Changes in Feature Importance for European Stock Market Determinants Caused by the COVID-19 Pandemic Shock

This paper is a study of the changes in the importance of selected market time series for the determination of the performances of the European equity index caused by COVID-19 pandemic. Equity assets are not the only asset class on the market and their prices interact with other instruments. As a comparable environment the selected commodity prices based on futures, foreign exchange quotations, and bond yields were chosen and their influence on Stoxx 600 performance was investigated. The problem of measuring the impact of the input variables on the dependent one can be approached in several ways, however, the permutation importance was chosen as it is universal for linear and non-linear models. The importance of the inputs varies over time and depends on the model. In this paper, the three models were chosen: linear regression, elastic net as regularized regression, and the random forest as a non-linear ensembled model. The results show the increased relationship between the equity market and the commodity futures prices after the COVID-19 shock, however, the general feature importance varies among the models.