Stochastic integrals in additive processes and application to semi-Lévy processes (Q1431089)

An \(\mathbb{R}^d\)-valued stochastic process \(X_t\) is called additive if it has independent increments and is continuous in probability. (It is not required to have stationary increments as a Lévy process.) Its law \(\mu_t\) admits a Lévy-Khinchin representation, with generalized drift term, say \(\gamma= (t\mapsto \gamma_t)\). \(X_t\) is known to be a semi-martingale when \(\gamma\) has a locally bounded variation. The author shows first that this happens if and only if \(\widehat\mu_t\) admits a factorization \[ \widehat\mu_t(z)= \exp\Biggl[\int^t_0 \log(\widehat\rho_s(z)\,\sigma(ds)\Biggr], \] where (mainly) \(\sigma\) is some atomless Radon measure and \(\rho_s\) is some infinitely divisible law on \(\mathbb{R}^d\). This happens in particular if \(X_t\) is self-similar. Such so-called ``natural'' additive processes \(X_t\) are then shown to be in one-to-one correspondence with random measures mapping disjoint sets on independent variables, which leads of course to a simple construction of the stochastic integrals \(\int f(s)\,dX_s\) (with deterministic integrands \(f\)). Finally, the author shows (under the hypothesis that \(X_t\) is natural, has a finite log-moment, and has periodically stationary increments) that for any positive definite matrix \(Q\), the integral \(\int^\infty_0 e^{-sQ} \,dX_s\) converges.