Why parlay math lies to you
Multiply the legs. That's the rule everyone learns, and it's fine for a Bills moneyline stacked with a Lakers moneyline, because nothing that happens in Buffalo affects Los Angeles. Multiplying assumes independence.
Now put four legs in one game. Atlanta's quarterback drops back forty-five times. That single fact simultaneously pushes Drake London toward five catches, pushes Kyle Pitts toward a touchdown, pushes the total over, and pushes Bijan Robinson's rushing yards down. One underlying variable is quietly steering every leg on your ticket at once.
So correlation isn't an exotic risk. It's the entire difference between a same-game parlay and four separate bets, and it runs in both directions.
Positive: the shootout cluster
Positive correlation means the legs rise and fall together. In football almost all of it traces back to two shared variables: how many plays the game has and how much of it goes through the air. Anything downstream of those two moves as a block.
The map for one game
Here's what the pairwise correlations actually look like for the Atlanta side of ATL @ CAR. Yellow is positive, purple is negative, and the intensity is the size.
300+ ydsLondon
5+ recPitts
ATDBijan
60+ rushGame
o44.5 QB 300+ yds — +.54 +.41 −.22 +.49 London 5+ rec +.54 — −.11 −.18 +.31 Pitts ATD +.41 −.11 — −.06 +.24 Bijan 60+ rush −.22 −.18 −.06 — +.06 Game over 44.5 +.49 +.31 +.24 +.06 —
Read the top row and the bottom row and you have the whole story. QB passing yards and the game total are the two hubs; almost everything else is a spoke off one of them. If your slip contains three spokes from the same hub, you have made one bet three times.
Ask yourself what single sentence would make every leg on your ticket cash. If you can write that sentence in under ten words — "Atlanta throws it forty-five times" — your legs are correlated, and you should be looking at the price of that sentence rather than the price of four separate props.
Negative: target competition and game script
Negative correlation is the one that quietly destroys tickets, because it looks like diversification and isn't. Three common sources:
Target competition
A football only goes to one player at a time. London 5+ receptions and Pitts 5+ receptions are competing for the same finite pool of passes. In a heavy passing game both can hit — that's why the correlation is only mildly negative at −0.11 rather than strongly so — but you are, in part, betting against yourself.
Game script
This is the big one and it's where most bettors get caught. Take the two most popular legs in America: a favorite on the spread and that favorite's running back over on rushing yards. Those feel like the same bet, and directionally they are — teams with leads run the ball. But push it further:
- Favorite covers a big number → garbage time → starters sit → the RB's fourth-quarter carries evaporate.
- Favorite wins a nail-biter → RB gets the whole game → over cashes, but the cover might not.
- Favorite falls behind → team abandons the run entirely → both legs die together.
The correlation between a −10.5 favorite covering and their lead back going over is around +0.09 — nearly nothing, and far weaker than the +0.30-ish that most bettors assume when they build the stack. At −7 or shorter it's meaningfully positive. Above two touchdowns it inverts.
The one nobody prices
Kicker points and the game under. A team that drives efficiently but stalls in the red zone produces field goals, a low total, and a happy kicker. That pair sits around +0.26, it's genuinely useful, and we have never once seen a book charge a correlation tax on it.
How the book charges you for it
Books aren't naive about this. When you build a same-game parlay, the app doesn't just multiply the leg prices — it re-prices the combination using its own correlation model, and then takes a cut of the correlation benefit for itself. That's the haircut.
Is 37% too much? That's the only question worth asking, and it has a real answer: it depends entirely on whether the true correlation is bigger or smaller than the correlation the book assumed. Books tune these models to the pairs that get bet constantly. They are much less careful about everything else.
Where the value hides
Two categories, and they're mirror images.
Over-taxed: the obvious stacks
QB with his WR1. RB with his own team's spread. Anytime TD with team total over. These are the combinations every casual bettor builds, every book has modeled to death, and the haircut is set generously in the house's favor because demand is inelastic — people are going to build the QB/WR1 stack regardless of price. If your slip is the same slip forty thousand people built this morning, assume you're paying a premium for it.
Under-taxed: the pairs nobody asks for
The haircut is applied from a lookup table, and the table is sparse. Pairs that get systematically under-taxed:
| Pair | True r | Priced as | Gap |
|---|---|---|---|
| Kicker 8+ pts · game under | +0.26 | 0.00 | +0.26 |
| Backup RB rush att · big favorite covers | +0.31 | +0.08 | +0.23 |
| Punts over · both team totals under | +0.34 | +0.15 | +0.19 |
| WR2 receptions · WR1 shadowed by CB1 | +0.22 | +0.05 | +0.17 |
| QB rush yds over · team total under | +0.18 | +0.03 | +0.15 |
| QB pass yds · WR1 rec yds | +0.61 | +0.66 | −0.05 |
That last row is the control: the most-built stack in football, priced slightly tighter than the true correlation. That's the whole pattern in one table.
These gaps are small, they close over a season as books update their tables, and a couple of them exist only at books with weak SGP engines. This is not a money printer. It's a reason to build the ticket nobody else is building, which is a better habit than a strategy.
Super Bowl LX: Seattle 29, New England 13 (SEA closed −4.5, won by 16). Stafford 4,707 yards / 46 TD and AP MVP. James Cook 1,621 rushing yards. Jaxon Smith-Njigba 1,793 receiving yards. Myles Garrett 23.0 sacks. Full graded plays live on the ledger.
Three rules
Before you place any same-game parlay, write the one sentence that makes every leg cash. If it's short, you've made one bet — so ask whether you'd take the price on that sentence directly, because there's usually a cleaner way to bet it.
A four-leg slip with three spokes off the passing-volume hub is a two-idea ticket wearing a four-leg costume. Diversification in a parlay isn't more legs, it's legs that answer to different variables — and remember that all of them still have to hit.
If the ticket in your head is the ticket in everyone's head, the haircut is set for that crowd. Same idea, different expression — the WR2 instead of the WR1, the receptions line instead of the yardage line — is frequently the same bet at a much better number.