Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Université Libre de Bruxelles - Solvay Brussels School of Economics and Management Centre Emile Bernheim de Recherche Interdisciplinaire en Gestion
ULB CP114/03 50, avenue F.D. Roosevelt 1050 Brussels BELGIUM [email protected] - Tel : +32 (0)2/650.48.64
CEBRIG Working Paper
The Ultimate Coasian Commitment: Estimating and Explaining Artist-
Specific Death Effects
Heinrich W. Ursprung, Katarina Zigova To extract part of their monopoly rent, Coase (1972) famously claimed that durable goods monopolists require some institutional device that allows them to restrict their output stream in a credible manner. We empirically test this proposition by applying it to the production of visual art. The ultimate commitment device in artistic production is the artist’s death. As living artists cannot commit to a pattern of restrained production, the prices of artwork increase when the artist dies. We identify with the help of a toy model the drivers of this so-called death effect and estimate individual death effects of a sample of famous visual artists who died between 1985 and 2011. Using data from art auctions that took place in a narrow bandwidth around the artists’ death, we apply several variations of the classical regression discontinuity design. The heterogeneity in death effects across artists turns out to be substantial. Up to 40% of the variation can be explained by age and reputation at death.
Keywords Coase conjecture, art auction prices, death effect, reputation, regression discontinuity. JEL Classifications C21, L12, Z11.
CEBRIG Working Paper N°21-013 August 2021
0
The Ultimate Coasian Commitment: Estimating and Explaining Artist-Specific Death Effects
Heinrich W. Ursprung* Katarina Zigova*
March 30, 2021
Abstract
To extract part of their monopoly rent, Coase (1972) famously claimed that durable goods monopolists require some institutional device that allows them to restrict their output stream in a credible manner. We empirically test this proposition by applying it to the production of visual art. The ultimate commitment device in artistic production is the artist’s death. As living artists cannot commit to a pattern of restrained production, the prices of artwork increase when the artist dies. We identify with the help of a toy model the drivers of this so-called death effect and estimate individual death effects of a sample of famous visual artists who died between 1985 and 2011. Using data from art auctions that took place in a narrow bandwidth around the artists’ death, we apply several variations of the classical regression discontinuity design. The heterogeneity in death effects across artists turns out to be substantial. Up to 40% of the variation can be explained by age and reputation at death.
JEL-Code: C21, L12, Z11 Keywords: Coase conjecture, art auction prices, death effect, reputation, regression discontinuity
*) University of Konstanz, Germany. We thank Kathryn Graddy, Arye Hillman, Tommy Krieger, Stephan Maurer, Jörn-Steffen Pischke, Guido Schwerdt and Tom Stanley for valued comments. We also thank Luc Renneboog for letting us discuss our project with him on a visit to Tilburg University in December 2018. We received helpful feedback from conference participants at the 20th International ACEI Conference on Cultural Economics in Melbourne 2018, the Ninth European Workshop on Applied Cultural Economics in Copenhagen 2019, and seminar participants at the University of Konstanz and Jadavpur University (Kolkata). An earlier version of this study circulated as “Diff-in-Diff in Death: Estimating and Explaining Artist-Specific Death Effects”, CESifo Working Paper No. 8181.
1
(B)y way of a preamble I want you to note that a great artist has never been acknowledged until after he was starved and dead. This has happened so often that
I make bold to found a law upon it.”
Mark Twain: Is he living or is he dead?
