Probability
Why Martingale Can Fail: The Cost of Repeated Doubling
Doubling after a loss can create many small recoveries and occasional large failures. The arithmetic of the next required stake explains why this pattern is not a guarantee of profit.
The appeal of the recovery calculation
Consider a fictional bet paying decimal odds of 2.00, with no fees or void outcomes. Starting at one unit, the sequence after consecutive losses is 1, 2, 4, 8 and so on. A win returns twice that round's stake.
If the fourth bet of eight units wins after losses of one, two and four, its net gain is eight. Subtracting the previous seven units lost leaves a one-unit profit for the sequence. This identity explains the system's appeal, but it assumes the next stake can always be placed.
Losses and required stakes grow together
After n consecutive losses from a one-unit start, cumulative loss is 2ⁿ − 1, and the next required stake is 2ⁿ. After six losses, 63 units are gone and the next stake is 64. After ten losses, the loss is 1,023 and the next stake is 1,024.
A starting balance of 100 cannot fund the seventh bet after six consecutive losses: only 37 remain. It is not enough that the starting balance exceeded any of the first six individual stakes. The earlier losses must also be funded.
Even a fair model supplies no free advantage
Take a fair independent 50/50 model and stop either after the first win or after six losses. The chance of six losses is 1/64. Every other sequence ends with a one-unit profit, while the all-loss sequence loses 63 units.
The expected result is (63/64 × 1) − (1/64 × 63) = 0. A high frequency of small successful sequences has not created positive expectation. The rare failure exactly offsets them in this idealised fair example.
Real constraints cannot be wished away
A maximum permitted stake can prevent further doubling even when funds remain. Odds below 2.00 also invalidate the simple one-unit recovery calculation. For example, a two-unit win at 1.90 yields a net gain of 1.80, not two.
Raising a starting stake changes the size of every amount; it does not remove the exponential growth. Nor does a previous losing run make the next round more likely to win in an independent model.
A report showing many completed one-unit recoveries is incomplete unless it includes failed sequences, maximum exposure and the stopping rules. This example explains the risk of the system, not a recommendation to use it.
Related reading
For adults aged 18 and over. Educational content, not a promise of profit. Read our responsible gambling guidance.
Back to the blog