On December 19, 2016, the Federal Reserve Bank of San Francisco published an Economic Letter utilizing Prattle central bank sentiment data to demonstrate that when the tone of FOMC minutes diverges from the tone of the corresponding FOMC statements, treasury yields are affected. The research note confirmed that Prattle data can be used to predict the treasury market response to the tone of FOMC meeting minutes. This is the second time the SF Fed has published economic research based on Prattle’s data.
Prattle vs the FSO
In the research note, Fed researchers Fernanda Nechio and Daniel Wilson utilize both Prattle sentiment data and Factiva Semantic Orientation (FSO) to quantify the variance in tone between FOMC statements and minutes from the same meeting. Both Prattle and FSO data indicate that differences in tone between the two types of communications cause a significant increase in treasury yield volatility. However, Prattle sentiment data proved more effective in exploring the relationship between the Fed’s tone and treasury yields in four distinct ways:
1. Prattle data is comprehensive and unbiased.
To produce its data, the FSO crawls news articles of all types in an effort to find references to the Fed and then connects those references to mentions of the word “hawkish” or the word “dovish.” This methodology does not look at primary source material, such as Fed minutes/statements, or account for qualifying language, such as “less hawkish.” Prattle’s methodology fully accounts for the complex language of primary source material and evaluates all words within every central bank communication to produce its data. In addition, Prattle’s data provides an objective baseline for the tone of each type of Fed communication and each speaker. These qualities make Prattle data more comprehensive and unbiased than FSO derived data—as well as easier to use in a study like the one performed by the SF Fed.
2. Prattle data clearly identifies tradable opportunities.
In this study, Prattle’s data more effectively predicted when treasury yield volatility would be high. When there was a marked difference in the tone of statements and minutes as measured and compared using Prattle’s data, the corresponding yield volatility was almost 1bps larger (see figure 1) than comparable results with the FSO data. For an investor, this suggests that Prattle data more clearly identifies tradable opportunities than the FSO and methodologies like it.
Figure 1: Prattle Data vs FSO Data
3.Prattle data results in intuitive findings, consistent with global macro trends.
Whereas the results using FSO data found roughly equal magnitude (~3bps) effects when FOMC minutes surprised both hawkish and dovish in relation to the corresponding FOMC statement, Prattle data found a larger magnitude change (~3.5bps) when the FOMC minutes surprised hawkish, but no statistically significant effect when the FOMC minutes surprised dovish.
This intuitively corresponds to the post-financial crisis era in which the bond market has dovish baseline expectations for Fed policy and is rarely surprised by a dovish policy shift. This post-financial crisis market psychology is particularly apparent when considering the fact that dovish moves were typically tied to weakening financial conditions—to which market participants are attuned. Because hawkish surprises are more commonly tied to labor market indicators, which are typically more opaque to market participants, it follows that hawkish surprises would cause greater market movement than dovish shifts.
Unlike the FSO data, the Prattle data accurately reflects the treasury market’s indifference to dovish shifts and attentiveness to hawkish surprises. In other words, Prattle data more accurately reflects market reaction.
4. Prattle data is available in real time.
FSO requires researchers to access a critical mass of relevant news coverage to glean the tone of these articles. This necessitates a delay of hours or even days for news articles to be written based on the relevant FOMC statement and minutes. While FSO might be useful for academic economic research, the availability of Prattle data in real time makes it far more useful to finance practitioners.
Conclusion
The SF Fed study confirmed that traders can use differences in the Prattle scores of a statement/minutes pair to predict treasury market volatility, but the study also highlighted the value of Prattle data as a broader research and trading tool. Nechio and Wilson specifically note that the tone of FOMC communications has an impact across asset classes, implying that the results of their study could be substantially similar for equity and currency markets.
Click here to read the full study.
The Signal Team
