Doubling After a Loss Changes the Risk, Not the Odds

The Martingale betting system (stake one unit, double after every loss, reset after a win) does not beat the house. It leaves the probabilities and expected value of every wager exactly where they were. What it changes is the shape of your results: frequent one-unit wins, paid for by a rare loss that grows exponentially with the length of the progression.

The numbers make the trade plain. With a $5 base stake and a six-wager cap, you expose $315 to win $5. On a single-zero roulette wheel with a ten-wager cap, about 99.8725% of sequences end one unit ahead, about 0.1275% end 1,023 units behind, and the average result is roughly −0.3056 units per sequence.

Enter your base stake and bankroll; the exposure, failure chance and expected result update below.

Every Winning Sequence Pays Exactly One Base Unit

Start with a $5 bet on red. If it loses, bet $10. If that loses, bet $20, and keep doubling after each loss. After a win, return to $5. If you reach a cap you set in advance, stop; nothing about the system lets you assume you can continue indefinitely.

An even-payoff win at any affordable level covers all earlier losses and leaves a profit equal to the original stake:

Sequence	Net result
Lose $5, win $10	+$5
Lose $5 and $10, win $20	+$5
Lose $5, $10 and $20, win $40	+$5

That accounting is real. The error is reading the recovery mechanism as evidence that the wager has improved. A staking pattern decides how much money is exposed to each outcome. It does not make red, black or zero any more or less likely to appear.

Negative expected value also does not mean every finite session loses. Short-term variance can leave a player ahead, and Martingale is unusually good at producing sequences that end with a visible one-unit gain.

The question that matters is how much the system wins when it succeeds and how much it loses when it fails. For the classic Martingale the first answer is one base unit. The second grows exponentially.

Ten Wagers Turn One Unit Into 1,023 Units Of Exposure

Call the opening stake the base unit. Wager 1 is the initial bet, and n means the total number of wagers permitted, not the number of times the stake is doubled.

Wager	Stake (units)	Cumulative exposure
1	1	1
2	2	3
3	4	7
4	8	15
5	16	31
6	32	63
7	64	127
8	128	255
9	256	511
10	512	1,023

The stake column is the next bet after consecutive losses. Cumulative exposure is the amount lost if every wager through that row loses. The totals follow a geometric series and match the standard finite calculation in this overview of the betting system: six wagers expose 63 units, seven expose 127 and ten expose 1,023, while a successful sequence still targets one unit.

A $5 Base Stake Puts $315 At Risk Over Six Wagers

With a $5 unit the six stakes are $5, $10, $20, $40, $80 and $160. Losing all six costs $5 + $10 + $20 + $40 + $80 + $160 = $315.

If the sixth wager wins, it recovers the preceding $155 of losses and adds $160 in winnings, leaving the player $5 ahead. If it loses and the player cannot or will not place a $320 seventh wager, the sequence ends $315 behind.

The opening stake disguises this commitment. A player may think of the progression as a “small $5 system” while its sixth wager is $160 and six straight losses cost far more.

Each added level doubles the next stake exactly and nearly doubles total exposure, from B × (2ⁿ − 1) to B × (2ⁿ⁺¹ − 1). Extending the cap does not remove the large loss. It makes that loss rarer within one sequence and bigger when it arrives.

Three Formulas Price The Trade-Off

Let B be the base stake, p the independent probability that a single wager loses, and n the total number of permitted wagers.

The formulas assume independent outcomes with a fixed loss probability, an even-payoff win, a full loss on a losing wager, doubling after each loss, and a reset after either a win or n consecutive losses. This is the finite-round model used in Peter Pflaumer’s statistical analysis of the roulette Martingale.

Complete Failure Requires n Straight Losses

The probability that a capped sequence fails completely is p raised to the power n. Adding a level makes failure less likely within that sequence, because one more loss is required. It never makes failure impossible.

Bankroll And Table Maximum Are Separate Tests

The permitted stakes run B, 2B, 4B and so on up to B × 2ⁿ⁻¹. If all n wagers lose, the total loss is B × (2ⁿ − 1). The final wager alone is B × 2ⁿ⁻¹.

That produces two affordability conditions. The bankroll must be at least B × (2ⁿ − 1), and the table maximum must be at least B × 2ⁿ⁻¹.

Both must hold. A player can have enough money to fund the sequence and still be unable to place the final wager because of the table limit. The reverse also happens: the table accepts the wager, but the player lacks the money to complete the progression.

Expected Profit Is B Times One Minus (2p) To The Power n

Every successful sequence earns B, and a complete failure loses B × (2ⁿ − 1). Expected profit for one capped sequence is therefore B × (1 − pⁿ) minus B × (2ⁿ − 1) × pⁿ, which simplifies to B × [1 − (2p)ⁿ].

This is how a high success frequency coexists with a negative average. All successful sequences earn the same small amount, while the unsuccessful outcome becomes exponentially more expensive as n rises.

For an unfavorable even-payoff wager, p is greater than one half, so 2p is greater than 1 and (2p)ⁿ grows with every added level. More permitted wagers reduce the frequency of capped failures and make the expected result per sequence more negative.

Single-Zero Roulette With A Ten-Wager Cap Loses 0.3056 Units Per Sequence

This example uses the conference paper’s specific model: a single-zero wheel with 37 equally likely outcomes, an even-money wager such as red with 18 winning numbers and 19 losing outcomes including zero, independent spins, a full loss on zero, a one-unit base stake and a maximum of ten wagers.

