Global motion filtered nonlinear mutual information analysis: Enhancing dynamic portfolio strategies

Area of Science:

• * Financial network analysis
• * Complex systems science
• * Quantitative finance

Background:

• * Complex financial networks are nonlinear and noisy, hindering market dynamics analysis and portfolio optimization.
• * Traditional methods often apply global motion filtering to linear matrices, with limited success in complex, nonlinear systems.
• * Noise interference complicates accurate market trend identification and investment strategy development.

Purpose of the Study:

• * To minimize noise in complex financial networks and enhance investment timing strategies.
• * To introduce and validate a new methodology using global motion filtering on nonlinear dynamic networks derived from mutual information.
• * To construct and evaluate investment portfolios focused on peripheral stocks in Chinese and American markets.

Main Methods:

• * Application of global motion filtering to nonlinear dynamic networks constructed from mutual information.
• * Utilization of eigenvalue patterns of global motion to identify collective market movement trends.
• * Construction of investment portfolios using peripheral stocks and comparative analysis against traditional methods (Pearson correlation networks).

Main Results:

• * Portfolios from global-motion-filtered mutual information networks showed superior Sharpe and Sortino ratios compared to Pearson correlation networks and unfiltered matrices.
• * The strategy demonstrated robust performance across various market conditions, including bearish, bullish, and turbulent periods.
• * Growth and decline patterns of global motion eigenvalues effectively identified market trends and influenced portfolio performance.

Conclusions:

• * Global motion filtering applied to mutual information networks is a novel and effective approach for noise reduction and portfolio optimization.
• * Peripheral stocks within these filtered networks yield enhanced risk-adjusted returns, outperforming traditional correlation-based portfolios.
• * The methodology offers significant implications beyond finance, applicable to biological, atmospheric, and neural sciences.