Glass Ceilings in the Art Market

Abstract

Using an exclusive data set consisting of the population of fine art auctions from 2000 to 2017for Western artists, with over 2.6 million auction sales, we provide strong empirical evidence of aglass ceiling for female artists. First, we show that female artists are less likely to transition fromthe primary (gallery) into the secondary (auction) market, where 96.1% of auction sales are bymale artists. This higher bar materializes in an average price premium of 4.4% for artworks byfemale artists at auction. Second, this premium is driven by a small number of women locatedat the top of the market, and manifests as a discount of 10% when we account for the number ofartworks sold per artist. Third, the superstar effect, where a small number of individuals absorbthe majority of industry rewards, prevails; at the top 0.1% of the market, artworks by femaleartists are traded at a discount of 9%. Moreover, the very top 0.03% of the market (where 40%of the sales value are concentrated) are currently unattainable for female artists selling at auction– revealing a glass ceiling. Our study has important implications for industries characterized bya superstar effect and illustrating how market structure impacts gender equality.

Full text

Glass Ceilings in the Art Market ∗
Fabian Y.R.P. Bocart
Artnet Worldwide Corporation†
Marina Gertsberg
Maastricht University‡
Rachel A. J. Pownall
Maastricht University§
August 20, 2018
Abstract
Using an exclusive data set consisting of the population of fine art auctions from 2000 to 2017
for Western artists, with over 2.6 million auction sales, we provide strong empirical evidence of a
glass ceiling for female artists. First, we show that female artists are less likely to transition from
the primary (gallery) into the secondary (auction) market, where 96.1% of auction sales are by
male artists. This higher bar materializes in an average price premium of 4.4% for artworks by
female artists at auction. Second, this premium is driven by a small number of women located
at the top of the market, and manifests as a discount of 10% when we account for the number of
artworks sold per artist. Third, the superstar effect, where a small number of individuals absorb
the majority of industry rewards, prevails; at the top 0.1% of the market, artworks by female
artists are traded at a discount of 9%. Moreover, the very top 0.03% of the market (where 40%
of the sales value are concentrated) are currently unattainable for female artists selling at auction
– revealing a glass ceiling. Our study has important implications for industries characterized by
a superstar effect and illustrating how market structure impacts gender equality.
Keywords: Art market, Auctions, Gender economics, Labour economics
JEL Codes: J24, J31, J71, Z11
∗We would like to thank Pierre-Andr´e Chiappori, Dakshina De Silva, Raffi Garcia, William Goetzmann, Claudia
Goldin, Kathryn Graddy, the seminar and conference participants at Bocconi University, ESSFT Gerzensee, the Inter-
national Industrial Organization Conference, the Conference on Auctions, Competition, Regulation and Public Policy
and the Yale Symposium on Art and Gender for valuable comments.
†Fabian Y.R.P. Bocart, Artnet Worldwide Corporation, 233 Broadway, New York 10279-2600, USA; Email:
FBocart@artnet.com
‡Marina Gertsberg (corresponding author), Maastricht University, P.O. Box 616, 6200 MD Maastricht, The Nether-
lands; Email: m.gertsberg@maastrichtuniversity.nl
§Rachel A. J. Pownall, Maastricht University, P.O. Box 616, 6200 MD Maastricht, The Netherlands; Email: r.
pownall@maastrichtuniversity.nl

1 Introduction
Market structures characterized by the superstar effect where a few individuals absorb the majority
of rewards can reinforce the lack of mobility of underrepresented market participants. In these
so-called “winner-take-all” industries lesser talent is an imperfect substitute for higher talent and
the gap in compensation can be disproportional to the gap in skill (Rosen, 1981). Eventually, the
size of rewards will depend more on rank than on talent. As stated by Frank and Cook (2013),
in “winner-take-all” markets sellers who are not located in the top of the distribution often earn
less than they could have earned in alternative occupations. At the same time, consumers in these
markets are often not able to perfectly evaluate the level of skill making meritocracy more difficult
to achieve. As a result, higher quality standards might be applied to market participants for whom
fewer track records are available and for whom it was historically more difficult to acquire legitimacy
due to lower status characteristics.1Furthermore, information cascades and herding behavior where
buyers imitate the decisions of preceding actors disregarding their own private information can result
in market inertia (Banerjee, 1992; Bikhchandani et al., 1992). In summary, the extremely large
competition for top ranks in combination with information asymmetries in these industries might
inhibit underrepresented groups from progressing. This may lead to a glass ceiling for them as
consumers try to minimize risk and avoid uncertainty in their consumption decisions.
In this large-scale empirical study, we are interested in whether the superstar effect encountered
in the art market translates into barriers for female artists at auction. In particular, we investigate
auction outcomes for artworks created by female and male artists and analyze whether the prevailing
market structure impacts artwork prices and materializes in a glass ceiling for female artists at the
top of the market.
The superstar effect is very well illustrated in the art market. The distribution of rewards in this
industry is highly skewed with the largest profits concentrated at the top. According to the latest
Art Basel and UBS art market report (McAndrew, 2018) only 1% of artists accounted for 64% of
auction sales in terms of value in 2017. The heterogeneous nature of art as a good makes it difficult
to evaluate its quality. The assessment criteria that make a good artist are not fully understood and
involve a lot of subjectivity in judgment. These information asymmetries give an important role to
the institutions in the art world (for instance, museums, art dealers, auction houses, art journalists
1Lower status characteristics might refer, for instance, to age, education race or gender.
2

and curators) in identifying talent and in determining the superstars in the market. Since the birth of
the modern art market in the 19th century in London little has change in terms of the identity of the
key instituions, sales channels as well as the structure of the competetive landscape (De Silva et al.,
2017). It is also documented that female artists are historically underrepresented in this industry.
The eminent female art historian, Linda Nochlin, was among the first to question the notion of the
male genius and to draw attention to the issue of gender discrimination in the art world in her 1971
landmark essay “Why Have There Been No Great Women Artists?”. Later in 1984, the Guerrilla
Girls started to create awareness for sexual discrimination by pointing out the underrepresentation
of female artists in the New York based Museum of Modern Art’s exhibition International Survey
of Painting and Sculpture where only 10% of all works were by female artists (The Guerrilla Girls,
2017).
The share of female artists decreases gradually along the career ladder indicating impaired mo-
bility. While women do not display less interest in pursuing an artistic career than men do (about
50% of all Master of Fine Arts (MFA) holders are female in the US), their share drops to 30% in
commercial US galleries (National Museum of Women in the Arts, 2017) and to 25% at art fairs
(McAndrew, 2018). According to the National Museum of Women in the Arts (2017), nowadays,
artworks by female artists represent only 3% to 5% of major permanent collections in the US and
Europe. Furthermore, based on our data, female artworks at auctions make up less than 4%.
The auction market accounts for about 50% of the total sales value in the art market (McAndrew,
2018; Pownall, 2017) and constitutes the secondary market where typically artworks are traded which
were previously sold on the primary market (i.e. through galleries). This implies that these artworks
have a resale value and are in demand by some buyers. The appearance of an artwork at auction
signals professional recognition (Goetzmann et al., 2016). Therefore, artists who reach this stage
are regarded as relatively established with their quality being certified by the market. Similar to
individual wages these prices represent market value of an artist. An artist’s track record within the
secondary market is highly visible to the public as opposed to prices in the primary dealer market.
Information on past auction results are frequently used by art collectors, experts, consultants and
insurers as input to determine the value and future potential of an artist. Therefore, while prices
fetched at auction do not directly accrue to the artists themselves they may have a large impact on
their careers and can also feed back into gallery prices (Galenson and Weinberg, 2000).
3

Given the low representation of women in the art market, we expect the superstar effect to create
larger barriers for female artists with the result that they are less likely to reach the secondary
(auction) market than male artists. However, conditional upon reaching this market, we predict
that the artworks by female artists should either command a price premium or show no significant
difference compared to prices for artworks created by male artists after controlling for quality. The
rationale for a premium for female artworks might arise due to a (perceived or actual) stricter quality
filter in earlier career stages which creates a higher bar for female artists. As a result, a female artist
will have to be on average better than a male artist in order to acquire legitimacy. Similar dynamics
have also recently been observed in a study involving a field experiment on a large online Q&A forum
(Bohren et al., 2017) as well as for the case of the referee process in academic journals (Hengel, 2017),
patent applications (Jensen et al., 2018) as well as investment advice (Botelho and Abraham, 2017;
Egan et al., 2017). However, if being traded on the secondary market is not sufficient to signal
quality (i.e. in order to resolve uncertainty with respect to the ability and/ or the future potential
of female artists) this would materialize in female artists being less likely to reach the highest ranks
of the market or a glass ceiling.
We have exclusive access to a exclusive auction data set representing nearly the whole population
of global art auction transactions in the time period between 2000 and 2017. From this global
database we consider only Western artists to facilitate an accurate identification of the gender of the
artists. Overall, we have auction results of 110,938 male and 5,612 female artists (with 2.6 million and
105,844 lots respectively). Furthermore, the data covers several art genres and media types allowing
us to segment and homogenize sub-samples resulting in improved comparability. This auction sample
allows us to compare the performance of female artists in the secondary market to the performance
of male artists. Our focus is on the output (the artistic product) instead of on the input of labor as is
often common practice in gender performance differential studies. This approach has the advantage
that our results are less likely to be driven by differences in the individual characteristics of the
artists. These may include better negotiation and self-promotion skills due to relative overconfidence
of men as documented in the literature (?). Artists can be considered entrepreneurs who take on a
large human capital risk by pursuing an artistic career. Therefore, their incentives are not distorted
and agency conflicts are immaterial in this setting. Additionally, we are fortunate to also employ
a unique primary (gallery) market data set to investigate the mobility between the primary and
4

the secondary (auction) market for male and female artists. Access to the opaque dealer market
data is extremely limited and has not been studied empirically within the art market literature.
Lastly, to test for the presence of a glass ceiling specifically and to observe the upper tail of the price
distribution where most of the rewards are concentrated, we perform a quantile regression analysis.
Our data reveal strong evidence for the presence of structural barriers faced by female artists
in the market for fine art. First, we find that women are less likely to be traded in the secondary
(auction) market. This selection bias may explain the average 4.4% price premium found for female
artworks at auction. Second, this price premium is likely to be due to a supply squeeze caused by
a small number of female artists located at the top 1% of the market. In particular, this effect
can be attributed to female artists from older generations (mainly pre-1950’s works of art) where
institutional barriers for women pursuing artistic career were stronger. At the same time, we observe
an increase in the share of contemporary artworks by women traded at auction. These artworks sell
at a price discount of 8.3% compared to contemporary artworks by men. This provides evidence of
lower barriers for female artists in more recent time periods. Further, consistent with results from
surveys and qualitative studies in the art market (Reis, 1995a,b) we find that being above the child-
rearing age has a more positive effect for a female than for a male artist. However, we also show that
market concentration is higher for women than for men accentuating stronger competition for top
ranks among female artists. This implies that the superstar effect, where rewards are concentrated
among a few individuals, appears to be even more prevalent for women. Third, we identify a second
hurdle faced by female artists which is manifested in their exclusion from the superstar league of the
market. In particular, the top 0.03% of the market, where 40% of the sales value is concentrated,
appear out of bounds for women and is entirely occupied by men. This is supported by the quantile
regression results which reveal that artworks by female artists are traded at a price discount of 9%
in the 99.9th quantile of the price distribution. Assuming that talent is evenly distributed across
gender the exclusion of women from the superstar league of the market points to important structural
barriers for underrepresented groups in the market for fine art.
We complement the literature on the influence market structures charactersized by the superstar
effect and their implication for the performance of underepresented groups in particular within the
area of the cultural and creative industries. There is a rich literature provding empirical evidence
5

for a gender gap across industries materializing in a pay inequality between men and women.2With
respect to gender inequality in the art world and in particular regarding performance differences
between male and female artists, anecdotal evidence as well as evidence from surveys (Throsby and
Mills, 1989; Throsby et al., 2003, 2010), case studies (Cowen, 1996; Reis, 1995a,b) and a small number
of empirical studies (Adams et al., 2017; Cameron et al., 2017; Rengers, 2002) can be found. Findings
stress high underrepresentaion, relatively lower sales revenues of female artists, as well as hurdles
experienced by women that impede their careers. With the exception of Adams et al. (2017), rather
small sample sizes impact the external validity and robustness of these results. As superstar markets
are characterized by a skewed distribution of income, it is of particular importance to be able to
analyze the tails of these distributions which calls for a critical sample size. For instance, using
a sample of 4,434 fine art graudates from Yale University from 1891 until 2014, a recent study by
(Cameron et al., 2017) provides partial empirical evidence of a higher bar for women. They show that
while female artists experience more obstacles than men when entering the market, their artworks
sell at a premium conditional on being traded in the market. This effect is mainly driven by pre-1983
graduates and disappears for post-1983 graduates. Even though these authors use a rather small
sample (auction results of 515 artists are examined) which is more likely to include higher skilled
artists (given the quality of the Yale School of Art), our results are consistent with these findings.
Another recent larger empirical study by Adams et al. (2017) employs a sample of 1.5 million global
transaction over a time period from 1970 until 2013 to study the performance of female artists at
auction. The authors document an average discount of 47.6% for female artworks before adjusting for
the quality of the artworks which is line with an unconditional average discount of 16.8% documented
in our data. This effect is moderated by the level of country-specific gender inequality. Our results
might be different from their findings due to differences in the sample composition. Given that we
have over 2.6 million auction transactions over a period of 18 years (2000-2017) for Western artists
only our data is more dense representing almost the entire population of auction transactions for
this sub-set of artists during the particular time span. Furthermore, our focus is specifically the top
end of the art market as opposed to average effects. There is some statistical support from other
industries subject to superstar economics showing that women encounter a glass ceiling when they
climb up the career ladder. This was found to be the case for top athletes (Kahn, 1991), movie stars
2See among other (Arulampalam et al., 2007; Blau and Kahn, 2000; Goldin et al., 2017).
6

