More sports
NFL predictions: win chances for every game
Each NFL game of the week gets a fair win chance from team ratings. The tool also shows the price and point spread that chance means, and the market chance next to it.
Week 8: 14 games
Finished games this week: the model’s favourite won 1 of 1; the market’s favourite 1 of 1.
Thu, Oct 29, 8:15 p.m. ET
Carolina Panthers1935%fair 2.84Green Bay Packers home2765%fair 1.54Model spread Packers −4.4. Market without margin: Packers 63%; spread Packers −3.5.
Model favourite won
Sun, Nov 1, 1:00 p.m. ET
Baltimore Ravens28%fair 3.59Buffalo Bills home72%fair 1.39Model spread Bills −6.9. Market without margin: Bills 59%; spread Bills −3. The model rates the Bills 3.9 points better than the market does.
Sun, Nov 1, 1:00 p.m. ET
Tennessee Titans18%fair 5.71Cincinnati Bengals home82%fair 1.21Model spread Bengals −10.9. Market without margin: Bengals 86%; spread Bengals −12.5. The model rates the Titans 1.6 points better than the market does.
Sun, Nov 1, 1:00 p.m. ET
Arizona Cardinals31%fair 3.18Dallas Cowboys home69%fair 1.46Model spread Cowboys −5.6. Market without margin: Cowboys 73%; spread Cowboys −7. The model rates the Cardinals 1.4 points better than the market does.
Sun, Nov 1, 1:00 p.m. ET
Minnesota Vikings53%fair 1.88Detroit Lions home47%fair 2.14Model spread Vikings −0.9. Market without margin: Lions 60%; spread Lions −3. The model rates the Vikings 3.9 points better than the market does.
Sun, Nov 1, 1:00 p.m. ET
Indianapolis Colts22%fair 4.54Jacksonville Jaguars home78%fair 1.28Model spread Jaguars −9. Market without margin: Jaguars 80%; spread Jaguars −10. The model rates the Colts 1.0 points better than the market does.
Sun, Nov 1, 1:00 p.m. ET
Las Vegas Raiders47%fair 2.12New York Jets home53%fair 1.90Model spread Jets −0.8. Market without margin: Jets 54%; spread Jets −1.5. Model and market agree within a point.
Sun, Nov 1, 1:00 p.m. ET
Cleveland Browns34%fair 2.97Pittsburgh Steelers home66%fair 1.51Model spread Steelers −4.9. Market without margin: Steelers 62%; spread Steelers −3.5. The model rates the Steelers 1.4 points better than the market does.
Sun, Nov 1, 1:00 p.m. ET
Atlanta Falcons45%fair 2.22Tampa Bay Buccaneers home55%fair 1.82Model spread Buccaneers −1.4. Market without margin: Buccaneers 54%; spread Buccaneers −2.5. The model rates the Falcons 1.1 points better than the market does.
Sun, Nov 1, 4:05 p.m. ET
Los Angeles Chargers23%fair 4.26Los Angeles Rams home77%fair 1.31Model spread Rams −8.5. Market without margin: Rams 71%; spread Rams −6.5. The model rates the Rams 2.0 points better than the market does.
Sun, Nov 1, 4:25 p.m. ET
Kansas City Chiefs37%fair 2.74Denver Broncos home63%fair 1.58Model spread Broncos −4. Market without margin: Broncos 49%; spread Chiefs −1.5. The model rates the Broncos 5.5 points better than the market does.
Sun, Nov 1, 4:25 p.m. ET
New England Patriots69%fair 1.46Miami Dolphins home31%fair 3.18Model spread Patriots −5.7. Market without margin: Dolphins 25%; spread Patriots −7. The model rates the Dolphins 1.3 points better than the market does.
Sun, Nov 1, 8:20 p.m. ET
Philadelphia Eagles45%fair 2.24Washington Commanders home55%fair 1.81Model spread Commanders −1.5. Market without margin: Commanders 71%; spread Commanders −6.5. The model rates the Eagles 5.0 points better than the market does.
Mon, Nov 2, 8:15 p.m. ET
Chicago Bears34%fair 2.93Seattle Seahawks home66%fair 1.52Model spread Seahawks −4.8. Market without margin: Seahawks 60%; spread Seahawks −3. The model rates the Seahawks 1.8 points better than the market does.
Season so far: model favourites 64 of 107 (60%); market favourites 68 of 107 (64%). Ties and pick’ems are left out.
Chances before kick-off; a finished game keeps the chance the model gave it. Fair odds = 1 ÷ chance, without a bookmaker margin.
Updated 2026-11-01 16:26 UTC
How the chances are calculated
Ratings. Every club has an Elo rating built from every game since 1999. Before a game the home side’s chance is 1 ÷ (1 + 10−(home rating − visitor rating + 50) ÷ 400): playing at home is worth 50 rating points, so between equal clubs the home side wins 57.2%; at a neutral site (London, the Super Bowl) nobody gets it. After each game both ratings move by 20 × (result − chance) × a margin factor: a rout moves them more than a field goal, and a favourite’s rout less than an underdog’s. Between seasons each rating keeps 50% of its distance from average.
Settings. The step (20), the home edge (50) and the carry-over (50%) are the best of a grid on the 2000–2022 seasons (lowest log-loss of the pre-game chances); 2023–2025 were kept aside to test them.
Spread and chance. A spread converts to a chance as P(favourite wins) = Φ(spread ÷ 11.7), Φ the normal curve; 11.7 is the value that fits the results of every closing spread of 1999–2022 best. A 3-point favourite wins about 60%, a 7-point favourite about 73%.
