Microsimulations in the Presence of Unobserved Heterogeneity

While econometric models routinely incorporate unobserved heterogeneity, microsimulations that are based on historical data rarely do so. At issue is that an individual’s history of outcomes contains information about that person’s heterogeneity draw, implying that the individual’s value may not be drawn randomly from the population distribution. Failure to account for heterogeneity may bias behavioral response estimates. This paper develops two approaches—the optimal assignment and conditional distribution methods—to incorporate (multivariate) heterogeneity into microsimulation models. We show that simulations with heterogeneity are clearly superior to those without. Thus enhanced, microsimulation models may better capture behavioral responses to policy change.