Analysis of Price Determinants in the Art Market

Analysis of Price Determinants in the Art MarketBACHELOR THESIS
Author: Elena Mizerakova
Academic Year: 2015/2016
The author hereby declares that she compiled this thesis independently, using
only the listed resources and literature.
The author grants to Charles University permission to reproduce and to dis-
tribute copies of this thesis document in whole or in part.
Prague, May 13, 2016 Signature
Acknowledgments
I am grateful especially to the thesis supervisor PhDr. Boril Sopov, MSc.,
LL.M. for his time and valuable advice throughout the work. Moreover, I
would like to express gratitude to my family for their endless support during
the university studies.
Bibliographic record
Mizerakova, E., 2016. Analysis of Price Determinants in the Art Market. Bach-
elor thesis. Charles University in Prague.
Character count: 75 201
Abstract
What qualities make the best-selling artworks worth so much? Does the in-
terest of the general public influence the probability that the art will be sold
in auction? The art market research focuses on various aspects that affect the
potential of art as an investment. The boom of big data offers a unique op-
portunity to utilize its global impact and improve the present models with a
novel measure. Into the econometric analysis of auction results the thesis im-
plements a change in the Internet searching volume provided by Google Trends
as a reflection of the taste and the state of mind of society. The subject of the
detailed discussion are not only the price determinants, but also the factors
that affect the selling probability. The findings lead to a conclusion that the
proposed measure based on Google Trends is significant for determining both,
the odds of selling the artwork and its price. Beside that, an important effect
on the price and the probability have auction houses, the personal brand of the
artist or the medium of artwork.
JEL Classification D44, C25, F23, Z10, Z11
Keywords art market, auctions, Google Trends, prices,
price determinants, odds of selling
Author’s e-mail [email protected]
Supervisor’s e-mail [email protected]
Ake kvality robia umelecke diela natolko hodnotnymi? Ovplyvnuje zaujem
verejnosti pravdepodobnost, ze sa dane dielo preda na aukcii? Vyskum trhu s
umenm sa sustred na rozne aspekty, ktore mozu ovplyvnovat potencial umenia
ako investcie. Rozmach tzv. big data ponuka jedinecnu prlezitost vyuzit ich
celosvetovy vplyv na vylepsenie sucasnych modelov pomocou novej charakteris-
tiky. V ramci ekonometrickej analyzy aukcnych vysledkov praca implementuje
zmenu v objeme internetoveho vyhladavania poskytovaneho sluzbou Google
Trends ako odraz zaujmu, ci postoja spolocnosti. Predmetom detailnej diskusie
su nielen cenove determinanty, ale i faktory, ktore ovplyvnuju pravdepodob-
nost predaja daneho obrazu na aukcii. Vysledky vedu k zaveru, ze navrhovana
velicina zalozena na Google Trends je skutocne vyznamna pre urcovanie ako
ceny, tak aj pravdepodobnosti predaja. Okrem toho sa ako vyznamne vplyvy
na cenu i pravdepodobnost ukazuju aukcne siene, osobna znacka umelca ci
charakter diela.
