> ## Content Index
> Fetch the complete content index at: https://longtalkshort.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Gambler Who Beat the Horses Explains Why Almost Everyone Else Loses
- URL: https://longtalkshort.com/bill-benter-what-are-my-odds-math-of-beating-the-odds/
- Published: 2026-10-25T16:20:03.000Z
- Updated: 2026-10-25T16:20:03.000Z
- Description: Bill Benter built one of the most successful betting operations in history on Hong Kong horse racing. In a 2004 lecture he explains the math behind it: expected value, the law of large numbers, Kelly sizing and a model of every horse. Plus the real money figures.
- Author: Truman
- Tags: Economics, Bill Benter

**Almost everyone who gambles loses over time. Bill Benter won, and his operation's winnings came close to a billion dollars.**

He says the difference started with a question a French gambler asked in 1654.

In December 2004, Benter stood before the International Congress of Chinese Mathematicians in Hong Kong and explained, in a talk called "What Are My Odds?", how he beat one of the hardest games on earth. Most of the lecture is 17th-century mathematics. The payoff is the computer model he used to bet on Hong Kong horse racing.

*In short: every gamble comes down to probability, expected value and the law of large numbers. Only bets with positive expectation win over time. The Kelly criterion tells you how much to stake. And horse racing can be beaten by modeling each horse's chance of winning from data.*

## 1\. Gambling invented probability

A Paris gambler, the Chevalier de Méré, was puzzled. Betting on **at least one six in four rolls of a die** paid off. Betting on **double sixes in 24 rolls of two dice** did not. Why?

He took the puzzle to Blaise Pascal. Pascal's letters with Pierre de Fermat laid the foundations of probability theory. Three ideas came out of that era:

- **Probability:** the share of equally likely outcomes that count as a win, or more generally the long-run frequency of success.
- **Expectation** (Christiaan Huygens): the probability-weighted average of the outcomes.
- **The law of large numbers** (Jacob Bernoulli): over many repeated trials, results converge on that expectation.

Those three ideas sound academic. Benter turns them into a single, brutal rule.

## 2\. Rule number one: only positive expectation wins

Take roulette. An American wheel has 38 numbers and pays 35 to 1\. So **each $1 bet has an expectation of minus 5.26 cents.**

Because expectation is additive, a whole night of bets is simply the sum of the bets. Hence his first rule:

> the only way to achieve a long-term expected profit in gambling is to make a series of bets that have a net positive expectation.

But if the math is that clear, why does anyone ever walk away a winner? Luck, in the short run.

Picture a coin that lands heads **51 percent** of the time, $1 won or lost per flip:

- **After 100 flips:** you're expected to be only about two units ahead, and swings of 10 or 20 either way are normal.
- **After 10,000 flips:** about 100 units up, and ahead roughly five times in six.
- **After 100,000 flips:** even a bad run leaves you ahead.

Flip the edge to the house and the same math works against you. A heavy gambler playing about a thousand hours a year in Macau, Benter says, "will certainly be behind."

![Three cards. Expected value at roulette: 1 in 38 to win 35 units, 37 in 38 to lose 1, so each bet loses 5.26 cents per dollar. Expected value for a coin at 51 percent: each even-money bet gains 2 cents per dollar. Kelly sizing: stake 2p minus 1, or 2 percent of the bankroll, $20 of $1,000, recalculated after every result. A footer states the rule that only a series of positive-expectation bets yields long-term profit.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/10/fig2-14.png)

Benter's two rules, using the talk's own roulette, coin and Kelly examples. Concept diagram. Made for Long Talk Short.

Having an edge is only half the problem. The other half is how much to risk on it.

## 3\. How much to bet: Kelly

Bet too little and an edge barely grows your money. Bet too much and a bad streak wipes you out. Benter's answer comes from **J. L. Kelly Jr.**, a Bell Labs researcher who in the 1950s applied information theory to how fast a gambler's capital can grow.

In an even-money game, **the Kelly stake is 2p minus 1 of your bankroll**, where p is your chance of winning. For the 51 percent coin, that's 2 percent: **$20 out of $1,000, recalculated after every win or loss.**

Theory was in place. What turned Benter from student to professional was a book.

## 4\. The book that sent him to Las Vegas

Blackjack's rules are simple, but tracking the cards left in the deck creates millions of combinations. In 1962 **Edward Thorp**, working with IBM computers, published *Beat the Dealer*, the first mathematical way to get a positive edge at a casino game.

