A Synergistic Approach to Volatility Forecasting: Combining GARCH and Attention-Based Neural Networks

This paper presents a hybrid model combining GARCH and ANN to improve volatility forecasting in financial markets. GARCH captures time-varying volatility, while ANN enhances forecast precision by learning temporal dependencies and using GARCH outputs as features. Empirical tests using daily closing prices from the S&P ۵۰۰, NASDAQ, and Dow Jones (۲۰۱۵–۲۰۲۴) show that the model outperforms traditional methods (e.g., GARCH, EGARCH, LSTM) on key metrics like MSE, MAE, and OSA. Additionally, computational efficiency innovations, such as attention weight pruning, reduce costs without sacrificing accuracy. These findings have significant implications for more reliable volatility predictions in financial markets.

کلیدواژه ها:

Volatility Forecasting ، GARCH Models ، Attention-based Neural Networks ، Hybrid Models ، Financial Time Series