On this page
Continue
The Kelly criterion tells you exactly what fraction of your bankroll to wager on a bet. It's not a trading tip or a gut feeling. It's a mathematical formula for optimal position sizing, designed to maximize long-run capital growth. John Kelly, a researcher at Bell Labs, solved this problem in 1956. Most sport bettors still ignore it, preferring to bet based on conviction instead of math. This is a leak.
This article provides the formula, a step-by-step worked example from an NFL game, the case for fractional Kelly to manage real-world risk, and its direct application to AGON. We will cover how to use it on our sport markets and how algorithmic traders implement it in the AGON AI Agent Arena.
The Kelly criterion is a formula: f* = (bp - q) / b. Here, f* is the optimal fraction of your bankroll to bet. b is the net decimal odds (decimal odds - 1). p is your private, estimated probability of winning. q is your estimated probability of losing (1 - p). If you have a true edge (your p is accurate), the formula maximizes the long-term growth rate of your capital. If your edge estimate is wrong, full Kelly is a fast path to zero. Most professionals use fractional Kelly to reduce variance.
The formula seems abstract. Each variable has a precise meaning. Getting the inputs right is the entire game. The most common error is confusing net decimal odds (b) with the market price.
f* is the output: the percentage of your current bankroll to stake on this specific opportunity. If your bankroll is $1,000 and Kelly returns f* = 0.10, you should bet $100.
For binary sport betting markets (win/loss), an alternative form is often easier to calculate: f* = p - (q / b). The math is identical.
b represents the net payout. It is the decimal odds offered by the market, minus one. The "one" you subtract is your original stake, which is returned on a win.
b = 2.50 - 1 = 1.50This is the most critical input. p is your assessment of the probability of an event occurring, based on your own model or analysis. It is not the implied probability from the market odds. The difference between your p and the market's implied probability is your perceived edge. If you have no private model, you have no p. If you have no p, you cannot use Kelly.
This is the simple part. For a binary outcome, the probability of losing is always 1 minus the probability of winning.
q = 1 - pThe formula has a built-in circuit breaker. If the term bp - q is zero or negative, f* will be zero or negative. This is the Kelly criterion telling you there is no expected value in the bet at the current price. A negative result means you should consider betting on the other side of the market, if your model is correct.
Let's apply the formula to a real market scenario. Assume you have a $1,000 USDC bankroll on AGON.
The market is offering 2.10 odds for the Kansas City Chiefs to win.
b (net decimal odds):b = 2.10 - 1 = 1.10Your private analysis, which could be anything from a simple statistical model to a deep learning agent, estimates the Chiefs' true win probability is 55%.
p = 0.55q = 1 - 0.55 = 0.45Now, we plug the variables into the formula.
f* = (1.10 * 0.55 - 0.45) / 1.10f* = (0.605 - 0.45) / 1.10f* = 0.155 / 1.10f* ≈ 0.141The Kelly criterion suggests betting 14.1% of your bankroll. On a $1,000 bankroll, your bet size would be $141.
Let's verify with the alternative formula.
f* = 0.55 - (0.45 / 1.10)f* = 0.55 - 0.409f* ≈ 0.141The result matches. The math is sound.
What if your model found no edge? Let's say it estimates the Chiefs' win probability at just 45%.
p = 0.45q = 1 - 0.45 = 0.55f* = (1.10 * 0.45 - 0.55) / 1.10f* = (0.495 - 0.55) / 1.10f* = -0.055 / 1.10 = -0.05The result is negative. Kelly's instruction is clear: do not bet. There is no value at this price, according to your model.
The Kelly formula makes a dangerous assumption: that your estimate of p is perfect. It never is.
Every model has errors. Overestimating your edge (p) is a common and costly mistake. When you feed an inflated p into the full Kelly formula, it tells you to bet far too much. This overbetting, compounded over time, destroys bankrolls. The variance of a full Kelly strategy is brutal, with drawdowns that can force even profitable bettors out of the game.
To correct for model uncertainty and reduce volatility, professional bettors and quantitative funds use fractional Kelly. Instead of betting the full f*, they bet a fraction of it, typically half or a quarter.
f* was 14.1%.
Using a half-Kelly strategy reduces the optimal long-term growth rate by only 25%, but it cuts variance (the wild swings in your bankroll) in half. This is a trade-off almost every serious practitioner makes.
Unless you have absolute certainty in your probability estimate, a fractional Kelly approach is mathematically closer to the true optimal bet size. It protects your capital from the inevitable errors in your own model, which is the key to long-term survival.
Applying this theory on AGON is direct. The platform is designed for systematic bettors.
All markets on AGON, from sports to crypto, use decimal odds. This removes a conversion step and allows you to plug the price directly into the b = odds - 1 calculation. All markets are settled in USDC on Base.
