A Hybrid GARCH-BiLSTM-KAN Model for Crude Oil Price Forecasting: Capturing Volatility, Temporal Dependencies, and...

Area of Science:

• Energy Economics
• Financial Forecasting
• Time Series Analysis

Background:

• Crude oil prices exhibit complex dynamics like volatility clustering and nonlinear responses, challenging existing forecasting models.
• Accurate crude oil price prediction is crucial for energy markets, strategic planning, and financial risk management.

Purpose of the Study:

• To develop and validate a novel hybrid framework for enhanced crude oil price forecasting.
• To address the limitations of existing models in capturing multifaceted price dynamics.

Main Methods:

• Integration of Generalized Autoregressive Conditional Heteroskedasticity (GARCH) for volatility.
• Application of Bidirectional Long Short-Term Memory (BiLSTM) networks for temporal dependencies.
• Utilization of Kolmogorov-Arnold Networks (KAN) for nonlinear pattern refinement.

Main Results:

• The proposed hybrid model demonstrated superior forecasting performance compared to benchmark models.
• Achieved lowest root mean squared error and mean absolute error, with the highest coefficient of determination.
• Statistical significance confirmed the model's outperformance across diverse market conditions.

Conclusions:

• The synergistic integration of GARCH, BiLSTM, and KAN offers a robust solution for crude oil price forecasting.
• This advanced framework provides significant implications for energy policy, risk management, and financial modeling.