<p>This dissertation studies the impact of mutual fund managers' beliefs on fund performance in the first chapter and focuses on the impact of firm behavior on stock performance in the second chapter. </p> <p><br></p> <p>In the first chapter, I utilize the Revealed Preference Theory to recover fund manager belief formation directly from their actual trading activities. By relating stock holdings in a fund's portfolio to past factor returns, I document three facts about managers' belief formation: 1. In contrast to belief extrapolation, a substantial fraction of mutual fund managers act as contrarian investors who expect lower factor returns after a good factor performance; 2. Whether a fund trades in an extrapolative or contrarian way is due to its managers' expectation biases rather than fund style investment strategy, fund catering strategy or fund risk preference; and 3. Contrarian managers generate superior performance, are more experienced investors, charge higher expense ratios, and manage smaller US equity funds. The top (contrarian) managers significantly outperform the bottom (extrapolative) managers by a return of 3.4% per annum after adjusting by FFC4 factor models.</p> <p><br></p> <p>The second chapter is co-authored paper with Yan Liu and Feng Zhang. In this chapter, we systematically replicate the bulk of long-run event studies conducted in the last three decades from 1990 to 2020 using extended samples and four long-run performance measures. The final sample contains 62 papers of long-run event studies and 148 corporate events. Our findings suggest that long-run return anomalies documented in the last three decades are not robust, and firms do not earn long-run abnormal returns following various types of corporate events. Only 2% of the 148 corporate events we replicate earn post-event abnormal returns that are statistically significant at the 5% level based on all the four performance measures, and the fraction further shrinks to 0% at the 1% significance level. Viewed together, our findings suggest that these long-run abnormal returns after corporate events are likely the result of data mining or "p-hacking".</p>