Jack Mitcham
BACK TO SELECTED WORK

PROJECT / FORECASTING

Putting uncertainty
to work.

I built two forecasting systems to estimate outcomes in prediction markets. The interesting part wasn’t finding one precise answer. It was understanding the range of outcomes and testing whether the probabilities were useful.

Passenger volume

A weekly model for average daily TSA checkpoint passenger counts. It matches future dates to historical analogs, adjusts for growth and holidays, and uses past forecast errors to turn point estimates into threshold probabilities. Weather disruption data provides another view of near-term conditions.

Tornado counts

A monthly model for preliminary tornado report totals. I compared historical analogs, statistical models, and random forests, with updates as reports arrived during the month. A dashboard placed the model’s estimates alongside live market prices.

The common thread

Both projects use rolling backtests and evaluate forecasts against the actual outcome thresholds. That discipline matters more to me than a model that merely sounds sophisticated.