About Market vs Reality [Football Season]
Market vs Reality [Football Season] is an independent research project studying how prediction markets forecast football games. It is a study of prediction markets, not an endorsement of or opposition to prediction-market trading or gambling.
The project records market probabilities before games, compares those forecasts with actual results, and reports the results whether the forecasts perform well or poorly. No real money is wagered by the Market vs Reality experiments described on this site.
Methodology
- Prediction-market observations come from Kalshi NFL game-winner markets.
- Market probability is calculated from the observed bid/ask midpoint. The two teams' midpoint probabilities are preserved as observed and are not forced to sum to 100%.
- Permanent snapshots are targeted at 7d, 6d, 5d, 4d, 3d, 2d, 24h, 12h, 6h, 1h, and 5m before kickoff when data is available.
- The final permanent snapshot closest to kickoff is used as the final recorded forecast in weekly scoring.
- NFL schedules and final game results are obtained from nflverse data.
- Weekly reports compare forecast accuracy with a 0.2500 Brier-score baseline representing a 50/50 forecast.
Brier formula used for two-team games
Here, p is the recorded probability and o is the observed outcome: 1 for the winner and 0 for the loser.
Definitions
- Probability
- A number from 0 to 1, or 0% to 100%, expressing how likely an outcome is judged to be.
- Prediction market
- A market whose contract prices reflect participants' trading around the likelihood of a future event.
- Bid/ask midpoint
- The point halfway between the observed best bid and best ask. Market vs Reality uses this midpoint as its recorded market probability.
- Brier score
- A measure of probabilistic forecast accuracy. Lower is better. A perfect forecast has a score of 0.
- 50/50 baseline
- For a binary outcome, a forecast of 50% produces a Brier score of 0.2500. We use it as a simple reference point.
- Calibration
- How closely forecast probabilities match observed frequencies over many predictions. For example, events forecast near 70% should occur about 70% of the time in a well-calibrated set of forecasts.
- Kelly criterion
- A mathematical method for sizing a hypothetical wager when you have an estimated probability and a market price. It depends heavily on the quality of the probability estimate.
- Fractional Kelly
- Using only a fraction of the full Kelly amount, such as one-quarter Kelly.
- Bet the House
- Market vs Reality's deliberately naive fictional benchmark. It divides the entire weekly fictional bankroll equally among eligible market favorites. It is a comparison experiment, not a recommendation.
Data Sources & Attribution
Prediction-market data: Kalshi. Market vs Reality independently records observed market bid/ask data and derives midpoint probabilities from those observations.
NFL schedules and game results: nflverse / nfldata. Applicable nflverse data is used with attribution under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
Market vs Reality [Football Season] is not affiliated with, endorsed by, or sponsored by Kalshi, nflverse, or the National Football League. Team names are used for identification and reporting.