Introduction to the Kelly Criterion

Among professional bettors and asset managers, the Kelly criterion is considered the holy grail of position sizing. It answers a deceptively simple question: given that you have an edge, how much of your bankroll should you bet on each opportunity?

The answer is a formula. Derived from information theory in the 1950s, the Kelly criterion tells you the exact position size that maximizes your long-term wealth while minimizing the risk of ruin.

The problem: for stock investors, using the formula at full strength is often a path to ruin instead.

The Kelly criterion is a mathematical formula for calculating the optimal position size given your historical win rate and edge. The formula is: f* = (bp - q) / b, where f* is the fraction of bankroll to bet, b is the odds, p is the win probability, and q is the loss probability.

Key Takeaways

• The Kelly criterion mathematically maximizes long-term growth. If your estimates are accurate, Kelly sizing generates superior compounding.
• Kelly assumes accurate estimates of edge and win rate. In practice, most investors dramatically overestimate both. Using full Kelly often leads to ruin.
• Half Kelly is often safer than full Kelly. Reducing Kelly-recommended sizing by half preserves most of the mathematical edge while dramatically reducing risk of ruin.
• Kelly works best for activities with high-frequency feedback. Blackjack, sports betting, and algorithmic trading have clear-cut wins and losses. Stock picking feedback is muddied by valuation changes and time horizons.
• Kelly is more valuable as a framework than as a literal formula. Understanding why Kelly recommends sizing can inform intuition about position sizing, even if you do not use the exact formula.

The Formula Explained

The Kelly criterion formula in its stock market form is:

f = (p × b - (1 - p)) / b*

Where:

• f* is the fraction of your portfolio to allocate
• p is your estimated probability of the investment outperforming
• b is the expected payoff ratio (wins divided by losses)

Example: You believe a stock has a 60% probability of outperforming the market over five years (p = 0.60). When you are right, it outperforms by 30%. When you are wrong, it underperforms by 15%. Your payoff ratio is 30/15 = 2 (b = 2).

f = (0.60 × 2 - 0.40) / 2 = 0.80 / 2 = 0.40 = 40%*

Kelly tells you to allocate 40% of your portfolio to this position.

Why Kelly Maximizes Growth

The logic is intuitive. When you have a strong edge (high p and high b), you should bet large. When your edge is weak, you should bet small.

Over many repeated bets, Kelly sizing asymptotically maximizes the geometric mean of your returns. This is the true measure of long-term wealth growth. It is superior to maximizing arithmetic mean returns, which ignore volatility.

Professional blackjack players use Kelly sizing. If they can count cards and detect a 1% edge, Kelly tells them how much of their bankroll to use. Professional sports bettors use Kelly. Over thousands of bets, Kelly sizing outperforms both aggressive (overbetting) and conservative (underbetting) approaches.

For stock investors, the logic is identical. If you can identify stocks with genuine edge, Kelly sizing maximizes compounding.

The Critical Assumption: Accurate Edge Estimation

Kelly has one critical assumption: you know your true edge and win rate.

This assumption rarely holds in stock investing.

When professional investors backtest strategies, they consistently overestimate historical edge. Factors they did not account for, data snooping, survivorship bias, and luck all inflate perceived edge. A strategy that tested at 2% annual alpha often delivers 0.2% alpha in live trading.

Similarly, most investors overestimate their win probability. When they estimate a 65% probability of outperformance, they average 55%. When they estimate 70%, they average 60%.

If you overestimate edge by 50% and feed inflated estimates into Kelly, you will be sized far too aggressively. A position that Kelly recommends at 25% (based on false estimates) might only deserve 10%.

The Ruin Risk Problem

Kelly sizing maximizes long-term wealth but permits temporary large drawdowns. In mathematical terms, Kelly accepts a 25% probability of a 50% drawdown at some point.

For professional bettors with access to new capital and the ability to resize instantly, this is acceptable. For individual investors with fixed capital and emotional constraints, it is not.

A 50% portfolio drawdown from Kelly-sized positions is psychologically catastrophic. Most investors abandon the strategy at the worst time (near the bottom). They lock in losses and never experience the recovery.

This is why half Kelly is popular among professionals who use any Kelly approach. Half Kelly eliminates 50% of the theoretical outperformance but reduces maximum expected drawdown from 50% to roughly 25%, and cuts the probability of ruin significantly.

Applying Kelly to Stocks

Translating Kelly to stock investing requires converting stock picks into betting language.

If you believe a stock will outperform the market by 30% over five years and underperform by 15% if you are wrong, you have:

• p (win probability): your estimated likelihood of outperformance
• b (payoff ratio): 1.30 / 0.85 = 1.53

The math is straightforward. The challenge is honest estimation of p and b.

For p, assume you are less confident than you think. If you estimate 70%, test whether 60% is more realistic. If you estimate 55%, 50% might be more honest. Professional research suggests most investors are overconfident by 10–15%.

For b, use the expected outcome if correct and the expected outcome if incorrect. Avoid best-case and worst-case scenarios. Use your median expectation for each path.

