Semiparametric Bayesian Modeling of Income Volatility Heterogeneity

Abstract

Research on income risk typically treats its proxy—income volatility, the expected magnitude of income changes—as if it were unchanged for an individual over time, the same for everyone at a point in time, or both. In reality, income risk evolves over time, and some people face more of it than others. To model heterogeneity and dynamics in (unobserved) income volatility, we develop a novel semiparametric Bayesian stochastic volatility model. Our Markovian hierarchical Dirichlet process (MHDP) prior augments the recently developed hierarchical Dirichlet process (HDP) prior to accommodate the serial dependence of panel data. We document dynamics and substantial heterogeneity in income volatility.