Author ORCID Identifier

Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)



First Advisor

Dr. Baozhong Yang

Second Advisor

Dr. Vikas Agarwal

Third Advisor

Dr. Zhen Shi

Fourth Advisor

Dr. Sean Cao

Fifth Advisor

Dr. Lin William Cong


In this paper, we apply a state-of-the-art deep learning model to understand and predict dynamic patterns in mutual fund returns. A long-short portfolio based on the model’s prediction generates a 2.8% annualized Carhart 4-factor alpha. This abnormal performance is persistent for up to four years. The model improves the prediction of future fund alphas substantially by increasing the R-squared by more than 25% in a predictive regression that includes other fund skill measures as well as fund and time fixed effects. The model’s predictive power derives from its ability to capture fund skills embedded in dynamic strategies. We construct model-based conditional skill measures that depend on the inferred informativeness of macroeconomic and fundamental variables. Such measures are predictive of fund performance in future periods when the conditioning variables are highly informative. The conditional performance of these measures is also persistent. Overall, our results suggest that mutual funds have various specific skills that generate superior returns when the time is right.


File Upload Confirmation