The St. Petersburg Paradox: A Fair Game Worth Infinite Money Nobody Would Play

Imagine a casino offers you a coin-flip game. You pay a one-time entry fee, and then a fair coin is flipped repeatedly until it lands heads. If it lands heads on the first flip, you win $2. If it takes two flips, you win $4. Three flips, $8. Every extra tail before the first head doubles your prize. The payout formula is simple: 2^k dollars, where k is the number of flips it takes to see heads.

Now the question: how much would you pay to play this game once?

Most people, asked honestly, say something like $10, maybe $20 if they're feeling generous. Yet if you do the textbook math — multiply every possible payout by its probability and add them all up — the "fair" price of this game isn't $10, or $20, or $1,000. It's infinite. An entry fee of any finite size, no matter how large, is technically a bargain. And yet nobody, offered this game for real money, would pay more than a modest amount to play it. That gap between what the math says and what any sane person would actually do is the St. Petersburg Paradox — one of the oldest and strangest puzzles in the history of probability, and one that quietly reshaped how economics thinks about money, risk, and decision-making.

The Concept

To see why the math breaks, look at the probabilities. There's a 1-in-2 chance the coin lands heads on the first flip, paying $2. There's a 1-in-4 chance it takes two flips, paying $4. A 1-in-8 chance for three flips, paying $8. In general, the probability of needing exactly k flips is 1/2^k, and the payout for that outcome is 2^k dollars.

Here's the strange part: multiply the probability of each outcome by its payout, and the 2^k in the numerator cancels the 2^k in the denominator every single time.

(1/2 × $2) + (1/4 × $4) + (1/8 × $8) + ... = $1 + $1 + $1 + ...

Every term in this sum contributes exactly one dollar, forever. There are infinitely many possible outcomes (the coin could, in principle, come up tails a thousand times before finally landing heads), so the sum of expected contributions is an infinite string of one-dollar additions. Add infinitely many dollars together and you get an infinite expected value.

In classical probability theory, going back to Blaise Pascal and Pierre de Fermat in the 1650s, the "fair price" of a gamble was assumed to equal its expected value — the average payout you'd get if you played the game an enormous number of times. By that logic, this coin-flip game is worth an infinite entry fee. But offer it to a real human being for even $10,000 and they'll laugh you out of the room. Something in the classical theory had gone wrong, and figuring out exactly what took mathematicians the better part of two centuries.

Why It Matters

The paradox isn't just a cute puzzle — it forced the invention of one of the most important ideas in economics: the idea that money itself doesn't have constant value to a person. The 100th dollar you earn doesn't feel as good as the 1st. Your 10-millionth dollar barely registers at all. This concept, called diminishing marginal utility, underpins modern insurance markets, portfolio theory, and the entire field of behavioral economics. Every time an insurance company prices a policy, or a bank assesses how much risk a client should take on, they are — whether they know it or not — using math descended directly from this 300-year-old coin-flipping puzzle.

The paradox also exposed a subtler and more disturbing fact: an "average" outcome can be a fiction that no actual person ever experiences. To collect anywhere near the infinite expected value of the St. Petersburg game, you'd need to play it an astronomical number of times — far more than one lifetime allows. A single play of the game will, overwhelmingly likely, pay you $2 or $4 and nothing more. The huge jackpots that make the average balloon toward infinity are so rare that a real person will essentially never see one. This distinction — between the average across many parallel universes of possible outcomes and the trajectory any one person actually lives through — turns out to be central to modern risk theory, and it resurfaces in debates about everything from lottery design to how hedge funds should size their bets.

The Details

Origins. The puzzle traces to a 1713 letter from the mathematician Nicolas Bernoulli to Pierre Rémond de Montmort, describing a dice-based version of the same idea. But it was Nicolas's cousin, Daniel Bernoulli, who gave the problem its lasting fame. Daniel worked at the Imperial Academy of Sciences in St. Petersburg, Russia, and published his analysis of the game in 1738 in the Academy's journal, Commentarii Academiae Scientiarum Imperialis Petropolitanae ("Papers of the Imperial Academy of Sciences in Petersburg") — which is where the paradox gets its geographically misleading name. Neither Bernoulli lived there permanently in any lasting sense tied to the puzzle's substance; the name is simply an artifact of where the paper was published.

