Martingale's Two Walls: The Table Limit and Your Bankroll
The pitch is clean: double after a loss, and one win returns everything plus one base unit. As long as you do not go broke, you always get it back.
The catch hides inside “as long as you do not go broke” — and the wall you hit first is usually not the one you were watching.
Wall one: the table maximum
However much money you bring, the table caps how many times you can double. With a $100 base:
| Table max | Multiple | Hands you get | Last bet | Down by then | P(that many losses) | Hands between hits |
|---|---|---|---|---|---|---|
| $1,000 | 10× | 4 | $800 | $1,500 | 6.598% | 29 |
| $2,000 | 20× | 5 | $1,600 | $3,100 | 3.344% | 59 |
| $5,000 | 50× | 6 | $3,200 | $6,300 | 1.695% | 118 |
| $10,000 | 100× | 7 | $6,400 | $12,700 | 0.859% | 234 |
| $20,000 | 200× | 8 | $12,800 | $25,500 | 0.435% | 464 |
| $50,000 | 500× | 9 | $25,600 | $51,100 | 0.221% | 917 |
| $100,000 | 1000× | 10 | $51,200 | $102,300 | 0.112% | 1,811 |
An ordinary $100–$10,000 table gives you seven hands. Lose the seventh and your next bet would be $12,800; the dealer will not take it. The progression stops there and the $12,700 stops being theoretical.
0.859% sounds safe. It is not a one-time probability — it applies to every run you start:
| Table max | Losses to break | At least once in 200 hands | In 500 | In 1,000 |
|---|---|---|---|---|
| $1,000 | 4 | 99.95% | 100.00% | 100.00% |
| $5,000 | 6 | 82.34% | 98.76% | 99.99% |
| $10,000 | 7 | 57.40% | 88.54% | 98.72% |
| $50,000 | 9 | 19.12% | 41.85% | 66.44% |
| $100,000 | 10 | 10.10% | 23.89% | 42.32% |
A shoe runs about 65 hands. Three shoes at a standard table and you are past even money to hit the wall.
Wall two: the bankroll — and it works backwards
Intuition says a bigger bankroll is safer. We ran it: bet Player, ties push, $100 base, walk away at +$1,000 net, 120,000 complete sessions per row.
| Bankroll | Table max | Hit target | Busted | Avg net per session | Avg wagered | Measured loss rate |
|---|---|---|---|---|---|---|
| $2,000 | $10,000 | 64.55% | 35.45% | −$63 | $5,059 | −1.254% |
| $5,000 | $10,000 | 81.91% | 18.09% | −$85 | $6,867 | −1.241% |
| $10,000 | $10,000 | 89.94% | 10.06% | −$106 | $8,700 | −1.220% |
| $20,000 | $10,000 | 94.60% | 5.40% | −$134 | $10,446 | −1.281% |
| $50,000 | $10,000 | 97.53% | 2.47% | −$260 | $19,924 | −1.303% |
| $20,000 | $1,000 | 92.29% | 7.71% | −$620 | $49,882 | −1.243% |
From $2,000 to $50,000:
- Sessions that end in profit: 64.6% → 97.5%
- Average net per session: −$63 → −$260
Half again as many winning nights, four times the money gone.
There is no contradiction. What the bankroll buys is the ability to absorb longer losing runs — which means playing longer and pushing more money across the felt. The cost per dollar of turnover does not move. Survive longer, wager more, pay more.
The last row makes it plainest. A $20,000 bankroll at a table capped at $1,000 hits the target 92.29% of the time, which looks excellent — but the capped stake stretches each session out, average turnover climbs to $49,882, and the average session loses $620. That is 4.6× the loss of the same bankroll at a $10,000 table.
Banker is worse, structurally
Banker pays 0.95 after commission, so plain doubling no longer recovers the sequence. To keep the property “one win repays everything and nets $100,” the ladder has to climb by about 2.105× a step:
Banker: $105 → $216 → $444 → $910 → $1,869 → $3,836 → $7,873 → $16,160
Player: $100 → $200 → $400 → $800 → $1,600 → $3,200 → $6,400 → $12,800At the same $10,000 table, Player reaches rung seven at $6,400; Banker’s rung seven is already $7,873 and rung eight is $16,160, outside the limit. The commission costs you a rung of cushion.
Banker’s house edge (−1.06%) really is lower than Player’s (−1.24%), so flat betting on Banker is the better bet. The moment you start doubling, that 0.95 payout flips from an advantage into a structural defect.
The number that never moved
Look again at the last column: −1.254%, −1.241%, −1.220%, −1.281%, −1.303%, −1.243%.
The exact house edge on Player is −1.2351% (full 8-deck enumeration, ties pushing). Seven combinations of bankroll and table limit, 120,000 sessions each, and every one landed within one standard error of it.
That is the whole story. Expectation is additive:
E[total] = E[bet 1] + E[bet 2] + … + E[bet n]Each term equals −1.2351% × that bet. Staking patterns, doubling rules, stop-losses, road-reading filters — all of them choose the sizes of the terms. None of them touches the sign.
Martingale does not raise your win rate. It reshapes the losses — trading “lose a little often” for “win a little often, lose enormously now and then.” The total is unchanged, and because turnover is amplified, the total is in fact larger.
Run it yourself
Every figure here is reproducible, and the code is public:
- Exact probabilities and house edge: github.com/baccaicom/baccarat-math — a full recursion over all 416 cards, not sampling
- Betting system simulator — change the bankroll, the limit, the system; it runs in your browser
- Risk-of-ruin calculator — bankroll and hand count in, exact ruin probability out
- Expected-value calculator — house edge for every bet on the layout
Method. Outcome probabilities come from a complete enumeration of the 416-card shoe (Banker 45.860% / Player 44.625% / Tie 9.516%), not an approximation. Tables 1 and 2 are exact (a Markov chain over losing-run length), not simulated. Table 3 is Monte Carlo: 120,000 complete sessions per row, fixed seed, error bars in the repository’s tests.
Read on: The mathematics of betting systems · Real probabilities and the house edge · How common are long streaks · How to read baccarat roads
FAQ
How big a bankroll does Martingale need?
There is no number that is enough. Going from $2,000 to $50,000 raises the share of winning sessions from 64.6% to 97.5% — and raises the average loss per session from $63 to $260. A bankroll buys a higher hit rate and a bigger loss, not an edge.
How many losses in a row does it take to break?
That is set by the table, not by your nerve. With a $100 base and a $10,000 maximum you can double six times; the seventh consecutive loss is turned away, and by then you are down $12,700. Seven losses in a row has probability 0.859% — but the chance of hitting one at least once in 200 hands is 57.4%.
Does Martingale work better on Banker?
It works worse. A Banker win pays 0.95, so plain doubling no longer recovers the sequence; the ladder has to grow by about 2.105× per step: $105 → $216 → $444 → $910 → $1,869. You reach the table maximum one rung sooner.
Is there any bankroll and limit that turns it positive?
No. Across seven combinations, 120,000 sessions each, every measured loss rate landed between −1.22% and −1.34%, scattered around the exact Player house edge of −1.2351%. Staking patterns change how fast money moves, not what each dollar of turnover costs.