase in employment rates of affected cohorts. Raising the ERA implies that the choice set of older workers is reduced and that the employment reaction of those who would have chosen to retire depends on the relative attractiveness of the remaining options. There is a large literature analyzing program substitution effects (for example, takeup of unemployment or disability benefits) in the context of pension reforms (for example, Duggan, Singleton, and Song 2007; Karlström, Palme, and Svensson 2008; Li and Maestas 2008; Coe and Haverstick 2010; Staubli 2011; Staubli and Zweimüller 2013; Borghans, Gielen, and Luttmer 2014; Atalay and Barrett 2015; Inderbitzin, Staubli, and Zweimüller 2016). Staubli and Zweimüller (2013) and Atalay and Barrett (2015) find that gradual increases in the ERA led to increased program substitution in Austria and Australia. Manoli and Weber (2016b) and Oguzoglu, Polidano, and Vu (2016) study the same reforms using different methods and find that there is no evidence for active program substitution from employment into unemployment or disability pension programs. Instead, previous findings of increased program substitution are mainly caused by people remaining in their respective labor market status and not by active transitions.
In contrast to previous studies, the reform we analyze is a large one-time change of pension rules. Usually changes of the retirement age are introduced in small steps over a longer time horizon, which generally requires stronger assumptions to separate the reform effect from time trends, other policy reforms, and cohort effects. The one-time policy change provides us with a clean quasi-experimental research design, which facilitates the identification of causal reform effects.
III. Institutional Background
In this section, we provide a brief overview of the German public pension system, the pension for women, and alternative routes to retirement.
A. The German Public Pension System
The public pension system covers nearly 90 percent of the German workforce.3 It provides old-age pensions, disability pensions, and survivors benefits. The system is financed by a pay-as-you-go scheme. The calculation of pension benefits is based on a point system that takes into account the entire earnings history and insurance record of each individual. A year’s contribution at the average earnings of contributors earns one pension point. Moreover, pension points can be acquired during other insurance periods (for example, unemployment, child-rearing, and while providing informal care). Pensions are roughly proportional to an individual’s average lifetime earnings and feature few redistributive properties. Public pensions are the most important source of income of the older population representing about 63 percent of total gross income (BMAS 2016). The remaining 37 percent include income from other mandatory pension schemes (14 percent), occupational pensions (8 percent), private pensions (8 percent), and other income.4
B. The Pension for Women
Depending on the length of the insurance record and other qualifying conditions, the age at which pension benefits can be claimed lies between 60 and 65.5 for the cohorts under study (1951–1952).5 Cohorts born before 1952 could claim the pension for women if they were female and fulfilled the following qualifying conditions: (i) at least 15 years of pension insurance contributions and (ii) at least ten years of pension insurance contributions after the age of 40.
These requirements ensured a minimum labor market attachment of eligible women. Our data show that almost 60 percent of all women born in 1951 are eligible for the old-age pension for women (Online Appendix Table B2). Out of those, about 35 percent retired through the early retirement program for women before their 63rd birthday.
Early retirement is associated with actuarial deductions of 0.3 percent per month before the NRA, which was 65 for the pension for women. That is, retiring at age 60 is associated with permanent pension deductions of 18 percent. The early retirement program for women born after 1951 was abolished in 1999. This reform raises the earliest retirement age for most eligible women to at least 63. At age 63, people with a long insurance record (≥35 years) can retire with deductions. Most women who qualify for the pension for women also qualify for this early retirement option (approximately 90 percent, see Online Appendix Table B2).
Due to the reform, women born in 1952 lose an important option to exit the labor market before age 63. For those who want to exit the labor market earlier, the remaining options are unemployment benefits, disability pensions, or inactivity.6
C. Alternative Routes to Retirement
Some women who would have otherwise claimed old-age pension benefits chose another social support program or withdrew from the labor force without claiming any benefits. In the following, we briefly describe the design of unemployment insurance and disability pensions in Germany, focusing on the potential for interdependencies and program substitution.
Unemployment benefits in Germany replace about 60 percent of previous net earnings. Moreover, people receiving unemployment benefits acquire pension points as if they earned 80 percent of their previous gross earnings. Eligibility and the entitlement period depend on age and previous work history. The maximum entitlement period for unemployment benefits does not change during our observation period. Specifically, the maximum entitlement period for individuals above the age of 57 is 24 months. Generally, there is a strong interdependence between unemployment benefits and pensions for older individuals. As documented in Giesecke and Kind (2013) and Engels, Geyer, and Haan (2017), some older individuals use unemployment benefits as a bridge into retirement. In particular, there is evidence that unemployed individuals exhaust their full entitlement period for unemployment benefits before entering retirement. The design of the institution provides strong incentives for this behavior: unemployment benefits are relatively generous, periods in unemployment increase pension entitlements, and, lastly, search requirements for unemployed persons close to retirement are very low. Therefore, an increase in the ERA is likely to affect the takeup of unemployment benefits in two ways. First, individuals have an increased incentive to postpone entry into unemployment, if unemployment benefits are indeed used as a pathway to retirement. This would lead to a shift in increased unemployment entry from 58 (cohort 1951) to 61 (cohort 1952) years, 24 months before reaching the cohort-specific ERA. Second, unemployment rates among women 60–62 years old may increase due to program substitution in response to the abolishment of the early retirement option because women who want to exit employment between the old and new ERA must take another path to exit the labor market.
The disability pension is the only pathway to retirement before reaching the ERA. Eligibility requires a long-term (at least six months) inability to perform an activity under normal labor market conditions for at least six hours (partial disability pension) or at least three hours (full disability pension) per day. The pension is calculated based on the previous insurance biography and amounts to the pension that would be paid had the individual continued to work until she turned 60. When reaching the statutory retirement age, the disability pension is converted into an old-age pension of at least the same level. In Germany, health-related eligibility criteria for disability pensions are relatively strict, especially since a 2001 reform. On average, about 60 percent of all applications are declined (Aurich-Beerheide, Brussig, and Schwarzkopf 2018). Therefore, using disability pensions as a pathway to a regular old-age pension is difficult and not typically an attractive option. Moreover, since 2001, actuarial deductions also apply to this type of pension. The pension is permanently reduced by 0.3 percent per month if retiring before the NRA. In 2012, the NRA of disability pensions was increased from 63 to 65, with deductions capped at a maximum of 10.8 percent. Virtually all of these pensions are reduced by maximum deductions because most people claim this pension before turning 60 (Deutsche Rentenversicherung 2015, p. 83).
