Pascal Lavergne - Toulouse School of Economics

"One-step nonparametric instrumental regression using smoothing splines", together with Jad Beyhum and Elia Lapenta

Abstract

This paper proposes a new estimator for nonparametric instrumental regressions. It relies on a minimum distance approach and smoothing splines. Unlike popular alternative estimation procedures, our approach does not rely on a first-step regression. This way the estimator is not affected by statistical variability from a first step and we avoid smoothness assumptions on the distribution of the endogenous variables conditional on the instruments The estimator is computationally advantageous because it has a closed-form expression and relies on a unique regularization/smoothing parameter that can be easily selected by cross-validation. We rely on the theory of reproducing kernel hilbert spaces to derive the rates of the convergence of the estimator and its first derivative. Simulations confirm the advantage of avoiding performing a first-step regression. We apply our new method to estimate Engel curves for leisure and catering.

Additional information:

  • Speaker: Pascal Lavergne
  • Time: Thursday, 04.05.2023, 16:00 - 17:00
  • Location: Faculty Lounge, Room 0.036 / Online via Zoom
  • Further links:
  • Organizer: Statistics Group
  • Contact: