After an NFL Score, How Many Seconds Until Polymarket Odds Catch Up?
We line up the start time of every 2025 NFL scoring play (touchdowns, field goals, safeties) with every on-chain Polymarket trade to measure how many seconds until trades lean toward the scoring team, when the odds move most, and how much they catch up within a minute.
- Result status
- Preliminary results
- Last updated
- Start date
- Duration
- Day 1 of 1
- Research team
- Mysports.AI Research Team
- Plan registered
Odds Reaction After NFL Scores: Key Findings
2,437 scores in the 2025 season with trades within 5 minutes (including field goals and safeties; range 14–15 seconds). Timing starts at the snap, seconds before the ball crosses the line. Compared with earlier no-score moments in the same game (1,323 matched cases: 15 seconds after a score vs 19 seconds), the gap is below the pre-set 5-second step, so the price path matters more: half of the 5-minute move is done by second 25. Touchdowns only: about 13 seconds.
- How the price moves: almost nothing on average in the first 10 seconds after the play starts, with the average move most concentrated in seconds 12–17 (the play itself, broadcast and order delays cannot be separated), half of the 5-minute move by second 25, about 90% by one minute.
- Main test (supported): within a minute, the odds catch up only about 57% of the public model’s jump (range 46%–69%); about 46% at 30 seconds and 66% at 5 minutes. Example: if the model moves the home team from 50% to 60% (odds 2.00 → 1.67), within a minute the odds reach only about 55.7% on average (odds about 1.80). This compares with a model; it does not say whether the market or the model is right.
- Touchdowns only: first trade toward the scorer at about 13 seconds, about 61% caught up within a minute; field goals react less (about 35%).
- When the less-favored side scores (a bigger surprise), the odds catch up more (about 74%, vs about 48% when the favorite scores) but still not fully; the two groups also differ in jump size and quarter, and after accounting for those the difference is only barely there.
- Whether the part not caught up after one minute is made up later: not seen. Whether the odds or the model 5 minutes after a score better predicts the final result: no difference seen.
- Playoff games are much faster: the median first trade toward the scorer comes at about 5 seconds; regular-season games also got faster later in the season (about 16–25 seconds early, about 11–15 seconds late).
In plain words
Why study this?
In-game is when Polymarket’s sports markets trade most. A score is news every viewer sees at the same moment, which makes it a clean ruler for “how fast do the odds catch up after news”—a ruler we can set beside our injury-news studies (R1, R12).
What do we want to know?
When a team scores a touchdown or field goal in an NFL game, everyone watching knows at the same moment. Polymarket runs a “who will win” market for every game, and people keep trading during the game. We ask: after a score, how many seconds until Polymarket’s odds start moving, and how long until they catch up?
How do we study it?
We matched the start time of 2,447 scoring plays (touchdowns, field goals, safeties) in 285 games of the 2025 season (nflverse play-by-play) with every on-chain Polymarket trade (about 1.41 million), measuring the second of the first trade leaning toward the scoring team and how the price moved; we used the public nflfastR win-probability model as a ruler for how much of the model’s jump the odds caught up. A Polymarket price converts to a win chance: $0.60 ≈ an implied 60% ≈ odds of about 1.67 (uncalibrated, before fees); lower odds mean more favored. The plan and the analysis code were timestamped before any results were computed.
What have we found?
After a scoring play starts (the snap or kick), half the time the first trade leaning toward the scoring team appears within 14 seconds. Compared with earlier moments in the same game with no score nearby (1,323 matched cases): 15 seconds after a score vs 19 seconds with no score—less than our pre-set 5-second step, so no difference can be claimed. The price itself barely moves on average for the first 10 seconds, with the average move most concentrated in seconds 12–17, completes half of its 5-minute move by second 25, and about 90% by one minute. Measured against the public win-probability model, though, the odds catch up only about 57% of the model’s jump within a minute and about 66% within 5 minutes. Touchdowns only: first trade toward the scorer at about 13 seconds, catching up about 61%. Playoff games are much faster (about 5 seconds).
How does this help with predictions?
In-game Polymarket odds are not instant: the average price move is most concentrated in seconds 12–17 after the scoring play starts (the play itself, broadcast and order delays cannot be separated), and they reflect only about 60% of the public model’s jump within a minute. This gives us a ruler for “how fast odds catch up after news” that we can set beside our injury-news studies. It does not mean there is a sure profit: spreads, fees and broadcast delays are not counted, and the model itself is not the right answer.
What can’t we say yet?
This is a retrospective study: the data are public and similar studies exist on other platforms, so it is not a blind test. Our time is when the scoring play starts; the ball crosses the line seconds later, plus broadcast and order delays, so the seconds are rough (we fixed 5-second steps in advance). We only have trades, not resting orders, so quotes may move earlier. The model is not the truth: odds catching up less could mean the market is cautious or the model jumps too much; their accuracy at predicting the final result shows no difference. This is one season, and late-season games were faster than early ones. The remaining 165 games of 2026 will be re-tested with the same rules.
Terms used on this page
- Scoring play
- The play on which a touchdown, made field goal or safety happens. The time we use is when the play starts (the snap or kick), not when the ball crosses the line.
- Odds (Polymarket)
- A Polymarket price converts to a win chance: $0.60 ≈ an implied 60% ≈ odds of about 1.67 (odds = 1 ÷ price; uncalibrated, before fees). Lower odds = more favored.
