A1312
Title: A conditional coverage test for forecast combination in Value at Risk prediction
Authors: Malvina Marchese - City University of London (United Kingdom) [presenting]
Abstract: The aim is to investigate the issue of Value at Risk forecasting for high dimensional portfolios from a Value at Risk perspective introducing a variation of a previous model averaging strategy. We consider several MGARCH and SV models with different decay rates and propose a general two step model diversification strategy valid for both classes. We rank the volatility models using several loss functions, robust to the choice of the volatility proxy, and we then construct the forecast combination using the Model Confidence Set approach. We consider several combinations of weights and implement a previous CGMM estimator for their estimation. Finally we establish the asymptotic and finite sample distribution of a conditional coverage test that adapts Christoffersen's conditional coverage test to our model averaging framework. The predictive Value at Risk forecasting performance of the models and of the candidate model-combinations is evaluated for a portfolio of 50 stocks traded on the NYSE over the period 2004-2016.