Individuals who are neither eligible for disability pension nor unemployment benefits may choose inactivity, meaning that they exit the labor force without benefit receipt. This is particularly relevant for women, who are often not the primary earner in their households.
D. Expected Reform Effects
Because the reform did not only reduce financial incentives for early retirement but abolished the option altogether, we expect large positive effects on employment rates of women aged 60–62 (up to 63rd birthday). However, not all women are able or willing to work until reaching the new ERA. Therefore, we expect unemployment, inactivity, and disability pensions to increase, which is the response to similar reforms in other countries (see Staubli and Zweimüller 2013; Atalay and Barrett 2015).
There are several reasons why we expect the 1999 reform to affect employment outcomes before the former ERA of 60. First, women born after 1951, who have a preference for early retirement around the age of 60, may choose alternative pathways to exit employment, perhaps even before their 60th birthday, instead of delaying until they reach the ERA. The ERA can serve as a reference age for retirement decisions (Blundell, French, and Tetlow 2016; Seibold 2017), which leads to bunching of retirement entries at the ERA and a reduced number of exits among individuals approaching the reference age. Consequently, an ERA increase may lead to an increase in employment exits of women approaching age 60 (unbunching).
Second, women may bridge the last one or two years prior to reaching the ERA with unemployment. As explained in the previous section, German employers and employees have an incentive to end employment relationships two years before the employee reaches the ERA because unemployment benefits are paid up to 24 months to older workers. An increase in the ERA leads to a shift in the period that can be bridged by unemployment benefits. If women voluntarily or involuntarily delay their (bridge into) retirement, we expect a negative effect on unemployment and disability pension rates for women approaching age 60, in particular among women 58–59 years old. Instead, we would expect women born after 1951 to enter unemployment at 61 or 62 years of age more often.
Third, the ERA increase was announced in 1999, although it only affects women turning 60 in 2012. Consequently, the ERA increase was already known when the first affected cohort was 47 years old. It is not obvious a priori how younger women adjust their labor supply in a response to the increased ERA. The reform can be interpreted as a strong negative wealth shock as it reduces the length of time that women receive pension benefits and thereby the total social security wealth. Standard economic models predict a positive effect on labor supply and the retirement age through the income effect (for example, Lazear 1986). If these women are forward-looking, they might have adjusted labor supply as soon as the legislation was passed. On the other hand, the ERA increase may discourage women from working because, due to the extended contribution period, the same level of pension benefits can be achieved with fewer contributions in younger years.
Fourth, the eligibility criteria of the pension for women are no longer relevant for cohorts born after 1951. However, early retirement at 63 requires the completion of a contribution period of 35 years, including child-rearing periods. It is not clear how the change in eligibility criteria for early retirement affects the labor supply of women between 47 and 59. For example, a woman in the 1951 cohort with 14 years of contributions has a strong incentive to work at least one additional year to qualify for early retirement. The incentive to accumulate 35 contribution years is higher for cohorts born after 1951.7 As documented in Seibold (2017), bunching around these thresholds is present, but it is very small, and we do not expect strong behavioral reactions to the changed incentives of the qualifying conditions.
IV. Data
We use high-quality administrative data from public pension insurance accounts (VSKT 2016: Versicherungskontenstichprobe; FDZ Deutsche Rentenversi-cherung Bund 2016).8 The VSKT is a stratified random sample of all pension insurance accounts of people aged 30–67. If the appropriate sampling weights are used, the VSKT is representative of the German population of public pension insurance accounts. Because the data are process-produced, recall errors due to memory gaps and wrong temporal assignment are avoided, while panel mortality is negligible (Fachinger and Himmel-reicher 2006). Furthermore, individual employment behavior and retirement entry are reported with monthly accuracy. A drawback is that socioeconomic variables are only recorded to the extent that they are relevant for the calculation of pension benefits. Consequently, information on education is missing in about one-half of the cases. Information about occupations is only available for the last occupation at the time of data collection, which may not be representative of entire employment histories. Furthermore, it is not possible to link spouses and other household members within the data.9
For our analysis we use the VSKT of 2016, the latest available wave at the time of analysis. We restrict the sample to women born in 1951 and 1952. Furthermore, we exclude all women who paid contributions to a special miners’ pension scheme (Knappschaftliche Versicherung) for at least one month, which applies to about 10 percent of all women. Another group excluded from our sample (about 7.5 percent of all women) consists of all women receiving an old-age invalidity pension at some point in their life (see Online Appendix A, for more details on this pension type). We argue that the invalidity pension is always superior to the pension for women because it is associated with lower pension deductions. Consequently, the group of women who claims invalidity pension benefits should not be affected by the 1999 reform. After dropping these groups, we are left with 7,023 women.10
The 1999 pension reform increases the ERA only for women who are eligible for the pension for women. Therefore, we focus on women who fulfill the eligibility criteria for the pension for women.11 About 60 percent of the women in our sample fulfill these eligibility criteria. Due to the traditionally stronger labor market attachment of women in eastern Germany, the share of eligible women amounts to more than 80 percent for women there. Our final sample consists of 3,477 women who fulfill the eligibility criteria of the pension for women. About 41 percent of the eligible women in the sample (cohort 1951) retire early through the early pension for women before their 63rd birthday.