1. Introduction
Ronald Coase’s 1972 note on “Durability and Monopoly” has unleashed a large theoretical
literature, which is still going strong. Some scholars who have contributed to this literature
motivated their efforts by claiming to reject or qualify the so-called “Coase Conjecture” that is
supposed to maintain that durable goods monopolists are unable to commit to restricted production,
which would enable them to capture part of the monopoly rent by raising prices above marginal
costs. Even though Coase did entertain this possibility as a limiting case in which market efficiency
would be achieved “in the twinkling of an eye (p. 143)”, his real objective was to point out that
monopoly power is much more limited in reality than in the distorted picture drawn in most
textbooks. After having discussed various market settings that may provide durable goods
monopolists with more or less market power, he concedes that “(a) full analysis of this situation
would be very complicated,” but adds that it would “not affect the main contention of this note,
that with durability some contractual or institutional arrangement of the type mentioned earlier may
be … the only way of achieving a monopoly price…” (p. 148). He then goes on to qualify:
“However, these practices have not, in fact, always be feasible. Furthermore, some of the
contractual arrangements will not be enforceable over a long period. In such circumstances, the
competitive outcome may be achieved even if there is but single a supplier (p. 149).”
Coase was thus well aware of the fact that competitive pricing of durable goods supplied by a
monopoly producer are rare cases and that even in the long-run convergence to competitive pricing
is far from a matter of course. This is why one should not interpret the literature deriving from
Coase’s seminal 1972 note as rejections or qualifications of a conjecture that is at any rate not more
than a cardboard-character. That literature rather provides generalizations to new market
2
arrangements or technical elaborations that couch Coase’s ideas in terms of modern-day methods.1
Interestingly, Coase’s own basic insights, but also the findings that derive from subsequent
generalizations and technical refinements, are seldom if ever subjected to empirical scrutiny. Two
reasons may explain this lack of empirical research in this area. First, if one acknowledges that
Coase envisaged in his note many market settings that give rise to pricing patterns ranging from
competitive pricing to perfect monopolistic price differentiation, any test would require
exceptionally rich micro datasets of prices, costs, and the institutional details pertaining to the
investigated market environments. Such datasets are hard to come by and persuasive empirical
studies of Coasian dynamics are, accordingly, scarce.2 Second, if one attempted to test the predicted
pricing behavior in a perfectly frictionless market environment that, according to the theory, lets
the initial price settle immediately and permanently at the competitive level, the empiricist is not
likely to find any real-world situation that corresponds to this “ideal” setting. Some scholars have
resorted in this case and variations thereof to laboratory experiments.3
Since we consider the basic idea of Coase’s note to be essentially about the dynamic consequences
of a monopolist’s commitment to a restricted production path, our empirical investigation focusses
on testing the efficacy of credible commitments to restrained production. The visual arts market is,
for various reasons, well suited for such an investigation. Works of fine art are, after all, durable
goods produced under monopolistic competition. The art market has also the advantage that we
can observe a large number of artist-producers, which allows us to discriminate between different
circumstances of production. Most importantly, however, we can exploit the artists’ deaths as
natural experiments that result in ultimate commitments to discontinue production.
We define the death effect as the causal influence of an artist’s death on the price of her works of
art. Given rational art market participants, the death effect works on impact, is discontinuous but
need not be persistent. Since living artists cannot commit to a severely curtailed total oeuvre,
supply-side induced changes in the price of their artwork will not occur as long as their creative
powers and ambitions remain unimpaired. Only when an artist dies, the final size of her oeuvre is
irrevocably determined. When the grim reaper arrives unsuspectedly, the final size of her oeuvre
1 Important early studies include Stokey (1981), Bulow (1982), Gul et al. (1986), Asubel and Deneckere (1989), and Bagnoli et al. (1989); recent contributions are Board and Pycia (2014) and Nava and Schiraldi (2019). 2 Murfin and Pratt (2019) show, for example, that captive financing of capital goods, i.e. financing provided by the manufacturer, helps manufacturers to commit to high resale prices in the future because they have an incentive to protect their collateral. The theoretical study by Bulow (1982) made the same point for leasing durable goods. 3 Güth et al. (1995), von der Fehr and Kühn (1995), Reynolds (2000), Cason and Sharma (2001).
3
is always lower than expected. It is this death-induced curtailment of an artist’s oeuvre, i.e. the
difference between the expected and the actual final size of an artist’s oeuvre, which causes a
sudden increase in the price of her artwork.
