Statistical ranking of different VaR and ES models by using Model Confidence Set approach for the banking industry: With an emphasis on Conditional Extreme Value Theory

Subject Areas : Financial engineering

Alireza Saranj ^{1 *} , marziyeh nourahmadi ²

1 - Assistant professor of management and accounting campus Farabi, Tehran University, Qom, Iran
2 - Graduate student, Faculty of Management and Accounting College of Farabi, Tehran University, Qom, Iran

Received: 2016-10-22 Accepted : 2016-12-07 Published : 2017-04-21

Keywords: Value at risk, Extreme Value Theory, Expected Shortfall, Model Confidence Set, Peak over Threshold Approach,

Abstract :

In this paper, we deal with the ranking of different VaR and ES approaches using daily banking industry index data over the period 2008 to 2016, with an emphasis on Conditional Extreme Value approach. In the first stage, we use Bernoulli coverage and independence of violation tests for VaR models and McNeil & Frey’s backtest for ES models to examine the validity of these models. In the second stage, we import the loss functions of the valid models remained from the first stage into the MCS function and rank statistically them. The loss function used for VaR models is Dowd loss function and the one used for ES models is Olsen loss function. The results show that the in both VaR and ES models, the conditional EV with normal standardized residuals, the conditional EV with student's t standardized residuals and GARCH with student's t residuals models are respectively ranked first to third.