Valuation multiples
A good P/E ratio: 22 of 127 have a line
A good P/E ratio is one of 22 thresholds in a registry of 127 metrics. The other 105 carry none, and twelve of the 22 are labelled convention, not evidence.
By Javier SanzPublished
Founder of ValueScreener. Built Ninety Nine, a securities services provider, and sold it to a tier-one global bank; ran operations and internationalization at Alpaca, the brokerage API.
Who writes thisEditorial processLinkedIn (opens in a new tab)X (opens in a new tab)
Of the 127 metrics in this product's registry that have a value behind them, read 2026-09-07, 22 carry a good or bad threshold and 105 carry none. A good P/E ratio is one of the 22, and twelve of the 22 are labelled convention rather than an estimated boundary.
Ask what a good price to earnings ratio is and you will be told 15, and the person telling you will rarely say where 15 came from. It came from chapter 14 of the 1973 revision of Graham's book, where it is one of seven criteria for a defensive investor buying large industrial companies in the United States. That is a specific claim with a specific provenance, and it is worth knowing which of the numbers on a screen have one and which do not.
Ours are countable. The metric registry holds 150 entries. Twenty-three of them name ideas rather than stored figures, so 127 are metrics with a value behind them. Of those 127, 22 carry a threshold and 105 carry none.
The 22 that have a threshold
| Metric | Where the line comes from |
|---|---|
| Price to earnings | Graham, The Intelligent Investor, rev. ed. 1973, ch. 14 |
| Price to book | Graham, ch. 14. The combined test is P/E times P/B at or below 22.5 |
| Price to Graham number | Graham, ch. 14 |
| Margin of safety | Graham, ch. 20. Buy at roughly two thirds of appraised value |
| Debt to equity | Graham, ch. 14 |
| Current ratio | Graham, ch. 14. At least 2 for the defensive investor |
| Interest coverage | Damodaran's synthetic-rating table, plus Graham's five times cover |
| PEG | Lynch, One Up on Wall Street, 1989. The rule of 1.0 |
| Return on equity | Damodaran's cost-of-capital dataset, mature-market median near 8 percent |
| Return on capital employed, after tax | Same anchor |
| Return on capital employed, pre-tax | Same anchor |
| Altman Z | Altman, 1968, Journal of Finance. The paper's own cutoffs |
| Piotroski F | Piotroski, 2000, Journal of Accounting Research |
| Beneish M | Beneish, 1999, Financial Analysts Journal. The minus 1.78 line |
| Net debt to EBITDA | Credit-analyst convention, where covenants commonly sit |
| ROIC less WACC | Definitional. Zero is where growth stops adding value |
| Dividend coverage | Definitional. Below 1.0 the dividend is not funded by the year's cash |
| Free cash flow conversion | Definitional. 1.0 is where reported profit fully arrives as cash |
| Quick ratio | Definitional. 1.0 is where liquid assets match short-term liabilities |
| Return on assets | Stated as convention, with no external source claimed |
| EV/EBITDA | Stated as convention, from transaction multiples |
| Free cash flow yield | Stated as convention, set either side of the long-run real return on equities |
Fourteen of the 22 cite a named book, paper or dataset. One cites a convention among credit analysts. The remaining seven state their own reasoning in a sentence and cite nobody, which is the honest form when the line is arithmetic rather than research: the point at which a dividend stops being covered is not a finding, it is 1.0.
Three of the 22 come from peer-reviewed papers and use the papers' own published cutoffs without adjustment: Altman's discriminant boundaries from 1968, Piotroski's score zones from 2000, and Beneish's minus 1.78 from 1999. Those three are the only thresholds here that were estimated from data by somebody who published the estimation.
Twelve of the 22 say so, in the interface
Each band set carries a flag saying whether it is a convention or an estimated boundary, and the flag prints above the citation. Twelve of the 22 are flagged as convention. Ten are not.
That distinction is the useful one. A convention is a number that a lot of people use, which makes it a common language and not evidence. An estimated boundary is a number somebody fitted to data and reported. Graham's 15 is neither, exactly: it is a rule an author proposed, published and defended, which is a third thing, and the registry files it under the citation rather than under the flag.
