pipette
ENEnglish

Debiasing Functionals of Conditional Distributions

Juan Carlos Escanciano, Gabriele Gazzei

Preprint

En palabras de los autores

Many economic parameters depend on estimated conditional distributions. We construct locally robust moments for these parameters by representing their first-step derivative and centering the representer conditionally on the covariates. We develop complementary conditional-measure and CDF linearizations, and derive corrections for control functions, counterfactual functionals, conditional rank-rank regression, and transformations of conditional -functionals. Under the CDF linearization, a Riesz-regression criterion estimates the correction automatically; its analysis distinguishes representer estimation from integration against the estimated conditional law. We establish asymptotic oracle equivalence, asymptotic normality, and consistent variance estimation for cross-fitted estimators under continuity and remainder conditions, without requiring functional weak convergence of the first-step estimators. An exact linear representation yields double robustness; for rank-rank regression this property is componentwise. Simulations report reductions in bias and RMSE relative to plug-in estimation, with double cross-fitting improving the finite-sample performance.

Resultado principalEl resumen no menciona limitaciones.

Apareció: viernes, 30 de octubre. arXiv. Preprint, todavía sin revisión por pares.