A year and half ago, I asked “Does it matter that yield curves (around the world) are sloping downward?” (October 12, 2007). I included this snapshot of term premia in the post:
The yield curves are different today because so many central banks are buying long-term debt with their freshly created Yuan, Yen, etc. They do not express what they did in the past.
At that juncture, I was also pretty circumspect myself about the predictive power of the yield curve, given the conundrum, the “saving glut”, and the Great Moderation. Ensuing events (e.g., worldwide recession) inspired Kavan Kucko and myself to write this paper (recently presented at this EABCN conference in Frankfurt). In the paper, we assay the predictive power of the yield curve for (i) growth of industrial production, and (ii) recession. Our key findings are:
- The term spread has significant predictive power when forecasting industrial production growth over a one-year time horizon.
- However, the predictive power for one-year growth is much weaker in the post-1997 period.
- Four out of six European models exhibited relatively high R-squared statistics when using data from 1998-2008 (and the adjusted R-squared actually increased in Italy and Sweden).
- The yield curve does have some predictive power for recessions (defined using ECRI criteria for non-US countries, and NBER criteria for the US).
- However, the predictive power is greatest for U.S., Germany, and Canada. Interestingly, in the latter case, the statistical significance of the term spread disappears when the short rate is included.
- We do not replicate Wright’s (2006) finding that adding in the short rate improves the fit of the equation for predicting recessions.
- The yield curve is hopeless in terms of explaining Japanese recessions.
The detailed results are contained in the paper. Figure 1 depicts the slope coefficients estimated over the full sample period for each country, while Figure 2 depicts a goodness of fit statistics.
Figure 1: Slope coefficient from regression of one year ahead growth on current term premium (10 year/3 month). Source: Kucko and Chinn (2009).
Figure 2: Adjusted R-squared from regression of one year ahead growth on current term premium (10 year/3 month). Source: Kucko and Chinn (2009).As remarked earlier, the probit models are not particularly successful in predicting recessions. Once again, recessions are best predicted in the U.S., Germany and Canada by the yield curve. But similarly, in the latter case, the significance of the yield curve is not robust to inclusion of the level of the short policy rate.
Unfortunately, given the small number of recessions in the latter subperiod, it’s not possible to make clear conclusions regarding how the predictive power of the yield curve has changed over time.
Of course, these are preliminary results. In particular, it’s not clear what lessons can be gleaned from the current yield curve for the US (and other countries) as the policy rates hit the zero lower bound. (On the other hand, Charles Goodhart argues that the yield curve has greatest power when uncertainty is high — and I can’t think of a time in the recent past where uncertainty has been higher.)
Moreover, as is obvious, we do not allow for some of the effects and complications laid out in this post by Jim.