Wealth Proxies in the PSID: A Wealth Series from 1969

Abstract

The Panel Study of Income Dynamics (PSID) records wealth from 1984. Pfeffer and Killewald (2018) and Fisher and Johnson (2023) predict the earlier waves from one or two of the sixteen characteristics the PSID records from 1969. They compare their wealth proxies against actual wealth on an intergenerational rank-rank slope alone. I build a gradient-boosting (GB) wealth proxy on all sixteen characteristics. Over the 16 waves recording wealth it outperforms both wealth proxies on the wealth rank distribution, intergenerational wealth mobility, wealth inequality and wealth accumulation trajectories, cutting the wealth rank error by 27 to 34% and more than halving the Gini distance. The wider set of input variables, not the estimator, explains the outperformance. The GB-proxy covers every pre-1984 household-year, giving PSID research a 1969 to 2023 wealth series with three biases of known size: it overstates intragenerational wealth persistence, understates wealth inequality and overstates wealth accumulation.