The assumption hiding in multiplication
Accumulator pricing looks trivially simple. Take the decimal odds of each selection, multiply them together, and the result is the combined price. Four selections at 2.00 produce 16.00.
That arithmetic is correct only under one condition: the selections must be independent. Independence means that knowing the outcome of one tells you nothing about the probability of the others.
Selections drawn from different matches in different competitions are close enough to independent that multiplication works. There is some residual relationship, since weather affects several matches in a region and league dynamics link results within a division, but the effect is small enough to ignore for pricing purposes.
Selections drawn from the same match are not independent at all, and treating them as though they were is one of the more expensive mistakes an unsophisticated sportsbook can make.
What correlation does to a price
Take a concrete case. A team is priced at 2.00 to win. Its main striker is priced at 2.50 to score at any time. Multiplying gives a combined price of 5.00, implying a 20% chance of both occurring.
But these outcomes move together. In matches where that striker scores, his team wins considerably more often than it does overall. Conversely, in matches the team wins, the striker has scored more often than his baseline rate suggests. The genuine probability of both happening might be 26% or 27%, corresponding to a fair price nearer 3.70.
A book offering 5.00 on a combination worth 3.70 is not merely giving up its margin. It is offering substantial positive value to the customer, and it will do so consistently, on every such combination, until it notices. Customers who understand this find these opportunities systematically.
Negative correlation produces the opposite error. Combining a team to win with the match ending under a low goals total involves outcomes that pull against each other, since winning usually requires scoring. Naive multiplication produces a price shorter than fair, which is safe for the book but uncompetitive and, if consistent, unfair to customers.
The general principle is that positive correlation makes joint outcomes more likely than multiplication suggests, and negative correlation makes them less likely. The size of the effect varies enormously with the specific combination.
Where correlation shows up
It is worth developing intuition for which combinations are strongly related, because the pattern is reasonably predictable.
Strongly positive. A team winning and its key attacker scoring. A team winning and a high goals total, in matches where the team is dominant. A player scoring and a player recording an assist within the same attacking side. A team leading at half time and winning. A high goals total and both teams scoring.
Strongly negative. A team winning and a low goals total. A team keeping a clean sheet and the opposition's forward scoring. A defensive player scoring and a low-scoring match, since one requires goals.
Approximately independent. Outcomes involving genuinely unrelated aspects of the match, such as a specific team winning and the number of corners in a period unaffected by their dominance, though even these usually retain some relationship.
The practical difficulty is that intuition establishes the direction of the relationship but not its size, and pricing requires the size. That is what the techniques below are for.
Pricing techniques
Simulation has become the dominant approach and is conceptually the cleanest. A model of the event generates many thousands of synthetic versions of the match, each producing a complete set of outcomes: the result, the score, which players scored, the number of cards, corners and so on. Any combination a customer proposes can then be priced simply by counting the proportion of simulations in which all its components occurred.
The advantage is that correlation is handled implicitly. The model does not need to be told that a team winning and its striker scoring are related; the relationship emerges naturally from simulating matches in which both are generated by the same underlying process. This makes simulation extensible to combinations nobody anticipated, which matters when customers can assemble arbitrary selections.
The requirement is a match model good enough that its synthetic matches resemble real ones in their joint behaviour, not merely in their marginal frequencies. A model that produces correct overall goal rates but unrealistic distributions across players will price combinations badly even while pricing single markets correctly.
Correlation matrices estimate the pairwise relationships between outcome types from historical data and adjust the naive product accordingly. This is computationally lighter than simulation and works reasonably for two-leg combinations. It degrades with more legs, because the joint behaviour of three or more correlated outcomes is not fully described by their pairwise relationships.
Conditional pricing builds the combination sequentially, pricing the first selection normally, then pricing the second given that the first has occurred, and so on. This is exact in principle and demanding in practice, since it requires conditional models for every combination path.