(Bielby and Bielby, 1996; Lincoln and Allen, 2004), top executives (Bertrand and Hallock, 2001)
and researchers (Barbezat and Hughes, 2005; Probert, 2005). However, these studies use salaries to
investigate reward differentials. This approach bears the risk of omitting individual characteristics
such as negotiation skills which might explain the gap in salaries. Our focus on auction sales enables
us to isolate any direct influence of the artist on prices.
Our results provide important insights on emerging market structures that coincide with the
presence of a superstar effect. It appears that the competitive pressure for high ranks paired with
uncertainty about the level of skill leads to higher quality standards for underrepresented market
participants and impairs their mobility. While we cannot provide clear evidence of the mechanism
that causes the barriers that lead to the observed performance differences between male and female
artists, we offer material input for a closer investigation of the underlying reasons.3Establishing
a comprehensive and detailed overview on the state of gender inequality in the art market beyond
anecdotal evidence is the first step towards its mitigation. Parallel dynamics are likely to be present
in other occupational areas characterized by the superstar effect such as high-end gastronomy, top
sports, academia, journalism as well as for leadership positions in general which are still largely
dominated by homogeneous groups. Therefore, this study might find valuable application across a
wider range of industries acting as a starting point for a founded and constructive discussion on
gender equality.
The paper proceeds as follows: In section 2, we present our conceptual framework and formulate
predictions which will be tested empirically. The data are described in section 3. Section 4 is
dedicated to the empirical analysis and results. We finish with some concluding remarks in section
5.
2 Conceptual Framework
Our conceptual framework combines the theory on statistical discrimination as laid out by Phelps
(1972) and a demand externality whereby the marginal utility of one person undertaking an action
is increased with number of other people engaging in the same behavior in order to explain how the
3One approach to disentangle the supply from the demand side would be to perform the analysis separately for common
names where there is a clear association with one gender (e.g Mary) and ambiguous artist names (e.g. Kim). However,
this is not a plausible assumption that auction houses or auction participants are uninformed about the identity of
the artist. For instance, auction catalogs typically provide some information about the artist using pronouns which
disclose the gender of the artist.
7

superstar phenomenon can impact the performance of underrepresented groups in the market. In the
context of this study, it will help us to formulate predictions concerning the performance of female
artists at auction.
An artist’s career typically starts with completion of an art degree4after which he or she seeks
gallery representation. A gallery acts as an agent with the goal to create a market for the artist in
exchange for a commission on the price of the sold artworks. This aligns the incentives of the artist
and his or her dealer and makes the initial hiring process very crucial. A sign for the artist’s market
establishment is that his or her artworks appear at auction sales (the secondary/ resale market)
trading for prices not below gallery level.5The sellers at auction are typically composed of private
individuals who previously acquired the artwork through the primary market (or through inheritance
as a family bequest).6With the exception of liquidity sales (these are famously known as the three
D’s: death, disease and divorce), an artwork is usually put up for sale at auction if there is a belief
that it has increased in value and that there is demand for this artist. There is an interdependence
between the gallery and the auction market. Trends in auction sales reflect in gallery representation
patterns and feed back into prices. At the same time, effective dealer marketing can boost the
demand of an artist in the secondary market. Therefore, we will focus on the dealer’s (art gallery’s)
as well as the buyer’s perspective at auction in order to formulate predictions for how the superstar
effect will influence the performance of male and female artist in the market.7
In this conceptual framework profit-maximizing art galleries and buyers (the sources of utility
for buyers will be discussed in more detail below) are unable to perfectly observe the ability of an
artist and instead observe a noisy signal. In order to interpret the quality signal art galleries use
an observable salient group characteristic (gender in this case) which is correlated with ability to
proxy for the missing information. The rationale for why gender may be employed as a lens to make
inferences about ability is due to the historically lower status ascribed to women (Ridgeway, 2001).
The distinctiveness of gender as an attribute provides a large amount of differentiation between groups
with little within-group variation (Hilton and Von Hippel, 1996). A posterior belief of the artist’s
4Not all artists are completing university art programs. According to Davis, Ben (2016), out of the 500 most successful
early-career artists 35% have no MFA degree and 12% are self-thought.
5There are also instance in which artists sell directly through auction houses (Damien Hirst being a very prominent
example). However, this is considered a risky strategy since only little control can be exerted over prices. Price stability
is of utmost importance in the art market as prices proxy the quality of an artwork and future potential of an artist
(Velthuis, 2007).
6It also happens that museums deaccession parts of their collections through auctions.
7Except for art dealers and collectors, museums and art critics can exert a major influence on the artist’s career.
8

ability will be formed using a prior (unbiased) belief about his or her gender’s ability. If prior beliefs
for male and female artists are not the same different quality standards will be applied to the two
groups. A crucial element is that due to the superstar effect and disproportional rewards to talent,
the relevant measure of group performance will be drawn from the top of the sales distribution at
auction instead of the average. The effect of demand externalities will come into play when explaining
why there will be a glass ceiling for female artists at the top of the art market despite higher hiring
standards in the primary market.
In both belief-based models that constitute the seminal contributions on statistical discrimina-
tion (Phelps, 1972; Arrow et al., 1973) hiring decisions are based on observing group identity, the
group’s past performance and a quality signal emitted by the applicant. While in the framework
of Arrow et al. (1973) group differences endogenously arise in equilibrium and the minority group
is rationally discriminated due to self-fulling negative ex-ante beliefs of the employers, in the model
by Phelps (1972) rational discrimination occurs due to imperfect information on employee ability
with a (correct) more negative belief about the ability of the minority group. Female artists might
indeed choose to underinvest in their skills upon observing the past performance of their female
peers as predicted in Arrow et al. (1973). Yet, since the number of professional female artists shows
an increasing trend (National Endowment for the Arts, 2011) and given the fact that information
assymetries are a key obstacle within the art market we base our predictions upon the statistical
discrimination model of Phelps (1972). Furthermore, we cannot exclude taste-based explanations as
laid out by Becker (1957) for the underperformance of one group in the market. However, it is in our
interest in this study to show that the superstar effect can evoke discrimination on rational grounds.
Art gallery hiring decision. In this study we cannot empirically observe the hiring decision
of an art gallery. However, conceptualizing this process important in order to understand the per-
formance of female and male artists at auction given that the gallery’s hiring decisions impact both
the supply and demand for artists in the secondary market. Galleries sales figures are not publicly
available. Nevertheless, there is evidence that the proportion between male and female artists rep-
resented by galleries is unbalanced (ranging at around 30%)8as opposed to the proportion among
MFA graduates (National Museum of Women in the Arts, 2017).
Imagine a population of identical artists who completed an art education program and who are
8In our gallery sample there are 12.5% female artists.
9

now seeking representation from a population of art galleries. While an art gallery cannot perfectly
observe an artist’s ability, it can observe group identity g∈ {F, M}denoting male and female artists.
An artist’s ability, a, is fixed and drawn from a normal distribution N(µg, σ2
g). As auction results are
public and art dealers have deep insights into other galleries we assume that art galleries hold correct
beliefs about the true distribution of ability across gender within the population (i.e. µgis estimated
without bias).9Overall, galleries observe group identity and a noisy signal of ability, θ=q+ε,
where εis a zero-mean error which is normally distributed with N(0, σ2
ε,g). In our case, this signal
may be an artist’s track record consisting of previous grades on university program projects, shows
and exhibitions, awards or other quality certifications. Art galleries will use the noisy signal, θ, to
infer the expected value of qcombining the information contained in his or her group identity and
the artist’s track record. If the signal would be precise (ε= 0) information on group identity would
not be required to determine an artist’s ability. However, assessment criteria that contribute to a
good artist are not fully understood and involve a lot of subjectivity in judgment. This condition
makes the art market highly susceptible to stereotyped thinking and a reliance on past group related
information. The art gallery will use the Bayes rule to form a posterior belief about the ability of an
artists where ability and the signal are jointly normally distributed. The conditional distribution of
qgiven θis normal with a mean equal to the weighted average of the signal and the unconditional
group mean:
E(q|θ) = σ2
g
σ2
ε+σ2
ε,g
θ+σ2
ε,g
σ2
g+σ2
ε,g
µg,(1)
where E(q|θ) is strictly increasing in θand µgmeaning that higher signals and higher expected
group ability result in a higher expected posterior estimate of the ability of an artist. The precision
of the signal ( σ2
g
σ2
ε+σ2
ε,g
) will determine how much weight is placed on the population averages versus
the signal’s value. With perfect information (ε= 0), an artist will only be evaluated based on his or
her signal and discrimination will not arise. As mentioned above, given that an artist’s ability cannot
be broken down into objective measurable criteria it is highly likely that a considerable amount of
weight is placed on population averages.
When relying on group averages in order to infer ability, two properties within the group distri-
9Bohren et al. (2017) also incorporate biased beliefs in their model of gender discrimination.
10

butions may lead to different prior beliefs which may result in differential standards in interpreting
the artist’s signals depending on group identity. First, an art gallery may (correctly) perceive that
the male population of artists has on average higher ability than female artists (µF< µM). This
will be the case if historically female artists underperformed male artists in terms of auction sales.
A curucial element is that the superstar effect prevalent in the art market implies that the largest
part of the sales value is concentrated among a small number of artists who are located at the higher
part of the price distribution. The prices at the top which eventually reflect in art gallery prices are
disproportional to the difference in rank between the artists. Therefore, art galleries will base their
beliefs about group ability in terms of past performance in the right tail of this skewed distribution
as opposed to its mean. In our context, this means that female artists will be subject to a disad-
vantage when there is imperfect information about their individual ability. Second, there is also the
possibility of differences in how informative the signals are emitted by the two groups (σε,F > σε,M ).
In particular, this is highly likely to be the case if one group (here women) was historically under-
represented wihtin the artist population. Art galleries will have collected less information on women
not being used to interacting with that group. As a result, galleries will attribute less confidence to
a pre-market signal emitted by a female artist than by a male artist.10 The resulting effect of both
sources of group differences (evidence of unequal abilities and precision of signals) is that female
artists (the disadvantaged/minority group) will be subject to a higher threshold than male artists.
In order to offset negative prior beliefs due to underrepresenation and less available information
they will have to emit higher signals given equal underlying ability in order to be hired by an art
gallery. This higher quality standard does not only imply that female artists will be represented by
art galleries at a lower rate11 but also that conditional on being hired a female artists will on average
be better than a male artist. This should reflect in higher prices for female artists once hired by
a gallery. While we do not model the supply side, it is also likely that higher standards for female
artists lead to a self-selection among women whereby only the most talented and persistent women
will pursue a professional artist career.12
Hypothesis 1 (H1).Conditional on being traded in the market female artworks will sell on average
10Bjerk (2008) considers both factors in his dynamic model showing that a sticky floor leads to a glass ceiling. In our
case, in addition to a sticky floor (hiring by an art gallery), the minority group is also subject to a barrier at the top
(at auction) due to the demand externality.
11If talent is randomly distributed across gender by nature, fewer women will be able to meet this higher threshold.
12For instance, Breen and Garcia-Penalosa (2002) show that the anticipation of lower revenues leads to a self-selection
among women which is responsible for gender segregation in occupations.
11

at a premium relative to the artworks by male artists.
Buyer’s purchasing decision. The buyers at auction sales are usually composed of private
collectors, art dealers as well as institutional buyers such as museums. In particular for private
collectors, purchasing motives may be driven by aesthetic, status and investment considerations.
Recently, art and other collectables such as wine or stamps have received a lot of attention for its
suitability as an financial asset class.13 It is reasonable to assume that after a certain monetary
threshold the ability of the artwork to act as a store of value is likely to be an important element.
While this threshold is subjective, we assume in our framework that the buyer derives utility from the
artwork’s conspicuous consumption value (status) as well its financial return as suggested by Mandel
(2009). Both motives imply that a buyer cares about the quality and future value appreciation of
the artwork. Furthermore, we adapt the assumption that consumer demand is not independent but
depends on the behavior of others. This introduces an externality or dis-functional utility to the
buyer’s demand function. Just like art galleries buyers cannot directly observe the ability of an artist
(or the quality of the artwork) and are subject to information asymmetries. Therefore, the buyer
will also make use of information contained in an artist’s group identity in order to evaluate his or
her underlying ability.14 In addition to group identity and the group’s past performance, art buyers
at auction are aware of the hiring policy of art galleries which set higher quality standards for female
artists. Moreover, a buyer at auction can also observe the purchasing behavior of all buyers before
him. Given that as a result of the gallery’s hiring policy female artists are underrepresented in the
market the buyer is more likely to observe a sale of an artwork by a male artist than by a female
artist providing him with more information.
When information asymmetries with respect to the talent of an artist are present, following the
crowd can be beneficial. First, herding may be rational when collecting own information involves
substantial costs (Banerjee, 1992). Second, even if private information is available adopting the ma-
jority’s choice and ignoring own information can be risk reducing since being wrong when everyone
else is wrong would be associated with a lower reputational cost than being wrong alone (Scharfstein
and Stein, 1990). Making a mistake becomes costlier the more capital is at stake. Lastly, if apprecia-
13There is a rich literature investigating the risk and return profiles of art, wine and stampts including among other
Ashenfelter and Graddy (2003); Baumol (1986); ?); ?); Goetzmann (1993); Mei and Moses (2002); Pesando (1993);
Renneboog and Spaenjers (2013).
14Since information on male artists tends to be more salient (Cameron et al., 2017) one might also imagine a scenario in
which buyers hold biased beliefs due a representativeness heuristic (Bordalo et al., 2016). For the sake of our analysis
we assume that buyers hold unbiased beliefs about the distribution of ability among male and female artists.
12

tion is the source of utility of consumption purchasing art which is consumed by a substantial number
of people can increase consumption capital leading to an even higher demand for a particular artists
(Adler, 1985; Stigler and Becker, 1977). Therefore, the buyer will maximize the following utility
function:
Ui=ai−P+ci(Zi,t−1),(2)
with ci>0. aiis a random idiosyncratic taste-shock incorporating income and other taste
parameters. Pis the exogenous price of engaging in the sales transaction. Lastly, Zi,t−1is the
proportion of the agent’s reference group buying an artwork of an artists of a particular gender at
time t−1. This means that the agent considers the external demand effect in the previous period.
The utility of agent, i, depends on his or her taste-shock, the cost of engaging in the transaction as
well the proportion of his or her reference group engaging in a certain purchasing pattern with respect
to the artist’s gender in the past period. An important implication is that price increases do not
result in falling demand since the additional utility from external consumption creates a bandwagon
effect that keeps the demand high. While we do not specify the unique sources of the utility derived
from the consumption of others, the point we want to stress is that the additional utility gained from
the increased consumption of other market participants might prevent the mitigation of statistical
discrimination despite the on average higher quality of female lots. The resulting glass ceiling for
female artists is then not necessarily an indicator for a distaste for female artists as the utility from
increased consumption can dominate a high quality signal. As a result there will be an imperfect
substitution between men and women and equally talented male and female artists will not have the
same chance of becoming superstars. This might lead to inertia and a persistent underrepresentation
of women among superstars in the art market as the advantage of one group over another accumulates
over time making it difficult to catch up.
Hypothesis 2 (H2).Female artworks will sell on average at a discount relative to the artworks of
male artists at the top of the price distribution at auction.
13

