1. Institutional setting of the 2017 pension reform in Finland

The Finnish pension system consists of three pillars. The statutory pension system (1^st pillar) comprises the employment-based earnings-related pension and the residence-based national and guarantee pensions. The latter are intended to provide basic income security. Earnings-related pensions represent a defined benefit system where pension levels are determined by the length of work history and past earnings. This is a mandatory pension system that covers all workers and virtually all earnings. Pensions are accrued throughout the individual’s working life, with contributions paid by both employees and employers. Pension insurance is also mandatory for self-employed persons with earnings above a specified threshold for confirmed annual income, up until they start to receive a full old-age pension.Footnote ⁵ Statutory pensions are broad in scope, and there is no ceiling to the amount of pensionable earnings or the pension amount.Footnote ⁶ Contrary to the case in many other countries, second-pillar and third-pillar supplementary pensions have only a minor role in Finland.

The Finnish earnings-related pension system was reformed in 2017. Prior to that reform, there was one universal statutory retirement age (SRA): 63 years. Like the United States and a couple of other countries, Finland has a flexible retirement age, which means that people can start to draw their pension within a certain age range. The SRA (or NRA) is the minimum age at which full pension benefits can be claimed. There are no ERAs in the Finnish pension system.Footnote ⁷ In connection with the 2017 reform, the decision was made to incrementally increase the SRA by three months for each birth cohort. The first cohort affected by the new rules was people born in 1955 (cut-off 1 January), whose SRA went up to 63 years and 3 months. People born in 1956 had an SRA of 63 years and 6 months, and so on. The retirement age increases by three months until it reaches 65 for the cohort born in 1962. For the cohort born in 1965, the old-age retirement age will be linked to life expectancy. For subsequent cohorts, the rise in retirement age is limited to a maximum of two months per cohort. Key aspects of the Finnish pension system before and after the 2017 reform are described in Table 1.

Table 1. Changes in pension rules in the 2017 pension reform

Individuals can claim an old-age pension starting from the month following the month they reach their retirement age. So, for a person whose retirement age is 63 years and 3 months, payment of the old-age pension can begin from the beginning of the month they turn 63 years and 4 months. However, people can choose to defer claiming their old-age pension, with no age limits. Employment contracts are automatically terminated at the end of the month in which the employee reaches the upper age limit for insurance obligation (68 to 70 years, depending on the cohort). The employer and employee can, however, agree to extend the employment contract beyond that age limit. Regardless of the claiming age, all employees are required to terminate their employment contract prior to drawing an old-age pension.

The amount of earnings-related pension accrued is calculated annually based on gross annual earnings from employment and self-employment. Before 2017, pension was accrued at the following rates: 1.5% at ages 18–52 years, 1.9 % at ages 53–62 years, and 4.5% at ages 63–68 years.Footnote ⁸ Since 2017, a single pension accrual rate of 1.5% of annual earnings has been applied across all age bands. The upper age limit for insurance obligation (which may be thought of as a mandatory retirement age) is set to rise from 68 to 70 years, meaning that new pension rights can be accrued from employment until this age. Those who do not claim their pension at the earliest retirement age will receive a 4.8% increment to their pension for every subsequent year (0.4% for each month), without any upper age limit. Periods of social security benefits for unemployment or sickness, for example, as well as periods of family leave, also count toward the pension. An individual’s accrued pension hence reflects both their earnings and their career length.Footnote ⁹

Some public sector employees were exempt from the scope of the universal reform in that they have a fixed occupational or personal retirement age that differs from the SRA. Occupational and personal retirement ages date back to the mid-1990s, when the SRA in Finland was 65 years.Footnote ¹⁰ Nurses, teachers, and police officers, for instance, have occupation-specific retirement ages that are below the SRA. Individuals with an occupational retirement age are entitled to a full pension upon reaching that age. Personal retirement ages are between 63 and 65 years (in other words, typically higher than the SRA). Individuals with a personal retirement age are entitled to additional pension that was accrued in the 1990s. Employees with a personal retirement age higher than the SRA may retire upon reaching the SRA, but if they decide to do so, their additional pension will be reduced, creating a significant financial incentive to continue working until the personal retirement age. Personal retirement ages have not been changed in connection with pension system reforms. They are still quite common: for example, 50% of public sector employees born in 1954 had a personal retirement age, which was 63 years and 9 months or above (Nivalainen, Reference Nivalainen, Nivalainen, Ilmakunnas and Laaksonen2023). Personal retirement ages only apply to people born before 1960. There are also a small number of individuals in the private sector who have an SRA of 63 years, regardless of their cohort. These ages are based on pre-2017 collective labor agreements and are quite rare.

As for early retirement options, full-time retirement before the SRA is possible only via the disability pension (with some exceptions for public sector employees; see above). The highest age of eligibility is tied to the retirement age, which means that a worker can transition to a disability pension up until they reach the SRA. Eligibility for (full) disability pension requires that the individual’s working capacity is permanently reduced by at least 60%. Another option is retirement on a partial disability pension if the individual’s working capacity is reduced by at least 40%. For older workers over age 60 years, more lenient criteria are applied in that their working capacity is assessed against their current job or occupation. In the case of younger persons, eligibility for disability pension is evaluated against any possible job. Most disability pensions awarded to persons over age 60 are full disability pensions (66% in 2019; Finnish Centre for Pensions’ statistical database). Persons receiving a disability pension are allowed to work (for the fraction of ‘remaining working capacity’), but this is rare (as a 60% reduction in working capacity is required). A disability pension is automatically converted into an old-age pension at the SRA. However, disability pensions where the onset of disability/sickness dates back to before 2017 are converted into an old-age pension at age 63 years (the pre-reform SRA).

For the long-term unemployed, the so-called unemployment pathway to retirement is available. This is an arrangement whereby earnings-related unemployment benefits are paid for an extended period until the individual is eligible to receive a full old-age pension. For the long-term unemployed born before 1958, a full old-age pension was available at age 62 years. However, the long-term unemployed can also postpone retirement and stay on unemployment benefits until age 65 years. For some unemployed people, the earnings-related unemployment benefit can be more than a full old-age pension, creating an obvious incentive to remain on unemployment benefits until age 65 years. Some of the unemployed make use of their right to a full pension at age 62 years, but a larger proportion retire at their cohort’s SRA, and a minority postpone retirement to a later age (see Figure 3). Those who are not eligible for the unemployment pathway to retirement can claim earnings-related unemployment benefits or basic unemployment allowance for a maximum of two years, after which their only option is a labor market subsidy from the Social Insurance Institution of Finland (Kela).

