Date of Award
2025
Degree Type
Dissertation
Degree Name
Doctor of Philosophy in Business Administration
Specialization
Finance
Department
General Business
First Advisor
Georges Tsafack
Abstract
This dissertation explores the role of frictions, regime dynamics, and predictive modeling in understanding persistent anomalies and managing financial risk in equity markets. Across three essays, I examine how transaction costs limit arbitrage opportunities, how market regimes shape the evolution of return anomalies, and how a hybrid econometric-machine learning framework can enhance risk forecasting.
The first essay investigates whether transaction costs and idiosyncratic risk constrain the profitability of long-term reversal and medium-term momentum strategies. Using a comprehensive dataset from 1927 to 2022, I estimate stock-level trading frictions through three distinct methods - Limited Dependent Variable models, Gibbs cost estimators, and bid-ask spreads. The results show that after accounting for transaction costs, reversal profits - especially among small-cap stocks - are substantially reduced or eliminated, and momentum profits also decline significantly. These findings highlight the central role of trading costs in limiting arbitrage and help reconcile the persistence of these anomalies with efficient market expectations.
Building on this, the second essay explores the dynamics of the momentum anomaly through the lens of market regimes. I document a notable post-2000 decline in momentum returns during positive market periods, contrasted with improved performance in down markets. To better understand and exploit these shifts, I develop a regime prediction framework that combines Markov Switching Models with machine learning algorithms, including decision trees, gradient boosting, random forests, and LSTM networks. This hybrid approach improves out-of-sample regime forecasting and enables adaptive strategy design, offering new insights into the evolving nature of return anomalies.
The final essay turns to the problem of risk estimation under regime uncertainty and volatility clustering. I propose a novel hybrid model - Markov-Machine Learning with Probabilistic Regime-Assigned GARCH (ML-MSM_GARCH) - that integrates regime-switching behavior with GARCH-based volatility estimation, dynamically weighted by machine-learning-predicted regime probabilities. Applying this framework to Fama-French factor portfolios (1963-2024), I find that the model outperforms standard VaR benchmarks, particularly at extreme quantiles, offering a more accurate and robust tool for tail risk management.
Together, these essays provide a comprehensive view of how frictions, structural shifts, and predictive modeling jointly influence asset pricing anomalies and financial risk, advancing both academic understanding and practical approaches in quantitative finance.
Recommended Citation
Aalipour, Nafise, "THREE ESSAYS ON INVESTMENT STRATEGIES, MACHINE LEARNING AND RISK MANAGEMENT" (2025). Open Access Dissertations. Paper 4481.
https://digitalcommons.uri.edu/oa_diss/4481