Retirement age and health trajectories: a multinational machine learning analysis

Abstract

In aging populations, retirement policies have major impacts on health outcomes; While many studies have looked at the health effects of retirement, many focus on short-term health outcomes. Moreover, studies on retirement rarely capture the heterogeneity in existing health trajectories of individuals. In recent years, research has branched out from the study of the event of retirement itself, to the study of how Statutory Retirement Age (SRA) affects the health of individuals. This study uses longitudinal data from the US, Europe, and South Korea to analyse the impact of higher SRA, on health outcomes, measured by physical frailty, mental frailty and non-communicable diseases (NCDs). To address the existing gaps in literature, we incorporate machine learning techniques to capture differences in health types of individuals, and detect structural breaks in long-term health outcomes. We then construct two difference-in-difference models to estimate the impact of SRA, particularly its increase, on the long-term health of individuals. With a more robust approach, this study provides a more well-considered picture of how SRA affects health in the longer term. Our study reveals significant differences in health outcomes impacted by SRA, where higher SRA is associated with increased physical frailty and incidence of NCDs. Our incorporation of machine-learning based clustering reveals heterogeneous effects of a higher SRA on individuals with different health types, with individuals with best health being most significantly affected. This paper presents a novel approach for evaluating retirement policies and their long-term consequences on health outcomes by combining machine learning and standard econometric approaches.