When it comes to sickness benefits, there are no eligibility conditions linked to the pension system. Sickness benefit is the main source of income for individuals who have been on sickness leave for a long period, which, depending on the employer, varies from two weeks to three months. During this period, the employer receives the sickness benefit via Kela, and the employee gets the same wage they would if they were working. After this period, the sickness benefit is paid directly to the worker. The maximum duration of sickness benefit is 12 months, and eligibility for disability pension will typically be considered after one year on sickness benefit.

2. Data and empirical method

2.1. Data

Our analysis of the impact of the SRA increase on labor market behavior uses rich administrative data from Finnish Centre for Pensions registers. These registers cover all individuals who are eligible for social security in Finland, with detailed information on individuals, including gender, date of birth, date of death, nationality, and municipality of residence. The registers also include information on labor market history and pensions, as well as information on educational attainment merged from Statistics Finland sources.

Our study population includes all individuals born between 1954 and 1956, and we further restrict the sample to Finnish citizens residing in the country at the end of the year they turned 59. At that point, we also define whether the individual worked in the private or public sector. To that end, we use the principle of the last insurer: this is the pension insurance company or the provider of public sector pensions who will eventually pay the individual’s pension. The last insurer (and hence the relevant pension act) is recorded in the Finnish Centre for Pensions’ registers, and therefore we are also able to designate the sector for individuals who are unemployed or outside the labor force, as well as differentiate between wage earners and self-employed persons.Footnote ¹¹ Moreover, we use career length and highest educational level (basic, secondary, lower tertiary, higher tertiary) as control variables.Footnote ¹² Likewise, the type of municipality of residence (urban, semi-urban, rural) and gender are controlled for. Our data include the individual’s full career history, which means we also know the amount of earnings-related pension they have accrued. We use this information to approximate the individual’s (post-retirement) income level, which is classified into three groups of equal size: low, medium, and high.Footnote ^13, Footnote ¹⁴ All these variables are measured at the end of the year in which the individual turns 59. In each cohort, there are 700–800 individuals for whom information on sector or municipality of residence is not available.Footnote ¹⁵ In addition, 4,236 individuals who died during the follow-up and before receiving an old-age pension are excluded. The final sample comprises 213,583 individuals. The number of individuals and observations for each cohort is presented in Table 2.

Table 2. Number of individuals in each cohort and the total number of observations in the data

Source: Finnish Centre for Pensions register data.

Our examination of the effect of the SRA increase on labor market outcomes uses monthly information on employment and the receipt of unemployment benefits, sickness benefits, disability pension benefits, and old-age pension benefits. The analyses are performed by age, and individuals are followed from the month they turn 60 years until the month they turn 64 years. The data thus comprise 10,465,567 person-month observations.

The labor market states included in our analysis are ‘retired’, ‘on sickness benefit’, ‘employed’, ‘unemployed’, ‘on disability pension’, and ‘otherwise inactive’. The states are defined in this order and are mutually exclusive. They reflect the situation at the end of each month. The state ‘on sickness benefit’ is second in order because an individual can be on sickness benefit at the same time as they are employed, unemployed, or otherwise inactive. We only consider sickness benefits that are paid directly to the individual. The state ‘employed’ is third in order, which means that those who are employed cannot simultaneously be unemployed or on a disability pension.

Persons born in 1954 are the control group, and those born in 1955–1956 form the treatment group. Table 3 presents the descriptive statistics for the cohorts we consider. The cohorts show rather similar characteristics in terms of career length, gender, income group, sector, self-employment status, and type of municipality. There are small differences with regard to education: a slightly higher share of those born in 1954 have only a basic level of education, while other education levels are somewhat more common among those born in 1955 and 1956. Overall, however, the three cohorts are quite similar.

Table 3. Descriptive statistics by cohort

Source: Finnish Centre for Pensions register data. Except for age and gender, variables are measured at the end of the year in which each cohort turns 59 years.

2.2. Empirical method

Following Staubli and Zweimüller (Reference Staubli and Zweimüller2013) and others, we estimate the effect of the increase in the SRA by comparing labor market states at the same age in cohorts facing a different SRA. All cohorts reach the SRA at a certain point, but due to the rising SRA, one cohort will have reached the SRA by a certain age, while the other cohort has not. We estimate the following differences-in-differences model:

(1)\begin{equation}{y_{iact}} = {\beta _0} + {\lambda _c} + {\theta _a} + {\gamma _t} + {\beta _1}UnderSR{A_{iact}} + X_{\text{i}}^{\prime}{\beta _2} + {\varepsilon _{iact}}\end{equation}

where ${y_{iact}}$ is a labor market state dummy equal to 1 if the individual i, at age a, from cohort c, at time t is in the relevant state. The labor market states under study are ‘retired’, ‘employed’, ‘unemployed’, ‘on disability pension’, ‘on sickness benefit’, and ‘otherwise inactive’.

In this specification, ${\lambda _c}$ are cohort dummies, ${\theta _a}$ are age dummies (age in months) and ${\gamma _t}$ are time effects. $X_{\text{i}}^{}$ includes a set of individual controls (gender, education, sector, length of career, and type of municipality).

The key explanatory variable is a dummy, $UnderSR{A_{iact}}$, which is equal to one if the individual is below the SRA applicable within their cohort, and zero otherwise. Because the increase in the SRA affects individuals at different ages depending on their cohort, the value of this variable changes with age, cohort, and time. For those born in 1954, this variable is equal to one until age 63 years, and zero after that. For those born in 1955, it is equal to one until age 63 years 3 months, and zero after that. For those born in 1956, it is equal to one until age 63 years 6 months, and zero after that.

