Disentangling the Predictive Power of Term Spreads under Inflation Targeting

Abstract

Under inflation targeting, S. Cho and J. Lee (2014, Inflation targeting and predictive power of term spreads. Seoul Journal of Economics, 27, 391–419), A. Estrella (2005, Why does the yield curve predict output and inflation? The Economic Journal, 115, 722–744) and P.L. Siklos (2000, Inflation targets and the yield curve: New Zealand and Australia versus the US. International Journal of Finance & Economics, 5, 15–32) have reported that the predictive power of yield spreads for future inflation decreases in inflation targeting countries. In this paper, we decompose the yield spread into the expectations hypothesis component and the term premium, and find that the decrease in the predictability is mostly due to the deterioration of information embedded in the expectations hypothesis component. Our finding reveals that if inflation targeting is successful in achieving its main goal, then the expectations for future inflation are anchored at the target inflation rate (or range), and thereby the predictive contents of the term spreads regarding future inflation decrease.

Key Words:

• Inflation targeting
• expectations hypothesis
• term premium
• term spread
• predictive power

JEL Classifications:

• E52
• E58
• G12

Notes

1 Support for the EH also exists, such as in Della Corte, Sarno, and Thornton (2008), Gerlach and Smets (1997), Longstaff (2000) and Shen (1998).

2 For instance, Rudebusch et al. (2007) compare the five TP measures of the US long-term bond yield with a 10-year maturity. They show that these five measures had a similar trend, but that their levels and volatilities were distinctly different from each other.

3 Equation (2)(2) is also estimated with the whole sample, not to be separated into two subsamples, but even in this case, the final result of this paper concerning the predictive power of the two components does not change significantly.

4 More precisely, in the case of the UK, the yield curve data published by the Bank of England are used instead of the actual yield data because the historical data of the government bond yields are supplied as 5-year, 10-year, and 20-year maturities.

5 The empirical results do not change significantly when s are assumed to be lower-triangular matrices.

6 The forecasting equation of Cho and Lee (2014) is , which is the restricted form of equation (11)(11) .

7 Cho and Lee (2014) consider the VAR model with the vector in the corresponding analysis.

8 Cho and Lee (2014) show that the estimates of the term spread coefficient are significantly positive over 9-to-20-quarter forecast horizons in period 1. However, contrariwise, those turn out to be significantly negative over 7-to-19-quarter forecast horizons in period 2.

9 Regarding the fitness of the estimated equation, it is mostly much higher when the two components are set to be independent variables than when the yield spread is included in the estimated equation. In the case of the equation for inflation in the US, the average adjusted R-squared in period 1 is only about 0.26 when the term spread is the independent variable, but it rises to about 0.39 when the two components are independent variables. Additionally, in Canada, the statistic sharply rises from about 0.1 to about 0.4 if the yield spread is separated into two components. Ang et al. (2006) and Favero et al. (2005) also report similar results from the equation for future real activities.