Randomization Inferences about the Local Causal Effects of Retirement on Human Capital

We estimate the local causal effect of retirement on a range of measures of human capital with a new randomization inference framework for panel data regression and kink discontinuity designs. We find that retirement does not significantly affect any of the dimensions considered. We contextualise these results with a dynamic programming model of retirement. The model emphasises that effective retirement policies must be built on an understanding of the behavioural implications of retirement. More importantly, any causal effect of retirement is mediated through the contributions of an individual's time use and own job environment (as opposed to occupation) on human capital. However, these parameters can vary widely across individuals, so that heterogeneity in population-wide samples is likely to be substantial. Longitudinal information about time use and the extent to which individuals' jobs contribute to their human capital is rare, what severely restricts the ability to ascertain the extent to which existing estimates are sensitive to sample composition. As a result, data on time use together with new data on jobs' contributions to human capital are essential to obtain informative estimates of the causal effect of retirement.