Statistics Seminar: Gerlach -- Bayes Factors for Assessing Dynamic Quantile Forecasts

May 3, 2013 Carslaw 373, 2pm-3pm 
(Please join us for lunch with our speaker at 12:30pm) 

Speaker: Richard Gerlach, The University of Sydney

Title: Bayes Factors for Assessing Dynamic Quantile Forecasts

Abstract:
This paper proposes Bayesian evaluation and Bayes factor methods 
for assessing dynamic forecasts of quantile levels. Bayes factor 
analogues of several popular frequentist tests for independence 
and correct coverage of quantile forecasts are developed. 
Multivariate quadrature methods and the standard asymmetric 
Laplace quantile likelihood function are employed, when analytic 
formulas are not available, to obtain relevant marginal 
likelihoods. The proposed methods are extensively assessed via 
simulations and compared to the relevant frequentist testing 
procedures. For the empirical study, the favoured methods from 
the simulation study are applied to test the adequacy of a 
range of forecasting methods for Value at Risk (VaR) in several 
financial market data series.