…using plain vanilla 10yr-3mo probit regression, over 1986M01-2019M08 period, using data shown below in Figure 1 [corrections to data, update results 9/5]
As noted in this post, based on the first half of August, the probability of recession in August 2020 was 47%. Using the entire August data, I obtain the following regression estimates:
Prob(recessiont+12) = -0.329 – 0.869 spreadt + ut+12
Prob(recessiont+12) = -0.329 – 0.895 spreadt + ut+12
McFadden R2 = 0.30
0.29, NObs = 392. Coefficients significant at 5% msl bold. The spread is in percentage points.
Here are the predicted probabilities:
Figure 2: 12 month ahead probability from probit regression on 10yr-3mo spread, (teal). NBER defined recession dates shaded gray. Forecast period shaded light green. [Probability using correction 1990M07 to be no-recession, 9/5] Source. NBER and author’s calculations.
Should one deliver a recession call? That would depend on the threshold. Over the last three recessions, a 40% threshold would catch all three recessions, and yield no false positives.
49% 51% > 40%.
Obviously, if one believes this time is different — so one should use a term premium adjusted spread — the implied probability of recession would be lower (as demonstrated in this post).