Time-Varying Factor Selection: A Sparse Fused GMM Approach

Guanhao (Gavin) Feng is an assistant professor of business statistics at the City University of Hong Kong. He is also the program leader of MSc. in Business Data Analytics, a faculty affiliate at the School of Data Science, and a PI in the Lab for AI-Powered FinTech. Gavin's research publications have appeared in the Journal of Finance and Journal of Econometrics. Gavin obtained his Ph.D. and MBA degrees from the University of Chicago in 2017. His research interests include Bayesian statistics, empirical asset pricing, financial technology, and machine learning in finance.

This paper proposes a sparse fused GMM approach (SFGMM) to estimate a sparse time-varying coefficient model for selecting factors with heterogeneous structural breaks. SFGMM offers an alternative estimation to the dynamic stochastic discount factor model, where factor risk prices are sparse and time-varying, employing a high dimension set of conditioning variables and test assets. Evaluating U.S. equity factors, we find that ours outperforms several benchmark models, improving asset pricing and investment performance and providing insights into time-varying factor selection. Our results indicate that risk factors have the strongest explanatory power when the aggregate dividend yield or default yield is high, but their effectiveness is reduced when market liquidity is low. Moreover, our study reveals that the selection of factors changes over time, with some previously successful factors such as momentum and idiosyncratic volatility disappearing in the recent decade, while new factors like betting-against-beta and expected growth have emerged.