Vlastnosti generalized autoregressive score modelů na konečných vzorcích

This thesis discusses the finite sample properties of generalized autoregressive score models (GAS). It provides a theoretical background to the area of time-varying parameter models as well as a description of GAS model and its properties. It shows the connection between GAS models and previously developed time-varying parameter models such as GARCH, ADC, and ADI models. Furthermore, it presents a simulation study of 29700 samples for the estimated GAS models on simulated data, using the R package gasmodel, across various probability distributions and model parameters. The simulation study results then empirically show the consistency and precision increasing with the sample size of some of the presented GAS models. The cases where the models have not converged are outlined, and a possible explanation is discussed.