How to find a trend in auction results

What each figure says

Four numbers, and each of them answers a narrower question than its label suggests.

Median, upper quarter, and the outlier

The median is the figure worth watching, for one reason: a single exceptional close moves an average and leaves a median where it was. On a set this size, one $80,000 name would drag a mean far enough to invent a rally out of nothing.

The top quarter is the second-most useful figure and the least read. It tells you where the strong end of the week actually sat, and unlike the maximum it does not move because one person got carried away.

The maximum earns its place as a pointer, not as a statistic. When it jumps, the useful response is to open that row and find out what the name was, which is usually either a genuinely excellent string or a domain carrying a website. A higher maximum is a notable result, not a trend.

Volume, and what the record leaves out

Everything here counts closes above a minimum price, across the marketplaces we read. Both halves of that shape the number.

The floor is doing more work than it looks like. The zero-bid and near-zero-bid tail is most of any marketplace's inventory, and it is worth nothing, so it is excluded. That makes every figure here a description of the contested end of the market. It also means the count is sensitive to what happens near the floor: a week where slightly more names scrape over it shows as a rise in volume without a single price having changed.

The source list matters just as much. If a marketplace we read has a quiet week, or an adapter breaks and one source stops arriving, volume falls and nothing about the market has moved at all. That is the first thing to rule out, not the last.

Comparing one period with another

The percentages compare the selected period against the period immediately before it, of equal length. Last 7 days against the 7 before. It is not year-on-year, and nothing here adjusts for seasonality, because there is nowhere near enough history to know what the seasons are.

Two consequences. A period containing a holiday is being compared against one that may not, and neither is labelled. And a period that is only offered once the record can also fill the period before it, which is why longer windows appear on their own over time rather than being available from day one. A 90-day label over nine days of collection would be a fact about us rather than about the market.

Source mix moves figures on its own

Different marketplaces carry different inventory. Registrar expiry auctions, drop-catch auctions between backorderers, and aggregated listings from other platforms are three different populations, and the mix between them is not constant week to week.

So a median can rise because one source with cheaper inventory contributed fewer rows. Nobody repriced anything. The composition changed.

The same applies to duplicates, which is why the record deduplicates before it counts. When an aggregator lists another platform's auction, the same close arrives twice at an identical price and time; counting both would report one auction as two and inflate volume in exactly the weeks the aggregators were busiest. Reading a single result covers how that works per row.

Names against former websites

Here is the trap that no amount of statistics will catch for you. A week in which several domains carrying long-running websites happened to close will show a higher median and a higher top quarter, and it will look exactly like a week in which the price of names went up.

They are not the same event, and only one of them tells you anything about the name you are holding. Before writing a conclusion about a period, open the archive on a handful of the top closes and on a couple around the median. Checking what a domain was before takes two minutes per name, and it is the only way to tell those two weeks apart.

Three readings, worked

Three shapes that come up constantly, and what each one is more likely to mean than the obvious reading.

Example

Writing a conclusion with a confidence level

If you are going to say something about the market, say how sure you are in the same sentence. Three levels are enough:

• Observed. The figure moved, and the sample did not. Say what moved and by how much.
• Suggested. The figure moved and there is a plausible market explanation, but the sample also changed. Say both.
• Not readable. The sample changed enough that the figure cannot be attributed. This is the most common answer on a short record, and saying so is worth more than a guess.

Nobody has ever regretted a market note that said "not enough data to tell". Plenty of people have regretted the other kind.

The trend-reading checklist

In order, because the early steps kill most conclusions before you invest in them:

1. Does the record cover the period, and the period before it?
2. Same minimum price on both sides of the comparison?
3. How did volume move?
4. How did the median move, and in the same direction?
5. What did the top quarter do?
6. Look at the highest closes separately, one row at a time.
7. Did the source mix change?
8. Are you comparing bids against bidders anywhere?
9. Check the archive on a sample of the top and middle rows.
10. Write the conclusion, with its confidence level attached.

Then go and look: the record is updated hourly, and the period control sits on the header.