= Noncausal stationary autoregression
{title2=$X_t=-\sum_{j\geq1}\phi^{-j}\varepsilon_{t+j}$}
A <weakly stationary process> satisfying an <autoregressive model> can depend on future driving <white noise>. For $X_t=\phi X_{t-1}+\varepsilon_t$ with $|\phi|>1$, the convergent in the sense of <mean-square convergence> solution is $X_t=-\sum_{j\geq1}\phi^{-j}\varepsilon_{t+j}$. Its <autocovariance> is $\sigma^2\phi^{-|h|}/(\phi^2-1)$. This is different from a <causal time series>.
Back to article page
