What Is the Kelly Criterion?
In 1956, a physicist named John L. Kelly Jr. published a paper at Bell Labs titled "A New Interpretation of Information Rate." Kelly was trying to solve a problem in telecommunications — how to maximize the rate of data transmission over noisy phone lines. What he discovered, almost by accident, was a formula that answered one of gambling's oldest questions: how much should you bet?
The Kelly Criterion tells you the exact percentage of your bankroll to wager on any given bet, based on two inputs: the odds being offered and your estimated probability of winning. The goal is not to maximize profit on a single bet — it's to maximize the long-term growth rate of your bankroll across hundreds or thousands of bets.
Since its publication, the Kelly Criterion has been adopted by professional sports bettors, poker players, blackjack card counters, and Wall Street hedge fund managers — including Ed Thorp (who beat Las Vegas casinos) and Jim Simons (whose Renaissance Technologies became the most profitable fund in history). It remains the gold standard for bet sizing because it balances two competing goals: grow your bankroll as fast as possible while never risking ruin.
Why It Matters
Most bettors lose money not because they pick the wrong side, but because they bet the wrong amount. The Kelly Criterion removes guesswork from bet sizing and replaces it with math.
The Kelly Formula
The Kelly Criterion can be expressed as a single, elegant formula:
f* = (bp − q) / b
f* = fraction of bankroll to wager
b = decimal odds − 1 (net profit per $1 wagered)
p = probability of winning (your estimate)
q = probability of losing = 1 − p
There is also a simplified way to think about it that makes the intuition clearer:
f* = Edge / Odds
Where Edge = expected profit per dollar wagered, and Odds = net payout (decimal odds − 1).
The output f* is a decimal between 0 and 1. Multiply by 100 to get a percentage. If f* = 0.05, you should bet 5% of your current bankroll. If your bankroll is $10,000, that means a $500 wager.
If f* comes out zero or negative, the formula is telling you there is no edge — don't bet. More on that later.
Worked Example
Let's walk through a real scenario step by step. Suppose a sportsbook offers +150 odds on the underdog in an NBA game. You've done your research and believe this team has a 45% chance of winning.
Step 1: Convert American odds to decimal odds.
Step 2: Identify the variables.
Step 3: Plug into the Kelly formula.
Step 4: Calculate the dollar amount.
Full Kelly recommends wagering $833 on this bet. That's a significant chunk of your bankroll — and that's precisely why most professionals use fractional Kelly instead.
Full Kelly vs Fractional Kelly
Full Kelly is mathematically optimal for maximizing long-term bankroll growth — but only if your probability estimates are perfectly accurate. In the real world, they never are. That's why the vast majority of professional bettors use fractional Kelly, betting only a fraction (typically 25–50%) of the full Kelly recommendation.
Let's compare using our +150 odds, 45% win probability example (full Kelly = 8.33%, bankroll = $10,000):
Full Kelly (100%)
Half Kelly (50%)
Quarter Kelly (25%)
Here's how those strategies compare side by side:
Full Kelly
Maximum growth rate but extreme variance. A 5-bet losing streak could cut your bankroll by 35%+. Only viable with perfect probability estimates.
Half Kelly
Achieves ~75% of full Kelly's growth rate with dramatically lower drawdowns. The sweet spot for experienced bettors with proven models.
Quarter Kelly
Achieves ~50% of full Kelly's growth rate with very manageable swings. Best for most bettors — absorbs estimation error gracefully.
Fractional Kelly Advantages
Absorbs estimation errors. Reduces drawdowns dramatically. Still compounds over time. Psychologically easier to stick with. Protects against model miscalibration.
Full Kelly Disadvantages
Assumes perfect probability estimates. Extreme short-term variance. Psychologically brutal during losing streaks. Tiny overestimation of edge leads to over-betting and bankroll destruction.
Misconception: "Full Kelly means maximum profit"
Full Kelly maximizes the growth rate, not raw profit on any single bet. And it only does so with perfect inputs. With imperfect estimates (i.e., always), fractional Kelly often produces better real-world results because it keeps you in the game during inevitable cold streaks.
When Kelly Says "Don't Bet"
If the Kelly formula returns zero or a negative number, it's delivering the most valuable signal it can: you have no edge on this bet. Walk away.
Let's see this in action. Same +150 odds (b = 1.50), but now you estimate only a 35% chance of winning:
The negative result means the sportsbook's implied probability is higher than your estimated probability. You'd be betting into a negative expected value situation. The Kelly Criterion is protecting your bankroll by telling you to pass.
This is arguably Kelly's greatest gift. Most recreational bettors lose money not because they always pick the wrong side, but because they bet on games where they have no edge at all. The formula enforces discipline — it forces you to quantify your edge before risking money.
