Edward Thorp, The Man Who Beat The Odds

Edward Thorp is an American mathematician who proved his ideas by making money with them. He was born on 14 August 1932 and took his doctorate in functional analysis at UCLA in 1958. He taught at the Massachusetts Institute of Technology, worked out how to beat blackjack, and built what is widely regarded as the first wearable computer with Claude Shannon. Then he ran one of the earliest quantitative hedge funds.

One question runs through all of it. It is not whether to bet, but how much to bet once you know the size of your advantage. The answer he helped make rigorous is the Kelly criterion, a formula that sets the stake which compounds money fastest and warns that anything larger ends in ruin.

He came to the casinos from pure mathematics, not the other way round

Thorp’s doctorate was in functional analysis, an abstract corner of mathematics with no visible connection to a casino. He taught at MIT from 1959 to 1961, then held professorships at New Mexico State University and the University of California, Irvine, before leaving academia to invest full time. The blackjack work came in the middle of it.

His 2017 memoir A Man for All Markets is the standard account of the whole career, and it reads like three or four careers stacked one on top of another. He turned 94 in August 2026.

Shannon handed Thorp the paper that made his career

Edward Thorp and Claude Shannon overlapped at MIT. Shannon’s Bell Labs colleague John Kelly had published a paper in 1956 showing how to use information to decide the size of a bet, and Shannon brought it to Thorp’s attention in 1960. The two then built a device to beat roulette. Edward Thorp had the mathematics of gambling, Shannon had the electronics, and between them they made something that had never existed before.

Shannon was not a passing acquaintance. The founder of information theory handed the young Thorp the specific result that would define his career, and then built hardware with him. Edward Thorp later took the same idea to the stock market, extending Shannon’s mathematics into a domain Shannon never worked in.

The Kelly criterion sets the largest stake that is still safe

The Kelly criterion is the technical heart of everything Thorp did. John Kelly derived it from Shannon’s notion of channel capacity in A New Interpretation of Information Rate, published in the Bell System Technical Journal in July 1956. It answers the question every gambler and investor faces, which is not whether to bet but how much to bet.

The criterion says to bet a fraction of your capital proportional to your edge, the amount by which the odds are in your favour. Bet that exact fraction and your wealth grows as fast as it possibly can over the long run. The remarkable part is what happens on either side of it. Bet less and you leave growth on the table. Bet more and you do not simply take on more risk for more reward, because past a point you guarantee your own ruin, however genuine the advantage was.

That asymmetry is the whole lesson. A real edge is necessary but not sufficient, because you can hold a true advantage and still lose everything by staking too much of your capital on it. Kelly’s formula is the line between disciplined confidence and reckless overreach, expressed as a number.

The roulette computer showed a plus 44 per cent expectation in tests

The roulette project produced a genuine landmark, widely regarded as the first wearable computer. Edward Thorp and Shannon built it and first used it in 1961. The cigarette-packet-sized device took four push-button clicks as its input, two timing the rotor and two timing the ball, and from those it predicted which section of the wheel the ball would favour.

Edward Thorp reported the results in his 1969 paper on optimal gambling systems. The expectation in tests was plus 44 per cent, meaning a gain of about 44 per cent per unit staked on a small neighbourhood of numbers on a slightly tilted wheel. That is a test figure rather than a realised casino edge. The same paper says one third of the Nevada wheels they observed had the tilt of at least 0.2 degrees the method needed. It also names the system’s ultimate weakness, since the house could forbid bets once the ball had been launched.

The machine mattered more as a proof of concept than as a money-maker. It showed that a game everyone assumed was pure chance was partly predictable to anyone willing to sit down and measure the physics of the ball and the wheel. That conviction went with him everywhere.

The 1962 book that proved blackjack was beatable

Edward Thorp’s most famous result came at the blackjack table. He set out the mathematics first in a short paper in the Proceedings of the National Academy of Sciences in January 1961, communicated by Shannon himself. Then came the 1962 book Beat the Dealer. It was the first book to prove that blackjack could be beaten by tracking which cards had been played, and by betting more heavily when the remaining deck favoured the player.

The key insight is that blackjack, unlike most casino games, has a memory. Cards already dealt cannot come again, so the composition of the remaining deck shifts as play continues, and when it shifts far enough in the player’s favour the house edge flips. Card counting is simply a way of knowing when that has happened, and the Kelly criterion tells you how much more to bet when it does.

