PRESENTATION OF A NEW APPROACH FOR ROBUST ESTIMATION OF VAR: A COMPARATIVE APPROACH

Developing of capital markets and decreasing of interest rates in commercial banks has caused to stock investing become as one of the most important opportunities to earn returns for individuals and rms. Since, the nature of capital markets is involved to abrupt shocks and volatility we have to allow risk. So, It must be predicted and controlled using appropriate models. One of the conventional models to measure and control the risks arising from uctuations in the capital market is using the concept of Value at Risk (VaR). Which is introduced as a standardized measuring risk tool not only for those nancial institutions that are large-scale commercial operations, but also for small banks, insurance companies, investment institutions and non-nancial businesses. Since Value at Risk is analogous to the methods, dierent assets and businesses, in recent years Value at Risk becomes prevalent as a new approach for measuring the risk among the managers and commercial investors. In the current nancial world abrupt and unexpected changes, even a little, has strong eective in predicting future uctuations so it can not be ignored. As a result, robust model should used to predict and control the uctuations that enhance the power and performance estimation and prediction models. According to the importance of the issue in this paper, the robust Cipra method with an optimal smoothing parameter is used to estimate Value at Risk (VaR) for normal statistical distributions and t-student. The data used daily logarithmic returns of the automobile industry index from Mach 2011 to September in 2015. In order to validate the model, the proposed model has been compared with conventional measuring VaR methods consisting of simple moving average, exponential moving weighted aver- age, and GARCH by using rst and second backtesting of Lopez loss function, and Blanco-Ihle backtesting. The results shows that performance of proposed method in normal distribution with condence levels of 95%, 97.5%, and 99% and also in t-student distribution with condence levels of 97.5%, 99% is better than others.