The main variables of interest in this analysis are whether or not an individual is employed, unemployed, inactive, or receiving a disability pension at any given age in months.12 A woman counts as employed if she has a job that is subject to social security contributions.13 The best approximation of inactivity is the residual category, which comprises all statuses other than employment, unemployment, or pension receipt. The residual category includes periods of education or training, insured selfemployment, noncommercial care for children or elderly family members, illness, and unknown status (missing value). By far the largest group within the residual category consists of women with missing employment status. These women could, in principle, be working as uninsured self-employed or as civil servants because these statuses are also recorded as missing values. However, we assume that this is unlikely in the sample we select: women older than 58 who qualify for the pension for women. That is, women who paid ten years of social security contributions after their 40th birthday.14
In order to analyze heterogeneous effects by income and health, we use information on average pension points and periods of sick pay to approximate income groups and health status. A woman is defined as belonging to the low income group if she is in the lowest half of the distribution of average pension points over all full contribution periods.15 The low income group is defined by having accumulated on average less than approximately 0.58 annual pension points in western Germany or less than 0.53 in eastern Germany. That is equivalent to 58 percent of average annual earnings or €20,988 for a western German woman in 2016.16 Note that we use individual pension points, which are based on individual earnings. We approximate poor health using periods of sick pay, which are only recorded if the sick leave exceeds six weeks (for the same illness within six months) or entails hospitalization for employed individuals. A person is defined as having a poor health status if she has at least one sick-pay spell between age 50 and 55, which holds true for about 21 percent of the sample of eligible women.17 We use a similar definition of poor health as Staubli and Zweimüller (2013), who define poor health by the number of days spent in sick leave between ages 45 and 49 for women and between ages 50 and 54 for men. Note that our measure of health is nonstandard and is therefore not comparable to other measures used in the literature. However, we do not observe a different measure of health using administrative pension account data.
Detailed characteristics of the cohorts in the sample of analysis (1951 and 1952) and earlier and later cohorts are displayed in Table B2 in Online Appendix B.
V. Empirical Strategy
The empirical identification of the effect of pension eligibility rules on labor supply and retirement behavior is challenging: employment histories and unobserved preferences for work and leisure affect both labor supply in old age and eligibility for early retirement. One way to circumvent this endogeneity problem is to exploit exogenous variation in the pension system over time or cohorts due to policy changes. Our empirical strategy makes use of the 1999 pension reform, which eliminates the option to retire at age 60 for women born in 1952 and thereafter. By making use of this exogenous variation in the ERA, we can estimate the causal reform effects on employment outcomes, regardless of unobserved preferences for work and leisure.
We employ a linear regression discontinuity research design to estimate the causal effect of an increase in the ERA on employment rates, unemployment rates, the fraction of older women receiving a disability pension, and the fraction in the residual category. The RD research design exploits the variation in the ERA by date of birth, relying on the assumption that the outcome variables of interest would be continuous at the cutoff date of birth (January 1, 1952) in the absence of the 1999 reform. We expect employment outcomes of the 1951 and 1952 cohorts to differ in absence of the reform because female labor supply increases over cohorts and because they face different macroeconomic environments when reaching age 60. In contrast to a simple OLS framework or a mean comparison of treatment and control cohorts, the RD research design is valid even if there is variation in outcomes across cohorts due to macroeconomic trends and other continuous changes in covariates over time.
The research design is implemented by the following empirical model:
1
where the indicator Di = 1, if the individual was born after January 1952. The subscript t refers to age in months and ranges from 721 to 756 (from the month after the 60th birthday to the 63rd birthday) in the baseline specification. The month of birth zi enters the empirical model in difference to the reform cutoff c, which is January 1952. In our baseline specifications, we include a linear trend in the running variable,f (zi – c) = zi – c. The specification allows for different slopes before and after the cutoff. All regressions include calendar month fixed effects and dummies for income groups,18 children, and western Germany, summarized in Xit. However, dropping Xit does not change the point estimates considerably (see Online Appendix D.5 for regression results without covariates). Regression discontinuity analyses are naturally prone to model misspecification. A nonlinearity in outcomes may falsely be interpreted as a discontinuity if it is unaccounted for. Therefore, we report linear regression results both with linear and quadratic trends in the running variable (RDD results with quadratic trends are displayed in Online Appendix D.6). Furthermore, we support our analysis by local linear regression estimations and plots.
Employment status data are recorded for each individual at every age in months t. Therefore, we need to specify a time window for the outcome variables of interest. In our baseline specification, we pool all observations from the month after the 60th birthday to the 63rd birthday (age 60–62). In order to account for correlation between observations for the same individual or individuals born in the same month, we cluster standard errors by month of birth. In the baseline specification, we estimate treatment effects for five outcome variables (employment, unemployment, disability pension, old-age pension, and the residual category). However, it may be of interest to estimate a more flexible model that allows for heterogeneous effects for every age in months t. Consequently, we analyze the reform effects for every age in months separately by including age–treatment interactions into the empirical model. The inclusion of age dummies and interactions with the treatment variable Di = 1 allows us to interpret the coefficient of the interaction term as the reform effect on a specific age group (see Section VI.C).
In the second part of the empirical analysis, we focus on active program substitution due to the ERA reform. In more detail, using a subsample of women who were employed on their 58th birthday (and their 60th birthday), we estimate outflows from employment and inflows into unemployment benefits, disability pension, and the residual category (see Section VI.D). If we look at the effects on the shares in different employment categories only, as in Staubli and Zweimüller (2013) and Atalay and Barrett (2015), we cannot distinguish between passive and active program substitution. In contrast, an analysis of employment outflows allows us to answer the question whether women increasingly used alternative social security programs to exit employment in response to the abolishment of the early retirement option.19 We circumvented the dynamic selection problem by conditioning on employment at a fixed age in months. Formally, we estimate the same regression discontinuity model as described in Equation 1; however, the outcome of interest is the probability to exit employment (into unemployment, disability pension, or inactivity) within the following two and five years, conditional on employment for at least six months at the 58th and 60th birthday. Conditioning on employment at a certain age is problematic if it is itself an outcome that is potentially affected by the reform. However, we can show that there is no discontinuity in the employment rate at the sample entry ages of 58 and 60 (see Section VI.D). Consequently, we argue that treatment effects on flow variables can consistently be estimated using the linear RD approach described above.
