All Chapters

The Missing Chapter

The Martingale Trap

The mathematically "perfect" betting system that guarantees ruin

An extension of Jordan Ellenberg's "How Not to Be Wrong"

Chapter 22

The System That Can't Lose

In the gilded gambling halls of 18th-century France, a whisper circulated among the regulars at the roulette tables — a système so elegant, so logically airtight, that it seemed to guarantee profit. The idea was devastatingly simple: when you lose, double your bet. Eventually you'll win, and when you do, you'll recover everything you lost, plus your original stake. It felt like finding a flaw in the fabric of mathematics itself.

The strategy came to be called the martingale, likely after the residents of Martigues in Provence, who were stereotyped as reckless gamblers.1 But the gamblers who adopted it didn't feel reckless at all. They felt rational. They had, they believed, found the one weird trick that casinos didn't want you to know about.

They were wrong. Not about the math — the math, in a very specific and useless sense, is correct. They were wrong about the universe they lived in. And the gap between the universe where the martingale works and the universe we actually inhabit contains one of the most important lessons in all of applied mathematics: any strategy that requires infinite resources isn't really a strategy at all.

How It Works (In Theory)

Let's make it concrete. You walk up to a roulette table. You bet $10 on red. If red hits, wonderful — you pocket $10 profit and start over. If black hits, you don't slink away. You double down: $20 on red. If red hits now, you win $20, which covers the $10 you lost on the first bet plus $10 profit. If black hits again? You bet $40. Then $80. Then $160.

The key insight is that whenever you finally win — and on a fair coin you will eventually win — your payout covers all previous losses plus your original bet. The math is tidy:

Total wagered after n consecutive losses

B + 2B + 4B + + 2n-1B = (2n − 1)B

When you win on round n+1, you receive 2nB — netting exactly B in profit.

B
Your initial bet ($10 in our example)
n
Number of consecutive losses before a win
2nB
The bet you must place on round n+1

So no matter how long your losing streak, the moment you win, you're up exactly one unit. It's like a mathematical rubber band — the further you stretch it, the harder it snaps back. The expected profit from any single "round" of the martingale is positive.2

This is the part where the 18th-century gambler smugly orders another brandy.

The Exponential Wall

But let's look at what happens to your bets after a few losses. Starting with just $10:

Loss #Bet RequiredTotal InvestedProfit If You Win
1$10$10$10
2$20$30$10
3$40$70$10
4$80$150$10
5$160$310$10
6$320$630$10
7$640$1,270$10
8$1,280$2,550$10
9$2,560$5,110$10
10$5,120$10,230$10
11$10,240$20,470$10

Stare at that last column. After risking $20,470 — after surviving ten consecutive losses and putting over twenty thousand dollars on the table — your profit, if you win, is ten dollars. You've risked a semester of college tuition to win back the cost of lunch.

Loss Number Bet Size ($) 1 2 3 4 5 6 7 8 9 $10 $20 $40 $80 $160 $320 $640 $1.3k $2.6k

Bet size grows exponentially — by loss #9, you're wagering $2,560 to win back $10.

This is the fundamental asymmetry of the martingale. You win often, but you win small. You lose rarely, but you lose catastrophically. It's a strategy that converts a series of coin flips into a game of Russian roulette.

The martingale doesn't eliminate risk. It concentrates risk into rare, devastating events — and then bets your entire fortune that those events won't happen today.

⁕ ⁕ ⁕
Chapter 22

Why "Eventually" Isn't Good Enough

The martingale's theoretical proof rests on a beautiful fact: the probability of losing forever is zero. On a fair coin, the probability of n consecutive tails is (1/2)n, which approaches zero as n grows. So with probability 1, you will eventually win.

But "eventually" hides a multitude of sins. It's like saying you'll eventually find a parking spot in Manhattan — true in theory, but you might run out of gas first.

