A Polymarket bot named automatedAItradingbot turned heads with ~$88k profit by trading daily high-temperature bucket markets across global cities. I dug into the wallet data (0xd8f8c13644ea84d62e1ec88c5d1215e436eb0f11), recent closed positions via public API, and the actual mechanics behind the viral story. Here’s the technical breakdown — shorter, data-focused, and actionable for devs and quant traders.
Wallet & Performance Reality Check
- Profile: Created Jan 2025, meteorologist + IT engineer bio, 3k+ predictions.
- Wallet P&L: ~$88,481 realized (profile aggregate). Total volume ~$3.3M.
- Recent sample (last ~2,000 closed positions): Only +$6k profit. Earlier big wins drive the headline number. API limits mean you need to aggregate historically for full picture.
Key asymmetric stat: 349 wins vs 1,537 losses (~18% win rate). Wins pay massively due to low entry prices.
The Legendary Paris Trade (The 20x+ Payoff)
Market: “Highest temp in Paris = 19°C on April 16?”
- Entry: Avg $0.05 per YES share.
- Position size: $32,010 → ~640k shares.
- Resolution: YES → +$30,409.50 profit.
- Math: $0.05 → $1 payout = 20x gross return (minus fees/slippage).
The bot bought when the crowd priced the outcome at ~5% probability, but tight forecast clustering made true prob much higher.
City & Price Bucket Analysis (Where the Edge Lives)
Cities in sample:
- Strong: Paris (+$3.6k), Shanghai (+$986), Istanbul (+$770).
- Weak: Seoul (-$4.2k), London (-$1k), etc.
- Strategy: Spray small bets globally; one big hit covers dozens of losers.
Entry price grouping (critical insight):
- $0.03 – $0.10: +$18k+ across 445 positions → sweet spot.
- ≥ $0.50 (“safe” bets): -$26k+ across 281 positions.
- Below $0.01: Small profit.
Lesson: The bot hunts drastically underpriced low-probability buckets, not high-certainty favorites. It exploits stale limit orders in illiquid markets.
How the Bot Likely Works (Operational Pipeline)
- Scan: Poll active weather markets (city, date, °C/°F bucket).
- Ingest forecasts: Pull professional models (ECMWF/GFS/etc.). Accuracy improves sharply 72h → 24h → event day.
- Probability model: Convert forecast distribution → bucket prob. If model says 45-55% but market = 5¢ → edge detected.
- Execute: Aggressively buy through order book (CLOB on Polygon).
- Resolve: Wait for official station data. (Resolution source knowledge = huge alpha.)
Why weather is exploitable:
- Forecasts converge predictably (unlike elections).
- Niche/illiquid → stale prices persist.
- Official station + rounding quirks create edge for those who know the exact resolution rules.
Failure Modes (Be Honest)
- Wrong station or 0.5°C rounding → instant loss.
- Late-day weather chaos (sea breeze, spike).
- Unit/timezone bugs in automation.
- Over-betting “safe” high-price buckets.
Bottom Line for Builders
This isn’t a “set and forget” weather printer. It’s a high-volume, low win-rate, asymmetric Kelly-style system that requires:
- Solid meteorological data pipeline.
- Real-time market scanner + order execution.
- Discipline to eat hundreds of small losses.
- Capital to size big on 5-10¢ edges.
Niche markets reward deep domain + automation. Weather buckets are one of the cleanest examples of inefficient prediction markets today.
Copy at your own risk — but study the data first.
If you have more questions, please feel free to contact me at any time: https://t.me/FatherSon97
Top comments (0)