Academic journal article Federal Reserve Bank of New York Economic Policy Review

Signal or Noise? Implications of the Term Premium for Recession Forecasting

Academic journal article Federal Reserve Bank of New York Economic Policy Review

Signal or Noise? Implications of the Term Premium for Recession Forecasting

Article excerpt

* Since the 1970s, an inverted yield curve has been a reliable signal of an imminent recession.

* Some have argued that the yield curve inversion in August 2006 did not signal a recession because it was driven by an unusually low level of the term premium rather than by changes in interest rate expectations.

* If the predictive power of the yield curve signal comes from interest rate expectations, then a forecasting model that separates this component from the term premium may be more accurate than the standard model.

* A comparison of forecasting performance finds that a model without the term premium provided recession signals similar to those of the standard model since the 1970s, but recently the two models' signals have diverged.


Ongoing efforts to find the best method for predicting recessions leave many questions unresolved. Existing papers propose a variety of indicators and modeling techniques, yet overall forecast accuracy has been mixed (see, for example, Stock and Watson [2003]). One approach with an excellent track record is the term spread model of Estrella and Hardouvelis (1991).1 When the yield on a three-month Treasury bill rises higher than the yield on a ten-year Treasury note, the model forecasts that a recession will begin twelve months in the future.

Why should a negative term spread predict a recession? The expectations hypothesis posits that long-term interest rates are determined by expected future short-term rates. Because short-term rates are governed by monetary policy, investors should expect declines as a phase of monetary tightening transitions to monetary easing. As expected future short-term rates fall below current short-term rates, the yield curve inverts. Estrella and Adrian (2008) show that the yield-curve inversion that comes at the end of a tightening cycle has historically been followed by a decline in real activity, which provides a compelling link between yield-curve inversion and an imminent recession.

However, a large body of literature (such as Dai and Singleton [2002]) shows that the expectations hypothesis does not provide a complete explanation of yield-curve behavior. In particular, yields also depend on the maturity of securities: Longer term Treasury securities are riskier and require a premium to compensate for this extra risk. These term premia vary over time as interest rate risk and investors' risk tolerance fluctuate.

Normally, the term premium provides a buffer that prevents minor variations in interest rate expectations from inverting the yield curve. But when the term premium is small and the yield curve is relatively flat, this buffer disappears. Previous research does not find a strong link between low term premia and recessions (see, for instance, Hamilton and Kim [2002] and Rudebusch, Sack, and Swanson [2007]), so this occasional sensitivity to small changes in expectations may reduce the accuracy of recession forecasts.

In August 2006, the yield curve inverted and the Estrella and Hardouvelis (1991) model predicted that a recession would begin in August 2007.2 This event drew renewed attention to the term spread recession forecasting approach. However, the term premium had also fallen to an unusually low level, which raised concerns that the observed yield-curve inversion might not in fact indicate an impending recession (see, for example, Dudley [2006]).

This article investigates whether changes in the term premium tend to distort the term spread's recession signals.3 We begin by decomposing the term spread into an expectations component and a term premium component, based on the Kim and Wright (2005) term premium estimates. Next, we construct recession forecasting models based on these components, following the approach of Estrella and Hardouvelis.

We find that the expectations component model is similar to the standard term spread model: Both models accurately predict all six recessions since 1961 when the signal threshold is set to a twelve-month recession probability of 25 percent. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.