Let Xt be a q-variate
stationary stochastic process. Let K be any set of t-values and let K′ be the
complement of K. If s ∈ K′ the problem of approximating Xs by linear combinations
of the Xt′s with t ∈ K and limit of such linear combinations is considered. The best
linear predictor and the mean square error matrix are evaluated in the following
cases: (1) t takes on all real values, K consists of the integers (2) t is interger-valued,
K consists of the odd integers.