Construction of the FRB Beige Book Corpus and Analysis

Abstract

We tackle constructing and preprocessing a corpus of the Federal Reserve's Beige Book, which is publicly available on the Federal Reserve's website. The Beige book is a collection of information on prevailing economic conditions and trends collected by each of the twelve regional Federal Reserve Banks. However, the format of the Beige Book varies depending on when it is published, and there is no readily available, cleaned data. On the other hand, the text data from the Beige Book has been acknowledged as a valuable source of information for economic analysis. Therefore, in this study, we aim to create a cleaned dataset that can be easily accessed and used by researchers for various analyses. Specifically, we divide the texts into individual sentences and tag each sentence with a "date", "distinct", "topic" and "finbert_score~(sentiment)" label. In addition, we conduct empirical analyses and present examples of applications of our corpus in reference to previous research that analyzed the Beige Book. Finally, we highlight the challenges associated with our corpus and empirical analyses.