The Setup
Bill Benter is a mathematician who earned approximately $1 billion betting on horse races in Hong Kong from 1987 to 2001, averaging roughly $70 million per year. His approach combined a statistical model with over 100 variables per horse (speed figures, jockey performance, track conditions, weather, post position bias, and dozens more) with strict Kelly criterion position sizing.
The model did not predict the winner of each race. It estimated the probability of each horse winning, compared those probabilities to the odds offered by the parimutuel betting pool, and identified situations where the model's probability was higher than the market's. These were the positive expected value bets.
Why Fractional Kelly Was Essential
Benter has publicly stated that even the best computer handicapping models can overestimate edge by a factor of 2. This means if the model says you have a 10% edge, the true edge might be only 5%. At full Kelly sizing, a 2x overestimate of edge leads to catastrophic over-betting. Benter used fractional Kelly (roughly half Kelly) to protect against this systematic overestimation.
The result was lower return than full Kelly would have produced in theory, but dramatically higher survival probability. Given that his career spanned 14 years of continuous operation, survival was clearly the right priority. Full Kelly would have produced higher returns in a hypothetical world of perfect probability estimation, but in the real world of uncertain edges, it would have likely blown up his bankroll at some point during those 14 years.
Lessons for Prediction Market Traders
Benter's approach maps directly onto prediction market trading. Build a model that estimates probabilities for events (not just horse races, but any prediction market contract). Compare your model's probability to the market price. Identify positive EV opportunities where your probability exceeds the market price by a meaningful margin. Size positions using fractional Kelly to account for model uncertainty.
The discipline required is the same: accept that individual bets will frequently lose, trust the process across hundreds of bets, size conservatively to survive the inevitable losing streaks, and continuously refine the model based on outcomes.
The $1 Billion Reinforcement
If there is one takeaway from Benter's career, it is that small, consistent edges compounded over thousands of bets produce extraordinary results. His per-bet edge was small. His per-race expected return was modest. But the compounding across thousands of races over 14 years, combined with disciplined sizing that prevented catastrophic losses, produced a billion dollars. The same mathematics applies to prediction markets, crypto trading, and any other probabilistic endeavor.
Explore these tools on Blockcircle: Prediction Markets Mispricing Engine