Structural characterization of taboo-stationarity for general processes in two-sided time. (Q2574531)

Let \(E\) be a Polish space, \(X^ * = (X^ *(t),\, t\in \mathbb R)\) an \(E\)-valued stochastic process in two-sided time, whose trajectories belong to the Skorokhod space \(D(\mathbb R;E)\). For \(t\in \mathbb R\), let us denote by \(\theta _ {t}X^ *\) the shifted process, \(\theta _ {t}X^ * = (X^ *(t+s),\, t\in \mathbb R)\). Let \(\varGamma ^ *\) be a nonnegative finite random time. The pair \((X^ *,\varGamma ^ *)\) is called taboo-stationary, if \(\mathbf P((\theta _ {t}X^ *,\varGamma ^ * -t)\in \bullet \mid \varGamma ^ * >t) = \mathbf P((X^ *, \varGamma ^ *)\in \bullet )\) for all \(t\geq 0\). First it is shown that \((X^ *,\varGamma ^ *)\) is taboo-stationary if and only if it arises as a total variation taboo limit, that is, if and only if there exist a process \(X = (X(s),\, s\geq 0)\) with paths in \(D(\mathbb R_ {+}; E)\) and a nonnegative finite random time \(\varGamma \) such that \(\mathbf P((X_ {t-h}, \varGamma -t)\in \bullet \mid \varGamma >t)\to \mathbf P((X^ *_ {-h}, \varGamma ^ *)\in \bullet )\) in the total variation norm as \(t\to \infty \) for all \(h\geq 0\). Then an interesting simple characterization of taboo-stationarity is given: \((X^ *,\varGamma ^ *)\) is taboo-stationary if and only if \(\varGamma ^ *\) is exponential and independent of \(\theta _ {\varGamma ^ *}X^ *\).