The predictive content of the term premium for GDP growth in Canada: Evidence from linear, Markov-switching and probit estimations

The objective of this study is to assess the predictive content of the term premium for future GDP growth in Canada over the past 55 years. The term premia for average long-term bond yields for four maturity categories are derived using the stationary vector-stochastic process for the expectation hypothesis of Campbell and Shiller (1987, 1991). The bivariate VAR model identifies term premia for the bond yields that are mean-stationary and increase on average to about 100 basis points. The coefficients on the expectations components of the term spreads in linear estimations are found to be equal to those on the term premium components for all maturity categories, suggesting that there is no addition information about future GDP growth in the term premium components. However, regime-switching estimations that stochastically divide the sample period into high- and low-variance regimes indicate that the equality of the coefficients on both components of the term spreads can easily be rejected for all maturities in both regimes. The expectations component is found to be more important than the term premium component when future output growth is volatile, while it is not found to be important during 'normal times' when the variability of GDP growth is relatively low. Probit model estimations suggest that the expectations component of the term spread (removing the term premium component) produce relatively more reliable predictions of future recessions than the term premium component, which has only marginal information for predicting future recessions.