VI. Results
A. Descriptive Evidence
First, we take a look at the cohorts of 1951 (pre-reform) and 1952 (post-reform) by age group. The distribution of employment statuses by age is displayed in Figure 1 for the 1951 and 1952 cohorts. The employment rates are relatively high because the sample includes only women who fulfill the eligibility criteria for the early pension for women. There are no significant differences between cohorts at ages 58 and 59. It can be seen that a large fraction (29 percent) of women born in 1951 receive an old-age pension when they are 60 and 61 years old. This fraction is shifts to the 62–63 age group if we look at the 1952 cohort, due to the ERA increase. Employment, unemployment, and inactivity (residual category) rates are significantly higher for 60- and 61-year-old women born in 1952 compared to women born in 1951. In particular, the employment rate for this age group amounts to 63 percent in the 1952 cohort, compared to 47 percent in the 1951 cohort. Even for 62- and 63-year-old women, average employment, unemployment, and inactivity rates are significantly higher in the post-reform cohort. There are no statistically significant differences in disability pension rates across cohorts.
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations.
A closer look at the fractions of women in different employment statuses by age reveals that women born in 1951 exhibit a large drop in employment rates when reaching age 60, while this discontinuity is not observed for the 1952 cohort (Figure 2a). Not surprisingly, the fraction of women receiving an old-age pension increases sharply at the ERA for the 1951 cohort (Figure 2b). At age 63, the employment rate of the 1952 cohort drops sharply, as the post-reform ERA is reached by most women. At the same time, the fraction of women receiving an old-age pension increases sharply with the 63rd birthday.20
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations.
Notes: Employment does not include marginal employment.
Figure 3a and 3b show that the fraction of women in marginal employment and the unemployment rate are lower for the 1952 cohort for all ages between 55 and 60. At age 60, however, we observe a drop in marginal employment and unemployment for the 1951 cohort. Presumably because women retire through the old-age pension for women. From age 60 to 62, the 1952 cohort is more likely to be marginally employed and unemployed. From age 63 onwards, marginal employment and unemployment rates of both cohorts are very similar. It can be seen in Figure 3d that the fraction of women in the residual category also drops sharply when reaching age 60, indicating that a large share of women who were previously in the residual category start receiving the pension for women. The fraction of women receiving a disability pension increases continuously with age for both cohorts (Figure 3c). It can also be observed that women born in 1951 are slightly more likely to receive a disability pension (Figure 3c) between 57 and 61 years of age. Note that differences between cohorts and fluctuations over time can be due to, for example, time trends or macroeconomic shocks. However, our empirical identification strategy is not threatened as long as differences between cohorts are continuous over the month of birth (see Section V for a detailed description of our empirical strategy.)
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations.
Notes: The residual category combines all statuses except (marginal) employment, unemployment, old-age pension, or disability pension receipt. Women can combine marginal employment and pension receipt.
While the sharp decrease in the proportion of women born in 1951 in several employment categories suggest an outflow into early retirement, an analysis of employment outflows is needed to gain further insights on employment exit behavior and potential program substitution effects. In particular, we cannot infer from Figure 3 whether the reform leads to increased inflow into unemployment, disability pension, or inactivity from employment.
Employment outflows are displayed in Figure 4a–4d. The employment exit hazard rate is defined as the fraction of women exiting employment at age t, conditional on survival in employment (excluding marginal employment) up to age t, out of all women in the sample who are employed for at least six months when reaching age 58. Unemployment, disability pension, and inactivity entry rates are defined as the probabilities to enter the respective category conditional on having survived until t, and employment (including marginal employment) for at least six months at their 58th birthday. Note that we do not condition on employment between age 58 and the first unemployment, disability pension, or inactivity entry event. We only consider the first exit or entry, therefore reentering the sample is not possible. It can be seen in Figure 4a that the employment exit hazard peaks at the pre-reform ERA of 60 (one month after the 60th birthday) for the 1951 cohort and at age 63 (post-reform ERA) and, to a lesser extent, at age 62 for both cohorts.
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations.
If women in our sample use 24 months of unemployment benefits as a bridge to retirement, we would expect a peak in unemployment entry at age 58 for the 1951 cohort, and at age 61 for the 1952 cohort—or at least higher entry rates in the two years before reaching the ERA. With respect to active program substitution due to the pension reform, we expect increased entry into unemployment and disability for the 1952 cohort at around age 60. However, neither are observed in Figure 4b. Therefore, it would be surprising if we discovered a large shift in unemployment entry or increased program substitution in the regression discontinuity analysis. The entry rates into disability pension and inactivity do not exhibit notable peaks, nor are there observable differences between cohorts (Figure 4c and 4d).
These descriptive results suggest that there is no increased substitution from employment into unemployment, disability pension programs, or inactivity due to the ERA increase. However, the hazard rates displayed here are descriptive only. At each age, the population that survives in employment is selective, based on previous hazard rates. Therefore, one cannot interpret the differences in hazard rates between cohorts in a causal sense. The results of a more rigorous empirical analysis are described in the following section.
B. Baseline Results
The results of the linear regression discontinuity analysis are displayed in Table 1. Figure 5a–5d visualize the results using local linear regression on both sides of the cutoff, a triangular kernel, and a bandwidth of 12 months.
Linear Regression Results, Age 60–62
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations.
Notes: Scatter plots display mean outcome values using monthly bins. Local linear regression plots are based on triangular kernel functions with a bandwidth of 12 and 6 months.
The increase in the ERA has a positive effect of 13.5 percentage points on the employment rate of women 60–62 years old women (see Column 1, Table 1). The coefficients can be interpreted as the average percentage point change in employment rates of all women in this age group due to the pension reform. Compared to the pre-reform mean the relative increase amounts to about 30 percent. The fraction of women receiving an old-age pension mechanically drops by 27.6 percentage points to zero (Column 5). About half of those women, who would have retired if they had the option, continue to work due to the reform. The remaining women split into unemployment and inactivity.
In addition to the effects on employment rates, we estimate the effects of the ERA increase on the unemployment rate (Column 2), the fraction of women receiving a disability pension (Column 3), and the fraction of women in the residual category (Column 4). The unemployment rate and the fraction of women in the residual category increase significantly, by 5.2 and 6.2 percentage points, respectively. The positive effect on the unemployment rate can be due to either a passive increase in the unemployment rate or an active program substitution from employment into unemployment. The zero effect on disability pension participation rates suggests that there is no program substitution into the disability pension program; consequently, the disability pension program is not used as an alternative pathway to enter retirement.