Klcova slova trh s umenm, aukcie, Google Trends,
ceny, cenove determinanty, pravdepodob-
2.2 Big Time of Big Data . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Google Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.9 Price Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Data Chapter 12
3.1 General Characteristics . . . . . . . . . . . . . . . . . . . . . . . 12
4.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Summary of Findings . . . . . . . . . . . . . . . . . . . . . . . . 29
Contents viii
5.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6 Conclusion 41
List of Tables
4.1 Probit model estimation results . . . . . . . . . . . . . . . . . . 24
5.1 Estimation results with Heckman correction . . . . . . . . . . . 35
A.1 Distribution of the auction results according to the year of creation I
A.2 Distribution of the auction results among the artists . . . . . . . II
A.3 Distribution of the auction results for auction houses and city
categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . III
A.4 Distribution of the auction results with respect to the medium . III
A.5 Frequency of correctly predicted . . . . . . . . . . . . . . . . . . III
B.1 Results of probit estimation in Heckman correction (Equation
5.1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IV
B.2 Estimation results from the main regression (Equation 5.2) . . V
List of Figures
3.2 Illustration of potential bias . . . . . . . . . . . . . . . . . . . . 17
3.3 Comparison of average changes using 2 week data vs. using 3
week data on searching the term ‘Marcel Duchamp’ . . . . . . . 18
4.1 Receiver operating characteristics (ROC) curve for estimated pro-
bit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
OLS Ordinary Least Squares
ROC Receiver operating characteristics
USD United States dollar
Proposed topic Analysis of Price Determinants in the Art Market
Topic characteristics In 2014 a record was broken in the art market - there
were sold artworks in total for 15,2 billion dollars in auctions. Moreover, in
May 2015 the painting by Pablo Picasso, Women of Algiers Version O, at-
tained the position of the most expensive artwork ever bought on auction by
being sold for 179,4 million dollars. Art belongs to very specific types of the
investment, bringing uncertainty, low liquidity, however, often higher profit.
Thus a question arises, what determines the price of art in the art market?
The main purpose of the proposed thesis is to analyse the factors, to model the
prices and to solve the puzzle whether it is possible to predict the price of art
short time before the auction more precisely providing the information about
the interest of the society in the author.
Hypotheses
1. The state of mind in society reflected by the quantity of searching influ-
ences the price of artworks.
2. Older pieces of art are sold for higher prices than contemporary artworks.
3. There is a higher dispersion in prices in case of swift increase in popularity
of author.
Methodology The empirical part of the thesis consists of the econometric
analysis using the hedonic model in which the sales data from indices of auction
houses Christie’s and Sotheby’s are processed with application of Google Trends
results. I aim to improve the explanatory power of the hedonic models by the
inclusion of search data (Google Trends).
(b) Art as an investment and the value of art
4. Empirical part
(b) Prediction
5. Conclusion
Core bibliography
1. Towse, R. (2003): “A Handbook of Cultural Economics.” MPG Books Ltd. ISBN
1840643382.
2. Mossetto, G. & M. Vecco (2002): “Economics of Art Auctions.” Franco Angeli,
Milan. ISBN 9788846441645.
3. Heilbrun, J. & C.M. Gray (2001): “The Economics of Art and Culture.” Cambridge
University Press. ISBN 9780521183000.
4. Pownall, Rachel A.J. (2007): “Art as a Financial Investment.” Tilburg Univer-
sity. Maastricht University. Available at SSRN: http://ssrn.com/abstract=978467 or
http://dx.doi.org/10.2139/ssrn.978467.
5. Beckert, J. & J. Roessel (2004): “Kunst und Preise.” KZfSS Koelner Zeitschrift
fuer Soziologie und Sozialpsychologie. 56(1): pp. 32–50.
6. Frey, N. (2011): “Betriebwirtschaftliche Kunstbewertung.” Springer–Verlag. ISBN
9783834931092.
7. Goetzmann, W., E. Mamonova, C. Spaeniers (2014): “The Economics of Aes-
thetics And Three Centuries of Art Price Records.” Cambridge, MA: NBER Working
Paper Series.
Author Supervisor
Chapter 1
Introduction
From time to time the public is hit by a shocking news about a new record in
the highest price paid for a painting, following with a discussion whether such
price tag is reasonable. The controversies, increases in the standards of liv-
ing in the emerging countries or curiosity provoke the research focusing on the
art market.(McNulty 2014) The art as an investment opportunity, its risk and
return are frequent subjects of the studies, analysing the influence of various
aspects, from the death effects to the relationship between the financial and
art market.(Chanel 1995) Even though there exists enough literature looking at
the art market from various points of view, as Renneboog & Spaenjers (2013)
point out, a characteristic that would capture the artist’s reputation is missing.
However, the niche might be in the future gradually filled up - utilizing the new
ability to collect enormous amounts of data from all around the world, that
creates the big data.
