Seminar Vol. 151

Title: A Machine Learning Approach to Weighted Least Squares Estimation

Speaker: Hongjun Li, Capital University of Economics and Business

Time: May 6th, 2019 15:00–16:15

Venue: Conference Room 106B, Zhonghui Building (IESR, JNU College of Economics)

About the speaker:

Hongjun Li is an Associate Professor at the International School of Economics and Management, Capital University of Economics and Business (China). He received a Ph.D. in Economics from Texas A&M University in 2014. His main research interests are nonparametric econometrics, social network, and high-dimensional data. Hongjun Li has published in such international journals as Journal of Econometrics, Economics Letters, Econometric Reviews, and Empirical Economics.

Abstract:

The traditional feasible weighted least square (WLS) estimation methods for linear regression models with conditional heteroscedasticity are unsatisfactory in certain sense, i.e., misspecification for parametric approach and poor finite-sample performance for nonparametric method. In this paper, we propose a machine learning approach to WLS estimation that can have better finite-sample performance than classical nonparametric approach while avoiding misspecification problem for the conditional variance. Similar to the nonparametric estimation, the new method also takes two steps. In the first step, we estimate the conditional variance using machine learning tools. In the second step, we estimate the linear coefficients using feasible WLS by plug in the unknown variance with its estimate. We show that the new estimator is consistent and asymptotically normally distributed. Also, we demonstrate the finite sample performance of the proposed method using kernel ridge regression.