For that model the paper reports a per-unit expectation of approximately −0.027027, a ten-wager sequence expectation of approximately −0.3056 units, a variance of 1,335.7 and a standard deviation of approximately 36.54 units per capped sequence.

The probability of losing one red wager is 19 ÷ 37. Expected profit per unit staked is 18/37 × 1 plus 19/37 × (−1), which equals −1/37, or about −0.027027.

The ten permitted stakes are 1, 2, 4, 8, 16, 32, 64, 128, 256 and 512 units.

Quantity	Value
Final stake	512 units
Loss if all ten lose	2¹⁰ − 1 = 1,023 units
Chance of complete failure	(19/37)¹⁰ ≈ 0.1275%
Chance of a one-unit win	≈ 99.8725%
Expected result	1 − (38/37)¹⁰ ≈ −0.3056 units
Standard deviation	≈ 36.54 units

Those percentages explain the system’s appeal. Nearly every capped sequence in this model ends with a visible win. The remaining outcome is not a loss of one or two units; it is a loss of 1,023.

The negative mean and the large standard deviation describe different properties. The mean says the player loses about 0.3056 units per comparable sequence on average. The standard deviation reflects the imbalance in possible results: many one-unit gains sitting beside a rare 1,023-unit loss.

These figures are not universal roulette results. Double-zero roulette has a different loss probability. Single-zero tables using la partage or en prison treat some even-money outcomes differently. A different game, payout, base stake or sequence cap requires a new calculation.

A Bigger Bankroll Or A Higher Limit Buys Rarer, Larger Losses

A larger bankroll supports more levels and makes a capped failure less frequent within any one sequence. It also increases the amount lost when the longer progression fails.

The target does not grow with the risk. Six successful levels produce one base unit, and so do seven or ten. What changes is the amount required to keep pursuing that unit.

A higher-limit table has the same effect. It may permit another doubling, but it cannot improve the probability of the next independent spin. Moving between tables or casinos extends the available betting range; it does not turn previous losses into information about the next outcome.

A losing streak therefore does not make the chosen color “due.” Under the independence assumption, the next spin has the same probabilities whether the preceding spins were evenly divided or included an unusual run of one color. Martingale reacts to past results by changing the amount wagered, not by predicting the next result.

The unlimited-capital intuition holds only in a narrow sense. If a player has arbitrarily large capital, faces no stake limit, has potentially unlimited time, continues until the first win, and has a positive chance of winning each trial, then the probability of eventually recording a win approaches one. A University of Toronto probability text identifies access to arbitrarily large capital as the hidden requirement behind “double until you win.”

Practical gambling has finite bankrolls, finite stakes and finite playing time. Table limits expose that weakness, but they are not the source of the negative expectation. Even with no table limit, changing stake sizes would not improve what each unit wagered is worth.

Martingale And Flat Betting Share The Same Expected Value Per Unit

Flat betting means wagering the same amount on every trial. Martingale changes the stake in response to the previous result. Neither changes the probabilities or payout of the underlying wager.

Comparison	Martingale	Flat betting
Stake pattern	Doubles after losses; resets after a win or cap	Same stake every wager
EV per unit staked	Same as the underlying wager	Same as the underlying wager
Short-term profit frequency	Depends on the cap and comparison design	Depends on the wager count and comparison design
Loss pattern	Concentrated in a capped losing sequence	One fixed stake at a time
Bankroll pressure	Exponential during a losing streak	Linear with wagers lost

A precise profitability comparison has to hold something constant: the same number of spins, the same total amount wagered, the same playing time, the same starting bankroll, or the same profit target and loss limit. These are not interchangeable. Ten Martingale sequences may involve different numbers of spins and much higher turnover than ten flat bets, so comparing only the percentage of profitable sessions hides how much was wagered and how large the losing sessions became.

In the cited finite roulette analysis, Martingale showed a higher probability of short-term positive profit than constant-size betting in the paper’s modeled comparison, together with much larger losses when the progression failed. That is not a universal ranking for every choice of bankroll, wager count, turnover and stopping rule.

Flat betting does not prevent losses or remove the house edge. It simply avoids raising the next stake because earlier wagers lost. The two methods distribute short-term results differently, and both retain the expected value of the underlying wager per unit staked.

The Betting System Is Not The Martingale Of Probability Theory

In probability, a martingale is a stochastic process satisfying integrability and information conditions whose conditional expected next value, given everything observed so far, equals its current value. That is a mathematical property of a process, not an instruction to double a wager after a loss.

Berkeley’s lecture notes on stopping times and martingales also explain why optional-stopping conclusions require conditions: bounded stopping times support results that do not automatically extend to every unbounded stopping rule.

Evaluate The Wager First, Then Fix The Loss Limit

For a recreational player, the order of analysis is the wager first and the staking pattern second. Examine the probabilities, payout, rules and expected value. A progression cannot remove the cost built into an unfavorable wager.

Bankroll caps and stop-loss limits can contain losses; they cannot turn Martingale into an advantage. Set those limits before play, use only money that can be lost without harm, and treat the full gambling bankroll as entertainment spending rather than investment capital.

Martingale makes a small win appear frequently while the required exposure grows to 63, 127, 1,023 or more units for the same one-unit target. Anyone who chooses to gamble is better served by selecting a lower-cost underlying wager and deciding the maximum acceptable loss in advance than by treating a doubling progression as a winning system.