Using Sentiment Analysis on Fed minutes to Understand Future Federal Funds Rate Changes.
This was my undergraduate thesis in Economics. I have included both the final write-up and the code.
This study investigates the information value of the Federal Open Market Committee (FOMC) meeting minutes and their potential to provide systematic indications for future policy decisions. I apply Natural Language Processing techniques to the meeting minutes to identify information that is not apparent through traditional quantitative measures. I then estimate a Taylor-Rule model that incorporates the sentiment of the meetings to analyse Fed Funds Rate changes between March 1996 to February 2023. My findings suggest that this information can explain future interest rate changes beyond what is explained by traditional macroeconomic variables.
1. Scraper - Employs multi-threading to find the Fed minutes release dates and the subsequent releases (publically available here). This method downloads all historic & recent minutes and adds paragraph labels for later cleaning. Relies heavily on the work of JonnyFLDN with only some minor adjustments.
2. Preprocessing - This program removes punctuation, tokenises, tags & lemmatizes the entire content.
3. Quant Analysis - Final cleaning, removing junk text. Performance of basic sentiment analysis using the dictionary method. Finally, estimating an ordered probit model with robustness tests (macroeconomic specifications (i.e Taylor Rule comparisons), Alternative Sentiment Measurements (FinBERT), and Shifts in Textual Sample over Time (ZLB exclusion)).