VaR modeling and back testing of short and long positions according to in Sample and out of Sample: application of family models Fractionally Integrated GARCH

Subject Areas : Financial engineering

Mansour  Kashi ¹ (Master of business –financial management, Islamic azad university , sistan o balochestan branch)
S. Hassan  Hosseyni ² (PH.D Student of business management of Isfahan university.)
A. Sadat  niyazkhani ³ (Master of accounting. Islamic azad university,Kashan branch)
S. Amin  Abdollahi ⁴ (M.A student of industrial management, Islamic azad university,Najaf abad branch)

Received: 2016-03-05 Accepted : 2016-05-07 Published : 2016-12-21

Keywords: Value at risk, HYGARCH, FIGARCH, short and long positions, back testing test,

Abstract :

In this study, In addition to calculate the short and long trading positions, we examined In Sample and Out of Sample VaR to assess the quality forecast model is considered. To estimate VaR Result, family models Fractionally Integrated GARCH (long term memory) shows that the model HYGARCH (1, d, 1) with the distribution skewed Student-t similar to the result for FIGARCH (1, d, 1) with skewed Student-t distribution for fat-tail phenomenon exhibits. A comparison of the two models with different distribution model HYGARCH (1, d, 1) with skewed Student-t distribution based on AIC criteria and maximum log-likelihood model was superior. failure rates,  , and duration-based tests where were prepared for back testing in Sample VaR, Indicates that the VaR model of the student-t HYGARCH (1, d, 1) acceptable performance than other distributed models HYGARCH (1, d, 1) and the FIGARCH (1, d, 1) will be . So to estimate Out of Sample VaR by student-t HYGARCH (1, d, 1) has been paid. Like the analysis of the in sample VaR, Out of Sample VaR was compared with the observed output and results were evaluated by and DQ tests. Ultimately resulting VaR-based loss function at all levels quintile (either long or short term trading positions) shows that the model that has the characteristics of long memory in the conditional variance, minimum losses and better performance in assessment Forecast offers.