“Two portfolios can post the same number. Only one is standing here.”
Picture two bridges across the same canyon.
The first hangs from a single cable, engineered to perfection and rated to carry ten thousand tonnes. The second hangs from a thousand smaller cables, no one of them remarkable, together rated to carry the same ten thousand tonnes.
Looking only at the load rating, they are identical. If all you were handed was the engineer’s number, you could not tell them apart.
But nobody would confuse them. Sever the cable on the first bridge and it is gone. Sever one cable on the second and nine hundred and ninety-nine still hold. The rating measures the load. It says nothing about what happens when one of your assumptions fails.
The Sharpe ratio is that rating.
It measures how much return a strategy earns for each unit of volatility it puts you through. A higher number means more reward for the same turbulence, and for decades it has been the standard by which investors and allocators rank one return stream against another. Two strategies can share the same Sharpe ratio and have radically different odds of surviving the moment their assumptions fail. Like the load rating, it tells you the capacity that was reached. It tells you nothing about the structure that reached it.
Consider two portfolios. Both post a Sharpe ratio of 1.5. The first is a single concentrated position, levered, resting on one estimate. The second is a broad collection of small, loosely connected positions, no one of them load-bearing. Same number. The metric calls them equals. They are not equals. One is the single cable. The other is the thousand.
It is tempting to treat this number as fair but incomplete, accurate as far as it goes and only silent on the rest. That is too generous. The Sharpe ratio fails twice over. It is blind to how the number was built. And the number itself is shakier than it looks, because the risk it divides by is the one quantity the markets make hardest to measure. We will come to that denominator, and it is the worse of the two. For now hold the first failure: the ratio reports the altitude a strategy reached per unit of risk taken, and it cannot report how the strategy was built to reach it. Under real conditions the construction is what decides whether the number is a foundation or a single point of failure.
The Two Roads
A high Sharpe ratio can be built two ways, and they could not be more different. The ratio records only the destination. It says nothing about which road you took to arrive.
The first road starts with an edge you believe in. An asset, a signal, a strategy that looks strongest on the evidence in front of you. You estimate how strong it is, then size it to make the most of that estimate, scaling the position and adding leverage until the apparent edge has been pushed as far as the estimate permits. The high Sharpe is the result of concentrating your confidence in a single estimate and pushing it to its limit.
The second road starts with many streams, no one of them outstanding. Each carries its own modest edge, and none is asked to carry the result alone. Because they do not all move together, their individual swings partly cancel, and the combined stream is steadier than any piece of it. The high Sharpe is the product of that cancellation. No single estimate was pushed anywhere. The number emerged from the way the pieces fit together.
These are not two ways of measuring. They are two ways of placing your confidence. The first road concentrates it. Everything the portfolio earns, and everything it risks, traces back to one estimate being right. The second road distributes it. No single estimate is load-bearing, because the result comes from the relationship between many.
You have already met these two bridges. The first road builds the single cable. The second builds the thousand. Both arrive at the same number, and the metric, recording only the arrival, is blind to the route. Why it is blind is built into what the ratio is.
Why the Metric Cannot See the Road
The Sharpe ratio is a ratio. Return on top, volatility underneath. Scale the strategy up or down, double the position or halve it, and the number does not move: return and risk grow together, so their quotient holds fixed. That is what makes it useful, a small position and a large one judged on the same footing, because size has been divided out.
But the same property is a blindness. The volatility underneath the ratio is a single number, the total risk of the finished stream. It is the same number whether that risk sits in one concentrated position or is spread across a thousand that partly offset. The ratio sees the total. It cannot see how the total was assembled. By the time it is calculated, the construction has been summarised into the two numbers it works with. The road has been paved over.
The same blindness afflicts Kelly, the formal version of the first road’s move. When the first road said size the estimate to its limit, the limit has a precise name: the bet that compounds wealth fastest, if your estimate is exactly right, is the Kelly fraction, and a higher Sharpe justifies a larger one. But Kelly, like Sharpe, takes the finished stream as its input and reports a property of it. Hand it a single levered position and it returns a size; hand it a thousand offsetting streams combined into one book and it returns a size for that too. Neither number looks inside at how the stream was built.
