Lévy integrals and the stationarity of generalised Ornstein-Uhlenbeck processes (Q2568302)

Let \((\xi,\eta)=(\xi_t,\eta_t)_{t \geq 0}\) be a bivariate Lévy process. The generalised Ornstein-Uhlenbeck process \((V_t)_{t \geq 0}\) is defined as \(V_t = e^{- \xi_t} (\int_0^t e^{\xi_{s-}} \,d\eta_s + V_0)\), \(t \geq 0,\) where \(V_0\) is a finite random variable, independent of \((\xi_t,\eta_t)_{t \geq 0}.\) The relationship between the stationarity of the process and the convergence of the Lévy integral in the general case when \(\xi\) and \(\eta\) are not necessarily independent is investigated. The authors precise this relation in the general case, showing that the conditions are not in general equivalent, though they are for example if \(\xi\) and \(\eta\) are independent. Characterizations are expressed in terms of the Lévy measure of \((\xi,\eta)\). Sufficient conditions for the moments of the strictly stationary distribution to be finite are given, and the autocovariance function of \((V_t)_{t\geq 0}\) is calculated and shown always to decrease exponentially with the lag. The authors also consider the tail behaviour of the stationary solution \(V_{\infty}\) showing that it has heavy (Pareto-like) tails under some conditions. Some discussion on related results and examples are given.