Carlo Favero, Sebastian Vismara and Iryna Kaminska
The slope of the yield curve has decreased in the US and the UK over the last few years (Chart 1). This development is attracting significant attention, because the yield curve slope (i.e. the difference between longer term government bond yields and shorter term government bond yields) is a popular business cycle indicator, and a fall of longer term yields below shorter term yields (i.e. an ‘inversion’ of the yield curve) has historically been considered as a powerful signal of recessions, particularly in the US.
Chart 1: Slopes of the US and UK government bond yield curves (10-year yields minus policy rates)
Since Estrella and Mishkin (1998), a number of studies have documented that the slope of the yield curve has significant power in predicting economic slowdowns (see e.g. the recent survey by Claessens and Kose, 2017). However, some – including FOMC members (e.g. Chair Powell) – have suggested that the slope of the yield curve is likely to contain less signals on the growth outlook than it has done in the past due to a number of changes that have taken place in interest rates markets.
To answer the question, in this post we analyse how the yield curve has performed as a predictor of GDP growth over time in the US and the UK, focussing on the performance of the different components of the yield curve: expected short interest rates and term premia.
How has the yield curve slope performed as a predictor of GDP growth over time?
We start by comparing the predictive power of different measures of the US and UK yield slope for GDP growth one year ahead. Specifically, in a similar spirit to Stock and Watson (2003), we predict the annual growth in real GDP one year ahead with the relevant yield curve slope, controlling for the latest available outturns of annual growth rate in GDP in real time and two additional lags. In other words, we condition the GDP forecast 1-year ahead only on the information available at each point in time. We use quarterly data since 1966 for the US, and a shorter series for the UK – since 1982, because of data availability. Within this framework, we analyse forecasts produced using a rolling estimation scheme (with a starting window of 15 years), and compare them with the forecasts obtained from autoregressive models that are based only on lagged GDP growth variables.
We find that, historically, yield curve slopes contain some predictive power in forecasting the level of economic growth in the US and the UK, over and above what is predicted by lagged GDP growth. However, over time, and across different measures, the US and the UK yield curve slope have gradually lost some of their predictive power (Chart 2). In the UK, due to the shorter sample window, we cannot say whether there was a longer term trend in place as in the US case.
Chart 2: Adjusted R2 of different GDP forecasting models
Yield curve slopes measured relative to policy rates (i.e. Fed funds rate for the US and Bank rate for the UK), appear to do a better job at predicting GDP growth rates than other measures. And it appears that there is no statistical difference over the whole sample between using various measures of slope when measured relative to policy rates (we looked at the 1-year 1-year forward rate, the 5-year and the 10-year yields relative to policy rates and also at 10-year yields relative to 2-year yields).
These measures show that if we focus on the most recent period, the predictive power of the slope of the yield curve has been picking up, albeit marginally. Using a 15-year rolling estimation window and looking at the rolling coefficient of the spread between
5-year yields and the policy rate, we can see that it had become insignificant and decreased towards zero around the early to mid-2000s but has recently increased back to positive levels (Chart 3). In other words, a steeper (flatter) yield curve is again associated with higher (lower) future GDP growth.
Chart 3: Rolling coefficient on the spread between 5-year yields and the policy rate for one-year ahead GDP growth predictive regressions.
So where is this increase in predictive power coming from?
Decomposing the slope: is the predictive power in expected short interest rates or term premia?
To better understand where the change in the predictive power of the slope of the yield curve is coming from, we decompose interest rates into their components – expected short interest rates and term premia (the compensation investors require for holding longer-term bonds). We also use a real time decomposition of the yield curve slope (Similar to Carriero, Favero and Kaminska 2004), allowing us to capture the true real time investor views on the state of the economy. This is in contrast to much of the previous work which has used in-sample forecasting performance (see, among others, Ang, Piazzesi, and Wei (2006)). This method assumes that the forecaster knows the full-sample information on the term structure of interest rates (i.e. including future yield curves) and so does not capture the true real-time investor views on the state of the economy. In addition, as Rudebusch and Williams (2009) show, yield curve slope models for predicting recessions that only use real-time yield curve information perform better than other models based on in-sample decompositions. We follow a similar approach.
Therefore, we focus on the out-of-sample forecasts and a decomposition of yields into expected short interest rates and term premia based on a term structure model by Adrian, Crump, and Moench (2013). To obtain the monthly decompositions in real time, we estimate the model recursively, starting with the sample 1985-1995 for both the US and the UK.
Our results show that the policy rate expectations component of the slope tends to be positively correlated with future GDP growth over the whole sample (Chart 4). The decomposition also sheds light on the causes of the disappeared predictive power of the slope in the 2000s. During the crisis, when policy rates reached the zero lower bound and were expected to stay there, the policy rate expectations component of the slope was unable to predict the magnitude of the recession and ceased to be informative about one-year-ahead GDP growth. The timings and the patterns of the coefficient loadings are remarkably similar in the UK and US. More recently, the expectations components seem to have regained some of their predictive power, with the loadings becoming significant and higher although the uncertainty bands are wider than in the pre-crisis period.
Chart 4: Rolling regression coefficients and the term premia and expected slope components of the spread from
The relationship between term premia and growth appears to be more idiosyncratic. In the US, the contribution of term premia was negative and significant prior to the crisis. There are several reasons why that might be the case. Around the turn of the century, the search for safe assets by foreign official investors became a key factor in the US treasury market, contributing to the reduction in US real term premia prior to the global financial crisis (see Kaminska, Vayanos and Zinna (2011) and Kaminska and Zinna (2018)). Therefore prior to the crisis, US term premia seemed to offset the positive and significant signal about GDP growth provided by the expectations component of the yield curve and hence tarnished the predictive power of the slope as a whole. Recently, the role of US term premia has changed, becoming positive and significant in predicting GDP growth. This may partly reflect QE, which has an expected pro-cyclical impact on the term premium via the portfolio balance and signalling channels. Instead, in this simple forecasting regression, the UK premium has been largely insignificant throughout the sample, although (as in the US case) the predictive relationship has become positive in recent years.
In sum, since the 90s, the slope of the yield curve has indeed become less efficient in predicting GDP growth over time in both the US and the UK. More recently however the yield curve slope appears to have regained some of its predictive power, as policy rate expectations are again positively and significantly associated with future GDP growth while the term premia has stopped obscuring their signal. This suggests that while the warning sign from the yield curve slope should be taken with caution, it should not be discarded completely.
Carlo Favero works for Bocconi University, Sebastian Vismara works in the Bank’s Market Intelligence and Analysis Division and Iryna Kaminska works in the Bank’s Monetary and Financial Conditions Division.
If you want to get in touch, please email us at firstname.lastname@example.org or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.