We have seen that the metric cannot see the construction. But before we walk the first road, there is the second failure, the one promised earlier, and it is the worse of the two. The number itself, the risk it divides by, is one you cannot trust.
The Risk It Divides By
Look only at the denominator. The Sharpe ratio divides return by volatility, and volatility means one specific thing: the standard deviation of the returns. It is the yardstick the whole ratio rests on, and it deserves more suspicion than it gets.
Standard deviation is built by squaring each deviation, averaging, and taking the root. The squaring is the problem. It weights large moves far more heavily than small ones, so the number is governed by its biggest observations and barely notices the rest. Take a long run of small moves and drop a single enormous one into it: the standard deviation reports almost nothing but that outlier, while the typical day all but vanishes. The yardstick is mostly measuring its own tail.
That would be a curiosity if markets were calm. They are not. Every financial return stream anyone has looked at is fat tailed, and under fat tails the yardstick fails twice. First, it cannot be measured well: the quantity it depends on is more tail-driven than the returns themselves, so it needs far more data to settle than anyone ever has, and computed on the history in front of you it is an estimate dressed as a fact. Second, and worse, in the calm stretch before a tail event the standard deviation is low precisely because the tail has not yet arrived to be counted. It is a flood gauge that reads the river at its calmest and calls that the height of the bank. The risk has not gone anywhere. It is sitting in the one place the yardstick cannot see until it fires.
Put the denominator back into the ratio and watch what happens. A low, calm-period standard deviation sits beneath the return, and the quotient comes out high. The strategy with the most dangerous buried tail, the one whose risk is most thoroughly hidden, posts the most reassuring Sharpe of all. This is not a number that fails to warn you. It is a number that reassures you, right up until the moment it cannot. The hedge funds with the proudest Sharpe ratios going into 2008 were not the survivors; if anything the ranking ran the other way.
So the ratio is compromised on both sides of the bar. The denominator is an unreliable measure of the risk that was there, and the whole ratio is blind to the construction that produced it. Hold both as we walk the first road, because they are about to meet.
When the First Road Becomes a Cliff
So we walk the first road. We size the bet to its limit. That limit is a precise place. The Kelly fraction is the largest bet that still maximises growth: stake less and you leave growth on the table; stake more and the bet begins to destroy the very thing it was built to compound.
So the optimal bet has no margin. It cannot, because margin is the thing it gave up in order to be optimal. To bet the limit is to stand one step from the drop, by design.
None of this is a problem if the estimate is exactly right. But the estimate is never exactly right, and not because the work was sloppy. The first reason is simpler than drift: the estimate is built from a finite sample, and the quantity it most wants to know, the edge itself, the average reward, is the very hardest to pin down. The data hands you a noisy photograph even when the thing photographed holds perfectly still. The differences you are trying to measure are small, the noise around them is large, and the estimate arrives uncertain before the world has changed at all. And the world does not hold still either. Markets are non-stationary; correlations, volatilities, the very shape of returns shift from one regime to the next. So on top of a photograph already blurred, the scene itself moves. For a slow edge that drift is gentle, a settling rather than a lurch, but gentle is enough, because the bet had no margin to give.
So the bet built to the edge of one distribution, itself only ever seen through noise, is carried fully sized into a different one. The margin it never had is the margin it now needs. This is the cliff. A bet sized to the limit of an estimate is, by construction, the bet least able to survive the estimate being wrong, and the estimate is always wrong by some margin, whether because it was noisy from the start or because the regime it described has moved on.
Two examples show the mechanism rather than prove a frequency.
Long-Term Capital Management levered narrow convergence trades to a size that was optimal while the assumed relationships held. When those relationships broke, the size that had been optimal was the size that was fatal. The leverage did not cause the loss; it removed the margin that would have let the firm survive being wrong.
