Good return on assets differs by sector

On 2026-09-07, across 4,499 US companies with filed statements on record, the median return on assets was 1.03 percent. Eleven sectors sit under that number and none of them look like it.

All figures are percent, from GET /api/public/distribution/roa on the build dated 2026-09-07. One deduplicated company per row, drawn only from companies whose filed statements we hold. Percentiles are the endpoint's own values rounded to two decimals, halves away from zero. The eleven sectors between them account for 4,498 of the 4,499 companies in the all-market cohort.

The level is the smaller half of the problem

Read the medians alone and the conclusion is that healthcare is a bad business and utilities is a decent one. Read the quartiles and a different thing is happening.

The gap between the 25th and 75th percentile is the range the middle half of a sector occupies. In utilities that range is 1.9 percentage points wide, from 1.27 to 3.16. In healthcare it is 68.7 points wide, from -70.25 to -1.53. One sector is thirty-six times more spread out than the other.

That is what makes a single cross-market threshold useless rather than merely imprecise. A rule reading "return on assets above 5 percent" puts every utility in the cohort on the same side of the line, because the whole middle half of the sector sits below 5 and the 90th percentile only reaches 4.71. The same rule inside healthcare separates the 90th percentile from the 75th, which is a real distinction between real companies. One number is doing two completely different jobs depending on which column it lands in.

Why healthcare reads the way it does

A negative median return on assets across 924 companies is not a claim that the healthcare industry destroys capital. It is a claim about what is listed.

Return on assets is net income over total assets. A clinical-stage biotech has raised money, holds it as assets, and has no product revenue, so the numerator is the year's research spending with a minus sign and the denominator is the cash raised to fund it. The ratio is arithmetically correct and describes nothing about operating performance, because there is no operation yet. The 25th percentile of -70.25 is the shape of a sector where a large minority of the listings are companies spending a balance sheet down on trials.

Utilities sit at the other end for a structural reason that is just as mechanical. A regulated utility carries an enormous rate base and earns an allowed return on it. Both terms of the ratio are large, both are stable, and the regulator is in effect setting the answer. A tight distribution is what a regulated return looks like when you measure it.

Neither of those facts is visible in the median. Both are visible in the quartiles, which is the argument for publishing quartiles.

What the numbers here are, exactly

Four things about the construction, because a percentile is only as good as the cohort behind it.

One row per company, not per listing. A company with two share classes appears once. The pool the site screens over is a list of listings and it is a different number, which is why a coverage count and a cohort size never match and should not be made to.

Only companies with filed statements on record. The cohort is built at the verified tier, so a listing whose indicator values exist without the statements behind them is out. That is a smaller and more defensible population than the whole universe.

Winsorised at the 1st and 99th percentile before the breakpoints were taken. Values are clamped to the observed edge, never dropped, so the cohort size is the cohort that produced the numbers. Across the whole market that clamp sat at -338.45 and 31.62. Every sector carries its own clamp: the consumer defensive cohort's lower edge is -1,076.49, which tells you something about its worst rows on its own.

Exact zeros excluded as coerced missing data. Seven companies were dropped market-wide on that rule. A return on assets of precisely 0.00 is almost always a null that a pipeline turned into a number somewhere upstream, and counting it as a real zero would pull the median toward a value nobody filed.

The second term, which is not published here

The usual next move is to split return on assets into net margin and asset turnover, because the same return can come from selling a little at a large markup or a lot at a small one. That decomposition is the right instinct and this page does not contain it.

Asset turnover is held for 5,196 of the pool's listings on the same day this was computed, so the data exists. What does not exist is a published cross-sectional distribution for it, and asserting sector quartiles for a figure nobody has computed the cohort for would be the kind of number this site is built to refuse. The decomposition is a page that can be written when the distribution is built, and it is not this page.

What a reader can do with this

Compare a company against its own sector column, not against the market median, and check the width of that column before deciding whether the comparison carries information. In utilities, being at the 75th percentile rather than the 25th is worth 1.9 points of return on assets and is a narrow finding. In technology, the same move is worth 20.4 points and is a large one.

None of the above says which companies those are, and none of it is a view on any security.

What this leaves out

These are cross-sectional percentiles from one build, dated 2026-09-07. They describe where a figure sat among comparable companies on that day. They are not a forecast, not a ranking, and not a view on any company.

Return on assets uses the year's net income over the period's total assets as this product computes them. Another data provider using average assets rather than period-end assets, or an adjusted earnings figure rather than net income, will produce a different number for the same company from the same filing. That is a real disagreement about definitions and not an error in either.

The cohort is US listings only, at the verified tier, one company per row. Companies whose statements we do not hold are not in it, and a sector cohort below thirty companies is refused rather than published. No sector was refused for this metric on this build.

Nothing here separates a low return on assets caused by a weak business from one caused by a large and recently raised balance sheet. Those are different facts with the same ratio, and telling them apart needs the statements, not the percentile.