Pre-sale forecasts
Evening sales, forecast before the hammer.
For every marquee evening sale at a house whose archive records its bought-ins, TheBasis.Art publishes the expected sell-through of the catalog before the sale — issued nightly as the catalog fills, frozen on the sale date, and scored against the recorded results afterward at the same URL. The model predicts whether a lot sells, not what it makes; guarantees and irrevocable bids are not in the record. Method →
0forecasts scoredissued before the sale
—mean errorprinted once 10 live cards are scored
0 of 0inside the 80% range
43retrospectives37 of 43 inside the range (86% vs 80% nominal) · mean error +3 pts
Upcoming issued nightly while the sale is ahead; soonest first
| Sale | Date | Lots scored | Expected sell-through | 80% range | Status |
|---|---|---|---|---|---|
| Christie's, Paris — Graziella Patiño de Ortiz Linares Collection - Evening sale | 2026-09-22 | 58 +2 unscored | 92% | 83–100% | upcoming |
| Sotheby's, Hong Kong — Modern & Contemporary Evening Auction | 2026-09-28 | 61 | 91% | 82–100% | upcoming |
| Phillips, Hong Kong — Modern & Contemporary Art Evening Sale | 2026-09-30 | 20 | 86% | 66–100% | upcoming |
Scored live forecasts held to the recorded results
No live forecast has reached its sale date yet. The first scored cards appear here once results are recorded for at least half of a card's lots.
Retrospectives out of sample for the model, but built after the sale
Retrospective — scored after the sale on the results as recorded; lots withdrawn before the sale are not in the record; the model that scored it was trained on data through 2025-09-15, before this sale. They show how the current model would have done; they are kept out of the live track record above.
How the model is validated the newest year of panel sales is held out; the table is what happened to lots it had never seen
All held-out lots n = 920,195 · AUC 0.713 · Brier 0.193
| Scored | Sold | n |
|---|---|---|
| 35% | 35% | 92,019 |
| 49% | 42% | 92,019 |
| 58% | 59% | 92,019 |
| 69% | 65% | 92,019 |
| 72% | 69% | 92,019 |
| 74% | 75% | 92,019 |
| 82% | 73% | 92,019 |
| 86% | 80% | 92,019 |
| 88% | 84% | 92,019 |
| 91% | 91% | 92,019 |
Held-out evening-sale lots n = 3,830 · AUC 0.773 · Brier 0.146
| Scored | Sold | n |
|---|---|---|
| 67% | 44% | 383 |
| 68% | 49% | 383 |
| 69% | 68% | 383 |
| 73% | 74% | 383 |
| 80% | 88% | 383 |
| 85% | 93% | 383 |
| 88% | 90% | 383 |
| 90% | 91% | 383 |
| 91% | 93% | 383 |
| 93% | 95% | 383 |
Each row is a tenth of the held-out lots by modeled score: the mean score in that tenth against the share that sold. Model trained on 4,926,256 panel lots through 2025-09-15, validated on 920,195. Sell probability →
Method and citation
- Eligibility. A sale whose title says "evening" and not "day", "afternoon" or "morning"; an art category; a house in the reporting panel this year or last; at least 20 lots with a printed estimate, summing to $5.0M of low estimates or with one lot at $1.0M; within 60 days.
- Issue and freeze. Re-issued nightly while the sale is ahead, every issue kept in the card's ledger; frozen on the sale date. A card never changes after that except for its outcome.
- Scoring. Over the lots the card scored, once half have a recorded result: actual sell-through against the forecast, in points; whether the result fell inside the 80% range; the Brier score against a flat forecast at the card's own mean.
- Cite as: TheBasis.Art, "Pre-sale forecasts", https://thebasis.art/research/forecasts. Index built 2026-09-16.
Press and data enquiries: [email protected]. Every card's payload is served by the data API.