Roulette Systems Under the Math Lens: Martingale, Fibonacci, D'Alembert, Labouchère & James Bond

'This system beats roulette.' — a claim first documented in 1796, the year of the original Martingale. Hundreds of variants have followed. Not one has ever beaten the house edge. This article works through the seven best-known systems mathematically and explains why each must fail — and which illusion each one creates.

The setup: European roulette has 37 pockets (0–36). An even-money bet (red, black, odd, even) wins with probability 18/37 = 48.65%. The house edge is exactly 1/37 = 2.70%. Per €1 staked you lose an expected 2.7 cents. This expected value is **additive** across all stakes: turning over €10,000 means losing an expected €270 — whether across 10 spins or 10,000, regardless of system.

The core theorem that kills every system: the **linearity of expectation**. E(A + B) = E(A) + E(B). If every individual stake has negative expected value, **every combination** of stakes has negative expected value — proportional to total turnover. No betting pattern, no doubling, halving or bridge rule changes that. It's not belief, it's algebra.

**System 1 — Martingale (negative progression):** Double after every loss, return to base stake after a win. Sounds watertight: you do win EVENTUALLY. Correct — but: with a €500 table limit and €5 base, after 7 losses (5→10→20→40→80→160→320) you hit the max bet; the 8th doubling step (€640) is illegal. Probability of 7 losses in a row: (19/37)⁷ ≈ 1.17%. Across 1,000 attempts this happens ~12 times. Each of those 12 costs you 5+10+20+40+80+160+320 = €635. The 988 successful rounds bring 988 × €5 = €4,940. Net: 4,940 − 12 × 635 = **−€2,680**. Expected value per Martingale cycle = exactly −2.7% of turnover, always.

**System 2 — Reverse Martingale / Paroli (positive progression):** Double after every win, return to base after a loss. Tries to ride hot streaks. Problem: a 5-win streak has probability (18/37)⁵ ≈ 2.7%. Cash-out gain at 5 wins: 5 × 2⁵ − 5 = €155 (on €5 base). But: 97.3% of cycles end earlier, losing the current stake. Expected value: same formula as Martingale, same −2.7%. You swap 'many small wins with one big loss' for 'many small losses with one big win'. Different distribution — identical mean.

**System 3 — Fibonacci:** Stakes follow the Fibonacci sequence (1-1-2-3-5-8-13-21-34-55-…). Step forward one after a loss, step back two after a win. Gentler than Martingale, same fate: after 9 losses in a row (prob. (19/37)⁹ ≈ 0.31%) you've lost 1+1+2+3+5+8+13+21+34 = 88 units. Bankroll drawdown is slower, table limit hits later — but it hits. Expected value over 1,000 rounds at €5 base: roughly **−€135** (simulation), matching the unavoidable −2.7% × turnover.

**System 4 — D'Alembert:** +1 unit after a loss, −1 unit after a win. The most popular 'gentle' method. Deceptively attractive because drawdowns look small. Mathematically it rests on a false premise — the **equilibrium expectation**: 'long-term, red and black balance out.' True in ratio, **not in absolute numbers**. After 1,000 spins the expected difference Red−Black is ±31 (standard deviation). Betting on convergence means betting on something that never arrives. Expected value: same, −2.7% of turnover.

**System 5 — Labouchère (cancellation):** Write down a sequence (e.g. 1-2-3-4-5), stake the sum of the outer two (1+5 = 6). On a win cross both off, on a loss append the stake at the end. When all numbers are crossed off, you've won the starting total (15). Sounds clever — collapses explosively on losing streaks: after 6 losses from 1-2-3-4-5, the list contains values like 16, 22, 28, 34, with the next stake at 35 units. Without an infinite table limit, you fail. Expected value: −2.7%. Math has no patience.

**System 6 — James Bond (combination):** Flat coverage across the table: €14 on High (19–36), €5 on the six-line 13–18, €1 on 0. €20 per spin, covers 25 of 37 numbers. Win distribution: 19–36 (prob. 18/37) → +€8. 13–18 (prob. 6/37) → +€10. 0 (prob. 1/37) → +€16. 1–12 (prob. 12/37) → −€20. Expected value: (18×8 + 6×10 + 1×16 − 12×20)/37 = (144+60+16−240)/37 = **−€0.54 per spin** = exactly −2.7% of €20. Elegantly constructed, mathematically identical to any other roulette stake.

**System 7 — Parlay (let-it-ride):** Re-stake the entire win on the next spin until you hit a target multiple. €5 → €10 → €20 → €40 → €80. Probability of 4 wins in a row: (18/37)⁴ = 5.6%. Expected payout: 0.056 × €80 + 0.944 × €0 = €4.48 on €5 stake = **−10.4% 'efficiency'**. Higher variance than every other system, same expected value relative to turnover. Condenses maximum hope into minimal volume.

What all seven share: they reshape the **distribution of outcomes**, not its mean. Martingale: high per-round win probability, small amounts, rare massive losses. Parlay: low per-round win probability, large amounts, frequent small losses. Both distributions have the same expected value: −2.7% of staked volume. You pick your risk profile, not your profit.

**Comparison after 1,000 spins at €5 base (simulation, EU roulette, €500 table limit):** Martingale: median −€50, 5%-quantile −€1,200, 95%-quantile +€85. Paroli: median −€135, 5% −€280, 95% +€420. Fibonacci: median −€90, 5% −€650, 95% +€110. D'Alembert: median −€135, 5% −€410, 95% +€95. Flat bet (constant €5): median −€135, 5% −€350, 95% +€75. **Expected loss is the same everywhere.** What differs: how often you experience extreme outliers.

The psychological hook: betting systems produce many small, frequent confirmations ('It works! I'm up!'). That confirmation IS the business model. The rare catastrophes that wipe everything out get filed under 'bad luck' — as if the system 'worked, but not today'. Classic survivorship bias: people who win 9 sessions with Martingale tell the story. People who lose 4 months of gains in session 10 stay silent.

**What actually reduces losses?** Three mathematically verifiable levers: (1) game choice — French roulette with La Partage has 1.35% instead of 2.7% house edge on even-money bets, halving expected loss. (2) Less turnover — fewer spins = less volume = less absolute loss. (3) Bankroll management with a firm stop-loss — doesn't change expected value, but limits loss concentration.

**When ARE systems useful?** As a **discipline tool**, not a profit tool. A fixed staking scheme prevents emotional escalation ('now I'll show them'). Someone who escalates by gut after losses risks bigger drawdowns than a Martingale player with a hard table limit. A system provides structure — structure beats gut. But structure is not an edge. It's just an orderly path to the same expected value.

Tools on Casinokeller: the bankroll simulator shows 1,000 players using the same system side by side — the distribution of end balances makes the 'same median, different variance' effect immediately visible. The house edge calculator gives expected loss for any game and setup. Both tools are ad-free, no affiliate links.

Related articles: 'Why do I always lose at roulette' (house-edge math), 'House edge explained — the complete guide' (pillar), 'Variance vs RTP' (distribution effects). External sources: expected value on Wikipedia, Wizard of Odds (roulette math).

Bottom line: roulette systems are not strategies, they're staking patterns. They redistribute a mathematically fixed loss differently over time — more rare large losses or more frequent small ones. Understanding that means giving up the search for the perfect system and seeing the game for what it is: paid entertainment with a measurable price. Not understanding it means funding the next generation of system sellers. The most honest strategy is to stop looking.