Fair odds are 1 ÷ chance: the price with no bookmaker margin. The market line is the consensus closing (or latest) price published by nflverse, shown without its margin. A rating knows only past results – not injuries, weather or who starts at quarterback.
How often the closing favourite won, by spread (1999–2022 regular seasons):
| Closing spread | Games | Favourite won | Formula |
|---|---|---|---|
| 0.5–2.5 | 1,191 | 52.7% | 55.7% |
| 3 | 993 | 57.5% | 60.1% |
| 3.5–6.5 | 2,047 | 66.1% | 65.9% |
| 7 | 427 | 76.6% | 72.5% |
| 7.5–9.5 | 672 | 76.9% | 76.3% |
| 10–13.5 | 609 | 82.8% | 82.9% |
| 14+ | 182 | 90.7% | 90.5% |
How good is the model
Hold-out test. The settings were fixed on 2000–2022; then the ratings forecast every one of the 816 regular-season games of 2023–2025 before kick-off. Lower log-loss and Brier scores are better; “always the home side” gives the home club its 2000–2022 win rate, 56.4%.
| Forecast | Winner picked | Log-loss | Brier |
|---|---|---|---|
| Our ratings (Elo) | 63.8% | 0.636 | 0.223 |
| Market: closing moneyline without the margin | 68.3% | 0.608 | 0.210 |
| Always the home side | 53.9% | 0.691 | 0.249 |
| Coin flip (50/50) | 50.0% | 0.693 | 0.250 |
Honest result: the ratings beat a coin and the home side, but the closing market beats the ratings – it knows injuries, weather and starters. Use our chances as a second opinion, not as a reason to bet.
This season, for real: over its 108 regular-season games so far the ratings picked the winner in 60% (market 63%, home side 51%); log-loss 0.664 (market 0.660, coin 0.693).
Calibration on the hold-out games: when the ratings gave the home side a chance in a range, how often it won.
| We gave the home side | Games | Home side won |
|---|---|---|
| 0%–30% (on average 24%) | 41 | 34% |
| 30%–40% (on average 36%) | 79 | 34% |
| 40%–50% (on average 45%) | 163 | 38% |
| 50%–60% (on average 55%) | 183 | 50% |
| 60%–70% (on average 65%) | 171 | 66% |
| 70%–80% (on average 74%) | 130 | 72% |
| 80%–100% (on average 84%) | 49 | 86% |
In-sample, 2000–2022 (5,903 games): ratings 64.2% winners picked, log-loss 0.630; market 66.3%, 0.612.
NFL playoff picture: seeds, odds and tiebreakers →NFL power rankings by the numbers (Elo) →Anytime touchdown scorer odds: what they mean →
Data: nflverse (schedule, scores, closing lines; licence CC BY), checked every 6 hours; nothing is published unless every club has its games, the division games match the alignment and the playoff games fit the bracket from the tie-breakers. Rules: NFL tie-breaking procedures, Wikipedia: NFL playoffs.
How to read a game row
- Find the matchupEach row is one game with its two teams. Finished games show the final score instead of chances.
- Read the win chanceEach team gets a chance from Elo ratings. The two chances together make one whole.
- Check the fair priceThe price that matches the chance, in decimal and American odds. No margin is added.
- See the matching spreadThe point spread that fits the chance. It comes from how spreads and results related in past games.
Fair price from a win chance
Fair decimal price = 1 ÷ win chance (chance written as a share of one)
Example: A 50.0% chance gives a fair price of 2.00, or +100. Convert any price.
Model chance vs market chance
Model chance
- Built from Elo ratings of past game results
- Adjusted for margin of victory, with a home edge
- Super Bowl played as a neutral-site game
- Knows nothing about news, injuries or lineups
Market chance
- Built from the closing moneyline in the nflverse data
- Closing line: the last price before kickoff
- For example, two −110 sides add up to 104.8%
- With the margin removed, each side is 50.0%
- Remove the margin yourself for any market
How to read the accuracy card
| Comparison | What it checks | What it tells you |
|---|---|---|
| Coin flip | Model vs even chance for both sides | Chances closer to results than 50/50 mean the ratings carry information |
| Always the home team | Model vs a forecast that always favours the home team | Doing better means the ratings add more than the home edge |
| Market | Model vs closing lines, margin removed | How close the chances come to the closing market's chances |
| Calibration | Forecast against result, grouped by chance | Do teams given a chance win about that often? |
All checks use seasons the ratings did not see. The settings were fitted on older seasons only.
What the ratings do not know
Injuries
A key player out for the game does not change the rating. The market can react to this news.
Quarterback changes
A new starting quarterback can change a team a lot. The ratings only see results after the games.
Weather
Wind, snow or rain are not in the model. Each game is rated the same way.
Rest and schedule
Short weeks, bye weeks and travel are not counted. Only the home edge is.
Questions
How are these NFL predictions made?
From Elo team ratings built on past game results, adjusted for margin of victory, with a home edge. The settings were fitted on older seasons and checked on recent seasons the model did not see.
Are these NFL picks?
No. Each game gets a win chance, not a pick. A team with the higher chance can still lose. For teams ranked by rating, see the power rankings.
Where does the game data come from?
From nflverse, open NFL schedule and results data under licence CC BY 4.0. It refreshes by itself several times a day. If the data is too old, the tool hides the numbers.
Why is the market chance different from the model?
The market comes from closing moneylines and reacts to news the ratings do not see, like injuries. The model uses only past results. A gap is not a signal to bet.
Where can I see NFL playoff chances?
The playoff picture shows seeds, playoff chances and tiebreakers for the current season.