Fun formalization

[Published April 2026]

The VoteHub 2026 Midterm Forecasting Model

Zachary Donnini, Ryan McComb, Varun Vishwanath, Austin Park, Luke Wines, Garrett Herrin

Abstract: We present VoteHub’s 2026 midterm forecasting methodology for U.S. House, Senate, and gubernatorial elections. The model combines fundamentals, polling, expert ratings, prediction markets, and correlated simulation into a modular forecasting system designed to produce coherent race-level and chamber-level probabilistic forecasts. We estimate two model specifications: a Legacy model that incorporates fundamentals and polling data, and a Complete model that additionally integrates expert ratings and prediction market information. Back-testing suggests that prediction markets have demonstrated a strong recent track record in forecasting electoral outcomes. The structural core of the system is a stacked regression framework trained and validated by holding out entire election years rather than random individual races, better reflecting the real forecasting problem of generalizing from past cycles to a new electoral environment. Polling is incorporated through a weighting scheme that accounts for recency, pollster quality, repeated releases, sponsorship type, and the effective number of polls (ENOP), which measures the information content of each race’s polling average. Point estimates are then refined using expert race ratings and liquidity-adjusted prediction market data, with market-implied win probabilities transformed into implied margin estimates. To model joint uncertainty across races, the model includes a continuous similarity-based correlated simulation framework. The result is a temporally validated forecasting model that remains responsive to current-cycle information while preserving interpretable structure across races, offices, and chambers.

Cite as:slampaper.mccomb.ca/papers/votehub-2026-methodology
Source:votehub.com