ABSTRACT
This study represents a review of relevant literature as well as the theoretical background of stock volatility and neural networks. Subsequently, the study data and variable selection were described. The traditional statistical models, generalized least squares, generalized autoregressive conditional heteroscedasticity and logit, used in the benchmark comparisons were employed and their results were analyzed and evaluated using the standard statistical tests. In addition, a feed-forward neural network was defined and implemented for the empirical study. Company-specific as well as macroeconomic data are used as independent variables in the traditional statistical models. Those variables include a log of total assets, earning per share, mean monthly stock return and market-to-book value as company-specific variables and interest rate, inflation rate, exchange rate and log of the gross domestic product as macroeconomic variables. These variables were used as input to the neural network model. The monthly standard deviation of stock return – a measure of volatility – was used as the dependent variable in the traditional models and as the output variable in the neural network model. The results of the study revealed that the neural network model has proven be outperform the traditional models in the prediction of stock return volatility. The study contributes to literature as it is used as an artificial neural network in two functions (prediction of stock return volatility and classification of the volatility to high volatility and low volatility). Also, few studies are concerned with stock return volatility in developing countries, especially Egypt.