We estimate individual death effects of visual artists and explain why these artist-specific death
effects differ across artists. In explaining the size of the death effects, we focus on two
determinants: the artist’s age at death and her reputation in the art scene when she dies. The death
effect is expected to vary negatively with age at death because the older an artist is when she dies
the smaller is the unexpected death-induced decrease in the expected size of her oeuvre and thus
the death effect.4 This asset pricing argument applies to all artists, independent of their reputation.
However, reputation may play a role, especially when a young artist dies. Expected future increases
in reputation may not materialize if death cuts an artist’s career short before she had time to live
up to her potential. Such frustrated expectations will induce a fall in the price of the deceased
artist’s work. An artist’s death effect is thus composed of a positive effect deriving from the
unexpected curtailment of her oeuvre and a negative effect deriving from frustrated expectations
of future reputation gains. Since it does not happen very often that artistic greatness is recognized
late in an artist’s life, we expect the total effect of an artist’s death to be negative only in some
cases of artists who die relatively young.
To test our hypotheses, we focus on a sample of famous visual artists whose artwork sold at
auctions sufficiently often to allow estimating individual death effects. Ready availability of
auction data further restricted us to focus on artists who died between 1985 and 2011. Since art
prices are, in the medium run, subject to non-observable changes in fads and tastes, we only use
auction price data from a relatively narrow time widow around the respective artist’s death and use
regression discontinuity (RD) strategies to estimate the individual death effects. In a second step,
we then go on to test to what extent age at death and reputation (preferably also at the time of the
artist’s death) determine the estimated artist-specific death effects. We conduct these tests by
regressing the individual death effects on those two explanatory variables.
Our results agree well with our theoretical predictions: death effects of artists who enjoyed a full
life are mostly positive and decrease with increasing age at death. For artists who died before their
time, we find that the death effect can be negative if these artists did not already enjoy a firmly
4 A potential demand-side effect may amplify this supply side effect because an older artist has already satisfied a larger fraction of the stock demand than a younger artist has. For a theoretical study producing this result, see Itaya and Ursprung (2016).
4
established reputation when they died. Moreover, we find for short-lived artists that the death effect
increases with increasing reputation. We show that these results are robust to using various
measures of reputation.
Our baseline estimates presuppose that the death of an artist comes as a surprise for the art market.
Technically speaking, we assume that an artist’s death changes the information set of the market
participants – and this is, of course, exactly the kind of change that causes asset prices to jump in
a rational expectations environment. Death, however, does not always come as a surprise. If, for
example, it is publicly known that an artist is terminally ill, her death can often be timed with some
accuracy. These exceptional cases give rise to an additional test of Coasian dynamics. We therefore
examine in some detail the cases of three artists whose death was anticipated well in advance:
Keith Haring, Elisabeth Frink, and Jörg Immendorff. For Haring and Frink we do indeed find clear
price surges at the time when those artists’ dire health status came to the public’s notice and hardly
any suspicious price movements when they actually died.5
Apart from the industrial organization literature that developed in the wake of Coase (1972), our
study fits into the sizable literature on art price formation that is nicely surveyed in Ashenfelter and
Graddy (2006). Ekelund et al. (2017) survey the empirical literature on the death effect. Our study
is most closely related to the study by Ursprung and Wiermann (2011) whose dataset of over
400,000 observations covers the 1980-2005 period and comprises all auction sales of oil paintings,
drawings, and prints reported in Hislop’s Art Sale Index. The empirical strategy employed by
Ursprung and Wiermann uses hedonic regressions that include artist and time fixed effects, a set
of explanatory variables that are commonly used in hedonic art price regressions, and, for recently
deceased artists, dummy variables that are interacted with the age at death. This specification
allows estimating death effects that are contingent on the age at death. The results indeed reveal
death effects that follow the Coasian prediction, i.e. prices jump immediately after the death of the
artists and the estimated death effects vary for older artists negatively with the age at death and for
younger artists positively, i.e. the death effect curve is hump-shaped across age at death. Etro and
Stepanova (2015) reproduced this result by exploiting a marvelous self-collected historic dataset
of almost 90,000 Paris auction sales of paintings sold in the 75 years straddling the periods of
Rococo (1720-1780), Neoclassicism (1770-1840), and Romanticism (1800-1850). Because of the
5 An empirical illustration of the difference between the price effect of an expected death (Haring) and an unexpected one (Basquiat) is to be found in the catalogue of a museum exhibition on artists who died young (Ursprung 2015).