3 Data
3.1 Sample
The data employed in this study were provided by Artnet AG (Artnet thereafter). The Berlin-based
company is an online platform offering trading as well as research and analytic services within the art
market. Their price database dates back to the year 1989 and has over ten million price quotation
records.15 Artnet collects all global art auction transactions which reach a hammer price equal to 500
US Dollars and above. As a result, our data set can be considered to represent the entire population
of art auction transactions worldwide.
In this analysis, we focus on the fine art sector. The category includes photography, prints and
multiples, works on paper, paintings, installations, design objects and sculptures totaling 6,140,774
auction transactions. We exclude installations from this study. The market for installations is
slightly different from the market for other more traditional object types as installations are more
difficult to maintain, store and exhibit for collectors. Furthermore, as Artnet gradually increased
the comprehensiveness of its price database between 1989 and 2000, we restrict our sample period
to the years 2000 (January) to 2017 (April) resulting in a very high degree of representation since
the millennium.
The database provides information on transaction characteristics including the name of the auc-
tion house and its location, the date of the sale, the lot number, the price pre-sale estimate of the
auction house and the hammer price in US Dollars before transaction costs. We deflate all prices
using the US consumer price index (CPI) provided by the OECD using 2017 as our base year.16
With respect to the artists’ attributes, the database records name, date of birth, living status and
nationality. At an artwork level, we have information on the title of the work, its size and object
type. Additionally, we categorize all auction transactions into movements based on the birth year
of the artist. Consistent with the classification in the Tefaf report (Pownall, 2017), we distinguish
between Old Masters and Impressionists (1250-1874), Modern (1875-1910), Post War (after 1911
and deceased) and Contemporary (all living artists). The artworks where the artist’s birth year
was not available are subsumed under “other”. We do not consider artists born before 1250. It is
important to mention that while we have artists in our sample from different artistic movements and
15This includes decorative art (antiques, ceramics, furniture, jewelry, and watches) which us excluded from this analysis.
16The employed consumer price index can be found under https://data.oecd.org/price/inflation-cpi.htm.
14

generation, we observe their sales only in the time period from 2000 until 2017. This implies that
while opportunities for these artists differed across time, we do not expect the perceptions of buyers
with regard to gender performance differences to vary too much during the period of the past 17
years.
Our variable of interest is the artist’s gender. Since Artnet’s price database does not indicate
the gender of the artists, we identified female artists by matching them to a name list. In order to
ensure accuracy and increase the homogeneity of the artists in our sample in terms of opportunities
such as access to resources and education, we focus on Western artists who are based in Europe and
North America (the US and Canada). Whenever there were two nationalities attributed to an artist,
the name was included in the sample if either of the nationalities was European or North American
(e.g. the male artist Zao Wou-Ki who is French-Chinese). We use two name lists available from the
US Social Security Administration17 and the German computer magazine Heise18. The former list
contains North American baby names, while the latter provides a name dictionary with a focus on
European names by country. In cases where the name was unisex (e.g. Jessy, Joan and Kim), we
manually researched the identity of the artist. Instances where the artist consisted of more than one
person (e.g. Christo and Jeanne-Claude) were dropped from the sample.
As a result, we were left with a sample size of 4,387,393 observations. We drop observations
where information on the dimension (size) of the object is missing which is the case for 58,166
transactions. Lastly, we exclude bought-in lots from our main analysis.19 Our final sample consists
of 2,677,190 auction transactions. To the best of our knowledge, this represents the largest and most
comprehensive art market auction transaction sample so far employed in such a study.
Additionally, we have exclusive access to primary market data provided by Artnet. Primary
market data identifies which artists are represented by which galleries and is highly confidential and
therefore difficult to obtain. As a provider of art market services, Artnet also provides an online
platform for art galleries to sell their work. This data set will be applied to examine the presence of
entry barriers into the secondary market for female artists. It contains the names of the galleries and
the names of the artists they represent as well as the artist’s year of birth over the time period from
17The list is available at https://www.ssa.gov/oact/babynames/limits.html.
18The list is available at ftp://ftp.heise.de/pub/ct/listings/0717-182.zip.
19In auctions, a buy-in takes place when an artwork is not sold as it fails to meet the seller’s reserve price. The buy-in
rate in our sample is 37.73% (1,622,019 observations) which is in line with the commonly observed buy-in rates in
auction sales.
15

2000 until 2017. Due to disclosure limitations, we pool the data and treat it as a cross-section instead
of a panel. In total, there are 1,281 galleries in Artnet’s international gallery network representing
15,121 unique artists. Again, we focus on Western artists only. Furthermore, as we are interested in
the transition from the primary to the secondary market we restrict our sample to the population of
living (contemporary) artists. This leaves us with an overall sample of 4,754 artists.
The following subsection will introduce the properties of our data set and provide some first
evidence for gender differentials within our sample based on univariate analysis.
3.2 Descriptive Statistics
Table 1 illustrates the extent of the concentration within the secondary market for art based on our
data for the whole sample period (2000 until 2017). It depicts for different shares of the market
(in terms of US Dollar value) the percentage (number) of artists who account for it. First, the
market is highly concentrated with only 2.2% of the artists accounting for 90% of the overall sales
value. Second, artwork sales of female artists amount to only 3.4% of the total auction market
($121.4 billion). Third, whereas the female segment is smaller in size it is more concentrated than
the male market. While 19.9% of male artists are responsible for 99% of the sales value, only
15.5% of all female artists occupy the same share within their segment. These numbers suggest that
the art auction market resembles a superstar market where rewards are concentrated among a few
individuals.20 This appears to be amplified for the segment of female artists.
Table 2 shows the summary statistics for auction prices for men and women with detailed statistics
by artistic movement, object type, region and living status. The last column presents the difference
between mean male and female prices. Overall, 96.1% (2,572,346) of all artworks sold at auction can
be attributed to male artists. Hence, the proportion of female artworks in terms of volume is slightly
higher (3.9%) than their share in terms of value (3.4%) in our sample. Figure 1 shows how the total
sales value and volume evolved for both genders over the sample period as well as over different
generations. We chose these two dimensions since while attitudes toward gender might not have
changed profoundly over the last 18 years, the market might perceive gender differently across artist
generations due to the improvement of conditions for women pursuing an artistic career. As shown in
Figures 1(a) and 1(c) sales volumes have clearly increased for men and women with a larger relative
20Another attribute of the superstar effect in the sense of Rosen (1981) is that rewards are disproportional to talent.
However, in this study we are unable to exactly determine the level of talent of each artist.
16

increase for women. While female artists increased their overall sales volume by a multiple of 1.95
(from 3,714 artworks in 2000 to 7,247 artworks in 2016), male artists increased sales by a multiple of
1.68 (from 97,807 artworks in 2000 to 164,936 in 2016). Similarly, total sales values have increased
for both genders despite a dip following the financial crisis. From the year 2000 until 2017, female
artists increased sales value by a multiple of 6.0 while male artists only increased sales by a multiple
of 2.8. Nevertheless, female artists remain a small fraction of the overall market in terms of volume
and value (4.2% in terms of volume and 5.0% in terms of value in 2016). For both genders, sales
numbers highly increased for artists born after 1875 as depicted in Figures 1(b) and 1(d). This is
more pronounced for female artists and is likely to reflect a higher supply of contemporary artworks
and indicates lower entry barriers for female artists born in later generations.
With respect to the number of artists, men clearly dominate the auction market occupying
95.2% of the market. While there are 110,938 male artists, there are only 5,612 female artists. The
proportion of female artists is highest for Contemporary art (9.3% are female) and smallest for the
Old Masters period (2.9% are female). Figure 2 shows the evolution of the number of male and
female artists during the sample period and over generations. From Figures 2(a) and 2(c) which
depict the number of distinct male and female artists in every year, we can observe that there is
an increasing trend for both groups over the years. However, the trend is stronger for the female
sub-group with an almost three-fold increase from 165 artists in 2000 to 456 artists in 2016. The
number of male artists at auction per year less than doubled from 4,303 to 7,815 artists over the
same time period. As a result, the male-to-female ratio improved by 40% over time from 0.03 in
2000 to 0.05 in 2017 (see Figure A2 in the Appendix). This trend is also reflected in Figure 2(d)
which shows a steady increase in the number of female artists over the generations with a clear peak
for the generation that was born between 1975 and 2000. The number of male artists remains rather
stable for the generations born after the year 1875. The rising market entry by female artists points
to a potential improvement in conditions and higher market acceptance making the artist profession
more attractive for women.
An interesting observation is that while the average prices of female artworks are significantly
below the average price for male artworks ($39,065 versus $45,614)21, the median price is with $3,931
higher for women than for men ($3,649). This is also reflected in Figure 3 which shows how these
21This is equivalent to an average discount of 16.8% which is below the unconditional discount of 47.6% documented
by Adams et al. (2017).
17

numbers have evolved over time and through generations of artists. In Figure 3(a) we can observe
that mean artwork prices tend to be higher for men, whereas median prices (3(c)) appear to be
higher for women after 2002 with a widening gap after 2011. The hedonic price indices based on the
respective time (year) dummies for both genders in Figure A1 in the Appendix show that sales of
female artists have overall outperformed male artists (1(a)). However, this seems to be driven by
artists from older generations since the financial performance of contemporary female artist (1(b))
appears to be significantly worse than the performance of contemporary male artists.
Paintings are with 42.3% the most frequent object type in our data set for both genders while
Photographs are the least frequent object type. Mean artwork prices are significantly lower for women
for Paintings and Works on Paper while median prices are only lower for Prints and Multiples.
In terms of national residency, it is noteworthy that mean artwork prices of female artists are
slightly higher in North America ($58,929 versus $58,234) and significantly higher in Eastern Europe
($68,258 versus $40,758). Only in Western Europe median prices for female artists are lower than
for male artists.
With respect to living status, the share of artworks by deceased artists is lower for the female
sub-sample (64.9%) than for the male sub-sample (78.5%). Furthermore, artworks by both living
and deceased female artists fetch significantly higher mean and median prices than artworks by male
living and deceased artists.
Lastly, Tables A1 and A2 in the Appendix provide an overview of the top 25 male and female
artists and reveal some first insights on the rank of female artists in the market. With a sales value
of $393 million the highest selling female artists, Joan Mitchell, does not even reach the total sales
value of any of the male artists in the top 25.
In summary, the univariate analysis reveals three important facts about gender differences in the
secondary art market. First, with a share of less than 4% female artists are extremely underrepre-
sented but relatively more concentrated in terms of sales value in the secondary art market. Second,
although mean prices are less median prices appear to be higher for women. This might be indicative
of a selection mechanism where a higher bar is applied to female artists admitting only the most
talented ones. Third, it appears that those women, who do break through the initial barrier to the
market, still lag behind top male artists in terms of sales value. In the following section, we will
perform an in-depth multivariate analysis in order to investigate the performance of female artists
18

in the secondary art market.
4 Empirical Analysis
4.1 Auction Participation
Starting at the art gallery level, we want to investigate whether female artists are less likely to enter
the secondary art market than men. As shown in the descriptive statistics, female artists are highly
underrepresented in the secondary market with a share of less than 4%. At the same time, it is
reported that the number of female students pursuing MFA degrees is not below the number of men.
This indicates that there appears to be a large drop out rate of women between these two career
stages. However, it might be the case that not all students attending fine art schools are interested
in pursuing professional careers as artists. Therefore, in order to make conclusions with respect to
the mobility of female artists, we need to observe the share of women present in the primary market
where less established and younger artists are represented by galleries.
If an artist is present in the primary data set, it means that this artist is represented by at
least one gallery during the sample period. Having gallery representation is the first crucial step in
an artist’s career after the completion of an art education program. A gallery provides the artist
with access to its network of buyers as well as marketing activities to improve his or her visibility
in the art market. While galleries can represent emerging as well as more established artists, good
representation is particularly important for new, unknown artists. Reasons for selling an artwork at
auction can be liquidity related or in order for its owner to realize a financial return following a value
appreciation. Hence, if an artist is not traded at auction it might be suggestive of an insufficient
value appreciation and/ or demand to make it attractive enough for its owner to sell.
In order to determine how many male and female artists move from the gallery to the auction
market, we check whether the artists in the contemporary primary market sample are also present
in our main (auction) sample of living artists. Table 3 shows that out of 4,180 male artists, 96.9%
(4,050 artists) can also be found in the secondary market sample. However, only 93% (534 artists)
out of the 574 female artists made this transition. The difference in proportions test is statistically
significant at a 1% level. It is also notable that the share of women decreases from 13.7% in the
primary market to 11.6% in the secondary market within this sample. This amounts to a drop
19

of 15%. The result of this univariate analysis provides us with a first evidence that female artists
progress slower to the secondary market. In order to analyze the mobility of women in a multivariate
setting, we employ a probit model on the entire primary market sample with a binary dependent
variable indicating whether an artist from the gallery sample is traded at auction. The model takes
the following form:
Zj=α1+δ1Dj+λ1Aj+1j,(3)
where Zjis a binary variable that takes the value 1 if an artist jparticipates at auction and 0 oth-
erwise from the population of Na= 4754 living artists in our sample. Djdenotes the discrimination
coefficient which is a gender dummy taking the a value 1 whenever the respective artists is a woman.
Ajis a 1 ×92 vector that denotes the artist characteristics including the artist’s nationality22, his
or her year of birth as well as a dummy for every gallery an artist is represented by as a gallery’s
reputation is known to have a high impact on an artist’s success.23
The result of the probit model is shown in Table 4. It provides evidence for a small but statistically
significant barrier for female artists at the transition from the primary into the secondary market. The
presented coefficients are the marginal effects at the mean. The coefficient on the female dummy
indicates that female artists are 2.2% less likely to participate at auction compared to men. For
example, a female artist might have interrupted her artistic career which would negatively impact
the market demand for her existent artworks. This can also be the result of self-selection whereby
female artists decide to cease their artistic endeavors in anticipation of less success. It could also be
the case that galleries underinvest in female artists as they estimate their likelihood to succeed to be
lower. An alternative explanation might be that buyers of female artworks differ from other other
buyers with respect to their buying motive. These buyers could prefer to hold on to their purchases
being less interested in realizing financial returns. Indeed, art dealers prefer selling to collectors
who agree to not sell (“flip”) the artwork after a short period of time at auction in order to avoid
unforeseen price fluctuations. This is in particular important for emerging artists who do not have a
22Nationality is defined on country level and includes all countries in Europe and North America totaling to 53 countries.
Due to collinearity concerns, 5 of these nationalities were included in the regression model.
23In order to avoid overparameterization, galleries that represented less than 100 artists were subsumed under the
category ‘others’. This resulted in 23 gallery dummies. Due to collinearity concerns, 9 of these galleries were included
in the regression model.
20