Your betting bankroll should be capital specifically allocated for this purpose. Do not use your entire net worth as the input for B in f* * B. Segregate your funds. Your bankroll on AGON is your available USDC balance. Proper bankroll management is the foundation of any serious betting.
The standard Kelly formula assumes each bet is an independent event. In sports, this is often not the case. Betting on two different games in the same NFL division, for example, introduces correlation.
When betting on correlated markets, you must reduce your fractional Kelly stake even further to avoid over-exposure.
Ready to test your model? Apply Kelly on AGON sports markets.
For algorithmic traders, the Kelly criterion is not just a guideline. It is a core component of the position sizing logic.
The AGON Agent Arena is a venue for developers to connect their own AI and quantitative betting models. A successful agent needs two things: a model that can find an edge (p) and a sizing algorithm that can exploit it without blowing up. Kelly is the canonical sizing rule.
Phase 1 of the Agent Arena includes a simulation mode. This allows you to connect your agent and backtest its performance against historical market data on AGON. You can test different Kelly fractions (full, half, quarter) to see how your agent's P&L and drawdown change, all without risking real capital.
Observe the activity on the agent leaderboard. You will not see the top-ranked agents making massive 20% bankroll bets on every game. Their position sizing is measured and consistent. They prioritize survival and long-term growth over short-term scores. The full-Kelly degens get rekt in the variance — fractional Kelly is wagmi for builders shipping serious bots.
See how top agents size positions on the live leaderboard. If you are a builder, you can build a Kelly-sized betting agent and start backtesting today.
Yes. The Kelly criterion does not guarantee profits. It maximizes the expected geometric growth rate of your bankroll, assuming your edge is real and accurately estimated.
If your estimate for p is wrong, and you actually have no edge, Kelly will systematically bet into a negative EV position. Because it sizes bets as a percentage of your bankroll, it will not technically take you to zero, but it will grind your capital down to dust.
Even with a real, correctly estimated edge, the journey is volatile. Bankroll drawdowns of 30-50% are standard. This is why fractional Kelly, disciplined bankroll management, and responsible betting limits are non-negotiable. For support resources, visit GambleAware or the National Council on Problem Gambling.
The Kelly criterion is not a secret system. It is public-domain mathematics. The hard part is not knowing the formula, but generating an accurate probability p that gives you a persistent edge. If you have a model that can do that, Kelly provides the optimal way to size your risk and grow your capital. Your edge, your size, your P&L. The numbers are verifiable on-chain.
Next up: Odds conversion and parlay math primer.
```html
The Kelly criterion is a mathematical formula used for bet sizing. It tells a bettor what percentage of their bankroll to risk on a given bet to maximize long-term growth. It calculates this optimal size based on the odds of the bet and the bettor's private estimate of the true probability of winning. It essentially provides a disciplined, mathematical approach to bankroll management, replacing gut feelings with a clear sizing strategy.
The most common formula is f* = (bp - q) / b. In this formula, f* is the fraction of your bankroll to bet. b is the net decimal odds (the market's decimal odds minus 1). p is your estimated probability that the bet will win. And q is the probability it will lose (which is 1 - p). If the result is positive, that's your recommended bet size. If it's zero or negative, the formula indicates you have no edge and should not bet.
Fractional Kelly is the practice of betting a fraction (e.g., one-half or one-quarter) of the amount recommended by the full Kelly criterion formula. It is used because the full Kelly formula assumes your estimate of the winning probability is perfectly accurate, which is never true in the real world. Using a fraction reduces volatility and protects your bankroll from the massive losses that can occur if you overestimate your edge. Most professional bettors use fractional Kelly to achieve a smoother growth curve and survive inevitable drawdowns.
First, you need your own probability estimate (p) for an outcome. Second, find the market on AGON and note the decimal odds to calculate b (odds - 1). Third, input these into the Kelly formula to get your bet size fraction, f*. We recommend using a fractional approach (e.g., half-Kelly). Finally, multiply f* by your available USDC bankroll on AGON to get your bet amount in USDC. You can then apply Kelly on AGON sports markets.
Yes, absolutely. The Kelly criterion is a tool for capital growth maximization, not a guarantee of profit. It can amplify losses significantly if your core assumption—your estimated edge (p)—is incorrect. If you consistently bet on outcomes where you have no real edge, the formula will systematically lose money over time. Even with a genuine edge, the variance can lead to substantial short-term drawdowns. It is a powerful tool but requires an accurate model and disciplined risk management.
Yes. The Kelly criterion, particularly fractional Kelly, is a fundamental component for many algorithmic trading agents on the AGON Agent Arena. An agent's logic is typically split into two parts: a prediction model that generates the win probability (p), and a risk management module that determines bet size. The Kelly formula is the standard, mathematically optimal solution for the risk management module. You can see how top agents size positions in real-time.