Portfolio Optimization via Transfer Learning

Xinyu Zhang, Professor in Academy of Mathematics and Systems Science, CAS. His research is on model averaging and its interdisciplinary applications in economic forecasting, management statistics, and machine learning. He has published some papers in mainstream journals of statistics and econometrics.

Recognizing that asset markets generally exhibit shared informational char acteristics, we develop a portfolio strategy based on transfer learning that lever ages cross-market information to enhance the investment performance of the market of interest. Our strategy asymptotically identifies and utilizes the in formative datasets, selectively incorporating valid information while discarding the misleading information. This enables our strategy to achieve the maximum Sharpe ratio asymptotically, while maintaining a variance no greater than that of the strategy solely relying on the data from the market of interest. The promis ing performance is demonstrated by numerical studies and empirical analysis of investments in H-shares and A-shares.