A discussion on the Wiener-Kolmogorov prediction principle with easy-to-compute and robust variants

The Wiener-Kolmogorov principle of minimizing the mean square estimation error is discussed in the framework of prediction theory, from both theoretical and practical points of view. Alternatives for suboptimal solutions, more easily computable and not requiring the explicit knowledge of the covariance function, are proposed. A robust version of the principle is also discussed and illustrated by a small numerical example. A plan for future work is suggested.