Bernoulli's fix. Daniel Bernoulli's key insight was that people don't value raw dollars — they value the utility, or personal satisfaction, that those dollars provide, and that utility grows much more slowly than money itself. He proposed that utility increases logarithmically with wealth: doubling your money doesn't double your happiness, it adds a constant, diminishing increment. Under this model, the huge-but-vanishingly-rare payouts from the coin game contribute almost nothing to your expected utility, even though they contribute a full dollar to your expected cash. Run the calculation with logarithmic utility instead of raw dollars, and the infinite sum collapses into a small, finite, entirely reasonable number — consistent with the fact that real people only offer a modest amount to play. A Swiss mathematician named Gabriel Cramer had actually stumbled onto a similar idea a decade earlier, in 1728, proposing that utility grows like a square root of wealth rather than a logarithm; Bernoulli credited him in his own paper.

Why utility alone doesn't fully close the case. For nearly 200 years, the logarithmic-utility explanation was treated as the final word. Then in 1934, the economist Karl Menger proved something unsettling: no matter what utility function you choose — logarithmic, square-root, or anything else that keeps growing — you can always construct a modified version of the St. Petersburg game whose payouts grow just fast enough to reproduce the paradox all over again. Utility functions could tame this specific game, but not the underlying phenomenon. The problem was more structural than anyone in the 18th or 19th century had realized.

The practical resolutions. Two more down-to-earth arguments round out the picture. First, real casinos and real players have finite bankrolls — nobody has infinite money to pay out or to wager. If you cap the maximum payout at, say, the total wealth of a country or even the entire world economy (on the order of $80–90 trillion, per recent global GDP estimates), the expected value of the game becomes finite and surprisingly small: capping payouts at world GDP works out to a fair entry price of roughly $46, not millions or billions. Second, the 18th-century naturalist Georges-Louis Leclerc, Comte de Buffon, argued — using actual coin-flip trials he ran by hand — that rational people simply discount probabilities below some threshold (he suggested roughly 1 in 10,000) as effectively impossible. Ignore those vanishingly unlikely, astronomically large payouts, and the "fair price" of the game drops to around $13.

A modern twist: time versus average. In the 21st century, physicist Ole Peters revisited the paradox using a concept from statistical mechanics called ergodicity — essentially, the question of whether an average taken across many different parallel players is the same as the average one single player experiences by playing repeatedly over time. Peters showed the two are not the same for games like this one: the "expected value" mixes together outcomes from many hypothetical parallel worlds, most of which never happen to the one real person actually playing. When you instead calculate how a single player's wealth grows, on average, over repeated real plays through time, you get a result that lines up neatly with Bernoulli's centuries-old logarithmic solution — but without needing to invoke psychological "utility" at all. It turns out the paradox was partly a mismatch between two different, legitimate-sounding notions of "average," and mistaking one for the other is what made the infinite expected value so misleading.

Where the same logic shows up today. The tension the St. Petersburg paradox surfaced — huge, rare payoffs vs. modest, likely ones — is everywhere once you know to look. Lottery jackpots are priced the same way: the "expected value" of a lottery ticket is often touted as reasonable, but that average is dominated by the astronomically improbable jackpot, not by what any individual buyer will realistically experience. Venture capital portfolios run on a similar logic in reverse — funds deliberately make many small bets, because the average return across the whole portfolio is dominated by one or two enormous, rare successes, even though most individual investments fail. And the Kelly criterion, a formula used by professional gamblers and quantitative investors to size bets, was explicitly developed to avoid the trap of maximizing expected wealth naively — it instead maximizes the long-run growth rate of wealth over repeated bets, which is much closer to Peters's time-average resolution than to Bernoulli's original expected-value framing.

Takeaways

The St. Petersburg game has a textbook expected value of infinity, yet no rational person would pay more than a small, finite amount to play it — a gap that classical 17th-century probability theory couldn't explain.
Daniel Bernoulli's 1738 solution — that money's utility grows much more slowly than money itself (he proposed logarithmically) — became the foundation of modern economic risk theory, insurance pricing, and decision-making under uncertainty.
Karl Menger later showed in 1934 that utility functions alone can't fully kill the paradox: any growing utility function admits a rigged version of the game that resurrects it.
Practical constraints — finite bankrolls, and rational people discounting vanishingly small probabilities — bring the "fair price" down to a modest, believable number (roughly $13–$46 under different reasonable assumptions).
The modern "ergodicity" resolution reveals a deeper lesson that outlives the specific coin game: an average computed across many hypothetical parallel outcomes can be wildly different from what one real person, living through one real timeline, actually experiences — a distinction that matters for lotteries, investing, and any decision involving rare, extreme payoffs.

Resources: - St. Petersburg Paradox — Stanford Encyclopedia of Philosophy - The time resolution of the St. Petersburg paradox — Ole Peters, Philosophical Transactions of the Royal Society A