Should he? FRB Minneapolis Fed President was at UW Madison today, and stated:
“The fact that the yield curve is now uninverted gives me some indication that we have taken the brakes off the economy,”
Should we rest easy? In some sense, the answer is “yes”.
Take the plain vanilla term spread model; run a probit regression of recession 12 months ahead on the simple 10yr-3mo spread, and one obtains an 50% probability of recession in August 2020, declining to 39% by October. This point supports Kashkari’s view. In addition, as many have pointed, the historical correlations may not provide a reliable guide to recession probabilities, given the drastic change in the size and sign of the term premium, and the implied recession probability is actually substantially lower.
Simply subtracting off the term premium from the term spread an estimated 10 year term premium should in principle isolate the pure expectations hypothesis of the term structure component. Define spread as the 10yr-3mo term spread in percentage points, and tp10 as the ten year term premium, as reported by the New York Fed. (Data in XLS file here.)
Prob(recession=1)t+12 = -0.442 – 0.794(spreadt-tp10t)
McFadden R2 = 0.18, SSR = 64.76, Nobs = 622, sample 1967M01-2019M10. Bold face denotes significance at 1% msl.
The predicted probability of the recession peaks at 5% in April 2020.
However, one could let the data speak, and allow the coefficients on the term spread, and on (the negative of) the term premium be unconstrained to be the same. Doing so leads to the following regression estimates.
Prob(recession=1)t+12 = -0.442 – 0.794(spreadt) + tp10t)
McFadden R2 = 0.32, SSR = 53.85, Nobs = 622, sample 1967M01-2019M10. Bold face denotes significance at 1% msl.
Notice the McFadden R2 increases substantially, while the sum of squared residuals drops. A Wald test for equality of coefficients on spread and (negative of) tp10 rejects wildly (p-value = 0.0000).
In Figure 1, I show the various spreads (unadjusted, adjusted, and unconstrained adjusted).
In Figure 2, the associated probabilities are shown. The unconstrained specification confirms that Kashkari should be reassured; the probability of recession falls from 35% to 26% in October. But maybe not too relaxed.