Strategy
Staking plans: flat, percentage, confidence scale and the systems to avoid
A staking plan cannot turn a losing bet into a winning one. It can, however, turn a winning bettor into a losing one — if it is the wrong plan.
- 4
- Plans that work
- 4
- Plans that destroy
- 1,024×
- Martingale stake after 10 losses
- 0
- Plans that create an edge
The plans that work
1. Flat staking
The same amount on every bet, 1–2 % of the bankroll. Advantages: simplicity, clean yield measurement, maximum resilience to losing runs. Disadvantage: it does not exploit bankroll growth until the next review.
2. Percentage
A fixed percentage of the current bankroll. Mathematically never hits zero, compounds profits. Disadvantage: recovery from a drawdown is slower (you need +43 % to recover −30 %).
3. Confidence scale (1–3 units)
Flat with three levels: 1 unit standard, 2 for strong positions, 0.5 for doubtful ones. Works only if your log proves that “strong” positions really perform better. Most people discover they don’t.
4. Fractional Kelly
Stake proportional to the estimated edge — in detail under the Kelly criterion. Needs a quantitative model.
Simulation: 1,000 bets at 2.00, true probability 52 % (yield +4 %)
| Plan | Final bankroll (median) | Max drawdown | Probability of ruin |
|---|---|---|---|
| Flat 1 % | +40 % | −18 % | 0 % |
| Flat 2 % | +80 % | −33 % | 0 % |
| Percentage 2 % | +115 % | −30 % | 0 % |
| ¼ Kelly (≈2 %) | +110 % | −31 % | 0 % |
| Flat 5 % | +190 % | −62 % | 2 % |
| Martingale (base 1 %) | +9 % or ruin | −100 % | 78 % |
| Fibonacci (base 1 %) | +15 % or ruin | −100 % | 45 % |
| D’Alembert (base 1 %) | +25 % | −55 % | 8 % |
Indicative Monte Carlo results; variance between runs is large, the ranking is not.
The plans that destroy
Martingale
Double after every loss, back to base after a win. Each cycle wins 1 unit. After 10 losses the stake is 1,024 units and total exposure 2,047. Over 1,000 bets at 2.00 the probability of meeting a 10-loss streak is ~62 %. Bookmaker stake limits make the system impossible even with infinite capital.
Fibonacci
A “gentler” Martingale — the stake follows 1, 1, 2, 3, 5, 8, 13… Same problem, slightly slower: after 12 losses you stake 233 units.
D’Alembert
+1 unit after a loss, −1 after a win. Does not go bust as fast, but in a losing run exposure grows linearly and recovery needs more wins than losses — which don’t exist without an edge.
Labouchère (cancellation)
Write a sequence, stake the sum of the ends, cross off on a win, add on a loss. More complex, same fate.
The common problem
All progressive systems trade many small wins for one huge loss. At every point the expected value equals flat staking — but the distribution is exactly the wrong one for a human with finite capital and finite composure.
How to choose
- Fewer than 300 logged bets → flat 1 %.
- Proven positive yield → percentage 1–2 % or ¼ Kelly.
- A model with quantitative probabilities → ⅓ Kelly with a daily exposure cap.
- Thinking about Martingale → read the table again.
Bankroll management explains how to set the unit and when to review it.
FAQ
But surely Martingale 'must' win eventually?▾
Yes — and it wins one unit each time. Until the 10–12 loss streak, which is statistically certain within a few hundred bets and requires a stake of 1,000–4,000 units. No bankroll and no bookmaker limit allows it.
Is there a plan that improves yield?▾
No. Yield (profit per unit staked) depends only on the quality of your picks. The plan affects variance and the probability of ruin.
Which plan to start with?▾
Flat 1 %. It is the plan that reveals most clearly whether your picks have value.
Invitation
Simple plan, clean data
The more complex the system, the more easily it hides that you are losing.
21+ | Regulated by the HGC | Play responsibly