The main parameter of interest is ${\beta _1}$, which captures the SRA increase-induced difference in the probability of being in a certain state between different cohorts. In practice, we compare the probability of a certain state at the same age for persons belonging to different cohorts with a varying SRA. The central identifying assumption is that, without the increase in the SRA, the probability of being in a certain state at the same age would be similar across cohorts, controlling for the cohort- and age-fixed effects, individual-level variables, and time effects (so-called common trends assumption). If this assumption holds, ${\beta _1}$ can be interpreted as the causal effect of the SRA increase. We estimate this parameter with a linear probability model and, as a result, ${\beta _1}$ can be interpreted as a percentage point difference in the probability that an individual is in a certain state in the cohort that has not yet reached its SRA (treatment group) compared to the cohort that has already reached its SRA (control group).

The 2017 pension reform has affected different cohorts at different ages, and different cohorts reach a certain age at different points in time. Therefore, we also want to control for the time effects (or cyclical effects) to minimize their impact on labor market states. To circumvent the identification issue of perfect linearity between age, period, and cohort, we use different time steps for each (see e.g., Atav et al., Reference Atav, Jongen and Rabaté2019). We measure cohort at a yearly level, age at a monthly level, and control for the business cycle at a quarterly level. This means we assume that the cyclical effects are the same for individuals observed in the same quarter.

We assume that there are no significant differences in the probability of being in a certain labor market state between cohorts before the SRA increase. Following Atav et al. (Reference Atav, Jongen and Rabaté2019), we formally test this in the case of employment by estimating the following interaction model:

(2)\begin{equation}{y_{iact}} = {\alpha _0} + {\lambda _c} + {\theta _a} + {\gamma _t} + {\beta _{c,a}}\left( {{\lambda _c}\,x\,{\theta _a}} \right) + X_{\text{i}}^{\prime}{\alpha _2} + {\varepsilon _{iact}}\end{equation}

where ${\lambda _c}\,x\,{\theta _a}$ is the interaction between cohort and age. The interaction coefficient ${\beta _{c,a}}\,$indicates how the probability of employment differs at a certain age between the treatment and control cohorts. We assume that the interaction coefficient is small and statistically insignificant before the SRA increase and statistically significant at ages where different cohorts face different SRAs. We show that we cannot reject that the common trends assumption holds, meaning there are no significant differences in employment (conditional on the control variables) between cohorts before the SRA of the control cohort.

We validate our differences-in-differences results by implementing an RD design (see e.g., Geyer and Welteke, Reference Geyer and Welteke2021) and estimate the following model:

(3)\begin{equation}{y_{ija}} = {\beta _{0ja}} + {\beta _{1ja}}Treate{d_i} + {\gamma _{0ja}}f\left( {{Z_i} - {c_j}} \right) + {\gamma _{1ja}}Treate{d_i}f{\left( {{Z_i} - {c_j}} \right)_{}} + X_{\text{i}}^{\prime}{\beta _2} + {\varepsilon _{ija}}\end{equation}

where ${y_{ija}}$ is a labor market outcome of individual i, given the SRA increase j, at age a. The indicator $Treate{d_i} = 1$ if the individual i was born at or after the cutoff ${c_j}$. We have two cutoffs for the SRA increase: cutoff1 is 1 January 1955, and cutoff2 is 1 January 1956. In the case of cutoff1, individuals born in 1954 form the control group, and individuals born in 1955 form the treatment group. In the case of cutoff2, individuals born in 1955 form the control group, and individuals born in 1956 form the treatment group. In practice, a local average treatment effect is estimated, which means that the effect of treatment is valid for people near the cutoff. We pool the observations and estimate the effect at age: old SRA + 1 month. This means that in the case of cutoff1, we estimate the effect at age 63 years and 1 month, and in the case of cutoff2 at age 63 years and 4 months.

The month of birth ${Z_i}\,$enters the empirical model as difference to the cutoff ${c_j}$. We include a quadratic trend in the running variable $f\left( {{Z_i} - {c_j}} \right)$. Specification (3) allows for different slopes before and after the cutoff. We do not require additional covariates to identify a treatment effect (Lee and Lemieux, Reference Lee and Lemieux2010). However, additional covariates can improve the accuracy of estimates (Calonico et al., Reference Calonico, Cattaneo, Farrell and Titiunik2017). We therefore include a set of individual controls (gender, education, sector, length of career, and type of municipality) in the model.

The validity of the RD design requires that individuals cannot manipulate the treatment assignment variable (month of birth) (Lee and Lemieux, Reference Lee and Lemieux2010). Clearly, it is impossible for individuals or their parents to have manipulated the date of birth in anticipation of a reform that took place some 60 years after birth. Moreover, we are not aware of any discontinuous changes in the incentive to give birth in December 1954/January 1955 and December 1955/January 1956. An analysis of monthly births for those born in 1954–1956 shows that there are no unusual jumps in the number of births between December 1954 and January 1955, or between December 1955 and January 1956 (see Online Appendix 2, Figure A1).

3. Descriptive evidence

Figures 1 and 2 present the shares of retired and employed individuals from the month they turned 60 to the month they turned 64. The SRA for the control cohort, 1954, was 63, and the SRAs for the treatment cohorts, 1955 and 1956, were 63 years 3 months and 63 years 6 months, respectively. This means that individuals born in 1954 could claim their pension in the month after they turned 63, and those born in 1955 and 1956 could claim theirs in the month after they turned 63 years and 3 months and 63 years and 6 months, respectively. The vertical lines in the figures mark the SRAs of the different cohorts. One should keep in mind that the outcomes are measured at the end of the month.

Figure 1. Share of retired individuals in different cohorts by age (%). Vertical lines indicate the SRA of different cohorts. Source: Authors' calculations based on register data of the Finnish Centre for pensions.

Figure 2. Share of employed individuals in different cohorts by age (%). Vertical lines indicate the SRA of different cohorts. Source: Authors’ calculations based on Finnish Centre for Pensions register data.

