Methodology
What counts as a call, how it gets graded, and what someone with no research at all would have scored on the same lots.
A price range on a specific lot, published before that lot closes, with a confidence level and a class attached. Nothing gets added, removed or adjusted after a sale. If I revise before the hammer, the original stays on the page next to the revision.
A blind call is made before any bid exists — sometimes weeks before the sale opens, off provenance and comps alone. A live-bid call is made after bidding starts, with a standing bid visible. Live-bid is easier and gets graded separately. Blending them would make the record look better than it is.
Conviction means there's a real chain of comparable sales behind it. Tape means I logged it so a surprise has a baseline, but the evidence is thin and I'm saying so. Across 60 graded lots, conviction calls beat the house 73% of the time and tape calls 55%. So the labels are carrying real weight.
Every figure is all-in — hammer plus buyer's premium — because that's what you actually pay. The houses don't agree with each other, and they don't all display the same thing, so everything gets converted before it's compared.
| House | Premium | What they show you |
|---|---|---|
| Sotheby's | 28% | Hammer. 28% to $2m, then 22%, then 15% |
| Heritage | 22% | Hammer. Guides are floors, not ranges |
| Goldin | 22% | All-in |
| Grey Flannel | 22% | Hammer |
| SCP | 20% | All-in |
Tax and shipping are left out. They change with the buyer and the destination, and including them would make two lots incomparable.
One wrinkle worth knowing: these rates move. Heritage ran 25% in 2017, 20% in 2023, 22% now. So an old comp has to be converted at the rate in force that year, not today's.
It's a hit if the final price lands inside the band. Beyond that, three things get recorded, because hit-or-miss on its own throws away the useful part.
Signed error — how far the midpoint sat from the final price, and which side. You can't fix "I was wrong." You can fix "I'm always wrong in the same direction."
Beat the house — was I closer than their estimate? Only possible where a house publishes one. Several don't.
Band width — a wide enough range catches almost anything, so precision gets tracked separately from accuracy.
A hit rate on its own means nothing. The real question is what someone with no research and no judgement would have scored on the same lots.
| Approach | Beat the house | Typical error |
|---|---|---|
| These calls | 70% | 44% |
| No-skill rule — scale the estimate by a constant | 15% | 91% |
| No-skill rule — always predict the high estimate | 12% | — |
| The auction house's own estimate | — | 73% |
One result I'd rather not print. A purely mechanical rule — take the standing bid the day before close and add 25% — beats the house 72% of the time. Marginally better than all of the research above. It uses information that doesn't exist when a blind call is made, so it isn't a fair fight. But it does suggest the standing bid a day out may be worth more than any amount of comp work.
The sample leans heavily basketball. Five houses, not fifty. Live-floor sales — where a room bids against the internet — resist the method, because the visible standing bid stops meaning much. Several categories have samples small enough that anything I say about them is directional at best. And where a house publishes no estimate at all, there's nothing to score a call against except the outcome.
I'd rather say that here than have someone else point it out.