5
long observation period, this study allows identifying sufficiently many repeated sales (about 1.5%
of all recorded transactions) to conduct, apart from the usual hedonic regressions, also repeated
sales regressions.
The hump-shaped pattern of the death effects can be explained if one assumes that an artist’s
reputation increases with the size of her oeuvre and thus with (career) age. If an artist dies young,
her reputation will never reach the level that the art market participants had good reason to expect.
These frustrated expectations will have a negative price effect that may or may not compensate the
positive effect associated with the death-induced reduction in the size of the oeuvre. The trouble
with this argument put forward by Ursprung and Wiermann (2011) is that the correlation between
reputation and age is far from being perfect. Some artists enjoy already substantial reputation at an
early age. Many of those artists are conceptual innovators (Galenson and Jensen 2001), i.e. artists
who work deductively by applying methods that are suitable to immediately transform a given
innovative idea into the preconceived artistic output. Since this method of operation does not
require accumulating expertise by incremental experimentation, conceptual innovation replaces
technical prowess, with the consequence that even very young masters can achieve artistic
reputation. The empirical studies by Galenson and Weinberg (2000 and 2001) indeed show that
successful conceptual innovators produced their most valuable and important work much earlier
than aesthetically motivated experimentalists. These insights clearly show that career age is not an
ideal measure for reputation. We therefore explain in this study the size of the identified death
effects with two variables: age at death and reputation.
The paper unfolds as follows. In section 2, we present a toy model that formalizes the
characteristics of the death effects that we then test empirically in the remainder of the study.
Section 3 describes the criteria for selecting the artists in our sample, presents the art auction data,
elaborates on our measures of artistic reputation and provides an impression of how the auction
prices developed around artists’ death. The estimates of the individual death effects are presented
in section 4. In section 5, we highlight the cases of three artists who were terminally ill. In these
cases, our theory predicts that the death-induced commitment will increase the prices of the work
of the ailing artists before they actually die. Using the entire sample of artists, we then test the
theory-based predictions concerning the size of the death effects in section 6. Section 7 presents
the sensitivity analysis and section 8 concludes.
6
2. A toy model of the death effect
To describe the mechanism driving the death effect in more detail and to suggest determinants of
its heterogeneity, we present a very stylized “toy” model of Coasian dynamics. The narrow focus
of the model derives from our sole aim to highlight the four characteristics of death effects that we
then go on to empirically identify. Those characteristics are the following. First, the death of a
visual artist usually has a positive impact on the price of her artwork. Second, the death effect is
larger for reputed artists than for less reputed ones. Third, the death effect decreases with age at
death and converges to zero for artist dying at a very high age. Fourth, negative death effects may
occur, especially for promising artists who die very young and who have, therefore, been prevented
to live up to the expectations of becoming highly reputed.
Following Coase (1972), we proceed from the observation that the price 𝑝 of a durable good (Coase
uses as his example land and the term “complete durability” to describe the extreme case of
longevity) depends on the available, i.e. final, stock of the good. In our case this final stock is the
artist’s entire oeuvre 𝑋: 𝑝 𝐷 𝑋 . Consider an artist whose artistic production begins at career
age 𝑎 0 and whose career will certainly end if she should reach some high age denoted by 𝑇. We
assume the artistic output accumulates over the artist’s lifetime in a linear manner, i.e. the size 𝑥
of the oeuvre at career age 𝑎 amounts to 𝑥 𝑎. Artistic production is thus not motivated by
commercial motives; artists are modeled as being driven by a creative urge.6 Let, finally,
𝜋 𝑡 denote the probability density function that describes, for an artist who has reached career
age 𝑎, the likelihood of dying at time 𝑎 𝑡.
