Stock market volatility forecasting at the Nairobi Securities Exchange: a comparison between asymmetric GARCH models and neural networks
Rugut, Diana Chemutai
In this study, we conducted a comparative study on the volatility forecasting models focusing on the asymmetric GARCH models and the Artificial Neural Networks. The study focused on the Nairobi Securities Exchange and used the daily data from the NSE-20 Share Index for a period between January 2012 - February 2021. We took advantage of the year 2020 to investigate whether the models are still effective or they break during a period of turbulence. The parameters of the asymmetric GARCH models were estimated using Maximum Likelihood Estimation (MLE). The asymmetric GARCH models, namely the EGARCH and GJR-GARCH models were compared using AIC and BIC to determine the model with the best fit when it comes to modelling and based on the results of the AIC and BIC, the EGARCH was the better model. The study then compared the forecasting ability of the asymmetric GARCH models to that of the Artificial Neural Networks in terms of the standard deviations as a measure of the actual volatility to determine the better model. The RMSE and MAE measures were used to evaluate the forecasting abilities of the models and based on the results of the RMSE and MAE from the study, the Artificial Neural Networks outperformed the asymmetric GARCH models when it comes to forecasting.
Research thesis submitted to Strathmore University in fulfillment of the requirements for the Master of Science in Mathematical Finance
Stock market volatility, Nairobi Securities Exchange, GARCH models, Neural networks