When Mutual Fund Managers Write Confidently

Abstract

This paper studies the information content of a subtle writing style, the expression of confidence. We use unsupervised machine learning to generate new lexicons for three aspects of a writer’s level of confidence in expressing opinions: certainty, intensity of expression, and emotional overstatement. Using them to capture mutual fund managers’ confidence in letters to shareholders, we show confidence contains important information about skill: underperforming managers writing with strong certainty significantly outperform other underperforming managers in the next six months, whereas top performing managers writing with strong intensity or emotional overstatement underperform. Further analyses reveal our confidence measures are informationally distinct from the human-based confidence measure and tone. Analyses of investors’ capital flows show they are largely incapable of detecting the information embedded in textual confidence. Our findings suggest that writing styles contain import forward-looking information.