Machine learning portfolios for US stock prices: Directional forecasting before and during the COVID-19 pandemic

Resumen

In this study, we evaluate the performance of five Machine Learning (ML) portfolios—logistic regression, random forest, decision tree, gradient boosting, and adaptive boosting—against that of the Dow Industrial Average -passive approach. We consider an active portfolio management approach, employing out-of-sample backtesting to simulate the strategy performance as a categorical approach. We employ as predictors the opening price, the highest price, the lowest price, the closing price, the Williams %R and a the 13-week T-bills. During the whole period, before COVID-19, and during the pandemic, in all cases, at least one ML portfolio beats the index. These results suggest that overall, investors obtain positive outcomes if they use ML portfolios instead of investing passively in the index, obtaining the most benefits in times of greater uncertainty, such as the peak of the pandemic.