Our results suggest that the linear trend in month of birth does not affect the outcome on either side of the cutoff. Note that the reform effect estimates do not change considerably if we drop linear trends in the month of birth or if we do not include covariates (see Tables D8 and D9 in Online Appendix D.5). Nevertheless, we include linear trends and covariates to account for potential differences between cohorts and increase the econometric validity of the results. Furthermore, including a quadratic function of month of birth also does not alter the result (see Table D10 in Online Appendix D.6).
Regression discontinuity research designs are prone to model misspecification. Therefore, a local linear regression analysis, which relies on fewer functional form assumptions, is presented in Table 2 and Figures 5a–5d. The local linear regression point estimates do not differ significantly from the OLS baseline estimation results, and the graphical evidence suggests that a linear regression model is a plausible choice.
Local Linear Regression Results, Age 60–62
C. Effects Across the Age Profile
In order to shed more light on the effects of an ERA increase on employment outcomes at different ages, we interact the entire age profile (in months) with the right-hand side of our regression equation. Thereby, we allow for heterogeneous treatment effects by age in months. The resulting coefficients for each age from the month of the 58th to the 64th birthday are displayed in Figure 6a–6d.
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations
Notes: The coefficients of the treatment dummy interacted with the age profile are estimated using a linear regression model including age fixed effects, linear trends in month of birth and the interaction with age in months, calendar month fixed effects, income groups, and a dummy for western Germany. Confidence intervals of clustered standard errors are displayed using error bars.
The effect on employment rates of women 60–62 years old is positive and slightly increasing with age. The gradual increase after age 60 is due to pre-reform pension entry past age 60. As expected, we can observe a positive effect on unemployment and inactivity rates from age 60 to 62 due to the elimination of the option to retire early.
As described in Section III.D, we expect a decrease in the unemployment rate of 58- and 59-year-old women and an increase in unemployment rates for 61-year-old women, if women are bridging the last 24 months before retirement entry with unemployment benefits. However, we do not find evidence for bridging behavior (see Figure 6b). Pooled regressions for women aged 58 and 59, displayed in Table C1, Table C2, Figure C1, and Table C3 in Online Appendix C, confirm that there is no significant increase in unemployment rates for this age group. Furthermore, we do not find evidence for cohort differences in employment and inactivity rates before age 60, as shown in Figures 6a and 6d. We conclude that, even though the ERA increase was long anticipated, there was little or no adjustment in labor supply in anticipation of the ERA increase.
The effects on employment, unemployment, and the residual category are insignificant for women after their 63rd birthday, the age at which a large fraction of the 1952 cohort could make use of the early pension for long-term insured individuals. The results of a pooled linear regression analysis on employment outcomes of 63-year-old women are displayed in Table C4 in Online Appendix C. It can be seen in the last column of Table C4 that the ERA increase has a negative effect of 9.5 percentage points on the fraction of 63-year-olds receiving an old-age pension. This effect corresponds roughly to the fraction of women affected by the ERA increase who are not eligible for retirement for long-term insured individuals with age 63. Because not all women can retire at age 63, we find positive (but insignificant) effects on the fraction of 63-year-old women in employment, unemployment and the residual category. Our results suggest that the fraction of women receiving a disability pension does not increase for any age.
D. Employment Outflows and Program Substitution
The results described in the previous sections suggest that the ERA increase leads to increased program substitution into unemployment. Furthermore, we find evidence for increased inactivity of women 60–62 years old as a response to the reform. This could be caused by passive or active substitution from employment into another labor market status. In order to distinguish between different types of program substitution, we estimate the effect of the ERA increase on the probability to exit employment (and enter unemployment, disability pension program, or residual category) in a specific age window, conditional on employment for at least six months at the start of this window. In particular, we condition on employment for at least six months at the 58th birthday and estimate the effects on the probability to exit employment in the following two and five years. Furthermore, we estimate the probability to exit employment between age 60 and 63, conditioning on employment for at least six months at the month before the 60th birthday. For identification of the treatment effect, we have to assume that employment rates at age 58 and 60 are unaffected by the reform. A test of discontinuity in the employment rate at the 58th and a month before the 60th birthday are displayed in Figure 7a and 7b. There is no statistically significant discontinuity in the employment rates at the cutoff in both cases.21
Source: FDZ Deutsche Rentenversicherung Bund (2016), own calculations.
Notes: The scatter plot displays mean outcome values using monthly bins. The local linear regression plot is based on triangular kernel functions with a bandwidth of 12 months.
The results for the employment outflow analysis are displayed in Table 3, where the coefficients in Columns 1, 4, 7, and 10 can be interpreted as the reform effect on the probability to exit/enter the respective category between the 58th and 60th birthday. The coefficients in Columns 2, 5, 8, and 11 correspond to the effects on the probability to exit/enter the respective categories in the five years between the 58th and 63rd birthday. Furthermore, we estimate the probability that women exit employment before their 63rd birthday, conditional on employment for at least six months at their 60th birthday. The results are displayed in Columns 3, 6, 9, and 12.
Effects on Employment Outflows, Conditional on Employment with Age 58 or 60
We find a large negative effect on the probability to exit employment between age 58 and 62, which is entirely driven by exits from age 60 onwards. The probability to exit employment between age 60 and 62 decreases by 23.7 percentage points due to the reform. Furthermore, we find small positive effects on unemployment rates of 58- and 59-year-old women who are employed at their 58th birthday. This could be explained by a small unbunching effect, meaning that women in the 1951 cohort are less likely to enter unemployment or inactivity in the months immediately before reaching the ERA. Women in the 1952 cohort enter unemployment and the residual category with a slightly higher probability between their 58th and 63rd birthday compared to the 1951 cohort. However, the small effects are only statistically significant at the 10 percent level if we look at the aggregated effect over the entire age window and can therefore not be interpreted as solid evidence for active program substitution.
