In this video, the creator shares their experience using an AI agent to discover profitable trading strategies on Polymarket, a prediction market platform. They set up a 24/7 running AI agent designed to scan markets for arbitrage opportunities, specifically targeting scenarios where buying both sides of a market guarantees a profit regardless of the outcome. One highlighted strategy involved trading on market caps around IPO day, where the agent identified trades that could yield a roughly 993% return by exploiting price discrepancies between “yes” and “no” shares, effectively locking in a guaranteed profit through arbitrage.
The creator explains how they configured the AI agent using tools like Codex, Claude, and Kim K3, emphasizing that the choice of AI model is flexible as long as it can handle coding, data analysis, and math. They leveraged Polymarket’s free API and websocket data to feed the agent with real-time market information. The AI was tasked with researching new features like “combo markets” on Polymarket, aiming to generate hypotheses for potentially profitable strategies. The agent was programmed with a clear goal to find three viable strategies, ensuring the research process had a defined endpoint.
A significant part of the process involved data collection and execution refinement. The creator notes that while the AI could identify promising strategies, successful implementation required precise execution to avoid risks such as only filling one side of an arbitrage trade, which could lead to losses. They spent several days fine-tuning parameters and improving the system’s ability to enter both sides of trades simultaneously, reducing risk and increasing profitability. The video also touches on the importance of collecting comprehensive market data, including quote lifecycle and order book details, which sometimes necessitated building custom data pipelines beyond publicly available historical data.
The creator highlights the synergy between data, code, and math as the core strengths of large language models (LLMs) in quantitative trading research. By combining these capabilities, the AI agent can filter and analyze data, write code to test hypotheses, and apply statistical math to evaluate expected value. This trifecta enables systematic and algorithmic trading research that is largely data-driven rather than speculative. The creator is still in the research phase, using paper trading to validate strategies before committing real capital, and finds the process both educational and enjoyable.
In conclusion, the video serves as an inspirational guide for others interested in leveraging AI agents for trading research. The creator encourages viewers to experiment with similar setups, emphasizing the accessibility of tools and data on platforms like Polymarket. They plan to continue refining their approach and sharing updates, particularly around execution challenges and new research findings. Overall, the video showcases how AI can be harnessed to discover and test quantitative trading strategies in a largely automated and passive manner, potentially generating steady side income while deepening one’s understanding of finance and data science.
