Kicker Forecast Methodology
The dashboard estimates full-game team field goals made, then compares each available Kalshi Yes and No purchase price with that estimate. It is an experimental research tool. The historical check did not establish a material predictive advantage over a league-average baseline, and no historical trading-profit test was performed.
Team contracts include field goals by replacement kickers. They are not equivalent to player-specific contracts, extra-point totals, total kicking points, first-field-goal markets, or longest-field-goal bets. Only contracts whose team, game, numerical threshold, and market type can be matched are priced by this model. The live contract rules are shown on each card. [1]
Opportunity and conversion
The inputs are earlier regular-season team games from nflverse. Each prediction excludes the entire target week and all later weeks, including games already played earlier in the target week. This sacrifices some current information to make the timing rule simple and reproducible. The bundled history covers 2022 onward; the browser attempts to refresh the current season when the page loads. Zero-attempt games are retained. [2]
The model uses up to 17 earlier games for the offense and 17 games in which its upcoming opponent faced an opposing offense. The recent league baseline uses the last 544 team-game observations. At least eight offense games, eight opponent games, and 100 league observations are required. Team-name aliases are normalized before matching.
Expected attempts are the geometric mean of the offense's historical attempt rate and the opponent's allowed-attempt rate. Both rates are shrunk toward the league average with eight equivalent league-average games. The expected make rate combines the offense's makes and attempts with 40 equivalent league-average attempts. Multiplying expected attempts by expected accuracy produces the expected number of made field goals.
The made-field-goal count follows a Poisson baseline. If the expected count is λ, the probability of exactly k makes is exp(−λ) × λ^k / k!. A 2+ contract uses the probability of two or more makes; its No side uses the probability of fewer than two. The forecast is a distribution, not a rounded average.
These fixed shrinkage settings are modeling choices, not values claimed to be optimal. The conversion estimate is a team-level historical make rate. It does not separately adjust each kick for its distance or the identity of the current kicker. The team/opponent rates indirectly reflect earlier usage, but this version does not separately estimate coach aggressiveness, drive outcomes, or game score scenarios.
Historical results
The fixed model was checked on all 544 regular-season team games in 2025. Each game's inputs came strictly from earlier weeks. The comparison model used the same earlier league window with a league-average Poisson rate. Lower Brier scores are better; this metric averages the squared difference between a predicted probability and whether the outcome occurred.
| Outcome | Kicker model | League baseline | Mean forecast | Observed rate |
|---|---|---|---|---|
| 1+ field goals | 0.13819 | 0.13843 | 81.3% | 83.5% |
| 2+ field goals | 0.24834 | 0.24973 | 50.1% | 52.8% |
| 3+ field goals | 0.17580 | 0.17540 | 24.0% | 22.6% |
| 4+ field goals | 0.08337 | 0.08344 | 9.3% | 9.2% |
The differences are small, and the model is slightly worse at the 3+ threshold. This check does not support an assertion that the model reliably beats market prices. The thresholds share the same games and are not independent samples. An overall average calibration comparison does not establish accurate probabilities for individual games or subgroups. Statistical uncertainty in performance differences has not been estimated.
The calculations use today's historical dataset, which may include later statistical corrections. They are not an archive of exactly what every source published before each kickoff. No archived executable Kalshi quotes, fill probabilities, fees, or closing prices were available for the historical check. Accordingly, the site makes no backtested return or win-rate claim for its highlighted selections. Machine-readable results · Historical inputs.
Interpreting a highlight
A high-probability candidate has a baseline probability of at least 75%. This threshold is a display preference, not a finding that the selection is profitable. The board selects at most one highlighted contract side per team per game to avoid presenting correlated outcomes as separate independent opportunities.
The sensitivity range recalculates the probability with expected makes reduced and increased by 20%. This is a transparent stress scenario, not a fitted confidence interval. For No contracts, the upper-makes scenario produces the lower probability. The cautious price cap is the largest whole-cent purchase price that leaves a positive gap using the lower scenario probability after estimated fees.
For a standard $1 binary payout, the conditional expected net value per contract is probability − purchase price − entry fee per contract. This arithmetic assumes ordinary Yes/No settlement and holding to resolution. Postponement, cancellation, or fair-price settlement can produce different payouts; the site displays those rules but does not estimate the probabilities of those exceptional outcomes.
The fee estimate uses the currently returned series multiplier and the published general taker formula 0.07 × multiplier × quantity × price × (1 − price). It rounds the fee upward to a whole cent for a conservative display estimate. The current schedule's rounding language and worked fee table differ in precision, so the actual checkout fee controls. The tool assumes one fill at one price; split fills and broker-specific charges can differ. It does not assume a free maker fill. [3]
Kalshi order books return bids on each side. A Yes ask can be derived from one minus the best No bid, and vice versa, at the corresponding size. The page checks available order books for up to the first 64 matched contracts on refresh. Unchecked or missing depth is labeled. Prices and depth may change before execution, and a larger order can cross multiple price levels. The quote retrieval time is displayed separately from market metadata, which is not treated as a quote-update timestamp. [4]
A “Model / price gap” label requires a baseline probability of at least 75%, positive cautious scenario value, a recognized fee schedule, a non-stale quote, and enough reported offer size. It remains a provisional research flag because model error, unusual settlement, stale context, or a changed order book can eliminate the apparent gap. Missing data never creates a positive recommendation.
Research checks and next improvements
Expanded cards show recent attempts and makes, a count distribution, current roster kickers, injury entries, relevant team news where supplied, and city-area weather near kickoff. A listed roster kicker is not confirmation of game-day participation. Missing injury entries do not imply good health. A roof designation or city forecast does not measure wind at the uprights.
Weather, injuries, and coaching information are displayed as context and do not currently alter the numerical forecast. Distance and environmental factors have support in field-goal research, but a successful model of individual kick accuracy is not sufficient to predict how many attempts a team will take. Historical relationships must also be retested after changes in players, coaches, and league rules. [5]
The next substantive model would estimate drives reaching kicking range, the chance the coach chooses a field goal, the attempt-distance distribution, and make probability conditional on distance, kicker, and conditions. Extra points would need a touchdown and two-point-decision model; longest kicks would need a distance-tail model; first-field-goal markets would need possession order and early-drive modeling. These are proposed extensions, not features claimed to be trained here.
Future evaluation should keep whole later seasons untouched, compare against contemporaneous market probabilities, measure calibration by probability band, and archive ask prices and available sizes before kickoff. The standard is improved prediction and obtainable net value after costs, not a persuasive story about a matchup.
Sources
- Kalshi. KXNFLFG series metadata and individual market rules linked from each dashboard card. Read September 14, 2026.
- nflverse. Team weekly statistics, 2022–2026; team-statistics documentation. Downloaded September 14, 2026.
- Kalshi. Fee schedule, effective July 7, 2026; checked September 14, 2026.
- Kalshi. Quick Start: Market Data. Public market, series, and order-book access; checked September 14, 2026.
- Torin K. Clark, Aaron W. Johnson, and Alexander J. Stimpson. Going for Three: Predicting the Likelihood of Field Goal Success with Logistic Regression. Historical study using the 2000–2011 NFL seasons; abstract reviewed.
- ESPN. NFL schedules, game summaries, injury entries, roster and news feeds. Live data accessed when the page loads and a research panel opens.
- Open-Meteo. Forecast API and Geocoding API. Used for city-area hourly conditions around kickoff.