katbrink/polymarket-weather-trading-bot

This is a TypeScript bot that trades weather markets on Polymarket, an online prediction market platform. Polymarket hosts questions like whether a specific city will reach a certain temperature range on a given day. The bot looks for situations where the crowd's pricing of those outcomes does not match what weather forecast models actually predict, then places bets on the mispriced outcome. The bot pulls temperature data from three sources: ECMWF and HRRR/GFS models for forecasts and METAR readings from airport weather stations for real-time observations. It covers 20 cities across multiple continents, including New York, Chicago, Miami, London, and Tokyo. An important detail the README highlights is that Polymarket resolves these temperature markets using airport weather stations specifically, not city-center measurements, so the bot is careful to use the correct station for each city. Before placing any trade, the bot calculates an expected value: if the math does not show a positive edge, it skips the market. Position sizes are determined by a formula called Kelly criterion, which scales the bet size up or down based on how large the edge appears to be. The bot also applies a stop-loss at 20% of position value and a trailing stop that locks in gains after a 20% gain. It skips markets where the gap between buy and sell prices is wider than $0.03. The bot runs on a loop, scanning all available markets on a regular interval and storing every forecast snapshot and trade result in JSON files. It uses that history to calibrate which forecast sources tend to be accurate for which cities. By default the bot runs in paper trading mode, logging simulated trades without spending real money. Live trading on Polymarket requires setting a private key and proxy wallet address in the configuration file. The README suggests paper trading until the results look consistent before switching to live mode.