price history. In some instances art dealers will offer to buy back the artwork in order to avoid that
it is flipped at auction at a low price (Velthuis, 2007). If maintaining price stability is more crucial
for female artists galleries might select different types of buyers for their female lots. However, in
this case we should observe a price premium on female contemporary artworks conditional on them
being traded at auction as a sale through auction should only take place if the risk of the hammer
price being below the gallery price level can be ruled out.
Overall, the magnitude of the chance with which female artists are less likely to transit from the
primary into the secondary market (-2.2%) appears to be low. However, it is worth keeping in mind
that this estimate is on the lower bound given that we are only considering Western contemporary
artists. It is also worth mentioning that Artnet’s gallery network does not capture the whole pop-
ulation of galleries. Our sample consists of more successful artists as being a member of an online
gallery network requires resources smaller galleries might not have. As a result, it might be the
case that the artists in our gallery sample represent the top of the market, experience a large value
appreciation and have a larger chance to progress into the secondary market.
4.2 Performance at Auction
In the previous section we have found that based on our gallery sample female artists are less likely
or slower to progress from the primary into the secondary market. We will now turn to our main
(auction) sample to investigate the overall performance of female artists on artwork level at auction.
The basic regression model has the following specification,
log Pit =α+ψWi+βXi+ηHi+τt+it , i = 1, . . . , N;t= 1, . . . , T ; (4)
where log Pit indicates the log of the real price of an artwork, i, which is sold at a given time t.
N= 2,677,190 artworks in our sample over T= 72 seasons between 2000 and 2017. Widenotes the
discrimination coefficient which is a gender dummy taking a value of 1 whenever the respective artists
of a given artwork, i, is a woman. This regression specification estimates the differences between
the actual sales price for an artwork of a female artist and the value of an artwork by a male artist
with the same characteristics. All artwork characteristics are captured in Xi, a 1 ×276 vector that
includes the object type (the base category is paintings), the auction house where it was sold and the
21

size of the artwork.24 Hiis a 1 ×5 vector that denotes the artist characteristics of a given artwork,
i, including region of the artist’s nationality (the base category is North America) 25 and a dummy
for the living status of the artist at the time of the transaction (the base category is deceased). Due
to collinearity between the artist names and the gender dummy, we exclude artist fixed effects from
the regression in our main analysis. τrepresents time fixed-effects for the years 2000 until 2017. ψ,
βand ηare time-independent parameters. αis a constant term. Lastly, it denotes the error term.
Table 5 reports the regression results when estimating parametrs using the OLS methodology.
The highly statistically significant female dummy coefficient in our base regressions shows that art-
works by female artists are on average 4.4% more expensive than the artworks of male artists given
the quality of the artworks. While this difference appears to be rather small, this depicts merely
the average effect. It is also consistent with findings by Bertrand and Hallock (2001) who studied
gender salary differentials for the case of top executives. This result implies that there is a premium
on artworks created by women which is supportive of evidence for the presence of a selection mech-
anism whereby female artists that make it to the secondary art market are on average better than
male artists. It could also be indicative of a potential supply squeeze. Due to the limited supply of
high-quality female artworks, collectors are willing to pay a premium for these rare lots.
All other coefficients are in line with expectations. Sculptures are the most expensive objects,
while prints and multiples display the highest discount relative to paintings. Artworks of artists
from Southern Europe sell highest. This is not surprising given that many of the top artists such
Picasso, Modigliani, Miro and Fontana originate from there. Lastly, there is a premium on deceased
artists (due to a fixed and established market values). The R-squared of the regression is 0.42 which
is within the usual range for hedonic models in the field of art market economics (Ashenfelter and
Graddy, 2002). All coefficients remain unchanged independent of whether the nominal or the real
artwork price is used as the dependent variable.
Male and female artists in different time periods were subject to different conditions especially
with respect to access to education and the general acceptance of women as creators of cultural goods
and part of the workforce. Assuming societal barriers as the only source of performance difference
between men and women, our base regression results might pick up unobserved quality differences
24In total, there are 1,522 auction houses in our data set. Due to collinearity concerns we subsumed auction houses
below the 90th quantile in terms of number of transactions under “other”. This resulted in 270 different categories.
25All countries are split into five regions: North America, Eastern Europe, Northern Europe, Southern Europe and
Western Europe.
22

between the artworks produced due to unequal opportunities granted to women. For instance, less
support for female artist in the Old Masters movement could have led to an even higher bar for
women during this time period reinforcing a potential selection mechanism. In turn, this would
imply that women who succeeded to pursue careers as professional artists had to be on average
better than male artists resulting in a higher demand for these lots. As opportunities and beliefs
held in society with respect to gender roles have shifted throughout time, the selection mechanism
is expected to become less pronounced resulting in a convergence in the supplied quality between
artworks produced by men and women in later time periods. As a result, any observed performance
differences at auction between the genders are expected to be to a lesser extent due to factors related
to differences in access to opportunities.
In order to test this hypothesis, we estimate the model specified in equation (4) for each artistic
movement separately. The last four specifications in Table 5 present the results. Interestingly, the
coefficient on the female dummy is positive and statistically highly significant for each movement
with the exception of Contemporary art where we observe a negative and statistically significant
coefficient. The Post War era yields the largest premium (14.9%) for female artworks. While it is
difficult to imagine that opportunities were worse for female artists in the mid-20th century than in
the mid-19th century, it could be the case that this era produced a small number of female artist
considered large superstars (e.g. Agnes Martin, Helen Frankenthaler and Joan Mitchell) for whose
lots competition among buyers is high. The discount on contemporary female lots is more likely
to be driven by a gender bias present in the market given the improved opportunities for women
pursuing an artistic career which is also manifested in a relatively higher proportion of women in
this period (9.3%). However, it is also indicative of a lower bar and a larger demand for female art
whereby also artworks of lesser quality enter the secondary market.26
Even though our data set contains artworks created by artists in different time periods, all sales
of these works take place over a time period of about 18 years (2000 to 2017). While we do not
expect large shifts in the market attitude towards female artists or strong differences in the quality
of the artworks by men and women available in the market, we are interested in investigating the
persistence of the difference in performance found in our baseline regression in Table 5 over time.
26To further homogenize our sample, we also consider every cohort of artists separately and run regressions for each
generation of artists whereby one generation is defined as a time period of 25 years. The results are presented in Table
A3 in the Appendix. Consistent with the previous results, we observe a premium on female lots for the generations
active before the year 1850 and a discount for more recent generations born after 1950.
23

Therefore, we split our data into four different time periods for which we run separate regressions.
The results are shown in Table 6. For all four periods a premium for female lots persists ranging
from 1.9% to 7.4%. The premium appears to be smaller for the years after 2010. However, this
might be due an increased supply of artworks in the market in later years by female artists (see
Figures 2(a) and 2(c)). Since we observed a discount on female Contemporary lots (Table 6), Table
7 separately reports the results for the sub-sample of Contemporary artworks for the four time
periods. Interestingly, it appears that the discount on female lots intensifies throughout time. While
the marginally statistically significant discount amounts to 3.5% in the period from 2000 until 2004, it
increases to 12.6% for the years from 2015 until 2017. These results also reject the participation rate
hypothesis which states that extreme performance outcomes are less likely for women for statistical
reasons as they are fewer in number. It provides an intriguing explanation why women are excluded
from top ranks in occupations with a high concentration of men. However, as our results show,
increasing the ratio of women or lowering their barriers to entry does not defeat a gender gap in
performance. This suggests that these additional female artists fall into the lower price quantiles
lowering their average performance which could be interpreted as a sign of lower barriers for female
artists.
A frequent reason underlying the gender gap is found to be the woman’s child-rearing responsibil-
ity (Reis, 1995a,b). In order to more closely investigate this potential explanation for a difference in
performance we estimate the model specified in equation (4) for the sample of contemporary artists
and include a binary variable which is equal to one if the artists is 40 years old or above at the time
of the transaction. We also interact this variable with our female dummy. Indeed, Table A4 in the
Appendix shows that being above 40 years old has a positive effect on price and that this effect is
stronger for women than for men. Motherhood and the risk of a career break might be a potential
concern for collectors.
Another possible explanation for the difference in performance of female and male artists might
be due to a lower productivity of women. A certain supply of artworks needs to circulate in the
market in order to satisfy demand in case of a price appreciation. While we do not have an overview
over all artworks ever created by every artists in our sample, we compared the amount of artworks
sold by each artist per year.27 Table 8 shows that the mean annual sales volume is a lot higher for
27These are typically listed in a catalogue raisonn´e which would need to be obtained for every artist on our sample.
24

male artists (105 versus 38). However, this appears to be driven by some male artists who have an
extremely large trading volume (e.g. Picasso) since median male and female yearly sales volumes are
very close to each other (14 versus 16). Furthermore, in terms of trading frequency female artists
appear slightly more active than male artists with an average of 165 days (versus 177 days for men)
in between two consecutive sales.
A better way to control for the unobserved quality characteristics of the artworks which are
not explicitly captured by our hedonic variables, would be to modify the dependent variable (the
artwork price) by dividing it by the mid-point of the auction house pre-sale estimate.28 This way, we
would analyze whether gender can explain the auction house’s estimation error which is equal to the
deviation of the final hammer price from the auction house price estimate. However, this assumes
that auction house estimates are unbiased measures of quality. We believe that this is unlikely
given that auction house experts incorporate buyer preferences and tastes in their valuations of the
artworks. In addition, there are over 1,500 auction houses in our sample with diverging valuation
procedures. The summary statistics in Table A5 in the Appendix show that the auction house
estimation error is on average higher for male artworks (2.43 versus 1.78). Proportionally, the share
of undervalued artworks are slightly higher for women than for men (and vice versa). For robustness,
we performed a regression using the model specification in equation (4) with the nominal price scaled
by the auction house pre-sale estimate as the dependent variable. The results of the OLS regression
can be found in Table A6 in the Appendix. The female dummy coefficient together with most of
the other coefficients on our hedonic variables becomes statistically insignificant. This means that
auction houses do not systematically over- or underestimate the artworks in relation to the final
hammer price and potentially account for a gender bias among buyers.
Even though Nochlin (1971) argues that topics are more correlated within artistic periods as
opposed to within gender, other potential explanatory variables that could be correlated with the
female dummy are certain colors or themes. For instance, it could be the case that female artists
are more likely to focus on family themes in their artworks. These topics might in turn be valued
higher or lower by the market. Furthermore, the artist’s identity is known to be the strongest
predictor of artwork prices capturing important not directly observable quality characteristics such
as an artist’s reputation. However, given that gender is a time-invariant characteristic of an artists,
28Before an auction takes place, auction houses typically publish a catalogue listing all lots that will be for sale with
their own estimated value of these lots.
25

a gender dummy cannot be included together with artist fixed-effects in the same regression model.
In order to incorporate the information contained in the artist’s identity we repeat the regression
model specified in equation (4) with artist fixed-effects instead of the gender dummy.29 We then
test for differences in the means of the distribution of male and female artist fixed-effect coefficients.
Figure 4 shows the density distribution of the male and female artist coefficient. The graph clearly
highlights the dominance of the female distribution of the fixed-effect coefficients. The two-sample
Kolmogorov-Smirnov test for equality of distribution (unreported) provides statistically significant
evidence that the male artist’s coefficient distribution is smaller than the female artist’s coefficient
distribution.30
Overall, the analysis in this section showed that there is an average premium on female artworks
which might be the consequence of a higher bar and a limited supply of high-quality female lots.
This effect is in particular driven by artworks of older generations where different possibilities for
men and women prevailed. At the same time, we observe an increasing discount over time for
Contemporary female lots which suggests lower barriers for women in recent time periods. Given
the underrepresentation of female artists and the superstar effect in the art market, our conjecture
is that the premium observed for older generations of artists is driven by a small number of female
artists whose artworks demand very large prices. These top artists could be causing a supply squeeze
as in a “winner-take-all” market demand will be concentrated around these few individuals. Further,
a potentially skewed distribution of sales is not taken into account by OLS estimation which focuses
on the average effect. Therefore, the next section will take a closer look at the distribution of sales
for male and female artists.
4.3 Distribution of Rewards
The analysis so far showed that the share of female artists decreased as they moved from the primary
into the secondary market. The shortage of supply of high quality female artworks at auction
is reflected in their outperformance in terms of auction prices. We hypothesize that this average
effect results from a supply squeeze for the most popular female artists who attract the highest
29Due to computational limitations we only allocate individual artist dummies to artists for which there are at least
45 sales transactions. This corresponds to the 25th percentile in the sales volume distribution on artist level. All other
artists are subsumed under two IDs (one for male and one for female artists). This results in 9,584 distinct male and
366 distinct female artists.
30The results of this test are available upon request.
26

demand. These women constitute the superstars in the “winner-take-all” market and drive the
observed average premium. Results on artwork level are distorted if a large amount of artworks is
sold by a small amount of female artists located on the top of the price distribution.
Table 9 depicts the distribution of lots of male and female artists. Overall, we can see that a
smaller number of female artists accounts for a relatively larger amount of lots in terms of volume
and value than it is case for male lots. For instance, 95% of the female lots sold at auction stem from
36.7% of female artists accounting for 99.5% of the overall value of female lots. For the male sample,
95% of all lots are covered by 40.0% of the male artists who absorb 99.0% in terms of artwork value.
Given the difference in the concentration between the male and the female market, the artwork level
OLS is likely to be not informative about the true performance of female artists at auction.
In order to correct for the large number of lots by the most expensive artists, we estimate equation
(4) using weighted-least-squares (WLS). The applied weights equal the inverse of the square root
of the total number of artworks sold per artist at auction throughout the sample period. This
transformation results in an equal weight for every artists in our sample. The results of the WLS
regression are presented in Table 10. As expected, the female dummy coefficient turns negative and
now yields an average discount of 10% on female lots given the characteristics of the artworks. Again,
this result remains unchanged when the nominal price is employed as the dependent variable. This
finding lends supportive evidence for the conjecture that the positive coefficients on artwork level
derived in the previous section is due to the presence of a small number of very popular female artists
who attract the largest demand and the highest prices from collectors in a “winner-take-all” market.
As in the previous section, we perform a robustness check for the different sub-samples consisting
of the four artistic movements. The discount on female lots observed in base specification seems to be
driven by Post War and Contemporary lots which yield a female dummy coefficient of -13.1% and -
4.2% respectively. While the price discount becomes smaller compared to the OLS baseline regression
in Table 5 for Contemporary female artists, it flips the sign for the case of women attributed to the
Post War era. It appears that particularly in this movement buyers compete for a very small number
of female superstars (such as Joan Mitchell and Agnes Martin) with a large discount for the average
female artist active during this period. The Post War era is characterized by abstract geometric
forms which might more likely be associated with male attributes. The relative reduction in the
magnitude of the discount for contemporary female lots implies that this movement is in particular
27

