What Is a Valuation Percentile?
Most valuation tools answer the question “is this stock expensive compared to other stocks?” The valuation percentile answers a different and usually better question: “is this stock expensive compared to itself?”
The definition, in one paragraph
A valuation percentile takes a stock’s current valuation multiple (its trailing P/E, P/S, EV/Sales, or EV/EBITDA) and ranks it inside the distribution of that same multiple over the stock’s own past five years. We sample the multiple weekly, which gives roughly 260 observations per stock, and ask: what fraction of those past readings were below today’s? A percentile of 4 means the stock has traded at a richer multiple than today about 96% of the time over the past five years. A percentile of 90 means today’s multiple is near the top of its own historical range.
That’s the whole mechanism. No peer group, no sector average, no analyst model. Only the stock’s own record, and where today sits inside it.
Why self-relative beats cross-sectional
The classic screen (“show me stocks with a P/E under 15”) is a cross-sectional comparison: every company measured against every other company on one number. The problem is that a multiple compresses an enormous amount of information into a single ratio: growth rate, margin structure, capital intensity, cyclicality, balance-sheet risk, competitive position. A software company at 40× earnings and a regional bank at 9× are not mispriced relative to each other; they are different organisms. Cross-sectional screens systematically surface the companies with the worst futures, because a low multiple is often the market’s accurate verdict on quality.
Ranking a stock against its own history holds most of those differences constant. Five-years-ago-you and today-you are the same company far more than you and your sector peers are the same company. Business model, accounting conventions, competitive moat, and management usually persist across five years, so a big move in the multiple isolates a change in one thing: what the market is willing to pay for this specific stream of fundamentals. The percentile turns that into a comparable 0–100 scale that means the same thing for a hyperscaler as it does for a utility.
The two approaches answer different questions, and the self-relative one is usually the one investors actually mean. “NVDA looks cheap” almost never means “cheaper than Coca-Cola”; it means “cheaper than NVDA has been.” The percentile makes that intuition precise.
What a 4th-percentile P/E does and does not mean
What it does mean:
- On this multiple, the market is paying close to the least it has paid for this company’s trailing fundamentals at any point in five years.
- Either the price has fallen, the fundamentals have grown into the price, or both. The percentile doesn’t distinguish; the band chart does.
- Something changed. A stock doesn’t drift from its median to its 4th percentile without a reason, and finding the reason is the actual research task.
What it does not mean:
- It is not a buy signal. A low percentile is a fact about price history, not a prediction about returns. Stocks can sit at low percentiles for years, or go from the 4th percentile to a new, permanently lower range.
- It is not “undervalued.” If a company’s growth has permanently slowed, a multiple at the bottom of its old range may be exactly right, or still too high. The percentile ranks against the past; it doesn’t validate the past.
- It says nothing about the future path of fundamentals. The multiple uses trailing twelve-month numbers. If next year’s earnings will be sharply lower, a low trailing percentile can coexist with a high forward one.
Where the method breaks
Self-relative ranking is a sharp instrument with known failure modes. The four that matter most:
| Pitfall | What happens | What to check |
|---|---|---|
| Regime change | The business is materially different from its five-year-ago self: a growth phase ended, a segment was divested, margins reset. The old distribution no longer describes the current company. | Compare revenue growth and margins today vs. the start of the window. If the trajectory broke, discount the percentile. |
| Thin history | Recent IPOs and spin-offs have too few observations for a distribution to mean much. A “10th percentile” inside 18 months of data is mostly noise. | StockResearch excludes series shorter than about two years from ranking; treat 2–3 year histories with extra skepticism. |
| Negative earnings | P/E is undefined when trailing EPS is zero or negative, and near-meaningless when EPS is barely positive; a P/E of 900× at the “50th percentile” tells you nothing. | Switch multiples. P/S and EV/Sales stay defined through unprofitability; that is what they are for. |
| Distorted base years | If the five-year window contains a bubble or a crisis, the distribution is stretched. A stock that spent 2021 at 45× sales can sit at the 5th percentile while still being expensive by any longer standard. | Look at the shape of the history on the band chart, not the number alone; it tells you whether the range itself was unusual. |
The mechanics, precisely
- Sampling: weekly closes, so short-lived spikes don’t dominate and the distribution has ~260 points over five years.
- Trailing fundamentals only: all ranked multiples use trailing twelve-month figures from company filings. Forward P/E appears on StockResearch as a labeled context stat, but it is never mixed into percentile math; estimates have no comparable history.
- Per-multiple ranks: each multiple gets its own percentile. A stock can sit at the 8th percentile on P/S and the 40th on P/E at the same time. That is usually a margin story, and worth noticing when it happens.
- Alongside the percentile you’ll see the five-year median, the interquartile range, and the gap between the current multiple and the median: the same numbers the band chart draws.