Kelly in Practice: Why It Fails

Imagine you apply Kelly carefully and estimate:

• p = 0.65 (65% chance of 5-year outperformance)
• b = 1.60 (payoff ratio of 1.6)

Kelly gives: f* = (0.65 × 1.6 - 0.35) / 1.6 = 0.69 / 1.6 = 0.43 = 43%

You allocate 43% to this position. Three years later, the stock has declined 40%. Your portfolio is down 17%. Your conviction wavers. You sell.

What happened? Several factors:

1. Your edge estimate (based on future prospects) collided with current price movements
2. Your probability estimate (a five-year outcome) could not be verified over three years
3. Kelly sizing created a large enough drawdown to trigger emotional selling

This is why Kelly works better in algorithmic systems with monthly feedback and instant rebalancing than in individual stock investing with multi-year time horizons.

Half Kelly: A Practical Compromise

A simple rule: use half of Kelly's recommended size.

If Kelly says 40%, allocate 20%. If Kelly says 20%, allocate 10%.

The mathematical cost of halving is small. You preserve 75% of Kelly's expected outperformance while cutting risk of ruin by much more than 50%.

Some sophisticated investors apply fractional Kelly: three-quarter Kelly, two-thirds Kelly, etc. The choice depends on risk tolerance.

A useful decision tree:

• If your portfolio is in tax-advantaged accounts and you have stable income: Consider full Kelly or seven-eighths Kelly
• If you are withdrawing from your portfolio: Use half Kelly or one-third Kelly
• If you are new to this approach: Start with one-third Kelly

Flowchart

Common Mistakes

Mistake 1: Calculating Kelly once and never updating. Your edge and win rate change as your skill improves or market conditions shift. Recalculate quarterly.

Mistake 2: Using worst-case scenarios in payoff ratio. If you think a stock could decline 50% if wrong, and increase 150% if right, do not use 150/50 = 3.0. Use realistic median outcomes: perhaps 80% gain if right, 20% loss if wrong, giving 4.0. Worst-case thinking inflates b.

Mistake 3: Assuming Kelly is precise. Kelly is a guide, not gospel. Using Kelly to size at 23.6% instead of 20% gives false precision. Round to sensible levels: 20%, 22%, 25%.

Mistake 4: Ignoring portfolio context. Kelly assumes fresh capital and no other constraints. In reality, portfolio tax status, liquidity needs, and correlation with other positions matter. Kelly provides a starting point, not a final answer.

Mistake 5: Applying Kelly to low-conviction positions. Kelly assumes you have genuine edge. If you estimate p = 55% (barely better than random), Kelly may suggest a tiny allocation. This is correct. Do not force sizing based on Kelly if conviction is weak.

FAQ

Q: Can I use Kelly with index funds? A: Technically, yes. If you estimate index funds will outperform other asset classes with 60% probability, Kelly can size your index allocation. In practice, most investors use fixed strategic allocations for index funds and Kelly primarily for individual stocks.

Q: What if I have negative edge (p < 50%)? A: Do not buy that stock. Kelly will recommend zero allocation. A stock you think has less than 50% chance of outperformance should not be in your portfolio at any size.

Q: How do I update Kelly sizing after each investment outcome? A: Track historical win rate and average payoff. If you originally estimated 60% win rate but achieved 50% over 50 picks, adjust p downward. Recalculate Kelly quarterly with new data.

Q: Should I use Kelly for bonds and other assets? A: Yes, if you estimate edge. But most bond investors are not trying to outperform; they are holding for stability. Kelly applies primarily to discretionary outperformance bets.

Q: Can Kelly sizing apply to sectors or asset classes? A: Yes. If you estimate a 60% probability that tech stocks will outperform the market with a 1.5x payoff ratio, Kelly tells you to allocate a certain percentage to tech. This is macro-level Kelly.

Q: What is the difference between Kelly and mean-variance optimization? A: Mean-variance optimization (like Markowitz models) tries to maximize return for a given risk level. Kelly tries to maximize long-term geometric growth. For long-term compounding, Kelly is superior. But Kelly requires more accurate input estimates.

Related Concepts

• Edge — A structural advantage allowing better-than-random performance
• Expected value — The weighted average outcome of a bet (edge × payoff)
• Risk of ruin — The probability of permanent portfolio loss
• Geometric vs. arithmetic returns — Geometric means account for volatility; Kelly optimizes geometric growth

Summary

The Kelly criterion is the mathematically optimal position sizing formula for anyone with a documented edge. It converts edge into a specific bet size that maximizes long-term wealth.

However, Kelly demands brutal honesty about your actual edge and win rate—something most investors lack. When estimates are inflated, Kelly sizing becomes catastrophic.

For practical long-term stock investors, half Kelly or fractional Kelly is a safer implementation. It captures most of Kelly's benefits while dramatically reducing risk of ruin. Even more practically, understanding Kelly's logic—larger positions for higher-edge ideas, smaller for weaker ones—can inform intuitive sizing without rigid formula adherence.

Next: Setting a Maximum Position Size

Mathematical frameworks like Kelly are useful, but they must operate within guardrails. We explore how to set hard limits on individual position sizes that work regardless of conviction or formula output.