We opened the Magic Formula's engine. Its ranking barely works.

Every claim in this piece rests on one idea: the Magic Formula ranks companies, and the top of the ranking is supposed to be better than the bottom. That is the entire product. So we did the obvious test nobody bothers with. We sorted every stock into ten buckets by the formula's own score, from the best ranked to the worst, and asked a simple question: did the better buckets actually earn more?

Does the ranking even sort stocks?

If the ranking worked, returns would step down cleanly from the top bucket to the bottom. They do not. The top ranked bucket was middling. The eighth bucket did best. The very bottom bucket, the stocks the formula rates worst, beat several buckets above it. Across the full decade the ordering was essentially random.

Put a number on it and the rank correlation between the formula's score and the decade's returns is about −0.14, barely off zero, when a genuinely useful ranking would sit near −1. Four of the nine bucket to bucket steps went the wrong way. The sort that the whole strategy depends on is, statistically, hardly there.

Half the formula is noise

It gets more specific. The Magic Formula has two ingredients: how much a business earns on its capital, and how cheap it is. We measured each one's own correlation with the returns that followed, across 4,785 stock observations. A useful signal would score well above zero. Here is what they actually scored.

The combined score correlates with future returns at about 0.03, which is to say not at all. Worse, returns on capital, half the formula, the "quality" half everyone praises, came in negative: in this decade, in this market, higher quality on that measure went with slightly lower returns. Only cheapness carried a faint positive signal. The famous two factor engine was really running on one and a half cylinders, and one of them was firing backwards.

And the little signal there is comes and goes

Maybe the ranking works in some periods and not others. It does, but not usefully: its strength swings wildly from one starting point to the next and averages out near nothing. Below is the rank correlation for every one year window over the decade. A genuine sort shows up as a negative number, heading toward −1; near zero, or positive, means no real sort. Green marks the windows it sorted correctly, red the windows it sorted backwards.

The more you trusted the pick, the less you made

If the ranking barely works, concentrating into its top names should hurt, and it does, brutally. Hold the broad basket of about fifty names and you got the full decade return. Follow the book's thirty and you gave a little back. Trust the ranking completely and hold only the top ten, and the strategy nearly halved its return while deepening its worst fall to three quarters of your capital.

How often did it actually beat the market?

Strip away the averages and ask the plain question: pick a random starting month, hold for a while, and how often did you actually come out ahead of a plain index fund? Even the best version of the strategy is close to a coin flip, and the concentrated version is a near guaranteed way to lose to the index.

Read the bottom row and the whole illusion collapses. Holding the ten "best" stocks, the ones the formula is most confident about, beat the index in about two of every hundred five year stretches. The confidence and the performance ran in opposite directions.

And it only worked in the right weather

A ranking that genuinely finds superior businesses should earn its keep in most conditions. This one was fussy. Split the decade by macro backdrop and the strategy made most of its money when growth was expanding, inflation was easing and rates were falling; in the opposite states it roughly halved. That is the fingerprint of a market bet, not a stock picking one.

Strip the name off. What is it really?

Here is the quiet punchline. Ask which real, buyable fund this "formula" most resembles, and the answer is an ordinary small cap blend: month to month, its returns tracked a basket of small cap index funds at a correlation of about 0.90. Measured against its closest index, the Nifty 500 Equal Weight, it added nothing; its alpha was negative, roughly −1.6% a year. The elaborate two factor ranking delivered something you could have bought as a cheap small cap fund, minus a little.

So what actually worked?

Put it together and a quiet answer appears. The decade long edge was real but small, and it did not come from the formula's genius at picking winners, because the picking did not work. It came from holding a broad, diversified basket of ordinary companies through a decade that, on the whole, rewarded being invested. The formula's job was to sound clever while a wide net did the earning.

That is not an argument against the Magic Formula. It is an argument for checking. A famous name and a confident backtest told you one story; the data, opened up, told another. The only way to know which one is true for any idea you believe is to run it against real history yourself.

So we took it to Krest, and ran it through the whole test.