چکیده:
The purpose of this study is estimation of daily Value at Risk (VaR) for total index of Tehran Stock Exchange using parametric, nonparametric and semi-parametric approaches. Conditional and unconditional coverage backtesting are used for evaluating the accuracy of calculated VaR and also to compare the performance of mentioned approaches. In most cases, based on backtesting statistics Results, accuracy of calculated VaR is approved for historical, Monte Carlo and Volatility-Weighted historical simulation methods. It is also approved for GARCH type of volatility models under normal distribution and Riskmetrics model under student-t distribution. On the other hand, it is observed that parametric approach measures VaR value more than non-parametric and semi-parametric approaches. This result indicates that GARCH type of volatility models under student-t distribution overestimate magnitude of value at risk. Finally, four volatility models of parametric approach including NARCH, NAGARCH and APGARCH under normal distribution and Riskmetrics under student-t distribution are selected best methods to measure accurate value of VaR.
خلاصه ماشینی:
(See Bollerslev, 1986; Nelson, 1991; Glosten & Jagannathan Runkle, 1993; Engle, 1982 & 1990; Engle & Bollerslev, 1986; Engle & V.
K. Ng, 1993; Higgins & Bera, 1992; Granger & Engle, 1993).
Big value for Jarque–Bera statistic also shows that null hypothesis of normality for return distribution of Tehran stock exchange is rejected at probability level of 1%.
Daily VaR of total price index based on parametric approach (including Riskmetrics model and nine GARCH type volatility models under normal and student-t distributions) and two other simulation methods along with expected and real failures at two confidence levels of 95% and 99% are reported in Table 4.
It can be observed that based on unconditional coverage test of POF, for confidence level of 95%, null hypothesis is only accepted for some volatility models (including GARCH, AGARCH, NARCH, EGARCH,NAGARCH APGARCH under normal distribution and Riskmetrics under student-t distribution) and Monte-Carlo simulation method.
Results show that based on TUFF test the accuracy of estimated VaR at confidence level of 95% is approved for semi-parametric approach, nonparametric approach and all nine GARCH type models under normal distribution.
But, at 99% confidence level accuracy of calculated VaR is accepted based on this test via all GARCH family models under normal distribution and Riskmetrics under student-t distribution.
But, at confidence level of 95%, accuracy of calculated VaR is approved only for Monte-Carlo simulation and some of volatility models including GARCH, AGARCH, EGARCH, NARCH, NAGARCH and APGARCH models under normal distribution and Riskmetrics model under student-t distribution.