Northwestern researcher launches election forecasting model using prediction market data
A new data science project by Northwestern University’s Thomas Miller utilizes real-time prediction market data to simulate potential U.S. Senate election outcomes.
Thomas Miller, director of the master’s in data science program at Northwestern University, has developed a 'NowCasting' tool that updates every 10 minutes to forecast election results. The model processes data from prediction markets Kalshi and Polymarket, running one million simulations to assess the probability of outcomes across 35 U.S. Senate races.
The methodology adjusts betting market prices using financial and econometric statistical models. Miller notes that while prediction markets provide high responsiveness to news, raw market prices can be distorted by large trades or platform fees. By applying data science techniques, the model seeks to convert speculative market prices into more stable probability estimates.
Miller’s history with this approach includes correct predictions for the 2020 U.S. Presidential election and the 2021 Georgia Senate runoffs. As of Monday morning, the platform assigned Democrats a 90% probability of gaining control of the Senate.
While prediction markets are increasingly viewed as a supplement to traditional polling, critics note that market dynamics can fluctuate significantly based on short-term news cycles. The model remains active as the final counts proceed toward Election Day.