Statistics Seminar（2014-15）

Topic:Estimating a Large System of Seemingly Unrelated Regressions Using Penalized Maximum Likelihood Estimation with Applications on Asset Returns

Speaker:Qingliang Fan, Xiamen University

Time:Thursday, 23 October, 14:00-15:30

Location:Room 217, Guanghua Building 2

Abstract:It is well known that the estimation of unrestricted covariance matrix of the panel data model is difficult because the dimension of the parameters of unconstrained covariance matrix could grow quadratically as the cross sectional dimension increases. In this paper, we proposed a shrinkage-based method, penalized maximum likelihood estimation, to estimate a seemingly unrelated regression model with the number equations comparable to the number of available sample size. To solve the difficulties of estimating the covariance matrix mentioned earlier, we employ a shrinkage method by Bien and Tibshirani (2011) which is permutation invariant. The new shrinkage method can capture most information of the covariance matrix and it can improve the prediction accuracy. In this paper, we show the asymptotic properties of penalized maximum likelihood estimator. It is illustrated in Monte Carlo simulations that the proposed estimator compares favorably in terms of statistical efficiency with other commonly used methods. We also apply our estimation approach to the study of asset price returns in which the number of equations is comparable to the sample size using Chinese market (Shanghai Shenzhen CSI 300 index) data.

About the speaker:Dr. Qingliang Fan received his PhD in Economics from North Carolina State University in 2012. He is currently an assistant professor of eonomics in the Wang Yanan Institute of Studies in Economics (WISE) at Xiamen University. Dr. Fan's reseach interests are theoretical econometrics with focus on high dimensional econometrics, model selection, microeconometrics and its applications in fields including cross-sectional asset prcing, industrial organization, etc.