Under these assumptions, the expected size of the oeuvre of an artist at career age 𝑎 amounts to
𝐸𝑋 𝑎 𝜋 𝑡 𝑡 𝑑𝑡. Without discounting, the price will therefore settle at 𝑝 𝐷 𝐸𝑋 .
The death effect 𝑑 for an artist dying at career age 𝑎 then amounts to
𝑑 𝐷 𝑎 𝐷 𝐸𝑋 .
Figure 1 illustrates the death effect for uniform probability density functions 𝜋 𝑡 𝑇 𝑎
and two different linear demand functions, 𝐷 𝑋 𝐴 𝛽𝑋 𝛽 1 for highly reputed artists and
𝐷 𝑋 𝐵 𝑋 for merely reputed artists. Notice, that X indicates the final stock of an artist’s
oeuvre. The bold lines 𝐷 𝑋 and 𝐷 𝑋 are thus demand functions for the work of deceased artists.
6 A model with endogenous artistic production is to be found in Itaya and Ursprung (2016).
7
High reputation implies that the works of such artists command a higher price than the works of
merely reputed artists, i.e. 𝐷 𝑋 𝐷 𝑋 . Moreover, the demand function for artwork of highly
reputed artists is steeper than the demand function for artwork of merely reputed artists because
the artwork of highly reputed artists is less substitutable than the artwork of minor artists: a Monet
is a Monet and a painting by some minor French impressionist is just another impressionistic
painting. The prices 𝑝 𝑎 𝐴 𝑎 𝑇 charged for the artwork of highly reputed, living
artists at career age 𝑎 (whose total artistic output so far amounts to 𝑥 𝑎) and the corresponding
prices 𝑝 𝑎 𝐵 𝑎 𝑇 for the artwork of the merely reputed living artists are indicated in
Figure 1 by the bold grey lines.
Consider now a highly reputed artist dying at career age 𝑎 . Immediately before her death, the price
for her artwork amounted to 𝑝 𝑎 , after her death it becomes apparent that the final stock of her
artwork turned out to be much smaller than expected (𝑎 instead of 𝑎 𝑇 /2) and the price
increases on impact to 𝐷 𝑎 . The death effect amounts to 𝑑 𝛽 𝑇 𝑎 /2, decreases with
increasing age 𝑎 at death, and disappears for artists who die after their productive period of life at
the age 𝑎 𝑇. For a merely reputed artist, dying at the same career age 𝑎 , the price increases from
𝑝 𝑎 to 𝐷 𝑎 and the death effect amounts to 𝑑 𝑇 𝑎 /2 which is smaller than the death
effect of the more reputed artist who is dying at the same stage of her career. In Figure 1, the fat
arrows LM and BC indicate those two death effects.
To portray a negative death effect, we need to make an additional set of assumptions. Assume that
one can identify those young artists who just may have what it takes to eventually become highly
reputed. Those who do not show that kind of promise will follow in Figure 1 the path ABCD.
Promising young artists will become highly reputed with a (small) probability 𝜃, and those who do
not make that cut will join the ranks of merely reputed artists. For simplicity, assume that high
reputation is bestowed or denied at career age 𝑎∗. Since we are concerned with individual artists,
that critical age a* should be interpreted to be artist-specific. This implies that up to 𝑎∗, the prices
𝑝 𝑎 of the artwork of promising artists corresponds to the weighted average of 𝑝 𝑎 and 𝑝 𝑎
(indicated in Figure 1 by the dashed line). If a promising young artist now dies before reaching
career age 𝑎∗, say at career age 𝑎 , she will never achieve high reputation (because her oeuvre is
too slight to warrant such an august status) and the prices of her artwork drop after her death from
8
𝑝 𝑎 to 𝐷 𝑎 . In Figure 1, the fat (downward pointing) arrow FG indicates that negative death
effect and the path EFGH describes the price development of such short-lived artists.
