Evaluation of Non-linear Combination Method (Neural Network) For Value-at-Risk Forecasting in Market

Please enable javascript in your browser.

Rashnavadi, Leila | 2012

724 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 43611 (44)
University: Sharif University of Technology
Department: Management and Economics
Advisor(s): Barakchian, Mahdi
Abstract:
Value at risk of an asset, is the asset’s expected maximum loss for a certain period of time and at a specified confidence level. Value-at-Risk can be calculated in the bank with its inter-nal method or standardized method. when a method have more violation number then bank need to keep more daily capital requirements. under the Basel 2 agreement if the violation of method more than 10 times in year, the Bank uses the standardized method.
There are trade off Between daily capital charge and violations. Therefore, existing methods for calculating the value at risk, usually lead to much daily capital charge or many violations. Studies show with combination of different methods to calculate value-a- Risk, we found a way to have acceptable value of two criteria. In this study we use of non-linear combination (neural network) for forecasting value-at-Risk during the crisis and after crisis 2008 by using the SP500 Index. We investigate the performance of a variety of single and combined VaR forecasts in term of Kupiec test, Chistoffersen test, lopez test and daily capital requirements. Result show neural network provide stable result across different criteria and periods
Keywords:
Forecast Combination ; Artificial Neural Network ; Value at Risk ; Backtesting