Title: Generalized autoregressive score models for conditional quantiles
Authors: Petra Tomanova - University of Economics, Prague (Czech Republic) [presenting]
Abstract: New specifications of conditional quantile models for value at risk are proposed. Our approach is based on Generalized Autoregressive Score (GAS) framework, also known as Dynamic Conditional Score (DCS) models, allowing parameters to vary over time and capturing the dynamics of time-varying parameters by the autoregressive term and the scaled score of the conditional observation density. Individual conditional quantiles are estimated directly without imposing distributional assumptions on returns. We compare our models to the originally proposed specifications of Conditional Autoregressive Value at Risk models. We pay a special attention to the estimation procedure. We compare various techniques how to initialize the estimation and address the issue of finding reasonable starting values since the optimization surface related to our problem is challenging. We also focus on assessing different link functions to derive models for conditional quantiles from GAS models, an approximation of quantiles of student t distribution so the estimation of conditional quantiles is computationally tractable, and other related computational issues.