Title: Estimating non-causal VARs using multivariate all-pass filters
Authors: Bernd Funovits - University of Helsinki (Finland) [presenting]
Abstract: An estimation strategy is proposed based on multivariate all-pass filters for a possibly non-causal VAR system driven by non-Gaussian i.i.d. shocks. Multivariate all-pass filters are matrices whose elements are rational functions and which preserve the second moment properties of the solution of the (transformed) VAR system. Applying a (non-trivial) all-pass filter to a non-Gaussian i.i.d. process results in an uncorrelated (white noise) process which is, however, not independent across time. First, we obtain the finite set of VAR systems whose solutions have the same second moment properties but whose determinantal roots are different. Subsequently, we minimize an objective function (using higher moments) which is minimal for the true VAR system.The obtained estimator can serve as an initial estimate for maximum likelihood estimation.The algorithm is implemented in the R-package varAllPass.