On the use of variograms for the prediction of time series (Q1060798)

Let \(\{y_ t\), \(t\geq 0\}\) be a discrete stochastic process, with mean \(E(y_ t)=m\). Define the variogram of the process to be \(\gamma (t,t- \tau)=E\{(y_ t-y_{t-\tau})^ 2\},\) assumed to be stationary, i.e., \(\gamma (t,t-\tau)=\gamma (\tau).\) The paper derives different minimum variance unbiased expressions for \((d+1)\)-step ahead predictions of the \(\{y_ t\}\) process based on variograms, under a variety of assumptions.