Equity sector rebalancing via machine learning

Abstract

In this dissertation the author will analyze whether supervised machine learning models namely Artificial Neural Networks, Support Vector Machines and Logistic Regressions can predict shifts in equity returns on a sector basis. Typically, in asset pricing linear factor models with a small number of variables are used. However, due to market efficiency, equity returns are highly influenced by unforecastable events making this task more challenging. Simple linear regressions also have difficulty incorporating a larger number of predictor variables, which the literature has accumulated over the decades, creating an opportunity for machine learning techniques. The Machine Learning models will be used to forecast whether the excess return of each equity sector over a period of one month will be positive or negative. Then using the model’s predictions capital will be allocated between the sectors and treasury bonds, building different portfolios namely an equal weighted, a value weighted portfolio. After all portfolios are built their performance will then be compared against the benchmark, namely the S&P500 index being back tested over a period of 25 years. The portfolios built using the forecasts from the ML models lead to an increase in absolute and risk­adjusted returns beating the benchmark. The implemented strategies were shown to protect investors against larger market declines, showing the potential of Machine Learning as an investment tool.

Date of Award	2 Feb 2023
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Dan Tran (Supervisor)