The thesis develops a new approach to generate long memory models by defining the class of fractional Lévy processes (FLPs) and investigates the probabilistic and sample path properties of FLPs. As for a fairly large class of Lévy measures the corresponding FLP cannot be a semimartingale, classical Ito integration theory cannot be applied. In the thesis we give a general definition of integrals with respect to FLPs. This integration theory is then applied to continuous time moving average processes in the sense that the driving Lévy process in the moving average integral representation of short memory processes is replaced by a FLP. It turns out, that the so-constructed process exhibits long memory properties. But an even more important result is that this process coincides with the moving average process (driven by the ordinary Lévy process) which is obtained by a fractional integration of its kernel function. This is a new method to generate fractionally integrated continuous time ARMA (FICARMA) processes. So far only univariate CARMA and FICARMA processes have been defined and investigated. In the second part of the thesis multivariate analogues of both models are developed by constructing a random orthogonal measure which allows for a spectral representation of the driving Lévy process. Furthermore, the probabilistic properties of multivariate CARMA and FICARMA models are studied. Like in the univariate case, the multivariate FICARMA process has two kernel representations which lead to the same model.
«
The thesis develops a new approach to generate long memory models by defining the class of fractional Lévy processes (FLPs) and investigates the probabilistic and sample path properties of FLPs. As for a fairly large class of Lévy measures the corresponding FLP cannot be a semimartingale, classical Ito integration theory cannot be applied. In the thesis we give a general definition of integrals with respect to FLPs. This integration theory is then applied to continuous time moving average proces...
»