The Kelly Criterion in Plain English
The Kelly Criterion is a mathematical formula that answers one of the most practical questions in investing: How much of your capital should you risk on a single bet to maximize long-term compounding? Invented by John L. Kelly Jr. in 1956, the formula reveals that the intuitive answer—"bet everything on your best idea" or "bet equally across all ideas"—is mathematically wrong. The optimal bet size is a precise fraction of your bankroll that maximizes compound growth while avoiding catastrophic ruin.
Most investors and traders have never heard of the Kelly Criterion, yet it explains why some portfolio managers compound wealth while others blow up. It also explains why diversification works and why concentration fails, not through moral platitudes but through math.
Quick definition: The Kelly Criterion is f* = (bp - q) / b, where f* is the fraction of bankroll to bet, b is the odds you receive, p is your win probability, and q is your loss probability (1 - p). It tells you the exact percentage of your capital to allocate to each bet to maximize compound growth rate.
Key takeaways
- The Kelly Criterion formula determines optimal bet sizing for any repeated bet with known odds
- Betting more than the Kelly percentage increases ruin risk exponentially
- Betting less than Kelly is safe but leaves compounding gains on the table
- Most successful investors use a fraction of Kelly (half-Kelly, quarter-Kelly) due to estimation error in real-world probabilities
- Diversification emerges naturally from Kelly math when you have multiple independent bets
- The formula explains why "bet everything on your best idea" destroys wealth despite good expected value
- Professional hedge funds and money managers use Kelly-inspired sizing in algorithmic trading
The Kelly Formula and What It Means
The Kelly Criterion formula is deceptively simple:
f* = (bp - q) / b
Where:
f* = Fraction of bankroll to wager
b = Odds ratio (payout ratio)
p = Probability of winning
q = Probability of losing (1 - p)
Let's decode this with an example.
Scenario: A coin flip with 2-to-1 payoff
You're offered a bet: flip a fair coin. Heads, you win $2 for every $1 wagered. Tails, you lose your $1 wager. Your bankroll is $10,000.
- b = 2 (you win 2x your bet)
- p = 0.5 (50% chance of heads)
- q = 0.5 (50% chance of tails)
Applying Kelly:
f* = (2 × 0.5 - 0.5) / 2
f* = (1.0 - 0.5) / 2
f* = 0.5 / 2
f* = 0.25 (or 25%)
Interpretation: Bet exactly 25% of your bankroll on each coin flip to maximize long-term compound growth. On a $10,000 bankroll, that's $2,500 per flip.
This is counterintuitive. Your intuition might say "bet it all—you have positive expected value!" Or "bet 50-50 across your options." But Kelly math says: bet exactly 25%, no more, no less.
Why 25%? The Compounding Mathematics
Betting more than Kelly hurts you through volatility drag. Betting less than Kelly leaves money on the table.
Consider what happens if you bet different fractions against the same 2-to-1 coin flip over repeated rounds:
Betting 50% per flip (above Kelly):
- Win: $10,000 → $15,000
- Lose: $10,000 → $5,000
- Win again: $15,000 → $22,500
- Lose: $22,500 → $11,250
- Two wins, two losses (breakeven sequence): $11,250
Final result: You lost money despite winning 50% of flips with positive expected value.
Betting 25% per flip (Kelly):
- Win: $10,000 → $12,500
- Lose: $10,000 → $7,500
- Win again: $12,500 → $15,625
- Lose: $15,625 → $11,719
- Two wins, two losses: $11,719
Final result: You made money even in a 50/50 sequence.
Betting 10% per flip (below Kelly):
- Win: $10,000 → $11,000
- Lose: $10,000 → $9,000
- Win again: $11,000 → $12,100
- Lose: $12,100 → $10,890
- Two wins, two losses: $10,890
Final result: You made less money than Kelly.
The difference compounds. Over 100 flips (50 wins, 50 losses), Kelly generates roughly 2.5x wealth, while 50% of bankroll generates a loss, and 10% generates minimal gains.
Why? Because when you bet too much, losses hit a larger base. A 50% loss on $15,000 (after a win) brings you to $7,500. Recovering to $15,000 requires 100% gains. But when you bet Kelly (25%), losses are smaller, and the asymmetric recovery compounding doesn't work against you as hard.
