The Effects of the Content of FOMC Communications on US Treasury Rates

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2096–2102, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics The Effects of the Content of FOMC Communications on US Treasury Rates Christopher Rohlfs1 and Sunandan Chakraborty2 and Lakshminarayanan Subramanian2 1Morgan Stanley 2New York University New York, USA Christopher.Rohlfs@morganstanley.com, Sunandan@cims.nyu.edu, lakshmi@cims.nyu.edu Abstract This study measures the effects of Federal Open Market Committee text content on the direction of short- and medium-term interest rate movements. Because the words relevant to short- and medium-term interest rates differ, we apply a supervised approach to learn dis- tinct sets of topics for each dependent variable being examined. We generate predictions with and without controlling for factors relevant to interest rate movements, and our prediction results average across multiple training-test splits. Using data from 1999-2016, we achieve 93% and 64% accuracy in predicting Target and Effective Federal Funds Rate movements and 38%-40% accuracy in predicting longer term Treasury Rate movements. We obtain lower but comparable accuracies after control- ling for other macroeconomic and market fac- tors. 1 Introduction This study uses the verbal content of Federal Open Market Committee (FOMC) public communications to predict the directions of interest rate movements on the days those communications are released. The FOMC, who determines government policies rel- evant to interest rates, meets roughly eight times a year and releases a statement after each meet- ing. The FOMC is known to be an important mover of markets, and economic research has found that equity and interest rate markets tend to move when FOMC communications are released (Farka and Fleissig, 2012; G¨urkaynak et al., 2005; Mueller, 2015; Rosa, 2011) that the policy actions alone do not explain these responses (and thus the content of the text must be responsible) (G¨urkaynak et al., 2005), and that the directions of market movements coincide with a human-coded measure of the sen- timent expressed in the texts (Rosa, 2011). Writ- ers in the finance industry and in the popular press have also examined word clouds of FOMC min- utes (Cofnas, 2010; Durden, 2011) and have dis- cussed the market implications of the total number of words included in FOMC minutes (Fitz-gerald, 2014; Kennedy, 2014; Wynne, 2013). A growing body of research applies NLP meth- ods to understand the market effects from the con- tents of these texts. Researchers have applied La- tent Semantic Analysis (LSA) to describe the key topics covered in FOMC minutes, obtaining insights into the FOMC’s deliberation process (Hansen et al., 2015; Fligstein et al., 2014; Schonhardt, 2013). Ad- ditionally, researchers have used NLP-derived mea- sures of the content of FOMC minutes to predict equity and interest rate volatilities; (Boukus and Rosenberg, 2006) use LSA-defined topics in a re- gression context, and (Zadeh and Zollman, 2009) apply a dependency-based measure of text content to an expert-classified set of financially relevant words and then use both regression and SVM to predict volatility. Papers have found temporary effects of the sentiments from company-specific news articles and message board postings …

Content truncated to 500 words