University of Edinburgh
School of Mathematics and BioSS
Date: Friday 19th May, 15:05 Location: JCMB 6201
Speaker: Agniezka Borowska, Vrije Unviestiy of Amsterdam
Title: Bayesian Risk Evaluation using Importance Sampling
Abstract:
An accurate and efficient approach to Bayesian estimation of two financial risk measures, Value at Risk and Expected Shortfall, for a given volatility model, is presented. The key insight behind the proposed importance sampling based method is the construction of the importance densities as mixtures of Student's t distributions sequentially. By oversampling the extremely negative scenarios and punishing them by lower importance weights, a much higher precision in characterising the properties of the left tail is achieved. Two methodologies are developed for two distinct applications: first, to long run risk evaluation for observation driven models; second, to short-run risk evaluation for parameter driven models. The former are more tractable and thus commonly applied by practitioners. Here the focus is on precise forecasting of the tail of the distribution of returns not only for the standard 10-days-ahead horizon but even for long ones, like one-month or one-year ahead. The latter are flexible and theoretically sound, yet typically do not allow for closed form expressions for the likelihood. Here, the emphasis is on augmenting the parameter space by the latent volatility factor and modelling the augmented parameter subspace corresponding to high losses. For both methodologies substantial accuracy gains are reported in empirical studies on financial daily logreturns.
This seminar is a part of Maxwell Institute seminar series.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|