We find a small positive effect on disability entry between age 58 and 62. The effect can be interpreted as a two percentage point increase in the probability of entering the disability pension program between age 58 and 62. The results suggest that the positive effect stems mostly from increased entry of women older than 60.22 While the effect on disability pension entry is large compared to the small baseline entry probability of 3.3 percent, it is not large enough to have a significant effect on the fraction of women receiving a disability pension. We conclude that there is evidence for active program substitution into the disability pension program; however, the effect is very small in an economic sense.
In general, our findings suggest that passive substitution, rather than active substitution, is the main driver of the estimated labor market effects. Interestingly, this finding is consistent with the findings for Australia and Austria. Differences in point estimates are mainly caused by pre-reform employment rates of women, the bunching at the ERA, and institutional differences of the social security system, in particular unemployment and disability benefits. The magnitude of our point estimates is slightly larger than those found by Staubli and Zweimüller (2013), who estimate an increase of female employment by 9 to 11 percentage points. Atalay and Barrett (2015) report an increase of about 12 percentage points. Cribb, Emmerson, and Tetlow (2016) find an increase of 6 percent for the UK, where financial incentives to continue working do not change as much as in the other countries. Staubli and Zweimüller (2013) estimate an increase of the unemployment rate by more than 11 percentage points but negligible increases in disability benefits and inactivity. Atalay and Barrett (2015) estimate an increase of 13 to 23 percentage points in the receipt of disability benefits. This relatively large effect is consistent with the Australian pension system because people can retire on welfare with almost no impact on pension benefits (Oguzoglu, Polidano, and Vu 2016).
E. Effect Heterogeneity by Subgroups
In order to fully understand the impact of the ERA increase, it is necessary to learn more about the group affected by the reform. A comparison of women who retire early through the pension for women with those who retire with 63 or later, displayed in Table B1 in Online Appendix B, provides insights on the characteristics of the group affected by the pension reform. Women who retire early collect fewer pension points on average; they have lower average earnings during the periods in which they were employed and contributed to their pension insurance accounts. The sum of contribution years at age 60 and the sum of contribution months after age 40 are also lower for early retirees. Women who retire late are more likely to be employed and less likely to be unemployed when they reach age 60.
If women who make use of the early retirement option differ from those working longer, we expect the abolishment of the early pension for women to have heterogeneous effects on different subgroups. Therefore, we split our sample into several subsamples to evaluate whether the reform had heterogeneous effects. In particular, we distinguish between eastern and western Germany.23 Furthermore, we distinguish women with low and high income, poor health, and women with and without children.24
The results for the analysis of different subgroups are displayed in Table 4 (and Table C3 in Online Appendix C for women 58–59 years old). Women in eastern Germany are much more likely to be eligible for the woman’s pension. Consequently, we find larger, although not significantly, employment effects for eastern Germany than for western Germany. While the reform effect on unemployment rates of women 60–62 years old is negligible in western Germany, there is a large positive effect of about 15.5 percentage points on the unemployment rate of women in eastern Germany. This is due to larger overall unemployment rates in the East.
Subgroup Analysis—Linear Regression Results, Age 60–62
We expect women to suffer disproportionately by an ERA increase if they have a stronger preference to retire early than the average population. However, we do not find significantly larger effects on employment or unemployment rates for the subgroup of women with low average earnings.
Women with poor health can be expected to have strong preferences for early retirement and inelastic labor supply at high ages. Consequently, we expect these women to move into alternative employment-exit paths more often when the ERA is increased. In particular, we expect larger unemployment rates and an increase in disability pension participation rates. For the subgroup of women with poor health, the effect on the disability pension rate is positive but insignificant. Our results show that the disability pension rates did not increase significantly for any subgroup as a response to the ERA reform.
Overall, we do not find conclusive evidence that the ERA increase affected certain groups heterogeneously, which may be due to the small sample size of subgroups and consequently low power. We do find, however, that unemployment rates of women 60–62 years old women increase more in eastern Germany than western Germany. Descriptive evidence further suggests that women who make use of early retirement programs before age 63 have accumulated less pension wealth than those who retire later. Combined with the finding that women who are inactive or unemployed spend more years without receiving pension benefits and without accumulating more pension income, we conclude that the ERA increase might have undesirable consequences for the distribution of pension wealth.25
VII. Discussion and Limitations
There are several concerns regarding the internal validity of our research design. The RD design is only valid if women cannot manipulate the treatment assignment variable (Lee and Lemieux 2010), which is the month of birth in our research design. It is impossible that women or their parents manipulated the date of birth in anticipation of the policy change, as the reform was introduced long after the cohorts in question were born. Furthermore, we are not aware of any changes in the incentive to give birth in 1951 as opposed to 1952.26
One of the most important assumptions of our analysis is that any discontinuities in the outcome variables at the cutoff are solely due to the 1999 pension reform. In particular, we need to assume that the differences between the cohorts in question are not caused by other policy changes. Two other pension policy changes also became effective for individuals born after January 1, 1952. First, the old-age pension for the unemployed was abolished for all individuals born after 1951 as part of the 1999 pension reform. However, the ERA for this pension was already at 63. Therefore, this change did not affect women at age 60. Second, the ERA of the invalidity pension program was increased from 60 to 63 in monthly steps starting with individuals born in January 1952. We exclude all women who received an invalidity pension because the ERA for the invalidity pension was also changed for the same cohorts as for the pension for women. It can be assumed that women eligible for either pension choose the invalidity pension due to the significantly more generous pension benefits. Nevertheless, excluding all women who received an invalidity pension may induce a selection bias because women born earlier are older at the point of data collection and, therefore, are more likely to receive an invalidity pension. However, inflow rates are so small that we do not expect this to be a problem.