- Taker trade
- The side that takes someone else’s resting order; whether it bought or sold is recorded on the blockchain, not inferred.
- nflfastR win-probability model
- A free public NFL model that estimates the home team’s chance of winning before every play from the score, clock, possession, field position and pre-game line. It is a benchmark, not the truth.
- Implied win chance, percentage point
- The win chance implied by the odds; 1 percentage point is, for example, 60% to 61%.
Detailed results
Share of the public model’s jump the odds catch up after a score
2025 season; timed from the start of the scoring play. The share is a regression slope: how much the odds’ implied win chance moves, on average, per 1-point jump in the model.
The pre-registered main test is the “within 1 minute” row (range 46%–69%); other rows are side-by-side or secondary analyses. Only events where the model moved at least 1 point and no other score occurred in the window. This is research, not betting advice.
HypothesesFor deeper reading
- H1One minute after a score, the odds have not fully caught up with the public win-probability model → supported: for each 1-point jump in the model's win chance, the implied win chance from the odds moves only about 0.57 points on average (range 0.46–0.69). Example: if the model moves the home team from 50% to 60% (odds 2.00 → 1.67), within a minute the odds reach only about 55.7% on average (odds about 1.80)Supported in this sample
- H1bWhether the part not yet caught up after one minute gets made up later → not seen (only 269 events could be fairly compared with “no score” moments)Insufficient evidence
- H2Whether the odds 5 minutes after a score or the public model better predicts the final result → no difference seenInsufficient evidence
- H3The odds react differently when the less-favored side scores (a bigger surprise) → supported: these scores see the odds catch up a larger share of the model’s jump within 5 minutes (about 0.74, vs about 0.48 when the favorite scores), though still not fully; the two groups also differ in jump size and quarter, and after accounting for those the difference remains but only barelySupported in this sample
Odds Reaction After NFL Scores: Research Methods
Research questions
- How many seconds after a scoring play starts does Polymarket see its first trade leaning toward the scoring team?
- Within a minute of a score, how much do the odds move compared with a public win-probability model (nflfastR)? What about 30 seconds and 5 minutes?
- During games, are the odds or the model more accurate? Do the odds react differently when the less-favored side scores?
Method summary
- Events: every score in 285 regular-season and playoff games of 2025 (1,459 touchdowns, 976 field goals, 12 safeties); times are nflverse play start times (UTC).
- Trades: every on-chain Polymarket trade (Polymarket-v1 dataset); whether the taker bought or sold is recorded on-chain, not inferred; only each game’s “who will win” market.
- “Share caught up” is a regression slope: how much the odds’ implied win chance moves, on average, per 1-point jump in the model; this is the only main test, and the other three are corrected together for multiple testing.
- All nine data checkpoints passed (285/285 games matched, home/away not swapped, no date errors in timestamps, etc.); the plan had four and the results two rounds of external code review; the analysis code was timestamped before running, and three code–plan mismatches found in review were fixed and re-timestamped (main numbers and formal tests unchanged).
Data and limitations
Data we use (all public and free)
- Polymarket trade-by-trade data (on-chain records; Polymarket-v1 dataset on Hugging Face, CC BY 4.0; Boka Qin, Rui Yang)
- NFL play-by-play, play start times, nflfastR win probabilities and schedules: nflverse (CC BY 4.0; underlying rights belong to the NFL)
Limitations
- Retrospective, not blind: the data are public and similar studies exist on other platforms.
- Times are when the scoring play starts, not when the ball crosses the line; broadcast, block and order delays cannot be separated, so seconds are read in rough 5-second steps.
- Trades only, no order book; the public win-probability model is not the truth.
- One season (2025) only; late-season games were faster than early ones.
- Spreads and fees are not deducted; this is not betting advice. This page shows only aggregated numbers.
Latest progress
Latest：Main results are in: odds barely move for 10 seconds after a score and complete half their move by second 25; within a minute they catch up only about 57% of the public model’s jump (main test supported); the plan and analysis code were timestamped first, and results were externally reviewed
Next step：Collecting the remaining 165 games of the 2026 season; re-test with the frozen rules in mid-March 2027
Study timeline
- Novelty checked; plan finalized and timestamped after four external code reviews
- Analysis code timestamped before running; all nine data checkpoints passed
- Main results in; corrected after external code review (main numbers and formal tests unchanged)
- Collecting the remaining 165 games of 2026; re-test with the same rules in mid-March 2027In progress
Update log
Version history
- v1.0-results
Main results: median first trade toward the scorer at 14 seconds, half the price move by second 25, about 57% of the model’s jump caught up within a minute; three code–plan mismatches fixed after external review, main numbers and formal tests unchanged
- v1.0
Plan finalized and timestamped (T0 = 2026-10-29 02:39 UTC); retrospective, not blind, with the rest of 2026 as a prospective test
Study record
9e335fe4a75d534f0d94f4a76af12d0148dfa3b27c3ae6794bef6efa091436cf01c131367ae58d6bf824b8073e72ea189bf515efaf8fa3a7f80934c4be7498ce01fc82d2fc5f437763946ce4064d8cb5197700dacdfe5b25a6ecb61e99ce28d68ec4e4da77a2249b3e32979e08ba9b70ed74f4e0108a395ebcffaa01a4079f38This page is a research log for reference only and is not betting advice or a guarantee of profit.