11 March 2025
Paul Ehling and Anne Wdowiak
The recent yield curve inversion at the end of 2022 has raised attention to the increased likelihood of a recession. This is because the term structure of interest rates, or yield curve, is known as a predictor of economic activity. An upward-sloping curve is a sign of good economic conditions while an inversion of the curve, a sign of an economic slowdown, or recession. For each of the last 8 recessions since the 1970’s, the yield curve was inverted in the 12 months before the recession.
In the US, the National Bureau of Economic Research (NBER) is responsible for defining and dating business cycles. A recession is defined as the period between a peak and a trough in economic activity, meaning that economic growth is negative. While we are waiting for news from the NBER, let us recall that the yield curve inverts when short-term rates exceed long-term rates. Higher short-term rates cause delays in business investments and tighter credit constraints, ultimately slowing economic activity, up to reaching negative growth. Driven by investors’ expectations of forward rates from short-term to long-term horizons, the yield curve can be thought of as simply the difference between long-term and short-term interest rates.
Finding a specific spread that signals recessions has received attention from academia and policy circles. Ideally an indicator should have very low rates of false positives and false negatives. A false positive is defined as an indicator signaling a recession when a recession does not occur in a period of time (usually 12 months ahead). A false negative is a recession occurring without being signaled by the predictor. While the most often used proxies of the term structure are the 10-years minus 3-months and 10-years minus 2-years spreads, here we also use forward spreads, that might be even better predictors.
Economists have identified and constructed other business cycles indicators based on leading, lagging and contemporaneous variables. Comovements of such macroeconomic variables integrated in a dynamic factor model with regime switching generate recession probabilities that perform both in-sample and out-of-sample. Hence, in addition to the term structure we use the following macroeconomic variables: (i) nonfarm payroll employment, (ii) industrial production, (iii) real manufacturing and trade sales and (iv) real personal income excluding transfer payments.
We then combine information given by the yield curve and the state of the economy, in a non-linear probit model to assess whether the recent yield curve inversion signaled a recession. We use the monthly time series of the 53-months ahead 1-month forward rate minus the 1-month yield (the forward spread) shown in Figure 1, and we control for the four real-time macroeconomic variables. We also compare our results with the more commonly used 10y minus 3m term spread.
Figure 1: We plot the 53-months ahead 1-month forward rate minus the 1-month yield from 1961 to 2023. The spread is smoothed using a 12-months moving average. Shaded areas represent NBER recessions.
Inspecting the forward spread suggests that the spread turns positive before the actual date of the peak, defined by NBER as the starting month of the recession. We therefore entertain the view that a recession starts when the yield curve has already turned back to its normal – upward – slope. Hence, we construct a dummy indicator that peaks when the spread turns positive.
Figure 2: The plots show positive spread indicators for the 10Y-3M spread and the forward spread. Shaded areas are NBER-dated recessions.
From Figure 2 we see that the 10Y-3M spread indicator peaks before each of the 8 recessions, bringing additional evidence to our hypothesis. In contrast, the forward spread indicator reveals different evidence; the forward spread doesn’t turn positive neither before or during the 1980 recession, and it peaks after the official start of the recession for half of them. Nevertheless, in-sample estimates indicate a significant positive effect of the indicator, for both spreads. These findings suggest that the slope of the yield curve is back to its upward shape in the 12 months preceding the beginning of a recession.
Figure 3: Out-of-sample one-quarter-ahead recursive predictions of the probability of a coming recession in the following 12 months, for the 10Y-3M term spread and the 53-months ahead forward spread.
Out-of-sample forecasts also confirm these results: we run recursive predictions, with initial training period of 5 years increasing monthly.
Both the term and the forward spreads models result in approximately similar and reasonably accurate predictions as can be seen from Figure 3. However, two false positives are observed in the mid-1990’s and early 2010’s in the two models, while the term spread predicts a false negative in 1990.
The forecasted probabilities from both models peak in 2022-2023, reaching higher level than any of the false positives. This suggests that either a recession has recently occurred, or the data has produced the strongest false positive so far. We emphasize that our predictions are one-year ahead forward-looking, whereas the NBER adopts a backward-looking recession dating process. This can result in a lag – sometimes up to 20 months – between the actual economic growth slowdown and the official recession start identification.