Sentiment analysis in FOMC minutes using LDA topic model

Abstract

In today's low interest rate environment, a central bank places importance on the communication with the markets about future monetary policy and it has been recently observed that the market prices are affected by forward guidance published by a central bank. Therefore, it is important to reveal the impact of the forward guidance on expectations of market participants and asset prices. In this study, we classified FOMC meeting minutes into 8 topics using LDA, and calculated sentiment value for each topic considering subtrees of dependency structure. As a result, we revealed that the sentiments extracted by our method were largely related to macroeconomic indicators comparing with conventional methods.