CogAlpha—GIM’s research on multi-agent factor discovery—has been accepted to the ACL 2026 main conference. The paper, Cognitive Alpha Mining via LLM-Driven Code-Based Evolution, was selected from 12,148 submissions and recommended for an Oral presentation by a Senior Area Chair.
The recognition places a capital-markets research problem within one of the world’s leading conferences for computational linguistics and language-model research.
Beyond one-step formula generation
Many attempts to use language models in investing stop at asking a model to generate a formula. CogAlpha instead combines code-level alpha representation, language-model reasoning, and evolutionary search in a system that can refine, mutate, and recombine candidate signals.
Multi-stage prompts and financial feedback help the agents search more broadly while preserving logical consistency, structural diversity, and economic interpretability.
Evidence across markets
Experiments across five stock datasets and three markets showed that CogAlpha could discover signals with stronger predictive accuracy, robustness, and generalization than the comparison methods evaluated in the paper.
The results support a shift from isolated model outputs toward research systems that can explore hypotheses, learn from measurable feedback, and improve their search process over time.
A step toward agentic investing
For GIM, the acceptance is both research recognition and a step toward a broader objective: building agentic systems capable of conducting increasingly complete and accountable investment research.