The main aim of the thesis is to introduce an innovative measure that would
catch the current changes in tastes of potential buyers, because the possible
shifts of interest could contribute to the poorer performance of art as an in-
vestment. Moreover, such measure could be a valuable instrument in analysis
of art auctions and could bring to light some phenomena covered from the view
that occur in the market. The analysis conducted in the thesis further tries to
find a possibility of improving the hedonic model used in the research for the
study of art value in the variable that could reflect the trends that may not
be captured otherwise. Such option could potentially lead to more accurate
price setting before the auction takes place and as an indicator of a deal for the
prospective buyer. For the measure described, the thesis proposes the change
1. Introduction 2
in the volume of searching recorded in the service Google Trends. The further
focus lies on investigation of the factors influencing the price of art and giving
credit to the effects of artwork’s properties, which is not present in the litera-
ture in such detail. The fundamental research question considers the relevance
of the change in public interest for the odds of selling an artwork and also for its
price, and the analysis tests the hypothesis of the significance of this measure.
In addition to that, the another hypothesis states that the old pieces of art are
being sold for more than the contemporary art.
The thesis is structured as follows: Chapter 2 provides a motivation for big
data analysis with examples of their application. Further, it outlines the gen-
eral features of the art auctions, and gives a review of the research conducted
on the art market, focusing on the investment potential and on the relation-
ship of various art characteristics and price. Chapter 3 describes in detail the
both parts of the dataset used, Blouin Art Sales data and Google Trends. It
comments the sample and sketches the adjustments made. The probability
that the artwork will be sold is analysed in Chapter 4. The chapter describes
the methodology used, discusses the results in an elaborate way and provides
a brief summary of findings. In a similar manner is arranged Chapter 5, that
analyses the effects of various characteristics on the price of artwork. The spec-
ification of the methods used is followed by the interpretation of the results.
Chapter 6 summarizes the findings with an emphasis on the measure of public
interest and highlights the main points of the analysis.
Chapter 2
Literature Review
This chapter gives reasons for analysing big data, shows its applications on
various industries and continuously indicates a possible utilization in another
field - in the art market. Further follow the basic characteristics of art auctions,
and an overview of previous studies regarding the art market and the effects
influencing the price.
2.1 Why Could Be Big Data Useful?
At the beginning of the information evolutionary chain, forming the base for
knowledge, is data. Categorisation of the world into classes, observations, ex-
traction through computations and experiments, breakage of the presence into
elements - according to Kitchkin (2014) all the mentioned creates data, repre-
sentative in its very nature. Every kind of measurement, numbers, symbols,
images, sounds, bits, supply a small piece which perfectly fills up a bigger
picture giving us the information. Yet not only the presence of a particular
element provides some clue for understanding the world, data can be implied
even by the absence of an element or action. Existing data can be utilized
to derive new ones - as an example can serve a subtraction of values at two
different points in time or comparison of part and the whole - changes over
time and percentages, respectively.
The importance of data brought to light and currently further accentuated
lies right within their utilization - data serves as a basis for analysis and mod-
els, which allow people on the right places to design innovations and policies.
In the world with causality, cause has its effect, every action has its reaction.