dominated by a number of women with a large amount of lower priced artworks. However, even
if we take this distribution into account, the discount persists. The female dummy coefficient for
Modern lots is negative but statistically indistinguishable from zero. Female Old Masters lots retain
a premium, however the coefficient is only statistically significant at the 10% level. Thus, it appears
that the premium found in the previous section is likely to be driven by a supply squeeze for a few
superstar female artists in the case of Post War art while it was due to a higher quality standard in
the case of Old Master works.
As an additional robustness check and in order to account for sales of extraordinary magnitude,
we split our sample into a sub-sample including only mega transactions (defined as artworks that
yielded above $1 million at auction) and a second sub-sample which excludes these large transactions.
The regression results for the baseline OLS model and the WLS model can be found in Tables A7
and A8 in the Appendix. It seems that the observed discount for female artworks is driven by a
number of male superstars who are responsible for a large number of high priced artworks. When
adjusting for the large amount of artworks sold by these artists (WLS regression) the performance gap
between men and women becomes statistically insignificant. In unreported results we also performed
a Wilcoxon-Mann-Whitney test based on the average auction price for male and female artists. This
non-parametric test does not rely on a normal distribution of the dependent variable. The result
suggests that there is a statistically significant difference between the underlying distributions of the
average price of male and female artists. The sum of the female ranks was lower while the sum of the
male ranks was higher than expected. Thus the male group had higher rank. However, the difference
in the distribution of average prices is not statistically significantly different between the two groups
for the sub-sample of contemporary artists.31
Further investigating the concentration within the female segment of the market. Table 11 shows
the percentages of male and female artists at every quantile of the sales value distribution on artist
level. As defined in the section above, the sales value equals the sum of the value of all sold lots
throughout the sample period per artist. The most interesting observation is that the female sub-
market is more concentrated at the top (top 10%) and less concentrated at the bottom (bottom
50%) than the male sub-market. The latter effect becomes more amplified the further we move
down the sales value distribution. While an expected share of 10.1% (5,650) of the male artists can
31The results of these tests are available upon request.
28

be found in the top 10% of the sales value distribution, only 7.5% (178) of the female artists are
located there. At the 50th quantile of the sales value distribution, only a total of 38.9% of the female
artists can be found as opposed to an expected share of 50.6%. Moving further down the sales value
distribution, 9.7% of all male and as many as 15.9% of the female artists are situated at the bottom
10% of the sales value distribution. Overall, this implies that female artists are more likely be found
at the bottom in the sales distribution than men. The superstar effect wherein a small number of
individuals absorbs all industry rewards (Rosen, 1981) applies even more to the female sub-group
than to the male segment.
4.4 The Superstar Effect
If being traded in the secondary market is not sufficient to signal quality and legitimacy for female
artists, we expect that this should materialize in a glass ceiling on the top of the market where the
largest rewards are concentrated.
Table 11 does not only exemplify the concentration of sales within the female sub-segment, but
also provides first critical evidence for a barrier for female artists at the top of the market. In
the 99.97th quantile of the sales value distribution no single female artist can found. This quantile
corresponds to a market share of 40% in terms of value which entirely accrues to a core of 40 top male
artists. As the most expensive female artist, Joan Mitchell, can be found in the 99.96th quantile.
With $393 million in total sales, she is ranked 43rd in the list of top artists.
Furthermore, Table 12 shows for different quantiles of the sales value distribution the respective
brackets for male and female artists as well as the number of artists per bracket. The key takeaway
is that the sales value is significantly lower for female artists than for male artists in every quantile
with the exception of 99th quantile where the sales value bracket is $9.3 million for male artists
and $12,.4 million for female artists. However, at the very top, namely at the 99.91th quantile, the
sales value per artist for men elevates again above the sales value level of women. Moving from $9.3
million in the 99th quantile to $176.8 million in the 99.1th quantile, represents a sizable jump. At
this sales level 99 male artists and a mere of 5 female artists can be encountered. While the overall
1:20 male-to-female ratio is preserved at this quantile, the increase in sales values is disproportional.
It represents the part of the distribution where the superstars of the art market are located who
absorb the largest chunk of the rewards. Lastly, we compare the distribution of maximum (record)
29

artwork prices achieved by male and female artists. Figure 5 highlights how in particular at the top
of the price distribution men overshoot women. This univariate artist level analysis shows that in
order to reach the sales level of male artists, a woman needs to be at the top of the distribution. At
the same time, she is precluded from entering the league of the superstars of the art market which
appears to be reserved for the male population of artists.
In the following step, we aim to investigate whether taking into account the skewed distribution
of prices in the art market will open up a more granular view on gender differences in the art market.
Capitalizing on our comprehensive data set that allows us to dig into the tails of the price distribution,
we estimate parameters of equation (4) with a quantile regression technique as laid out by Koenker
and Bassett Jr (1978). Quantile regression models consider every price segment separately focusing
on parts of the distribution other than the conditional expectation. Table 13 provides an overview of
the number of male and female artworks located in every price quantile. Only 67 artwork by women
lie above the 99.9th quantile yielding prices above $4.4 million. Table 14 presents the regression
results and offers a very clear perspective on gender effects in our sample which are in line with the
findings from the univariate analysis. For illustration, the female dummy coefficient is plotted in
Figure 6. At the 25th quantile, we observe a premium on female artworks which steadily increases
from 4.5% to 6.7% in the 95th quantile. A discount of 1.8% emerges for artworks by female artists at
the 99th quantile and amplifies to 9.1% at the 99.9th quantile, which represents the very top of the
secondary art market. This is supportive evidence for the presence of a glass ceiling that precludes
women from participating in the high-end of the art market. We have repeated the same regression
for the sub-sample contemporary artists. The coefficients presented in Table 15 mirror the results for
the full sample. The discount on female lots becomes more pronounced along the price distribution.
At the top of the distribution there is premium of 3.7% for female lots. This could be regarded as
consistent with a supply squeeze for a small number of female superstars and an amplified superstar
effect among female artists.
Overall, the preceding analysis showed that at the very top of the market the idea of the male
artistic genius still prevails. The “winner-take-all” effect in the art market appears to create a
structure which precludes female artists from reaching the top of the market.
30

5 Concluding Remarks
This is the among first large-scale study to provide a convincing empirical illustration of how a
market structure characterized by the superstar effect reflects on the mobility and performance
of an underrepresented group. Using a sample of data that is only fractionally smaller than the
true population enables us to accurately describe and analyze gender performance differences in the
secondary market for fine art. The analysis goes beyond establishing the average effect of being a
woman on the price of artworks. Instead, we closely look into the upper tail of the sales value and
price distribution where most of the rewards are concentrated.
First, we show that female artists are still highly underrepresented in the primary (gallery) as
well as in the secondary (auction) market. While the share of women in art schools pursuing MFA
degrees is reported to be equal to the share of men, we encounter a proportion of only 13.7% in
our primary art market sample of contemporary artists which decreases to 11.6% in the secondary
market. Overall, across movements and generations, female artists make up a share of less than 4%
in terms of number of artists as well as number of lots. This could be interpreted as supporting
evidence for higher quality standards for female artists at the gallery hiring stage. Further, the
probit model results show that women are 2.2% less likely to progress into the secondary market.
This could be due to a slower process for women in establishing themselves as artists in the market.
Alternatively, it could be because of a difference in motives between the buyers of male and female
art with the latter preferring to hold on to their acquisitions.
In line with a higher bar explanation, we observe an average price permium of 4.4% on artwork
level for female artists which is driven by artists of older generations where opportunties were less
equal presumably allowing only the best and most presistent women to pursue an artistic career.
These findings are consistent with the most recent working paper by Cameron et al. (2017) who find
a premium for female artworks traded at auction within a sample of Yale graduates. A higher bar for
women was also found in recent studies within the area of academia where journal papers abstracts
by women had to be better written than the abstracts of male authors (Hengel, 2017) and science
where patent application by women were found to be subject to higher scrutiny (Jensen et al., 2018).
Furthermore, Bohren et al. (2017) found that women are subject to higher standards in an online
Q&A platform when no task history certifying their reputation is available. At the same time, our
results show that the share of contemporary female artists has increased and that they are subject
31

to a price discount at auction. This might be indicative of lower entry barriers for women in recent
years.
Second, we provide empirical evidence that the superstar effect which is a characteristic of the
art market is more prevalent within the group of female artists than within the male segment. The
observed average price premium for female artworks turns into a 10% price discount after correcting
for the number of lots per artist. This discount is driven by Post War and Contemporary artists
and implies that the observed price premium is due to a small number of female artists who account
for a large share of expensive lots. Additionally, we find that the top end of the market is more
concentrated in the female sub-sample than in the male sub-sample. The women located at the
top of the sales distribution appear to be responsible for the price premium and potentially cause
a supply squeeze for their limited amount of lots. Furthermore, relatively more female than male
artists are located in the lower tail of the value distribution. In every quantile of the distribution,
the total sales value for men is higher than the one for women with exception of the 99th quantile.
This implies that unless a female artist reaches the top, her sales will remain below the sales level
of a male artist in the secondary art market. This has also been shown for the case of women in top
executive positions and for women in higher salary quantiles in general (Bertrand and Hallock, 2001;
Garcia et al., 2001; Kuhn, 1987).
Third, we reveal that the top end of the art market is still dominated by a core number of male
artists. In terms of total sales values, the 99.97th quantile which corresponds to 40% of the market
by value is entirely occupied by male artists where no single women can be found. This is supported
by the quantile regression results which show that within the 99.9th quantile of the price distribution
a discount of 9% for female artworks emerges. This result is in line with empirical findings in other
industries where the superstar effect prevails including the market for top athletes (Kahn, 1991),
movie stars (Bielby and Bielby, 1996; Lincoln and Allen, 2004), high-level executives (Bertrand and
Hallock, 2001) and researchers (Barbezat and Hughes, 2005; Probert, 2005) and might be the result
of the given market structure which prevents the mobility of historically underrepresented groups.
For the case of contemporary artists we observe an increasing discount as we move through the price
quantiles. This could be regarded as evidence for a lower mobility for female artists to the top. It
appears that being traded in the primary or secondary market is not sufficient for a female artist to
signal quality and establish legitimization.
32

Overall, our results suggest that gender still plays an important role in the art market. While
the art industry has grown substantially over time with increasing rewards for artists located at the
top of the market as documented by the recent Art Basel and UBS Art Market Report (McAndrew,
2018), these rewards are still concentrated among a core that purely consist of male artists. This can
induce self-selection mechanism for female artists. The anticipation of lower sales might discourage
women from pursuing professional artistic careers leading them to drop out of the market. Only
female artists with lower opportunity costs and greatest talent may be willing to remain in the
market. While this might be efficient, there is no explanation why the same filter is not applied to
the population of male artists. On the other hand, we show that the number of artworks by women
traded in the secondary market has increased. The price discount on female contemporary lots can
be interpreted as a sign that the bar for female artist to enter the market became lower over the
years and demand for their lots has increased.
Our study provides important lessons for gender differentials in labor market outcomes within
markets characterized by the superstar effect. Our results suggest that an inertia towards existing
market structures can hamper the mobility of historically underrepresented groups. While in other
occupational areas (e.g. orchestra auditions) blinding the identity of the individuals has mitigated
gender inequality, this is less feasible in the art market. Part of an artwork’s value consists from its
historical value which is difficult to detach from an artists’ identity. Therefore, to establish gender
equality, it is important that institutions in the art world deviate from the current paradigm in order.
At the same time such non-conformance needs to be made worthwhile for institutions giving also an
important role to policymakers. Recent trends such as a growing number of women among high-net-
worth-individuals increasing the share of female art buyers, as well media attention and interest in
female artists by major museums might contribute to an organic alleviation of the status-quo. As
more representation of female artists occurs, art dealers and buyers will be more confident about
the quality of artworks created by women so that less reliance on historical group statistics will be
necessary potentially increasing the number of women at the top of the market.
33

References
Adams, R. B., Kr¨aussl, R., Navone, M. A., and Verwijmeren, P. (2017). Is gender in the eye
of the beholder? identifying cultural attitudes with art auction prices. Available at SSRN:
https://ssrn.com/abstract=3083500.
Adler, M. (1985). Stardom and talent. The American Economic Review, 75(1):208–212.
Arrow, K. et al. (1973). The theory of discrimination. Discrimination in Labor Markets, 3(10):3–33.
Arulampalam, W., Booth, A. L., and Bryan, M. L. (2007). Is there a glass ceiling over europe?
exploring the gender pay gap across the wage distribution. International Labour Review, 60(2):163–
186.
Ashenfelter, O. and Graddy, K. (2002). Art auctions: A survey of empirical studies.
Ashenfelter, O. and Graddy, K. (2003). Auctions and the price of art. Journal of Economic Literature,
41(3):763–787.
Banerjee, A. V. (1992). A simple model of herd behavior. The Quarterly Journal of Economics,
107(3):797–817.
Barbezat, D. A. and Hughes, J. W. (2005). Salary structure effects and the gender pay gap in
academia. Research in Higher Education, 46(6):621–640.
Baumol, W. J. (1986). Unnatural value: or art investment as floating crap game. The American
Economic Review, 76(2):10–14.
Becker, G. (1957). The Economics of Discrimination. University of Chicago Press.
Bertrand, M. and Hallock, K. F. (2001). The gender gap in top corporate jobs. Industrial and Labor
Relations Review, 55(1):3–21.
Bielby, D. D. and Bielby, W. T. (1996). Women and men in film: Gender inequality among writers
in a culture industry. Gender & Society, 10(3):248–270.
Bikhchandani, S., Hirshleifer, D., and Welch, I. (1992). A theory of fads, fashion, custom, and
cultural change as informational cascades. Journal of Political Economy, 100(5):992–1026.
Bjerk, D. (2008). Glass ceilings or sticky floors? statistical discrimination in a dynamic model of
hiring and promotion. The Economic Journal, 118(530):961–982.
Blau, F. D. and Kahn, L. M. (2000). Gender differences in pay. Journal of Economic Perspectives,
14(4):75–99.
Bohren, J. A., Imas, A., and Rosenberg, M. (2017). The dynamics of discrimination: Theory and
evidence. Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstractid= 3081873.
34