The book was a sensation. Casinos changed their rules, introduced multiple decks and more frequent shuffling, and began barring players they suspected of counting. Edward Thorp had done something unusual for a mathematician, which was to prove a theorem so practical that an entire industry restructured itself in response.

The same method made 250 million dollars in the markets

Edward Thorp treated the stock market as a larger, richer version of the same problem. The move was deliberate. In 1967, with Sheen Kassouf, he published Beat the Market, showing how mispricings in warrants and convertible securities could be found and traded with a statistical edge.

In 1969 he put the theory to work, co-founding with Jay Regan the hedge fund that became Princeton/Newport Partners. Edward Thorp calls it the first market-neutral derivatives hedge fund. It was run on mathematical models rather than judgement about companies, finding small and reliable pricing discrepancies and harvesting them in volume while hedging the broader market risk away.

The results were substantial and quiet. Writing in Wilmott in 2003, Thorp records that Princeton/Newport Partners made some 250 million dollars for its partners, with half or more of that coming from convertible hedging. In the same article he reports the results of the statistical arbitrage programme he launched in August 1992. Through October 2002 it compounded at 26 per cent a year before his performance fee, and 20 per cent a year net to investors. Treating markets as a source of measurable statistical edges is the approach that later firms, including D. E. Shaw, built into an industry.

High cards left in the deck are what flip the house edge

The blackjack method is the clearest example of what Thorp means by an edge. The player assigns a small value to each card as it appears, keeping a running tally that rises when low cards are dealt and falls when the high cards leave the deck. It is one number, not a memorised deck.

The reason it works is that high cards favour the player. They make blackjacks, which pay extra, more likely, and they make the dealer bust more often. When the running count shows that the remaining deck is rich in high cards, the odds have tipped toward the player, and that is the moment to raise the bet. When the deck is poor, the player bets the minimum and waits.

No system wins every hand. The technique makes the player the favourite over many hands, and only the disciplined sizing of bets across thousands of hands turns that small edge into reliable profit. Individual results still swing wildly. The gap between having an edge and realising it is the same gap that separates a promising quantum result from a useful quantum computer.

He had the option pricing formula before Black and Scholes published

One episode captures how far ahead of the field Thorp often was. In his 2003 Wilmott article he says he had guessed the Black-Scholes formula in late 1967 and used it to trade warrants from that year on. Black, Scholes and Merton published theirs in 1973, about six years later.

He never published his version. The published result earned lasting fame, and in 1997 the Nobel Prize went to Scholes and Merton, since Fischer Black had died in 1995 and the prize is not awarded posthumously. Edward Thorp used the mathematics quietly instead, to make money. In the same 2003 article he raises no question of credit, and lists Myron Scholes among the interesting people he met along the way.

The same arithmetic told him Madoff could not be genuine

Edward Thorp’s analytical eye also worked in reverse, detecting edges that were too good to be real. By his own account in his memoir A Man for All Markets, he was asked to evaluate an investment manager whose reported returns were suspiciously smooth, and he examined the numbers. His conclusion was that Bernie Madoff’s operation could not be genuine. That was years before the scheme collapsed in 2008 and was exposed as the largest Ponzi scheme in history.

The reasoning was the same discipline in a different direction. Madoff’s reported returns implied an edge and a consistency that no honest strategy could produce. In his 2003 Wilmott article Thorp says the two risk questions he asks of any investment are what the factor exposures are and what the risks from extreme events are. Reputation is not on that list. When a claimed advantage is statistically impossible, the correct conclusion is not that a genius has been found but that something is wrong.

He could always state the size of his edge as a number

The popular image of Thorp as a gambler undersells the mathematician. He came at blackjack and at markets with the tools of a working researcher, and not with those of a man selling a system. That distinction is where his credibility comes from.

This matters for how his results should be read. Beat the Dealer was not a collection of tips but a proof, worked out with the early computers available at MIT, and its conclusions were checked and reproduced by others. When Thorp claimed an edge he could state its size and defend the calculation, which is why his claims held up when countless gambling systems did not.

It is the same standard that separates real from illusory results in quantum computing. The field moves on proofs and on measured resource costs, and an impressive demonstration on a convenient example is neither of those things. The claims that survive are the ones whose authors can say exactly what was shown and how large it was. Edward Thorp always could.

