How I singlehandedly made a sports betting company change their payout format

Preface: I do not encourage involvement in sports betting/any forms of gambling. However, if you do choose to engage in such an activity, it is crucial to gain a complete understanding of how the industry works, as well as the math behind everything. I am not necessarily an opponent of sports betting, but I am an opponent of uninformed sports betting, predicated on a false perception that it is easy to win long-term without any consideration of the math behind everything. The industry is built on deception and information asymmetry. Gamble responsibly.

For the past few years, I have been quite an avid fan of professional League of Legends. And on one fateful day in 2022, I discovered I could bet on the games I watched. Since then, I have been on and off betting League of Legends in my free time.

There are several markets to bet on - a player’s kills, deaths, or assists among the most popular. However, the market that I specifically focused on, fantasy score, was a combination of the three, along with a player’s creep score (the amount of minions they last hit in a game), with positive recorded stats increasing fantasy score, and negative stats decreasing it. If you think about it a little bit, it is pretty evident that many of the outcomes are reliant on whether a team wins the game or not, but because of the way fantasy score works, it is even more dependent on the outcome of the game than any other market. The reason is quite obvious, if a player wins they are significantly more likely to record more kills and assists, and significantly less likely to die. This singular fact is where I gained the most leverage.

The site I first bet on, PrizePicks, had fixed payouts - meaning that the payouts were the same no matter what, irregardless of whether the props I took were correlated or not.

PrizePicks initially only offered lines for player kills. However, it was still very possible to create correlation with the limited props offered. For instance, taking two player’s kills over from team 1 and two player’s kills under on team 2 would be an example of a slip correlated to the outcome that team 1 smashes team 2. And for a while, this would pay out the same amount as an uncorrelated play. I think it is fairly obvious how there is an edge here.

Back in Worlds 2022, I believe I was one of the first people to do this. However, over time, as people began to realize the edge and “correlation bets” caught wind, the payouts eventually were me reduced. Around late 2023 to early 2024, the payouts for the aforementioned correlation slip with two overs and two unders shifted from 10x the initial wager to 7x, essentially establishing the notion that a bet like this should have 1 to 7 odds rather than 1 to 10, which is textbook evidence that those bets used to have immense edge, and I was one of the first people on it. 

I stopped betting on League of Legends for a while after the payout reductions because I thought that correlating players was the biggest edge there was, and after it was taken away, I did not see as much of an opportunity to profit.

That was until Underdog Fantasy enabled betting on multiple players on the same team, something that was not previously allowed. 

The way Underdog used to account for correlation was simply not allowing users to bet multiple players from the same game, which was quite effective. However, in early 2024 I realized that it was finally possible to do this, albeit with even more extreme payout shifts, having that same aforementioned four player bet paying out somewhere in the neighborhood of 4.5x the wager instead of 10x. But, the most crucial discovery I made while doing some experimentation was their anticorrelation payouts: if you were to pair two anticorrelated props (for example, overs on kills on two players playing against each other), the payout on a two player bet would shift from 3x to 3.6x. And the even more crucial discovery was that the payout shifts were even more extreme for anticorrelated fantasy score plays, since, as I mentioned earlier, fantasy score is more correlated with the outcome of the game than anything else so Underdog thought that making correlation/anticorrelation payout shifts even more extreme for these props would account for that. The issue with that was, it was possible to create massive payout multipliers with anticorrelation that were much higher than I thought they were supposed to be.

After some more messing around, I figured out that if you bet with anticorrelation in certain ways, it was possible to slips that paid out a wild 175x the initial wager that relied on outcomes whose odds I thought were way shorter than 1 to 175. And that didn’t even factor in the payouts for getting 4 or 5 out of 6 picks correct, which would pay out around 1.2x and 11.5x respectively. My hypothesis was that the payouts for these “anticorrelation slips” were way, way too high. And so I put my hypothesis into practice, by placing mainly two types of slips that I thought had the most edge:

Let’s assume there are two games: Team A vs Team B and Team C vs Team D, and each team has five players, Players 1-5.

Slip 1 (3/3 stack, now pays 58.98x/6.13x/0.58x):

Team A Player 1 over

Team A Player 2 over

Team A Player 3 under

Team C Player 1 over 

Team C Player 2 over

Team C Player 3 under

Slip 2 (5/1 stack, now pays 54.35x/5.65x/0.54x): 

Team A Player 1 over

Team A Player 2 over

Team A Player 3 over

Team A Player 4 under

Team A Player 5 under

Team C Player 1 over/under (doesn’t matter, uncorrelated)

Each of the slips in these examples used to have payouts of (175x/11.5x/1.2x) for getting 6, 5, or 4 picks correct respectively. I am going to focus on the first type of slip, 3/3 anticorrelated two team stacks, for my explanations because those were what I primarily used. 

