The Martingale System and Why Betting Systems Cannot Work
Martingale, d'Alembert, Fibonacci and Labouchère produce identical expected losses to flat betting. The proof, the bankroll arithmetic, and what table limits actually do.
The Martingale is the oldest betting system in gambling: after every loss, double your stake; after every win, return to the base bet. Each win recovers all previous losses plus one unit.
It works exactly as described. It also loses money at precisely the same rate as flat betting, and this page shows why.
The sequence
Base bet $10 on red at a European wheel.
| Spin | Bet | Result | Cumulative |
|---|---|---|---|
| 1 | $10 | Lose | −$10 |
| 2 | $20 | Lose | −$30 |
| 3 | $40 | Lose | −$70 |
| 4 | $80 | Lose | −$150 |
| 5 | $160 | Lose | −$310 |
| 6 | $320 | Win | +$10 |
Six spins, $630 wagered, $10 profit. The recovery is real and it is why the system is seductive: you win, over and over, until you don’t.
Why the expectation does not move
Expected value is linear. For any sequence of bets:
E[total] = E[bet 1] + E[bet 2] + ... + E[bet n]
This holds whether the bets are independent or not, and whether their sizes are chosen in advance or based on earlier results. It is a property of expectation itself, not an assumption about the game.
Every single bet on a European wheel has expectation −2.70% of its size. So:
E[total] = −0.027 × (total amount wagered)
That is the entire proof. The only thing a betting system can change is how much you wager. Martingale increases it dramatically after losses, which increases expected loss in absolute terms while leaving the rate untouched.
In the six-spin sequence above, the $630 wagered carries an expected loss of $17.01 — more than the $10 the system was trying to win.
The failure, priced
Martingale’s profit is small and near-certain; its loss is large and rare. Those two facts do not offset in your favour, and here is the arithmetic.
On a European wheel, red loses with probability 19/37 = 51.35%.
| Consecutive losses | Probability | Next bet needed | Total staked |
|---|---|---|---|
| 4 | 6.96% | $160 | $150 |
| 6 | 1.84% | $640 | $630 |
| 8 | 0.485% | $2,560 | $2,550 |
| 10 | 0.128% | $10,240 | $10,230 |
| 12 | 0.0337% | $40,960 | $40,950 |
A run of eight is not rare. It happens roughly once in 206 sequences — which, at 40 spins an hour, is a few hours of play. To survive it you need $2,550 in the bankroll and a table that accepts a $2,560 bet, all to protect $10 of profit per cycle.
The table limit is what actually kills it. A $10 minimum table typically has a $500 or $1,000 maximum. From $10, doubling reaches $640 on the seventh bet and $1,280 on the eighth. You are cut off after six or seven losses — an event with probability around 1.8%, so it happens roughly every 55 cycles.
Play 55 cycles: you win about $550 in $10 increments, then lose $630 in one sequence. Net, you have wagered a great deal of money to lose 2.70% of it.
The other systems
All of them are Martingale with different arithmetic, and all fail for the identical reason.
Grand Martingale — double and add one unit. Larger wins, larger losses, same expectation.
Anti-Martingale (Paroli) — double after wins instead. Frequent small losses, rare large wins. Mirror image; same expectation.
D’Alembert — increase by one unit after a loss, decrease by one after a win. Gentler progression, slower ruin, same expectation. Its premise — that wins and losses tend to equalise — is the gambler’s fallacy stated as a rule.
Fibonacci — step along 1, 1, 2, 3, 5, 8 after losses. Grows slower than doubling, so it lasts longer and recovers less. Same expectation.
Labouchère — write a sequence, bet the sum of the ends, cross off on a win, append on a loss. The sequence grows on losses and the required bets grow with it. Same expectation.
Oscar’s Grind — increase only after a win, target one unit of profit per cycle. Very low variance, very long cycles, same expectation.
Every one of these has been tried by very determined people for three hundred years. The result is a theorem, not an opinion: no strategy for varying stakes on independent negative-expectation bets produces a positive expectation.
The gambler’s fallacy underneath
Most systems rest on the belief that a run of losses makes a win more likely. It does not. The wheel has no memory, the shoe is not owed anything, and the dice do not track their history.
After ten reds, the probability of black on the eleventh spin is 18/37 — exactly what it was on the first spin. The 50-50 balance people expect emerges over millions of spins by dilution of early results, not by correction of them. A ten-red streak is never cancelled; it is simply outnumbered.
What actually changes the odds
There are exactly three categories, and none of them involve bet sizing on a fixed game:
- Card counting in blackjack — the composition of the shoe genuinely changes as cards are dealt, so the odds of the next hand change. Bet sizing then works, because you are betting more when the bets themselves are better.
- Positive-expectation machine games — a full-pay Deuces Wild machine at 100.76%, or a progressive jackpot above its break-even point. The bet itself is favourable.
- A mispriced line — sports betting or poker, where you are wagering against someone else’s judgement rather than a fixed mathematical structure.
All three change the odds of the underlying bet. That is the only thing that works.
If you use a progression anyway
Some people prefer the shape of the results and accept the cost, which is a defensible choice made honestly. If that is you:
- Know your table maximum before you start, and count how many doublings it allows.
- Size the base bet so the full progression fits inside your session bankroll.
- Understand that you are trading many small wins for a rare large loss, and that the trade is slightly negative, every time.
- Do not increase the base bet to “make back” a completed failed cycle. That is Martingale applied to Martingale.
See also: house edge across every game · responsible gambling