Risk prediction plays an increasing role in financial risk management. This study aims to investigate existence of asymmetry and long memory volatility in Tehran Stock Exchange Index daily data over period of 1998-2006.
1467 daily index returns are used for volatility modeling via GARCH (Long & short Memory) processes for both normal and t-student innovations. The specification and forecasting performance of competing volatility models are compared by standard criteria. Considering the evidence of long memory, ARFIMA models are developed for conditional mean and both long and short memory models are used for conditional variance. We find that long memory models (particularly with normal distribution of innovations) perform more accurately. Also empirical results indicate that GARCH models have confidential performance with t-student innovation. In sample and out–of-sample Value at Risk calculation resulted by FIGARCH models are more accurate than those of generated by traditional GARCH, particularly in 2.5% critical region.
JEL Classification: C22, C53, G15