Bordalo, P., Coffman, K., Gennaioli, N., and Shleifer, A. (2016). Stereotypes. The Quarterly Journal
of Economics, 131(4):1753–1794.
Botelho, T. L. and Abraham, M. (2017). Pursuing quality: How search costs and uncertainty magnify
gender-based double standards in a multistage evaluation process. Administrative Science Quarterly,
62(4):698–730.
Breen, R. and Garcia-Penalosa, C. (2002). Bayesian learning and gender segregation. Journal of
Labor Economics, 20(4):899–922.
Cameron, L., Goetzmann, W. N., and Nozari, M. (2017). Art and gender: Market bias or selection
bias? Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstractid= 3025923.
Cowen, T. (1996). Why women succeed, and fail, in the arts. Journal of Cultural Economics,
20(2):93–113.
Davis, Ben (2016). Is getting an mfa worth the price? https://news.artnet.com/art-world/
mfa-degree-successful-artists-620891.
De Silva, D. G., Gertsberg, M., and Pownall, R. A. (2017). Market evolution of art dealers. Available
at: https://papers.ssrn.com/sol3/papers.cfm?abstractid= 2866949.
Dimson, E., Rousseau, P. L., and Spaenjers, C. (2015). The price of wine. Journal of Financial
Economics, 118(2):431–449.
Dimson, E. and Spaenjers, C. (2011). Ex post: The investment performance of collectible stamps.
Journal of Financial Economics, 100(2):443–458.
Egan, M. L., Matvos, G., and Seru, A. (2017). When harry fired sally: the double standard in
punishing misconduct. National Bureau of Economic Research.
Frank, R. H. and Cook, P. J. (2013). Winner-take-all markets. Studies in Microeconomics, 1(2):131–
154.
Galenson, D. W. and Weinberg, B. A. (2000). Age and the quality of work: The case of modern
american painters. Journal of Political Economy, 108(4):761–777.
Garcia, J., Hern´andez, P. J., and Lopez-Nicolas, A. (2001). How wide is the gap? an investigation
of gender wage differences using quantile regression. Empirical Economics, 26(1):149–167.
Goetzmann, W. N. (1993). Accounting for taste: Art and the financial markets over three centuries.
The American Economic Review, 83(5):1370–1376.
Goetzmann, W. N., Jones, P. W., Maggioni, M., and Walden, J. (2016). Beauty is in the bid of the
beholder: An empirical basis for style. Research in Economics, 70(3):388–402.
Goldin, C., Kerr, S. P., Olivetti, C., and Barth, E. (2017). The expanding gender earnings gap:
evidence from the lehd-2000 census. American Economic Review, 107(5):110–114.
35

Hengel, E. (2017). Publishing while female. are women held to higher standards? Available at:
https://www.repository.cam.ac.uk/handle/1810/270621.
Hilton, J. L. and Von Hippel, W. (1996). Stereotypes. Annual Review of Psychology, 47(1):237–271.
Huang, J. and Kisgen, D. J. (2013). Gender and corporate finance: Are male executives overconfident
relative to female executives? Journal of Financial Economics, 108(3):822–839.
Jensen, K., Kov´acs, B., and Sorenson, O. (2018). Gender differences in obtaining and maintaining
patent rights. Nature Biotechnology, 36(4):307.
Kahn, L. M. (1991). Discrimination in professional sports: A survey of the literature. Industrial and
Labor Relations Review, 44(3):395–418.
Koenker, R. and Bassett Jr, G. (1978). Regression quantiles. Econometrica, pages 33–50.
Kuhn, P. (1987). Sex discrimination in labor markets: The role of statistical evidence. The American
Economic Review, pages 567–583.
Lincoln, A. E. and Allen, M. P. (2004). Double jeopardy in hollywood: Age and gender in the careers
of film actors, 1926–1999. In Sociological Forum, volume 19, pages 611–631. Springer.
Mandel, B. R. (2009). Art as an investment and conspicuous consumption good. The American
Economic Review, 99(4):1653–63.
McAndrew, C. (2018). The Art Market 2018. Art Basel and UBS.
Mei, J. and Moses, M. (2002). Art as an investment and the underperformance of masterpieces. The
American Economic Review, 92(5):1656–1668.
National Endowment for the Arts (2011). Artists and arts workers in the united states research note
no.105. https://www.arts.gov/sites/default/files/105.pdf.
National Museum of Women in the Arts (2017). Get the facts. http://www.nmwa.org/advocate/
get-facts.
Nochlin, L. (1971). Why have there been no great women artists? The Feminism and Visual Culture
Reader, pages 229–233.
Pesando, J. E. (1993). Art as an investment: The market for modern prints. The American Economic
Review, pages 1075–1089.
Phelps, E. S. (1972). The statistical theory of racism and sexism. The American Economic Review,
62(4):659–661.
Pownall, R. A. J. (2017). TEFAF Art Market Report 2017. European Fine Art Foundation.
Probert, B. (2005). I just couldn’t fit it in: Gender and unequal outcomes in academic careers.
Gender, Work & Organization, 12(1):50–72.
36

Reis, S. M. (1995a). Older women’s reflections on eminence: Obstacles and opportunities. Roeper
Review, 18(1):66–72.
Reis, S. M. (1995b). Talent ignored, talent diverted: The cultural context underlying giftedness in
females. Gifted Child Quarterly, 39(3):162–170.
Rengers, M. (2002). Economic lives of artists: Studies into careers and the labour market in the
cultural sector. PhD thesis.
Renneboog, L. and Spaenjers, C. (2013). Buying beauty: On prices and returns in the art market.
Management Science, 59(1):36–53.
Ridgeway, C. L. (2001). Gender, status, and leadership. Journal of Social Issues, 57(4):637–655.
Rosen, S. (1981). The economics of superstars. The American Economic Review, 71(5):845–858.
Scharfstein, D. S. and Stein, J. C. (1990). Herd behavior and investment. The American Economic
Review, pages 465–479.
Stigler, G. J. and Becker, G. S. (1977). De gustibus non est disputandum. The American Economic
Review, 67(2):76–90.
The Guerrilla Girls (2017). The guerrilla girl’s complete chronology. https://www.guerrillagirls.
com/chronology/.
Throsby, C. D. and Mills, D. (1989). When are you going to get a real job? an economic study of
australian artists.
Throsby, D., Hollister, V., et al. (2003). Don’t give up your day job: An economic study of professional
artists in australia. Technical report, Surry Hills, NSW: Australia Council.
Throsby, D., Zednik, A., et al. (2010). Do you really expect to get paid? an economic study of
professional artists in australia. Technical report, Strawberry Hills, NSW: Australia Council for the
Arts.
Velthuis, O. (2007). Talking prices: Symbolic meanings of prices on the market for contemporary
art.
37

Figures
100 150 200
Volume of artworks (in thousands)
2000 4000 6000 8000 10000 12000
Value of artworks (in millions)
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
Value Volume
(a) Number and value of artworks by men by years
0 200 400 600
Volume of artworks (in thousands)
0 5000 10000 15000 20000 25000 30000
Value of artworks (in millions)
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Generation
Value Volume
(b) Number and value of artworks by men by generation
2 4 6 8
Volume of artworks (in thousands)
100 200 300 400 500 600
Value of artworks (in millions)
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
Value Volume
(c) Number and value of artworks by women by years
0 5 10 15 20 25
Volume of artworks (in thousands)
0 200 400 600 800 1000 1200
Value of artworks (in millions)
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Generation
Value Volume
(d) Number and value of artworks by women by generation
The year 2017 is omitted in Figures a) and c) as we only use the first four months of this year. Overall, there were 35,860
artworks by male and 1,787 artworks by female artists in this year. The value of these artworks is $1,521,769,000 and
$53,611,000 respectively. Due to missing data on the year of birth not all artists could not be allocated to a generation.
Figures b) and d) omit these artists. Overall, there are 89,888 artworks by male and 2,199 artworks by female artists
in this omitted category. The value of these artworks is $761,310,000 and $7,780,000 respectively.
Figure 1: Evolution of sales by male and female artists
38

4000 5000 6000 7000 8000 9000 10000
Number of artists
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
(a) Number of male artists by years
0 5,000 10,000 15,000
Number of artists
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Generation
(b) Number of male artists by generation
100 200 300 400 500 600
Number of artists
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
(c) Number of female artists by years
0 200 400 600 800 1,000 1,200
Number of artists
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Generation
(d) Number of female artists by generation
The year 2017 is omitted in Figures a) and c) as we only use the first four months of this year. Overall, there were 6,171
male and 167 female artists in 2017. Due to missing data on the year of birth not all artists could not be allocated to a
generation. Figures b) and d) omit these artists. Overall, 21,748 male and 1,113 female artists could not be allocated
to a generation.
Figure 2: Evolution of number of male and female artists
39

25 30 35 40 45 50
Mean artwork price (in thousands)
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
Males Females
(a) Mean artwork prices by year
0 20 40 60 80 100
Mean price (in thousands)
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Generation
Male Female
(b) Mean artwork prices by generation
3 3.5 4 4.5 5
Median artwork price (in thousands)
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
Males Females
(c) Median artwork prices by year
0 2 4 6 8 10
Median price (in thousands)
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Generation
Males Females
(d) Median artwork prices by generation
The year 2017 is omitted in Figures a) and c) as we only use the first four months of this year. Overall, the mean
(median) value is $42,436 ($3,681) for artworks by male and $30,001 ($4,306) for artworks by female artists in this
year. Due to missing data on the year of birth not all artists could not be allocated to a generation. Figures b) and d)
omit these artists. Overall, the mean (median) value is $8,968 ($1,992) for artworks by male and $3,542 ($1,182) for
artworks by female artists in this omitted category.
Figure 3: Evolution of mean and median artwork prices for men and women
40

0 .1 .2 .3
Density
-4 -2 0 2 4
Artist coefficient
Female artist Male artist
Figure 4: Density distribution of the artist fixed-effect coefficient for male and female artists
0 .05 .1 .15
Density
0 5 10 15 20
Maximum real price
Female artist Male artist
Figure 5: Distribution of maximum prices achieved at auction for male and female artists
41

-.1 -.08 -.06 -.04 -.02 0 .02 .04 .06
Female dummy coefficient
20 30 40 50 60 70 80 90 100
Quantile
Figure 6: Coefficient on female dummy in quantile regression
42

Tables
Table 1: Concentration in the auction market (2000-2017)
Share of market value
Share (number) of artists 50% 75% 90% 99% Total value
All artists 0.07% (80) 0.43% (497) 2.18% (2,563) 19.67% (22,926) $121.4bn
Male artists 0.07% (73) 0.41% (453) 2.16% (2,401) 19.89% (22,065) $117.3bn
Female artists 0.27% (15) 0.89% (50) 2.41% (135) 15.54% (872) $4.1bn
43

Table 2: Summary statistics for men and women
Men Women
Price N N artist mean median sd N N artists mean median sd ∆ in means
Overall 2,572,346 110,938 45,614 3,648 686,070 104,844 5,612 39,065 3,931 330,635 16.8%∗∗∗
Buy-in rate* 0.378 0.376 0.027 0.361 0.361 0.034 4.7%
Movement
Contemporary 388,070 19,917 38,025 3,146 432,129 30,434 2,031 28,502 4,011 170,396 33.4%∗∗∗
Postwar 532,238 34,173 41,047 3,090 635,913 24,280 1,863 54,262 4,121 405,305 -24.4%∗∗∗
Modern 819,923 21,281 51,358 3,542 860,069 34,920 961 41,332 3,701 409,739 24.4%∗∗∗
Old Masters 525,405 19,806 48,503 3,776 678,067 13,781 595 29,921 3,928 223,868 62.1%∗∗∗
& Impressionists
Other 306,710 15,761 38,641 5,743 494,310 1,429 162 34,316 5,418 135,044 12.6%
Object type**
Design 212,709 9,250 12,848 3,269 67,873 11,141 521 19,934 4,009 17,703 -35.5%∗∗∗
Sculptures 169,704 15,306 70,600 5,032 852,476 8,132 807 88,341 11,419 483,140 -20.1%∗∗∗
Paintings 1,132,403 78,184 75,343.8 4,903.5 951,020.9 33,064 3,663 72,025 5,142 486,108 4.6%∗∗∗
Works on paper 453,729 36,161 24,543 3,090 315,271 16,477 1,646 18,470 3,797 93,931 32.9%∗∗∗
Prints and multiples 477,203 15,050 2,117 2,241 172,222 19,371 711 6,630 1,895 100,688 -68.1%∗∗∗
Photographs 126,598 6,822 15,477 3,572 76,292 16,659 603 20,475 5,125 107,243 -24.4%∗∗∗
Region
North America 545,239 24,641 58,234 3,946 803,389 34,751 1,727 58,929 4,525 467,225 -11.8%∗∗∗
Northern Europe 463,192 19,162 29,560 3,033 593,016 25,195 1,310 27,827 3,625 263,459 6.2%
Western Europe 1,099,021 44,143 43,114 3,571 594,711 35,243 1,673 24,473 3,394 140,217 76.2%∗∗∗
Southern Europe 337,164 14,049 57,251 4,246 912,561 4,040 329 25,012 5,695 111,384 128.9%∗∗∗
Eastern Europe 127,730 8,943 40,758 4,247 453,220 5,615 573 68,258 4,258 491,061 -40.3%∗∗∗
Living status
at time of sale
Deceased 2,018,743 65,760 49,159 3,893 748,360 68,033 2,263 44,659 3,941 390,600 10.1%∗∗∗
Alive** 553,603 47,175 32,686 2,864 380,340 36,811 3,454 28,728 3,909 170,957 13.8%∗∗∗
All prices are in constant 2017 $.
*The buy-in rate is the share of lots of all lots offered per artist that is not sold at auction.
In total, 156,761 male lots and 59,258 female lots were bought in.
**Multiple attributions for a single artist are possible.
***The difference in mean prices between men and women is statistically significant on a 1% significance level.
44

Table 3: Summary statistics for men and women: primary market sample
Variables Men Women
N mean sd N mean sd
Auction participation 4,180 (4,050) 0.969*** 0.174 574 (534) 0.930*** 0.255
Total sales value (in $) 4,050 3,381,389 41,400,000 534 1,536,746 8,015,190
Year of birth 4,180 1955 15.622 574 1958 14.990
The primary market sample consists of Western, contemporary artists only.
***The difference in proportions of the auction participation rates between men and women is
statistically significant on a 1% significance level.
All prices are in constant 2017 $.
Table 4: Auction participation - Artist level regression results (primary market)
Variables Auction participation
Probit model
Female -0.022***
(0.006)
Year of birth -0.001***
(0.000)
Artist Nationality Effects Yes
Gallery Effects Yes
Observations 4,754
Standard errors in parentheses. p<0.01, ** p<0.05, * p<0.1.
The probit model shows the marginal effects at the mean.
The primary market sample consists of
Western, contemporary artists only.
45

