The Kelly Criterion: The Math of Betting Big When the Odds Are in Your Favor

Overview

The Kelly criterion is one of the most famous sizing rules in finance and gambling because it answers a very specific question: what fraction of capital should you bet to maximize long-run growth?

That is the core idea. Not what to buy. Not when to buy. How much to bet once you believe you have an edge.

The rule originated in a 1956 Bell System paper by J. L. Kelly Jr., and was later pushed into investing and gambling by Ed Thorp. The central result is that Kelly sizing maximizes the expected logarithmic growth rate of wealth over repeated favorable bets.

This is the core Kelly visual. The curve shows the expected log growth rate g(f) for different bet fractions, given a 55% win probability and 2:1 payout odds. Growth rises to its peak at the Kelly fraction f*, then turns down sharply. The overbetting zone makes the asymmetry obvious: betting too much is far more destructive than betting too little.

This simulation runs 300 sequential bets at 55% win probability with 2:1 payout odds under three sizing strategies. The median equity path shows that full Kelly maximizes long-run geometric growth, half Kelly grows slower but much more smoothly, and double Kelly often severely underperforms or blows up.

Same three strategies, same edge — but now showing the median drawdown path instead of equity. Half Kelly keeps drawdowns shallow. Full Kelly endures meaningful but recoverable dips. Double Kelly suffers persistent, deep drawdowns that make the strategy nearly impossible to stick with.

For a simple bet with probability of winning p, probability of losing q = 1 − p, and net odds b to 1, the Kelly fraction is f* = (bp − q) / b. This tells you the fraction of capital to risk.

If the result is negative, the bet is not favorable and Kelly says do not bet. That is already one of the deepest insights in the framework. Kelly is not aggressive by default. It is only aggressive when the edge is real.

For an even-money bet where b = 1, the formula simplifies to f* = 2p − 1. So if your win probability is 55%, Kelly says bet 10% of capital.

Kelly is not trying to maximize expected profit in a single bet. It is trying to maximize the compound growth rate of capital across many repeated bets.

That means the objective is max E[log W] — not max E[W]. That difference is everything.

A strategy that maximizes expected dollar profit can still blow up. A strategy that maximizes log wealth automatically respects the fact that large losses are much more damaging than large gains are helpful.

Consider: a −50% loss followed by a +50% gain does not get you back to breakeven. You end up at 0.75 of your starting capital. To recover from a 50% loss, you need +100%. Kelly internalizes that asymmetry directly through the log function.

Kelly says there is a tradeoff between edge and overbetting. If you underbet, you grow too slowly — positive growth, but not maximal. If you overbet, the volatility drag becomes so strong that long-run growth actually falls.

At some point, overbetting becomes catastrophic. This is the real lesson of Kelly: even with a genuine edge, betting too much can destroy the advantage.

That is why Kelly is one of the most insightful sizing frameworks ever developed. It separates having an edge from surviving long enough to exploit it.

In investing, the clean casino formula usually gets replaced by a continuous approximation. The growth-optimal allocation for a risky asset in a simplified setting is f* = (μ − r) / σ², where μ is expected return, r is the risk-free rate, and σ² is return variance.

The intuition is simple: optimal leverage rises when expected excess return rises and falls when variance rises. So Kelly is really saying size is proportional to edge divided by risk.

That is why it remains so influential in quantitative finance.

Kelly is mathematically elegant, but in practice it is very easy to misuse because the formula depends on inputs you do not know perfectly: p, μ, σ, and correlations.

If your edge estimate is too high, Kelly tells you to bet too much. If your volatility estimate is too low, same problem. Estimation error is one of the biggest practical limitations of the Kelly criterion.

This is why many investors prefer fractional Kelly — typically half Kelly or quarter Kelly. Fractional Kelly sacrifices some theoretical growth in exchange for shallower drawdowns and more robustness to bad estimates.

This is maybe the single most important Kelly insight. If you bet less than Kelly, growth declines gradually. If you bet more than Kelly, growth can deteriorate much faster.

You can think of the growth function as g(f) = E[log(1 + fX)]. It rises until f = f*, then turns down. The key is that g(0.5f*) is usually only slightly less than g(f*), while g(1.5f*) can be much worse.

That asymmetry is why half-Kelly is so popular. That is the practical reason many professionals treat Kelly more as an upper bound than as a mechanical target.

This is another critical point people often miss. Kelly does not tell you how to find alpha. It assumes you already have a positive expected value bet.

So the real stack is: signal + edge estimate + Kelly sizing. Not: Kelly = trading strategy.

That is why Kelly pairs well with things like sports betting models, stat arb, options mispricing, systematic trading signals, card counting, and portfolio optimization. It is a capital allocation framework, not an alpha source.

One reason Kelly feels uncomfortable in real life is that full Kelly can produce severe drawdowns even when it is mathematically optimal for long-run growth.

Growth-optimal does not mean smooth. Kelly can have very large equity swings. This makes Kelly hard to use for client capital, institutional mandates, high-volatility strategies, and anyone with behavioral limits.

In practice, many investors prefer a lower-growth, lower-drawdown path. The tradeoff between maximum long-run growth and interim comfort is real and unavoidable.

The Kelly criterion is one of the clearest frameworks ever created for thinking about position size. Here are the core principles that make it so powerful.

A clean summary of the Kelly philosophy captures why this framework still matters decades after its invention.

The Kelly criterion is one of the clearest frameworks ever created for thinking about size. Its core message is: bet more when edge is larger, bet less when uncertainty and volatility are larger, and never overbet just because you are confident.

That last point is the deepest one. Kelly is a warning against ego as much as it is a formula. It turns position sizing from guesswork into a mathematical discipline.

The practical takeaway is not to use full Kelly mechanically. It is to understand that there is always an optimal sizing level for any edge, and going beyond it — no matter how convinced you are — makes your results worse, not better. That insight alone is worth the entire framework.