The asymmetric volatility model of Indonesian crude oil price with artificial neural networks hybridization

Abstract

Volatility is a measure of changes in security prices that can be used to measure investment risk. Volatility can be detected through a variance model that contains heteroscedasticity effects in the form of ARCH/GARCH. If volatility contains asymmetric effects, it can be overcome by exponential and threshold models, namely the EGARCH and TGARCH models. Although it can overcome heteroscedasticity and asymmetric effects, this variance model needs to maintain model stability and accuracy levels, so the neural networks of multilayer perceptron (MLP) and radial basis function (RBF) are proposed as a neuroinformatics approach. This study uses Indonesian crude oil price (ICP) data from January 2011 to December 2024. The modeling results show that the variance model with the hybrid MLP and RBF neural networks approach is able to provide better accuracy, although the EGARCH and TGARCH models can still overcome asymmetric volatility. The hybrid neural networks approach can provide more precise information on variance models that can be used as an alternative in investment decisions related to ICPs that have an asymmetric volatility structure with irregular price fluctuations.