The Fair Game That Kills You

A question worth three quarters of a million dollars

Imagine a game you cannot lose on the first move, yet are mathematically destined to lose in the end. Now imagine a second game where the odds are tilted in your favour, and you go broke anyway. The first is a famous probability puzzle. The second is how most investors quietly misunderstand compounding. The surprising part, and the reason this article exists, is that the second game is far more dangerous than the first.

The first game circulates around the trading desks of firms that pay seven figures for the right kind of mind. It is short enough to fit on a napkin.

A drunk man stands one step from a cliff edge. Every second he steps forward or back with equal odds. What is the probability that he eventually falls off?

The fast answer is fifty percent. It is also wrong, and the way it is wrong is more instructive than the question itself. The slower answer, that he falls with certainty, is right, but most people reach it through reasoning that would not survive a follow up question. The truly interesting answer sits behind both of those, and it is the reason this little puzzle has anything at all to teach a systematic trader.

I came across the drunkard’s walk in a post on X, presented as a clever interview brain teaser. It is a fine puzzle. But it stops one step short of the thing that should actually frighten a trader. If you think the drunkard’s cliff is unfair, wait until you see what happens to wealth. So let us walk to the edge and look down properly, and then keep walking, because the cliff is the gentle part.

Why the obvious series is a lucky accident

A common attempt runs like this. He falls on the first step with probability one half. He falls on the second step with probability one quarter. On the third, one eighth. Sum the geometric series, one half plus one quarter plus one eighth and onward, and it converges neatly to one. Certain ruin. Done.

The arithmetic gives the correct total. The reasoning is broken.

You cannot fall in exactly two steps. Start him one pace from the edge, with the cliff at position zero and the man at position one. On the first step, forward ends him, and backward moves him to position two. To fall on the second step he would need to reach zero from position two in a single pace, and a single pace from two lands on one or three. Neither is the edge. The probability of falling in exactly two steps is not one quarter. It is zero.

The mistake is subtle but fatal to the reasoning. The neat geometric series arrives at the right answer through faulty reasoning. The destination is correct. The path is not.

Right answer, wrong path. An interviewer pays attention to the path.

The real reason he falls

The honest reason has nothing to do with a convenient series. A symmetric random walk on a line is recurrent. This is Polya’s theorem, and it says something almost unsettling. With a fair coin and unlimited room to wander behind him, the man is guaranteed to revisit every point on the line infinitely often. Including the one that ends him.

It is gambler’s ruin against a house with infinite patience. The game is fair, the edge is absorbing, and recurrence does the rest. He does not fall because the odds are against him. 

“He falls because the odds are exactly even and the wall never moves.”

The detail that makes it ominous

Here is the part that the fifty percent answer and the lucky series both step over without noticing.

He falls with certainty.

Yet the expected time until he falls is infinite.

The walk is null recurrent. The event is certain. The waiting time has no finite average. Certainty arrives on no schedule at all. You can watch this man for a thousand years and find him still pacing, and nothing about that long survival lowers his eventual fate by a hair.

That gap, between something that almost surely happens and something that happens in finite expected time, is the first lesson worth carrying out of the puzzle. Fairness is not safety. A symmetric process is harmless on an open line and lethal the moment one outcome becomes irreversible. The symmetry is exactly what hides the threat. An absorbing barrier turns a coin flip into a sentence, and it does so quietly, with no drift, no bias, and no warning in the odds themselves.

But the warning is fragile

If the story ended there, the lesson would be cleaner than it deserves to be, because the drunkard’s doom is delicate. It needs three things to stand at once.

It needs a fair coin. Tilt the odds even slightly away from the edge and his probability of eventual ruin falls below one. The chance of falling becomes the ratio of step toward the edge to step away, raised to his starting distance, and as long as he leans away from the cliff he can pace forever and never go over.

It needs a low dimensional world. In one dimension and in two, the symmetric walk is recurrent and return is certain. In three dimensions it becomes transient. The probability that a drunken bird ever finds its way home drops to roughly thirty four percent. Add a dimension and certainty evaporates, even with a perfectly fair coin.

It needs a fixed, reachable, absorbing edge. Move the wall, or make it permeable, and the trap relaxes.

Remove any single leg and the man lives. So the drunkard is ominous, but escapable. You get out by changing the odds or by changing the geometry. This matters, because it sets up the contrast that should keep a trader awake. The additive walk is the gentle cousin. There is a worse machine, and it is the one running your account.

Now stop adding. Start multiplying.

Wealth does not move in fixed paces. It does not step one metre forward or one metre back. It scales. A good month multiplies the account by some factor above one, a bad month by some factor below it. Returns compound. And the instant you move from adding steps to multiplying them, the escape routes that saved the drunkard close behind you.