In practice, two iron walls block the martingale's path:

You don't have infinite money. If you start with $1,000 and bet $10, you can survive at most 6 consecutive losses ($10 + $20 + $40 + $80 + $160 + $320 = $630). On the 7th loss, you'd need $640 you don't have. At that point you've lost $630 — 63% of your bankroll — chasing a $10 profit.

Casinos aren't stupid. Every table has a maximum bet, typically 100–500× the minimum.3 A table with a $10 minimum and $5,000 maximum kills the martingale after just 9 doublings. The casino doesn't need to ban the strategy; the table limits do it for them.

And here's the kicker: those losing streaks aren't as rare as they feel. The probability of losing 10 fair coin flips in a row is (1/2)10 ≈ 0.1%. That sounds tiny. But if you play 1,000 rounds of the martingale (each "round" being a sequence that ends in a win or ruin), you'll face a 10-loss streak with surprisingly high probability. It's the birthday paradox of gambling — rare events become likely through repetition.4

The Expected Value Mirage

Defenders of the martingale sometimes appeal to expected value. And they're right: for any finite sequence, the expected gain per round is positive (on a fair game). But expected value is doing something sneaky here. It's averaging across all possible universes, including the ones where you have infinite money. In the universe where you just lost your rent money, the fact that a parallel-universe version of you is doing great offers cold comfort.

This is a case where expected value — the workhorse of probability theory — actively misleads. The median martingale player goes broke. The mean martingale player does fine, propped up by a vanishingly unlikely universe where the streak never comes.5

The Core Insight

The martingale's expected value is a lie told by averaging across impossible scenarios. In any world with finite resources, the strategy has negative expected value — because when you hit the wall, you lose everything you've slowly accumulated.

⁕ ⁕ ⁕
Chapter 22

Watch It Collapse

Theory is one thing. Watching a martingale player's bankroll over time is something else entirely. The chart has a distinctive sawtooth pattern: a steady climb of small wins, each one adding a single unit of profit, interrupted by sudden, vertical drops where a losing streak wipes out dozens or hundreds of rounds of patient accumulation.

Try it yourself. Set your bankroll, your initial bet, and watch the strategy play out over hundreds of rounds. Pay attention to the shape of the curve — those long, gentle upslopes that end in cliffs.

Martingale Simulator

Watch the double-down strategy play out. Small wins, then catastrophic collapse.

Balance Ruin Starting bankroll
$0 $500 $1k Ruin Small steady wins ↗ ← Crash ← Crash

The martingale's signature sawtooth: long gentle climbs punctuated by devastating drops.

⁕ ⁕ ⁕
Chapter 22

How Long Can You Survive?

The key question for any aspiring martingale player isn't "will I win?" — it's "how long until I'm ruined?" Given a bankroll and a bet size, you can calculate exactly how many consecutive losses it takes to wipe you out, and then ask: what's the probability of that streak occurring over a given number of plays?

Bankroll Survival Calculator

How many consecutive losses until ruin — and how likely is that streak?

Consecutive Losses to Ruin
7
You'd need $1,270 on the table
Ruin Probability
63.2%
Over 500 rounds
Prob of Single Streak
0.78%
Max Bet Before Ruin
$640
Total at Risk
$1,270
Profit If You Win
$10
⁕ ⁕ ⁕
Chapter 22

The Martingale Is Everywhere

You might think the martingale died with the powdered-wig set, but it's alive and thriving. It just wears different costumes now.

Crypto and Forex "Grid Bots"

In the cryptocurrency and forex markets, automated "grid bots" and "DCA bots" implement thinly disguised martingale strategies. The pitch: when a coin drops in price, the bot buys more, lowering your average cost. When it bounces back, you profit. It's the same logic — keep doubling down, and the inevitable recovery will bail you out.6

This worked beautifully in 2021, when everything bounced back. It worked less beautifully for holders of LUNA, FTT, or any of the dozens of tokens that went to zero. A martingale strategy on an asset that can go to zero isn't just risky — it's a wealth destruction machine that accelerates your losses at the exact moment you should be cutting them.

