It’s kind of limiting to look only at recessions as predicted by the yield curve. What about growth?
I run regressions of 4 quarter GDP growth (in log terms) as a function of one year lagged 10 year-3 month Treasury spread, the spread augmented by one year lagged economic policy uncertainty (EPU, from Baker, Bloom and Davis), and spread augmented by EPU and contemporaneous four quarter growth in potential GDP.
Notice all forecasts suggest declining growth rates going forward — although the simple yield curve model predicts the least decline, and has the lowest R-squared at 2% over the 1986-2018Q3 period.
The yield curve augmented with economic policy uncertainty has greater explanatory power — R-squared of 8%. And this specification incorporating the estimated growth rate of potential GDP had an R-squared of 33%. Interestingly, these last two models forecast the same y/y growth rate in 2019Q4: 1.5%
All of these forecasts are accompanied by large prediction errors. Figure 2 shows the forecast from the yield curve EPU augmented model, and the 60% prediction interval (+/- 1 standard error). Notice the prediction interval encompasses negative values.
Many permutations are possible, including an estimate of the term premium, allowing a level term from the yield curve, and including a credit risk term. All this is just illustrative of what kind of slowdowns are implied by the standard model, and that incorporating one measure of economic policy uncertainty.