Enhancing estimation of expected returns in modern portfolio theory through machine learning

Abstract

In the context of Modern Portfolio Theory (MPT), this study delves into the application of Machine Learning techniques to enhance the accuracy of expected return estimations. The research encompasses a comprehensive three-fold objective that traverses multiple dimensions of financial prediction and portfolio optimization.

Beginning with an exploration of Machine Learning methods commonly employed in stock return forecasting and time series forecasting domains, the study draws from an extensive literature review to spotlight key models, including Random Forest, Gradient Boosting, Long Short-Term Memory, and Transformers. This cataloging of methodologies forms a foundation for the subsequent investigation.

Building upon this framework, the study shifts its focus to assess the potential advantages of Machine Learning methodologies when compared to traditional methods for expected return estimation. By leveraging historical price data as input, the research examines the extent to which Machine Learning enhances accurate estimations of expected returns. A thorough analysis subsequently underscores the remarkable predictive capabilities of Machine Learning models, with Transformers, the architecture well known for its involvement in ChatGPT, emerging as a particularly potent contender. These findings can significantly influence financial practices, signaling a broader transformation toward data-driven decision-making. This paradigm shift is essential in an era where the complexity and scale of financial data require innovative predictive tools. The implications extend beyond individual strategies, ultimately reshaping the landscape of investment and risk management practices across the industry.

In a final facet, the study probes the financial implications inherent in enhanced expected return estimations, particularly when incorporated within the context of Modern Portfolio Theory. It is here that the research illuminates a counter intuitive insight: while better-predicted expected returns are crucial, they do not invariably translate to superior cumulative realized portfolios. The study uncovers the intricate interplay between estimation precision and the complexities of portfolio construction, highlighting the nuanced nature of investment decisions within a broader financial ecosystem.