Don Johnson: $15 million from Atlantic City, and the 20% rebate that flipped the edge

Twelve hours at the Tropicana

The hand everybody retells is the $800,000 one. Johnson bet $100,000, was dealt two eights against a dealer five, split them, was dealt two more eights, split again, and then doubled on all four hands when the small cards came: $800,000 on the felt. The dealer drew to 15 and busted with a ten. He did not pause; he signalled for the next hand. By his own account he was playing almost a hand a minute for twelve hours, and in a three-hand stretch took $1.2 million. When the session ended he was up $5.8 million, which the casino’s own filing to the state turned into a public story.

That filing is the part of the legend that can be checked. Atlantic City casinos report monthly win to the Division of Gaming Enforcement, and the April 2011 release shows what one player did to one casino’s month. The Tropicana’s table-game win was $3.19 million against $7.00 million in April 2010, a 54.4% fall, while the state’s eleven casinos together were down 15.4% at the tables. Its slot win fell 5.6%. The Press of Atlantic City connected the dots from those numbers, and the headline said a single player had ruined the casino’s month.

Casino win is what the house kept after paying players. The Tropicana’s April table win was the second-lowest of the eleven casinos. Johnson’s $5.8 million came out of this line; the Borgata and Caesars sessions fell in earlier months and are not visible here.

The run was not one night. In December 2010 he had won $4.23 million at Caesars, where the dealer eventually declined to refill the chip tray and the pit boss pointed him to the main floor. Over about five months he took roughly $5 million from the Borgata. Caesars, he said, cut him off and then effectively barred him from its casinos worldwide; the Borgata and the Tropicana let him keep playing but withdrew the terms. Mark Giannantonio, the Tropicana’s president, who had authorised the $100,000-a-hand limit, was gone weeks after the April filing.

The deal was signed before the first card

Johnson was known to the casinos. Their own research had him as a skilled, well-funded player, and he was watched closely enough for card counting to be ruled out. He had spent his working life around betting: a jockey at fifteen, then racetrack management, then regulator in four states, then a company doing computer-assisted horse wagering. He had played blackjack seriously for about ten years without ever beating the Borgata. What changed in late 2010 was not his play. It was Atlantic City’s revenue, down year-on-year for 32 consecutive months by April 2011, and the discounts its marketers were willing to offer to keep the biggest players coming.

A high-roller “discount” is a rebate on losses: lose $100,000 under a 10% discount and you owe $90,000. Johnson had earlier held a “lifetime” discount at the Borgata, which is worth little to a winning player, because the rebate only starts once you have lost back all your previous winnings plus the threshold. In late 2010, he says, the Borgata called and offered to put him on a trip-to-trip basis instead: every visit starts at zero. Caesars and the Tropicana, competing for the same player, matched. That one change is most of the story.

From Johnson’s account to The Atlantic (he would not disclose every clause of the e-mailed agreement). A standard six-deck game with ordinary rules runs about 0.4–0.6% against basic strategy; the engine figures for Betkyo’s own rules are in the verified block below.

Notice what he did not ask for. No free chips, no suites, no jet: the Tropicana’s CEO called him a cheap date. Every concession he took was a number he could put into the edge or into the rebate, and he could compute what each was worth. The casinos had models for this too. The Tropicana’s was, in its CEO’s words, elaborate: plug the customer in, estimate the house advantage from the game and how he plays it, set the discount from that. Asked how the models produced a deal he himself called a huge edge, Johnson’s answer was that somebody missed the math.

Why a rebate on losses is not a smaller edge but a different sign

Start with the game alone. At a 0.25% house edge, a $100,000 hand costs $250 in expectation; at a hand a minute that is about $15,000 an hour, $180,000 across a twelve-hour session. Against that, the standard deviation of a blackjack hand is about 1.15 bets, so 720 hands at $100,000 swing by roughly $3.1 million either way. Johnson’s +$5.8 million is a little under two standard deviations above the expected −$180,000: a one-in-forty night, lucky but not absurd. The edge is real and the casino would collect it over enough nights. The rebate changes what “enough nights” costs.

A rebate on losses pays only when you lose, so its value depends entirely on how you stop. Johnson’s rule was explicit: if the session reaches −$500,000, stop and pay $400,000; if the session goes well, keep playing, because you are now betting the casino’s money with its own discount behind you. Treat the game as a fair coin (the 0.25% is small against these swings) and ask what one attempt is worth if you quit at a target T or at −$500,000, whichever comes first.

With no edge, the chance of reaching +T before −L is L ÷ (L + T). A loss is capped at $500,000 and rebated to $400,000; a win is T. The expected value is 0.2 × L × T ÷ (L + T), which is positive for every target and approaches $100,000 (20% of the $500,000 threshold) as the target grows. The 0.25% edge subtracts about $250 per hand played, tens of thousands over the few hundred hands a session typically lasts: enough to trim these numbers, not enough to change their sign.

Read the table slowly, because it says something uncomfortable about what a winning player looks like. Under the $5.8 million target, Johnson loses $400,000 on more than nine attempts in ten. He is not a man who wins most nights; he is a man whose rare winning nights are enormous and whose common losing nights are capped. The rebate is an option on variance, and an option is worth more the wilder the underlying. That is why the small bet limit and the small edge were not the point: $100,000 hands at a quarter-percent edge give the maximum swing for the minimum drag, and the discount pays for the swing.

Two details of the deal matter more than any rule of play. The threshold had to be per trip, so that each session starts the option fresh instead of first eating previous winnings. And the money won at one casino was never exposed to that casino’s rebate: what he was up at the Borgata could not be lost back at the Borgata on rebated terms, because he stopped there and moved to Caesars, then to the Tropicana, each offering its own $500,000 floor. A rebate that only applies to losses is a rebate you use by leaving while ahead.

What the model missed, and what a fixed edge cannot give away

The Tropicana’s model, as its CEO described it, priced the player: his game, his skill, his expected loss per hour. From that it derived a discount. What it did not price was the player’s stopping rule. A loss rebate assumes the customer plays on through the losses, so that the rebate is a share of a loss that will happen; a customer who stops the moment the rebate vests and keeps playing only when the house is paying has turned the discount into an option, and options are priced on variance, not on expectation. The 20% looked like a fifth of a predictable loss. It was a fifth of a $500,000 floor under an unbounded ceiling.

There is also a plainer failure: the discounts had crept from 10% to 20% as revenues fell, and the limit had been raised to $100,000 a hand, each decision made by a marketer competing with the casino next door. Johnson’s own summary was that the casinos had started swinging for the fences. The same month the filings showed what a swing and a miss costs, the executive who signed the limit was gone, and by the time The Atlantic wrote the story in 2012 most Atlantic City properties would not deal to him at all. The Tropicana still did, with modified discounts; he says he won another $2 million there that October on the same rules without a rebate.

That last sentence is the reason this story belongs in a casino-math journal rather than a sports page. Johnson did not beat blackjack; he beat a negotiation, and the arithmetic says anyone offered the same terms and the same discipline would have had the same positive expectation. A provably fair original has nothing to negotiate. Its edge is a line in the derivation, the same for a $1 hand and a $100,000 one, and its result is a function of a seed committed before the bet. A rebate on wagers, which pays whether you won or lost, scales that edge down by a fixed fraction and can never flip it. A rebate on losses, as Atlantic City learned in April 2011, is a different instrument entirely, and the only safe way to offer one is to price it the way Johnson did: as an option, on the variance, for a player who will stop exactly when it tells him to.