Robust difference-in-differences analysis when there is a term structure

Abstract

For variables with a term structure, the standard difference-in-differences (DiD) model is predisposed toward misspecification, even under random assignment, because of heterogeneity over the maturity spectrum and imperfect matching between treated and control units. Estimated treatment effects that are false, biased, or hard to interpret become a concern. Neither unit fixed effects nor standard term-structure controls resolve the problem. Solutions that overcome imperfect matching involve estimating the term structure of hypothesized treatment, which is also what is economically interesting (regardless of matching efficiency). These issues are not unique to DiD analysis, but are generic to group-assignment settings.