University of Auckland

The DIC as a model comparison criterion for Stochastic volatility models

Email address: andreas@stat.auckland.ac.nz

Abstract:
Due to computational progress and the recognition of MCMC methods in Bayesian statistics the demand for promising model selection criteria has grown continually.A recently developed proposal by Spiegelhalter, Best and Carlin (1998) is the DIC (Deviance Information Criterion), which is distinguished by simplicity in calculation and has a wide range of applicability. Although it is strongly related to the AIC, treated in the context of a Bayesian analysis, the DIC stands out with its flexibility for many kinds of models. The problem of identification of the number of parameters in complex models which is required for BIC and AIC applications has been elegantly overcome with the introduction of the principle of effective number of parameters.This model comparison criterion has already been applied to demanding models in the field of medical statistics (Zhu et al.(2000)), but the use of this criterion for financial time series isn't common yet. The aim of this article is therefore to introduce the DIC and to show how to use it for a rigorous and common model for modelling econometric data, namely the stochastic volatility model.The performance of the DIC is shown for various SV models fitted with a set of stock market data, namely the S\&P 100 index. Since the calculation of the DIC is already implemented as a tool in the BUGS (Bayesian Inference Using Gibbs Sampling) software package, the Bayesian practitioner can easily take advantage of this model checking criterion without computational difficulties.

PDF file of paper: berg.pdf

Session: Econometric Methods in Finance

Time: Saturday, 7 July, 8am - 9:30am

Room: A