Portfolio optimization based on machine learning predictions

Abstract

I use machine learning stock return predictions to improve minimum variance and Sharpe ratio maximization portfolio optimizations. I forecasted stock returns using Ridge regression, Lasso regression, Random forest, Gradient boosted tree and artificial neural network models with 89 variables. I chose the best model, Gradient boosted tree, based on R2 for portfolio optimization. Using machine learning forecasts improve minimum variance portfolio optimization, when short-sales are not allowed. However, machine learning Sharpe ratio portfolio performs poorly due to finding local instead of global optima. This thesis contributes by providing a new methodology to optimize minimum variance portfolios without short sales for practitioners.