Creating alpha with Machine Learning algorithms

Abstract

This thesis investigates the application of machine learning (ML) and traditional financial models in portfolio optimization, focusing on the OBX index. The research aims to determine whether ML algorithms can outperform traditional models in forecasting returns, estimating volatility, and optimizing portfolio weights. The study employs advanced ML techniques such as Random Forest, Support Vector Machines, Gradient Boosting Machines, and k-Nearest Neighbors alongside traditional models, including ARIMA for return prediction and various GARCH frameworks for volatility modeling. Performance is evaluated using risk-adjusted metrics such as the Sharpe Ratio, Sortino Ratio, and Fama-French-Carhart regressions to assess the alpha generated by each model. Results reveal that ML-based portfolios significantly outperform the benchmark OBX index in both risk and return. Notably, the Random Forest model with a nine-week rolling window achieved the highest annualized return of 17.83% and a cumulative total return of 97.71% over 200 weeks, while maintaining lower volatility than the benchmark. Traditional models also performed well, with the IGARCH-based portfolio showing strong results, although they fell short of ML-based approaches.