The volatility sellers of early 2018 sized their positions for a calm regime, and the mechanics that delivered smooth returns in that calm were the same mechanics that forced catastrophic unwinding when volatility returned. The position built for the assumed world was the one least able to survive the actual one.
Note what moved in both cases. Not a slow edge quietly decaying, but correlations and volatilities lurching: the fast-moving structure of markets, which shifts regime far more violently than any average return ever does. The slow first-moment edge may sit relatively still for decades; the dependence between assets does not, and it is the dependence that fails all at once.
And this is where the two failures meet. In each calm stretch the standard deviation beneath the Sharpe was low, so the ratio was high, so the bet looked not merely safe but excellent. The high Sharpe was not a missing warning. It was the measure confirming the trade, in the language allocators trust most, right up to the day the dependence broke and the risk it had never counted arrived all at once. Everything rested on the single cable, and the single cable is the one that, when it goes, takes the bridge with it.
Why the Second Road Holds
The second road never sizes a bet to the edge of a single estimate, so it never stands where the first road fell.
The difference looks like it is about volatility. It is really about where the burden of accuracy sits. On the first road, one estimate carried the whole result, sized as though it would hold. On the second, each stream contributes a little and the outcome comes from how they combine; if one estimate is wrong, the others are still standing. The portfolio does not need any particular stream to be right. It needs only that they are not all wrong in the same way at the same time, which is a far weaker thing to ask, and a far easier thing to get. The lower volatility everyone talks about is the visible by-product. The burden-shift underneath is the mechanism.
Notice what this does not claim. It does not promise a higher return, a smoother one, or victory in any particular year. Those are outcomes, and outcomes cannot be read off the construction in advance. What the construction decides is the failure mode. The first road fails by being wrong once, in the wrong place, at the wrong size. The second fails only when something goes wrong everywhere at once. That is what robustness is: not a guarantee of winning, but a shape that survives being wrong.
These are the thousand cables. Not stronger than the single cable. Just arranged so that no one of them carries the bridge.
The Clues, and Their Limits
So the construction matters more than the number. The temptation is to reach for a quick rule: count the positions, and a high count means safety. It does not, and it is worth seeing why, because the failure here is the one that fools careful people.
Picture a thousand ropes. Tied to a thousand separate anchors driven into solid rock, they are the second bridge, and cutting any one leaves the rest holding. But tie all thousand ropes to the same anchor and you have something that only looks like the second bridge. The ropes are real, the count is real, the spread across the cliff face is real. Underneath, there is one anchor. The day it pulls, all thousand ropes go together, and the count that looked like safety turns out to have described the surface and nothing below it.
That is the trap in the position count. A single levered position is honestly the first road; the concentration is visible, and the instinct to distrust it is right. The real danger is the book that is diversified in name and concentrated in fact. Borrow short and lend long across a dozen markets and you hold one interest-rate bet, not twelve. Sell volatility across a dozen instruments and you hold one bet on calm, not twelve. Streams that look independent in the quiet share a single anchor, and the quiet is exactly when you cannot see it. The day the common cause arrives is the day everyone learns how many anchors they were really tied to.
The count answers an easier question than the one that matters. Not how many ropes there are, but how many anchors. And that cannot be read off the finished numbers; it has to be read from the construction. What does each stream rest on? Where do those foundations overlap? What single event takes down more than one at once? The answers live in how the book was built, not in the statistics it produces.
The Second Road, Built on Purpose
The construction that answers those questions well is not an accident of good intentions. It is the second road built deliberately, and the classic trend follower has been building it for decades. The central decision is the one that looks least sophisticated: he sizes every market to the same risk. Not the markets he likes most, larger; not the signals he trusts most, heavier. Every stream, the same risk budget, at entry.
It is easy to mistake this for a refusal to think. It is not. He chooses his markets, his timeframes, his rules, and those choices are the whole craft. What he declines is the step after: ranking the markets by how strong each looks right now and tilting the book toward the leaders. That tilt is an act of selection, and selection means trusting that this market’s edge really is larger than that one’s, right now, by enough to bet the difference. That trust cannot be earned to the precision the bet requires. The differences between edges are smaller than the noise in measuring them. You would be reading a rank order off a signal too faint to order reliably, an estimate uncertain before the world has even moved, and what little it pins down does not stay put. Tilt toward the apparent leaders and you have rebuilt the single cable.
