Where public domain sale records come from
Every published sale record was volunteered by someone with a reason to volunteer it. Understanding that reason is most of the skill in using comps.
Public sales data is not a market feed. It is a set of disclosures, collected from parties who chose to disclose, filtered through platforms that chose what to publish. Nothing in it is audited, and the parts that are missing are missing for consistent reasons.
The sources, and what each contains
- Marketplace and auction results. Platforms publish the closing figures of sales concluded on their own systems. Coverage of those sales is close to complete, prices are as recorded rather than as claimed, and the sample is limited to what happened on that platform. Auction results skew toward expiring names and investor bidding rather than end user purchases.
- Trade press sales charts. Published lists compiled from figures supplied by brokers, marketplaces and sellers, usually with a floor below which sales are not listed. These capture larger deals across many venues, which is their value, and they exclude everything under the floor, which is most of the market.
- Escrow-derived aggregates. Summary reporting drawn from completed escrow transactions. Prices here reflect money that actually moved, but names and parties are often withheld, which limits their use as comparables for a specific term.
- Aggregator databases. Searchable archives that combine the above into one history per name. Convenient, and inheriting every bias of the feeds behind them, including duplicates when one sale is reported by two sources.
- Dispute decisions and litigation. Published decisions occasionally state what was paid or asked for a name. Rare, but reliable, since the figures were entered as evidence.
- Company disclosures. Where an acquiring company reports the purchase, the figure is verifiable. This happens almost exclusively at the top of the market.
What never appears
Private sales between two parties who never used a public venue account for a large share of end user transactions, and they are invisible unless one side announces them. Sales bound by confidentiality clauses are absent by contract, and those clauses are most common precisely where the price was unusual.
Also missing: instalment deals recorded at the first payment rather than the total, part-cash arrangements, sales where a website or a business was included with the domain, and transactions below a reporting floor. Names that failed to sell are absent entirely, which removes the denominator from any calculation of how a category performs.
The biases that follow
| Bias | Effect on a comp list |
|---|---|
| Selection by the reporter | Sellers report successes and stay quiet about weak sales, lifting the apparent average |
| Reporting floors | Small sales are excluded, so the visible market looks stronger than the whole one |
| Venue mix | Auction-heavy sources show wholesale levels, broker-heavy sources show retail levels |
| Currency and date | Figures converted at an unknown rate, or recorded on the reporting date rather than the closing date |
| Duplication | One transaction appearing twice under slightly different figures |
| Unverifiable entries | Self-reported sales that were never independently confirmed |
Using the data without being misled
Match on type before matching on price. A comp is only useful when the term type, the extension, the buyer population and the date are close to the name being valued, and the buyer population is the one most often ignored. An investor to investor sale and an end user purchase of the same name are different transactions at different levels, and mixing them produces a number that describes neither. The method is set out at greater length in comparable sales.
Read the distribution rather than the average. A handful of large outcomes drags any average upward, and the median of a well matched set is closer to what a seller should expect. Where only a few comps exist for a term type, treat them as a range with wide edges rather than as a price.
Check the age of every record. Extension level demand shifts, and a category that traded actively several years ago may have no active buyers now, which is a recurring pattern in newer extensions covered in new gTLDs. A comp from a period when an extension was being promoted heavily says little about its present market.
Finally, treat automated appraisals built on these feeds as inheriting all of the above, since a model trained on reported sales cannot see the sales that were never reported. Their limits are examined in valuation tools and limits.