The Squawka Signal NFL model is our in-house projection system. For every team it produces a probability of reaching the playoffs, winning its conference and winning the Super Bowl. For every game it produces a win probability and a projected margin. We then set those numbers against the price at a licensed Canadian sportsbook, so you can see where the model and the market disagree. Here is how it works, in plain terms.
Team power ratings
At the core is a power rating for every team, built with the Elo system that chess has used for decades and FiveThirtyEight popularized for football. Each team carries one number for its strength. After every game the winner takes rating points from the loser. Beat a strong team by a wide margin and you gain more. Lose to a weak one and you drop further.
The ratings are built on every NFL game since 2021 and update after each week. Home advantage is baked in. At the start of each season every team is regressed part of the way back toward the league average, because last year’s form only tells you so much about this year’s roster.
The quarterback adjustment
The biggest single lever in projecting NFL games is the quarterback, and it is where a pure team rating falls short. A team can look strong on paper and start a backup. Another can be middling and have an elite passer.
So on top of the team rating we apply a quarterback adjustment. Every starter’s play is scored from the box score, taking passing and rushing production, touchdowns, interceptions and sacks into a single value figure, with a rolling measure of current form. That figure is regressed toward the quarterback’s longer-term level each offseason, so a hot streak in December does not distort the following September.
The effect is that Signal rates a team on whoever is actually starting, and separates a quarterback’s quality from his team’s results.
Simulating the season
With every team rated, we simulate the rest of the season tens of thousands of times. Each run plays out every remaining game, seeds the playoffs using the NFL’s real tiebreaker rules, then runs the bracket and a neutral-site Super Bowl.
Counting how often each team reaches each stage produces the probabilities you see on our pages. For a single game, the rating gap becomes a win probability and a projected margin.
Model against the market: how we read value
Alongside our number we show the same market implied by DraftKings, an Ontario-licensed sportsbook, converted to a clean probability. That conversion matters. A posted price carries the book’s margin, so the raw implied probabilities across all 32 teams add up to well over 100. We strip that margin out proportionally, which leaves a field that sums to 100 and two columns that can be compared directly.
From there the read is mechanical. Where our model rates a team at least 35% higher than the de-vigged price, we flag it as value. Where it rates them 20% lower or worse, overpriced. In between, fair.
One caveat belongs here, because it changes how those flags should be read. Our model currently backs a narrower field than the market does, concentrating more probability in the top handful of teams. That means most of that group screens as value by construction. It is a disagreement about how open the season is, not a row of edges waiting to be collected, and the model can simply be wrong about the strength of the field.
How we know it works
We test the model on games it had not seen. Across four seasons of out-of-sample results, Signal called game winners correctly about 65% of the time, clearly better than backing the home team every week. Calibration matters just as much: teams it rates a 70% chance go on to win about 70% of the time. Adding the quarterback adjustment measurably improves both, which is why it is built in.
What Signal is, and what it is not
Signal is a model read, not a pick. Treat it as a data-led second opinion to set beside the market rather than a guarantee. The NFL is a high-variance sport and any single game can go either way.
The numbers move as games are played, injuries land and starting quarterbacks change. Use it to understand where value might sit, not as a promise of what will happen.
Where to see the model
The model runs at the top of our NFL hub, where each team’s probability sits next to the DraftKings price with a value read on it, plus a projected pick on every game in the current week.
- Super Bowl race, model against market
- Power rankings for all 32 teams, with playoff and title probabilities
- This week’s games, each with a model pick and line
- Super Bowl odds for all 32 teams, the model against the market on the season’s biggest market
Season-long markets are broken out on our NFL futures page, and the weekly slate is called on NFL picks and predictions.
Signal model FAQ
An in-house projection model. It gives every team an Elo-based power rating, applies a quarterback adjustment, then simulates the rest of the season tens of thousands of times to produce playoff, conference and Super Bowl probabilities.
Tested on games it had not seen, across four seasons it called winners correctly about 65% of the time. Its probabilities are also well calibrated, meaning teams rated a 70% chance win about 70% of the time.
DraftKings, which is licensed in Ontario. We strip the book’s margin out of the posted price first, so our probability and the market probability can be compared on the same basis.
No. It is a model read rather than a pick, and it is one opinion to set beside the market. The NFL is high variance and any single game can go either way.
Ratings update after each week of games, and the market column moves whenever prices do. Injuries and a change of starting quarterback can both move a team’s projection sharply.
Written for analysis and entertainment. This is not financial advice. 19+ (18+ in Alberta, Manitoba and Quebec). If betting stops being fun, help is free and confidential: ConnexOntario on 1-866-531-2600, or your provincial support line.