Impulse response analysis in infinite order cointegrated vector autoregressive processes (Q1372925)

Some general asymptotic results for infinite-order cointegrated VAR processes are derived, e.g. the joint asymptotic distribution of the estimated VAR coefficients and the white noise covariance matrix including a consistent estimate of the covariance matrix of this joint asymptotic distribution. These results enable to investigate asymptotic properties of estimated instruments which are used for interpreting VAR models: impulse responses, forecast error variance decompositions, persistent profiles (in general, these instruments are nonlinear functions of the VAR coefficients). The asymptotic approach used in the paper is based on the assumption that finite-order VAR processes are fitted to the observed time series but the order of the fitted process is assumed to increase with the sample size. The results are illustrated by a real data analysis of a German money demand system relating \(M1\), GNP and a short-term interest rate.