Prediction with incomplete past of a stationary process. (Q1766054): Difference between revisions

Let \((X_k)\) be a zero mean weakly stationary stochastic process. Let \(\hat X_0\) and \(\hat X_0'\) be the best linear mean square predictor of \(X_0\) based on \(\{X_k, k\leq -1\}\) and on \(\{X_k,\;k\leq -1,\;k\neq -n_1,\dots ,-n_N\}\), respectively. The author introduces formulas for \(X_0-\hat X_0'\) and for \(\text{var}(X_0-\hat X_0')\). The autoregressive representation of \(\hat X_0'\) is derived and the processes where \(\hat X_0=\hat X_0'\) are characterized. Some properties of \(\hat X_0'\) when \((X_k)\) is either an AR\((p)\) process or an MA\((q)\) process are established.