Thèse numérisée par la Direction des bibliothèques de l'Université de Montréal.

The first essay examines the relation between fund performance and stock selection process. I classify mutual funds into two groups according to their distinctive stock selection approaches: tire kickers who rely on fund managers' personal judgment and fundamental analysis to pick stocks, and quant jocks who use computer-based models to select stocks. I examine how the stock selection approach affects mutual fund performance and economies of scale. I document an increasing trend of quantitative techniques used by mutual funds, in addition to some unique characteristics of quant jocks. Quant jocks and tire kickers have similar factor-adjusted alphas, but quant jocks have higher Sharpe ratios. Quant jocks tend to be much smaller than tire kickers. I explore possible explanations for the size difference. I find that although quant jocks can cheaply screen a large universe of stocks, the stocks that quant jocks invest in are smaller and less liquid, which results in higher transaction costs and limited scalability of quantitative investment strategies. The second essay investigates mutual fund managers' private information about future stock returns as revealed in their portfolio holdings. Specifically, we develop three different stock alpha estimators to predict stock returns based on portfolio compositions and past performance of mutual funds. We find that investment strategies based on our stock alpha estimators perform well, when using information on recent fund holdings and fund purchases. This evidence suggests that fund managers' stock selection skills are quite persistent, and vary widely in the cross-section. We also compare our strategies with 12 quantitative investment signals based on market anomalies, and find that our strategies are not subsumed by these quantitative signals. Thus, our stock alpha estimators reflect private skills of active fund managers that are unrelated to known anomalies. Finally, we develop a conditional stock alpha estimator using information on stock characteristics and fund characteristics. Investment strategies based on the conditional stock alphas deliver further improved performance.