For US, Germany, UK and Canada:
figure 1: For the US (black), Germany (brown), the UK (green), and Canada (red), the recession probability for the specified month is estimated based on the 10-year to 3-month spread with a 12-month lag. The probability of decay for a probabilistic model. Source: Federal Reserve calculations via FRED, OECD, NBER and authors.
Of course, most concerns are due to inversions of 10- to 2-year spreads (eg, Deutsche Bank) as described in Chin and Cook (2015), 10-year to 3-month spreads have stronger predictive power. However, the difference in forecasts for the US case (compared to Germany, UK and Canada) can be seen in Figure 2 below.
figure 2: Recession probabilities for the months shown are estimated based on the 10-year to 3-month Treasury spread trailing the US by 12 months (blue) and the 10-year to 2-year spread (brown). The probability of decay for a probabilistic model. Recession dates as defined by NBER are shaded from peak to trough in gray. Source: Federal Reserve calculations via FRED, OECD, NBER and authors.
I’m not ready to get 2-year government bond yields for the other three countries (for a snapshot of several G-20 countries as of March, see here), so here is an estimate based on a 10-year to 3-month spread. remember, Chin and Cook (2015) The spread was found to work well for the US and Germany in the 1990s and 2000s, but not so well for the other countries examined.
image 3: Based on the 10-year to 3-month spread, the estimated probability of recession for the given month is 12 months behind Germany (blue). The 3-month yield is the interbank offered rate. The probability of decay for a probabilistic model. ECRI-defined recession dates are shaded in gray. Source: OECD via FRED, ECRIand the authors’ calculations.
Figure 4: Based on the 10-year to 3-month spread, the UK (blue) has a 12-month lag estimate of recession probability in the given month. The 3-month yield is the interbank offered rate. The probability of decay for a probabilistic model. ECRI-defined recession dates are shaded in gray. Source: OECD via FRED, ECRIand the authors’ calculations.
Figure 5: Based on the 10-year to 3-month spread, the estimated recession probability for the given month is 12 months behind Canada (blue). The 3-month yield is the interbank offered rate. The probability of decay for a probabilistic model. ECRI-defined recession dates are shaded in gray. Source: OECD via FRED, ECRIand the authors’ calculations.
How well does the model fit? It can be seen that depending on the threshold used, the German model misses the most recent recession (i.e. a false negative), while the UK model gives a false positive. The Canadian model would mark a very low bar for the last two recessions. The McFadden pseudo R-squared for Germany, UK, and Canada were 0.21, 0.33, and 0.37, respectively. The corresponding figure for the United States is 0.26, 1960-2022M03 (0.23 for 10yr-2yr, 1976M06-2022M03).
While the current fashion is to focus on the 10- to 2-year spread (for the US, see discuss here), Engstrom and Sharpe (March 2022) This spread is considered to have no incremental predictive power for near-term forward spreads (6-month forward 3-month yield minus 3-month yield). Miller (2019) Examine a number of spread portfolios and find that during 1984-2018, the highest AUROC (predicted to true positive ratio) for the 12-month period were the 10-year federal funds and the 5-year federal funds, followed by 10-year to 3-month dissemination.