We observe a clear parallel trend in both retirement and employment before age 63 years. Due to occupational retirement ages in the public sector, some individuals have retired before age 62 years. There is also a small jump in the retirement rate at age 62 years, partly due to the occupational retirement ages in the public sector and partly due to the eligibility of the long-term unemployed to receive a full old-age pension at age 62 years. After age 63 years, we see a considerable increase in retirement and a decrease in employment among individuals born in 1954. For individuals born in 1954, bunching at the SRA is around 22 percentage points (the hazard rate – the drop in the share of employed persons at the SRA over the share of employed persons just before the SRA – is 0.43). It should be noted that the share of retired individuals increases right after age 63 years also among those born in 1955–1956. This is, for the most part, due to disability pensions where the onset of disability was before 2017, as these pensions automatically become old-age pensions at age 63 years (see Figure 3). Furthermore, there is a small drop in employment at age 63 years among those born in 1955–1956. This is mainly due to two factors: a small number of private sector employees have an SRA of 63 years due to collective labor agreements, and those who are working while on a disability pension are defined in the analysis as employed. If the disability pension changes into an old-age pension at age 63, then those working while receiving a disability pension transition from work to old-age pension in the analysis. After age 63, the trends for cohorts facing different retirement ages diverge. The increase in retirement and the drop in employment shift to the right in line with the increase in the SRA, that is, by three months for the cohort of 1955 and again by three months for the cohort of 1956.

Figure 3. Share of unemployed, disabled, sick, and otherwise inactive individuals in different cohorts by age (%). Vertical lines indicate the SRA of different cohorts. Source: Authors’ calculations based on Finnish Centre for pensions register data.

In Figure 3, we observe rather parallel trends in the proportion of individuals in unemployment, disability, sickness, and inactivity before age 63, and a clear drop after this age for those born in 1954. The share of unemployed persons decreases in all cohorts at age 62 years because the long-term unemployed are entitled to a full old-age pension at age 62 years. There are some differences in the share of unemployed persons around age 62 years and thereafter, especially between those born in 1956 and other cohorts, but these differences are no more than 2 percentage points. In the case of disability, there is a considerable drop after age 63 years also for those born in 1955 and 1956, which is explained by pre-reform disability pensions. It should be noted that after age 63 years, the share of individuals on a disability pension reflects those disability pensions where the disability onset occurred in 2017 or later. The period from 2017 to the new SRA is shorter for those born in 1955, and the share receiving a disability pension among this group has increased only modestly. The share among those born in 1956 has increased more. Otherwise, for those born in 1955, the largest drop in these states occurs after age 63 years and 3 months, and for those born in 1956, after age 63 years and 6 months.

The figures provide an initial idea of the effects we want to analyse. They also support the assumption of common trends for different cohorts before the increase in retirement age, which is at the center of our identification strategy. Later, we will formally demonstrate that we cannot reject the common trends assumption.

4. Main results

Table 4 presents linear regression estimates for the impact of the SRA increase on retirement, employment, unemployment, disability, sickness, and inactivity using equation (1). We only show the differences-in-differences coefficients. All estimates are significant at the 0.1% level. We observe that the SRA increase has decreased retirement by 32 percentage points on average for the months between the old and new SRA. Before the reform, 64% were retired. The SRA increase is accompanied by a 19 percentage point rise in employment. Compared to the pre-reform baseline (29%), the relative increase in the employment rate amounts to nearly 70%.

Table 4. Average effect of an increase in SRA on different labor market states, percentage points

*** Pre-reform: those born in 1954 at age 63 years and 1 month. Linear regression includes cohort dummies, age dummies, quarter dummies, and sociodemographic and regional controls (gender, education, sector, career lengths, region), with standard errors clustered by month of birth. Robust standard errors are in parentheses. Coefficients and standard errors are multiplied by 100 and should be interpreted as percentage points. P < 0.001; **P < 0.01; *P < 0.05. Source: Finnish Centre for Pensions register data.

At the same time, the SRA reform has increased unemployment by 6 percentage points on average. This is a large effect, given that the proportion of unemployed individuals before the SRA increase was less than 5%. Likewise, receipt of disability pensions has increased by 2 percentage points. The share of otherwise inactive persons has increased by over 4 percentage points, which means that this share has almost tripled because of the reform. Sickness plays a minor role in that it has increased by only 0.5 percentage points (but again, compared to the pre-reform baseline, the increase is quite large).

Employment accounts for 60% of the decrease in the retirement rate (19.4/32.1), meaning that this share of individuals who delay claiming pension benefits in response to the SRA increase are working. At the same time, 19% of those who would have already retired without an increase in the SRA are unemployed, and 6% are on a disability pension. In total, 15% are on sickness benefits or otherwise inactive due to the reform.

The absolute employment effect (19 pp) is around the same size as the highest effects reported in previous studies in the Netherlands and France. However, the employment effect relative to the reduction in retirement (60%) is much larger than in these studies, suggesting that the SRA increase in Finland has been more effective in terms of employment. In the Netherlands, the employment effect of the SRA increase was 21 percentage points, representing 36% of the decrease in the retirement rate (Rabaté et al., Reference Rabaté, Jongen and Atav2024). The corresponding figures in France were 21 percentage points and 44% (Rabaté and Rochut, Reference Rabaté and Rochut2020).

The SRA reform has hence not only increased employment considerably but also produced large substitution effects from retirement to other labor market states. We can expect that part of these effects is due to passive substitution: when the SRA increases, individuals who would otherwise have retired stay longer in their respective states. Another part of the effects may derive from active substitution, meaning that some employed individuals, for example, may become unemployed before reaching the new retirement age. We will investigate substitution effects in more detail under the section ‘Passive or active substitution?’

The 2017 pension reform has contributed to reducing old-age pension expenditure and increasing tax revenue from income. However, due to substitution effects, expenditure on unemployment benefits, disability pension benefits, and sickness benefits has increased. The same applies to possible income transfers to inactive individuals. Therefore, the overall effect on public finances is not straightforward. We follow Rabaté et al. (Reference Rabaté, Jongen and Atav2024) and estimate equation (1) on monthly gross income (including zeros) from each labor market state.Footnote ¹⁶ The results show that savings in old-age pension expenditure average 546 euros per person for each one-month increase in the retirement age, and income from employment increases by 565 euros per person on average. At the same time, unemployment benefit costs increase by 90 euros, disability pension costs by 28 euros, and sickness benefit costs by 8 euros, totaling 126 euros. Assuming a marginal tax rate of 40% for pensions and benefits, the net savings equal (1–0.40)*(546–126) = 252 euros per person.Footnote ¹⁷ Assuming a marginal tax rate of 40% on the additional income as well, tax revenue from income increases by 0.40*565 = 226 euros. In total, the net fiscal gain is 252 + 226 = 478 euros per person on average for each one-month increase in the retirement age. Given the cohort size of around 71,000 persons, this translates into 34 million euros per cohort for each one-month increase in the retirement age.

