Guides / Quant Modeling

LSTM for Prediction Market Data

LSTM is a recurrent architecture suited to sequential orderbook or price-path data. Applied to prediction markets, it's typically trained on engineered features — orderbook imbalance, spread, recent trade flow, time-to-resolution — pulled from historical snapshots, with the market's eventual resolution (or a forward price move) as the label.

Where the data comes from

ProbSights doesn't run LSTM for you — it provides the orderbook and trade history across Kalshi and Polymarket that a model like this consumes: BTC/ETH/SOL Up/Down snapshots at 5m–24h resolution on the Pro plan, and Polymarket + Kalshi data more broadly on Builder. Pull it via the API and feed it into your own training pipeline.

Related

• Order book imbalance analytics
• How to backtest a Polymarket strategy
• Prediction market backtesting API

Related guides

• Anomaly Detection in Prediction Market Order Books
• Bid-Ask Bounce and Microstructure Noise in Prediction Markets
• Feature Engineering From Prediction Market Order Book Data
• Iceberg Order & Hidden Liquidity Detection in Prediction Markets

Data from Kalshi and Polymarket. Search live · API docs · Pricing