A generalized least squares estimation method for invertible vector moving average models

We propose a new GLS procedure for estimating VMA models. Its main feature is to consider the stochastic structure of the approximation errors arising when lagged VMA innovations are replaced with lagged residuals from a long VAR.