Predicting Recessions at the Zero Lower Bound

Ralf Fendel - April 14, 2021

Tips for practitioners

Prior to the Covid-19 pandemic that triggered a recession that was impossible to predict, the Eurozone experienced a rather classical economic cycle characterized by the build-up of imbalances and overheating before the Global Financial Crisis, followed by a structural weak economy lingering in a liquiditytrap thereafter. Observers have witnessed different monetary regimes ranging from orthodoxy-induced monetary policy to unconventional monetary policy of the ECB with the main refinancing rate pinned down at the so-called zero lower bound (ZLB). In addition, the brief existence of the Eurozone makes it unique compared to other currency areas. These particularities of the Eurozone require that predictors of recessions need to be robust enough to detect the state of economic cycles irrespective of central bank policies.

The yield curve as a recession predictor

Traditionally the yieldcurve has been among the most reliable indicators for predicting recessions. According to the expectation theory of the term structure, a traditional upward-sloping yield curve becomes flatter and may even inverts its slope before the inception of a recession. This phenomenon is usually expected because the central bank hikes up the short-term interest rates to tame inflationary pressure of a (still) booming economy while long rates simultaneously decline on account of financial markets’ expectations of an impending slowdown or recession and a more accommodative monetary policy in future. In the empirical literature, the described theoretical phenomenon has been supported convincingly, and the slope of the yield curve (modelled by term spread) has been associated with paramount predictive power for the future state of the business cycle.

At the ZLB, however, the shape of the yield curve as a predictor of the future state of the business cycle is compromised by at least two developments. First, this is due to the effect of unconventional monetary policy such as forward guidance and quantitative easing on the yield curve. Specifically, on account of the unconventional monetary policy, financial market expectations has lowered the interest rates along the entire term structure. While the conventional calibration of recession models using the term spread would be signaling elevated probabilities of an economic downturn, the cause for the flattening is however the recently implemented unconventional monetary policy. Second, the increased relevance of the market segmentation hypothesis at the ZLB alters the dynamics of the yield curve. At the ZLB, bond purchases by the ECB have an asymmetrically stronger effect on the long end of the yield curve than on the short end.

The declining predictive power of the traditional term spread 

In a recent publication, me and my co-authors include 43 variables to analyze the recession probabilities for the Eurozone. A benchmark model based on univariate (i.e., single-variable) probit regressions is estimated and complementary extensions are developed to find reliable predictors of Eurozone recessions. Figure 1 shows the predictive power of the traditional term spread (black line: Difference between the 10year interest rate and the 3months interest rate) by plotting the signaled recession probability as indicated by the respective indicator. The two recessions are indicated by the shaded areas. While the recession of 2008-2009 still serves as a prime example of the term structure’s ability to forecast an economic downturn with sufficient lead-time, the indicator fails to capture the second recession beginning in 2011. On the contrary, the signal inches north in the aftermath of the European Sovereign Debt Crisis and remains elevated for the time period between 2015-2017 in the wake of a flattening of the yield curve induced by the Expanded Asset Purchase Program by the ECB starting in 2015.

A reformulation of the term spread

In an attempt to restore the informational content of the yield curve, we suggest a modified version of the term spread that incorporates the so-called shadow rate as a substitute for the observed short-term interest. The shadow rate is an artificial short-term rate that aims to incorporate the unconventional monetary policy measures. It is estimated as a linear function of three latent variables called factors, which follow a first-order autoregressive process. As quantitative easing at the ZLB can be perceived as a substitute for further rate cuts, this revised indicator is composed of a variable front-leg that switches from the 3m Euribor to the shadow rate as soon as the ECB’s policy rate hits the ZLB. The modified term spread is shown as the blue line in Figure 1. Indeed, the ECB’s unconventional monetary policy leads to a “shadow” steepening of the yield curve as the short end of the yield curve turns deep into negative territory. Hence, the average recession probability for the period between January 2015 and December 2017 recedes to 1 percent compared to about 40 percent as indicated by the traditional term spread. The modified term spread even indicates higher probability of recession in months running before the two recessions.  

Alternatives to the term spread itself

In our research, we also present alternative indicator to the term spread indicator. The best performing indicator in our analysis is the growth rate of narrow money (12m change of real M1). In the run-up to both recessions, the indicator crosses the 50 percent threshold probability 12 months before the start of each economic downturn. A cohort of another better performing variables is composed of indicators from sentiment indices, the so-called purchasing manager index (PMI), refinancing conditions [Figure 4], and measures of competitiveness like the terms of trade and the exchange rate. All those indicators were able to signal the recession.  

n our research, we also present extensions of the univariate model and show that combinations of several indicators can be used to form an indicator with better predictive power. This is done by extracting the first principal component for categories representing several specific transmission channels. The idea behind this approach is not necessarily to outperform the results of the benchmark model, but rather finding a specification that removes idiosyncratic vulnerabilities of single variable-based indicators. A further modification adds the relative state of a variable versus its long-term trend to identify what Keynes termed “Animal Spirits”. The subsequent Figure 6 for example shows the predictive power of PMI from Figure 3 can be improved by adding its derivation from its trend. In particular, the first recession of the Eurozone could have been predicted earlier and with higher precision.

Original publication

• Fendel, R./Mai, N./Mohr, O. (2020): Recession probabilities for the Eurozone at the zero lower bound: challenges to the term spread and rise of alternatives, forthcoming in: Journal of Forecasting.

Further reading

• Chauvet, M./Potter, S. (2005): Forecasting recessions using the yield curve, in: Journal of Forecasting 24(2), p. 77-103.
• Estrella, A./Hardouvelis, G. A. (1991): The term structure as a predictor of real economic activity, in: The Journal of Finance 46(2), p. 555-576.
• Rudebusch, G. D./Williams, J. C. (2009): Forecasting recessions: The puzzle of the enduring power of the yield curve, in: Journal of Business & Economic Statistics 27(4), p. 492-503.
• Wu, J. C./Xia, F.D. (2016): Measuring the macroeconomic Impact of monetary policy at the zero lower bound, in: Journal of Money, Credit, and Banking, 48(2-3), p. 253-291.