A1263
Title: Automatic doubly robust forests for heterogeneous IV effects
Authors: Tomoshige Nakamura - Juntendo University (Japan) [presenting]
Abstract: Estimating heterogeneous treatment effects (HTE) with instrumental variables (IVs) presents challenges due to endogeneity and high-dimensional nuisance functions. The Automatic Doubly Robust Forest (DRRF) algorithm, a method that combines automatic Riesz representer-based debiasing with non-parametric, forest-based estimation, is extended. This strategy leverages DRRFs core strengths: automatic construction of the debiasing term, double robustness in estimating each component, and computational efficiency for multiple query points. The final IV-HTE is the ratio of these estimates, offering a flexible approach to causal inference with instrumental variables in complex, high-dimensional settings.
