On a characterization of optimal predictors for nonstationary ARMA processes (Q2640300)

The authors deal with a nonstationary ARMA process \(\{y_ t\}\) given by \(A_ k(q^{-1})y_ k=C_ k(q^{-1})e_ k,\) where \(q^{-1}\) is the back-shift operator, \(A_ k(q^{-1})=1+a_ 1(k)q^{-1}+...+a_ n(k)q^{-n}\), \(C_ k(q^{-1})=1+c_ 1(k)q^{-1}+...+c_ n(k)q^{- n}\), and \(\{e_ k\}\) is an orthogonal-valued process such that E \(e_ k=0\), \(0<m\leq E | e_ k|^ 2\leq M\). It is assumed that there exist limits lim \(a_ j(k)=a_ j\), lim \(c_ j(k)=c_ j\) as \(k\to - \infty.\) An equation for the h-step predictor \(\hat y_{k+h| k}\) is derived and a weak law of large numbers is proved.