The Kelly Criterion
The bet-sizing formula Thorp adopted from information theory — and the single idea that runs through every chapter of his career, in casinos and in markets alike.
Having established that a real, positive edge existed, Thorp faced a second, separate problem: even with a genuine edge, betting too large a fraction of a bankroll on any single hand risks ruin from an unlucky streak, while betting too small leaves most of the edge's value uncaptured. His solution was the Kelly criterion, a formula originated by Bell Labs scientist John Kelly for a different problem (information transmission rates), which Thorp recognized could be applied directly to bet sizing: bet a fraction of the bankroll proportional to the size of the edge, which mathematically maximizes the long-run growth rate of the bankroll while keeping the probability of ruin at zero.
The book explains why this matters beyond blackjack: Kelly sizing became Thorp's central risk-management discipline for the rest of his career, including decades later in his hedge funds — the same underlying logic (size a bet in proportion to edge and bankroll, never bet so large that a string of losses can wipe you out) applied whether the "bet" was a blackjack hand or a market position, which the book presents as the throughline connecting the seemingly very different halves of Thorp's career.
The book is candid that Kelly sizing is psychologically uncomfortable in practice: the formula still permits large, painful-looking drawdowns on the way to its long-run growth guarantee, and many players who understood the math intellectually still deviated from it under real pressure, either over-betting during winning streaks out of overconfidence or under-betting after losses out of fear — both of which sacrifice the very growth the formula is designed to protect.
The full formula generalizes to any edge/odds combination; the core principle is that bet size should scale with the size of the edge relative to bankroll — betting more than this fraction increases risk of ruin faster than it increases long-run growth, and betting less leaves growth on the table.
A subtle but critical point the book returns to repeatedly: naively maximizing expected value on each individual bet, without regard to bankroll, can actually lead to ruin over a long enough series of bets, because a single catastrophic loss can end the game entirely, at which point no future edge can be captured. Kelly sizing instead maximizes the long-run compound growth rate of the bankroll — a subtly different and, for anyone playing repeatedly over time, more important goal, since it is specifically constructed never to risk complete ruin regardless of how unlucky any specific stretch gets, as long as the underlying edge estimate itself is correct.
Because full Kelly sizing can still produce uncomfortably large swings even when the math is correct, the book notes that Thorp and most other practitioners typically bet a fraction of the full Kelly amount — commonly somewhere around half — deliberately trading away some theoretical long-run growth for meaningfully smaller drawdowns and a much lower sensitivity to an overestimated edge. This matters because the formula assumes the edge estimate itself is exactly correct; in the real world an edge is always an estimate, and betting full Kelly on a slightly-too-optimistic edge estimate can push actual risk of ruin meaningfully higher than the formula promises, which fractional Kelly partially protects against.
- The Kelly criterion sizes each bet as a fraction of bankroll proportional to the size of the edge, maximizing long-run growth while keeping ruin risk at zero.
- This became Thorp's central risk-management discipline for his entire career, applied identically to blackjack hands and market positions.
- Maximizing long-run compound growth is a different, and for repeated betting more important, goal than naively maximizing expected value on each individual bet.
- Full Kelly sizing is psychologically hard to stick with — many players deviate under real pressure, which sacrifices the growth the formula is designed to protect.
- Most practitioners, including Thorp, bet a fraction of full Kelly to protect against an edge estimate that turns out to be too optimistic.