A valuation percentile places today’s multiple inside the distribution of every multiple the same company has traded on over some window. If a company’s P/E sits at the 12th percentile of its own ten-year history, it has been cheaper than this on only about one day in eight over that decade.
The point of the measure is that it removes the comparison nobody can make honestly. “A P/E of 14 is cheap” is a claim that depends on the industry, the era and the interest rate; “a P/E of 14 is at the 12th percentile of this company’s own decade” is a statement of fact about one business, and it is the same statement whether the business is a bank on 9 times or a software company on 40.
Low is cheap for a price multiple — and the other way round for a dividend yield, where a high reading is the cheap one, so the percentile has to be inverted before it can be read like the rest.
Suppose a company has traded between 11× and 29× earnings over the past ten years, with a median of 18×, and is on 13× today:
Note what the percentile adds over the median. The median alone says the shares are 28% below their usual multiple; the percentile says the market has rarely priced this company this low — a stronger statement, and one that is not distorted by a couple of extreme years at the top of the range.
Always read a percentile with its window attached. The same company can be at the 12th percentile of ten years and the 60th of one, and neither figure is wrong — they answer different questions. A verdict quoted without the period it was measured over is not a verdict.
Look at several multiples before concluding anything. A company cheap against its earnings history but dear against its cash flow history is telling you the two have diverged, which is usually more interesting than either reading on its own.
In the app, the Fair Value tab computes this percentile for each of its seven lenses and turns it into a verdict, always drawn beside the window it was judged over.
It is at its most useful on a company with a long, stable record, where the historical range describes a business that still exists. It is also the cleanest way to compare “cheapness” across companies that could never be compared on the raw multiple — a percentile is a percentile whether the denominator is earnings, cash flow or book value.
It assumes the past was normal. A business that has changed shape — shifted to subscriptions, sold its cyclical division, taken on far more debt — is being measured against the multiple history of a company that no longer exists, and a permanent re-rating downwards will read as an ever-deepening bargain all the way down. A short window can put a company at the 5th percentile of a range it only entered last year, and a percentile says nothing whatever about whether the denominator is about to fall, which is the usual reason a multiple looks low.