Empirical adjustment applies observed corrections to naive prices based on how combinations of that type have historically performed. It is crude, requires substantial historical data and offers no guidance on novel combinations, but it is better than nothing and is how many operators started.
Margin compounding
Set correlation aside for a moment and consider a separate property of accumulators that explains their commercial importance.
Each leg of an accumulator carries margin. When legs are combined by multiplication, their margins multiply too.
Take four legs each priced with 5% margin, so each is offered at roughly 95% of its fair price. Combining them produces a price at approximately 0.95 to the fourth power of fair, which is around 81.5%. The combined margin is therefore roughly 18.5%, far more than the 5% on any individual leg.
Extend to eight legs and the compounding becomes dramatic. This is the mathematical reason accumulators are the most profitable product in a sportsbook, and it explains a great deal of commercial behaviour: the promotional emphasis on multiples, the prominence of bet builders in product design, and the willingness to price single markets thinly while relying on combinations for margin.
It is worth being clear-eyed about this. Accumulators are appealing to customers because a small stake can produce a large return, and they carry margin several times that of the single bets they are built from. Both statements are true simultaneously, and an honest account of the product includes both.
The risk side of combinations
Combination products concentrate risk in a way single markets do not, and the concentration is correlated across customers.
When a popular favourite is included in tens of thousands of accumulators, the book holds an aggregate position on that team that far exceeds its exposure in the match odds market alone. If that team wins, a large proportion of those accumulators advance to the next leg, and the book's liability across all of them increases simultaneously.
The pattern compounds through a weekend. Each round of favourites winning advances the same accumulators, and by the final leg the book may face substantial liability concentrated on a small number of remaining selections held by a very large number of customers.
This is the mechanism behind results that materially affect an operator's reported earnings, and it is invisible in any individual market view.
The controls are portfolio-level. Aggregate exposure monitoring by underlying selection, cutting across singles, accumulators and bet builders, is the basic requirement. Maximum payout limits cap the return on any single bet regardless of the odds achieved, which bounds the tail. Leg limits restrict how many selections may be combined. Combination restrictions prevent specific pairings that concentrate risk or that the pricing cannot handle reliably. And selection limits cap how much total liability may accrue against any one outcome across all products.
Cash out and its economics
Related to combinations, because it is most valuable to customers holding them, is cash out.
The mechanism offers a customer the opportunity to settle a bet before the event concludes, at a price reflecting the bet's current standing. A customer holding an accumulator with three legs won and one to go can take a guaranteed return rather than risk everything on the final selection.
Its commercial properties are worth understanding. The cash out price includes margin, typically wider than the margin on the original bet, so the operator earns twice on the same stake. It reduces the operator's outstanding liability, which trading teams value particularly when the remaining exposure is concentrated. And it is popular with customers because it provides a sense of control over an outcome that would otherwise be entirely out of their hands.
The pricing requirement is real-time valuation of an outstanding bet, which for a multi-leg combination with legs still in play means continuously repricing the remaining selections and computing the combination's current worth. Errors here are directly exploitable, and cash out has been a source of costly mistakes where valuations lagged the underlying markets.
There is a fairness dimension worth noting. Because the margin on cash out is generally wider than on the original bet, customers who use it frequently receive materially worse value than those who let bets run. Whether that is adequately understood by customers is a reasonable question, and the clarity with which cash out terms are presented has attracted regulatory interest in several markets.
What this means commercially
Pulling the threads together, combination products sit at the centre of the modern sportsbook for three connected reasons.
They are what customers want, because the appeal of turning a small stake into a large return is genuine and durable. They carry the best margin in the product set, through compounding. And they are the hardest thing to price, which means they are a genuine point of differentiation between operators with real pricing capability and those without.
An operator that prices correlated combinations well can offer a broad, flexible bet builder confidently, at competitive prices, and earn properly from it. An operator without that capability faces an unattractive choice: restrict the product to combinations it can price safely, which limits its appeal, or price defensively across the board, which makes it uncompetitive. A third option, pricing naively and hoping, is chosen more often than it should be and is reliably expensive.
