Anchoring Effects: Evidence from Art Auctions

Suppose two Matisse paintings with identical characteristics were put up for sale. However, one of the paintings had previously been sold in a “hot” market for a high price and the other had been sold in a “cold” market for a low price. Would the painting that previously sold in the “hot” market fetch a higher price at auction? Anchoring is one of the most powerful and well-established biases in laboratory experiments. A classic example is given in Amos Tversky and Daniel Kahneman (1982) when subjects are asked to estimate a number (for example the percentage of African countries in the United Nations). The experiment begins by the subjects being given a number between 1 and 100 that is determined by a spin of a wheel. The subjects then show a bias in their final estimates toward the number they were originally given. Such overweighting of often irrelevant initial cues in judgement has been documented in a wide range of laboratory experiments. Yet very little work has been done to establish the presence of the anchoring heuristic in real economic situations. We test for anchoring effects both on actual prices in art auctions, and on experts’ presale valuations. In order to separate anchoring (where an irrelevant message has an effect on the outcome) from rational learning (where past prices are not irrelevant because they may be used to estimate unobservable quality) we use the following strategy. We first develop a hedonic prediction of price for both the current sale (e.g., the 2006 sale) and the previous sale. These hedonic predictions are based on observable characteristics of each painting and an index of overall prices. We then regress actual sale price on our hedonic prediction, on the difference between the actual price in the previous sale and our hedonic prediction (this expression tests for anchoring), and on the difference between the actual price in the previous sale and our hedonic prediction of price in the previous sale (this expression controls for unobservable characteristics). The test for anchoring is the extent to which the actual price in the previous sale affects the current sale price. We can identify anchoring from other effects because the demand for art, which is captured by the average overall price index, changes over time, whereas the unobservable component of quality is assumed to remain constant between auctions. This allows us to control for unobserved quality characteristics. As long as something drastic has not happened between sales—such as a painting has been deemed a fake, which is a very rare occurrence—the assumption of constant quality is a realistic one.