4.1. Sensitivity analyses

Next, we present robustness tests for our results. We first test the sensitivity of the employment effects to alternative specifications. The results are presented in Table 5. Baseline is the preferred specification that was used in Table 4. In the first specification, there are no controls and no clustering of standard errors. In the second specification, the standard errors are clustered by the month of birth. The third specification includes quarter dummies to control for the business cycle. The fourth specification presents the employment effect only for the first increase in the SRA, that is, when only those born in 1954 and 1955 are included. The estimated effects of the change in the employment rate are very similar, independent of the specification used.

Table 5. Sensitivity tests for the employment effect of the increase in SRA

*** Baseline is the specification used in Table 4. The specifications 4 and 5 are similar to the baseline. Pseudo reform includes those born in 1953 and 1954 and assumes that the SRA started to increase from the cohort of 1954. Coefficients and standard errors are multiplied by 100 and should be interpreted as percentage points. P < 0.001. Source: Finnish Centre for Pensions register data.

Another way to check the validity of our model is to conduct a placebo test, i.e., to test whether there is an effect when we would not expect to see one. Therefore, we estimate the baseline specification using those born in 1953 and 1954 and assume (incorrectly) that the increase in the SRA started from the cohort of 1954 (specification five). The estimate of this pseudo reform is statistically significant but negative and very small in magnitude.

A central assumption in our analysis was that there are no significant cohort differences in employment before the SRA increase. We formally test this assumption by estimating equation (2) on employment. The coefficients of interaction between cohort and age are presented in Figure 4. The vertical lines indicate the SRA of different cohorts. The development of employment before age 63 years (the SRA of those born in 1954) is very similar across the cohorts; the coefficients of the interaction terms are small and statistically insignificant. After this age, on the other hand, the coefficients are positive and statistically significant. For example, at age 63 years and 1 month, the difference in employment between the 1955 and 1954 cohorts is 0.17 (17 percentage points), and at age 63 years and 4 months, the difference in employment between the 1956 and 1954 cohorts is 0.21 (21 percentage points). Based on this analysis, we cannot reject the assumption that the common trend holds.

Figure 4. Employment in cohorts 1955 and 1956 relative to cohort 1954 by age, coefficients of interaction between cohort and age, with 95% confidence intervals. Vertical lines indicate the SRA of different cohorts. Source: Finnish Centre for Pensions register data.

The interaction coefficients estimated above offer a simple way to calculate the reform-induced change in the average exit age from work (see Atav et al., Reference Atav, Jongen and Rabaté2019; the formulas are given in Online Appendix 3). An intuitive way of thinking about this is that the interaction coefficient gives the proportion used for working for every respective month of increase in the SRA. Calculating the sum of interaction coefficients between the old and new SRA, we see that a three-month increase in the SRA has delayed exit from work by 0.6 months, and a six-month increase by 1.2 months. This magnitude of delay in the exit age is comparable to the findings of Rabaté et al. (Reference Rabaté, Jongen and Atav2024), who observed in the Netherlands that a three-month increase in the SRA leads to a 0.6 month increase in the average exit age.

To further confirm the validity of our results, we replicate the analyses presented in Table 4 using an RD design. First, graphical evidence of the effects is presented in Figure 5. As can be seen, there is a clear discontinuity at the cutoff (SRA increase) in each of the outcomes. Other outcomes remain quite constant after the SRA increase, but the incidence of disability clearly increases the farther away the month of birth is from the cutoff. This is quite natural as those who are born at the end of the year have had more time to transition to disability pension between the beginning of 2017 and their SRA.

Figure 5. Mean labor market outcomes by month of birth from cutoff (SRA increase), RD-design. Source: Finnish Centre for Pensions register data. The figure presents the mean outcome values using monthly bins for different birth cohorts at the old SRA + 1 month. The SRA increases at 0 (cutoff). Quadratic fit and 95% confidence intervals are presented.

The estimation results from Equation (3) are presented in Table 6. The results are very similar to the differences-in-differences results shown in Table 4. All estimates remain significant at the 0.1% level, and the difference between the results is around 1 percentage point at most. The results do not change significantly if a linear trend is used in the running variable instead of a quadratic trend. Likewise, dropping the control variables does not significantly alter the estimates.

Table 6. Average effect of an increase in SRA on different labor market states, percentage points, RD-estimates

*** All linear regressions include sociodemographic and regional controls (gender, education, sector, career length, region) and quadratic trends in the running variable on both sides of the cutoff. Standard errors are clustered by month of birth. Robust standard errors are shown in parentheses. Coefficients and standard errors are multiplied by 100 and should be interpreted as percentage points. P < 0.001; **P < 0.01; *P < 0.05. Source: Finnish Centre for Pensions register data.

5. Passive or active substitution?

As mentioned earlier, the employment and substitution effects shown in Table 4 may result either from the persistence of a certain labor market state (passive substitution) or from a transition between states (active substitution) due to an increase in the SRA. To investigate this issue in more detail, we estimate equation (1), conditioning on the initial labor market state at age 62 years and 6 months. At this age, 54% of the study population were employed, 11% unemployed, 17% on a disability pension, and 1% on sickness benefit. Visual analysis of the development of these states in the control and treatment cohorts after age 62 years and 6 months supports the assumption of common trends before the increase in the SRA. This is particularly evident in employment, which is the most common labor market state (Online Appendix 4). When it comes to unemployment, there are some small differences, but there were differences in this respect in the main sample too (see Figure 3). Linear regression results are presented in Table 7. To study the sensitivity of the choice regarding the timing of the initial state, we estimated the models measuring the initial state also at age 62 years and 9 months. The results are very similar to those reported here.

