Financial Market Prediction During the COVID-19 Pandemic Using Machine Learning

The study examines the predictability of S\&P 500 stock movements during the COVID-19 pandemic. It presents a comparative analysis of random forest and logistic regression models and evaluates the importance of different COVID-19-related features and control variables for the prediction task. The results show that all examined models significantly outperform a random classifier at a significance level of 1\%. The random forest and logistic regression forecasts increase significantly in accuracy when a feature set including COVID-19-related data is utilized compared to a benchmark feature set without COVID-19-related variables. The random forest model trained on the full features set yields the statistically most accurate market forecasts. A feature importance analysis of the most accurate model reveals that the predictive power is not concentrated on a single COVID-19-related feature type but spreads over multiple different features. Due to its empirical nature, this study represents a snapshot between July 2020 and December 2021. Hence, future research may examine its application in future market environments. The paper introduces a machine learning-based market prediction framework during the COVID-19 pandemic. The presented results suggest that COVID-19-related features improve market forecasts during the COVID-19 pandemic and should be included in contemporary asset pricing models. This paper presents a machine learning-based market prediction framework and sheds light on market predictability in the changing market environment of the COVID-19 pandemic.