How do you run a value screen, and what does the raw result get wrong?
The quant-coding domain built a screener as a tool; this lesson runs one as a factor sort and reads what comes back, because the raw result of a value screen is a lesson in what the screen cannot know.
The screen
A first value screen, on everything listed in the US: market capitalisation above $2 billion, dividend yield above 3%, earnings per share above zero, sorted largest first, the top hundred. Large enough to hold, paying a dividend, earning money.
Three of the four value ratios the sorts lesson named cannot be screened on directly: the screener behind this lesson has no price-to-book or price-to-earnings filter. What each row does show is the price and the earnings per share, so the earnings multiple is one division away — the screen narrows the universe, and the ratio is computed on the rows it returns.
What came back, measured
Run on 2026-09-04, the screen returned the hundred largest names passing the filter, and the top of the list is instructive in the wrong direction. The first row was an over-the-counter line of Tencent — a Chinese company, quoted in dollars on the US OTC market, with a reported yield of 8.5%. The second and third were two lines of the same Chinese bank, one at a price of 19.37 and the other at 0.96, the second showing a "yield" of 33%. Sixth and seventh were two preferred share lines of JPMorgan, each carrying the common stock's earnings per share of 13.48 against a preferred price near 24, which divides to a price-to-earnings of 1.8. The first ordinary large US company on the list was Chevron, fourth, at a market capitalisation of about $415 billion, an earnings multiple near 20 and a yield of 3.3%.
Nothing in that output is a data error. An OTC line's yield is computed from a foreign dividend against a thin dollar quote; a preferred share inherits its issuer's fundamentals. The screen did what it was asked. What it was asked was too little.
Cleaning the screen
Four rules turn the raw list into a value sort. Restrict the exchange to the NYSE and Nasdaq rather than every US listing, which drops the OTC lines. Drop the non-common lines — a screener row does not say what kind of security it is, so the survivors need a second look at each company's record, which says common or preferred stock; a ticker ending in -P is the tell before you look. Preferred shares and depositary receipts are not what a factor sort means by a company. Sanity-cap the ratios: a price-to-earnings under 3 or a yield over 15% on a large company is a data artefact until proven otherwise. Compute the sort variable yourself from figures that belong to the same line — price over earnings per share from the row, or price-to-book from each survivor's own record.
After cleaning, the list is a set of large, profitable, dividend-paying US companies with low earnings multiples, and that is a value sort's top group. It is also, on most days, heavy in banks, energy and telecoms — the risk story's list, and the sector concentration a factor index's constraints exist to cap.
In the data
The raw screen's first twenty rows, as it returns them today:
Read the tickers before the numbers: foreign lines quoted over the counter, and preferred shares, sit among the ordinary companies. The second look for a survivor is its own record. Verizon's reported on 2026-09-04 an earnings multiple of about 13, a yield of about 5.6% and a return on equity of about 16% — a value name that also passes a quality floor:
Try it now
Count how many of the twenty raw rows above are OTC lines or preferred shares. The cleaned screen, NYSE only in place of every US listing, is one click away in the EODHD Terminal — open it there — so count again on its first twenty rows and compare.
For the first five ordinary companies on the Terminal's list, compute price divided by earnings per share by hand and write the five multiples down beside the market's own — the S&P 500 fund traded near 25 times trailing earnings in 2026. Lower is the sort working.
The market the sort is drawn from, so the cheap fifth has a scale: