Trading Correlation: From Parlays to Dispersion | Vinicius Esposito<br>Trading Correlation: From Parlays to Dispersion<br>August 19, 2026
It’s the summer of 2026 and betting is very cool. You just do not want to be caught watching a World Cup game without having done a 17-leg parlay. You get to say things like “this match is so mispriced” or “I’m going to buy insurance on my country losing”. For max street cred, during hydration break, drop a “my analyst at D.E. Claw just pinged me on Telegram, they’ve found a sick +EV trade”.<br>However, people trading stuff in seemingly complex ways and bizarre jargon is not a new phenomenon as volatility traders have been around for a long time. In fact, institutionally dealing with contracts that depend on the outcome of more than one event, such as Kalshi’s combos and Polymarket’s parlays, is among the main sources of income for French people in NYC and London.<br>It turns out, pricing these bets is all about correlation. And correlation trading is something vol traders think a lot about, particularly for one kind of strategy called dispersion.<br>Pricing a parlay<br>A parlay is simply a bet on two or more events - it pays off if all its legs hit, so its price must depend on the joint probability distribution of its legs.<br>Take the simplest parlay: two legs, A and B, each a coin flip. We can visualize the whole outcome space in a 2x2 grid, seen below. Green denotes the probability that the parlay hits as both events land, so its price should mostly be determined by this area. Change the correlation between these events by dragging the slider or choosing a number directly to see how it impacts the price.<br>P(both legs win)<br>25.0%<br>Fair parlay odds<br>4.00×<br>Legs multiplied (ρ = 0)<br>4.00×
Parlay payoff area as one quadrant of the joint outcome squareindependencegrid (ρ = 0)A losesA winsA loses · B wins25.0%Both lose25.0%Both win25.0%A wins · B loses25.0%Correlation ρ
0.00<br>−1<br>−0.5<br>+0.5<br>+1
Each leg is a single binary outcome with a fixed win probability, and the correlation ρ links them.<br>The default fixes both legs at 50% probability, the case where ρ can span all of [−1, 1]; unequal probabilities shrink that range.As is to be expected, higher correlation makes the green area larger and therefore increases the price of the parlay (there is no free lunch, since you also have a higher probability of winning). But also notice how the vertical divider never moves: correlation is about how the diagonals compare to each other, while the columns are about marginal probabilities. Said otherwise, individual leg prices contain no information about parlay price beyond the independence baseline.1<br>What does this mean for your sports betting? Well, a bet on “Lakers win vs Warriors AND Celtics win vs Knicks” has a fair correlation of basically zero, as the games happen in indoor stadiums, at different times, in different parts of the country. A bet on “France wins vs Argentina AND Mbappé scores” should not have a fair correlation of zero, as the probability of France winning is not the same given an Mbappé goal as opposed to no goal. Of course, whoever is selling you the parlay knows this and they will offer you that correlation at a premium to fair to make some money.2 In fact, to spot this in the wild, go ahead and price a two-leg parlay on some NBA games and back out the implied probability off of the individual leg prices. You just paid for a market maker’s lunch!<br>What does this have to do with options?<br>Now, instead of two coin flips, consider an option on a basket of two stocks, each with 50% weight. Though it may sound like an intimidating jump, they are fundamentally similar in that both a coin flip and a stock’s return are random events, with their respective potential outcomes and attached likelihoods. The key difference is that we move from a discrete world (of binary events such as “heads or tails” or “Knicks win or lose”) to a continuous one (a stock’s return is described by a probability distribution). The neat thing is that the main learnings from the previous section still stand and, notably, the largest difference is that instead of looking at a 2x2 grid we now look at a “cloud” of dots, all of which represent a draw from the joint probability distribution of the two stocks.<br>P(basket call pays)<br>17.3%
Basket vol<br>21.2%<br>Each leg’s vol (fixed)<br>30%
Basket call payoff region over the joint return cloud of two stocksfixed marginalsbasket callpays here(basket up 20%)−60%0+60%+60%0−60%Stock A returnStock B returnCorrelation ρ
0.00<br>−1<br>−0.5<br>+0.5<br>+1
Each dot is one joint outcome for the two stocks’ one-year returns (both 30% vol,<br>mean zero); the same random draws are reused as ρ changes, so the slider reshapes<br>one cloud rather than resampling it. A call on the 50/50 basket pays...