Volatility forecasting using asymmetric EGARCH model with Neural Network models and Fuzzy Inference system: the case of Nairobi Securities Exchange
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Strathmore University
Abstract
This research undertook a comparative analysis of volatility prediction models, with a particular emphasis on asymmetric EGARCH model with Artificial Neural Networks and Fuzzy Inference system. The rationale was to produce more accurate volatility forecasts for the Nairobi Security Exchange (NSE). To this end, we combined the EGARCH model with artificial neural networks in two distinct ways. Firstly, we modeled the return series of the stock index with an asymmetric EGARCH model using the Augmented Dickey Fuller (ADF) test and Jacque-Bera test. In the next phase, we estimated an ANN model and then developed two hybrid models, the Fuzzy-EGARCH-ANN and the EGARCH-ANN models. The forecasts generated by each of the four models (asymmetric EGARCH model, ANN model, and the two hybrid models) were compared in terms of their proximity to the realized volatility. The research utilized daily time series data from the NSE-all Share Index, spanning from April 1st, 2011 to July 31st, 20220. Out of these, 940 observations (approximately 90% of the total observations) were used for estimations (training set), and 100 observations (approximately 10% of the total observations) were used for out-of-sample forecasts (test set). The parameters for the asymmetric EGARCH model were determined through maximum likelihood estimation (MLE). The asymmetric EGARCH model was evaluated using AIC and BIC to ascertain the best-fitting model. The RMSE and MAE measures were employed to assess the forecasting abilities of the models. The results from the RMSE and MAE indicated that the artificial neural networks surpassed the asymmetric EGARCH models.
Key words: NSE, MLE, EGARCH-ANN.
Description
Full - text thesis
Keywords
Citation
Kuri, B. (2025). Volatility forecasting using asymmetric EGARCH model with Neural Network models and Fuzzy Inference system: The case of Nairobi Securities Exchange [Strathmore University]. https://hdl.handle.net/11071/16898