COMPARISON OF ASYMMETRIC GARCH MODELS WITH ARTIFICIAL NEURAL NETWORK FOR STOCK MARKETS PREDICTION, A CASE STUDY
Samreen Fatima1*, Mudassir Uddin1
Much efforts have been done for modeling of financial data theoretically and empirically for the international
stock markets, for example: Asia, Europe and Australia etc. But no frequent research has been done for the SAARC
countries stock markets. Therefore, bench mark Index of Pakistan; Karachi Stock Exchange (KSE-100) and Bombay
Stock Exchange (BSNSE) of India are selected as case study. They are not only the member of SAARC but also sharing
the common border, due to this they are also involving in bilateral trading. We used closing indices of daily share
price for the period of 1st January, 2010 to 15th January 2016. This study compares the forecasting performance
and also investigates more volatile stock markets using Asymmetric GARCH (A-GARCH) models and non-parametric
method (Artificial Neural Networks). In the A-GARCH; EGARCH and PGARCH models are used. Firstly, suitable
Asymmetric GARCH (A-GARCH) model was developed for forecasting and investigating leverage effect. Secondly,
an Artificial Neural Networks model was developed for the said stock markets. Lastly, forecasting performance of the
FA-GARCH and ANN models both in and out sample were evaluated using root mean square error. In the A-GARCH;
EGARCH (1,1) performed better than PGARCH(1,1) in both stock market data. However, when comparing A-GARCH
with ANN, it was found that ANN gave minimum out sample forecasting error as compared to A-GARCH models.
Therefore, ANN out played other studied models.