Radial Basis Functions With Adaptive Input and Composite Trend Representation for Portfolio Selection.

Area of Science:

• Computational Finance
• Machine Learning
• Financial Engineering

Background:

• Portfolio selection relies heavily on asset price trend representation.
• Current methods often use single trend types, lacking composite representation.
• Effective trend analysis is crucial for robust financial predictions.

Purpose of the Study:

• To develop novel radial basis functions (RBFs) with adaptive input and composite trend representation (AICTR).
• To improve portfolio selection (PS) by integrating multiple trend representations.
• To enhance prediction accuracy and robustness in financial markets.

Main Methods:

• Proposed novel RBFs capable of processing multiple trend representations.
• Implemented an adaptive input mechanism that selects the best trend representation based on performance.
• Developed a new objective function for combining RBFs and portfolio selection.

Main Results:

• The proposed RBFs effectively combine diverse trend representations.
• AICTR demonstrated state-of-the-art performance in investing and risk control across six datasets.
• The system proved robust against transaction costs and computationally efficient.

Conclusions:

• AICTR offers a significant advancement in portfolio selection strategies.
• The adaptive and composite trend representation approach enhances financial prediction capabilities.
• AICTR is suitable for real-world financial applications due to its performance and efficiency.