Deep Learning Mutual Fund Disclosure: Risk Sentiment, Risk Taking, and Performance
47 Pages Posted: 18 Apr 2022
Date Written: June 27, 2021
Abstract
We use a deep learning model to extract syntactic structures from mutual fund managers’ narrative discussion and construct forward-looking risk sentiment measures. These measures capture the manager’s sentiment and consciousness about the risks facing a mutual fund. Managers with a more negative (positive) risk sentiment are more likely to reduce (increase) their portfolio risk in the subsequent period. Managers who are conscious about negative risk generate superior risk-adjusted return and higher Sharpe ratio, and are more likely to have higher intra-quarter trading skills, measured by return gap. Risk-conscious managers also obtain higher Morningstar ratings and attract greater flows from sophisticated investors, giving incentives to managers to report their risk sentiments. Given their forward-looking nature, our new measures can inform investors and researchers about fund managers’ risk management and investment decisions. Our study also calls for more applications of deep learning models in textual analytics that can reveal and analyze linguistic features previously inaccessible to researchers.
Keywords: Deep Learning, Machine Learning, Mutual Fund, Disclosure, Textual Analysis, Sentiment
JEL Classification: C45, G11, G14, G23
Suggested Citation: Suggested Citation