3
After the literature review, I proceed in chapter 3 with a brief description of how the international
art market works. I present some data by the research firm Kusin & Co. that confront total
auction sales and average prices between different Countries on a global scale. I also discuss
about regulations concerning the art trade. Of course, I put an accent on the trade of Italian art.
Many of the characteristics of the art market that I describe in chapter 3 will also be present in
my dataset, which I describe in the first part of chapter 4. Finally, in the second half of chapter 4 I
present the results of my analysis (regressions) and in chapter 5 I discuss the conclusions.
4
2. Literature review
2.1 Introduction
The aim of this chapter is to illustrate the vast literature that studied art in economical terms. Art
is considered as an alternative asset class for investments and many surveys observe the
performance of the art market. Typically, these studies build some indices to represent the trend
on the art market and its yield. However, what I want to observe in this study in particular are the
pre-sale estimates (which are published in the catalogues prior to the auction) and the existence
of any bias in those estimates. Thus, I will start my literature review highlighting those studies
that concern the estimates. There has been huge debate about art since more than twenty years
and the art market has been studied from every possible perspective. Exhibit 1 illustrates a
summary of all the studies that I present in this chapter.
In the first section I cite those researches that concentrate explicitly on the reliability of the
estimates. Many authors ask themselves whether the estimates are a good predictor of prices or
not. Methodologies and conclusions couldn’t be more diverse. Another topic related to the first
one, is the width of the estimate range, which is an indicator of the uncertainty of the auction
house in setting the estimates. The wider the range, the less precise are the estimates.
The sections that follow illustrate some other issues that have emerged in several papers and
that still relate to the way estimates are built and their trustworthiness. One paragraph discusses
whether auctioneers influence the estimates for their own interest. Different authors make
different assumptions. Some affirm that it is the responsibility of the auction houses to provide
honest estimates in order not to compromise the yield on the long term. But others argue that
auctioneers do manipulate them.
Next, I refer about the so called “law of one price”. There are some studies that look for price
differences across auction houses and geographical markets. Although these papers focus on
prices and not estimates, they are interesting because they propose a methodology that we
could use to find similar differences in the pre-sale estimates. Another subject is the declining
price anomaly, which has been first addressed by Ashenfelter (1989). It’s the observation that
prices and estimates tend to decrease during subsequent auctions of identical items. I will show
how this paper generated some discussion. One last topic of interest is the presence of
5
anchoring effects. I report some studies that observe whether the presence of a previous price
(in the case of repeated sales) does influence the sale price and/or the estimates in the second
sale.
Finally, I present the vast literature that has been published on art market indices. The vast
majority of surveys on the art market compare the performance of art to that of stocks and/or
bonds. The general conclusion is that art yields lower returns in the long run, but it’s a good way
to diversify a portfolio. Then, in the last two sections I cite two other subjects which relate to the
dynamics of auctions. The first is whether paintings become more difficult to sell following an
auction in which they didn’t reach the reserve price (whether paintings are “burned”). The
second is the habit of keeping the reserve price secret. There are papers both supporting and
discarding this practise.
2.2 Reliability of the pre-sale estimates
2.2.1 Are estimates biased?
The issue that I want to address in this paper isn’t about computing the yield on art investments.
Instead, I try to analyze how precise are the pre-sale estimates. Auction houses always publish
a catalogue before each sale with a description of each painting. These catalogues describe all
features of the paintings: the artist name, the period, the media, the size… Also, auctioneers
provide buyers with an estimate of the monetary value of the artwork. The estimate is usually
expressed as a range in which the experts forecast the final price will fall. The seller plays some
role in influencing this range because it is agreed that the low estimate should be higher than his
reserve price (the minimum price at which he is willing to sell). The reserve price is usually kept
secret. If the seller has a high reserve price, it may force the auction house to post a higher
estimate. Though this may result in a higher probability of the painting remaining unsold, this
bias exists. Some studies discuss whether it is good that the reserve price is kept secret or not. I
will illustrate some of these later on.
Many papers observed the reliability of the estimates using different datasets and covering
different periods. Ashenfelter (1989) affirms that estimates are usually truthful being highly
correlated with the prices achieved, though the estimates do not consider all the relevant
information. Abowd and Ashenfelter (1988) conduct an empirical study of how well auctioneers
predicted the prices fetched by Impressionist paintings in London and New York in auctions held
by Sotheby’s and Christie’s in the years 1980-1982. This dataset shows that pre-sale estimates
6
are usually a good predictor of hammer prices2 (these frequently fall within the estimates range).
However, a drawback of this study is that it considers sold items only, and I will show that
observations vary substantially when unsold items are included in the sample.
