Dynamic Bunching Estimation with Panel Data

Abstract An increasingly common technique for studying behavioral elasticities uses bunching estimation of the excess mass in a distribution around a price or policy threshold. This paper shows how serial dependence of the choice variable and extensive-margin responses may bias these estimates. It then proposes new bunching designs that take greater advantage of panel data to avoid these biases and estimate new parameters. Standard methods over-reject in simulations using household income data and over-estimate bunching in an application with charities. Designs exploiting panel data provide unbiased bunching estimates, improved heterogeneity analysis, and the ability to estimate extensive-margin responses and long-run effects.