A hybrid neurogenetic approach for stock forecasting

Area of Science:

• Computational Finance
• Artificial Intelligence in Finance

Background:

• Stock market prediction is complex, with traders using technical indicators.
• Existing prediction models often lack adaptability to market dynamics.

Purpose of the Study:

• To develop and evaluate a hybrid neurogenetic system for enhanced stock trading prediction.
• To investigate the efficacy of a context-based ensemble method for dynamic model adjustment.
• To assess the system's performance across a diverse range of stocks and historical data.

Main Methods:

• Utilized a recurrent neural network (NN) with one hidden layer for prediction.
• Employed a genetic algorithm (GA) for optimizing NN weights via 2-D encoding and crossover.
• Implemented a context-based ensemble method for dynamic NN adaptation.
• Parallelized the GA on a Linux cluster using message passing interface for efficient data processing.

Main Results:

• The hybrid neurogenetic system demonstrated significant average improvement over the buy-and-hold strategy.
• The context-based ensemble further enhanced the predictive accuracy and trading performance.
• Analysis revealed varying predictability across different companies, suggesting portfolio construction applications.

Conclusions:

• The proposed hybrid neurogenetic system offers a robust approach to stock trading prediction.
• Dynamic adaptation through context-based ensembles improves trading strategy effectiveness.
• The system's ability to identify company-specific predictability opens avenues for financial portfolio optimization.