So the US yield curve has inverted in earnest. Inverted yield curves, and compression of term spreads in general, predict US recessions. This is because bank earnings are a linear function of the term spread — their business model is to borrow short and lend long. When spreads get compressed, the present value of future bank earnings falls along with forward-looking measures of banks’ net worth. Banks therefore must shed risk, yielding a contraction of bank lending, and thus a slowdown in the real economy. This is the original risk-taking channel of Adrian and Shin.
It is not, of course, lost on Democrats that a recession ahead of the 2020 election would bolster their chances up and down the ticket. A recession could turn a narrow victory or loss into a landslide.
So what exactly are the implied recession probabilities for next year? We present the results of two probit models. We define a recession as negative quarterly growth in US real GDP. The baseline model simply models the probability of recession as a function of lagged term spread. After testing various term spreads it was found that the 4-qtr lagged difference between the yield on the 5 year note and the going rate on the 3-month bill in secondary markets, was the best predictor of both the US growth rate and US recessions. Figure 1 displays the predicted recession probabilities as a function of the spread. Note that the spread is now at -0.22 percent. So according to this model, there is a 32 percent chance that the US economy will fall into a recession in the first quarter of 2020. The United States is, of course, not an isolated economy. US growth depends on the growth of the world economy as a whole. The next figure shows that the growth rates of the United States and the OECD are contemporaneously correlated (r=0.766). The data is for 1962Q1-2018Q4 as reported by the OECD.
We find that after controlling for contemporaneous growth in the OECD as a whole, the 4-qtr lagged term spread predicts US recessions. The next figure displays the recession probabilities under three scenarios. Assuming the OECD as a whole does not slowdown or accelerate, with the term spread set at -0.22 percent, the model predicts a 32 percent chance of recession in 2020Q1, same as the baseline model. If there is an acceleration in the OECD, the recession probability falls to 22 percent. If there is a slowdown in the OECD, the recession probability for 2020Q1 jumps to 44 percent.
These probabilities are much higher than those predicted by FRED. The problem with that model seems to be that it does not take into account forward-looking variables that contain information on market expectations. That’s nothing new. Macroeconomists are always many steps behind market participants in anticipating recessions.
These probabilities are, in fact, understated, for the obvious reason that they are estimates for 2020Q1 instead of what we are really interested in, the probability that there will be a recession in time for the election. For instance, if the recession probability in each of last four quarters before the election is 30 percent, then the probability that there will be a recession in any of those four quarters is 76 percent. Of course, this is not the best way to add up these probabilities. The way to do that would be to create an indicator variable for a recession in any quarter of the coming year. Let’s try it. The next figure shows the recession probabilities for 2019Q4-2020Q3 under the same assumptions as before.
Under the baseline scenario the probability of a US recession/US real GDP falling on a quarterly basis, in 2019Q4-2020Q3 is estimated at 80 percent. Under the OECD slowdown scenario, this rises to 84 percent. On the other hand, an acceleration of the world economy reduces the probability of US real output falling in 2019Q4-2020Q3 to 75 percent. Note that the NBER defines recessions as two quarters of falling output. We are using a weaker (although no less valid) definition of recession.
Things are looking good for the Dems. The market odds are pretty good that there will be a recession in time for 2020.
Postscript. Let’s use the NBER definition for the sake of interpretation. That is, What is the probability that 2020Q1 the economy would be considered to be recession by the NBER? The 2-factor baseline estimate of that probability, assuming no change in OECD growth rate is 60 percent. Under the acceleration scenario, the probability falls to 50 percent; under a global slowdown scenario it jumps to 69 percent. See next figure.
The simple linear model with the term spread as the predictor works well with NBER recessions. See next figure. The estimated probability is 46 percent. Note that both these figures are again probabilities that the US economy, in the quarter four quarters from the current one, will be considered to being in recession by the NBER. Neither tells us the probability that the US economy will fall into recession in 2019Q1-2020Q2.
If we look at the probability that the US will fall into recession in 2019Q4-2020Q3 — presumably our proximate macro predictor of the election — we find a baseline single factor estimate of 69 percent. Roughly speaking, the yield curve is telling us that there is a 70 percent chance that there will be a recession in the United States in time to have an impact on 2020. We weren’t far off the mark in our earlier estimates.
The 2-factor probit model, with lagged term spread and contemporaneous OECD growth, yields a baseline estimate of 76 percent. Under the pessimistic scenario, this goes up to 79 percent. Under the optimistic scenario, to 72 percent. These estimates are in line with our original estimates, suggesting that our definition was not that far from the NBER’s. We were right the first time. The odds are closer to three-fourths than two-thirds that there is a US recession in time to sway the 2020 election.