Forensic scores
The median Piotroski F-Score is 4 of 9
Piotroski's F-Score across 3,659 US companies on 2026-09-07: a median of 4 of 9, a 75th percentile of 6, and 14 listings in the whole pool scoring nine.
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)
On 2026-09-07, across 3,659 US companies with filed statements on record, the median Piotroski F-Score was 4 and the 75th percentile was 6. Fourteen listings in the whole pool score 9.
Piotroski's F-Score is nine binary tests, one point each, and every explanation of it describes a scale from 0 to 9. The scale is not evenly used.
| Percentile | F-Score |
|---|---|
| 5th | 1 |
| 10th | 2 |
| 25th | 3 |
| 50th | 4 |
| 75th | 6 |
| 90th | 7 |
| 95th | 7 |
From GET /api/public/distribution/piotroski_f, build dated 2026-09-07. One
deduplicated company per row.
The 90th and 95th percentiles are the same number. On an integer scale that means the bucket at 7 spans both, so fewer than one company in ten scores 8 or 9. Counted directly, 169 listings do. The scale is nine points wide and the middle eighty percent of the market occupies four of them.
The two ends, counted exactly
| Score | Listings |
|---|---|
| 9 | 14 |
| 8 | 155 |
| 8 or 9 | 169 |
| 0 | 39 |
Counted with POST /api/screener/search on 2026-09-07. Listings in the pool of
5,257 that day holding a stored F-Score, of which there were 5,083. Fourteen
out of 5,083 is 0.28 percent, and 169 is 3.3 percent.
Read when this page is served rather than when it was written: the pool holds 5,323 (counted ) listings and 5,024 (counted ) of them carry an F-Score.
The counts at 9 and at 0 came back directly, because the public endpoint reports at most fifty matching rows and a query that comes back under fifty is reporting a true count. The count at 8 came back at the cap, so the pool was split by sector and then by market capitalisation until every part came back under fifty, and the parts were added. The two methods agree where they overlap: 155 at exactly 8 plus 14 at exactly 9 is the 169 that the same splitting returned independently for 8 or above.
A perfect nine requires all nine tests to pass in the same fiscal year: positive return on assets, positive operating cash flow, return on assets up on last year, operating cash flow above net income, leverage down, current ratio up, no shares issued, gross margin up, asset turnover up. Six of those nine compare this year against last, so a nine is a statement about one year's direction rather than about the level of anything. That is why it is rare, and it is also why it does not persist: a company that improved on six axes last year has to improve on them again to score nine twice.
Direction is also what the score contributes at the end of a screen built on levels. Asking for 7 or better after three cheapness and solvency filters is the step that separates cheap and improving from cheap and deteriorating, and it is the step that leaves almost nothing standing.
Why the low end is fuller than the explanation suggests
The 25th percentile is 3 and the 5th is 1. That is a large share of the market failing most of nine tests, and the reason is mechanical rather than editorial.
Six of the nine tests are year-over-year comparisons. A company with only one filed year on record scores all six as failures, because there is nothing to compare against. A company with no revenue cannot pass the gross margin test or the asset turnover test, so two more are unavailable to it before anything is judged. In a market where a large minority of listings are clinical-stage biotechnology companies, a floor of low scores is built into the arithmetic.
The healthcare cohort shows it: 902 companies, 25th percentile 2, median 3, against a market median of 4.
| Sector | Companies | 25th | Median | 75th | 90th |
|---|---|---|---|---|---|
| Healthcare | 902 | 2 | 3 | 5 | 6 |
| Technology | 629 | 3 | 5 | 6 | 7 |
| Industrials | 547 | 4 | 5 | 6 | 7 |
| Consumer Cyclical | 428 | 4 | 5 | 6 | 7 |
| Energy | 236 | 4 | 5 | 6 | 7 |
| Basic Materials | 224 | 3 | 5 | 6 | 7 |
| Consumer Defensive | 186 | 4 | 5 | 6 | 7 |
| Financial Services | 182 | 1 | 2 | 4 | 6 |
| Communication Services | 176 | 4 | 5 | 6 | 7 |
| Utilities | 103 | 4 | 5 | 6 | 6 |
| Real Estate | 45 | 3 | 4 | 4 | 6 |
Seven of the eleven sectors have a median of exactly 5. Strip out healthcare and financial services and the score barely discriminates between sectors at all, which is the correct behaviour for a measure designed to compare a company against its own prior year rather than against its neighbours.
What Piotroski's paper was actually about
The score was published as a way of separating, among stocks that were already cheap on book value, the ones whose fundamentals were improving from the ones whose were not. The population it was tested on was the cheapest quintile of the market by book to market, not the market.
That matters for how these percentiles should be read. A high score on an expensive company is not the finding the paper reported, and the market-wide median of 4 is a fact about the market rather than a benchmark anyone should be scored against. The number that means something is the change: this year's score against last year's, on a company that is already cheap.
Every test is also binary. A gross margin that rose by one hundredth of a point scores identically to one that doubled. Two companies both scoring 6 can have arrived there from completely different places, and the score carries none of that.
Where the score is withheld
740 companies were excluded from this cohort as not applicable: banks, capital markets firms, insurers, equity REITs and mortgage REITs. At least one of the nine signals is undefined for each of them, and a score out of six rescaled to nine would be a different statistic wearing the same name. Piotroski's own study excluded financial firms.
The 182 companies still shown under Financial Services are the ones the gate does not exclude: brokers of insurance, payment networks, ratings businesses and other financial companies that file ordinary commercial statements. Their median of 2 is the lowest of any sector, and it is worth treating as a question rather than a finding until somebody has read the nine signals on a few of them.
37 companies were dropped for holding a score of exactly zero, on the rule that an exact zero on this figure is usually a missing value that a pipeline coerced into a number. That rule is worth stating because a real zero is possible here in a way it is not for most ratios, so the exclusion is a deliberate trade of a few true zeros against a larger number of false ones.
What this leaves out
These are cross-sectional percentiles from one build, dated 2026-09-07, on US listings only, at the verified tier, one company per row. They describe where a score sat. They are not a view on any company.
The two exact counts are counts of listings holding a stored value, not of deduplicated companies, and they include listings for which the product would refuse to display the score. A screener filter matches the stored column before the applicability gate runs. The cohort behind the percentiles is built after that gate, which is why 5,083 and 3,659 are different numbers measuring different things.
Counts between the two extremes are not published here. The public screener reports at most fifty matching rows to an anonymous caller, so a count for any value in the middle of the scale cannot be reproduced by a reader from the public endpoints, and this page does not state figures a reader cannot check.
The score depends on what is on file. A company with one filed year scores the six year-over-year tests as failures, and that is a property of our records as much as of the company.