Start with the simplest version, no formulas at all. Lose fifty percent. Then gain fifty percent. You are not back where you started. You are down twenty five percent. The loss took you to half, and a fifty percent gain on half only returns you to three quarters. Multiplication is not addition. A gain and a loss of the same size do not cancel, they leave a hole. That single asymmetry is the entire engine of what follows.

Now consider a bet you would take all day. A fair coin. Heads pays you fifty percent. Tails costs you forty percent. The odds favour you, plainly. The arithmetic expectation is a gain of five percent per round. Line up ten thousand players, let them each play, average their wealth, and the crowd grows richer every single round. The ensemble prospers.

The typical individual player goes broke.

The growth rate that a single trajectory actually experiences through time is not the arithmetic mean. It is the geometric mean: the square root of one point five times zero point six, which is the square root of zero point nine, roughly zero point nine four nine. The representative path loses about five percent of itself every round and compounds steadily toward zero. The odds were in his favour the entire time. It made no difference.

“Being right on average is no defence at all.”

Why this is worse than the cliff

Look closely at how this machine differs from the first, because the differences are the whole argument.

The drunkard needs a fair coin to be doomed. The investor is destroyed by a favourable one. Positive expectation was never his protection, because positive expectation was never the disease.

The drunkard escapes by tilting the odds in his favour. The investor cannot escape that way, because his odds were already tilted in his favour and he was ruined regardless. You cannot fix a problem that was never about the sign of the expectation.

The drunkard’s edge sits a finite distance away. It is visible. He can see it and step back from it. The multiplicative barrier at zero is approached asymptotically by the typical path. There is no single step that does the damage, no dramatic moment of falling. Just a slow compounding descent that the crowd’s average will cheerfully insist is not happening, right up until the individual account is gone.

“The drunkard dies because a fair game met a wall. The investor dies because volatility eats compounded returns even when the game is rigged in his favour.Same outcome, certain ruin. Entirely unrelated machinery.”

The trajectory you actually live in

The reason this is not a parlour puzzle is that a trading account is the second machine, never the first.

An account compounds. It does not live in the crowd’s average. It lives in its own single trajectory through time. The expectation computed across a thousand hypothetical versions of you is a comfort that the one real version of you never collects. The arithmetic mean is what the ensemble earns. The geometric mean is what you earn. When volatility is large enough relative to drift, those two numbers part company, and the account follows the smaller one down.

This is the precise error that blows accounts up. It is the belief that a fair edge, or even a favourable one, makes you safe. It does not. The drunkard intuition, that a fair game must balance out in the end, is an additive intuition smuggled into a multiplicative world where it has no standing. The account does not balance out. It compounds, and compounding is unforgiving of variance.

“The account does not live in the crowd’s average. It lives in its own single trajectory through time.”

What the systematic trader is actually defending against

Seen this way, the core machinery of disciplined trend following stops looking like a collection of habits and starts looking like a single coherent answer to one problem: keep the geometric mean above one.

Equal dollar risk allocation is not tidiness. It is a refusal to let any one position carry enough variance to drag the compounded path under water. Normalising position size by volatility, sizing each market by its own range rather than by conviction, is the same instinct expressed in the language of the instrument. Treating closed balance equity as an inviolable defensive line, the cut back rule, is a direct answer to the asymmetry the drunkard taught us, that an irreversible loss is categorically different from a recoverable one. And the deep suspicion of leverage is nothing more than the knowledge that leverage widens the multiplicative spread, lifts the arithmetic mean, and quietly pushes the geometric mean below one. It is precisely the move that makes a favourable bet lethal.

The trend follower is not chasing the arithmetic average. He is protecting the time average. Everything defensive in the methodology exists because the account is a single path through time, and a single path through time is governed by the geometric mean, and the geometric mean is murdered by uncontrolled volatility regardless of how favourable the odds appear.

The two warnings, side by side

The fair walk warns you that irreversibility makes symmetry lethal. That an absorbing barrier can turn an even contest into certain ruin without any bias at all. It is the gentler of the two, because you can escape it by changing the odds or the geometry.

The multiplicative process tells you something far more uncomfortable. That being right on average is no defence. That a bet stacked in your favour can still destroy the one who takes it, round after round, with no visible moment of failure. And that there is only one way out. Keep the geometric mean above water. Control the volatility. Size the position. Refuse the leverage that drowns the time average while flattering the ensemble.

The drunkard is the warning you can see coming. The compounding machine is the one you cannot.

The cliff is not what should worry the trader. Everyone sees the cliff. Everyone respects the cliff. The real danger is the gentle slope that looks safe, the favourable bet that quietly compounds beneath its own cost of volatility, the slow descent with no obvious moment of failure.

The drunkard falls because he can see only one edge.

Most investors fail because they never realise there are two.