فهرست نشریات علمی دسترسی آزاد ایران

Accurate volatility forecasting is vital for effective decision-making in financial markets, yet remains a complex challenge due to the noisy, non-stationary nature of financial time series and the diversity of volatility definitions. This study proposes a novel hybrid framework that combines statistical, signal processing, and machine learning techniques to enhance volatility prediction. The approach begins with wavelet transformations to extract multi-scale features from raw financial data, effectively addressing non-stationarity. These features are then evaluated using multiple volatility estimators to determine their predictive relevance. The framework integrates GARCH models, wavelet-derived inputs, and deep learning architectures, with Particle Swarm Optimization (PSO) employed for optimal parameter tuning. Leveraging S&P 500 data from 2000 to 2024 and incorporating multi-source inputs, the model achieves a more holistic representation of market dynamics. Empirical results demonstrate that the hybrid method significantly reduces prediction errors and consistently outperforms baseline models and established benchmarks. To validate its practical utility, we developed trading strategies based on the predicted volatility. Backtesting results show substantial performance gains.