Similarity analysis of federal reserve statements using document embeddings: the Great Recession vs. COVID-19

Abstract

AbstractThe coronavirus pandemic has already caused plenty of severe problems for humanity and the economy. The exact impact of the COVID-19 pandemic is still unknown, and economists and financial advisers are exploring all possible scenarios to mitigate the risks arising from the pandemic. An intriguing question is whether this pandemic and its impacts are similar, and to what extent, to any other catastrophic events that occurred in the past, such as the 2009 Great Recession. This paper intends to address this problem by analyzing official public announcements and statements issued by federal authorities such as the Federal Reserve. More specifically, we measure similarities of consecutive statements issued by the Federal Reserve during the 2009 Great Recession and the COVID-19 pandemic using natural language processing techniques. Furthermore, we explore the usage of document embedding representations of the statements in a more complex task: clustering. Our analysis shows that, using an advanced NLP technique in document embedding such as Doc2Vec, we can detect a difference of 10.8% in similarities of Federal Open Market Committee (FOMC) statements issued during the Great Recession (2007–2009) and the COVID-19 pandemic. Finally, the results of our clustering exercise show that the document embeddings representations of the statements are suitable for more complex tasks, which provides a basis for future applications of state-of-the-art natural language processing techniques using the FOMC post-meeting statements as the dataset.