The Martingale system and why it cannot change the house edge

The Martingale system doubles the stake after each loss on an even-money bet, so the first win recovers all losses plus one base stake. It does not beat the game: every bet keeps its house edge, so the expected loss is the edge times the total staked. The system trades frequent small gains for a rare, very large loss.

By Nairobi Ultimate Editorial (prepared with AI tools under our editorial policy) · Last updated 2026-10-08

The rule of the system

The Martingale system is a stake-sizing rule for bets that pay even money, such as red or black on roulette. Start with a base stake. After a loss, double the stake; after a win, return to the base stake.

The appeal is a piece of arithmetic that is true. The first win after any run of losses always leaves the player exactly one base stake ahead of where that sequence began. The question is whether that makes the game beatable, and the answer depends on what the rule does to the average, which is what the expected value of a bet measures.

Step 1: why one win repairs every loss

Take a base stake of KES 20, which is also the smallest online bet allowed by section 71 of the Gambling Control Act, 2025. Suppose the first k bets are lost and the next one wins.

The losses so far are 20 + 40 + 80 + ... + 20 x 2^(k-1) = 20 x (2^k - 1). The next stake is 20 x 2^k, and an even-money win pays profit equal to the stake. The net result is 20 x 2^k - 20 x (2^k - 1) = KES 20, whatever the value of k.

Step 2: the chance of each bet

A single-zero wheel has 37 pockets: 1 to 36 and a zero. Red covers 18 of them, so the chance of red on one spin is 18/37, about 48.65%. The chance of losing is 19/37, about 51.35%, because the 18 black pockets and the zero all lose a bet on red.

Spins are independent, so the chance of losing n bets in a row is (19/37) multiplied by itself n times. The chance that a sequence ends with a win within n bets is one minus that figure.

Step 3: the cost of a long losing run

A sequence that is allowed at most n bets ends in one of two ways. With probability 1 - (19/37)^n it ends with a win of KES 20. With probability (19/37)^n every bet loses, and the whole 20 x (2^n - 1) is gone.

The average result is 20 x (1 - L) - 20 x (2^n - 1) x L, where L is the chance that all n bets lose. For n = 10 it is about 20 - 20,480 x 0.001275 = -KES 6.11. Figures are rounded and use a single-zero wheel with 18 winning pockets.

The frequent small win and the rare large loss are one and the same bargain. A long run of apparent success is what the system produces most of the time, which is why it can look reliable until the rare run arrives.

Step 4: the edge is untouched

Every bet in the sequence is a bet on red with the same chance of winning. Its expected value is (18/37) x 1 - (19/37) x 1 = -1/37 of the stake, about -2.70%, as set out in the guide to the roulette house edge.

Expected values add up. The expected result of the whole sequence is therefore -1/37 times the expected total amount staked in it. The stakes change from bet to bet, but the factor 1/37 does not, so the total can never reach zero while anything is staked.

The check agrees with Step 3. For n = 5 the expected amount staked is 20 x 37 x ((38/37)^5 - 1), about KES 105.5, and one thirty-seventh of that is about KES 2.85, matching the result above.

What stops the doubling in practice

Real play meets two ceilings. One is the balance: the stake 20 x 2^k must be available after k losses. The other is the maximum stake a game allows, which is part of that game's rules.

The Act and Regulations read for this guide give a minimum online bet but no general maximum. Regulation 20(2) of the 2026 Regulations requires a licensee to display "rules of games, odds, house edge and average return to the player", so a game's own limits should be readable before staking.

The same problem in other progressions

Doubling is one rule among many. Others add a fixed amount after a loss, follow a number series, or raise stakes after wins instead. Each is a function that turns past results into the next stake.

None of them changes the chance on the next spin, which depends only on the pockets and the payout. Each bet has the same negative expected value, so any total of such bets has one too. The pattern differs, with long quiet stretches and sudden large swings, which is what RTP and house edge hide when a game is described only by its average.

Why it still feels convincing

• Short samples. Most sequences end with a gain, so a few dozen sequences can show no loss at all.
• Hidden size of the loss. Many gains of KES 20 are easy to count. One loss of KES 5,100 or more is a single event and is easy to treat as bad luck.
• The belief that a win is due. A long run of losses does not make the next spin more likely to win. This is the gambler's fallacy, covered in the gambler's fallacy and session limits.
• A bigger bet after a loss is chasing. Doubling is chasing with a formula, and it raises the amount staked exactly when the balance is lowest.

What a limit does and does not do

A loss limit or a session limit does not improve any bet. It caps how much money meets the edge, which is the only part of the cost a player can control.

Regulation 87(2) of the 2026 Regulations requires licensees to provide, as a minimum, deposit limits on a daily, weekly and monthly basis, loss, session and expenditure limits, reality checks showing duration of play, easy self-exclusion for at least twenty-four hours, a link to the national self-exclusion register and a player savings tool. Where gambling has stopped being a choice, the responsible gambling page sets out the regulator's toll-free line and the self-exclusion process. Gambling is addictive: play responsibly and only with money that is not needed for anything else.