Überprüfung stochastischer Modelle mit Pseudo-Residuen

TL;DRAbstract

In this thesis we investigate the application of pseudo-residuals for assessing the fit of probability models. Pseudo-residuals enjoy three important advantages over traditional residuals: They are applicable to a wide variety of different types of models. For discrete-valued observations they are defined as intervals, thereby avoiding the approximations associated with point residuals. Their exact distribution under the model is known (when the model parameters are assumed to be known). This enables one to objectively examine them either jointly (to assess the general fit of the model) or individually (to identify outliers). Pseudo-residuals have a meaningful and simple interpretation, and are easy to compute using standard statistical software. Their application is considered in detail, and illustrated, in the contexts of generalized linear models, financial time series models and multivariate distributions. In each case it is shown how the analysis of pseudo-residuals provides detai

Chat with Paper

AI Agents for this Paper