Benter read it, **left university, moved to Las Vegas and became a professional gambler.**

Cards follow fixed rules, though. His real fortune came from a game that doesn't.

## 5\. Horse racing: modeling the real world

Racing, "nearest and dearest to my heart," is harder because there's no closed rulebook. **You have to estimate each horse's chance of winning from data.**

His model gives each horse an expected performance, a weighted sum of factors, perhaps a hundred in all, such as:

- Average past finishing position
- Days since its last race
- Weight carried
- The jockey's win rate

Actual performance is that figure plus random noise. **A horse's chance of winning is the chance its result comes out best among overlapping bell curves,** and the probabilities in each race must add up to one.

The weights are fitted to thousands of past races, finding the values that would have given the actual winners the highest odds. He then shows, "just because I'm proud of it," his cumulative profit over about 6,000 races, roughly 11 years of Hong Kong racing.

In the Q&A he adds two practical points:

- **The track doesn't gamble.** Bookmakers and the Hong Kong tote don't use such models; the tote takes a fixed cut of each pool and carries no risk.
- **Bet late, with a cushion.** Odds shift until the race starts, so he bets as late as possible and builds in a margin, since odds on likely winners "usually drop."

![Concept diagram with illustrative numbers. A model card estimates a horse at 25 percent from past form, days since its last race, weight carried and jockey record. A crowd card shows the tote odds imply 18 percent. A combined card blends them into a best guess of 22 percent, weighted by past accuracy. A terracotta edge card says to bet only if that probability times the payout beats the cost after the track's cut.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/10/fig1-14.png)

How a model and the public's odds combine into a betting edge, as described in Benter's 1994 paper. Concept diagram, illustrative numbers. Made for Long Talk Short.

## Weighing it: what holds up

**The money headlines are off.** The re-upload circulating now is titled "Bill Benter turned $180K into $900M," but Benter never mentions his winnings in the talk. Bloomberg's 2018 profile and Wikipedia say Benter and partner Alan Woods started in Hong Kong with $150,000 and had lost $120,000 of it by the end of their first season in 1986, before splitting up. The "close to a billion dollars" is Benter's own figure, as quoted by Bloomberg, and covers his operation's total winnings, not all of it his. We found no reliable source for "$900M" or "$180K."

**The talk skips his best-known idea.** His 1994 paper, "Computer Based Horse Race Handicapping and Wagering Systems: A Report," says the model alone wasn't enough. He combined its probabilities with the public's odds, treating the two as "two experts," and reported significant profits over five years.

**Copying him isn't realistic.** Bloomberg puts the Hong Kong tote's take at about 17 percent, so a model must beat the crowd by a wide margin just to break even. Benter's operation had a staff, years of hand-collected data and huge stakes. Others now run models too, and he says late odds shifts eat into profit. And Kelly assumes your probabilities are right; overestimate your edge and full Kelly leads to deep drawdowns.

**The real lesson cuts the other way.** His own charts show that negative-expectation games are near-certain losers over time. That is the position of almost everyone who bets.

## Long Talk Short

- Every gamble rests on three old ideas: probability, expected value and the law of large numbers.
- Expected value adds up across bets, so only a series of positive-expectation bets can profit over time.
- Luck dominates the short run; over thousands of bets, results converge on the expectation, for the house or for you.
- Kelly sizing stakes a fraction of your bankroll in proportion to your edge: 2 percent for a 51 percent even-money bet.
- Benter's racing model estimates each horse's chance from many weighted factors; his 1994 paper blends it with the crowd's odds.
- The reported figures are a $150,000 start and close to $1 billion in operation-wide winnings, not "$180K into $900M."

**The one thing to remember:** don't ask whether you'll win the next bet; ask whether the price is wrong in your favor, and by how much.

Source: ["What Are My Odds?"](https://www.youtube.com/watch?v=yvfdvqWKMJY&ref=longtalkshort.com), a talk by William Benter at the International Congress of Chinese Mathematicians, Hong Kong, 2004, as re-uploaded on YouTube by Victor Ivanov. This is an independent summary of that talk and is not affiliated with the speaker, the congress or the channel.

*This article is for general information only and is not gambling, financial or investment advice. Most gamblers lose over time. If gambling is causing you harm, contact a local support service; in the U.S., call or text 1-800-GAMBLER.*