Identification of Finite Dimensional Linear Systems Driven by Levy processes

TL;DR

This paper develops a novel method for identifying finite-dimensional linear systems driven by Levy processes using the empirical characteristic function, focusing on noise characterization via characteristic functions rather than densities.

Contribution

It introduces an adapted empirical characteristic function method for Levy system identification, providing error analysis and asymptotic efficiency results.

Findings

Proposed a new identification method based on empirical characteristic functions.

Derived error bounds and asymptotic covariance matrices for the estimators.

Demonstrated the asymptotic efficiency of the method in Levy process contexts.

Abstract

Levy processes are widely used in financial mathematics, telecommunication, economics, queueing theory and natural sciences for modelling. A typical model is obtained by considering finite dimensional linear stochastic SISO systems driven by a Levy process. In this paper we consider a discrete-time version of this model driven by the increments of a Levy process, such a system will be called Levy system. We focus on the problem of identifying the dynamics and the noise characteristics of such a Levy system. The special feature of this problem is that the statistical description of the noise is given by the characteristic function (c.f.) of the driving noise not by its density function. As an alternative to the maximum likelihood (ML) method we develop and analyze a novel identification method by adapting the so-called empirical characteristic function method (ECF) originally devised for estimating parameters of c.f.-s from i.i.d. samples. Precise characterization of the errors of these estimators will be given, and their asymptotic covariance matrices will be obtained. We also demonstrate that the arguments implying asymptotic efficiency for the i.i.d. case can be adapted for the present case.