Predicting non-stationary processes (Q2425399)

The authors are concerned with a problem of prediction of a probability measure given another one, when both are living on the space of one-way infinite finite-alphabet sequences. This means, if a sequence was considered to be generated by the first measure, when do its conditional probabilities w.r.t. the second reference measure converge in some sense. The main result are sufficient local absolute continuity criteria assuring prediction and the generalization of several notions to identify classes of measures for which prediction turns out to be feasible.