A morphological perceptron with gradient-based learning for Brazilian stock market forecasting

Area of Science:

• Computational Finance
• Artificial Intelligence
• Time Series Analysis

Background:

• Traditional stock market forecasting models face limitations due to the random walk dilemma (RWD), causing time phase distortions in predictions.
• Existing techniques struggle to accurately reconstruct stock market phenomena due to inherent delays in forecasting.

Purpose of the Study:

• To propose a novel model, the increasing morphological perceptron (IMP), inspired by mathematical morphology (MM) and lattice theory (LT).
• To address the random walk dilemma (RWD) and eliminate time phase distortions in stock market time series forecasting.

Main Methods:

• Developed the increasing morphological perceptron (IMP) model integrating concepts from mathematical morphology and lattice theory.
• Implemented a gradient steepest descent method for IMP design, adapting back-propagation (BP) to handle non-differentiable morphological operations.
• Incorporated an automatic correction procedure within the learning process to mitigate RWD and time phase distortions.

Main Results:

• The IMP model demonstrated effective performance in overcoming the RWD and correcting time phase distortions in stock market forecasting.
• Experimental analysis on Brazilian stock market data and natural phenomena time series showed the IMP's forecasting capabilities.
• The IMP achieved competitive results compared to recently proposed models in the literature.

Conclusions:

• The increasing morphological perceptron (IMP) offers a robust solution for stock market forecasting by addressing the random walk dilemma.
• The proposed model shows promise for improving the accuracy and reliability of time series analysis in financial and natural domains.
• The integration of mathematical morphology and lattice theory provides a novel approach to enhance predictive modeling.