There’s a 60.9% Chance Interest Rates Won’t Change Next Month - Bayesian Classification of Fed Statements
I remember hearing in an Economics class long ago that the Federal Reserve has to give hints to the public about future interest rate changes since surprising financial markets is usually a bad thing.
I got to wondering recently if there were certain key words the Fed might use in its statements to warn the markets about upcoming changes in rates.
Coincidentally I also recently came across a very easy to use Python based Bayesian classifier called Reverend. (Here and here are some good introductions to the concept of Bayesian classification.)
So I came up with a plan put all of the FOMC statements I could locate into a Bayesian classifier and see if I could use words in past statements to predict future rate changes.
Here is the data I compiled:
FOMC Statements with Rates (Includes two files, a Gnumeric format and a wierd CSV format)
Next, I put together a little Python script to read in the data I compiled, and train two Bayesian models.
Here’s everything you need to run the program for yourself.
One model will try to predict the exact rate change e.g., -.5, -.25, 0, .25, or .5 percent.
The other model tries to predict only the direction (or no change) of an interest rate change.
To test these models, I used 10 fold cross validation on a randomly shuffled data set and then I did 10 runs of that.
From these runs:
The direction-only model succesfully predicted the next rate change direction an average of 72.17% of the time with a standard deviation of 18.8 across all of the folds for all of the runs.
The model to predict the exact rate change was correct an average of 69.5 % of the time with a standard deviation of 17.28.
I don’t have any other methods to compare these results to so I’m not sure if they are good or not. But it looks like the predictions are correct a fair amount of the time, so perhaps this method is at least plausible.
What words in FOMC statements signal rate changes?
I’ll pull out a few of the high probability words for somes of the classes (rate change amounts) so you can have a look.
Lowered rate by .5% at next meeting:
(mostly words associated with a poor economy, pretty neat!)
Raised rate by .5% at next meeting:
(Maybe words associated with an overheating economy?)
Left rate unchanged:
And finally, my predictions for interest rates for July, 2007:
Both models predict a 60.99% chance of no change to the Fed funds rate.
Here are all of the two models’ predictions:
Exact Rate Change model predicts:
Direction only model predicts:
‘No Change’, 0.60992435592948235