Why two stock screeners return different lists for the same screen

How it works

A screen looks like one question put to the whole market. It is really four decisions taken before the question is asked, and two screeners take each of them differently.

1. The universe. Which securities are rows at all.
2. The figures. Which vendor extracted which number from which filing, and when it did so.
3. The derivation. How quarterly and annual filings become the trailing, per-share and ratio fields the filters actually test.
4. The snapshot. Which price and which share count a ratio is divided by on the day you run it.

The filter states none of this. "P/E under 15" does not say whose earnings, over which twelve months, per which share, against which close. Few screeners in this catalogue even name the vendor behind their fundamentals. Koyfin says it licenses S&P Capital IQ for fundamentals, estimates and valuation. eSignal sells FactSet fundamental data as an $8 add-on. justETF names Trackinsight, etfinfo, Xignite, gettex and FactSet, and computes its screener figures as of the previous close or month end. Two screeners built on two vendors are two datasets, and each list is the product of its own set of choices.

The universe is decided before any filter runs

Start with what "every US-listed stock" means. Nasdaq publishes a daily symbol directory in two files: one for securities listed on Nasdaq, and one for everything listed elsewhere, NYSE, NYSE American, NYSE Arca, Cboe BZX and IEX among them. The files dated 25 September 2026 carry 13,288 symbols between them. 5,738 are flagged as ETFs and 37 as test issues. Of the 7,515 left, counting by security name, roughly 470 are warrants, 370 units, 460 preferred issues and 280 American depositary shares. That last count is a floor: there is no ADR flag, and TSMC's line reads only "Taiwan Semiconductor Manufacturing Company Ltd.". Among the Nasdaq listings, 347 carry a financial-status flag other than normal: 320 deficient against continued listing requirements, 18 delinquent in their filings, and 9 both.

Each of those is a decision a screener takes on your behalf. Is a warrant a stock? A SPAC unit? A preferred share, which has a price but no earnings of its own? A company that has missed a filing deadline and has stale fundamentals? The catalogue shows how far apart the answers land. Finviz Elite screens NASDAQ, NYSE and AMEX, so non-US companies appear only as US-listed ADRs. With no filter applied, it returned 11,617 tickers on 14 September 2026. Stock Rover puts Premium, its first paid tier, at 8,500+ stocks, and adds OTC screening and about 6,000 more tickers only on Ultimate. TrendSpider covers NYSE, NASDAQ, AMEX and OTC. Koyfin's screener runs over a global universe the vendor puts at 100,000+ securities. The same "profitable, market cap over $2 billion" screen can return four different lists before a single fundamental disagrees, because different securities were allowed in.

Share classes are the quiet case. Alphabet trades as GOOGL and GOOG, and Berkshire Hathaway as BRK.A and BRK.B. All four are ordinary lines in the directory. A screener that treats each listing as a row shows each company twice. One that de-duplicates shows it once, and which line it keeps decides which price every ratio uses.

Twelve months that no filing reports

Trailing twelve months, or TTM, is the most common period on a screener, and it appears in no US filing. It is computed, and the computation has two places where vendors part ways.

The fourth quarter does not exist as a filing. Form 10-Q is filed after each of the first three fiscal quarters, "but no report need be filed for the fourth quarter of any fiscal year". The 10-K reports the full year. So Q4 exists in a database only as the annual figure minus the nine-month year-to-date figure. Every TTM number that spans a fiscal year end contains a quarter the vendor produced by subtraction. If the 10-K restated or reclassified something the third-quarter 10-Q did not, the whole difference lands in that derived Q4.

Quarterly cash flow does not exist either. Under Regulation S-X Rule 10-01, a 10-Q reports its income statement for the most recent quarter and for the year to date. The statement of cash flows is required only "for the period between the end of the preceding fiscal year and the end of the most recent fiscal quarter". No second- or third-quarter cash flow is reported anywhere. Quarterly operating cash flow, capital expenditure and free cash flow are all differences of year-to-date figures, and a TTM free cash flow is a sum of four such differences. The same rule allows, but does not require, a cumulative twelve-month presentation. A few companies print the TTM figure themselves. Most leave it to whoever is computing it.

Then comes the definitional layer, where a vendor chooses and rarely says what it chose:

• Basic or diluted earnings per share.
• GAAP or adjusted. The earnings release is where non-GAAP measures live, and a vendor that reads the release rather than the financial statements can pick them up.
• Which way the per-share figure is built. TTM EPS can be the sum of four quarterly EPS figures, or TTM net income divided by a current share count. The two differ whenever the share count has moved during the year.
• What a negative denominator does. A loss-making company can be excluded, shown blank, or given a negative P/E that passes a "P/E under 15" filter.

A fiscal year is not a year

Companies choose their own fiscal calendars. NVIDIA's first quarter of fiscal 2025 ended on 28 April 2024, so most of NVIDIA's "fiscal 2025" happened in calendar 2024. "Revenue growth, last fiscal year" on a screen compares years ending in January, September and December, and treats them as one period.

The alternative is to align everything to calendar quarters, which is an approximation of a different kind. SEC's own XBRL frames API, which assembles one fact per company per calendar period, describes the trade. It takes the fact "that most closely fits the calendrical period requested", because company calendars start and end on any day and some quarters change length with the day of the week, and it warns that "data users should be mindful different reporting start and end dates for facts contained in a frame." A screener that calendarises and one that does not are both defensible. They will not agree on any company whose year does not end in December.