An edge is measured before the money goes down

Underneath every part of Thorp’s career is a precise idea of what an edge is. It is a measurable and repeatable statistical advantage, quantified before any money is staked, and it is not a hunch or a story about why something ought to work. The word gets abused constantly.

Card counting is an edge because the deck composition can be counted exactly. The roulette computer created one because the physics of the ball can be timed. The convertible-bond trades were edges because the mispricings could be measured against a model. In every case Thorp knew the size of his advantage as a number before he committed anything. His account of his first Nevada trip shows the other side of it, since his backers proposed a bankroll of 100,000 dollars and he cut it to 10,000 “as a personal safety measure”.

This is a stricter standard than most claims of advantage meet, in any field. It rules out the plausible story, the demonstration on a convenient example and the confident projection, since none of those is a measured edge. Few claims survive it, which is the point.

His questions are the right ones to ask of a quantum computer

Quantum computing is, from an investor’s or a reader’s point of view, a question about edges. Where does a quantum computer hold a real, measurable advantage over the best classical alternative, how large is that advantage, and on which specific problems does it exist? Those are Thorp’s questions, transposed into a new domain.

Read that way, the honest picture becomes clear. There are a few places where quantum computing has a genuine and in some cases provable edge, such as factoring and the simulation of quantum systems, discussed across our coverage of quantum algorithms. There are many more places where the claimed edge is unproven, unmeasured, or has evaporated on closer inspection, which is exactly the situation Thorp trained himself to detect and avoid.

The value of his example is that it supplies a test rather than an opinion. The question is not whether quantum computing is exciting. It is where the measured edge sits and how big it is, which is the only question that separates a real opportunity from an expensive story. Understanding what quantum advantage actually requires is the same discipline applied at a different table.

Overbetting a real edge still ends in ruin

The second half of Thorp’s lesson is easier to forget. Finding an edge is necessary, but sizing the commitment correctly is what keeps you in the game long enough to collect on it. Edward Thorp is blunt about this in his 2003 Wilmott article, where he writes that the Kelly criterion “shows that consistent overbetting eventually leads to ruin”.

Applied to a technology, this argues for proportion. A narrow but genuine quantum advantage justifies a narrow but genuine commitment, not the all-in enthusiasm that treats every advance as the one that changes everything. Overcommitting to a real edge is still a route to ruin, whether the edge is a card count or a class of quantum algorithm. The history of technology is full of sound ideas backed so recklessly that the backers did not survive to collect.

Thorp’s career is a long demonstration that the disciplined middle, real edges backed in proportion, beats both timidity and excess over time. It is an unglamorous conclusion, and it is the correct one.

The betting analogy breaks where the odds keep moving

An honest profile should mark where the comparison stops, because a metaphor pushed too far misleads. Betting and technology assessment are not identical, and the Kelly criterion is a tool for repeated bets with known odds, whereas a bet on a technology is often a one-off with genuinely unknown probabilities.

Quantum computing also differs from a casino game in that the odds themselves are moving. A problem with no quantum advantage today may gain one as algorithms improve or hardware changes. The edge is not a fixed quantity to be measured once, but a shifting one to be reassessed, and Thorp’s world of stable calculable edges is cleaner than that.

These limits do not undo the lesson, they sharpen it. Precisely because the probabilities are harder to pin down, the discipline of demanding a measured edge and sizing commitment to it matters more, not less. The danger in a fast-moving field is to abandon the discipline altogether and bet on the story, which is the one move Thorp’s entire career argues against.

The method travels further than the subject

Thorp’s discipline is a short list. Insist on a measured edge rather than a story, and establish exactly how large that edge is before committing anything at all. Refuse the bet when the advantage cannot be quantified. Even with a real edge, size the commitment so that being wrong is survivable. None of that is a quantum idea, and all of it is the fastest way to read a quantum claim honestly.

The chain begins with Shannon, and that is not a coincidence. The same mathematics that underpins quantum information theory also produced, by way of Kelly and Thorp, a method for deciding how much to believe any claim at all. The theory and the scepticism come from one source.

Biographical and performance details here come from Edward Thorp’s own website, his memoir and his 2003 Wilmott article on quantitative finance. His papers on the roulette computer and the Kelly criterion are collected on his articles page.

Frequently asked questions