Forecasting recessions in the United States with the yield curve

There is a large literature on the relationship between the yield curve and recessions that started in the 1980s as a result of the inability of macroeconomic models to explain sudden downturns in economic activity. In this box, we review the literature on the predictive power of the yield curve, with a particular focus on the United States, and compute the current implied probability of a recession using data ranging from 1953 to 2018. We find that a recession in the US is not likely soon, but the recent flattening of the yield curve should be monitored carefully.

Recessions have often been associated with an inversion of the yield curve, moving from a positive slope to a negative one. A positive slope of the yield curve comes from the fact that investors require a premium for holding longer maturity bonds (the term premium) or expect the short-term rates to be higher in the future. A negative slope is a more unusual event – it has occurred less than 10 per cent of the time in the US in the past 65 years – reflecting the fact that the economy is probably in a transitory phase. In that situation, investors expect the future short-term rates to be lower than the current ones, according to the expectations hypothesis. Figure 1 displays the spread of the 10-year Treasury note yield minus the 3-month Treasury bill yield. The figure shows indeed that recessions have often been preceded by a negative value of the spread.

Figure 1: Yield spread between 10-year Treasury note and 3-month Treasury bill

Researchers from the Federal Reserve, in particular Arturo Estrella, played a large role in developing this literature. Laurent (1988) and Estrella and Hardouvelis (1991) first showed that the spread in yield between longer-dated Treasury notes and shortdated T-bills could help predict future real GNP growth. Harvey (1988) and Estrella and Hardouvelis found that the yield spread can also be used to help forecast other economic variables such as consumption and investment growth. Comparing the role of the yield spread to other financial and economic indicators, Estrella and Mishkin (1998) concluded that while stock market and Stock-Watson indicators have good predictive power one quarter ahead, the yield spread dominates at longer time-horizons, in particular one year ahead, to forecast recessions. International evidence is however more mixed than for the United States; Chinn and Kucko (2015) found that the yield spread performed relatively well predicting recessions in Germany and Canada but it performed less well in Japan and Italy.

The yield curve sent a strong signal ahead of all the past recessions

We reproduce the probit methodology of Chinn and Kucko to estimate the probability of a recession in the US at any point within the next twelve months. As is common in the literature, we define the yield spread as the difference in yield between the 10-year Treasury note and the 3-month Treasury bill and use the recession dates from the NBER. All data are of monthly frequency, between April 1953 and March 2018. Figure 2 shows the implied recession probabilities from the model and in the shaded areas the actual recessions. A reading above 50 per cent indicates that a recession is likely to be either ongoing or about to start in the next twelve months. Looking at the statistical power of this model, we can see that it has good ‘precision’ – when the model identifies that a recession month is likely during the following 12 months, it is correct in 69 per cent of the cases – but a rather low ‘sensitivity’ – when a recession month actually happens during any of the following 12 months, the model identifies it in only 35 per cent of the cases. So the model is far from perfect because it gives some false negatives. However, the model fares much better at predicting the onset of a recession period. For all the nine recession periods in our sample (we exclude the first one in 1953–4 because we don’t have yield information one year in advance of the beginning of the recession), the indicator correctly rose above 50 per cent in the 12 months before the beginning of the recession. The signal came sometimes earlier and sometimes later. In the four recessions of 1970, 1980, 1990 and 2008, the indicator was already at 50 per cent or above more than 12 months before the first month of the recession and in the remaining five recessions the signal appeared between 5 and 10 months before the beginning of the recession.

Figure 2: Probability of a recession in the US within the next 12 months, implied by the yield curve

The latest data point for March 2018 is a yield spread of 1.1 per cent, which using the model translates into an estimated probability of recession within the next twelve months of 30.9 per cent. While this may appear quite high to forecasters, who look at a range of economic indicators, it is only marginally higher than the unconditional probability of 27.8 per cent, and well below the 50 per cent threshold. What is more interesting is that the indicator is on an upward trend, and in November 2017 it reached a 10-year high before levelling off. The main reason for the flattening of the yield curve in recent years is that the 3-month yield increased from 0 to 1.7 per cent while the 10-year yield stayed broadly flat. In short, the indicator does not indicate an imminent recession, but it would be wise to monitor if the yield curve flattens further.

ZLB and QE artificially pushed down the yield spread

Interpreting these results should be done with a degree of caution for several reasons. First, the severity of the Great Financial Crisis has pushed the Fed into unprecedented monetary actions; Fed funds rates reached the Zero Lower Bound (ZLB) and the Fed bought a large quantity of long-dated government bonds as part of its Quantitative Easing (QE) programme. Both ZLB and QE have probably artificially reduced the yield spread by respectively pushing up the short rate and pushing down the long rate compared to what they would otherwise have been. As a result, the information content of the yield curve may have been temporarily blurred. Looking forward, the current tightening of monetary policy by the Fed is likely to have ambiguous effects on the slope of the yield curve; increasing Fed funds rates increases the short-term rates but reducing its balance sheet (composed mainly of long-term T-notes) also increases yield at the long end. In that regard, the current situation is different from what happened in previous business cycles. Secondly, the coefficients of the probit regression are not stable with regard to the estimation period sample, which means that out-of-sample forecasting performance is likely to be less good. As an example of such a structural break in the data, the average yield spread has nearly doubled since the mid-1980s as can be seen in figure 1: between 1953 and 1985, it averaged 1.1 per cent and since 1985 it has averaged 1.9 per cent. Indeed, the ‘Great Moderation’ led to short rates decreasing more than long rates on average.

To conclude, using the yield curve to predict upcoming recessions is an easy and model-free way of extracting some of the information contained in the government bond market to forecast an event that is otherwise very difficult to predict. Our research suggests that the possibility of a recession in the US has risen somewhat over the past year but it is still far from our central case outlook.

References
Chinn, M. and Kucko, K. (2015), ‘The predictive power of the yield curve across countries and time’, International Finance, 18(2), pp. 129–56.
Estrella, A. and Mishkin, F. (1998), ‘Predicting U.S. recessions: financial variables as leading indicators’, The Review of Economics and Statistics, MIT Press, 80(1), February, pp. 45–61.
Estrella, A. and Hardouvelis, G. (1991), ‘The term structure as a predictor of real economic activity’, Journal of Finance, 1991, 46, 2, pp. 555–76.
Harvey C. (1988), ‘The real term structure and consumption growth’, Journal of Financial Economics, 22, 2, pp. 305–33.
Laurent, R.D. (1988), ‘An interest rate-based indicator of monetary policy,’ Economic Perspectives, Federal Reserve Bank of Chicago, January, pp. 3–14.