Correlated bets: why linked picks are riskier
Two bets that tend to win or lose together carry more risk than they look. How correlation works, using our three Saturday college football cards.
On this page
Two bets are correlated when their results tend to move together. That sounds harmless, but it changes your risk: linked bets lose together, so two of them behave less like two small bets and more like one large one. This piece explains how, with our own Saturday cards as the example.
Independent bets first
If two results have nothing to do with each other, the chance both happen is one chance multiplied by the other. Today's Desk has three college football calls for Saturday 31 October:
| Game | Our pick | Our chance | Best price seen | Edge |
|---|---|---|---|---|
| Texas Tech v Arizona | Arizona | 53.5% | 2.53, Pinnacle | +14.0 pts |
| South Alabama v Louisiana Tech | South Alabama | 52.2% | 2.99, Pinnacle | +18.7 pts |
| Oregon v Northwestern | Northwestern | 55.6% | 2.35, Bovada | +13.0 pts |
If the three games were fully independent, our numbers would give:
- all three win: 53.5% × 52.2% × 55.6% = about 15.5%,
- all three lose: 46.5% × 47.8% × 44.4% = about 9.9%,
- expected wins: 53.5% + 52.2% + 55.6% = about 1.6 of the three.
Most days you would get a mix, and the mix is what keeps a spread of small bets steady.
What correlation does
Correlation moves chance away from the middle and into the ends. If the results are positively linked, "all three win" and "all three lose" both become more likely than the products above, and the steady mix becomes less likely. The average number of wins does not change. The spread of outcomes does, and the spread is what hurts a bankroll.
The obvious kind is two bets on the same game: a team to win and the same team's striker to score, say. When one lands, the other is far more likely to. Bookmakers know this, which is why they limit linked selections in a normal accumulator and price bet builders differently; see accumulator odds explained.
The less obvious kind: a shared model
Our three Saturday games are in different places with different teams, so on the field they are close to independent. But all three calls come from the same model, and all three back the side the market makes less likely. If our model has one common error, for example rating outsiders too highly in college football this season, it would be wrong in the same direction in all three at once. That is correlation too, and it does not show up in any single card. We look at that pattern honestly in why this week's cards all back the outsider.
What to do about it
- Treat linked bets as one bet. Add their stakes together and ask whether that total is a size you would put on a single bet.
- Cut every stake, not just the last one, when a day's bets share a cause: the same game, the same weather, the same model's view.
- Keep the day's total small. Our guide to staking several bets on one day shows the arithmetic.
The numbers in the table are estimates, not promises, and so is the independence assumption behind the sums.
18+. Bet only what you can afford to lose.
Keep your record and your bankroll in one place
Record every bet with its stake, odds and result, and watch your profit, your return and your total stake move as each one settles.
Free plan: 5 assessments a month. No card.
Common questions
What are correlated bets?
Bets whose results tend to move together. If one winning makes the other more likely to win, they are positively correlated, and holding both is riskier than it looks.
Why are correlated bets riskier?
Because the bad outcomes arrive together. Two linked bets behave more like one bigger bet than like two separate ones, so your real exposure is larger than each stake suggests.
Can I put correlated picks in an accumulator?
Most bookmakers stop you combining obviously linked selections from the same match in a normal accumulator, and price bet builders to account for the link. Our accumulator explainer covers how acca odds are built.
Are BAI8's picks correlated?
Different games are mostly independent, but they share one thing: our model. This week our three Saturday college football cards all back the side the market makes less likely, so if the model has a common error, those three bets fail together.