The stretches that make people quit, and the measures that mislead

The underperformance is a feature, which does not make it survivable

In the book's own test period the formula trailed the market for stretches measured in years, not months — and Greenblatt presents this as the mechanism that preserves the edge. A published strategy that worked smoothly would attract capital until it stopped working at all. One that periodically humiliates its followers sheds them instead, and the mispricing it feeds on regenerates.

This argument is elegant and it is also a genuine cost. 'The bad years are why it works' is true and consoling in year one of a bad stretch, and nearly impossible to act on in year three. No refinement of the formula shortens the stretch — anything that did would remove the moat.

When return on capital measures the accounting, not the business

Return on capital divides operating profit by recorded operating assets — and 'recorded' is the weak word. A company whose real assets are brands, software or research shows a small capital base, which makes its return on capital enormous. Some of those companies are genuinely great businesses; others are ordinary businesses whose assets the balance sheet cannot see.

• Intangible-heavy companies rank as higher quality than they are, because the capital that built the intangibles was expensed years ago and no longer appears in the denominator.
• Off-balance-sheet obligations cut the other way — a company financing its operations through structures the balance sheet does not carry shows less capital employed than it truly uses.
• One-off spikes in operating profit rank as quality for exactly one year, which is precisely the year the formula buys them.

The basket is the only defence the method offers, and it is a real one: accounting artefacts are scattered across the list rather than systematic, and thirty names absorb them. What the basket cannot absorb is an era in which the artefact becomes systematic — a market whose largest quality businesses are all intangible-heavy is a market where the ranking's idea of quality has quietly drifted.

Threshold implementations fail differently from the book

The product's single-symbol version — ROIC above 15%, EV/EBITDA below 10 — inherits the formula's spirit and a different failure profile. In an expensive market the thresholds pass nothing, so a strategy that was designed never to be out of the market becomes one that can sit in cash for years. In a cheap market they pass nearly everything, and the selectivity the ranking provided is gone.

The best-documented failure is the investor editing the list

Greenblatt later reported a natural experiment from accounts offered through his own firm: clients could either take the formula's picks automatically or approve each purchase themselves. The self-managed accounts ran the same formula and did measurably worse — because given the choice, people skipped the picks that looked most frightening, and the frightening picks were where the returns were concentrated.

A stock reaches the top of the combined list by being cheap, and stocks are rarely cheap while the news about them is good. Every name the formula hands you arrives with a plausible reason not to buy it. Vetoing those names feels like prudence and is, on the recorded evidence, the single most reliable way to underperform your own system.