Kalshi Prediction Market Forecasting

Python Forecasting Statistics Prediction Markets

I’ll write a detailed case study here soon. In the meantime, here’s the short version.

I built two independent forecasting systems to price contracts on Kalshi, a regulated prediction market exchange.

TSA Passenger Volume Forecasting

A weekly forecast for average daily TSA checkpoint passenger counts. The model uses a holiday-aware analog ensemble approach: it maps future dates to historical analogs, scales them by year-over-year growth trends, and converts point forecasts to threshold probabilities using empirical out-of-sample errors. A second layer incorporates weather disruption data from airport observations to adjust forecasts based on trailing national disruption conditions.

Tornado Count Forecasting

A monthly forecast for preliminary tornado report totals. The system compares multiple model families – historical analog ensembles with ENSO conditioning, ridge-based negative binomial models, and random forest approaches with local residual bootstrapping. It supports both month-ahead and mid-month updating as reports come in, and includes a live Streamlit dashboard that pulls real-time Kalshi market prices for comparison.

Both projects are Python-based and include rolling backtests evaluated on Brier score, MAE, and RMSE across the actual Kalshi threshold bins.


← Back to Projects