Modelling interest rates with a cointegrated VAR-GARCH model

We use a bivariate VAR model to model and predict the joint evolution of short term and long term interest rates. We introduce a GARCH effect on the innovations of the model in order to account for the changing volatility of the series. We test the cointegration of the two interest rates, which is implied by a present value relation between the rates. The cointegration test is done both with and without taking account of the GARCH effect. The empirical results for five countries (Belgium, Germany, France, Great Britain and the USA) point to the same conclusions: i) the incorporation of the GARCH part allows to conclude more clearly that a cointegration relation exists; ii) GARCH effects are quite present; and iii) the models are useful for short term predictions of interest rates.