Healthier when retiring earlier? Evidence from France

ABSTRACT

This paper contributes to the literature on the health-retirement relationship by looking at the effect of retiring before legal age on health in later life in France. To account for the endogeneity of the early retirement decision, our identification strategy relies on eligibility rules to a long-career early retirement scheme introduced in France in 2004 that substantially increased the proportion of older workers leaving their last job before the legal age of 60 years. We find a positive correlation between early retirement and health problems among male retirees. However, we fail to find any significant causal effect of early retirement on poor health once we account for the endogeneity of the decision to retire before the legal age. Controlling for working conditions does not influence the effect of retirement and occupying a demanding job is harmful to health after retirement regardless of the retirement date. Similar results are found for female retirees.

KEYWORDS:

• Early retirement
• self-assessed health
• working conditions

JEL CLASSIFICATION:

• J14
• J26

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ See http://www.oecd.org/els/emp/Summary_1970%20values.xls for an overview of time-series data concerning the average effective age of retirement in OECD countries.

² According to OECD (2015), the average effective age of retirement for the 2009–2014 period was 59.4 years for men and 59.8 years for women. The difference with the normal retirement age is 1.8 years for men and 1.4 years for women, respectively.

³ With the exception of Heller-Sahlgren (2017), instruments used to account for endogeneity are constructed using both the early and legal retirement ages in all the aforementioned studies. However, we can expect that early and regular retirement have different effects on health.

⁴ However, using data from the Health and Retirement Study, Miah and Wilcox-Gök (2007) show the reverse story for chronically ill men and women: they retire later because of lower asset accumulation.

⁵ According to Bingley and Martinello (2013), the estimator of the retirement effect is severely biased and the magnitude of the endogeneity bias depends on the correlation between schooling and the retirement instruments.

⁶ In Ireland, the negative effect of retirement on mental health is significant for involuntary or forced retirement, but not for voluntary retirement (Mosca and Barrett 2016).

⁷ Gorry, Gorry, and Slavov (2015) find a positive effect of time spent on retirement on mental and physical health (especially several years after the retirement date). They put forward that these benefits are cumulative over the retirement period.

⁸ Using data from the HRS, Falba, Gallo, and Sindelar (2009) show that divergence between the subjective probability of working full-time at age 62 and actual labour participation at that time affects the risk of depression. Assessing the effect of a Dutch pension system reform, De Grip, Lindeboom, and Montizaan (2012) find that depression rates increase by about 40% for the 1950 cohort affected by a reduction in pension rights compared to the 1949 cohort.

⁹ The definition of the reference wage for the general regime has changed over time. Before 1993, it was based on the average earnings of the best 10 years. Since then, it has been gradually raised with an increase of one year for each cohort from generation 1933 to generation 1948. For civil servants, the reference wage corresponds to the average earnings of the 6 best months.

¹⁰ A distinction has to be made between two types of quarters, i.e. quarters of contribution and assimilated quarters. A quarter of contribution to the system is validated if the annual earnings are at least equal to 200 h of minimum wage (1886 euros in 2013). The number of quarters of contribution cannot exceed four in one year. Under some conditions, some missing quarters of contribution may be purchased by individuals to compensate incomplete years or high exit age from the schooling system. An assimilated quarter may be validated even though the individual is not employed, for instance, due to sickness leaves, unemployment schemes, maternity or disability.

¹¹ See https://www.legifrance.gouv.fr/affichTexte.do?cidTexte=JORFTEXT000000781627.

¹² Among the other changes, the number of quarters purchased by workers to compensate incomplete years or high number of schooling years have been excluded since 2009 from the total number of validated quarters required to be entitled to the retirement before 60. Also, the possibilities of overstatement for contributed years have been reduced. Age of the first contribution requires a formal evidence of work since 2009.

¹³ See http://dares.travail-emploi.gouv.fr/IMG/xls/series_internet_caa_2003-2014_sexe_age_2015_23_11.xls.

¹⁴ The PRE data files are available to researchers using the French portal Réseau Quetelet for data in the humanities and social sciences (http://www.reseau-quetelet.cnrs.fr/spip/).

¹⁵ There is no other health indicator (like mental health) in the PRE survey.

¹⁶ We exclude the 1942 birth cohort as the number of individuals born in 1942 (N = 81) is much lower compared to that of other cohorts. This choice has no effect on our results.

¹⁷ In particular, 21 respondents claim having benefited from the RACL scheme (which means that they retire before the legal age of retirement), but do not indicate in the survey that they retire before 60.

¹⁸ In terms of working conditions, RACL recipients report more often having experienced short repetitive tasks (+6.3 points) and physically demanding work (+4.6 points) compared to other early retirees. Conversely, they indicate less often a high pace of work (−5.9 points).

¹⁹ Since we do not have longitudinal data, we are not able to account for unobserved heterogeneity at the individual level. That is why it is important to account for the role of working conditions in our regressions since those covariates may affect both the decision to retire early and the health status.

²⁰ In our empirical analysis, we will investigate whether our results are robust when considering subsamples of individuals more comparable in terms of age or age at labour market entry.

²¹ For respondents who did not benefited from the RACL, only 31.5% received a pension the year they left and 18.3% benefited from generous financial incentives to leave early.

²² We have also estimated Probit models for each health outcome and obtain very similar marginal effects of the various explanatory variables. We have also estimated a trivariate Probit model explaining the three health outcomes jointly. As expected, we find large positive significant coefficients of correlation between residuals of each health equation.

²³ Additionally, we estimated models including industry and region-specific fixed effects. The former corresponds to the industry occupied before retirement with the four following categories: primary sector, secondary sector, construction and tertiary sector. We also included the unemployment rate of males working in the private sector in a defined industry-region cell for the year respondents left their last job. This allows controlling for economic conditions at the time individuals withdrew from the labour market. These additional covariates have no effect on our empirical results.

²⁴ We have also estimated bivariate Probit models to estimate a non-localized treatment effect exploiting the functional form as an additional source of identification. These additional estimates, available upon request, lead to very similar results compared to those obtained with the linear IV regressions.

²⁵ The standard errors are 0.161, 0.166 and 0.144 when considering subjective bad health, having chronic problems and health limitations as dependent variable, respectively. Under exogeneity, the standard errors were equal to 0.027, 0.028 and 0.025.

²⁶ We note that the standard errors remain quite stable, respectively, equal to 0.041 and 0.036 for chronic problems and health limitations.

²⁷ The marginal effects are −7.6 percentage points for poor subjective health, +0.5 points for chronic health problems and +1.2 points for health limitations.

²⁸ However, there is no information on early retirement (as well as on date of retirement) in the ESPS survey.

²⁹ For the third outcome (health limitation), the question is about limitation in activities that people do usually because of any health problem.

³⁰ The ratio is 2.16 for bad health (2931 euros for respondents in bad health against 1360 euros for those in good health), 2.20 for chronic problem (2931 euros for respondents in bad health against 1360 euros for those in good health) and 2.29 for health limitation (2931 euros for respondents in bad health against 1360 euros for those in good health). The average amounts are calculated on a subsample of 566 respondents for whom information on health expenditures is available.

Additional information

Funding

This work was supported by the CEPREMAP;