Agentic Market-Making Strategy for Polymarket

The Breakthrough

The author developed an autonomous agentic strategy for Polymarket that replaces taker-side betting with maker-side limit orders, using an AI-derived fair value model to capture a consistent 4-cent edge while avoiding transaction fees and slippage.

What Actually Worked

• Fair Value Calculation: The agent continuously calculates a fair value price for binary outcome markets (e.g., BTC 5-minute price movement) by analyzing historical market snapshots.
• Resting Order Placement: Instead of executing market orders, the agent places limit orders at a 4-cent discount relative to the calculated fair value price. If the fair value is 0.51, the agent bids 0.47 for an 'up' share.
• Data-Driven Calibration: The strategy relies on a robust dataset to minimize overfitting. The author calibrated the model using 144,000 graded snapshots, 2,000 resolved markets, and 170 hours of live data.
• Autonomous Monitoring: The agent runs as a background process, adjusting bid/ask spreads dynamically as the fair value price shifts, ensuring the strategy remains in a positive expected value state.

Context

Trading on prediction markets like Polymarket often results in eroded margins due to taker fees and slippage, especially when the edge is small. The author shifted from active betting to a market-making approach, using AI to determine the true probability of an outcome versus the market-quoted price. By consistently bidding below the fair value, the agent captures profit from impatient traders who cross the spread to exit positions. The author emphasizes that the success of this strategy is entirely dependent on the accuracy of the fair value model, which requires significant historical data collection via API.

Content References

• tool: Polymarket, https://polymarket.com, mentioned
• tool: Codex 5.5, mentioned