SAME GAME PARLAY
CORRELATION

In a normal parlay across different games, multiplying each leg's probability is roughly right, because the games have nothing to do with each other.

In a same game parlay it is wrong. Every leg is being decided by one football game, and that game has a single script, a single pace, and a fixed number of plays to hand out. Legs that both benefit from the same script are more likely to hit together than the multiplication suggests. Legs that need the game to go two different ways are less likely.

Sportsbooks know this and price it in. The practical consequence is that a same game parlay usually pays less than the same legs would across separate games, and the size of that haircut is not shown to you.

Independent legs multiply

If two outcomes have nothing to do with each other, the chance of both happening is the product of their individual chances. A 50% leg and a 50% leg combine to 25%. That is the rule every parlay calculator uses, and across different games it is fine.

Same-game legs share one football game

Now put both legs inside one game. A quarterback's passing yards and his receiver's receiving yards are literally made of the same throws. If one lands high, the other is more likely to land high, because they are partly the same event counted twice.

The multiplication rule has no way to represent that. It treats every leg as a separate coin, and these are not separate coins.

What correlation means here

Correlation is just a number between −1 and +1 describing how two outcomes tend to move relative to each other. Positive means they tend to be high together or low together. Negative means one being high tends to go with the other being low. Zero means knowing one tells you nothing about the other.

For parlays, the sign matters more than most people realise, and the size matters more than most people expect.

A worked example you can check

Take two legs priced −110 each. That price implies 52.38%, but −110 on both sides of a market carries the sportsbook's margin. Strip it out and each leg is a coin flip: 50%.

Treated as independent: 0.50 × 0.50 = 25%, which is fair odds of +300.

For two legs sitting right at their midpoint, there is a clean formula for the true joint probability once you allow them to be linked: one quarter, plus the inverse sine of the correlation divided by two pi. You can check the arithmetic yourself.

Using an illustrative +0.30 link, that gives 29.8% — fair odds of about +235, not +300. Using an illustrative −0.20 link, it gives 21.8%, or about +359.

So the same two legs, at the same prices, are worth anywhere from +235 to +359 depending only on how they relate to each other. If you price them as independent you are wrong in one direction or the other every time.

Correlation is not a statistical abstraction here. It comes from four concrete football mechanisms.

Game script

The biggest one. A team that falls behind throws more and runs less; a team protecting a lead does the opposite. One scoreboard decides this for both teams at once, which is why a single event — an early turnover, a long touchdown — can push a dozen player props in predictable directions simultaneously.

Pace and total plays

A fast, no-huddle game with few punts produces more snaps for everyone. A slow game with long drives and clock-draining produces fewer. Total plays is a shared ceiling that every counting stat in the game sits underneath.

The shared touch pool

This is the mechanism most parlay advice ignores. A team throws a finite number of passes in a game. Every target that goes to receiver A is a target that did not go to receiver B. Within one game, teammates at the same position group are competing for the same limited supply.

Conditions

Heavy wind suppresses deep passing for both teams. Rain raises fumble risk and pushes offenses toward the run. Conditions apply to everyone on the field at once, so they move whole groups of props together.

WinForge fits correlations between player outcomes from several seasons of play-by-play data and uses them to price parlays jointly rather than multiplying legs. The directions and rough strengths below come from that fit. The exact fitted values are not published; the signs and the relative ordering are what matter for deciding whether a slip makes sense.

Read "moves together" as: if the first leg goes over, the second is more likely to go over too.

Two receivers on the same team, both to go over, is a NEGATIVELY correlated bet. Stacking them makes the slip less likely to hit than the multiplication suggests, not more.

Why the fantasy intuition misleads

Look at a full season and two receivers on a good passing offense obviously look correlated. They both had big years. They were both on the team that threw a lot. Season-long fantasy points for teammates are positively related, and that is where most people's intuition comes from.

But a parlay does not settle over a season. It settles on one Sunday.

Season-long and single-game are different questions

Across a season, the thing that varies most is how good the offense is. Good offense lifts everyone, so teammates look correlated.

Within a single game, offensive quality is roughly fixed — it is the same team, the same week, the same opponent. What varies is who got the targets. And that is close to zero-sum. The week receiver A goes for 110 is disproportionately the week receiver B goes for 40.

Once you remove the shared season-long talent and offense effects and look only at game-to-game variation, the sign flips. Same-team skill players compete more than they correlate.

What that means for your slip

The classic "stack both receivers from the shootout game" slip is quietly worse than it looks. It is not unbettable, but it is being priced by you as if the legs help each other when they mildly hurt each other.

