Date of Award

Fall 2026

Abstract

This dissertation explores whether active equity mutual fund managers are skilled enough topick stocks and create portfolios that outperform the market on average. Much of the existing literature finds marginal or negligible evidence of skill which is in line with the standard theory of efficient market hypothesis. In this study I explore the role of investment style in assessing skill among active mutual funds. In chapter 1 of my dissertation, I use cross section method to investigate skill among active mutual funds. In this method, a bootstrap analysis is performed to estimate the distribution of outperformance that arises purely by random chance of luck. This estimated luck distribution is then compared with the actual distribution of outperformance. If the actual distribution exceeds the luck distribution, it provides evidence of managerial skill. I rely on the most updated bootstrap method by Harvey and Liu (2022) to estimate the distribution of fund performance attributable to luck. I find that around 17% of active mutual funds possess stock picking skill to create portfolios that outperform the market and covers all trading costs and about 7% of funds that outperforms the market and covers all cost. I also find that the evidence of skill is strongest among styles Small Cap Blend and Small Cap Growth.

In Chapter 2 of my dissertation, I examine the relationship between fund characteristics and future fundperformance using both machine learning models such as random forest and XGBoost model and a linear regression framework. If fund characteristics are able to predict future performance, this provides evidence of managerial skill. I find that when random forest model is used the out of sample performance of the superior funds is 4.4 basis points higher than those generated by linear models such as OLS per month. This is because machine learning models can capture nonlinear relationships among fund characteristics, enabling them to identify complex patterns in the data and generate more accurate predictions of future performance. I also find that the out of sample abnormal performance improves significantly by 1.7 basis points per month when investment style is incorporated in the machine learning model as predictor variables.

In chapter 3 of my dissertation, I evaluate the importance of style in addressing whether activemutual fund managers are skilled using the persistence approach. This approach tests whether outperformance in one period predicts outperformance in the subsequent period. This is one of the most used approaches to address the question. Existing studies that used persistence approach have mostly found none or short term such as one year persistence among active mutual funds. In this study I build on the chapter 2 of my dissertation and use the winner funds that has been identified by the random forest machine learning model and investigate if any of the winner funds persistently remain in the top decile year after year. I use two non-parametric approaches – First, comparing the actual distribution of winners with that of a random binomial distribution. Second, I compare the actual distribution of winner funds with a theoretical probability distribution that I derive mathematically based on the likelihood of a fund appearing in the top decile consecutively for a given number of years. I find that some winner funds identified by machine learning models remain winners for as long as 12.5 years out of 18 years in my sample. I also find some of the winner funds remain winners for about 7 years consecutively. I also do the same analysis for each of 9 investment styles, and I find that large cap blend and large cap growth style funds are the most persistent winners while small cap blend and small cap growth style funds are the least.

Document Type

Dissertation

First Advisor

Michael D. Goldberg

Second Advisor

Yin Germaschewski

Third Advisor

Jay Horvath

Department or Program

Economics

Degree Name

Doctor of Philosophy

Download

Share

COinS