Compare methods on one frozen decision set
Use identical events, probabilities, accepted prices, results, settlement order, and starting bankroll. Changing the selections as well as the stakes makes it impossible to attribute the path difference.
| Method | Stake input | Main benefit | Main limitation |
|---|---|---|---|
| Level cash | Fixed amount | Simple and reproducible | Does not shrink automatically after loss |
| Level units | Fixed unit count | Normalizes reporting | Cash exposure can change when unit value changes |
| Current-bankroll percentage | Bankroll and fraction | Scales cash stake with balance | Reset and denominator choices matter |
| Kelly-style | Probability, payoff, bankroll | Connects stake to a growth objective | Highly dependent on model inputs and assumptions |
| Fractional Kelly | Kelly output and fraction | Reduces model-derived exposure | Fraction remains a risk choice, not an error correction |
Worked three-bet path
Starting bankroll is GBP 500. Method L stakes GBP 5 each time. Method P stakes 1% of the current bankroll after each settlement. The illustrative outcomes are loss, win at 2.00, loss.
| Step | Level bankroll | Percentage bankroll |
|---|---|---|
| Start | 500.00 | 500.00 |
| After loss | 495.00 | 495.00 |
| After win | 500.00 | 499.95 |
| After loss | 495.00 | 494.95 |
For Method P, the second stake is GBP 4.95 and the third is GBP 5.00 after rounding to pennies. Small path differences arise from compounding and rounding; they do not establish superiority.
Evaluate more than final profit
Record total stake, maximum single stake, maximum open exposure, maximum drawdown, stake concentration, and cash required. Use the same settlement and rounding rules. OpenStax's expected-value and variance treatment explains why an average expectation and the spread of possible paths are separate properties.
Probability-led variation needs validated probabilities
Kelly's original model assumes known probability and payoff inputs. Sports-betting research that explicitly treats the win probability as unknown can produce materially different, generally smaller fractions in the studied cases (Chu, Wu and Swartz). Do not interpret a confidence label or recent win rate as a validated Kelly input.
Decision rule
Choose the least complex method that meets the documented objective and can be followed through drawdowns. Freeze it for the evaluation period. A method change starts a new series; it should not be applied retrospectively to make historical results look smoother.
Reverse the outcome order
Run the same wins and losses in at least two plausible settlement orders. Level cash staking produces the same arithmetic total when prices and stakes are fixed, while current-bankroll percentage stakes can produce a different ending value because later cash stakes depend on earlier outcomes. This is path dependence, not evidence that one order is more likely.
Report how rounding and simultaneous settlement are handled. If several bets are placed from one snapshot, calculate all stakes from that snapshot rather than allowing an arbitrary database settlement order to change them. A fair comparison fixes this policy in advance and uses it for every method.
Next step
Use Flat Staking Strategy for the next part of this topic.
Continue learning
- Next guide: Bet Sizing Under Probability Uncertainty
- Related guide: Can a Betting Bankroll Support Income? Evidence and Limits
Assumptions and limitations
The worked path is illustrative, uses sequential settlement, and ignores commission and dependence. No staking comparison proves that the underlying betting method has positive expectation. Affordability, operator limits, and harmful-gambling controls remain separate release conditions.
GambleAware
GamCare
Gambling Therapy
GAMSTOP