What the Private Market Does to the Public Record

The measured market is mostly not an auction

The Art Basel and UBS Global Art Market Report 2026, the tenth edition, written by Clare McAndrew of Arts Economics, estimates global art sales at $59.6bn in 2025, up 4 per cent year on year across some 41.5 million transactions. The composition matters more than the total. The dealer sector accounted for $34.8bn, up 2 per cent. Public auction sales came to $20.7bn, up 9 per cent. Private sales conducted by the auction houses came to just under $4.2bn, down about 5 per cent. The United States held 44 per cent of value at $26.0bn, the United Kingdom 18 per cent at $10.5bn, and mainland China 14 per cent at $8.5bn.

Set the auction column against the rest and the shape of the evidence base becomes clear. Something on the order of $39bn of the estimated $59.6bn passed through channels that publish nothing: no lot number, no estimate, no hammer, no premium-inclusive total, no date beyond what the parties choose to say afterwards. The $20.7bn of public auction is the only part of the market that generates a citable record as a matter of routine, and it is the smaller part.

It is also worth being precise about what the larger figure is. Dealer and private sales totals in the report are estimates built from survey work, not from a register of transactions. They are the best available measurement of a market that does not report itself, which is a different thing from a count. Any statement that begins with a share of the global art market rests on that distinction, and should say so.

The houses' own numbers say the same thing more sharply

The Art Newspaper reported the 2025 results of the two largest houses in December of that year. Sotheby's took $7bn in total, of which $5.7bn came from public auctions, up 26 per cent, with private sales at $1.2bn, or 17 per cent of the total. Christie's took $6.2bn, of which $4.7bn came from public auctions, up 8 per cent, with private sales at $1.5bn, or 24 per cent of the total. Christie's added a single sentence with more consequence than the totals: its top three sales of the year were made privately.

Consider what that means for the record of 2025. The largest public transaction of the year is documented exhaustively. Gustav Klimt's Portrait of Elisabeth Lederer sold at Sotheby's New York on 18 November 2025 in the Leonard A. Lauder evening sale at a hammer price of $205,000,000, giving $236,360,000 with buyer's premium. The sale was a white-glove result across twenty-four lots totalling $527.5m against a pre-sale estimate of $379.2m to $412.5m. Anyone can check the lot, the date, the estimate and the arithmetic of the premium.

Christie's three largest transactions of the same year have no lot number, no date, no price and no counterparties. They are larger than almost everything in the public record, and they are absent from it entirely. Any statement about 2025 as a year in the market that rests on auction results is a statement about the visible remainder after the largest deals have been removed.

The private channel moves against the public one

The relationship is not merely one of scale, it is counter-cyclical, and that is what makes it corrosive to any index. Arts Economics reported global sales of $57.5bn in 2024, a fall of 12 per cent. Within that, public auction sales fell 25 per cent while private sales conducted by the auction houses rose 14 per cent. The dealer sector fell 6 per cent. The public market did not simply shrink in 2024; a portion of it migrated.

The reasoning is not mysterious. A consignor with a work of consequence and no urgent need to sell will not accept the risk of a public failure in a weak season. A lot that is bought in is not merely unsold, it is publicly marked, and the trade remembers. A private sale carries none of that exposure: no estimate to fall short of, no room to read, no printed record if the price disappoints. So the works most likely to define the top of the market withdraw from the public channel precisely when the public channel is weakest.

The consequence for measurement is a bias that runs in the same direction as the cycle. A downturn measured on auction data alone will look worse than the market was, because the strongest supply has left the sample. A recovery measured the same way will look stronger than it was, because that supply returns. The first half of 2026 illustrates the swing: ArtTactic figures reported by The Art Newspaper put combined Christie's, Sotheby's and Phillips auction sales at $6.8bn including fees, up 70 per cent year on year, with London at $1.42bn, up 131 per cent. Some of that is genuine recovery. Some of it is supply returning to a channel it had left.

What an index can and cannot see

Ashenfelter and Graddy set out the problem in their survey of the empirical literature on art auctions, and the argument has not been answered since. All the prices in an art index are drawn from auction records, so only works that have been re-auctioned enter a repeat-sales calculation. That excludes both ends of the return distribution. Works that fall sharply in value are generally not brought back to auction, and works that go to museums never reappear at all. Whether an owner decides to sell at auction is itself conditioned on whether the work has appreciated. The sample is selected on the outcome being measured.

Two further gaps compound it. Transaction costs are excluded from index returns, and they are not small: the authors put the combined effect of buyer's and seller's premiums at as much as 25 per cent of the value of the object. And unsold lots have no price at all. Sale rates vary widely by category, and their own figures from Christie's London in 1995 and 1996 show 71 per cent of Impressionist and modern lots sold by number against 66 per cent for Victorian pictures. An index of hammer prices is an index of successful sales, and the failures leave no trace beyond a gap in the numbering.

Add to this the private market and the picture is complete. The works that never come to auction, the works that come and fail, and the works that sell privately at figures nobody publishes are all outside the series. What remains is a real and useful record of one channel, and it is routinely quoted as though it were a record of the market.

What 'sold privately' in a provenance actually tells you

Compare the information content of two provenance lines. An auction line gives a house, a city, a date, a sale title, a lot number and a result that can be checked against a published record, and it locates the work in a room on a specific afternoon in front of witnesses. A line reading that a work was sold privately, or acquired from a named dealer, gives a channel and at best a year. It cannot be checked. There is no document a third party can consult, and often no document at all beyond an invoice held by two parties who have contracted not to discuss it.

That silence is contractual, not incidental. Sotheby's private sale conditions of business for buyers extend the same five-year authenticity guarantee that applies at auction, sell the property as is, and require the buyer to pay any applicable artist's resale right royalty as part of the purchase price. They also restrict either party from disclosing the terms without written consent, subject to the usual carve-outs for advisers and legal compulsion. A private buyer therefore receives a warranty comparable to an auction buyer's and, in exchange, a duty of silence about the transaction that will one day be the provenance.

The largest reported private transactions show what survives that silence. Willem de Kooning's Interchange of 1955, sold by the artist through the Sidney Janis Gallery for $4,000, is reported to have been sold in September 2015 by the David Geffen Foundation to Kenneth C. Griffin for $300m, together with Jackson Pollock's Number 17A at a reported $200m, a package widely described at $500m. Those figures come from press reporting and encyclopaedic compilation rather than from any document either party has published, and a serious provenance should record them as reported and not as established. The primary record, if one exists, has never been produced.