Once upon a time, I built a model for predicting recessions.
It was built on the work of a Fed paper, plus some inspirations from the Bank Credit Analyst.
It looks mainly at the yield curve slope and the commercial paper spread.
The signal state was calculated once a week.
Since it was intended as a market timing tool, not a recession predictor, I measured the average
forward return four weeks later for the S&P index adjusted for inflation. (ignored dividends).
Of course, there are very few recessions in history, so it’s easy to overfit the data.
So, I did!
But…that was a long time ago.
When I built the model, in May 2006, these were the results in the in-sample period:
CAGR 1964-2006 (average four week hold, annualized) starting from normal periods: +11.1%/yr
CAGR 1964-2006 (average four week hold, annualized) starting from “recession indicated” periods: -3.9%/yr
Seems like a pretty decent model, on average.
Here are the results for the out of sample period, 2006 to date:
CAGR starting from normal periods: +10.4%/yr
CAGR starting from “recession indicated” periods: -15.8%/yr
Even if many individual weeks were terrible, that’s a pretty good average result, nicely validated.
I mention all this because it started indicating the likelihood of a recession on Feb 25 this year, and is still saying that.
So, the omen is bad. Also, the real index has dropped -14.3% since is started saying that.
Jim