Time:

13:30-15:00, Wednesday, March 12th, 2025

Location:

Mathematics Building, Room 216

Speaker:

Qingfeng Liu

Speaker Bio:

Qingfeng Liu, Professor

Department of Industrial and System Engineering, Hosei University, Japan

PhD(Economics),Kyoto University, 2007

Research Associate, Department of Operations Research and Financial Engineering, Princeton University, *. April 2007-October 2008.

Professor of Economics, Department of Economics, Otaru University of Commerce. Japan, October 2009-2022.

Visiting Scholar, institute of Economic Research, Kyoto University, Japan, April 2015-March 2016.

Associate Professor of Economics, Department of Economics, Otaru University of Commerce.

Japan, April 2009-September 2015.

Visiting Scholar, Department of Statistics, Columbia University, *, October 2018 -August 2019.

Research fields: Econometrics, Statistics, Machine Learning.

He published dozens of papers at Journal of Business & Economic Statistics, Econometric Reviews, Econometrics Journal and other international leading professional journal of measurement and statistics.

Abstract:

This study proposes a tying maximum likelihood estimation(TMLE) method to improve the estimation performance of statistical and econometric models where most time series have long sample periods while others have significantly shorter periods. TMLE achieves this by tying the parameters of the long time series to those of the short ones, facilitating the transfer of valuable information to improve the parameter estimation accuracy for the short series. We establish the asymptotic properties of TMLE and derive its finite-sample risk bound under a fixed tuning parameter that determines the strength of the tying. Further, we propose a bootstrapping-based method for selecting the tuning parameter and provide finite-sample theories to guide the effective execution of this procedure. Extensive simulations and empirical applications demonstrate that TMLE exhibits outstanding performance in terms of both point estimates and forecasts.

JEL Classification: C00, C13, C30, C50

Keywords: Parameter tying, dependent data, small sample, maximum likelihood estimator, bootstrapping 17, 2024