In an earlier posting I showed that the nominal GDP growth rate minus the federal funds rate, called the policy rate gap, is an easy-to-use measure of the stance of monetary policy and does a decent job predicting NBER recessions. A question that kept bugging me after doing that posting is what would happen if both the policy rate gap and the yield curve spread (10yr-3m Treasuries) were used in the probit model? Well, instead of working on a paper that needs to be completed for an upcoming conference, I tinkered around with the data this morning and got some exciting results on this question. First, take a look at a probit model estimated with four lags for the period 1956:Q1 through 2007:Q2:
Again, a rough and dirty cut in need of further refinement, but I got a pseudo R-squared of 74.5%! That is a huge improvement over what I had before and a better fit than probit models that use the yield curve spread alone. Obviously, my interest was peaked so I went ahead and plugged the numbers into a cumulative standard normal distribution and came up with the following graph:
Wow! This graph shows a 100% probability for every NBER recession that happened and does a much better job with the two periods that plagued the policy rate gap results, 1984 and 1994. The only miss is 1966-1967 which is a common miss for yield curve spread models. Edward Leamer, in his KC Fed symposium paper, notes that the year 1967 should have seen a recession--like today, there was a major housing bust dragging down the economy at the time--had it not been for the increased government spending on the Vietnam War. Of interest to us now, though, the model shows there to be an almost 50% chance of a recession.
Look forward to a paper from me that further develops this approach. These results are too promising not to do a paper.
An exciting result.ReplyDelete
What is it forecasting for probability of recession in 2008 ?
"Wow" is an understatement. What is most interesting about this model is that it is effective in eliminating false positives, except for the 1966 blip, at any level above 25% probability. Call that the noise threshold?ReplyDelete
through 2007:Q2 it has a probability of recession near 47.97,not to far of some of the calls being made now. For example, the intratrade contract on a recession in 2008 as of today near 55%. It will be interesting to update it when Q3 numbers are released...but by then it may be obvious we are in a recession.
Interesting. I don't fully understand it, but interesting nonetheless.ReplyDelete
I'm only missing one step now:
(1) Read about an economic-forecasting tool at Macro and Other Market Musings.