Fractional Kelly Betting: Calculation and Drawdown Trade-offs

Calculation

Fractional stake = alpha * full-Kelly fraction * bankroll

where alpha is between zero and one under the usual fractional-Kelly convention.

Using the illustrative full-Kelly output 8.333% from decimal odds 2.20 and p = 0.50:

Values are rounded. The GBP amounts are mathematical examples, not recommendations.

What the fraction changes

Reducing alpha reduces the cash exposed to both correct and incorrect model signals. Under the assumed model, that usually gives up some expected log growth in exchange for a less aggressive path. It does not establish a specific real-world drawdown probability.

Busseti, Ryu and Boyd formulate drawdown risk explicitly and compare their risk-constrained approach with fractional Kelly. Their work supports treating drawdown as a separate objective, not claiming that half Kelly has one universal risk level.

What the fraction does not fix

If the true probability is below the break-even probability, multiplying an erroneous positive Kelly estimate by one-quarter still produces a negative-expectation bet. A smaller wrong stake is smaller exposure, not a corrected decision.

Sports-wagering research that models uncertainty in p obtains fractions through specified loss functions and priors, with materially different results across methods (Chu, Wu and Swartz). Arbitrarily choosing half Kelly is therefore a risk convention, not evidence that estimation uncertainty has been quantified.

Portfolio check

Calculate full Kelly and the chosen fraction at the portfolio level where positions overlap. Adding five separate quarter-Kelly outputs can exceed the intended total exposure and can count the same team or match information repeatedly.

At minimum, record:

• record the position-level full-Kelly output;
• alpha and reason;
• stake after operator limits and rounding;
• total open stake and maximum liability;
• shared teams, fixtures, markets, and model features;
• bankroll snapshot and settlement order.

No-bet boundary

If the full-Kelly expression is zero or negative, fractional Kelly is also zero for a back-only decision. Do not apply a minimum stake that overrides this boundary, and do not replace an unavailable recommended stake with the operator maximum.

Stress both probability and fraction

Build a two-dimensional table with plausible probabilities in rows and alpha values in columns. For every cell, calculate the cash stake and total open exposure. This shows whether reducing alpha meaningfully controls the uncertainty in p or merely scales an unstable signal.

Record the output at break-even and at an adverse probability. If full Kelly changes sign, every positive fractional output based on the original point estimate should be flagged for review. Also compare the sum of fractions across simultaneous positions with a portfolio-level constraint. A quarter of five separate Kelly stakes is not automatically quarter-Kelly exposure for the combined bankroll, especially when outcomes share teams, matches, or model inputs.

Next step

Use Kelly Criterion Betting for the next part of this topic.

Continue learning

• Next guide: Kelly Criterion for Betting
• Related guide: Martingale and Fibonacci Betting Systems

Assumptions and limitations

The example assumes a binary payoff, known price, one currency, and no commission. Full and fractional Kelly optimize model-defined objectives, not affordability or wellbeing. Neither method proves that the forecast is calibrated, the price is executable, or gambling is suitable for the reader.