The 105 with no threshold, and why that is the answer
The rule the registry is written to is short. Where no published or conventional threshold exists, the metric gets no bands and the caveat says the peer percentile is the only defensible reading. Inventing a threshold would be worse than admitting there is not one.
Sixteen metrics carry the same sentence, word for word, as the reason: there is no cross-industry threshold worth printing, and the percentile against companies in the same industry is the only defensible reading. Gross margin is on that list. So are operating margin, net margin, EBITDA margin, free cash flow margin, asset turnover, inventory turnover, days sales outstanding and revenue per employee.
A gross margin of 35 percent is excellent for a grocer and poor for a software company. There is no number that is good for both, so printing one would mean printing something false for one of them on every page it appeared. The remaining 89 unbanded metrics are unbanded for related reasons: growth rates, streaks, stability measures and price-derived figures have no level that is good in isolation.
What Graham's 15 looks like against the market it now describes
The threshold is defensible and dated. Both halves of that matter, and the second is measurable, so the figures in this paragraph are read from the published distribution when the page is served rather than typed when it was written. Each one prints the day it was computed beside it, and that day is neither the day this page was written nor the day you are reading it.
Of the 5,323 (counted ) listings in the US equity pool, 2,746 (counted ) hold a price to earnings ratio at all; the rest either have no positive trailing earnings to divide by or have no filed figure on record. Across the 2,625 (built ) deduplicated companies in the published cohort, the median is 19.84x (built ) and the cheapest quarter of the market ends at 12.48x (built ). Set Graham's ceiling of 15 against those two marks and you can read for yourself what his price test admits today.
A single dated reading of the same question, for comparison: on 2026-09-07, 766 of the 2,359 listings holding a P/E and sitting in a named sector carried a multiple below 15, or 32.5 percent of them. That sentence is true of that day and of no other, which is the difference between the two kinds of figure on this page. Neither is an argument that Graham was wrong. Both are the reason a band is printed with its date and its source next to it rather than as a verdict: the reader can see that the line was drawn for a different market and decide what to do about it.
How to read a band
Three rules follow from the audit, and they are all about what a band is not.
A band is not always a judgement. Four of the 22 carry bands whose every label is neutral by design: price to earnings, price to book, PEG and EV/EBITDA. The registry will not call any level of a valuation multiple good or bad, because a low multiple can mean the market expects the earnings to fall, and that is not a fact the ratio contains. The other 18 do carry judging labels, and they are all measures of profitability, solvency or accounting quality, where a direction of goodness genuinely exists.
A band with a citation is a claim you can check. The citation names an author, a year and often a chapter. Disagreeing with it is a legitimate move, and it is a move the page gives you enough information to make.
No band at all is information. Where the registry prints no threshold, it is saying that anyone printing one for that metric across industries is making it up. The percentile against the company's own industry is the answer there, and it is a different kind of answer: relative, dated and specific to a cohort.
What this leaves out
The audit above is a count of the metric registry as it stood in the repository on 2026-09-07. It is source code, not a survey: it says what this product prints, not what the field agrees on.
Counting a threshold is not the same as endorsing it. A band flagged as convention is still printed on the page and still colours a figure. The flag tells you how much weight it carries; it does not remove the weight.
The market figures in that section are read when the page is served, not taken on one day, and they are two different counts from two different passes: a cohort size and a breakpoint are computed by the nightly build, while a coverage count is taken when the endpoint answers. Each prints its own date. The dated sentence beside them is a single reading on 2026-09-07 and has not been re-run since. All of them cover listings with positive trailing earnings only: a loss-making listing has no P/E and is absent from every one of these counts.
Several of the sources are books rather than papers, and a book's rule was never estimated from a sample. Graham's criteria in particular were written about large US industrial companies buying at 1973 interest rates, and the registry cites the chapter so the reader can weigh that themselves.
Nothing here says what threshold you should use. It says which ones we print and where each came from.