The Health Gradient and Early Retirement in Germany: Evidence from the German Socio-Economic Panel

Abstract

Rationale: Recent empirical studies and our own research have shown that the health gradient in income increases with age. This raises the question of whether a worker's relative position vis-a-vis the agegroup-specific health gradient can help to explain differential retirement behaviour at the individual level. Identification of a causal relationship between the health gradient and retirement behaviour could provide a new tool to predict future retirement behaviour on the basis of social and economic determinants. It would also establish a new rationale for health and educational policies that target the health gradient since the social costs of early retirement are likely to increase substantially amid population aging. The problem of early retirement can then no longer be addressed by pension policies alone, but must be addressed by investment in health and human capital. Objectives: We aim to understand the relationship between the health gradient - the positive correlation between income and health - and the timing of individual retirement behaviour in Germany, using data from the German Socio-economic Panel (GSOEP). We further aim to develop a method that promises better predictions of the long term impact of policies affecting the health gradient on workers' timing of retirement amid population aging. Methodology: We first estimate agegroup-specific health gradients and find that their slope increases with age. We then estimate a variety of parametric survival models, based on the Weibull distribution for early retirement. We use the results from stratified survival analysis, with strata defined on the basis of workers' position relative to the health gradient, to make out-of-sample predictions of workers' individual time to retirement, conditional on the included covariates. Results: We find that our new measure of workers' position relative to the agegroup-specific health gradient has about the same explanatory power as self-assessed health status and income together. It can therefore serve as a near-perfect substitute for these conventional covariates in survival analyses of individual retirement behaviour. Our parametric survival model specifications perform well and our basic findings are confirmed when compared to semi-parametric Cox survival models with the same set of covariates. We find that the lowest quintile of workers along the health gradient are approximately 30 percent more likely to retire during any given period of time than the second lowest quintile of workers - and almost twice as likely as the average worker among the 60 percent ranked highest. Conclusions: We have demonstrated the explanatory power of the health gradient for the timing of workers' individual retirement in Germany. This will allow us - in future research - to make quantitative predictions about the likely long-term implications of policies affecting the health gradient and workers' mobility on the health gradient during their lifecycle. More generally, our findings underline the importance of imperfect medical technology in reconciling the well-known human capital theory of health demand with the observation of more rapid declines in health among less educated workers.