Which kind of online business sells for the highest multiple?

Loading live medians from /api/stats…

A category needs at least five priced comps before we publish a median at all, so the rows at the bottom carry no multiple: domains are our largest block of sold deals and almost none of them publish an earnings figure to divide by. How many sold comps a valuation needs covers that floor. The price and earnings medians come from the comps behind each category's report band, so they are the same population the multiple is computed over, not three unrelated statistics parked in one row.

A multiple is a ratio, and a ratio can move because either end moved. Three things that have nothing to do with business quality push these numbers around, and we think they account for most of the spread between categories. We are reasoning about our own sample here rather than reporting a measurement — the table above is measured, the explanation is our reading of it.

1. The denominator is claimed, not audited. Every earnings figure we hold is what a source page stated. Where a category's listings routinely overstate what the business earns, the same sale price divided by a larger claim produces a smaller multiple. A category can therefore rank low precisely because its sellers are optimistic, which is not a fact about the businesses. What counts as earnings takes the denominator apart.
2. The categories are not the same size. Compare the price and earnings columns across rows: a category whose sales cluster in the low hundreds while its listings claim thousands in annual profit will land far below 1×, and one whose prices and claims are the same order of magnitude will land near or above it. That is a size and venue effect before it is a valuation effect — what makes a sold comp comparable is the same point applied to a single deal.
3. Different categories publish through different venues. Sold prices reach us from the places that state them, and those places are not evenly spread across asset types. A category seen mostly through one auction venue inherits that venue's economics; another seen through founder announcements inherits the reporting habits of pleased founders. Where sold-price data comes from is the full account of that skew.

“Which type of business should I buy or build?” is a question about future returns, and every column here is about past published sales of businesses that already existed. Our sample can tell you what a category's sales have cleared relative to their claimed earnings. It cannot tell you which category will still be earning in three years, how hard each one is to operate, or how many of the listings in each never sold at all — we only see the sales that happened and got published.

There is also a selection problem we cannot correct for, in either direction. Sales reach the public record because someone chose to publish them, and the choice is not random: an auction close is published by the mechanism, a private sale at a disappointing price is published by nobody. Where a category's comps lean on voluntary announcements rather than auction closes, its median is being computed over the deals someone was pleased to talk about. We can say that the bias exists and roughly where it points; we cannot say how large it is, and more rows will not fix it.