Multiscaled Neural Autoregressive Distributed Lag: A New Empirical Mode Decomposition Model for Nonlinear Time Series...

Area of Science:

• Machine Learning
• Statistics
• Econometrics

Background:

• Accurate time series forecasting is crucial for decision-making.
• Traditional econometric models are being enhanced by neural network advancements.
• Existing models struggle with complex nonlinear patterns in econophysical data.

Purpose of the Study:

• To propose a novel multiscaled Feedforward Neural Network (FNN) for multivariate time series forecasting.
• To develop an Empirical Mode Decomposition (EMD)-based Neural Autoregressive Distributed Lag (ARDL) model.
• To capture nonlinear patterns like trends, seasonality, and long-range dependency.

Main Methods:

• Utilizing Empirical Mode Decomposition (EMD) to break down time series into different resolution levels.
• Employing feedforward Neural ARDL models for each EMD component level.
• Extrapolating forecasts level-by-level and recombining them for a final output.
• Designing an optimal learning scheme for efficient training.

Main Results:

• The proposed EMD-based Neural ARDL model demonstrates effectiveness in capturing nonlinear patterns.
• Experiments on real-world data show competitive performance against benchmark models.
• The multiresolution approach enhances forecasting accuracy for complex time series.

Conclusions:

• The novel EMD-based Neural ARDL model offers a powerful tool for multivariate time series forecasting.
• The integration of EMD, nonlinearity, and neural networks provides superior pattern recognition capabilities.
• This approach advances the field of time series analysis, particularly for econophysical applications.