Earlier this month, Deutsche Bank warned that there’s a 60% probability of a US recession, based on the firm’s analysis of the Treasury yield curve. Neil Irwin at the NY Times wonders: “Can We Ignore the Alarm Bells the Bond Market Is Ringing?” But perhaps the better question is whether we can rely on any one signal—or model—for evaluating recession risk? No, we can’t, and fortunately we don’t have to. The one exception for this common-sense rule: combining recession-risk estimates from multiple methodologies and taking the median number as the closest thing to a single, relatively reliable measure of macro distress. On that note, allow me to introduce the Composite Recession Probability Index (CRPI).
CRPI reflects the median of recession probability estimates via a probit model for the following five benchmarks:
All of these indexes are regularly reviewed in the US Business Cycle Risk Report and starting with the next issue CRPI will be added to the lineup. The logic is straightforward and backed up by decades of research, which falls under the heading of forecasting/nowcasting literature (for a deep dive into the details, you might start with “Combining forecasts: A review and annotated bibliography” by Robert T. Clemen (International Journal of Forecasting)).
The key takeaway: more forecasts are better than fewer. Every nowcast/forecast is wrong, but they’re wrong in different ways. By carefully selecting different methodologies and datasets for nowcasting/forecasting, and then combining the results (in this case by taking the median probability estimate), we can enhance the reliability of the signal relative to any one of the component indexes.
As a quick test of that idea, consider CRPI’s recent history. In particular, note the moderate rise in the probability estimate in the first half of this year—peaking (so far) at around 8% in March. The current reading (as of July 11) is slightly lower, at roughly 6%. In other words, recession risk is low, although not quite as low in the first half of 2016 relative to last year. To be precise, the probability of an NBER-defined recession is about 6%, based on economic and markets data available through yesterday.
Recall that earlier in the year there was heightened concern that recession risk was much higher–and rising. The bulk of this concern, however, was driven by market-based analytics, which suffer from quite a lot of noise. By contrast, the hard economic data, in the aggregate, painted a brighter profile at the time. For now, the brighter profile has prevailed, even though a markets-based review suggested otherwise a few months earlier.
The lesson, of course, is that we should be wary of relying on market signals alone for estimating recession risk. Even broader-minded benchmarks are suspect to a degree. The good news is that combining nowcasts/forecasts from models of different designs and inputs can juice the reliability of the analysis, mainly by reducing the frequency of false signals.
There’s still no guarantee—nothing rises to that level for estimating recession risk in real time. But we can move a few steps closer to the ideal by reducing the noise potential via combining forecasts/nowcasts.
For the moment, CRPI tells us that the US economy isn’t in recession. But the 94% probability that this statement is accurate comes with a big caveat: it’s forever in flux as new data arrives. There’s a solution, of course: running fresh numbers regularly on CRPI (and all its component benchmarks) via The US Business Cycle Report.