Momentum Is Real. It's Also Completely Priced.

"Momentum" is the most overused word in football commentary, and the reflex among analytically-minded fans is to dismiss it as a narrative device — teams don't carry anything from last Sunday into this one.

That reflex is wrong. Momentum is real and it is large.

Teams that have won their last five games win the next one 65.7% of the time. Teams that have lost their last five win 32.0%. Across more than 8,900 team-games since 2002, that pattern is monotonic and about as statistically solid as anything in football.

Here's the part that matters: those same red-hot teams beat the point spread 50.2% of the time. The ice-cold ones beat it 52.0%. Both are indistinguishable from a coin flip.

Momentum exists. It just doesn't survive contact with the number.

What we measured

For every team, every game since 2002, we looked back at its previous five games within the same season — no window crosses from January into September, which would smuggle in a different roster — and then recorded what happened next.

That gives 8,930 games where a team had a full five-game history to be judged on. The whole sample covers regular season and playoffs, 2002 through 2025.

Two questions, deliberately separated:

• Does winning persist? Straight up — did they win the game, ignoring the spread entirely?
• Does covering persist? Against the spread — did they beat the number the market set for them?

They sound like the same question. They are not, and the gap between them is the entire article.

Question one: winning persists, dramatically

Sorted by how many of its last five a team won, here is how often it won the next one:

Perfectly monotonic across six buckets, spanning 34 percentage points. This is not noise — the extremes sit eight standard errors from even.

It is also not mysterious. Good teams win, good teams keep winning, and a five-game record is a decent read on whether a team is good. Momentum in the loose sense — recent results tell you something about future results — is simply true.

Question two: covering doesn't persist at all

Same teams, same games, same five-game windows. This time the question is whether they beat the point spread:

Flat. Every bucket sits within one standard error of even, and the overall figure across all 8,930 games is 4,464–4,466 — 50.0%, which is as close to a coin flip as a real dataset gets.

The same holds if you define "hot" by covering rather than winning. Teams that covered all five of their last five went on to cover 49.6% of the time. Teams that failed all five covered 52.5%. Neither is a real effect.

Whatever a five-game run tells you about a team, the point spread already knows it.

Question three: watch the number move

This is the part that makes the first two fit together, and it is visible in the data rather than inferred.

Here is the average spread each group faced in that next game — negative means favored, positive means underdog:

A ten-point swing between the hottest and coldest teams in the league.

That is the mechanism. The market is not ignoring the winning streak — it is pricing it aggressively, moving ten full points from one end of the form table to the other. And once it has, the hot team's chance of beating that adjusted number is 50.2%.

Ten points of adjustment, zero points of residual advantage. The market's read on momentum is, as far as we can measure, exactly right.

Define hot by covering instead and the same thing happens on a smaller scale: teams that covered all five of their last five were favored by an average of 3.07, teams that covered none were underdogs by 3.13 — a 6.2-point swing, and again no edge left afterwards.

What this means for the word "momentum"

Three claims usually travel together under that one word. They deserve separating, because two are true and the third is not:

• "This team is playing well and will probably keep playing well." True. 65.7% versus 32.0%.
• "Recent form is informative." True, and it is one of the inputs behind every number you see.
• "Recent form is an opportunity, because the market underrates streaks." Not supported. It moves ten points to account for them, and what remains is 50.0% across 8,930 games.

The third claim is the one that gets said out loud most often, and it is the one the data will not carry.

There is a faint hint of the opposite in the table — cold teams cover 52.0%, and teams that failed to cover five straight cover 52.5%. Both are inside noise (z = +0.97 and +0.84) and we are not going to dress them up. If anything is there, it is a mild over-adjustment to bad form, and it would take several more seasons of data to say so with a straight face.

The honest caveats

• These buckets aren't independent observations. A single team on a long run contributes several rows as its window slides forward, so the effective sample is somewhat smaller than the raw counts suggest. It does not change a result this flat, but the standard errors above are optimistic.
• The 5-0 and 0-5 groups are the smallest (662 and 569 games) and carry the widest uncertainty — which is why the story rests on the monotonic pattern across all six buckets rather than on the extremes.
• Windows stay inside a season, so a team's first five games each year never get judged against the previous January. That is deliberate; it also means early-season games are excluded entirely.
• Pushes are excluded from percentages, as everywhere on this site.
• This is about the closing number. It says nothing about whether a line moved during the week, only that where it ended up already contained the streak.

See it yourself

The last-five trend column is on the SpreadTrends Rankings page, and every team's full game log — with the spread it faced and how the game finished against it — is on its Breakdown page. The streaks are real. Check what the number did about them.

Data: 12,998 team-game rows from the SpreadTrends game database, 2002–2025; 8,930 with a complete prior-five-game window inside the same season. Rolling windows never cross seasons. Pushes excluded from percentages.