Title: Asymmetric-loss-based evaluation of daily value-at-risk models
Authors: Anna Titova - Institute for Statistics and Econometrics (Germany) [presenting]
Abstract: A comprehensive comparison of models for forecasting daily value-at-risk is presented. While most of the similar studies perform such analyses using only a few financial time series, the main goal is to rank forecasting performances of a multitude of models on a substantially larger dataset. The models are ranked according to statistical as well as regulatory criteria with guidelines suggested by the Basel accords. Modeling value-at-risk as a conditional quantile via heterogeneous quantile autoregression has shown the best overall results. Additionally, including external predictors containing market characteristics improves a models performance. The validity of the conclusions for expected shortfall forecasts is examined.