Do NFL Hot Streaks Carry Over? What Three Games Really Tell You

"They've won three straight, they're rolling." Recency is the most natural instinct in betting: the last thing you saw feels like the truest. We build our own power ratings and our own spread for every game on Boom Matchups, so we had to decide how much the latest games should count. We tested it instead of guessing.

Last season matters less every year#

How well does a team's final rating last season predict its final rating this season? We measured it for every season from 2020 to 2025 (1.0 would be perfect).

It has fallen steadily. Rosters turn over faster than they used to (free agency, trades, quarterbacks moving), so last year is a weaker guide every season.

Three games in, this season already takes over#

After just three games, our ratings from this season's plays matched teams' final ratings at 0.53 to 0.75, depending on the season. That's about as good as the betting market's own view heading into Week 4.

So three games aren't nothing. By early October, what teams have actually done on the field already beats last season as a guide.

But hot streaks inside a season don't carry#

Here's the surprise. We tested counting a team's most recent games more heavily than its earlier ones, the way "they're rolling" thinking does. It predicted the rest of the season worse, not better.

The best setting was the simplest: count every game this season the same, with last season's rating worth 25%. That's what our ratings use now.

Why recency misleads#

• Three games is a small, noisy sample. A fluky fumble or a backup quarterback can swing a week. Leaning on the latest weeks means leaning on the noisiest part of the sample.
• Opponents vary. A three-game win streak against weak teams looks like momentum and isn't.
• What's real shows up in every game. A genuinely good team is good in Week 1 and in Week 3. Counting every game equally catches that without chasing the noise.

What it means for your bets#

• Be skeptical of streak narratives. "Won three straight" tells you less than how they played in all their games. (Same goes for "he always kills this team": we tested that one too.)
• Don't over-trust last season in October. By then, this season's plays are the better guide.
• Big disagreements with the line are usually noise early on. When our spread disagreed with Vegas by 3 to 6 points, our side covered only 44% of the time (399 games). The line was the better bet. When we disagreed by 6 or more, our side covered 56%, but on just 72 games, too few to tell an edge from luck. We show that record next to every projection.