Differently, Beggs and Graddy (1997) find out that there are frequent over- and under-
valuations. They find out that for Contemporary, Impressionist and Modern art more recently
executed artworks are more commonly overvalued. I will refer to this study in more detail later in
this chapter. One explanation for these findings may simply be that auction houses are
unwittingly overestimating consumer demand (and hence willingness to pay) for recent
Contemporary art. Beggs and Graddy do not believe that auction houses manipulate the
estimates on purpose, but other authors make hypothesis about auction houses being
opportunistic in setting price estimates.
Some other studies focus on particular markets, as in Ekelund, Ressler and Watson (1998),
where the authors concentrate on the Latin American art market. They observe the sales of
6.378 oil-on-canvas sold at Sotheby’s and Christie’s of New York between 1977 and 1996. The
authors calculate the mean of the low and high pre-sale price estimates for each piece (they call
this number the “guess”) and the percent that that estimate varies from the actual sale price
(“bias”) for all pieces that sold. When the bias is negative, this means that the auction house
underestimated the price. Most of the times the hammer price is higher than the guess, which
means that auction houses have a tendency to underestimate the price.
McAndrew and Thompson (2003) look at French Impressionist paintings sold at auction between
1985 and 2001 by a cohort of 130 international auction houses (Sotheby’s and Christie’s
account for 85% of total sales). It’s a total of 4.174 paintings (2.925 were actually sold). For
those works that remained unsold they consider 75% of the low estimate, in order to include the
full sample in the analysis. 75% of the low estimate represents an approximation of the seller’s
reserve price. In order to see whether estimates are biased or not, the authors use a different
approach. They compute the “hammer ratio”, which they define as the ratio between the hammer
price and the arithmetic average of the high and low estimates. This ratio should be equal to one
if pre-sales estimates are unbiased. If this is true, it means that the value in the middle of the
estimate range is an exact predictor of the hammer price.
However, when considering only sold items, the distribution of the hammer ratios appears to be
skewed to the right, the mean hammer ratio being 1,14. Often, the hammer price is bigger than
the estimate, which suggests there is a tendency to underestimate hammer prices. Hence, there
is some bias in the estimates. The opposite is true when the whole sample is considered
(including unsold works). In this case the average hammer ratio becomes 0,94, which is less
2
The price at which a piece of art is “hammered down” or sold. It’s the highest bid during an auction.
7
than unity and thus indicates that over the full sample auction houses tend to overestimate
works of art. But since the ratio becomes much closer to unity, this may suggest that pre-sale
estimates are much more precise than one would suppose looking at sold items only. This is due
to the fact that unsold items make the hammer ratio decrease.
At this point, McAndrew and Thompson ask themselves whether pre-sale estimates should be
considered biased or not. They suggest that when all items put up for sale are included in the
analysis, the estimates can be considered unbiased. Hence, they suggest that auctioneers
consider not only expected hammer prices, but also reserve prices when setting their estimates.
Indeed, a drawback of the vast majority of this kind of studies is that they usually consider sold
items only. According to McAndrew and Thompson, this may influence the observations and
make it easier for estimates to appear biased. Actually, the latter is the most common
conclusion, as I illustrate in this paper. The benefit of considering all items is that of conducting a
more reliable analysis.
In another paper Bauwens and Ginsburgh (2000) observe 1.600 pieces of English silvery sold
at auction by Sotheby’s and Christie’s between 1976 and 1991. They also find some errors in
the estimates, though not enormous. They have gathered both the hammer price (thus
considering only sold items) and the estimates for all items. Then, they regress the hammer
price on the estimate and find out that both houses have a tendency to underestimate the most
expensive pieces. A possible explanation they give is that auction houses may have an interest
in making most expensive pieces more attractive through reducing the estimates. But finally
Bauwens and Ginsburgh show that the errors in the estimates are due to the fact that the
experts aren’t using all available information when making their forecasts. They use hedonic
regressions, which are one of the most common methods to build art indices. They allow you to
build an estimate of the price based on all observable characteristics of the painting.
Different considerations are proposed by Czujack and Martins (2004), who seem to agree with
Ashenfelter (1989). The authors show that given all available information, auction houses
couldn’t have proposed better estimates. The dataset used in this paper consists of 675 Picasso
paintings sold between 1975 and 1994 (499 were actually sold) at Christie’s and Sotheby’s. Two
questions are answered. The first one is how precise are the estimates and the second one is if
there are any differences between the two auction houses (if one is better than the other at
predicting prices). The advantage of this study is that it considers both sold and unsold items.
They observe that on average both Sotheby’s and Christie’s overestimated 50% of items (those
that remained unsold or were sold at below the range) and underestimated 20% (those that
were sold at above the range). No significant differences exist between the two houses.
Ricevi informazioni sui nostri servizi, sulle offerte e non perdere news e consigli su università e lavoro.