Even in the absence of other reform changes, women born in 1952 may still be different from women born earlier due to time trends in employment outcomes. Employment rates of women have been increasing over the past decades for every age. Including trends in birth dates should resolve this issue in a RD research design, as long as we can assume that women who were born close to the cutoff are not different from each other. This is tested by checking for discontinuities in covariates, using the same regression discontinuity framework. Results from the test for covariate discontinuities are displayed in Table D1 in Online Appendix D.1. We do not find significant discontinuities in covariates that are not inherently influenced by the 1999 reform. Furthermore, we perform a difference-in-discontinuities analysis in order to test whether our results are caused by a turn of the year effect. Reassuringly, the results of the difference-in-discontinuity analysis, displayed in Table D2 and Table D3 in Online Appendix D.2, do not differ significantly from our baseline results.
Several studies emphasize the importance of interdependencies of retirement decisions within couples. For example, Blau (1998), Gustman and Steinmeier (2000), and Michaud and Vermeulen (2011) estimate structural retirement models and show that complementarity in spouses’ leisure is a likely explanation for joint retirement behavior. Other studies use exogenous variation in pension age rules or incentives to identify joint retirement behavior in couples (for example, Atalay, Barrett, and Siminski 2019; Lalive and Parrotta 2017; Selin 2017). While studies document sizable spillover effects of pension reforms on female employment, most studies find insignificant results for men; see, for example, Lalive and Parrotta (2017) or Selin (2017). An interesting question would therefore be whether the 1999 ERA reform for women led to spillover effects within couples. If men are indirectly affected by the ERA increase for women, the overall reform effects would be even larger. Furthermore, women might be indirectly affected by simultaneous pension reforms for men. Unfortunately, we are not able to answer these questions because we cannot link spouses using the administrative pension account data.
Another concern arises due to the selection of the sample by the eligibility criteria of the pension for women. Specifically, women born in 1951 may select into the sample by extending their pension contribution period in order to be eligible for early retirement. In contrast, women born in 1952 do not have the same incentives to fulfill the eligibility criteria. We discuss the problem of sample selectivity in Online Appendix D.3. We argue that the potential bias due to selection is negligible because there is no change in the fraction fulfilling early retirement eligibility criteria due to the reform.
Overall, we are convinced that the internal validity of our results is high. Nevertheless, external validity is questionable, as in most empirical papers based on treatment evaluation methods. We claim that our results are valid for the cohorts and groups included in the analysis, namely women born in 1951 and 1952 who fulfill the eligibility criteria for the women’s pension.27 We focus on eligible women only because this is the group that could potentially be affected by the reform. By design, eligible women have higher labor market attachment and therefore differ from the full sample of all women, as displayed in Table B2 in Online Appendix B for the cohorts born 1949–1954. We repeat the analysis without sample restrictions using all women born in 1951 and 1952 regardless of pension eligibility criteria. The estimated effect can be interpreted as an intention to treat effect. The point estimates are smaller and correspond to the estimated effects of the eligible sample multiplied by the share of eligible women (see Online Appendix D.4).
Generalizations to other populations are inherently difficult for several reasons. First, institutional settings differ across countries. For example, the increase of the ERA in the UK (Cribb, Emmerson, and Tetlow 2016) or Australia (Atalay and Barrett 2015) occur in a noncontributory pension scheme. However, the results are relevant for countries with a similar institutional context, such as Austria (Staubli and Zweimüller 2013; Manoli and Weber 2016b). Second, the way an ERA reform is implemented matters. The one-time increase of three years in the ERA for women in Germany is unprecedented in other countries. Third, the effect of an ERA increase can differ across time, depending on macroeconomic factors and changes in the characteristics of cohorts. Among other trends, women’s labor force participation increased over time. However, the cohorts in our study (1951 and 1952) do not differ systematically from earlier and later cohorts. Table B2 in Online Appendix B shows that there are no large mean differences in covariates across cohorts born between 1949 and 1954.
VIII. Conclusion
This paper provides novel insights into the causal effects of pension reforms on labor market outcomes. We exploit a large exogenous increase in the ERA for women. In more detail, we focus on the 1999 pension reform that increases the ERA by at least three years for women born after December 1951. Previous studies show that labor market exits increase significantly at the pension eligibility age. If women shift their employment exit to the new ERA, it might be an effective tool to increase old-age employment. However, it could imply that some women who are not able to extend their working life are adversely affected by this reform. The estimation is based on high-quality administrative data from the German pension insurance.
The sharp discontinuity in the ERA by cohorts allows us to analyze the behavioral responses using a regression discontinuity design. Our results show that employment rates among women between the old and new ERA increase by 13.5 percentage points, which corresponds to an increase of about 30 percent compared to the prereform employment rate. Employment rates before age 60 remain unaffected by the reform, even though the reform was long anticipated. This is also surprising because previous studies show that earlier cohorts often used unemployment benefits as a bridge to retirement.
Furthermore, we find a positive reform effect on the unemployment and inactivity rates rate of women 60–62 years old, which is mainly caused by passive rather than active program substitution. That is, women who lose the early retirement option remain in their respective labor market status, that is, in unemployment or inactivity, instead of retiring early. In order to distinguish between passive and active program substitution, we analyze the effects on employment outflows, and unemployment and disability pension inflows. We do not find significantly increased unemployment or inactivity or entry among women 60–62 years old. In other words, employed women of the 1952 cohort remain in employment, while unemployed or inactive women do not return to the labor market. We do, however, find a small increase in the probability of women 58–62 years old to enter the disability pension scheme. The change in inflows is not large enough to have a significant effect on the fraction of women receiving a disability pension.
The ERA increase might have undesired distributional effects, as the ability to work long and the remaining life expectancy may depend on socioeconomic status. In particular, workers with poor health and weak labor market position might be negatively affected by fewer retirement options. Consequently, we examine whether the behavioral reactions differ by region, income, health status, and whether or not women have children. We find that women in eastern Germany are more affected by the ERA increase than those in western Germany. In particular, unemployment rates of women 60–62 years old increase more in eastern than in western Germany. The main distributional effects of the reform result from the persistence of labor market statuses. Unemployed or inactive women remain in their respective status. For these women, the time between employment exit and retirement entry is simply extended, and the period of pension benefits receipt shortened while no or little pension wealth is accumulated.