Table 7. Average effect of an increase in SRA on labor market transitions from employment, unemployment, disability, and sickness, percentage points

*** Pre-reform: those born in 1954 at age 63 years and 1 month. Linear regression, including cohort dummies, age dummies, quarter dummies, and sociodemographic and regional controls (gender, education, sector, career lengths, region), with standard errors clustered by month of birth. Robust standard errors are in parentheses. Coefficients and standard errors are multiplied by 100 and should be interpreted as percentage points. P < 0.001; **P < 0.01; *P < 0.05. Source: Finnish Centre for Pensions register data.

We notice that, relative to the baseline transition for the control cohort, the rise in the SRA has increased employment persistence on average by 35 percentage points. This means that about 88% (34.5/39.4) of those employed individuals who would have retired given the option have continued to work due to the reform. At the same time, there has also been active substitution, as we see an increase in transitions from employment to inactivity (2.1 pp) and unemployment (1.6 pp), and, to a smaller degree, to sickness (0.7 pp) and disability (0.5 pp). The share of originally employed individuals who are outside work has increased in total by 5 percentage points between the old and new SRA. This means that, in total, around 12% (4.9/39.4) of employed individuals who would already have retired without the SRA increase are inactive, unemployed, on sickness benefit, or on a disability pension.

In the case of unemployment, the decrease in the retirement rate induced by the reform has mostly translated into continued unemployment. As a result of the SRA increase, unemployment persistence has grown by 40 percentage points. This indicates that 90% (40.2/44.8) of those unemployed who have delayed claiming pension benefits due to the SRA increase remain unemployed longer because of the reform. A small portion of unemployed people have returned to employment or become inactive. The increase in the SRA has also had some impact on disability pension or sickness enrolment among unemployed individuals.

Disability is an absorbing state in that there are practically no transitions from a disability pension to any other labor market states following the reform. Receipt of sickness benefits, on the other hand, seems to be a less absorbing state; persistence in this state due to the rising SRA shows an increase of less than 6 percentage points, which accounts for 8% (5.4/67.7) of the decrease in the retirement rate. Then again, a considerable part (53%) of those on sickness benefits who would have retired without the reform end up on a disability pension. Due to the SRA increase, 30% of individuals on sickness benefits return to employment instead of retiring.

Since the number of employed individuals is significantly larger than the number of individuals who are unemployed or on sickness benefits, transitions from employment to other labor market states clearly exceed those from unemployment or sickness to employment. This means that the SRA increase results in more negative (out-of-work) than positive (towards-work) active substitution. However, active substitution only plays a minor role, as most of the effects are due to passive substitution: when the SRA increases, individuals mostly remain longer in their pre-SRA labor market state.

6. Heterogeneity of effects by gender and socioeconomic group

Next, we split our data into several subgroups to evaluate whether the reform had heterogenous effects and estimate equation (1) separately for each group: men and women, the low-educated and high-educated, those with low and high income, the private and public sectors, and wage earners and the self-employed.

Table 8 presents the employment rate just before the SRA, bunching at the SRA, and the hazard rate at the SRA in different groups (for the 1954 cohort). There are no large gender differences in the pre-SRA employment rates, even though a slightly larger share of women are employed before the SRA, and men’s retirement hazard rate is somewhat higher than women’s. When considering education, income, and sector, the pre-SRA employment rates are clearly lower for individuals with a basic education, individuals in the low-income group, and those in the private sector. At the same time, the retirement hazard rate is clearly higher in these groups (around 50%) than in the other groups (23–37%), although the difference in the hazard rate between low-income and high-income individuals is smaller than the respective difference in the other groups. It is worth pointing out that, in the low-income group, the pre-SRA employment rate is very low. This stems from a high incidence of disability and a relatively high incidence of unemployment in this group.

Table 8. Employment rate just before the SRA, bunching at the SRA, and the hazard rate by subgroup, cohort 1954

Employment rate is measured just before the SRA. Bunching measures the share retiring at the SRA. Hazard rate represents the share retiring at the SRA relative to the employment rate just before the SRA. Source: Finnish Centre for Pensions register data.

Overall, the proportion of people retiring at the SRA is lower in the groups where the pre-SRA employment rate is higher, i.e., among the high-educated and high-income individuals and in the public sector. It is well-known that highly educated individuals tend to retire later (Riekhoff et al., Reference Riekhoff, Järnefelt and Laaksonen2020; Mäcken et al., Reference Mäcken, Präg, Hess and Ellwardt2022). In Finland, there is a culture of later retirement in the public sector, which is mostly attributable to a long history of personal retirement ages (which are higher than the SRA). The low retirement hazard rate in the public sector can be interpreted as signaling the importance of social norms and peer behavior.Footnote ¹⁸ The self-employed are a distinctive group: they have a very high employment rate before the SRA. However, the proportion retiring at the SRA, and consequently the retirement hazard rate, is the highest observed among the groups under study. This is because self-employed individuals who have taken out their old-age pension are not required to have pension insurance. Therefore, many self-employed people choose to claim their old-age pension at the earliest possible age, even when they continue with their business.

Table 9 presents linear regression estimates of the impact of the SRA increase on retirement, employment, unemployment, disability, sickness, and inactivity in different subgroups. We observe that there are no large gender differences in the effects of the SRA increase. Among men, the rising SRA has increased employment by 21 percentage points on average for the months between the old and the new SRA, and among women, by 18 percentage points. For both genders, there are also large substitution effects towards unemployment and inactivity, followed by disability, and a minor increase in sickness. Relative to the reduction in retirement, the substitution effects to labor market states outside employment are fairly similar: around 60% of both men (60%; 20.6/34.5) and women (61%; 18.3/29.9) who would have retired given the option remained in employment due to the reform. Previous investigations of gender differences are rare, but in the Netherlands, Atav et al. (Reference Atav, Jongen and Rabaté2019) found that the SRA increase had a somewhat larger employment effect for men than for women, while the substitution effects were considerably larger for women. According to the authors, this is because women in the Netherlands are more often out of the labor force in the years before the SRA. As seen in Table 8, this is not the case in Finland: older women are employed before the SRA even more often than men.