The genuinely positive same-team stack is the quarterback with one of his pass catchers, not two pass catchers with each other. That pairing shares the throw rather than competing for it.

Same game parlay odds are not multiplied

A standard parlay multiplies the legs. A same game parlay does not — the book applies its own correlation model first, then prices the adjusted joint probability, then adds margin. This is why building the same legs as an SGP and as a normal parlay produces different numbers.

The correlation tax

Because bettors overwhelmingly build positively correlated slips — quarterback with his receiver, a team's players in a game they expect to be high-scoring — the adjustment usually cuts the payout. That reduction is a real, if invisible, cost of building inside one game.

Blocked and restricted combinations

Where correlation is extreme, books refuse the combination outright rather than repricing it. If a leg pair is not offered as an SGP, that is information: it is usually the book telling you the correlation is strong enough to matter.

Why it stays opaque

You are shown a final price, not the correlation assumption behind it. You cannot tell whether the adjustment was conservative or aggressive, which means you cannot tell how much margin you are paying. Estimating the joint probability yourself is the only way to know whether the offered price is reasonable.

A copula, in plain English

Each leg gets its own probability from the player projection. Then all the legs are placed on a shared scale and linked together with a correlation matrix, so that when the simulation makes one leg come out high it makes the linked legs come out high too, by the right amount.

The simulation is then run many thousands of times and the answer is simply the share of runs in which every leg hit. No multiplication anywhere.

This is the same family of method sportsbooks use. The difference is not the technique; it is which correlations you feed it and whether you are willing to say where they came from.

Where the numbers come from

The correlations are fitted from several seasons of public NFL play-by-play data, with same-player and same-team relationships measured separately, and re-fitted rather than carried over from fantasy-scoring assumptions.

An earlier version pooled every player together, which mixed talent differences into the correlation and overstated how often same-player legs hit jointly. Measuring game-to-game variation within each player instead brought the estimates down. That correction is why the same-team result reads the way it does.

What it does not do

It does not price touchdown legs' correlation — most touchdown combinations are currently treated as unrelated, which is a known gap and is listed as one in the table above rather than hidden.

It does not tell you a parlay will win. Correlation changes the probability estimate; it does not remove the sportsbook's margin, the model's own error, a late inactive, or the fact that a four-leg slip at 20% loses four times out of five.

Most guides treat negative correlation as something to avoid. That is only true if you are trying to maximise the chance everything hits.

Negatively correlated legs are how you build a slip that survives more than one version of the game. If one leg needs the favourite to control the game and another needs the underdog to be throwing, they cannot both be at their best — but the combination is far less likely to be wiped out by a single early score.

There is a real cost: the joint probability is genuinely lower, so you need a price that compensates. The mistake is not building negatively correlated slips. The mistake is building them without noticing, and paying a positively-correlated price for them.

Estimate the joint probability before you look at the payout

If you decide what a slip is worth after seeing what it pays, the number you see will anchor you. Form the estimate first, then check whether the price clears it.

Prefer QB-to-pass-catcher over pass-catcher-to-pass-catcher

Within one team, the quarterback shares the throw with his receiver. Two receivers compete for it. If you want a same-team stack, the quarterback should usually be in it.

Treat same-player multi-stat legs with suspicion

A receiver's targets, receptions, and receiving yards are very strongly linked. Combining them barely diversifies anything — you are close to betting one outcome three times while being paid as though you bet three. This is the most overpriced structure in the category.

Fewer legs

Every added leg multiplies your exposure to the model being wrong, not just to the game being unlucky. Correlation analysis makes a bad six-leg slip slightly less bad; it does not make it good.

Correlation analysis is a pricing tool, not an advantage in itself. Knowing the true joint probability tells you whether an offered price is reasonable. It does not tell you the projections behind it are right, and it does not overcome the margin on a long slip.

The correlation directions described here are model output fitted to past seasons. Roles change, offenses change, and a fitted relationship is an average across many games, not a rule about the specific game you are betting.

WinForge is informational sports analytics. It is not a sportsbook, does not accept wagers, and none of this is betting advice. Users must be 21+ where required.

To see the projections these estimates are built on, start with the NFL prop projections hub or the player index.

To check the single-leg math yourself, the no-vig calculator strips the margin out of a price and the EV calculator compares your probability against it.

To judge whether the underlying projections deserve trust at all, read how WinForge works and the public accuracy ledger.

• nflverse — the public NFL play-by-play data these correlations are fitted from
• Copula (probability theory) — the method used to combine linked outcomes
• National Council on Problem Gambling