Table 5: Artwork level OLS regression results
Variables Log of real price
Real price Nominal price Old Masters Modern Post War Contemporary
Female 0.044*** 0.044*** 0.100*** 0.045*** 0.149*** -0.083***
(0.004) (0.004) (0.011) (0.007) (0.008) (0.007)
Design -0.219*** -0.219*** -0.012 -0.199*** -0.261*** -0.168***
(0.003) (0.003) (0.009) (0.006) (0.006) (0.009)
Photographs -0.688*** -0.688*** -0.707*** -0.788*** -0.718*** -0.494***
(0.004) (0.004) (0.014) (0.007) (0.008) (0.007)
Prints & multiples -0.918*** -0.918*** -0.897*** -1.017*** -0.962*** -0.804***
(0.002) (0.002) (0.006) (0.004) (0.005) (0.006)
Sculpture 0.330*** 0.330*** 0.322*** 0.406*** 0.341*** 0.393***
(0.003) (0.003) (0.008) (0.007) (0.007) (0.007)
Works on paper -0.409*** -0.409*** -0.379*** -0.383*** -0.371*** -0.325***
(0.002) (0.002) (0.005) (0.004) (0.005) (0.006)
Eastern Europe 0.014*** 0.014*** 0.441*** 0.168*** -0.528*** -0.359***
(0.005) (0.005) (0.010) (0.007) (0.011) (0.012)
Northern Europe -0.272*** -0.272*** -0.228*** -0.130*** -0.497*** -0.057***
(0.003) (0.003) (0.008) (0.006) (0.006) (0.007)
Southern Europe 0.149*** 0.149*** 0.107*** 0.539*** -0.228*** -0.085***
(0.003) (0.003) (0.010) (0.006) (0.007) (0.008)
Western Europe -0.043*** -0.043*** 0.010 0.120*** -0.284*** -0.100***
(0.003) (0.003) (0.006) (0.005) (0.005) (0.006)
Alive -0.381*** -0.381*** -0.370***
(0.002) (0.002) (0.004)
Log of size 0.181*** 0.181*** 0.186*** 0.144*** 0.188*** 0.240***
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Observations 2,677,190 2,677,190 539,186 854,843 556,518 418,504
R-squared 0.422 0.419 0.420 0.417 0.437 0.483
Year Effects Yes Yes Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
The base category for the object type is paintings.
The base category for the region is North America.
46

Table 6: Artwork level OLS regression results – by time period
Variables Log of real price
2000-2004 2005-2009 2010-2014 2015-2017
Female 0.074*** 0.069*** 0.019*** 0.039***
(0.009) (0.007) (0.006) (0.013)
Design -0.425*** -0.207*** -0.200*** -0.186***
(0.021) (0.007) (0.005) (0.011)
Photographs -0.945*** -0.722*** -0.563*** -0.570***
(0.009) (0.007) (0.006) (0.014)
Prints & multiples -1.175*** -1.074*** -0.736*** -0.661***
(0.005) (0.004) (0.004) (0.009)
Sculpture 0.406*** 0.444*** 0.289*** 0.190***
(0.008) (0.007) (0.005) (0.011)
Works on paper -0.439*** -0.448*** -0.378*** -0.346***
(0.005) (0.004) (0.004) (0.008)
Eastern Europe -0.016 0.140*** 0.000 -0.138***
(0.012) (0.008) (0.007) (0.015)
Northern Europe -0.260*** -0.264*** -0.247*** -0.273***
(0.007) (0.006) (0.005) (0.011)
Southern Europe 0.242*** 0.188*** 0.095*** 0.083***
(0.007) (0.006) (0.005) (0.011)
Western Europe 0.012** -0.049*** -0.048*** -0.084***
(0.006) (0.005) (0.004) (0.009)
Alive -0.487*** -0.397*** -0.357*** -0.279***
(0.005) (0.004) (0.003) (0.007)
Log of size 0.201*** 0.204*** 0.167*** 0.147***
(0.001) (0.001) (0.001) (0.002)
Observations 496,923 756,668 1,026,029 209,830
R-squared 0.452 0.443 0.424 0.402
Year Effects Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
The base category for the object type is paintings.
The base category for the region is North America.
47

Table 7: Artwork level OLS regression results – by time period (contemporary sample)
Variables Log of real price
2000-2004 2005-2009 2010-2014 2015-2017
Female -0.035* -0.040*** -0.093*** -0.126***
(0.020) (0.014) (0.011) (0.022)
Design -0.504*** -0.135*** -0.126*** -0.194***
(0.061) (0.019) (0.013) (0.028)
Photographs -0.583*** -0.514*** -0.435*** -0.520***
(0.018) (0.013) (0.010) (0.023)
Prints & multiples -1.061*** -0.993*** -0.659*** -0.654***
(0.015) (0.011) (0.009) (0.018)
Sculpture 0.488*** 0.471*** 0.399*** 0.271***
(0.021) (0.015) (0.011) (0.022)
Works on paper -0.249*** -0.313*** -0.309*** -0.375***
(0.017) (0.012) (0.009) (0.018)
Eastern Europe -0.286*** -0.221*** -0.384*** -0.489***
(0.040) (0.023) (0.017) (0.033)
Northern Europe -0.091*** 0.056*** -0.093*** -0.102***
(0.019) (0.014) (0.010) (0.021)
Southern Europe 0.050** 0.033** -0.211*** -0.120***
(0.024) (0.016) (0.012) (0.025)
Western Europe 0.034** -0.055*** -0.165*** -0.141***
(0.017) (0.012) (0.009) (0.019)
Log of size 0.257*** 0.265*** 0.234*** 0.205***
(0.004) (0.003) (0.002) (0.004)
Observations 44,731 106,980 185,573 43,267
R-squared 0.538 0.496 0.498 0.443
Year Effects Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
The base category for the object type is paintings.
The base category for the region is North America.
48

Table 8: Frequency of trading at auction
Variables Men Women
mean median sd max mean median sd max
Average number of sales (per year) 105 14 329 2,995 38 16 55 398
Average time between consecutive sales (in days)* 177 42 402 6,245 165 33 398 5,951
*Excludes artists who had only one transaction (amounting to 31.6% of male and 41.1% of female artists.
Table 9: Concentration of lots for male and female artists
Share of lots Men Women
Share of artists Share of value Share of artists Share of value
25% 0.30% (336) 51.69% 0.37% (21) 23.62%
50% 2.09% (2,316) 78.40% 1.66% (93) 65.65%
75% 8.87 % (9,838) 93.39% 6.56% (368) 93.36%
90% 23.96% (26,585) 97.99% 20.31% (1,140) 98.77%
95% 39.95% (43,318) 98.99% 36.72% (2,061) 99.54%
110,938 117,335,262,644 5,612 4,095,761,313
All prices are in constant 2017 $.
49

Table 10: Weighted least squares regression results
Variables Log of real price
Real price Nominal price Old Masters Modern Post War Contemporary
Female -0.100*** -0.100*** 0.049* -0.010 -0.131*** -0.042***
(0.010) (0.010) (0.029) (0.021) (0.020) (0.016)
Design 0.229*** 0.229*** 0.500*** 0.458*** 0.230*** 0.318***
(0.011) (0.011) (0.034) (0.023) (0.019) (0.023)
Photographs -0.376*** -0.376*** -0.183*** -0.360*** -0.347*** -0.202***
(0.011) (0.011) (0.041) (0.029) (0.022) (0.015)
Prints & multiples -0.672*** -0.672*** -0.758*** -0.522*** -0.564*** -0.503***
(0.009) (0.009) (0.026) (0.017) (0.019) (0.020)
Sculpture 0.437*** 0.437*** 0.649*** 0.604*** 0.348*** 0.513***
(0.010) (0.010) (0.023) (0.021) (0.018) (0.016)
Works on paper -0.342*** -0.342*** -0.392*** -0.219*** -0.271*** -0.195***
(0.006) (0.006) (0.012) (0.011) (0.013) (0.013)
Eastern Europe 0.247*** 0.247*** 0.560*** 0.290*** 0.260*** -0.013
(0.012) (0.012) (0.028) (0.023) (0.023) (0.022)
Northern Europe -0.003 -0.003 -0.104*** -0.116*** 0.072*** 0.002
(0.009) (0.009) (0.019) (0.023) (0.016) (0.019)
Southern Europe 0.493*** 0.493*** 0.405*** 0.194*** 0.541*** 0.084***
(0.011) (0.011) (0.024) (0.025) (0.021) (0.021)
Western Europe 0.153*** 0.153*** -0.023 -0.063*** 0.265*** -0.038**
(0.008) (0.008) (0.017) (0.017) (0.016) (0.017)
Alive -0.359*** -0.359*** -0.132***
(0.005) (0.005) (0.009)
Log of size 0.138*** 0.138*** 0.202*** 0.133*** 0.095*** 0.175***
(0.002) (0.002) (0.004) (0.004) (0.003) (0.003)
Observations 2,677,190 2,677,190 539,186 854,843 556,518 418,504
R-squared 0.362 0.354 0.378 0.316 0.332 0.356
Year Effects Yes Yes Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
The base category for the object type is paintings.
The base category for the region is North America.
The weights are based on the number of observations
per artist.
50

Table 11: Quantiles by total sales value for men and women
Men Women
Quantile Total sales value ($) N artists Cumulative N artists Cumulative
>99.97% 0.03% (40) 0.03% 0.00% (0) 0.00%
<99.97% 452,388,320 0.01%(17) 0.05% 0.02% (1) 0.02%
<99.96% 351,808,064 0.04% (43) 0.08% 0.05% (3) 0.07%
<99.1% 176,461,520 0.90% (994) 0.98% 1.19% (67) 1.27%
<99% 9,461,848 4.05% (4,490) 5.03% 3.06% (172) 4.33%
<95% 982,622 5.09% (5,650) 10.12% 3.17% (178) 7.50%
<90% 312,493 15.24% (16,908) 25.36% 10.23% (574) 17.73%
<75% 50,209 25.19% (27,949) 50.56% 21.19% (1,189) 38.92%
<50% 8,604 24.92% (27,644) 75.48% 26.60% (1,493) 65.52%
<25% 2,089 14.82% (16,442) 90.30% 18.55% (1,041) 84.07%
<10% 814 4.90% (5,435) 95.20% 6.99% (392) 91.05%
<5% 545 4.80% (5,326) 100.00% 8.95% (502) 100.00%
Total sales value 121,431,023,957
Table 12: Group-specific quantiles for men and women
Men Women
Quantile Total sales value ($) N artists Total sales value ($) N artists
>99.91% 99 5
<99.91% 176,750,048 1,010 135,153,952 51
<99% 9,342,266 4,437 12,382,016 224
<95% 992,138 5,547 724,759 281
<90% 318,364 16,641 168,534 842
<75% 51,854 27,735 23,281 1,403
<50% 8,801 27,734 4,557 1,403
<25% 2,147 16,640 1,288 841
<10% 831 5,546 581 281
<5% 554 5,549 403 281
Total sales value 117,335,262,644 4,095,761,313
51

Table 13: Quantiles for artwork prices for men and women
Quantile Price ($) No. of male artworks No. of female artworks
>99.9% 2,607 67
<99.9% 4,382,047 23,099 998
<99% 606,033 102,447 4,641
<95% 103,541 512,642 22,582
<75% 12,808 643,428 26,084
<50% 3,930 643,252 26,016
<25% 1,512 644,871 24,456
Table 14: Quantile regression results
Variables Log of real price
q25 q50 q75 q95 q99 q99.9
Female 0.045*** 0.049*** 0.054*** 0.067*** -0.018*** -0.091***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Auction House Effects Yes Yes Yes Yes Yes Yes
Region Effects Yes Yes Yes Yes Yes Yes
Alive Dummy Yes Yes Yes Yes Yes Yes
Size Effects Yes Yes Yes Yes Yes Yes
Artist Effects No No No No No No
Observations 2,677,190 2,677,190 2,677,190 2,677,190 2,677,190 2,677,190
P-values based on bootstrapped standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Table 15: Quantile regression results - Contemporary artists sample
Variables Log of real price
q25 q50 q75 q95 q99 q99.9
Female -0.042*** -0.062*** -0.090*** -0.094*** -0.120*** 0.037
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Auction House Effects Yes Yes Yes Yes Yes Yes
Region Effects Yes Yes Yes Yes Yes Yes
Alive Dummy Yes Yes Yes Yes Yes Yes
Size Effects Yes Yes Yes Yes Yes Yes
Artist Effects No No No No No No
Observations 418,504 418,504 418,504 418,504 418,504 418,504
P-values based on bootstrapped standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
52

Appendix
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
Index
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
Years
Males Females
(a) Index by gender - Full sample
.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
Index
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
Years
Males Females
(b) Index by genders - Contemporary sample
Figure A1: Price index by gender
.03 .035 .04 .045 .05
Female/Male ratio
2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
Figure A2: Evolution female-to-male ratio
53

Table A1: Top 25 male artists by value of sales
Rank Artist Origin Total sales Total sales Buy-in
value in $ volume rate
(market share) (market share)
1 Pablo Picasso Southern Europe 5,853,551,616 (4.99%) 37,386 (1.45%) 0.247
2 Andy Warhol North America 4,931,258,880 (4.20%) 19,028 (0.74%) 0.310
3 Claude Monet Western Europe 2,509,770,496 (2.14%) 493 (0.02%) 0.197
4 Gerhard Richter Western Europe 2,128,574,336 (1.81%) 3,587 (0.14%) 0.255
5 Francis Bacon Northern Europe 2,071,435,648 (1.77%) 1,372 (0.05%) 0.235
6 Alberto Giacometti Western Europe 1,661,223,808(1.42%) 1,991 (0.08%) 0.309
7 Jean-Michel Basquiat North America 1,604,688,384 (1.37%) 1,308 (0.05%) 0.288
8 Mark Rothko North America 1,589,495,040 (1.35%) 142 (0.01%) 0.184
9 Henri Matisse Western Europe 1,384,500,224 (1.18%) 5,157 (0.20%) 0.302
10 Roy Lichtenstein North America 1,365,195,904 (1.16%) 6,429 (0.02%) 0.247
11 Amedeo Modigliani Southern Europe 1,282,909,952 (1.09%) 502 (0.58%) 0.344
12 Marc Chagall Western Europe 1,246,740,480 (1.06%) 14,957 (0.57%) 0.294
13 Joan Mir´o Southern Europe 1,195,891,584 (1.02%) 14,781 (0.21%) 0.285
14 Willem De Kooning North America 1,144,317,696 (0.98%) 1,272 (0.06%) 0.272
15 Lucio Fontana Southern Europe 1,098,615,296(0.94%) 2,772 (0.11%) 0.266
16 Alexander Calder North America 1,088,666,752 (0.93%) 5,479 (0.05%) 0.238
17 Pierre-Auguste Renoir Western Europe 1,046,396,352 (0.89%) 3,766 (0.15%) 0.309
18 Zao Wou-Ki Western Europe 1,015,000,512 (0.87%) 4,045 (0.15%) 0.206
19 Fernand L´eger Western Europe 1,005,042,112 (0.86%) 2,978 (0.16%) 0.354
20 Cy Twombly North America 850,141,376 (0.72%) 881 (0.06%) 0.765
21 Jeff Koons North America 848,892,096 (0.72%) 1,646 (0.12%) 0.296
22 Paul C´ezanne Western Europe 791,902,080 (0.67%) 697 (0.05%) 0.299
23 Edgar Degas Western Europe 771,783,232 (0.66%) 1,274 (0.17%) 0.295
24 Ren´e Magritte Western Europe 734,759,296 (0.63%) 1,519 (0.03%) 0.235
25 Damien Hirst Northern Europe 705,134,592 (0.60%) 3,940 (0.03%) 0.406
All prices are in constant 2017 $.
54

