Efektywność predykcji zmiennych zakłócających

Abstract

The author Investigates the efficiency of prediction of noise variables in two cases, namely when these variables are generated by causal models and by autoregressive models. In the first case as the values of (random) explanatory variables in the horizon of prediction are not known, it is assumed that their extrapolated values are used instead. It is proved that the efficiency of such prediction is always smaller than the efficiency when autoregressive models are used.