A worked correlation example
Working the numbers through makes the effect concrete.
A football match has a home team priced at 1.90 to win, implying roughly 52.6% before margin removal. Its centre forward is priced at 2.20 to score at any time, implying roughly 45.5%.
Naive multiplication gives a combined probability of 0.526 × 0.455, approximately 23.9%, corresponding to a price of 4.18 before the operator's own margin is applied.
Now consider the conditional reality. In matches where this forward scores, his team wins perhaps 70% of the time rather than 52.6%. The correct calculation is the probability the forward scores, multiplied by the probability the team wins given that he scored: 0.455 × 0.70, approximately 31.9%.
The true price is therefore around 3.14 rather than 4.18. A book offering the naive price is offering roughly 33% more than the combination is worth.
Repeat that across every same-game combination a customer can assemble, at volume, and the loss is substantial. It is also entirely systematic rather than random, which means it does not average out. This is why same-game products were slow to appear despite obvious customer appeal: they were not viable until operators could price correlation properly.
Building the product responsibly
A final consideration that sits alongside the commercial one.
Combination products are engaging precisely because they offer large returns from small stakes, and because assembling a personalised bet is more involving than selecting a single market. Those same properties mean they warrant care in how they are presented.
The specific issues that have attracted attention are the display of potential returns without corresponding prominence for the probability of achieving them, promotional emphasis that presents multi-leg combinations as routinely achievable, and the interaction between combination products and cash out, where a customer holding a partially successful accumulator faces a decision under considerable psychological pressure.
Several regulators have addressed elements of this through rules on how odds and potential returns may be advertised, and through restrictions on promotional framing. Operators building these products sensibly present the mathematics honestly, avoid implying that large multiples are a reasonable expectation, and ensure cash out valuations and their margin are clear rather than obscured.
Where operators differentiate
Because pricing correlation is genuinely difficult, it has become one of the clearest points of separation between sportsbooks, and the differences are visible to customers who look.
Breadth of combination is the most obvious. An operator confident in its pricing allows customers to combine almost anything within a match. One that is not restricts the available selections, permits only pairs from a pre-approved list, or blocks combinations it cannot evaluate. Customers experience this as a product that either does what they want or refuses.
Price competitiveness on combinations is the second. Two operators offering the same bet builder may quote materially different prices, and the difference reflects both margin policy and pricing capability. An operator uncertain of its correlation handling protects itself by pricing defensively, which is safe and uncompetitive.
Speed of availability is the third. Combinations priced by simulation require the simulation to run, and operators differ in how quickly a customer sees a price after assembling a selection. Delay is a conversion cost.
In-play combination pricing is the hardest case of all, requiring correlated pricing that updates continuously as the match state changes. Relatively few operators do this well, and it is where the gap between the strongest and weakest pricing operations is widest.
For anyone assessing a sportsbook's technical capability, the same-game product is the most informative thing to examine. Single market prices are largely a matter of margin policy and can be copied from a feed. Correlated combination pricing cannot be, and it reveals whether there is genuine modelling capability behind the product or a supplied price list with a brand on it.
Summary of the pricing decision
Three questions determine how an operator should approach combination products.
Can we price correlation, and how well? This is a capability question with a factual answer. An operator running simulation-based pricing on a validated match model can price broadly. One applying pairwise corrections can price two-leg combinations reasonably. One multiplying naively cannot price same-game combinations at all and should not offer them.
What margin do we want, and where? Compounding means combinations naturally carry more margin than singles. An operator can choose to reduce per-leg margin on combinations to remain competitive, accepting less per bet in exchange for more volume, or leave it to compound.
What exposure will we accept? Combination liability aggregates on popular selections, so limits on payout, leg count and total exposure per underlying outcome need setting before the product goes live rather than after a costly weekend.
Answering these honestly usually produces a narrower product than commercial teams initially want and a considerably safer one, and it is a better position than launching broadly and discovering the pricing gaps through losses.