Policy Statements  |   Top 10 Lists  |   Charts  |   Of Interest  |   Sentiment Analysis  |   Spreadsheets  |   Notes
WhatTheFEDSaid.com (WTFS)
      by Bill McDonald
      Follow WTFS on Twitter: @WhatTheFedSaid
      Updated: Wed Dec 20 15:36:06 2017

WhatTheFedSaid.com (WTFS) provides a textual analysis of information released by the U.S. Federal Reserve's FOMC (Federal Open Market Committee). 1  Some interesting slices of the textual data are provided on this website, along with two files containing complete word-counts of the FOMC Minutes.  The WTFS website is updated immediately after the Policy Statements and FOMC Minutes are released on the FED's website and posted on WTFS with a three hour delay.   (If you're interested in subscribing to the instantaneous parses please e-mail). Analysis of the Policy Statements and FOMC Minutes is presented below.

Policy Statements

Given the nature of the FOMC Policy Statements (relatively short and following a stable template), parsing for word usage is not particularly useful. However, the template structure means that changes can be easily highlighted. The Policy Statements are released on the final day of the FOMC meetings. The segment below presents a markup of changes in the Policy Statement from the prior period to the current.


FOMC Minutes

Current Release:

Title:The Fed - Monetary Policy:
Release date:22-November-2017
# of words:6,896
Avg # of words per release from 2010 to current date:7,889.149

Top 10:

Words from the current FOMC minutes are ranked on frequency in the current period (%) with the frequency from the prior period (%(t-1)) reported in the fourth column. The averages reported are based on all FOMC minutes from 2010 to the current date. 2  The final two columns present the top 25 words ranked on their tf-idf score. The tf-idf score gives more weight to words that do not frequently appear in the minutes. 3  Because of the typical distribution of words ( Zipf's Law ), most of the action is in a small number of words – thus only the top 10 are reported.

Rank  Word%%(t-1)Average        Wordtf_idf
1  INFLATION1.64%1.13%0.84%        HURRICANES69.45
2  RATE0.75%0.91%0.69%        HURRICANE20.17
3  PARTICIPANTS0.71%0.62%0.63%        AFTERMATH10.62
4  PERCENT0.62%0.52%0.32%        STORMS9.47
5  TERM0.59%0.57%0.42%        BOOSTING8.23
6  LABOR0.54%0.42%0.40%        DISRUPTIONS8.15
7  CONTINUED0.49%0.33%0.38%        REFORM7.62
8  WOULD0.45%0.54%0.46%        VULNERABILITIES6.83
9  FUNDS0.42%0.41%0.30%        REBUILDING6.72
10  EXPECTED0.41%0.50%0.25%        TECHNOLOGICAL6.31

The following words in the current FOMC minutes have not appeared before: attracting, coalition, damaging, destroyed, disasters, displaced, earthquakes, headed, indebtedness, overhangs, plant, plausible, reversals, ruling, sharpened, spurring, susceptibility, unexplained, victory.


Word frequencies are plotted relative to the release date of the FOMC minutes. 4 


Of Interest:

Current debate centers around whether the Fed will increase, decrease, or maintain rates in their next outing. Perhaps the use of the word "gradually" might reflect their next move.


Sentiment Analysis:

Word frequencies are plotted relative to the release date of the FOMC minutes. Word lists for each sentiment category are taken from Loughran and McDonald (Journal of Finance, 2011).



  1. Textual analysis attempts to identify patterns in text based on artifacts such as word choice. In economics, finance, and accounting the method has been shown to provide useful information on the relation between economic announcements, SEC filings, earnings announcements, and stock prices, to name but a few. An overview of the method and recent literature in finance can be found here. Pablo Azar and Andy Lo provide evidence that textual information surrounding the FOMC meetings has predictive value for asset prices ( here). The method is not a crystal ball, but provides some insights beyond traditional information channels. The time series used in the analysis begins in 2010. Although FED documents are available over a longer history, the minutes became more predictably structured in 2010, which makes document comparisons more effective.

  2. The reported lists do not include generic stop words such as "the", "of", or "and". Also excluded are FOMC-specific stop words: "board", "committee", "economic", "federal", "governors", and "market".

  3. tf-idf = term frequency - inverse document frequency. Many text methods are based on simple counts of words, however in some cases a weighting scheme is used to create better measures of word impact. tf-idf multiplies the word count times the log of the ratio of the total number of documents divided by the number of documents in which the given word appears (more at Wikipedia). Thus an increased count in a common word has a score of zero (if it appears in all documents), while a word that appeaers for the first time has a much higher score.

  4. The "Inflation-related" chart reports counts of all words with a stem of "infla" or "deflat" or the words "price" or "prices". "Unemployment-related" words those with a stem of "employ" or "wage" or the word "labor". "Rates-related" words are "interest", "rates", "yield", and "yielding". "Energy-related" words are "crude", "energy", "gasoline" and "oil".