On Estimating Partial Effects after Retransformation
35 Pages Posted: 14 Apr 2016 Last revised: 15 Apr 2016
Date Written: April 7, 2016
This paper makes several contributions to the literature on the retransformation problem. First, we derive the asymptotic distribution of the exponential sieve estimator, which has been previously unavailable in the literature. Second, we develop an average structural function type of average partial effect (APE), which applies to the Duan-type retransformation problem even though the independence between the underlying error and covariates is violated. Third, we consider the conditional average partial effect (CAPE), which allows for endogenous explanatory variables and is computationally simpler than existing nonparametric estimators currently available. Finally, we present two non-trivial applications of CAPE estimator. The first is to the average treatment effect in a nonlinear model; the second is to establish the estimator of interaction effect of two dummy variables on the level of dependent variable in the log linear model. Both applications provide new and useful tools for empirical research.
Keywords: Average Partial Effect (APE), Average Structural Function (ASF), Conditional Average Partial Effect (CAPE), Exponential Sieve Estimator
JEL Classification: C10, C18, C21
Suggested Citation: Suggested Citation