Table 9. Average effect of an increase in SRA on different labor market states by gender, education, and income group, percentage points

*** Pre-reform: those born in 1954 at age 63 years, 1 month. Linear regression, including cohort dummies, age dummies, quarter dummies, and controls, with standard errors clustered by month of birth. Robust standard errors are in parentheses. Coefficients and standard errors are multiplied by 100 and should be interpreted as percentage points. P < 0.001; **P < 0.01; *P < 0.05. Source: Finnish Centre for Pensions register data.

The differences in the effects of the SRA increase are noticeable between the low-educated and the high-educated. Among the low-educated (qualification from basic education), employment has increased by 19 percentage points on average due to the SRA increase. This is a large relative increase, as the pre-reform baseline employment rate was 19%. We also notice a significant and greater-than-average rise of 7 percentage points in unemployment and 3 percentage points in disability. Inactivity has increased by 4 percentage points. In total, 56% of the low-educated who delay claiming pension benefits in response to the SRA increase are working.

Among the high-educated (higher tertiary education), the increase in the SRA has raised the employment rate by 15 percentage points, which in both absolute and relative terms is less than among the low-educated. This is partly explained by the higher pre-SRA employment rate and lower hazard rate at the SRA in this group compared to the low-educated. Consequently, a considerable share of high-educated individuals (53%) were working beyond the SRA even before the reform. The substitution effects to other labor market states are smaller than those observed among the low-educated. In this group, employment accounts for 61% of the decrease in retirement. This result of smaller substitution effects is quite in line with a previous finding from Germany, according to which an increase in women’s ERA had no substitution effects among highly educated women but large substitution effects for women with lower education (Geyer et al., Reference Geyer, Haan, Hammerschmid and Peters2020).

Responsiveness to the SRA increase varies between individuals in the lowest and highest income groups.Footnote ¹⁹ In the low-income group, the SRA reform has increased employment by 14 percentage points on average. The relative increase is substantial compared to the pre-reform baseline employment rate (16%). Unemployment has increased by 6 percentage points, inactivity by 5 percentage points, and disability by 2 percentage points.Footnote ²⁰ In total, only 50% of the low-income individuals who would have retired given the option are working due to the SRA increase.

In the high-income group, employment has increased by 20 percentage points on average following the SRA increase. In absolute terms, this is more than in the low-income group. However, since the pre-reform baseline was 38%, the relative increase is less than in the low-income group. The substitution effects to labor market states outside employment are smaller for high-income individuals: altogether, 64% of the high-income individuals who delay claiming pension benefits in response to the SRA increase are working. These findings contradict Geyer and Welteke (Reference Geyer and Welteke2021), who found that the employment effect of the ERA increase in Germany was larger than average among low-income women and clearly exceeded the substitution effects to other labor market states. High-income women were not studied. We found a smaller-than-average employment effect and large substitution effects for low-income individuals. In contrast to the data in Germany, our data includes both men and women, and the reasons for low income likely differ. In Finland, the employment rate for the low-income group is exceptionally low as early as age 60, with a very high incidence of disability and above-average unemployment. Consequently, individuals in this group have more fragmented labor market histories.

There is also heterogeneity in the effects of the SRA reform by sector (Table 10). In the private sector, the reform has increased the employment rate by 22 percentage points on average. The pre-reform baseline was 23%. We observe a significant increase of almost 8 percentage points in unemployment, 5 percentage points in inactivity, and 2 percentage points in disability. In the private sector, 58% of those who would have retired were it not for the SRA increase, were working.

Table 10. Average effect of an increase in the SRA on different labor market states by sector and self-employment status, percentage points

*** Pre-reform: those born in 1954 at age 63 years and 1 month. Linear regression including cohort dummies, age dummies, quarter dummies, and controls, with standard errors clustered by month of birth. Robust standard errors are in parentheses. Coefficients and standard errors are multiplied by 100 and should be interpreted as percentage points. P < 0.001; **P < 0.01; *P < 0.05. Source: Finnish Centre for Pensions register data.

In the public sector, the rising SRA has increased employment by 14 percentage points on average, which is both absolutely and relatively less than in the private sector. The reasons for this are similar to the differences observed between low-educated and high-educated individuals: a high pre-SRA employment rate, a low hazard rate at the SRA, and consequently a higher share (42%) continuing at work beyond the pre-reform SRA even before the reform. Substitution effects to other labor market states are much smaller than in the private sector. In the public sector, 69% of those who would have retired given the option were working due to the reform. To our knowledge, there are no directly comparable results in the literature. However, Atav et al. (Reference Atav, Jongen and Rabaté2019) found that in the Netherlands, the employment effect of an SRA increase among those employed at age 60 years was larger and the substitution effects were smaller in the public sector, both in absolute terms and relative to the negative retirement effect. In contrast, we found a smaller absolute employment effect in Finland’s public sector. This is primarily due to the culture of later retirement in the public sector compared to the private sector.

Among wage earners, the reform has increased employment by 17 percentage points on average. The pre-reform baseline was 30%. In total, 58% of the wage earners who delay claiming pension benefits in response to the SRA increase are working. As was shown above, the self-employed form a distinctive group. The increase in the SRA has raised employment among the self-employed by 36 percentage points, which is considerably higher in both absolute and relative terms (the pre-reform baseline was 26%) than in any other group. It is noteworthy that inactivity among the self-employed has increased by an exceptional 9 percentage points, and sickness has also increased more than average, by 1 percentage point. In total, 67% of the self-employed who would have retired were it not for the SRA increase were working. We are not aware of previous studies on the self-employed and increasing retirement age. However, the high hazard rate we observe for the self-employed contrasts with the results of Rabaté et al. (Reference Rabaté, Jongen and Atav2024), who found a significantly lower hazard rate for the self-employed compared to wage earners, suggesting the importance of demand-side factors. Our results do not support this interpretation. It is important to remember that in Finland, pension insurance is not mandatory for the self-employed after old-age retirement, and therefore many decide to start drawing their pension at the SRA while continuing to work.

