Approximating multiple itô integrals with "band limited" processes

From MaRDI portal
Publication:3681667





In this work, one considers a family of processes \(n^{\epsilon}(t)\), the bandwidth of which tends to infinity as \(\epsilon\) \(\to 0\) and one studies the convergence properties of the process \(x^{\epsilon}(t)=\int^{t}_{0}\int^{\tau}_{0}n^{\epsilon}(\tau) L(\tau,s)n^{\epsilon}(s)d\tau ds\) where L is some kernel such that \(L(t,s)=0\) for \(s>t\); it is assumed that \(L+L^*\) is nuclear - so that \(L(t,s)=\sum^{\infty}_{1}\lambda_ iM_ i(t)K_ i(s)\) for \(s<t\)- and that the process \(n^{\epsilon}\) can be constructed by \(n^{\epsilon}(t)=y(t/\epsilon^ 2)/\epsilon\) where y is a stationary process which is, roughly speaking, either Gaussian, either strongly mixing - actually it is noticed that this assumption can be weakened. Then it is proved that \(\int^{.}_{0}n^{\epsilon}(s)ds\) converges weakly to a Wiener process and that the finite-dimensional distributions of \(x^{\epsilon}(.)\) converge to those of a double Itô integral plus a correction term; moreover, with some additional assumptions, the convergence of the whole process \(x^{\epsilon}\) is obtained in the Skorokhod space. As an application of this result, the authors consider a partially observed system consisting of a signal solution of a linear stochastic differential equation, which is observed through a linear noisy channel; replacing the observation white noise by a large bandwidth one, they deduce from the previous theory a weak approximation result for the likelihood ratio associated with this problem and they explain how it can be used in statistical tests. Finally, in the last section, the results are extended to the case of multiple integrals - of order greater than 2.











This page was built for publication: Approximating multiple itô integrals with "band limited" processes

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3681667)