Equal risk weighting refuses to build it, and not because the maths is hard. It is the optimisation, run honestly. When you cannot reliably tell which edge is strongest, the truthful position is that your estimates of the differences are closer together than they look, and pressed to its conclusion that truthfulness returns equal weights. Equal weighting is not the absence of an answer. It is the answer the moment you take the uncertainty in your inputs seriously. This is why it so often holds up against the carefully optimised book out of sample: the optimiser bet on the precision of its estimates, and the estimates were noisier than they looked; the equal-weighted book made no such bet, so it had nothing to lose when the bet would have failed.
It would be too easy to stop there, because the second road rests on an estimate of its own. Its risk reduction comes from the offsets between streams that do not all move together, and those offsets are measured; the correlations they depend on are no better behaved than the volatilities we have already distrusted, and under stress they can climb toward one just when the cancellation is wanted. The second road’s volatility is as much an estimate as anything on the first. The difference is not that one estimates and the other does not. It is what rides on the estimate. The first road sized its whole bet to the precision of one number, so when the number was wrong the bet was fatal. The second never sized to its correlation estimate; it spread risk equally first, and treats the offsets as a benefit it collects when they hold, not a level it leans the book against. When the cancellation thins, the equal-weighted book loses a cushion. It does not lose its footing. The estimate failing costs it less because it was never load-bearing.
What this buys is not a promise to win. The equal-weighted book may trail a lucky concentrated bet for a year, or several. What it buys is the bridge that does not fall when one cable goes, because it was built from the start on the assumption that one of them would.
How the Number Was Made
We began with two bridges that carried the same load, and a number that could not tell them apart.
That was the whole problem in miniature. The Sharpe ratio reads the finished stream and reports what it carries. It cannot report what holds it up. And the weight it does report, it measures with a yardstick that goes blind in exactly the conditions that matter, reading low and calm right up to the moment the tail it never counted arrives.
So the question was never how high the number is. It was how the number was made. Built on one estimate sized to its edge, it is the single cable, and it stands until the one assumption it rests on gives way. Built on many streams that do not fail together, it is the thousand, and it holds through the failure of any one of them because it never needed that one to hold. The difference is not in the number. It is in the construction, and the construction can be read by anyone who stops asking how high and starts asking how it was built.
That is the shift The Geometry of Wealth set out to make. Not from one metric to a better one, but from the number to the thing underneath it. The investor who learns to read the structure is no longer at the mercy of a number that cannot see the cliff.
Two portfolios can show the same Sharpe ratio. One stands on a foundation of many footings. The other balances on a single point estimate. The number will never tell you which.
The construction always will.
Further reading
This essay extends an argument set out in the earlier series, The Geometry of Wealth: series synopsis.
That fat tails are universal across markets, appearing in every asset class and on every continent, is established empirically in The Fractals of Finance: the complete case.
The consequences for the Sharpe ratio, namely standard deviation as an unreliable measure of risk and the fragility of the second moment, draw on Nassim Nicholas Taleb, Statistical Consequences of Fat Tails (2025), Chapters 3 and 4.
Richard Brennan writes on systematic trading, complex adaptive markets, and the philosophical foundations of trend following at atstradingsolutions.com. His books include The Fractals of Finance, Complex Adaptive Markets, Carved by Impossibility and The Aussie Turtles Trend Following Guide.
Want to explore why structure exists at all?
Carved by Impossibility: What Remains When Everything Else Is Eliminated
The book explores the architecture of constraint, emergence, and reality itself, and what it means for how we understand markets, life, and the universe.
Available now on Amazon in paperback, hardcover, and Kindle.
Want the theoretical foundation for why trend following works?
The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets bridges complexity science with practical trading implementation. With a foreword by Jerry Parker, original Turtle Trader.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.