Volatility in Prediction Markets: A Structural Approach
Ciamac Moallemi
Session summary
The presenter describes joint work proposing a structural model of volatility in prediction markets that explicitly uses their bounded, binary-payoff structure and known resolution times. Using several years of hourly data from a large platform, the authors document that prices move more in the near term when markets are near 50% and as contracts approach expiry. They interpret this through a “variance budget” identity: for a binary payoff, total uncertainty must be allocated across remaining periods, motivating a baseline model where variance decays at an average rate over time.
They then augment this with an adverse selection channel based on a Glosten–Milgrom-style order-flow model, where the variance impact of trades scales with spreads and volume. Evaluated via Winkler scores for forecast confidence intervals, their structural models outperform standard ARCH/GARCH and a prior specialized benchmark, especially when combining deadline resolution, adverse selection, and GARCH. Performance varies across categories, with economics markets fitting better than sports, which appear more event-concentrated and violate uniform information-arrival assumptions.
The model is presented as a tool for understanding how information may be incorporated into prediction market prices, rather than as a deployed risk system.
Key takeaways
- The authors derive a variance-budget framework for binary prediction markets, showing that near-term volatility depends systematically on current price and time remaining to resolution.
- Adding an adverse selection channel that links volatility to spreads and trading volume improves volatility forecasts relative to both pure deadline-based and standard GARCH models.
- Forecast accuracy and information-arrival patterns differ substantially by market category, with economics markets closer to the model’s uniform-information assumptions than event-concentrated sports markets.