The Growth Rate Equation: Why Kelly Maximizes Wealth
The Kelly Criterion mathematically maximizes the compound growth rate, defined as:
G = p × ln(1 + f × b) + q × ln(1 - f)
Where:
G = Growth rate per bet
f = Fraction of bankroll wagered
p, q, b = Same as before
ln = Natural logarithm
This equation is the reason Kelly works. It balances:
- Upside: Larger bets (larger f) increase winning payoffs (1 + f × b grows)
- Downside protection: Larger bets increase ruin risk (1 - f shrinks toward zero)
Kelly finds the exact sweet spot where the growth rate is maximized.
For the 2-to-1 coin flip example, plotting G against different bet fractions reveals a peak at exactly f = 0.25. Betting less (0.20 or 0.15) yields lower growth. Betting more (0.30 or 0.35) also yields lower growth due to increased volatility drag. Only at 0.25 is growth maximized.
Real-World Example: Stock Portfolio Allocation
Kelly applies to investing, not just gambling. Consider a stock you've analyzed carefully.
Your assessment:
- 65% probability the stock outperforms the market over 5 years
- 35% probability it underperforms
- Expected outperformance if it wins: 40% total return
- Expected underperformance if it loses: -20% total return
Using Kelly:
- b = 0.40 (if you win, you gain 40%; if you lose, you lose 20%, so your odds ratio is 0.40/0.20 = 2.0, approximately)
- p = 0.65
- q = 0.35
f* = (2.0 × 0.65 - 0.35) / 2.0
f* = (1.30 - 0.35) / 2.0
f* = 0.95 / 2.0
f* = 0.475 (or 47.5%)
Kelly says: Allocate 47.5% of your portfolio to this stock. The remaining 52.5% should be allocated to other bets or held in diversifying assets.
This is radically different from concentrating your entire portfolio into your "best idea."
Half-Kelly and the Practical Implementation Problem
Professional investors rarely use full Kelly. They use half-Kelly, quarter-Kelly, or even lower. Why?
The estimation error problem. The Kelly formula works perfectly if you know p and b accurately. In reality, you don't. Your estimate of win probability is often wrong, sometimes drastically. If you're 65% confident a stock will outperform but actually it only has 45% chance, full Kelly sizing will ruin you.
Half-Kelly is more robust. It's slower (only 25% the growth rate of full Kelly), but it creates a buffer against probability misestimation. If your true odds are worse than estimated, half-Kelly still works. Full Kelly explodes.
| Kelly Variant | Growth Rate (as % of Full Kelly) | Ruin Risk Reduction |
|---|---|---|
| Full Kelly | 100% | Baseline |
| Half-Kelly (0.5 × f*) | 75% | 25% of full ruin risk |
| Quarter-Kelly (0.25 × f*) | 56% | 6% of full ruin risk |
| Eighth-Kelly (0.125 × f*) | 32% | Minimal ruin risk |
Most professionals operate at half-Kelly to quarter-Kelly, sacrificing 25–50% of theoretical growth for robustness to estimation errors.
Why Diversification Emerges from Kelly Mathematics
Suppose you have identified three independent investment opportunities:
- Stock A: Kelly allocation 35%
- Stock B: Kelly allocation 28%
- Stock C: Kelly allocation 22%
Total Kelly: 85%. You invest 85%, hold 15% cash/bonds for stability.
If you had identified four opportunities (A, B, C, D at 25% each), total Kelly is 100%. You go fully invested.
If you had identified ten opportunities at 8% each, you're diversified across ten positions at 8% allocation each.
Natural diversification emerges from Kelly math. When you have many independent positive-expected-value bets and you size them optimally, you're forced to diversify. This is why hedge funds with dozens of positions often outcompound concentrated bets—the math of Kelly compels diversification.
Conversely, if you have few high-confidence ideas, Kelly permits concentration. If you have many moderate-confidence ideas, Kelly demands diversification.
The Ruin Risk Curve: Why Overbetting Destroys Compounding
Betting above Kelly percentage creates exponential ruin risk. Consider the probability of ever going bust with repeated betting:
| Bet Size (as % of Kelly) | Probability of Ruin (100 flips) | Expected Final Wealth |
|---|---|---|
| 50% of Kelly | 0% | $27,000 |
| 100% of Kelly | 0% | $45,000 |
| 150% of Kelly | 0.01% | $65,000 |
| 200% of Kelly | 0.5% | $95,000 |
| 300% of Kelly | 5% | $150,000 |
| 400% of Kelly | 15% | $200,000 |
The ruin probability rises slowly at first, then exponentially. At 400% of Kelly, you have a 15% chance of losing everything over 100 bets. Importantly, the expected wealth is higher—but the tail risk (ruin) is catastrophic.
