A0826
Title: Neural network methods for volatility forecasting
Authors: J Miguel Marin - University Carlos III (Spain) [presenting]
Helena Veiga - Uned (Spain)
Hongfei Guo - University Carlos III of Madrid (China)
Abstract: Accurate volatility forecasting is essential for risk management and financial market investment decisions. The aim is to present a new framework that improves volatility prediction over various time horizons by combining neural networks and stochastic volatility models. The method uses information from both low- and high-frequency financial data and provides symmetric and asymmetric stochastic volatility specifications enhanced with jump components. Uncertainty is quantified robustly by estimating model parameters using Bayesian inference techniques. To improve predictive performance and stability, a Bayesian ensemble method is also created that combines forecasts from multiple models. Forecast accuracy is evaluated using two loss functions and conditional superior predictive ability tests when applying the suggested methodology to three significant international stock indices. Findings demonstrate significant improvements in volatility forecasting accuracy, highlighting the advantages of integrating machine learning and traditional methods within a unified Bayesian framework.