7. Conclusion

In this article, we have examined the effect of an increase in the statutory retirement age on multiple labor market outcomes – retirement, employment, unemployment, disability pension, sickness, and inactivity – using the cohort-based changes made to the SRA in connection with the 2017 pension reform in Finland. We had access to total register data on all Finns born in 1954–1956 and on labor market states recorded at a monthly level. For those born in 1954, the SRA was 63 years, and for those born in 1955 and 1956, 63 years and 3 months, and 63 years and 6 months, respectively. We exploited the exogenous variation that the reform induced in the SRA and used a differences-in-differences approach, which allowed us to isolate the causal effects of the reform. We verified our results with a regression discontinuity design.

Our results show that, due to the increase in the SRA, the employment rate among the cohorts affected has increased by 19 percentage points on average between the old and new SRA. Rather than the absolute increase in employment, a more meaningful measure for assessing the effectiveness of the SRA increase is the proportion of the reduction in retirement that has translated into employment. Measured this way, the employment response is quite substantial: the employment effect represents 60% of the decrease in the retirement rate induced by the reform. We also found large substitution effects to unemployment (6 pp), inactivity (4 pp), and disability (2 pp), and smaller effects to sickness (0.5 pp).

Further analysis showed that the increase in employment is mainly attributable to employed individuals continuing to work. The observed substitution effects are mostly caused by passive rather than active substitution: those facing a higher SRA stay longer in their respective labor market states rather than retire. For a considerable portion of the population, the increase in SRA hence means that unemployment or disability persists. This implies that increased employment comes with the social cost of those in a weaker position remaining in that position for an extended period. However, we also found evidence of active program substitution from employment to other labor market states. With the rising SRA, some employed individuals turn to alternative pathways to retirement, either voluntarily or involuntarily. Less frequent transitions from unemployment and sickness toward employment also occurred. Overall, the results indicate that fiscal savings in retirement benefit payments due to postponed retirement and additional tax revenues from increased employment are partly offset by the rising costs of other social security programs. Our calculations show that the total net fiscal gain from a one-month increase in the SRA has been around 34 million euros per cohort.

To identify potential heterogenous socioeconomic effects, we examined the effect of the SRA increase separately by gender, education, and income. We did not find large differences between men and women. To our knowledge, this is a new finding in the literature and most likely reflects the high degree of gender employment equality in Finland. In addition, we found that, for low-educated and low-income individuals, the substitution effects to labor market states outside work were larger than for high-educated and high-income individuals. Importantly, the results indicate that, for the low-educated and low-income groups, the SRA increase is less effective in the sense that the employment effect represents a smaller proportion of the reform-induced decrease in the retirement rate. As far as we know, our finding that the substitution effects to labor market states outside employment are larger for both low-educated and low-income individuals when considering both genders is a new one. Low-educated and low-income individuals use unemployment and disability as alternative pathways to retirement more often. This may lead to rising social and economic inequalities between low-educated and high-educated and low-income and high-income individuals via negative effects on their financial position, their pension accrual, and, consequently, their old-age security, not to mention potential adverse effects on their mental wellbeing (e.g., Barschkett et al., Reference Barschkett, Geyer, Haan and Hammerschmid2022). It is also worth pointing out that, when the SRA increases, individuals who die earlier lose a higher proportion of their lifetime pensions. This effect is particularly pronounced among individuals with a lower socioeconomic status, as they have a shorter life expectancy.

A limitation in this study is that we investigated the short-term effects of the SRA increase, that is, a situation in which the SRA has not yet increased very much. Then again, it seems that the largest effect of the SRA increase manifests itself at the exact point that the SRA increases, so in this sense, even a short-term analysis provides relevant insights into the effects of the 2017 pension reform. Keeping that in mind, it seems that the SRA increase in Finland has produced a larger employment effect than generally reported in the literature. Furthermore, it appears to have been a more effective way to improve employment than in many other countries, as a larger share of the reduction in retirement has translated into employment. However, it should be noted that there are country differences in pre-reform employment rates, retirement hazard rates, and bunching at the SRA. In Finland, the pre-SRA employment rate is fairly high and the retirement hazard rate relatively low. This combination means that bunching at the SRA – which largely defines the absolute employment effect – is fairly high and around the same magnitude as in France or the Netherlands, where high employment effects have been observed (Rabaté et al., Reference Rabaté, Jongen and Atav2024). The underlying mechanisms, however, are different: in these two countries, the pre-SRA employment rates are lower and the retirement hazard rates higher than in Finland.

The reform-induced decrease in the retirement rate in Finland has been attenuated for the cohorts under study. This is because disability pensions, where the onset of disability occurred before 2017, automatically changed into old-age pensions at age 63 years, the pre-reform SRA. This is one reason why the employment effect relative to the reform-induced decrease in retirement has been considerably high. In the future, as a larger share of disability pensions convert into old-age pensions at each cohort’s SRA, the effectiveness of the SRA increase in this respect will likely diminish. This effect will be pronounced in the groups where the incidence of disability is high: among low-educated and low-income individuals.

Even though we did find active substitution between labor market states as a result of the reform, the active substitution effects have not yet been very large. As the SRA continues to rise, it is likely that more individuals will exit work before reaching the SRA for reasons such as poor health or psychological well-being. This means that, in the future, increases to the SRA could lead to smaller employment growth and larger substitution effects than so far. In this situation, the socioeconomic gradient in the effects of rising SRA might become even more pronounced than today. For example, it has been shown that, in the case of rising retirement age, high-educated employees are less often concerned about their physical and mental capabilities to continue working, and they more often take action in order to be able to extend their careers (Oude Mulders et al., Reference Oude Mulders, Henkens and van Dalen2021).

An opposite trajectory of larger employment effects is also possible, as the increase in SRA may have wider a priori effects around the retirement age in the long run, or so-called ‘distance-to-retirement’ effects (Hairault et al., Reference Hairault, Sopraseuth and Langot2010). For example, the higher retirement age could prompt unemployed individuals to invest more effort in finding a new job. It is also possible that employers’ norms and behavior change over time. It has been observed that the views of Finnish employers regarding a suitable exit age from work move in the same direction with the SRA (Järnefelt et al., Reference Järnefelt, Riekhoff, Laaksonen and Liukko2022). Taken together, these may increase the pre-SRA employment rate in the long run. For the reasons given above, the effect of a rising SRA will continue to remain a relevant research topic in the future.