How accurate are auction house estimates? A market-wide assessment
Auction house estimates occupy a central role in structuring expectations, anchoring pricing, and signalling value within the art market. Yet their accuracy, defined here as the ability to predict realised prices within a stated range, remains limited and highly conditional.
Across a dataset of more than 1.18 million sold lots, only around one-third of works transact within their published estimate range, with the remainder split almost evenly between above- and below-estimate outcomes. Estimates frame expectations without tightly determining realised prices; they centre expectations, but outcomes disperse systematically around them.
Figure 1. Above- and below-estimate outcomes occur in nearly equal proportions, indicating that realised prices systematically diverge around published expectations.
Source: Artscapy | Visualisation: Darden Gildea
The divergence is not random. It resolves into a clear structural asymmetry. In high-value, high-demand segments, including major auction houses, top-tier artists, and core market centres, estimates are systematically conservative, with realised prices consistently exceeding expectations. In weaker and less liquid segments, the pattern reverses: estimates become persistently over-ambitious, with a disproportionate share of works failing to meet their low estimate.
These opposing outcomes suggest that in practice, the same estimate signal implies a different risk profile depending on where the work sits in the market. In stronger segments, estimate-setting is more demand-driven, positioned conservatively to stimulate competition within deep buyer pools. In weaker segments, estimates become more supply-driven, functioning to sustain or communicate value despite fragmented demand.
Other factors, including geography, category, and sale format, introduce variation, but their effects remain secondary to demand depth, liquidity, and market concentration. Estimates are therefore best understood not as uniformly reliable indicators, and more as context-dependent pricing tools.
How auction house estimate accuracy is measured
Any assessment of estimate accuracy must begin by defining “accuracy”. In the auction context, accuracy refers to the ability of estimates to predict realised prices within a stated range, which can be evaluated across three dimensions: the proportion of lots selling within the estimate range, the relationship between realised prices and midpoint estimates, and the directional bias of outcomes, or whether realised prices systematically exceed or fall below expectations.
Accuracy must also be distinguished from precision. Accuracy concerns whether estimates are correctly aligned with realised prices, while precision concerns how tightly outcomes cluster within the stated range. The two do not necessarily coincide. An estimate can be directionally accurate but imprecise if realised prices consistently exceed or fall below it, while tightly defined ranges can still fail when they are systematically misaligned with clearing levels. Variation in estimate performance is driven primarily by directional bias, with dispersion remaining broadly stable across the market.
At an aggregate level, the market appears broadly centred but weakly predictive. The median hammer-to-midpoint ratio of 0.96, suggesting a slightly aggregate tendency toward overestimation. Approximately 31.0% of lots sell below their low estimate, 34.2% fall within the range, and 34.8% exceed the high estimate. The near symmetry between above- and below-estimate outcomes indicates that estimates are directionally balanced in aggregate, but imprecise at the lot level.
Crucially, averages are heavily distorted by outliers. The mean hammer price of $61,700 contrasts sharply with a median below $2,000, reflecting the extreme concentration of value within the market. Mean-based measures overstate the typical price level and obscure how most works transact, so median-based analysis is therefore essential to capture the central tendency of the market and to assess estimate performance in a way that reflects the experience of the majority of lots.