The Startup Pivot Trap

Consider the founder who keeps "doubling down" on a failing startup. Each round of funding is bigger. Each pivot requires more resources. The logic feels identical: "We've invested so much, we can't stop now. Just one more round and we'll break through." This is the martingale combined with the sunk cost fallacy — a cocktail of cognitive biases that has burned through more venture capital than any market crash.7

The economist might call it escalation of commitment. The gambler calls it the system. They're the same thing.

Casino $10 → $20 → $40 Ruin Crypto Bot Buy more as it drops Ruin Startup Bigger rounds, pivots Ruin Same Pattern: Double down on losses Hope for recovery

The martingale strategy wearing three different costumes — same exponential trap underneath.

Nassim Taleb's Inversion

What makes the martingale especially insidious is that it looks like it's working most of the time. A martingale player will post a winning record — 90% of days profitable! — right up until the day they blow up. Nassim Nicholas Taleb calls these strategies "picking up pennies in front of a steamroller."8 The steady pennies are real. The steamroller is also real. You just haven't seen it yet.

The opposite strategy — small, frequent losses with occasional massive wins — is psychologically painful but mathematically sound. It's how insurance works. It's how venture capital works (most bets fail, but the winners pay for everything). The martingale inverts this by making you feel great daily while setting up catastrophe.

⁕ ⁕ ⁕
Chapter 22

The Deeper Lesson

The martingale trap isn't really about gambling. It's about the difference between theoretical and practical mathematics — between proofs that work on infinite paper and strategies that work with finite resources.

When someone tells you a strategy "can't lose," ask them: under what assumptions? The martingale can't lose if you have infinite money, infinite time, and no table limits. In other words, it can't lose if you're God. For the rest of us, it's a slow-motion catastrophe disguised as a sure thing.

Any strategy that requires infinite resources to work isn't really a strategy. It's a prayer dressed up in mathematics.

The 18th-century French gamblers weren't stupid. They understood the logic perfectly. What they missed — what we all miss, whenever we double down on a failing course of action — is that logic and reality are not the same thing. A proof that requires conditions the universe can't provide isn't a proof. It's a fairy tale.

And the house, as always, knows the difference.

Notes & References

  1. The etymology is debated. Some trace "martingale" to the inhabitants of Martigues in Provence; others to a type of horse harness that prevents a horse from raising its head. See Roger Mansuy, The Origins of the Word "Martingale", Electronic Journal for History of Probability and Statistics, 2009.
  2. On a fair game (50/50), the expected gain per martingale round is exactly 0. On a game with positive expected value per bet (e.g., a biased coin), it's positive. But expected value hides the variance — see the discussion of median vs. mean outcomes below.
  3. Modern Las Vegas roulette tables typically have minimum bets of $10–$25 and maximums of $500–$10,000 on outside bets. See Steve Bourie, American Casino Guide, 2023 edition.
  4. More precisely: the probability of a run of k consecutive losses in N independent trials with loss probability q is approximately 1 − exp(−N·qk/(1−q)) for large N. See Mark F. Schilling, "The Longest Run of Heads," The College Mathematics Journal, 1990.
  5. This is closely related to the concept of ergodicity. The time-average of a martingale player's wealth (what actually happens to one person over time) diverges from the ensemble average (what happens to the average across many players). See Ole Peters, "The Ergodicity Problem in Economics," Nature Physics, 2019.
  6. Popular crypto trading platforms like 3Commas, Pionex, and Bitsgap offer "DCA bots" that implement martingale-variant strategies. During the 2022 crypto crash, users reported losses of 80–95% on these bots when tokens failed to recover.
  7. Barry Staw's classic paper "Knee-Deep in the Big Muddy: A Study of Escalating Commitment to a Chosen Course of Action," Organizational Behavior and Human Performance, 1976, describes this pattern in organizational decision-making.
  8. Nassim Nicholas Taleb, Fooled by Randomness: The Hidden Role of Chance in Life and in the Markets, Random House, 2001. The "picking up pennies" metaphor appears in Chapter 5.