A backtest glows on the screen with a Sharpe ratio so high a professional would assume it’s broken. A headline announces an AI that returned some eye-watering number last year. Both whisper the same seductive lie: the future is a machine, and someone found the lever. We know the pull of that lie — because we recently built one ourselves. Hover or tap any underlined term.
One of our own trading models backtested at a Sharpe ratio of 40, winning 74% of its trades. On paper it was the best thing we’d ever built — several times better than the boldest “AI beat the market” headline you’ll ever read. We deleted it.
When we broke the results out year by year, the miracle dissolved. Two years of the underlying price data were corrupted — individual bars showed the market moving a hundred times more than physically possible — and those two junk years happened to produce the strategy’s biggest “wins.” Strip them out and the edge kept shrinking, until this year it turned negative. The Sharpe-40 wasn’t a discovery. It was a mirror reflecting our own hope back at us.
That experience is the whole essay — and it was described, precisely, decades before we were born.
Ludwig von Mises taught that value is subjective: it doesn’t live inside a thing, it lives in the mind of a person choosing, at a moment, at the margin. Water is worthless by a river and priceless in a desert. There is no objective “true value” that a price is obligated to return to — which is exactly why a market can stay “wrong” far longer than your model expects.
His sharper point cuts straight through modern quant finance. Mises split uncertainty in two:
Markets are overwhelmingly case probability — and here is the trap: a backtest secretly assumes class probability. It treats the future as more draws from the same deck. But the deck gets reshuffled: regimes change, data has flaws, the world stops resembling your sample. Our Sharpe-40 wasn’t a measurement of the odds. It was a roulette wheel painted onto a first date.
Frank Knight gave this its formal, respected name in 1921. Risk is measurable and insurable (class probability). Uncertainty is not (case probability). And here is the line every trader should tattoo somewhere: profit is the reward for bearing true, unmeasurable uncertainty — not for bearing quantifiable risk, which competition prices down to nothing.
Friedrich Hayek — Mises’s student — explained why a price is so hard to beat: it is a single number silently summarizing the dispersed, un-writable-down knowledge of millions of people. No model, however elaborate, contains that. In his Nobel lecture he named the modern disease directly: “the pretense of knowledge” — dressing up guesses about human behavior in the costume of physics.
A glowing backtest is the pretense of knowledge in its purest form. So is any “AI has solved markets.” The honest move Hayek leaves you is not to forecast harder — it’s to read prices, the market’s own information system, instead of your own narrative.
Benoit Mandelbrot spent decades showing that market returns are not the gentle bell curve that most risk models and Sharpe ratios assume. Real markets have fat tails — the rare, violent move is far more common than the equations predict. A Sharpe ratio measures an average calm; markets pay their bills in sudden storms. Every metric built on the bell curve is quietly under-counting the day that ends the fund.
Nassim Taleb put a modern face on all of it. Across thousands of parallel possible histories, someone posts a spectacular year by luck alone — and that someone gets the magazine cover. This is survivorship bias and being fooled by randomness: a single dazzling year is noise, not signal, and a track record with no visible risk of ruin is hiding the ruin, not avoiding it.
Two more voices, from very different chairs, sharpen the point that price is a moving human agreement, not a fact. John Maynard Keynes — a brilliant, occasionally wiped-out investor himself — described the market as a “beauty contest”: you don’t win by picking the best company, you win by guessing what everyone else will decide is the best company. His other line is the one that outlives every model: “the market can stay irrational longer than you can stay solvent.”
George Soros — a practitioner, not a professor — called the feedback loop reflexivity: prices don’t just reflect reality, they bend it — a soaring stock lowers a company’s cost of capital, which improves the fundamentals, which lifts the stock. There is no fixed equilibrium underneath for a backtest to anchor to. The ground itself moves when you step on it.
You don’t have to be an Austrian or a contrarian to reach the same humility. Eugene Fama’s Efficient Market Hypothesis — the establishment’s cornerstone — says prices already reflect what’s knowable, so persistent edge is rare and fleeting by construction. We take that seriously as our null hypothesis: we assume we have no edge until forward reality proves otherwise. That’s not pessimism. It’s the only honest place to start.
Here is the turn, and it’s the most useful part. If markets are case-probability, fat-tailed, reflexive, and already efficient — and if track records lie — that sounds like despair. It’s the opposite. It means the edge was never a secret formula. The edge is discipline.
That discipline is exactly what let us delete our Sharpe-40 fantasy before it deleted our capital. It’s the same lens we’d turn on any “an AI has solved markets” claim, ours or anyone’s: interesting — now show me the forward record, the risk of ruin, and the independent proof. Not as an attack on anyone. As the price of not fooling ourselves.
Nobody knows the price. Anyone who tells you they’ve solved it is offering the one thing markets never award: certainty.
Dragonfly Lens is built on the opposite of the “51% AI” pitch — every claim sourced, fact kept separate from hope, and the losers reported alongside the winners.
Join the Lens →Sources & further reading: Mises on subjective value and class-vs-case probability — Human Action (1949), Mises Institute; Knight on risk vs. uncertainty — Risk, Uncertainty and Profit (1921); Hayek, “The Pretense of Knowledge” (Nobel lecture, 1974) — NobelPrize.org; Mandelbrot on fat tails — The (Mis)Behavior of Markets (2004); Taleb on randomness, survivorship, and ruin — Fooled by Randomness (2001), The Black Swan (2007); Keynes’s “beauty contest” — The General Theory (1936), ch. 12; Soros on reflexivity — The Alchemy of Finance (1987); Fama’s Efficient Market Hypothesis — “Efficient Capital Markets” (1970), Nobel 2013. The deflated-Sharpe / multiple-testing correction we use to “correct for how many you tried” follows Bailey & Lopez de Prado.
Educational research, not personalized investment advice. Dragonfly Lens is not a registered investment advisor. The Sharpe-40 example is our own internal research; specifics are simplified for illustration. Past performance does not guarantee future results.