Term Spreads, Financial Conditions, Oil and Probit-based Recession Probabilities

With July in, it’s interesting to note that elevated probabilities of recession in 12 months come not from spreads, nor financial conditions, but oil prices.

Figure 1: Probability of recession from probit regression on 10yr-2yr spread (blue), on 10yr-3mo spread (green), and 10yr-2yr spread plus Chicago Fed Financial Conditions Index (tan), for 12 month horizon, estimated 1976M06-2022M07. NBER defined peak-to-trough recession dates shaded gray. Red horizontal line at 33%. Source: Authors calculations based on data in Figure 2.

While the probability for recession in 23M07 is now higher for the 10yr-3mo (due to the drop in this spread), it still remains below the 28.8% coming from the 10yr-2yr. Notice including the Financial Conditions Index reduces the probability.

Figure 2: 10yr-2yr Treasury spread (blue), 10yr-3mo Treasury spread (green), both in %, and Chicago Fed National Financial Conditions Index (tan). Three month Treasury is yield on secondary market, FRED series TB3MS. NBER defined peak-to-trough recession dates shaded gray. Source: Treasury, Chicago Fed both via FRED, NBER, and author’s calculations.

Why then are many forecasters indicating a high likelihood of recession. In some it’s foreign yield curves, inflation or unemployment configurations. One big determinant is oil prices; if one adds to the 10yr-2yr specification the relative price of oil (WTI to core CPI), then one adds to the forecasts the red line below.

Figure 3: Probability of recession from probit regression on 10yr-2yr spread (blue), on 10yr-3mo spread (green), 10yr-2yr spread plus Chicago Fed Financial Conditions Index (tan), 10yr-2yr spread plus relative price of oil (bold red), for 12 month horizon, estimated 1976M06-2022M07. July WTI oil price thru 7/23 and July core CPI is Cleveland Fed nowcast. NBER defined peak-to-trough recession dates shaded gray. Red horizontal line at 33%. Source: Authors calculations based on data in Figure 2.

Using a 33% threshold, this specification would miss the 1990, 2001 and 2020 recessions, but capture the others. Surely a wider search would yield a specification that captures the other recessions — a 3 month change in nominal oil prices captures 1990, 2001, but misses the 2007 and 2020 recessions (and comes close to predicting a recession in 2021M03).

The bottom line is that term spreads and financial conditions are not currently the reason that most are saying there’s a high likelihood of recession; oil prices are key to pushing probability estimates over the threshold.

(By way of contrast, note that I was getting over 40% probability of recession by 2020M06, back in 2019, using only 10yr-3mo term spread).