This article was originally published in Forbes
This piece kicks off our new series on the analysis of the Federal Reserve using Sentieo’s natural language processing power and flexible Doc Search technology. We will focus on bringing interesting ideas and surprising revelations derived from thousands of public federal reserve documents. Join us as we scrutinize meetings, congressional testimonies, and press conferences with some truly impressive technology; and see what’s coming up next in our series:
Sentiment Analysis Of FOMC Statements Reveals A More Hawkish Fed
Why Is The Fed Still Raising Rates? The Yellen Effect
Assessing Fed Chair Hopefuls With NLP Analysis Of Past Speeches
Predicting The FOMC Statement With Beige Book Sentiment Data
The Federal Reserve System’s Federal Open Market Committee (FOMC) meets eight times a year, at 2 p.m. Eastern Time in the basement of a nondescript, Washington, D.C. office building. The terse statements released after those meetings drive the direction of global financial markets and the meeting minutes are carefully scrutinized carefully by the media.
We parsed recent statements and minutes since 2012 using Sentieo’s natural language processing and sentiment analysis and found some interesting trends.
For the most recent statement 9/20, the strongest topic continued to be inflation, as highlighted in the unfiltered word cloud shown here.
The intensity was roughly equivalent to the prior statement, as the Fed continues to be vexed by an inflation shortfall versus expectations. Based on the statements alone, this analysis would suggest that Fed intentions have barely changed. However, when we apply sentiment analysis to the words in the documents using the Loughran-McDonald context-specific lexicon, which assigns a simple positive or negative value to words based on the financial services industry context, the 9/20 statement occurs as much more hawkish.
The statement highlights the impact of Hurricanes Harvey and Irma, showing a much stronger negative tone versus the 7/26 statement with these topics. However, a reading of the statement itself shows these effects were noted as being short-term and unlikely to alter the medium-term economic outlook. More striking, however, was the stronger array of positive words which made the 9/20 statement look more hawkish than the 7/26. Markets were surprised by this, and the probability of additional hikes this year went from 50% to 69% after the statement came out.
We used Sentieo’s redlining capability to compare the two documents to each other and found the biggest difference to be the comments on the hurricanes:
We next looked at FOMC meeting minutes (which will not be released until next month for yesterday’s meeting) since 2012, looking for trends.
First of all, we looked at the most commonly used words and found inflation is at number 1, good news for a central bank that doesn’t have an inflation target mandate (if you’re a monetary hawk).
Next, we looked at sentiment over time for the minutes.
A localized regression (LOESS) line shows relatively flat sentiment overall but a slight uptick in the past year. More interestingly the first fed meeting of each year has sharply higher sentiment (much more positive tone). We also looked at sentiment scored for ‘uncertainty’, rather than just positive or negative:
This seems to indicate that the Fed has become slightly more uncertain. To provide more transparency into the lexicon driving these outcomes, we looked at the frequency for each sentiment category, including a fourth, ‘litigious’.
In the next article, we’ll look into the FOMC meeting minutes and the unclear reasoning behind the continued rate hikes.