Even though the props seem anticorrelated, they actually make sense to play together because there are specific scenarios and drafts where the outcome of all three “anticorrelated” props hitting has a real chance to occur. For example, if the draft implies that a certain player would want to side lane more rather than teamfight, they would get a lower fantasy score, and they would have a chance to go under even in a game where the rest of their team goes over. Or if a team tends to strongside a certain lane and put another lane on an island, it makes sense for the player on an island to have a lower fantasy score relative to the rest of their team. So each of these three “uncorrelated” plays actually plays for the “correlated” scenario that a certain player is not as involved in fights compared to the rest of their team. These opportunities were not as rare as they seem; understanding the game, the meta, and playstyles of teams made finding opportunities to build anticorrelated stacks very possible.

Now that the theory is out of the way, let’s talk about the math behind it, using rough intuition to create percentages for each scenario to occur. Let’s say that in the games I chose to bet, a player has a 85% chance to go over their fantasy score line in a win. But, the player that I would take a shot on to go under has a 75% chance to go over their fantasy score, since I would only bet an under on a player that I think is less likely to go over than the rest of their team. And let’s just say that the teams that I think will win actually wins 70% of the time (this number seems high, but there are always some seriously lopsided matchups).

Then the odds of hitting a 175x would be 0.7(0.7)(0.85)(0.85)(0.25)(0.85)(0.85)(0.25), giving me an insane 1.6% chance to hit 175x my wager, which is already wildly positive expected value as is.

But even if the perfect outcome did not occur, there was still the cushion of the payouts for getting 5 out of 6 picks correct. 

The odds of hitting an 11.5x would be 2(0.7)(0.7)(0.85)(0.85)(0.75)(0.85)(0.85)(0.25) + 4(0.7)(0.7)(0.1)(0.9)(0.25)(0.9)(0.9)(0.25), an insane 10.5% chance to hit 11.5x, on top of the 1.6% chance to hit 175x. 

But the absolute craziest part was that you could miss even one more play and still get your money back plus a little more.

The odds of hitting 1.2x would be 6(0.7)(0.7)(0.15)(0.15)(0.25)(0.85)(0.85)(0.25) + 8(0.7)(0.7)(0.15)(0.85)(0.75)(0.85)(0.85)(0.25) + (0.7)(0.7)(0.85)(0.85)(0.75)(0.85)(0.85)(0.75), a 21.46% chance to get your money back, on top of the 10.5% and 1.6% to hit larger payouts.

So the expected value is effectively 0.016(175) + 0.105(11.5) + 0.2146(1.2) - 0.6644(1) = 3.60. Let me contextualize this result for you if you do not know what it means. For every $1 you put in YOU EXPECT TO RETURN $3.60. This is ABSOLUTELY crazy and almost completely unheard of in the betting industry. And I was even being generous, I likely deflated the percentage chances of each scenario relative to my results for these calculations, so in reality the expected value could be even higher.

I would say that it is very plausible to predict match winners at around 75-80%, and at those numbers you would expect to return $4.17 and $4.88 per $1 invested for 75% and 80% respectively. These numbers are just disgusting.

So with this in mind, I started exclusively betting these anticorrelation slips for a while because I thought they were my biggest edge. Over a few weeks, I hit a few 5 out of 6s, and the 11.5x payouts were nice, but I really wanted the 175x. And on July 2nd, 2024, I hit it. But the high was pretty short-lived, because when I woke up the next morning to do my research and place bets, the payouts for my anticorrelation bets had already gotten changed. They dropped from 175x/11.5x/1.2x to 58.98x/6.13x/0.58x for 6/5/4 picks correct, which is the number that they still stand at today. And while, in theory, the expected value is still positive ($1.05 for $1), I do not feel compelled to bet these anticorrelation slips anymore because the reward is so minimal for bets with such long odds.

I sort of knew that the moment I hit the 175x something like this was going to happen. A win that big would definitely catch the company’s attention and something would have to be changed. I just did not expect the change to be that fast, or the payout shift to be so severe. However, I do think I was fortunate to hit my 175x before Underdog noticed the incorrectly high profits I was making on 5 out of 6s to make payout changes.

It was at that moment that I finally knew I was a sharp bettor. I found a massive edge in the market, and the sportsbook nerfing their payouts so hard verified that my hypothesis was correct. And it felt quite rewarding to see my theory prove to be successful in real time.

Avarice is forever a legend in my book for creating massive imbalance in the second game of this slip by completely taking over botside to cash my 175x.