Given that most term spread models indicate a “sure bet,” if the economy isn't around the corner, what can we conclude? Impossible outcomes (this is a probabilistic world!), breakdown of historical correlations, omitted variables problem? To shed light on this question, I examine probabilistic estimates from (i) ordinary spreads, (ii) debt service ratio and foreign term spread widening, and (iii) term premium-adjusted spread specifications.
figure 1: Estimated recession probability over the next 12 months based on spreads and short-term rates (blue), spreads, short-term rates, debt service ratios, and foreign term spreads (tan), and term spreads adjusted for term premiums (green). NBER-defined recession peak-to-trough dates are in gray. Source: NBER and author's calculations.
The predicted probability of recession over the next 12 months is obtained using the Probit model. The first specification (blue line) is a vanilla term spread model estimated from 1990 to April 2023, assuming no US recession by April 2024.The second (tan line) is at Chin and Laurent (2024), abandoning an index of financial conditions that did not provide much incremental predictive power. The third specification (green) uses only term spreads, where long-term rates are adjusted to eliminate the term premium estimated according to Kim and Wright (1995).Some motivations for this modification are here.
Notably, the ordinary term spread model provides the highest recession probability estimates. The other two specifications predict no more than 50%. Interestingly, the highest probability occurs in May 2024. Lewis-Mertens-Stock/New York Federal Reserve Weekly Economic Index and Baumeister-Leiva-Leon-Sims Weekly Economic Conditions Indicator (Both include data released as of May 11), it's possible we just haven't seen the data yet. (I still remember April/May 2001, when many believed we had dodged a recession based on the data available at the time).