The Discipline Edge
The Kelly Criterion doesn't just tell you how much to bet — it tells you when NOT to bet. A Kelly value of zero or negative means stay out. The discipline to pass on no-edge bets is what separates profitable bettors from the rest.
Advantages of Kelly
The Kelly Criterion has endured for nearly 70 years because its mathematical properties are genuinely remarkable:
Maximizes long-term growth. Of all possible fixed-fraction betting strategies, Kelly produces the highest geometric growth rate. No other strategy will grow your bankroll faster over a large number of bets — this is mathematically proven.
Prevents ruin. Because Kelly sizes bets as a percentage of your current bankroll, you can never technically go to zero. As your bankroll shrinks, your bets shrink proportionally. This is a built-in survival mechanism that flat-dollar betting lacks.
Scales bets to edge size. Your biggest bets go on your strongest edges, and your smallest bets go on marginal spots. This is intuitively correct — you should risk more when you have more confidence — and Kelly quantifies exactly how much more.
Self-adjusting. Kelly automatically increases bet size when your bankroll is growing (compounding gains) and decreases bet size when it's shrinking (preserving capital). No manual intervention required.
Identifies no-edge situations. When f* is zero or negative, you know immediately that there's no profitable bet to make. This saves you from the most common leak in betting: wagering without an edge.
Misconception: "Kelly is only for gambling"
The Kelly Criterion is used far beyond sports betting. Hedge funds use it for portfolio allocation, venture capitalists use it for investment sizing, and professional poker players use it for tournament buy-in decisions. Any scenario involving repeated decisions under uncertainty can benefit from Kelly-based sizing.
Risks and Limitations
The Kelly Criterion is powerful, but it is not a magic bullet. Understanding its limitations is just as important as understanding the formula itself.
Danger: Overestimating Your Edge
The Kelly Criterion is only as good as the probability estimates you feed it. If you think a bet wins 55% of the time but the true number is 50%, Kelly will have you massively over-betting. Garbage in, garbage out. A small error in your model translates into a huge error in bet size.
Volatile short-term results. Even with perfect estimates, full Kelly produces extreme bankroll swings. Drawdowns of 40–60% are not uncommon and can last for weeks. Most bettors psychologically cannot handle this level of variance and abandon their strategy at the worst possible time.
Requires accurate probability estimates. The formula assumes you know the true probability of winning. In sports betting, you're always estimating. If your model is poorly calibrated, Kelly will amplify your mistakes rather than your edge.
Correlated bets are dangerous. Kelly assumes each bet is independent. If you're betting on multiple NFL games that are correlated (e.g., all overs in bad weather), you're effectively over-betting. The formula doesn't account for correlation between bets without modification.
Over-Kelly is catastrophic. Betting more than the Kelly-recommended amount doesn't just reduce your growth rate — it can turn positive expected value into negative expected growth. Even slightly over-Kelly betting destroys bankroll faster than it builds it over the long run.
Warning: The Over-Kelly Trap
If Kelly says bet 5% and you bet 12% because you "feel confident," you're not being aggressive — you're being destructive. Over-Kelly betting has a negative long-term growth rate even when your edge is real. Always err on the side of under-betting.
Practical Tips
Here's how to apply the Kelly Criterion effectively in your sports betting:
Start with quarter Kelly. Until you have at least 500+ tracked bets proving your probability model is well-calibrated, use 25% Kelly. You can always scale up later — you can't undo a blown bankroll.
Be conservative with edge estimates. If your model says a bet has a 58% chance of winning, mentally round it down to 54–55% before running Kelly. The cost of under-betting a good edge is small. The cost of over-betting a mediocre edge is massive.
Set a maximum bet cap. Even if Kelly says to bet 15% of your bankroll, cap your individual bets at 3–5%. Edges that large are extremely rare in sports betting, and a Kelly output that high usually means your model is overestimating your edge.
Recalculate as your bankroll changes. Kelly is a percentage of your current bankroll, not your starting bankroll. Update your bankroll number at least weekly so bet sizes automatically adjust to wins and losses.
Never override Kelly with gut feel. If the formula says bet 2% and your gut says bet 10%, trust the formula. Emotional overrides are the single fastest path to bankruptcy. The math doesn't care about your feelings — and that's its greatest strength.
Combine with bankroll management fundamentals. Kelly tells you how much to bet on each wager, but you still need the discipline of a separate betting bankroll, clear loss limits, and emotional detachment from results. Kelly is one tool in a complete system — not the entire system.
Track and review. Log every bet with your estimated probability, the Kelly recommendation, your actual bet size, and the result. Over time, compare your predicted win rates against actual outcomes to calibrate your model. If they diverge, adjust your estimates — not the formula.
The Bottom Line
The Kelly Criterion is the closest thing to a mathematically proven "answer" for bet sizing. Use fractional Kelly, be honest about your edge, and let the formula protect you from the two biggest killers in betting: over-betting when you're wrong and under-betting when you're right.