Fractional Lévy processes with an application to long memory moving average processes (Q2642806): Difference between revisions

Let \(\{ L(t) : t \in\mathbb R \}\) be two-sided Levy process on \(\mathbb R\) with \(EL(1)=0 , E[L(1)]^2 < \infty \) and without Brownian component. For fractional integration parameter \(0<d<0.5\) and \(t \in\mathbb R\), \[ M_d(t) = \frac{1}{\Gamma (d+1)}\int_{\mathbb R} [(t-s)^d_+ - (-s)^d_+] \,L(ds) \] is called a fractional Levy process (FLP). The author presents different methods for constructing FLP: as an integral w.r.t. a Poisson random measure; as having a continuous modification which is an improper Riemann integral; by using series representation for Levy processes. The latter construction can be used for simulations of FLP. The second order structure and sample path properties of FLPs are also studied in the present paper. FLPs exhibit long memory and they are not always semimartingales. Thus the author developes a stochastic calculus w.r.t. FLPs. She investigates moving average (MA) processes with the long memory property, and states that the Levy-driven long memory MA process with fractionally integrated kernel has a MA integral representation where the integrand is not fractionally integrated and the driving process is a FLP.