Efficient closed-form estimation of multivariate moving-average processes using higher order statistics

An improved algorithm is proposed for estimation of non-Gaussian, nonminimum phase, multivariate moving average (MA) processes using higher order cumulants. This algorithm improves upon earlier results and contains development beyond existing algorithms. It provides a closed-form solution to estimating the MA parameter matrices (up to a post-multiplication by a permutation matrix), and (under certain assumptions) eliminates the indeterminacy associated with scaling. The algorithm is theoretically derived and tested via computer simulations. In addition, it is shown that this algorithm is computationally more efficient than the one proposed by Tong, Inouye, and Liu (see ibid., vol.40. no.10, p.2547-2558, 1992). Finally, the effect of imperfect input data on our algorithm is tested via simulations