2. Literature Review 4
For achieving the most efficient results in any field, from government decision-
making to consumers’ decision, knowledge of possible reaction could be key
input into policy designing process. Empirical evidence represented by data
stands as a firm background behind any model describing the behaviour or ’re-
action’, and allows to create a valid argument for further reasoning. (Kitchkin
2014)
Another definition of data provided by Floridi (2008) describes collections of
abstract facts, collections of binary elements distinct from other data, that are
transmitted and processed electronically. Even though the word ’collection’
could subconsciously evoke in the reader some kind of structure, it may not al-
ways be the case. While structured data is easily organized, stored and used in
the model, one can distinguish also semi-structured data. These are character-
ized by irregular structure, however with consistent set of fields. Furthermore,
semi-structured data provide self-defining meta data because of separating con-
tent semantically. Qualities may be often difficult to analyse - even though all
elements share specific general structure, each piece of data may have a differ-
ent form. This kind of data can be queried and searched, but its nature makes
any computational analysis hard to perform. (Kitchkin 2014)
2.2 Big Time of Big Data
Although information is collected in numerous ways since long ago, the ex-
pansion of Internet usage was just the factor that empowered not only the
increased speed of collection, but the amount as well. These changes initiate
new challenges regarding the analysis, storage, processing, and applications of
big data. Laney (2001) in his report outlines the big data by defining their 3 key
characteristics often referred to as 3V - volume, velocity, and variety. Volume
illustrates the immense quantity of information consisting of terabytes (1 TB
= 1012 bytes) or (petabytes 1 PB = 1015 bytes) of data. Information is high in
velocity, being created in or near real-time, while still being diverse in variety of
types. At the same time, the information can be referenced to both, space and
time. In comparison to old database systems, which were constrained to two of
3V, a new database design with enhanced computational power made all three
simultaneously achievable. Other important characteristics of big data include
being exhaustive in scope by capturing entire population or system, flexibil-
ity in possibility of adding new fields, and ability to expand in size rapidly.
2. Literature Review 5
(Kitchkin, 2014)
By making connections between pieces of data, and does not matter if dealt
with an individual, his relationship to others or with structure of information
itself, big data is networked. As Boyd & Crawford (2011) point out, its value
comes from patterns which can be derived and analysed for relationships by
interlinking diverse sets of information - from personal, through social to spa-
tial. Information with the qualities mentioned could be obtained via various
sources, for instance by surveillance or measuring.
Smart cards for public transport which in many cities replaced classic paper
tickets enable authorities to track their holders. Even rubbish collection can
create information about the amount and composition of trash, that allows to
charge specific fee for each family. With photographies made by drone or by
traffic management system and CCTV footage supply, they create category
of directed and automated data. Every day use of smartphones, cameras and
other digital devices also gives rise to digital data collected by simple activities
like logging in, recording notes, pictures, etc. In many situations people may
not even realize how much information they reveal - simple swiping a credit or
loyalty card uncover when, where and what was purchased.(Kitchkin 2014)
With automated surveillance systems, sensors and agencies the fact that state
is the main generator and user of data may not be surprising. Better knowl-
edge of population could lead to cost savings through operational efficiency,
better informed citizenry accompanied with improvement in state administra-
tion. Advanced network of controls on borders, immigration screening and
other data sources contribute to ability of keeping state security and of fighting
against crime.(Kitchkin 2014)
Rationale for using big data for businesses lies in organizational efficiency as
well. A new view on information makes possible not just to enhance the com-
petitive advantage in the time of reducing the costs, but also to improve the
customer experience. Executives with detailed and timely insight can manage
actions in smarter, more flexible and innovative ways.(Manyika 2011)
2. Literature Review 6
2.3 Google Trends
The position of a global leading search engine gives Google an opportunity to
gather loads of data about its users. Keywords people use in order to find
desired information may reveal recent trends or situations that captured wide
attention. For this purpose Google released Google Trends service and in such
manner gave public access to percentages of web searches showing how many
queries were done over a certain period of time. (Google Inc. 2016b)
However, one cannot determine a precise amount which was typed into the
query box. First, it would be troublesome to keep pace with numbers chang-
ing from second to second with millions of people ’googling’ in real-time. And
second, Google adjusts the data to make them easier to compare. Some terms
may be very popular and have high absolute frequency which would make them
always rank first, therefore for each keyword data points are scaled with respect
to its own maximum. As Google Inc. (2016b) mentions, ’popularity is made
relative’ by division of data by total searches in the area and time it repre-
sents. In Trends environment anyone can follow in weekly intervals popularity
of chosen query, except searches made by very few people, words with special
characters and Google also omits searches repeated by the same person over a
short period of time. (Google Inc. 2016b)
2.4 Big Data Hubris
In 2008 Google…