Correlation in Same-Game Parlays

In a normal parlay the legs are independent and the odds multiply cleanly. In a same-game parlay they are not, and that is the entire subject.

Why correlation changes the math

If two outcomes are positively correlated, one happening makes the other more likely. Multiplying their independent probabilities therefore understates how often both occur — so a parlay priced as independent is underpriced.

That is the only sound reason to build a parlay. Everything else is paying compounding vig for a bigger number.

The problem is that books know. Same-game parlay pricing models apply correlation explicitly, which is why an SGP pays less than multiplying the individual legs would suggest. It is also where the margin hides: SGP hold routinely runs two to four times a straight bet's.

The measured coefficients

These are the baseline correlation values WeBall's engine carries. Almost nobody publishes a table like this, which is reason enough to put it here.

NBA

NFL

Soccer

How to read the table

Above 0.6 — a strong relationship that any competent pricing model already handles. Betting these together is not an edge; it is the thing the SGP product was built to price.

0.2 to 0.5 — the interesting band. Real relationships, weak enough that some models treat them as noise. This is where mispricing survives.

Near zero — legs that genuinely are independent. A parlay of these is just compounding vig.

Negative — actively bad. The classic is a team winning combined with the game going under: winning teams tend to be scoring, so the two work against each other at −0.28 in the NFL.

Where an edge can actually exist

Not in the headline stacks. Quarterback yards with his top receiver is the first thing every pricing model handles.

The candidates are relationships that sit in the middle band and cross categories the model may treat separately — a receiver's yardage with the game total (0.47) rather than with his quarterback, or a player's points with the game total (0.64) rather than with the team's win.

Even then, the hold has to be cleared first. On a market holding 15%, a correlation the book underrated by a few points does not get you to positive expected value.

What WeBall does and doesn't do

The coefficients above are live in the API and drive correlation lookups between any two legs. What does not exist yet is a user interface for it — there is no parlay builder in the product, so the data sits behind an endpoint with nothing calling it.

That gap is named on the roadmap. In the meantime the numbers above are the useful part, and they are the same ones the engine uses.