Inverse covariances of a multivariate time series (Q1067333)

By analogy with the univariate case the author introduces the inverse covariance function of a multivariate time series with spectral density matrix F(\(\lambda)\). Under the assumption that the spectral matrix \(F^{-1}(\lambda)\) of inverse processes exists the author derives a formula of linear interpolation or reconstruction of skipped values for the general case and multivariate ARMA models. The suggested constructions turn out to be equivalent to minimization of the variance of a certain linear filter. The problem of estimation of inverse covariances is also discussed.