The 10yr-3mo German spread, hitherto a pretty good predictor, inverted at 2022M12. Hitherto a pretty good predictor of German recessions (as labelled by ECRI), the implied probability was 70%, using a plain vanilla spread regression.
From Chinn and Ferrara (2026), here are estimated probabilities for a probit regression, estimated up to 2019M12 (black line in bottom panel), corresponding to the 10yr-3mo term spread.
Figure 1: Top panel, 10yr-3mo German term spread, % (black); bottom panel, probit estimate for regression on spread, short rate, pre-pandemic (black line), full-sample (red), including expected change in unemployment (blue), including debt service ratio (green). ECRI defined peak-to-trough recession dates shaded light blue. Source: OECD, Chinn and Ferrara (2026).
In Chinn and Ferrara (2026), we find the term spread and short rate specifications have relatively high proportion of variation explained, both for the pre-pandemic and full sample. As shown in Figure 1, all recessions save the 2020 recession, are predicted using a 40% threshold. But a false positive is also recorded for 2024 using this same threshold.
Including the expected change in unemployment yields results with slightly greater proportion of variation explained (0.29 relative to a basic specification estimated over the same sample period at 0.19). The recession predictions are much the same as the basic specification, however, with a false positive still registered for 2024, using the 40% threshold.
Finally, including the debt-service ratio, which results in much different results for the US, raises the pseudo-R2 relative to a basic specification (0.33 vs. 0.19). All coefficients enter with expected and statistically significant coefficients, including the debt-service ratio. And perhaps most importantly, there is no false positive for 2024, using a 40% threshold.
Does one has to appeal to the debt-service ratio to resurrect the term spread as predictor. One alternative is to argue that Germany did experience a recession in 2023-24. For instance, Kemp (2026) writes: ” Germany’s economy experienced a pronounced recession during 2023–24, with real GDP contracting and growth falling notably behind the rest of the euro area.” However, neither ECRI, nor the German Council of Economic Experts, have declared a recession post-pandemic.
To see why, consider some main indicators:
Figure 2: top panel, German employment +15yo; middle panel, German real GDP, bn.2010Euro SAAR; bottom panel, German industrial production ex-construction (3 month centered moving average). Light green shading denotes peak-to-trough movements. Source: OECD via FRED, OECD.
Notice there is almost no coherence between the peak-to-trough periods. Peak employment is 2024Q4, peak GDP is 2022Q3, peak industrial production is February 2023, while peak recession probability is 2024M12. That is why there’s been no declaration of recession, defined as a broad, sustained decrease in economic activity.
The most charitable interpretation is that the term spread caught the downturn in employment, but not in GDP nor industrial production.
By
