How to use a stock screener

A screener applies conditions to a database of companies and returns the ones that satisfy all of them. The screener here carries about a hundred metrics grouped by where they come from: the income statement, the balance sheet, cash flow, ratios, valuation, dividends, price performance and company information such as sector and country.

Each condition is a metric, an operator and a value, with greater than, less than and between available. Two conditions on the same metric are combined into a range, so a between filter is the same thing as a pair of bounds.

The output is a table you choose the columns for, and any result can be opened as a company page or added to a watch list. The screen narrows the universe; it does not evaluate anything.

The most useful first filters are the ones that define what kind of company you are willing to look at rather than what it currently costs: sector, country, and a floor on size.

A size floor does most of the work. Without one, a screen on any ratio returns micro caps whose figures are volatile, thinly covered and frequently wrong, and those are exactly the companies most likely to sit at the extreme of any sorted column.

Once the universe is set, one or two valuation or quality conditions are usually enough. The instinct to add a tenth filter is what produces empty screens.

Filters that look independent often are not, and a screen that returns three companies is usually a screen that has asked for a contradiction rather than one that has found something rare.

The table lists the combinations that most often produce an empty or unrepresentative result, and what is actually being asked in each case.

A condition on a metric a company does not report removes it from the results. That is correct behaviour and it is invisible, because a screen shows what matched and never what was dropped.

Gross margin is the clearest case: financial companies have no cost of revenue, so any filter on it excludes the entire sector. The same applies to filters on metrics that need several years of history, which quietly remove recent listings.

The way to notice is to run the screen with the filter and without it and compare the counts, or to add sector as a column and check that the mix is what you expected.

Sort by something other than the metric you screened on. Sorting by the same criterion puts the most extreme values at the top, and extreme values in financial data are disproportionately errors: a share-class mismatch, a currency mismatch, or a figure that was not computable and was recorded as zero.

Anything at the very top or bottom of a sorted column is worth treating as a data question first. A P/E of 0.01x or a free cash flow yield in the thousands of percent is a defect rather than an opportunity.

The screen is the beginning of the work. It narrows several thousand companies to a few dozen, and the company pages are where the question of whether any of them is interesting gets answered.

Adding filters until the result set is small, which optimises for a short list rather than a good one.

Screening on a single ratio, which returns the companies whose ratio is extreme rather than the companies that are attractive.

Comparing metrics across sectors inside one screen, when the same number means different things in different industries.

Trusting the top row. The most extreme value in any column is the one most likely to be wrong.

Treating the output as a list of candidates rather than as a list of companies to research.