Note · 31 July 2026
Nine more are about to test it, at one house, inside five weeks. I'm writing this before any of them close, so the prediction is on the record rather than assembled afterwards from whatever happened.
The strongest number on this site is that 63.5% of lots close below their low estimate. That's 219 lots across two Sotheby's sales, and I've now verified every one of them against the house's own results page. It's the closest thing I have to a law.
Except one sale ran the other way entirely. The Stephen Curry Collection, April, single-owner, sold direct from the athlete for his foundation. 61 of its 81 lots closed at or above low estimate. Median 1.50×. The whole sale landed at 1.60× its aggregate low.
Same house. Same season. Same premium. Opposite direction.
And the inversion is sharpest exactly where you'd least expect it. The cheap lots — the ones estimated at $8,000 or under — ran 2.44× low, far hotter than the sale overall. In a consigned sale the modest lots are where estimates go to die. In this one they were the engine.
I've been calling that an access premium: what people pay for the fact that it came from him rather than from a dealer. But I should be honest about the base. It is one sale. One sale is an anecdote with a spreadsheet attached.
Julien's is running nine sales through August, inside five weeks, and it publishes estimates on all of it — which almost nothing else I track does. Seven are single-owner; two are themed controls I'll come to. Heritage, by comparison, gives me 1,162 lots and cannot test estimate accuracy at all, because it publishes floors rather than ranges.
And it isn't a token amount of it. Every lot across all nine sales carries an estimate — 1,612 lots, and not one marked ‘estimate upon request.’ Before this capture, my entire estimate-bearing base was 300 lots across three sales. This is more than five times that, at one house, in five weeks — which is the only reason the test is possible at all.
So this is the first real chance to move the access finding off a sample of one. And the calendar happens to contain a distinction I couldn't have engineered.
Three living consignors. Tony Hawk, twice. Eminem. The person is alive and could, in principle, sell more tomorrow. Supply is open.
Four estates. Anna Strasberg. Allee Willis. Juice WRLD. Whitney Houston. Supply is closed permanently and everybody in the room knows it.
If the access premium is really about provenance — about the object having come from the person — both groups should run hot in roughly the same way. If it's about scarcity, the estates should run hotter, because no more is coming.
I don't know which. That's the point of writing it down now.
Julien's is also running ordinary multi-consignor themed sales in the same window — Street Culture, Ladies of the '80s. Those aren't access sales. Nobody's name is on the door.
They matter more than they look. Without them, any result I get is hopelessly confounded: if the access sales run hot, is that an access effect, or is it just that Julien's estimates conservatively? I'd have no way to tell, and I'd probably publish the flattering answer.
With them, the question separates. The themed sales measure the house. The single-owner sales measure the premium. The difference between them is the finding.
This is the correction I'd make to my own earlier work if I could. The 63.5% comes from two consigned sales at one house, and I still cannot fully separate the market underperforms estimates from Sotheby's estimates high. Here I've got the control group from the start.
The access sales beat their estimates and the themed sales don't. If the themed sales run below low the way consigned sales do, and the single-owner sales run above, the effect is real and it's about access rather than about the house.
The cheap lots outperform inside the access sales. That's the Curry pattern and it's the one I'd least expect to replicate, which makes it the most informative thing on this list. If it holds twice it stops being a curiosity.
Estates and living consignors run about the same. This is a genuine guess. My reasoning is that the buyer is paying for the connection to the person, and death doesn't change where the object came from. But I can construct the opposite argument in one line — closed supply is worth paying for — and I'd not be surprised to be wrong.
One more, which I'd rather flag now than explain later: the Eminem sale is 117 signed sneakers, and 111 of them are ordinary shoes whose estimate is almost entirely the signature. If access premium is really about the person, that block is the purest measurement of it anywhere in my data. I've called it at $115,000–175,000 hammer against a $110,000 aggregate low. That call is already published and it doesn't count toward the record, because it's a charity sale.
If the themed control sales run above estimate too, the access finding dies on the spot and what I've actually discovered is that Julien's estimates low. That would be a useful thing to learn and an embarrassing thing to publish, and it is a live possibility.
If the nine sales scatter with no pattern, Curry was a single hot room and I'll say so. I held back a milestone finding earlier this month for exactly that reason — strong in one sale, reversed in the next — and this would be the same call.
And if the estate and living-consignor groups split hard in either direction — on four estate sales and three living — I'll report it as an observation rather than a finding. That is not many either way.
I own three lots from the Stephen Curry Collection — the sale this entire finding rests on. They mostly predate the research and I'm not an active bidder, but you should weigh that when you read a result that says access sales run hot. The full position is on the conflicts page.
I'm making no calls on any of the nine sales, apart from the Eminem lots already published. These are captures, not predictions.