If there is no recession coming, what can we conclude, since most of the word spread models were showing “sure bets”? An impossible outcome (possible world!), a breakdown of historical correlation, an omitted variable problem? To shed light on this question, I examine probability estimates from (i) the vanilla spread, (ii) the debt service ratio and the advanced foreign term spread, and (iii) the specification of the adjusted term-premium spread.
Figure 1: Average 12 months prior to recession, based on spread and short rate (blue), spread, short rate, debt service ratio and external term spread (tan), and premium adjusted term spread (green). The NBER has defined the peak number of recession days in gray. Source: NBER and author’s calculations.
The 12-month forecast of recession is obtained using probit models. The first specification (blue line) is a plain vanilla nominal distribution model estimated for 1990-2023M04 assuming no recession in the US until 2024M04. The second (tan line) is the one described in Chinn and Laurent (2024), reducing the Financial Conditions Index which did not provide much upward predictive power. The third specification (green) uses only the spread term, where the long term is adjusted to remove the premium term estimated by Kim and Wright (1995). Another reason for this change is here.
Notably, the plain vanilla distribution model provided the highest estimated probability of recession. The other two specifications do not yield rates that exceed 50%. Interestingly, the highest probability is May 2024. Since we do not have data for May — except for the Lewis-Mertens-Stock/NY Fed Weekly Economic Index and the Baumeister-Leiva-Leon-Sims Weekly Economic Conditions Indicators (both include release data until the 11 May), we may not have seen the data yet. (I still remember that in April/May 2001, many were thinking how we would escape recession, based on the data available at the time).
Source link