Value investing method
Benjamin Graham stock screener: 12 left
A Benjamin Graham stock screener run against 5,257 US listings on 2026-09-07 leaves 84 companies, 12 of which pay a dividend. The biggest cut is missing data.
By Javier SanzPublished Updated : The price-staleness sentence is removed, because it was false. It said every price here was more than seven days old and that the market data had last been refreshed on 2026-08-26; the refresh ran on 2026-09-11 and the sentence went on saying otherwise, as a computed fact typed into prose will. The scope caveat it sat in is unchanged, and the build date on the funnel is unchanged at 2026-09-07. Previously, on 2026-09-13: Graham's twenty-year dividend record moved from untestable to testable when the streak became a per-share figure on 2026-09-11. That criterion is re-measured. The earnings-stability count was published without a date and had rotted, as an undated count on a nightly figure will; it is re-measured at 5,007 and dated. The funnel above is unchanged and still dated 2026-09-07.
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)
Run a Benjamin Graham stock screener over the 5,257 listings in this product's US pool on 2026-09-07 and 84 companies survive, 12 of which also have a dividend yield on file. Four of Graham's seven defensive-investor criteria can be tested this way at all, and in sequence they cut like this.
| Step | Listings remaining |
|---|---|
| Hold a current ratio at all | 4,399 |
| Current ratio at or above 2.0 | 2,044 |
| Debt to equity at or below 1.0 | 1,514 |
| Price to earnings at or below 15 | 160 |
| Price to book at or below 1.5 | 84 |
| Positive dividend yield on file | 12 |
Every count except the first was produced by POST /api/screener/search on
2026-09-07. The public endpoint reports at most fifty matching rows, so a query that comes back under fifty is reporting a true count. The pool was split by sector and then by market capitalisation until every part came back under fifty, and the parts were added. The first row is the coverage figure for
current_ratio from GET /api/screener/fields on the same day.
The biggest cut is not a valuation cut
The step from 1,514 to 160 removes 1,354 listings and looks like a savage price test. It mostly is not.
Of the 1,514 listings clearing the two balance-sheet tests, 593 hold a price to earnings ratio at all. The other 921 have none, because there are no positive earnings to divide the price by, or because the figure is not on file. A filter reading "P/E at or below 15" silently drops every one of them. It also drops the listings whose earnings were barely positive, for the same underlying reason: a denominator near zero.
So the valuation test, applied to the population it can actually be applied to, passes 160 of 593. That is 27 percent, not 11 percent. The difference between those two numbers is the difference between "Graham's price test is brutal" and "most of the companies with a strong balance sheet are not profitable", and only the second one is what the data says.
The same effect runs through the rest of the funnel, smaller each time. 157 of the 160 hold a price to book. 58 of the 84 hold a dividend yield figure, and 12 of those have a positive one.
| Step | Remaining | Of which hold the next figure | Clearing the next test |
|---|---|---|---|
| After debt to equity | 1,514 | 593 hold a P/E | 160 |
| After P/E | 160 | 157 hold a P/B | 84 |
| After P/B | 84 | 58 hold a dividend yield | 12 |
Twelve listings out of 5,257 is 0.23 percent. Read the funnel without the middle column and the conclusion is that Graham's criteria are impossibly strict in 2026. Read it with the middle column and the conclusion is narrower and more useful: the criteria are strict, and they are also being applied to a market where a large share of listings cannot be measured by three of the four tests at all.
What the four tests are, and where they differ from Graham's
Each of these is a proxy. None of them is exactly what chapter 14 asks for, and the gaps matter more than the thresholds do.
Current ratio at or above 2.0. This one is Graham's, as written. It is the first half of his financial-condition test.
Debt to equity at or below 1.0. This one is not. Graham's financial condition test asks, for an industrial company, that long-term debt not exceed net current assets. Debt to equity at 1.0 is the modern rendering of the same idea and it is what this product's rule catalog attributes to chapter 14. It is a different test, and on a company with large fixed assets and small working capital it gives a different answer.
Price to earnings at or below 15. Graham asked for the price to be no more than fifteen times the average earnings of the past three years. The figure used here is the trailing ratio on the latest earnings. On a cyclical business at a peak that is a materially looser test, and at a trough a materially tighter one. Averaging was most of the point, and what averaging ten years instead of one does to this market is counted separately. Where fifteen falls against the market is a separate reading again: it sits above the 25th percentile of US price to earnings ratios, 13.61, and below the published median of 21.16.
Price to book at or below 1.5. Graham offered this as one of two ways to pass his price-to-assets test, the other being that the product of the price to earnings ratio and the price to book ratio should not exceed 22.5. Applying both P/E at 15 and P/B at 1.5 as hard tests is stricter than what he wrote, because it removes companies that would have passed on the combined rule.
The two criteria that cannot be tested here at all
Adequate size. Graham's own figures were in the money of the early 1970s. Translating them forward is a judgement about which index to inflate by, and this page makes no such judgement rather than pretending a particular dollar threshold is his.
Earnings stability: some earnings in each of the past ten years. On 2026-09-13 this product stores a positive earnings streak for 5,007 listings in the pool, so the figure exists. It cannot be added to the funnel above from the public interface, which accepts at most three conditions on an anonymous caller in the expression form that reaches that field. This is a limit of the public API and not of the data.
Dividend record: uninterrupted payments for at least twenty years. This one could not be answered when the funnel above was run, and it can be now. Since 11 September 2026 the dividend streak is measured per share from the record of declared dividends rather than from ten filed cash flow statements, and on 2026-09-13 182 listings in the pool carry a streak of twenty years or more.
Two cautions before that number is used as Graham's test. Our streak counts years the dividend was held or raised, which is stricter than his, so it is a floor on how many listings would pass his wording. And 75 of the 182 are themselves floors, because the record begins inside the run. What the ceiling on a streak actually depends on is the source it was read from.
Earnings growth of at least a third over ten years, on three-year averages. Outside the archive. The stored growth figures run to five years, and no per-share record stands behind earnings the way one now stands behind dividends.
Why the survivor count is the least interesting number here
Twelve is a real count and it is reproducible from the requests above. It is also almost useless on its own, because the four filters interact with the coverage of four different figures in four different ways, and the result is as much a description of what is on file as of what is cheap.
The number worth carrying away is 27 percent: among listings that clear both balance-sheet tests and have a price-to-earnings ratio, that is the share priced at or below fifteen times trailing earnings. That statement has a denominator a reader can check, and it does not quietly count absence as failure.
What this leaves out
Every count is of listings in the pool as it stood on 2026-09-07, not of deduplicated companies, and the pool contains more listings than companies.
Screener filters match stored values before the applicability gate runs, so these counts include listings for which the product would refuse to display the figure on a company page. Banks in particular carry a stored debt to equity that the product declines to show, on the grounds that deposits are not debt.
Two of the four thresholds are Graham's own, from The Intelligent Investor, revised edition 1973, chapter 14. The other two are modern renderings of what he wrote, and the differences are described above rather than smoothed over. None of the four is a boundary this product estimated.
The price to earnings and price to book tests both depend on a share price, so both describe the market as it was priced when the funnel ran on 2026-09-07, not as it is priced on the day you read this. The two balance-sheet tests depend on filings and are unaffected. The date of the market data behind a figure is printed beside that figure, and /coverage carries the current one.
Passing four filters is not a finding about any company, and no company is named here. Graham's own point in the chapter these criteria come from was that the tests were a starting list for further work, not a conclusion.