Recursive Relations for Predictors of Non-Stationary Processes
- 1 September 1965
- journal article
- research article
- Published by Oxford University Press (OUP) in Journal of the Royal Statistical Society Series B: Statistical Methodology
- Vol. 27 (3) , 523-532
- https://doi.org/10.1111/j.2517-6161.1965.tb00612.x
Abstract
It is shown that if a process {xt} is generated recursively by a relation of type where ajt, bkt and λt, are time‐dependent sequences and {ϵt} is a process of unautocorrelated variables, then in general the linear least‐square predictor of xt+m based upon {xs; s≤t}, denoted vt, obeys a recursion of the type where r = max(p – 1, q – m). A method is given for obtaining the coefficients βkt, cjt and vt, and this solution is made explicit for various special cases.This publication has 0 references indexed in Scilit: