The Role of Health Expectations in Retirement Behavior
by Feiya Shao, University of Michigan
This dissertation aims to contribute to our understanding of subjective expectations and how they affect behavior and economic outcomes. It focuses on developing tools for harnessing subjective information from surveys for behavioral inference. Surveys remains the unique avenue for researchers to access subjective measures, while administrative data provides complementary objective measures.
The first chapter examines the impact of anticipated and unanticipated health changes on retirement timing, age of claiming Social Security payments and also hours worked. I use survey data to construct a novel, empirical measure of individual-level health expectations and estimate a distribution of heterogeneous health expectations in the population. I find that unanticipated changes in health cause an increased likelihood of retirement while anticipated changes do not. Poorer expected health is also correlated with earlier claiming of Social Security. However, both anticipated and unanticipated health declines result in reduced work hours conditional on not retiring entirely. This also provides an empirical test of the extent to which retirement is a planned event.
The second chapter directly estimates the relationship between health and labor supply using survey responses on expected labor supply, conditional on different health states. By leveraging on the counterfactual, we are able to study behavior response to health changes without observing realized shocks. Using a unique dataset from the Vanguard Research Initiative, we uncover the non-linear relationship between expected retirement wealth and health-contingent labor supply.
The third chapter investigates gender differences in retirement trajectories, with focus on the job transitions when nearing retirement. We pioneer a novel method of collecting employment histories by sampling up to two job spells from each respondent. We also field comprehensive batteries on job perception around the time of separation (if any) and any job search activities that did not result in an observed job spell. This enables us to compare not only the realized work histories, but also opportunity sets faced by male versus female workers late-in-life.