When the new number arrives

The deadlines set how long the market spends half-updated. A 10-Q is due 40 days after quarter end for large accelerated and accelerated filers and 45 days for everyone else. The 10-K is due 60, 75 or 90 days after the fiscal year end, by the same filer categories. Large companies report well inside those windows: Alphabet's 10-Q for the quarter ended 30 June 2026 is signed 22 July, three weeks in. The smallest filers can take the full 45 or 90 days.

So for six to thirteen weeks after every quarter end, the "latest quarter" is a different quarter for different companies in the same screen. Between two screeners it can be a different quarter for the same company, depending on when each vendor ingested the filing and when each screener rebuilt its derived fields from it. Neither screen is wrong on the day. They were computed from different filings, and a screen run in mid-March mixes fiscal years more than one run in June.

Share counts and market cap

Market cap is price times shares, and both factors are choices.

The cover page of every 10-Q states "the number of shares outstanding of each of the issuer's classes of common stock, as of the latest practicable date." Alphabet's second-quarter 2026 10-Q gives 5,868 million Class A, 835 million Class B and 5,527 million Class C shares as of 15 July 2026. Only Class A (GOOGL) and Class C (GOOG) have a line in the symbol directory, so there is no market price for Class B. The GOOGL price times the Class A count is less than half the company. The GOOGL price times all 12,230 million shares treats three classes as one security. Pricing each listed class separately still leaves the Class B shares to value somehow. Each is a defensible market cap, and a "market cap over $X" filter treats every company with more than one class this way.

The count also comes from a different date from the one behind earnings. EPS in the same 10-Q is computed on the weighted-average shares outstanding during the period. The cover count is taken on a date after the quarter closed. A vendor that divides TTM earnings by the cover count, and one that uses the reported per-share figure, disagree on every company that bought back or issued stock.

ADRs: one security, two denominators

An American depositary receipt "may represent the underlying shares on a one-for-one basis, or may represent a fraction of a share or multiple shares", in the words of the SEC's investor bulletin from August 2012. The ratio is set so that the ADR trades at a price typical of a US share. It is not visible in the ticker.

The issuer's financial statements do not follow the ratio. A foreign private issuer files an annual report on Form 20-F within four months of its fiscal year end. Beyond that report, the bulletin says, non-US companies "are generally only required to disclose what is required in their home country". The one set of financial statements the SEC requires arrives once a year, per ordinary share, in the home currency.

TSMC is the worked case. Its 20-F for fiscal 2025 states that each ADS represents five common shares. It reports basic earnings of NT$65.47 per share and NT$327.37 per equivalent ADS, with convenience translations of US$2.09 and US$10.44 at NT$31.37 to the dollar, the Federal Reserve rate for 31 December 2025. So there are three ways to get the P/E wrong. Divide the ADR's dollar price by US$2.09 and the P/E is five times too high. Divide it by NT$327.37 without converting and it is about thirty-one times too low. Convert the year's earnings at a rate from another date and you get a third number. A screen for low P/E across ADRs tests the vendor's handling of ratios and currencies at least as much as it tests the companies.

A sector is somebody's classification

Sector and industry filters look like facts and are licensed schemes, and schemes change. GICS, maintained by S&P Dow Jones Indices and MSCI, moved transaction and payment processing companies out of Information Technology and into a new Financials sub-industry after the close on 17 March 2023. On a screener using GICS, a technology screen run the day before and the day after returned different companies with nothing about them having changed. A screener that licenses GICS (Optuma bundles it into every plan) and one that uses another scheme or its own assign some companies to different sectors indefinitely. A screener that has not applied a revision disagrees with one that has.

What you can do about it

Reproduce one disputed row from the filing. Take a company that is in one list and not the other. Open its latest 10-Q or 10-K through SEC EDGAR and compute the field that decided it: TTM from the three 10-Qs and the 10-K, the per-share figure, and the share count and its date. Ten minutes on one company usually names the difference. It is rarely a bug. More often one tool measured a different period, share count or security, and you now know which.

Count the universe before you compare screens. Run each screener with no filters and note the row count, then add explicit filters for exchange, security type and country if the tool has them. If two tools start from different universes, no later filter will reconcile them.

Prefer reported fields when two tools must agree. A last-fiscal-year figure from a 10-K is one number in one filing. A TTM free cash flow is eight year-to-date figures and four subtractions. If the screen is meant to be reproducible, the fewer derivations it depends on, the fewer places two vendors can differ.

Treat ADRs and multi-class companies as a separate screen. Check one ADR's P/E against the per-ADS figure in its 20-F. Check one multi-class company's market cap against its cover page. If either is off, filter those securities out of any ratio screen on that tool, or check them by hand.

Know when in the reporting cycle you are screening. In the six to thirteen weeks after a quarter end, a fundamentals screen mixes quarters across companies. The same screen run just after the filing deadlines is comparing more like with like.

Export and diff rather than eyeball. Finviz Elite exports CSV and Excel. Stock Rover exports metered CSV and prints plan-by-plan universe sizes worth reading before comparing. Join two exports on ticker and the symmetric difference is your worklist. An export can have holes of its own: Barchart is not licensed to export fundamentals outside US equities, and those columns come out as N/L.

Do not backtest a screen on today's figures. A historical screen re-run on restated fundamentals and on today's list of survivors describes a market nobody could have screened. That needs point-in-time data and a universe that includes delisted names. The rest of the field is in charting platforms and screeners, and the deeper fundamentals tools are in fundamentals research.