1. History: John L. Kelly Jr. and Information Theory
The Kelly Criterion, formulated by John L. Kelly Jr. in 1956, originated at Bell Labs. Kelly was a physicist and researcher working on information theory alongside Claude Shannon. The core idea was to maximize the long-term growth rate of a bankroll in scenarios where an investor or bettor has an edge, drawing parallels between gambling and transmitting information over a noisy channel.
Unlike traditional staking systems, Kelly's approach mathematically guarantees the highest possible bankroll growth over infinite bets, provided the estimated edge is perfectly accurate.
2. The Kelly Formula and Derivation
The classic Kelly formula determines the optimal fraction of your bankroll (f*) to wager.
Where:
- b is the net decimal odds minus 1 (e.g., if odds are 2.00, b = 1).
- p is the true probability of winning (e.g., 0.55).
- q is the probability of losing (1 - p = 0.45).
Deriving this requires calculus, specifically maximizing the expected geometric growth rate (or the expected logarithm of wealth). Let W be wealth. We maximize E[ln(W)] = p cdot ln(1 + fb) + (1-p) cdot ln(1 - f). Taking the derivative with respect to f and setting it to zero yields the Kelly fraction.
3. Full, Half, and Quarter Kelly
Warning: Full Kelly is highly aggressive. A simple streak of bad luck, combined with imperfect probability estimation, can severely deplete a bankroll.
To mitigate volatility, bettors use fractional Kelly:
- Full Kelly: Maximizes growth but entails massive volatility.
- Half Kelly: Betting 50% of the recommended f*. Reduces growth by 25% but decreases variance by 75%.
- Quarter Kelly: Betting 25% of f*. Significantly smoother bankroll curve.
4. Worked Example
Imagine a scenario: Odds = 2.00, True Probability (p) = 55%, Bankroll = $1000.
First, calculate b = 2.00 - 1 = 1.
q = 1 - 0.55 = 0.45.
f* = (1 * 0.55 - 0.45) / 1 = 0.10.
The Full Kelly recommendation is to bet 10% of the bankroll, which is $100. A Half Kelly bettor would wager $50.
5. Risk of Ruin
The theoretical risk of ruin with Full Kelly is 0% if the bankroll is infinitely divisible and the edge is real. However, because minimum bet sizes exist, and edges are estimates, ruin is possible. Monte Carlo simulations consistently show that overestimating your edge is the primary cause of bankruptcy when using Kelly.
6. Why Professionals Use Fractional Kelly
Estimation error is inevitable. If a bettor believes their probability is 55% but it's actually 48%, a Full Kelly bet is destructive. Fractional Kelly (often 0.25 to 0.5) provides a safety cushion, allowing survival even if the model's edge is slightly miscalculated.
7. Common Mistakes
- Overestimating the edge: The most fatal error.
- Simultaneous bets: Standard Kelly assumes sequential bets. For simultaneous bets, complex covariance matrices are required.
- Parlays: Applying Kelly to parlays multiplies estimation errors exponentially.
8. Tilt Protection
Info: Kelly inherently protects against tilt because bet sizes scale dynamically with the bankroll. As you lose, your absolute bet size decreases.
9. Calculator Tool
Test these concepts using our Kelly Criterion Calculator.