Predicting stock returns: ARMAX vs. machine learning

Abstract

In the modern world, online social and news media significantly impact society, economy, and financial markets. In this chapter, we compared the predictive performance of financial econometrics and machine learning and deep learning methods for the returns of the stocks of the SP100 index. The analysis is enriched by using COVID-19 related news sentiments data collected for a period of 10 months. We analyzed the performance of each model and found the best algorithm for such types of predictions. For the sample we analyzed, our results indicate that the autoregressive moving average model with exogenous variables (ARMAX) has a comparable predictive performance to the machine and deep learning models, only outperformed by the extreme gradient boosted trees (XGBoost) approach. This result holds both in the training and testing datasets.