A1163
Title: On the use of random forests to estimate triangular two-level panel data models with individual fixed effects
Authors: Monika Avila Marquez - University of Bristol (United Kingdom) [presenting]
Abstract: The focus is on investigating the use of random forests to estimate triangular simultaneous equation models for panel data with an additive separable individual specific effect and an additive separable disturbance term. First, a model composed of a linear structural equation is considered with one endogenous variable. The endogenous variable presents a nonlinear relationship with the instrumental variables and the exogenous regressors. The parameter of interest is the structural parameter of the endogenous variable. The identification of this parameter is obtained under the assumption of available exclusion restrictions and using a control function approach. The estimation of the parameter of interest is done using two proposed estimators that are composed of two steps. In the first step, the nonlinear reduced form equation is estimated using random forests and the residuals are obtained. In the second step, the residuals are used as an estimated control function for the endogeneity in the structural equation. Later, the functional form assumption is relaxed in the structural equation and a semiparametric structural equation is considered. In this new setting, random forests are used to estimate the nonparametric component of the structural equation. A Monte Carlo simulation is performed to test the performance of the estimators proposed. It is concluded that the estimators perform well, provided that the nuisance parameter can be accurately learnt in the first stage.
