Why staking plans cannot work

2026-09-10

Almost every “baccarat system” reduces to one idea: raise the stake after a loss, take the money after a win. Martingale, cable, 1-3-2-4, D’Alembert — different shapes, same core.

They all work. Most nights. That is exactly why they keep selling.

Run a real one, unchanged

We reimplemented a staking plan that is actually in circulation, without changing a number:

• Capital $5,000, stop at +$1,000 net
• Win ladder: 100 → 200 → 300 → 100 → 200, change table after five wins
• Any loss drops to the loss ladder: 300 → 700
• Three losses in a row: walk, stake back to 100
• Ties push and count toward neither streak

Outcomes use real baccarat probabilities (Banker 45.86% / Player 44.62% / Tie 9.52%), with a 5% commission on Banker. 200,000 complete sessions, each run to target or bust.

The results

Four sessions in five end with a thousand dollars in hand. That is why it feels like it works — and the feeling is not an illusion. 79% is real and reproducible.

And the average session loses $179. Also real.

How both are true

Because the wins and the loss are wildly asymmetric:

• Win: +$1,000, 79% of the time
• Loss: −$5,000, 21% of the time

0.7908 × (+1,000) + 0.2092 × (−5,000) = 790.8 − 1,046.0 = −$255

(The simulation measures −$179, slightly better, because a busting session often cannot fund its final stake in full and bets only the remaining balance.)

Four small wins do not cover one large loss. A progression does not raise your win rate; it redistributes the loss — converting “lose small often” into “win small often, plus occasionally lose enormously.” The total is unchanged.

The number that decides everything

Look at the last row: realised loss rate −1.30%.

Against theory: a 60/40 Banker/Player mix has expectation −1.25%. The simulation returns −1.30%.

That is all of it. However the stakes are arranged, whenever you change tables, whatever streak you stop on, your money disappears at roughly 1.3 units per 100 of turnover. The only thing a staking plan influences is how fast the turnover accumulates — and raising stakes means faster, which means losing faster.

This is not a poorly designed plan. Expectation is additive:

E[total] = E[bet 1] + E[bet 2] + ... + E[bet n]

Every term is negative. Any arrangement, any trigger condition, any filter that “only bets on certain roads” merely selects which negative numbers get added up. The sum is always negative.

Would a bigger bankroll help?

No — it makes it worse. A larger bankroll:

• busts less often (it survives more losing runs)
• reaches the target more reliably in a single session
• but generates more turnover, at an unchanged loss rate → larger total losses

Then there are table limits. Every doubling sequence eventually meets the maximum bet, and at that moment every “it has to come back eventually” assumption fails at once.

Watch it run

The demo at the top of the home page is this exact plan, running live. There is a “fast-forward 5,000 sessions” button: press it and you will see the target rate settle around 79% while the cumulative total heads down and stays there.

Read next: The real probabilities and the house edge · How common are long streaks · How to read baccarat roads · The third card rule · Martingale’s two walls