Causalité de long terme et amélioration de la prévision : application aux courbes de taux d'intérêt

We propose a definition and a characterization of long-run causality between non-stationary, possibly cointegrated, series. In a VAR framework, a Wald test can be performed to test for long-run non-causality, with the statistics distributed as a chi-square, conditionally on the cointegration rank. This methodology is used to study long-run causal links between the interest rates of euro-currency term structures, for different countries, between 1983 and 1996. Contrary to the implications of the expectations hypothesis, the interest rates of different maturities do not play a symmetrical role, when contributing to the common trend of the yield curve. The common trend appears to be mainly led by longest term rates, which can be used as an indicator of market participants expectations about future monetary policy.