Table A2: Top 25 female artists by value of sales
Rank Artist Origin Total sales Total sales Buy-in
value in $ volume rate
(market share) (market share)
1 Joan Mitchell North America 392,962,816 (9.59%) 641 (0.61%) 0.213
2 Georgia O’Keeffe North America 211,702,064 (5.17%) 117 (0.11%) 0.204
3 Louise Bourgeois North America 197,968,512 (4.83%) 649 (0.62%) 0.289
4 Agnes Martin North America 193,711,040 (4.73%) 296 (0.28%) 0.249
5 Cindy Sherman North America 140,606,176 (3.43%) 1,269 (1.21%) 0.268
6 Barbara Hepworth Northern Europe 135,153,952 (3.30%) 616 (0.59%) 0.146
7 Tamara De Lempicka Eastern Europe 127,470,128 (3.11%) 313 (0.30%) 0.357
8 Natalia Sergeevna Goncharova Eastern Europe 127,109,512 (3.10%) 731 (0.70%) 0.463
9 Mary Cassatt North America 88,247,688 (2.15%) 832 (0.79%) 0.296
10 Helen Frankenthaler North America 79,406,904 (1.94%) 1,100 (1.05%) 0.253
11 Bridget Riley Northern Europe 78,610,368 (1.92%) 818 (0.78%) 0.189
12 Berthe Morisot Western Europe 76,978,256 (1.88%) 258 (0.25%) 0.340
13 Eileen Gray Northern Europe 75,399,800 (1.84%) 184 (0.18%) 0.326
14 Gabriele M¨unter Western Europe 67,722,952 (1.65%) 449 (0.43%) 0.231
15 Niki De Saint Phalle Western Europe 67,633,304 (1.65%) 1,849 (1.76%) 0.361
16 Maria Helena Vieira Da Silva Western Europe 62,461,532 (1.53%) 683 (0.65%) 0.320
17 Elisabeth Frink Western Europe 56,816,528 (1.39%) 1,212 (1.16%) 0.186
18 Camille Claudel Western Europe 47,351,292 (1.16%) 115 (0.11%) 0.275
19 Julie Mehretu North America 39,050,448 (0.95%) 117 (0.11%) 0.328
20 Marie Laurencin Western Europe 37,916,940 (10.93%) 1,633 (1.56%) 0.452
21 Germaine Richier Western Europe 36,489,668 (0.89%) 207 (0.20%) 0.310
22 Charlotte Perriand North America 36,297,372 (0.89%) 1,270 (1.21%) 0.367
23 Sonia Delaunay Western Europe 35,823,440 (0.87%) 2,414 (0.23%) 0.412
24 Zinaida Evgenievna Serebryakova Eastern Europe 46,413,028 (0.87%) 130 (0.12%) 0.272
25 Elizabeth Peyton North America 34,532,152 (0.84%) 305 (0.29%) 0.343
All prices are in constant 2017 $.
55

Table A3: Artwork level OLS regression results - by generation of artist
Variables Log of real price
<1700 <1800 <1825 <1850 <1875 <1900 <1925 <1950 <1975 <2001
Female 0.358*** 0.058 0.125*** 0.340*** 0.008 -0.008 0.124*** 0.080*** -0.105*** -0.097***
(0.069) (0.041) (0.031) (0.024) (0.013) (0.008) (0.008) (0.008) (0.011) (0.033)
Design -0.668*** -0.292*** 0.045 0.049*** -0.012 -0.171*** -0.248*** -0.312*** -0.068*** 0.180***
(0.092) (0.048) (0.041) (0.017) (0.011) (0.007) (0.007) (0.009) (0.016) (0.055)
Photographs -0.708* -0.469*** -0.231*** -0.775*** -0.689*** -0.787*** -0.658*** -0.756*** -0.406*** -0.461***
(0.371) (0.105) (0.020) (0.026) (0.017) (0.010) (0.008) (0.008) (0.010) (0.037)
Prints and multiples -1.267*** -1.441*** -1.154*** -0.999*** -0.851*** -1.003*** -1.010*** -1.025*** -0.742*** -0.738***
(0.013) (0.012) (0.019) (0.014) (0.008) (0.005) (0.005) (0.005) (0.010) (0.042)
Sculpture 0.225*** 0.061*** 0.118*** 0.352*** 0.354*** 0.400*** 0.452*** 0.314*** 0.483*** 0.103**
(0.026) (0.019) (0.017) (0.014) (0.011) (0.008) (0.008) (0.007) (0.012) (0.042)
Works on paper -0.601*** -0.467*** -0.402*** -0.475*** -0.348*** -0.390*** -0.355*** -0.367*** -0.248*** -0.326***
(0.013) (0.009) (0.010) (0.009) (0.006) (0.005) (0.005) (0.006) (0.010) (0.038)
Eastern Europe 0.392* 0.119** 0.567*** 0.513*** 0.427*** 0.134*** -0.182*** -0.499*** -0.431*** -0.399***
(0.201) (0.049) (0.031) (0.020) (0.012) (0.009) (0.010) (0.012) (0.019) (0.046)
Northern Europe 0.060 -0.537*** -0.659*** -0.209*** -0.224*** -0.145*** -0.221*** -0.476*** 0.058*** -0.197***
(0.189) (0.020) (0.018) (0.013) (0.010) (0.008) (0.007) (0.007) (0.010) (0.039)
Southern Europe 0.491*** -0.085*** -0.183*** 0.174*** 0.089*** 0.607*** 0.027*** -0.294*** -0.168*** -0.300***
(0.188) (0.021) (0.027) (0.018) (0.013) (0.007) (0.007) (0.008) (0.014) (0.059)
Western Europe 0.500*** -0.299*** -0.327*** 0.055*** -0.008 0.081*** -0.022*** -0.335*** -0.212*** -0.322***
(0.188) (0.020) (0.017) (0.011) (0.008) (0.006) (0.006) (0.006) (0.010) (0.038)
Alive -0.079*** -0.322*** -0.659***
(0.005) (0.004) (0.013)
Log of size 0.133*** 0.224*** 0.229*** 0.206*** 0.182*** 0.147*** 0.182*** 0.222*** 0.287*** 0.220***
(0.003) (0.002) (0.003) (0.003) (0.002) (0.001) (0.001) (0.001) (0.002) (0.008)
Observations 125,023 103,448 103,135 171,749 343,970 612,586 521,910 437,473 159,572 11,237
R-squared 0.455 0.440 0.415 0.445 0.413 0.419 0.412 0.481 0.533 0.589
Year Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1
The base category for the object type is paintings. The base category for the region is North America.
56

Table A4: Artwork level regression results – child-rearing
Variables Log of real price
OLS WLS
Female -0.140*** -0.103***
(0.025) (0.037)
Above 40 0.153*** -0.041**
(0.009) (0.016)
Female x Above 40 0.067** 0.073*
(0.026) (0.041)
Design -0.172*** 0.319***
(0.009) (0.023)
Photographs -0.497*** -0.203***
(0.007) (0.015)
Prints & multiples -0.808*** -0.503***
(0.006) (0.020)
Sculpture 0.391*** 0.513***
(0.007) (0.016)
Works on paper -0.327*** -0.195***
(0.006) (0.013)
Eastern Europe -0.347*** -0.014
(0.012) (0.022)
Northern Europe -0.053*** 0.001
(0.007) (0.019)
Southern Europe -0.086*** 0.085***
(0.008) (0.021)
Western Europe -0.099*** -0.037**
(0.006) (0.017)
Log of size 0.241*** 0.175***
(0.001) (0.003)
(0.118) (0.140)
Observations 418,504 418,504
R-squared 0.483 0.357
Year Effects Yes Yes
Season Effects Yes Yes
Auction house Effects Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
This sample includes contemporary artists only.
The above-40 is a dummy variable that equals one if the
artist is above 40 years old at the time of the transaction.
The base category for the object type is paintings.
The base category for the region is North America.
57

Table A5: Summary statistics- Auction house estimation error
mean median sd no. undervalued artworks no. overvalued artworks
Male artworks 2.43 1.03 328.7 1,229,644 (47.8%) 1,068,275 (41.5%)
Female artworks 1.78 1.05 126.7 52,931 (50.5%) 42,444 (40.5%)
The estimation error is defined as the artwork sales price divided by the mid-point
of the auction house pre-sale price estimate.
Table A6: Artwork level OLS regression results - price scaled by estimate
Variables Ratio nominal price to action house pre-sale estimate
Pooled Old Masters Modern Post War Contemporary
Female -0.667 -0.085 0.458 -1.777 -2.268
(1.070) (2.116) (1.932) (2.453) (2.764)
Design 2.915*** -0.353 4.617*** 5.109*** -4.058
(0.907) (1.670) (1.597) (1.956) (3.531)
Photographs 0.043 -0.151 0.794 -0.783 -0.505
(1.036) (2.573) (2.081) (2.373) (2.538)
Prints & multiples 0.905 -0.348 2.055* 0.157 -0.278
(0.654) (1.278) (1.203) (1.572) (2.262)
Sculpture Sculpture 0.952 0.229 1.787 -1.092 1.279
(0.935) (1.610) (1.963) (2.051) (2.755)
Works on paper 0.212 -0.160 0.616 -0.473 -0.600
(0.609) (0.917) (1.111) (1.601) (2.285)
Eastern Europe 1.383 -0.131 2.823 -0.712 -0.697
(1.235) (1.891) (2.037) (3.311) (4.501)
Northern Europe 0.084 0.448 -4.825*** 1.680 3.668
(0.841) (1.408) (1.753) (1.919) (2.641)
Southern Europe 0.055 0.040 -0.875 -1.158 -0.712
(0.880) (1.939) (1.637) (2.068) (3.093)
Western Europe 0.439 0.011 0.887 -0.724 -1.083
(0.686) (1.185) (1.329) (1.613) (2.345)
Alive 1.060** -1.356
(0.529) (1.162)
Log of size -0.232 0.086 0.035 -1.200*** -0.335
(0.148) (0.265) (0.289) (0.350) (0.507)
Observations 2,434,732 479,566 772,853 515,204 392,850
R-squared 0.001 0.000 0.003 0.001 0.004
Year Effects Yes Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
The base category for the object type is paintings.
The base category for the region is North America.
58

Table A7: Artwork level OLS regression results – mega transactions
Variables Log of real price
All All Contemporary Contemporary
Only mega transactions Excl. mega transactions Only mega transactions Excl. mega transactions
Female -0.184*** 0.046*** -0.179** -0.053***
(0.033) (0.004) (0.073) (0.007
Design -0.317*** -0.192*** -0.316 -0.146***
(0.068) (0.003) (0.250) (0.009)
Photographs -0.333*** -0.627*** -0.361*** -0.445***
(0.074) (0.004) (0.075) (0.006)
Prints & multiples -0.265*** -0.869*** -0.350*** -0.759***
(0.060) (0.002) (0.131) (0.006)
Sculpture -0.028 0.318*** 0.050 0.385***
(0.022) (0.003) (0.041) (0.007)
Works on paper -0.229*** -0.381*** -0.315*** -0.300***
(0.025) (0.002) (0.081) (0.006)
Eastern Europe 0.004 0.021*** 0.058 -0.337***
(0.034) (0.004) (0.146) (0.011)
Northern Europe 0.076*** -0.256*** -0.101** -0.052***
(0.025) (0.003) (0.045) (0.007)
Southern Europe 0.136*** 0.142*** -0.268*** -0.078***
(0.022) (0.003) (0.064) (0.008)
Western Europe -0.281*** -0.368*** 0.197*** -0.104***
(0.019) (0.002) (0.041) (0.006)
Alive -0.487*** -0.397***
(0.005) (0.004)
Log of size 0.084*** 0.169*** 0.096*** 0.228***
(0.005) (0.001) (0.012) (0.001)
Observations 15,881 2,661,309 2,270 416,234
R-squared 0.095 0.410 0.138 0.472
Year Effects Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Mega transactions are defined as transaction above $1,000,000 in real 2017 USD.
The base category for the object type is paintings.
The base category for the region is North America.
59

Table A8: Artwork level WLS regression results – mega transactions
Variables Log of real price
All All Contemporary Contemporary
Only mega transactions Excl. mega transactions Only mega transactions Excl. mega transactions
Female -0.049 -0.100*** -0.103 -0.041***
(0.074) (0.010) (0.080) (0.016)
Design 0.022 0.232*** -0.107 0.319***
(0.163) (0.011) (0.164) (0.023)
Photographs -0.170** -0.369*** -0.119 -0.199***
(0.068) (0.011) (0.074) (0.015)
Prints & multiples 0.032 -0.666*** 0.231 -0.502***
(0.238) (0.009) (0.310) (0.020)
Sculpture 0.016 0.438*** 0.016 0.511***
(0.069) (0.010) (0.060) (0.016)
Works on paper -0.059 -0.337*** -0.308*** -0.194***
(0.059) (0.006) (0.076) (0.013)
Eastern Europe -0.150** 0.248*** 0.083 -0.011
(0.066) (0.012) (0.121) (0.022)
Northern Europe -0.237*** 0.001 -0.062 0.002
(0.072) (0.009) (0.069) (0.019)
Southern Europe -0.133** 0.488*** -0.068 0.085***
(0.066) (0.011) (0.090) (0.021)
Western Europe -0.029 0.153*** -0.085 -0.037**
(0.055) (0.008) (0.059) (0.017)
Alive -0.306*** -0.356***
(0.049) (0.005)
Log of size 0.053*** 0.137*** 0.089*** 0.174***
(0.013) (0.002) (0.019) (0.003)
(0.137) (0.040) (0.342) (0.138)
Observations 15,881 2,661,309 2,270 416,234
R-squared 0.153 0.360 0.137 0.356
Year Effects Yes Yes Yes Yes
Season Effects Yes Yes Yes Yes
Auction house Effects Yes Yes Yes Yes
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Mega transactions are defined as transaction above $1,000,000 in real 2017 USD.
The base category for the object type is paintings.
The base category for the region is North America.
60

Table A9: Quantile regression on OLS residual
Variables OLS residual
q25 q50 q75 q95 q99 q99.9
Female 0.033*** 0.055*** 0.096*** 0.079*** -0.068*** 0.031
(0.000) (0.000) (0.000) (0.000) (0.000) (0.689)
Observations 2,677,190 2,677,190 2,677,190 2,677,190 2,677,190 2,677,190
P-values based on bootstrapped standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
The residual is based on a linear regression of the object type, auction house,
season, region, alive dummy, artists dummies and size are regressed on the
logarithm of the artwork price .
61