Blacklisted by US casinos, physicist Bill Benter used a 100+ variable model and Kelly sizing to win a billion dollars at the Hong Kong track over 17 years — on an information edge of just 2-5 percentage points.
Racing is a multi-outcome bet: the model gives each horse a probability, compared with market odds to find the underpriced ones. Kelly turns a tiny edge (model IC≈0.12 vs market 0.10) into a bet fraction — the point isn't any single bet, it's the disciplined compounding of hundreds of thousands.
Multi-outcome Kelly · allocatetypescript
// For each underpriced outcome i, allocate by edge/odds:
// f*_i ∝ (p_i · b_i − 1) / b_i (p model prob, b odds)
// Benter: 2-5% edge × 500k+ bets × Kelly sizing = $1B
◆The book's most romantic win
"It was those 2-5 points of edge — through hundreds of thousands of repetitions and Kelly's amplification — that compounded into a billion dollars."