This is why leveraged portfolios blow up. A 2-to-1 leveraged portfolio is equivalent to betting 200% of Kelly. Over a 20-year period with multiple market corrections, the ruin probability becomes significant. Many hedge funds using leverage experience spectacular failures despite positive expected value.
Kelly Bet Size Impact
Kelly in Practice: Professional Allocations
Renaissance Technologies (Renaissance Medallion Fund)
Jim Simons' legendary quant fund is rumored to use Kelly-inspired math for position sizing. The fund holds hundreds of small positions, each sized optimally based on estimated edge. This natural diversification through Kelly sizing enabled 66% annualized returns for three decades without catastrophic losses.
Berkshire Hathaway
Warren Buffett operates a Kelly-influenced strategy. When highly confident (Apple, Bank of America after 2008), he concentrates positions (25–30% of portfolio). When moderately confident, he diversifies. This corresponds to Kelly sizing—high confidence ideas get larger allocation.
Individual Trader Example: Coin Flip Bets
A trader has a profitable edge in a specific market with 60% win rate, 1.5-to-1 payoff.
f* = (1.5 × 0.60 - 0.40) / 1.5
f* = (0.90 - 0.40) / 1.5
f* = 0.50 / 1.5
f* = 0.333 (or 33.3%)
Kelly says: Risk 33% of bankroll per trade. With 100 trades, the trader expects roughly 60 wins (avg gain: 0.50 per unit risked) and 40 losses (avg loss: 1.0 per unit risked), compounding wealth at a consistent rate.
If the trader instead risks 60% per trade (double Kelly), they compound at higher speed initially but face exponential ruin risk. A bad streak of losses (which will eventually occur in any repeated betting) produces bankruptcy.
Common Mistakes
Mistake 1: Ignoring Kelly entirely. Many investors bet randomly, allocating equal weight across ideas regardless of conviction. Kelly math proves this is suboptimal. Equal-weighted portfolios compound slower than Kelly-weighted ones.
Mistake 2: Using full Kelly in estimation-error contexts. You never know true probability. Always use half-Kelly or quarter-Kelly unless you have decades of data proving your estimate accuracy.
Mistake 3: Betting above Kelly due to overconfidence. This is the fastest path to bankruptcy. Every professional trader has blown up at least once from betting too much. The market's job is to find your edge and exploit it with leverage until you're ruined.
Mistake 4: Focusing on wins instead of growth rate. A strategy with 70% win rate but massive overbetting compounds slower than a 55% win rate strategy with Kelly sizing. Growth rate is the metric that matters.
Mistake 5: Using Kelly for uncorrelated assets incorrectly. Kelly assumes independence. If your positions are correlated (multiple tech stocks, for example), you should treat them as a single bet and size accordingly, reducing Kelly allocation.
Mistake 6: Forgetting the cash buffer. Kelly optimal sizing assumes you can always place the next bet. Real portfolios need cash reserves for opportunities and to avoid forced selling during crashes. Practical Kelly allocation is usually 10–20% below mathematical Kelly.
FAQ
If Kelly is optimal, why do successful investors use less than Kelly sizing?
Estimation error. You don't know true probabilities. Using half-Kelly or quarter-Kelly provides a safety buffer. If your estimate of 65% probability is actually 45%, full Kelly ruins you, but quarter-Kelly still compounds. The professionals who survive long-term are those conservative enough to survive mistakes.
Can I use Kelly for my entire stock portfolio?
Partially. Kelly works well for tactical allocation (30% to your highest-conviction stock idea, 15% to your second-best, etc.). But your overall equity allocation (60% stocks, 40% bonds) should be based on time horizon and risk tolerance, not Kelly math on individual stocks.
What if I have more than 100% Kelly opportunities?
This is common. You've identified fifteen stocks, each with Kelly allocation of 10%. Total is 150%. You use proportional Kelly: allocate 10/15 = 6.7% to each stock, keep 30% in bonds/cash. This maintains Kelly's growth properties while staying fully invested.
Does Kelly work if I'm wrong about my edge?
No, it fails catastrophically. This is why practitioners use fractional Kelly. If your true edge is smaller than estimated, fractional Kelly still works; full Kelly ruins you. This is the Kelly Criterion's fatal flaw with real-world estimation error.