Footnotes
An earlier version of this paper was circulated under the title “Closing Routes to Retirement: How Do People Respond?” The authors thank David Blau, Flavia Coda-Moscarola, Gordon Dahl, Peter Haan, Markus Knell, Rafael Lalive, Maarten Lindeboom, Arthur Seibold, two anonymous reviewers, and numerous seminar and conference participants for insightful comments and suggestions. The authors are grateful for the support of Tatjana Mika, Wolfgang Keck, and their colleagues at the Research Data Centre of the German Pension Insurance. Clara Welteke acknowledges funding from the Research Network of the German Pension Insurance Fund through a doctoral scholarship (Forschungsnetzwerk Alterssicherung–FNA). The views expressed here are solely those of the authors. The usual disclaimer applies. This paper uses confidential data from the Research Data Centre of the German Pension Insurance. The data can only be used on-site by filing a request directly with the Research Data Center (https://www.eservice-drv.de/FdzPortalWeb/).
Supplementary materials are freely available online at: http://uwpress.wisc.edu/journals/journals/jhr-supplementary.html
* Supplementary materials are freely available online at: http://uwpress.wisc.edu/journals/journals/jhr-supplementary.html
↵1. We refer to the NRA as the earliest retirement age at which people can retire without deductions.
↵2. Note that in contrast, Asch, Haider, and Zissimopoulos (2005) find no evidence of excess retirement at statutory retirement ages in the context of U.S. federal civil service workers, which does not support the existence of societal-wide norms regarding retirement.
↵3. Civil servants have a separate tax-financed defined benefit scheme. Most of the self-employed are exempt from compulsory insurance.
↵4. Börsch-Supan and Wilke (2004) provide an extended overview of the German pension system.
↵5. See Online Appendix A for more details.
↵6. Working after retirement is uncommon in Germany. Similar to other countries, pensions are reduced at relatively high rates if people have income from earnings before reaching the NRA. Moreover, standard labor contracts end with retirement entry, and it is generally not allowed to re-employ a former employee after retirement on a temporary basis. The alternative permanent contract, however, would be very risky for the employer.
↵7. There is a large overlap between women who fulfill the criteria of the pension for women and those who fulfill the eligibility criteria for early retirement for the long-term insured. A graphic representation of the distribution of pension contribution years is displayed in Table D1 in the Online Appendix.
↵8. We use the full VSKT, which is roughly four times as large as the scientific-use file SUFVSKT and can only be accessed on-site. A detailed description of the data can be found in Himmelreicher and Stegmann (2008).
↵9. Similar to other studies looking at changes in the ERA and NRA, we focus on labor market behavior. This limitation follows directly from the administrative data set, which does not include data on consumption, wellbeing, or available household income. We are aware of only one study that analyses the effect of an increase in the ERA on poverty and nonfinancial measures after an increase of the ERA for women in the UK (Cribb and Emmerson 2019).
↵10. We account for regional differences in the empirical analysis because the employment behavior of women in eastern and western Germany differs markedly. About 17 percent of the sample collected most of their pension contribution points in eastern Germany. The assignment of a region to individual employment histories is straightforward because very few women in the relevant cohorts earned nonnegligible pension points in both eastern and western Germany.
↵11. We discuss potential bias due to sample selection in the Online Appendix D.3. Furthermore, we include an analysis using a sample of all women regardless of pension eligibility in Online Appendix D.4.
↵12. We define disability pension periods as months with pension receipt before reaching the ERA and by using the pension-type information for current pension spells. If a disability pension is converted into an old-age pension, the months of old-age pension receipt are recoded as disability pension periods.
↵13. We do not include marginal employment in our definition of employment in the baseline analysis. Marginal employment (geringfügige Beschäftigung) is defined as a tax-free employment relationship with earnings under a certain threshold (that is, until 2013 up to €400 per month, since 2013 €450 per month). Marginal employment is not subject to social security contributions.
↵14. Note that inactivity is used here to describe the status of out of the labor force and out of the social security system. Therefore, it includes, for example, housework or care for a family member, and should not be misinterpreted as leisure or idleness.
↵15. We used the pension points earned over employment periods only. Points earned in the East and West are treated equally; however, percentiles are constructed separately for eastern and western German women, using women who fulfill the eligibility criteria of the women’s pension.
↵16. 20,988 = 0.58 * 36,187, where 36,187 was the average gross earnings of all insured individuals in western Germany in 2016.
↵17. The subsample of women with poor health is more likely to be employed because sick-leave spells are only recorded during employment.
↵18. High and low income are defined as the highest and lowest third of the distribution of average pension points collected in employment periods for all women in the sample.
↵19. A drawback of survival data is that if one compares the hazard rates of two groups over time, the group composition changes if one group has a higher exit probability. This is called the dynamic selection problem. Consequently, we cannot estimate treatment effects by comparing the difference in hazard rates between cohorts over age in months.
↵20. There is also a small increase in the fraction receiving an old-age pension at age 62. This is due to a Vertrauensschutzregelung (protection of legitimate expectations), which grants some women access to a pension for long-term insured individuals with age 62, if they already had a partial-retirement agreement with their employer before 2007.
↵21. This is supported by a regression analysis that is not reported here.
↵22. Aggregating disability entries over several years enables us to discover effects that are too small to be statistically significant for the smaller age windows (see Columns 7 and 9 in Table 3).
↵23. A woman is defined as western German if she collected the majority of her pension contribution points in western Germany.
↵24. Low income is defined by the lowest half of the distribution of pension points collected in employment periods for all women in the sample. Poor health is defined as having at least one sick-pay spell of at least six weeks from age 50 to age 55. Note that women who received sick pay are more likely to be employed because sick leaves are only recorded during periods of employment. Due to data limitations, we cannot divide the sample into married and unmarried women, even though this would be another subgroup analysis of interest.
↵25. It would be interesting to estimate the reform effects on pension levels and distributions once retirement entry and pension levels are observed for the cohorts in question. However, this is not yet possible with the VSKT of 2016 because a nonnegligible fraction of women born in 1951 and 1952 are still working in 2016.
↵26. It can be shown that the number of observations is relatively stable across all months of birth.
↵27. The full VSKT is representative for the population of women who have a public pension insurance account.
- Received July 2017.
- Accepted March 2019.
