BIMSA >
BIMSA Digital Economy Lab Seminar
Can Group Lasso Capture More Information About Stock Returns? Based on the Perspective of the Industry
Can Group Lasso Capture More Information About Stock Returns? Based on the Perspective of the Industry
Organizers
Speaker
Time
Monday, October 21, 2024 3:20 PM - 4:20 PM
Venue
A3-2-303
Online
Zoom 230 432 7880
(BIMSA)
Abstract
We introduce a nonnegative group lasso (NNGLasso) penalty into the coefficient fitting of a cross-sectional stock-lagged return forecasting model and design an efficient algorithm based on second-order information to solve it. This penalty can exploit the priori grouping information of stocks in prediction with high sparsity. In addition, it can effectively screen out the strongly correlated groups of target forecasting stock. Therefore, we choose the stocks grouped according to the industries they belong to as candidate predictors and establish an NNGLasso sparse prediction framework. This framework first extracts the effective stock-related industries through NNGLasso and then uses the least absolute shrinkage and selection operator (lasso) estimation for the selected sector groups. The final empirical results show that our forecasting method with NNGLasso has better out-of-sample performance than using the lasso regression or ordinary least squares(OLS) estimation method alone. Furthermore, the Markowitz portfolio with expected return vector constructed by our framework shows a more stable Sharpe ratio than its competitors, the sample mean portfolio, and the lasso portfolio.
Speaker Intro: Shengyin Jia is PhD student at Beijing Institute of Mathematical Sciences and Applications. His research interests center on portfolio theory and asset pricing, employing the optimization technique in portfolio construction.