Neuron-adaptive higher order neural-network models for automated financial data modeling.

Area of Science:

• Computational finance
• Artificial intelligence
• Machine learning

Background:

• Real-world financial data exhibits nonlinearity, high-frequency components, and discontinuity, posing challenges for classical modeling approaches.
• Traditional neural networks struggle to autonomously identify optimal models and orders for approximating financial data.

Purpose of the Study:

• To introduce novel neuron-adaptive higher order neural-network (NAHONN) models for improved financial data modeling.
• To demonstrate the capability of NAHONNs to automatically determine optimal models and orders for financial data.

Main Methods:

• Development of one-dimensional (1-D), two-dimensional (2-D), and n-dimensional NAHONN models.
• Introduction of a learning algorithm for NAHONNs and establishment of network convergence and universal approximation capabilities.
• Introduction of NAHONN Group (NAHONG) models.

Main Results:

• NAHONNs and NAHONGs are demonstrated to be transparent ('open box') models, enhancing their acceptability to financial experts.
• These models automatically identify the optimal model structure and appropriate order for specific financial datasets.
• Convergence and universal approximation capabilities of NAHONNs are theoretically established.

Conclusions:

• NAHONN and NAHONG models provide a superior, transparent approach to modeling complex financial data compared to classical neural networks.
• The adaptive nature of NAHONNs allows for automatic determination of model complexity, addressing limitations of existing methods.
• The developed models offer a promising advancement for financial data analysis and prediction.