Predicting the reaction of financial markets to Federal Open Market Committee post-meeting statements

Abstract This article examines the impact of Federal Open Market Committee (FOMC) statements on stock and foreign exchange markets with the use of text-mining and predictive models. We take into account a long period since March 2001 until June 2023. Unlike in most previous studies, both linear and non-linear methods were applied. We also take into account additional explanatory variables that control for the current corporate managers’ and retail customers’ assessment of the economic situation. The proposed methodology is based on calculating the FOMC statements’ tone (called sentiment) and incorporate it as a potential predictor in the modeling process. For the purpose of sentiment calculation, we utilized the FinBERT pre-trained NLP model. Fourteen event windows around the event are considered. We proved that the information content of FOMC statements is an important predictor of the financial markets’ reaction directly after the event. In the case of models explaining the reaction of financial markets in the first minute after the announcement of the FOMC statement, the sentiment score was the first or the second most important feature, after the market surprise component. We also showed that applying non-linear models resulted in better prediction of market reaction due to identified non-linearities in the relationship between the two most important predictors (surprise component and sentiment score) and returns just after the event. Last but not least, the predictive accuracy during the COVID pandemic was indeed lower than in the previous year.