An innovative hybrid investment model using large language models and genetic algorithms

This study proposes a hybrid financial forecasting framework, GA-Contemplation (GAC), which integrates Large Language Models (LLMs) with Genetic Algorithms (GAs) to analyse short-horizon stock price movements in the Taiwan equity market. Using structured prompt-engineering techniques – Skeleton-of-Thought, Take-a-Step-Back, and Chain-of-Thought – LLMs extract semantically informed predictive factors from financial news through structured and stepwise prompt-guided reasoning, while GAs optimise factor selection and decision thresholds. Empirical results indicate favourable relative performance of GAC compared with a buy-and-hold benchmark. Additional analyses indicate that structured prompting is more beneficial in low-dimensional and interpretable factor configurations, with diminishing marginal effects as factor dimensionality increases. Overall, the findings provide methodological insights and exploratory empirical evidence on combining LLM-based reasoning with evolutionary optimisation for short-horizon financial analysis.