Methodology
- Estimate fair win probability from a model or no-vig market plus signal adjustment.
- Convert odds to decimal payout and calculate the Kelly fraction.
- Apply a fractional multiplier to account for uncertainty and correlation.
- Cap exposure by day, sport, market, creator, and correlated outcomes.
Example output
Fractional Kelly example
How a raw edge can become a smaller risk-managed stake.
| Input | Example | Sizing effect |
|---|---|---|
| Model probability | 55% | Creates estimated edge |
| Available price | -105 | Defines payout |
| Raw Kelly | 7.9% | Too aggressive for many users |
| Quarter Kelly cap | 2.0% | More conservative stake |
Fractional Kelly is still risky if the probability estimate is wrong.
Why full Kelly is usually too aggressive
The formula assumes your probability estimate is accurate. Sports models are uncertain, markets move, and correlated bets can quietly multiply risk.
- Use quarter Kelly or half Kelly when estimates are noisy
- Set maximum stake caps even when the formula suggests more
- Reduce exposure for correlated plays from the same game
- Pause or reduce sizing during drawdowns and model changes
How to present Kelly in a model report
A conversion page should show that the tool encourages discipline. Include the estimated probability, price, raw Kelly fraction, applied fraction, and final capped stake.
- Label every stake as educational analytics, not financial advice
- Show what happens when probability is reduced by one or two points
- Warn when edge disappears after line movement
- Make bankroll limits visible before any subscription prompt
Responsible-use note
Analytics should support disciplined decision-making, not guaranteed outcomes. Bet only where legal, never risk money you cannot afford to lose, and use limits before volume increases.
FAQ
What is fractional Kelly?
Fractional Kelly uses part of the full Kelly stake, such as half or quarter Kelly, to reduce volatility and model-error risk.
Does Kelly guarantee bankroll growth?
No. Kelly depends on accurate probabilities and long-run conditions. Bad estimates, limits, correlation, and variance can cause losses.
Should beginners use Kelly staking?
Beginners should be cautious. Flat staking or very small fractional Kelly may be easier to control while learning to evaluate edges.