JACSM - Volatility estimation plays a crucial role in financial modelling, risk management, and derivative pricing. Traditional approaches, such as GARCH models and stochastic volatility frameworks, often face limitations in capturing nonlinear patterns and adapting to changing market dynamics. This study explores a hybrid methodology that integrates artificial neural networks (ANNs) with state space models (SSMs) to enhance the accuracy and adaptability of volatility estimation. By leveraging the data-driven learning capacity of ANNs and the structured temporal modelling of SSMs, the proposed framework captures both nonlinear dependencies and latent volatility dynamics. Empirical evaluation is conducted on daily lithium price data from 2017 to 2024, comparing four models: GARCH, ANN, SSM, and the hybrid ANN–SSM. The findings show that the hybrid ANN–SSM model achieves the lowest error metrics (RMSE = 0.0068; MAE = 0.0048) and better information criteria scores, outperforming GARCH (RMSE = 0.9171), ANN (RMSE = 2.8625), and SSM (RMSE = 1.8010). While GARCH remains robust in modelling volatility clustering and persistence, and ANN captures nonlinear regime shifts, both struggle with structural breaks and extreme volatility spikes. The hybrid ANN–SSM successfully balances accuracy, robustness, and interpretability, offering a more reliable framework for volatility estimation in complex and rapidly evolving financial markets. This research underwrites to the budding convergence of artificial intelligence and statistics, presenting hybrid models as a powerful alternative to conventional volatility modelling.
Basira et al. (Wed,) studied this question.