How often should I rebalance under Kelly sizing?
When probabilities change or positions drift more than 20% from target. If you allocated 25% to a stock and it's now 30% due to gains, rebalance quarterly or semi-annually. Frequent rebalancing (monthly) adds transaction costs that eat into Kelly's gains.
Can I use Kelly for cryptocurrency, meme stocks, or other speculative assets?
Only if you have reliable estimates of probability and payoff. For assets with no historical edge data (most cryptos, meme stocks), your probability estimate is likely wrong, making Kelly dangerous. Use quarter-Kelly or lower, if anything.
What's the relationship between Kelly and the Sharpe ratio?
Both measure edge, but differently. Sharpe measures risk-adjusted return on a continuous scale. Kelly measures optimal bet sizing on a discrete (on/off) bet. For normal distributions, higher Sharpe ratio implies larger Kelly fraction.
Real-world examples
Example 1: The 2008 Financial Crisis and Leverage
Lehman Brothers used approximate leverage ratios of 30-to-1, far above Kelly sizing. Their estimated edge (mortgage bonds are safe, historically proven) led to massive overconfidence. When the edge was wrong (mortgage defaults spiked), the leverage destroyed the firm in days. Kelly sizing would have limited leverage to 2-to-1 or less, surviving the crisis.
Example 2: A Trader's Edge and Optimal Sizing
A trader backtests a strategy: 55% win rate, avg win $2, avg loss $1.50.
b = 2 / 1.5 = 1.33
f* = (1.33 × 0.55 - 0.45) / 1.33
f* = (0.732 - 0.45) / 1.33
f* = 0.282 / 1.33
f* = 0.212 (or 21.2%)
Using full Kelly (21% per trade) on a $100,000 account: risk $21,000 per trade. Over 100 trades, expected result is +$5,250 profit.
Using half-Kelly (10.5%): risk $10,500 per trade. Expected profit over 100 trades is still positive, but lower ($2,600), and ruin probability drops from 0.5% to 0.01%.
Example 3: Portfolio Allocation Under Kelly
An investor identifies three stock opportunities:
- Stock A: 70% win probability, expected 30% return, Kelly = 40%
- Stock B: 60% win probability, expected 20% return, Kelly = 22%
- Stock C: 55% win probability, expected 15% return, Kelly = 12%
Total Kelly: 74%. Investor allocates 40% A, 22% B, 12% C, keeps 26% in bonds.
A year later, Stock A has doubled (70% true case), probability now appears higher (perhaps 75%), but other estimates remain the same. Investor rebalances: now 45% A, 20% B, 10% C, 25% bonds.
Over 10 years, Kelly-adjusted rebalancing produces far higher wealth than any fixed allocation.
Related concepts
- Asymmetric Bet Outcomes and Compounding
- Why Diversification Works: The Math
- Leverage and Ruin Risk
- Position Sizing in Practice
- Edge, Win Rate, and Expected Value
- FINRA: Position Sizing and Risk Management
- SEC: Hedge Fund and Investment Advisor Regulations
- Investor.gov: Understanding Margin and Leverage
- Federal Reserve: Leverage and Financial Stability
Summary
The Kelly Criterion is a precise mathematical formula for optimal bet sizing. It tells you exactly what fraction of your capital to allocate to each investment opportunity to maximize long-term compound growth while minimizing ruin risk.
Full Kelly (exact formula output) is theoretically optimal but practically dangerous due to estimation error. Most professionals use half-Kelly, quarter-Kelly, or lower, sacrificing 25–50% of theoretical growth for robustness to probability misestimation.
Importantly, Kelly math reveals why diversification works: when you have many independent positive-expected-value bets and you size them optimally, you're forced to hold a portfolio, not concentrate. Concentration emerges when you have few high-confidence ideas; diversification emerges when you have many moderate-confidence ideas. Both are Kelly-consistent, depending on your edge assessment.
The investors and traders who blow up are those who bet above Kelly—often due to overconfidence, leverage, or incomplete understanding of their edge. Those who thrive are conservative in sizing, humble about probability estimation, and disciplined about rebalancing as new information arrives.
Kelly math explains why boring, diversified, moderately-leveraged portfolios often outcompound exciting, concentrated, highly-leveraged portfolios over decades. The math isn't subjective—it's probabilistic truth.