I let four MoE LLMs from different model families argue stocks to try and pick the best ones.

2026-05-12 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Capital Markets & Investment Management · Depth: Intermediate, quick

Summary

An AI trading experiment, "moefolio.ai," utilizes four local Mixture-of-Experts (MoE) Large Language Models from different families to debate bull and bear cases for stocks. A host model then evaluates these debates and issues BUY, SELL, or HOLD recommendations. The system currently operates on Alpaca paper trading accounts, drawing data from over 50 free sources, with no real money or paid APIs involved. The creator conducts postmortems on losing trades to identify model overconfidence, biases, and reasoning breakdowns. The technical stack includes a Mac Studio M3 Ultra for LLM inference, a Mac Mini with FastAPI for web app snapshots, and a ThinkStation PGX for media generation and YouTube transcription, with the entire system built using Claude Code.

Key takeaway

For AI Engineers exploring autonomous decision-making systems, this experiment highlights the value of multi-agent debates and post-mortem analysis. You should consider implementing a similar "host model" approach to synthesize diverse AI opinions and systematically debug failures, especially when dealing with high-stakes applications like financial trading. This iterative refinement process is critical for moving from paper trading to real-world deployment.

Key insights

An AI trading experiment uses MoE LLMs to debate stock picks, with a host model making final decisions.

Principles

• Debate-driven AI can enhance decision-making.
• Postmortems are crucial for AI system improvement.

Method

Four MoE LLMs argue stock cases, a host model grades the debate, and then issues a BUY/SELL/HOLD decision based on the consensus and reasoning.

In practice

• Use local LLMs for cost-effective experimentation.
• Integrate diverse data sources for robust analysis.

Topics

• MoE LLMs
• Algorithmic Trading
• Stock Market Analysis
• Local Inference
• Alpaca Paper Trading

Best for: AI Engineer, Machine Learning Engineer, Data Scientist

Related on AIssential

• Toward Expert Investment Teams:A Multi-Agent LLM System with Fine-Grained Trading Tasks — Fine-grained task decomposition in multi-agent LLM systems significantly improves financial trading performance.
• Longreads + Open Thread — Rigorous analysis across diverse domains often reveals counter-intuitive truths and challenges established narratives.
• Building Multiple Agentic AI Trading Portfolio Pods — Autonomous agentic pods enable diversified trading, reducing emotional bias and single-strategy risk.
• My Adventures Building an AI Money Machine — AI can enable arbitrage in "juicy" prediction markets by exploiting inefficiencies.
• UK wealth manager and price comparison site shares fall amid AI fears — AI advancements are rapidly disrupting traditional financial and comparison service sectors, causing significant market shifts.
• Funds Coin Expands AI Trading Infrastructure Across Gold, Forex, and Stock Markets With Multi-Agent — Multi-agent AI trading systems enhance market execution by distributing strategy processing and eliminating human emotional biases.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.