Figure 1: The toy model of the death effect
The prices of the work of promising young artists who reach the career age of 𝑎∗ will suddenly
increase from 𝑝 𝑎∗ to 𝑝 𝑎∗ for those lucky few who become highly reputed; most of the
promising young artists will, however, be denied high reputation and the prices of their artwork
will drop to 𝑝 𝑎∗ . Those reputation-induced price jumps are, of course, an artifact of our
assumption that the uncertainty of artistic reputation is resolved at one (artist-specific) point of
time. A more realistic portrayal of the process leading up to an artist’s becoming highly reputed
would certainly smooth out these price jumps. In any event, the price jumps at time 𝑎∗, indicated
in Figure 1 by the thin arrows JK and JA’ are not to be confused with the impact effects following
an artist’s death, which are the only matter of our concern. In our toy model, the price development
of the work of formerly promising artists who actually do become highly reputed follows the path
EJKLMN, and for those formerly promising artists who attain only a status of mere reputation the
prices follow a price path akin to EJA’BCD.
9
3. Data
Artists and Art Auction Data
We rely on art auction data reported by Hislop’s Art Sales Index and its successor, the Blouin Art
Sales Index.7 These auction records are electronically available from 1980 onward.8
Our sample of deceased artists is based solely on data availability. As we employ the regression
discontinuity method, we focus on the availability of auctions in a narrow window around each
sampled artist’s death. In the broadest window, we consider five years before and five years after
her death day, i.e. a time window of ten years. We therefore restricted the sample to artists who
died after 1985 and had a sufficient number of observations in this time window. Of the artists who
died between 1985 and 2011, we selected those who, in the 10-year window, had at least 20 sales
before, and 20 after the artists’ death.9 Those constraints led to the sample size of 106 artists with
a total number of 20,635 auction sales between 1980 and 2016.10 Table A1 in the Appendix lists
those 106 artists and provides some relevant descriptive statistics.
Figure 2 shows the distribution of the years of death (left panel) and the age at death (right panel)
for the 106 sampled artists. In the 24 years of our observation period 1985-2011, the number of
cases of death varies between zero and eight, but in most years between 2 and 6. The distribution
of the age at death is more concentrated: the mean age at death is 80 and the median is 82.11
The average auction prices do indeed increase after death; for 78 of the 106 artists in our sample,
the difference in the 5-year averages before and after death is positive. The size of this average
price difference varies between minus 1.5 and plus 1.4 million US dollars, indicating a sizable
variation across all of our sample artists (see Table A1). To be sure, art prices depend on a multitude
of factors. Before drawing conclusions with respect to the death effect, we therefore need to
account for those hidden effects that include the properties of the artwork, the auction house
7 Hislop’s Art Sales Index, CD-ROM, 2005, edited by Duncan Hislop. Art Sales Index, Ltd., Egham, Surrey, England www.artbusiness.com/revs1205asi.html. Blouin Art Sales Index: www.blouinartsalesindex.com/search.action 8 Hislop auction data cover all auctions of all artists between 1980-mid 2005. As of July 2005 we used Blouin auction data which we collected manually only for the deceased artists in our sample and only for the relevant interval around their death. This was possible thanks to an annual subscription from Bloiun giving us right to collect up to 20,000 entries. 9 In a few cases, we deviated from that rule to arrive at a larger number of artist who died before their time. 10 We excluded 61 auction houses outside Europe and the US because they handled only very few sales of our sample artists’ work. 11 The artist’s death pattern in our sample thus resembles the death pattern in developed countries. For example, in Sweden in 2000, the mean age at death was 80 and the median 83 (Canudas-Romo 2010).
10
handling the sale, and the year of sale. The development of the average raw hammer prices round
the artists’ death is illustrated in the left panel of Figure 3. For the average prices across the entire
sample, a clear price jump is apparent.
Figure 2: Distribution of death years and age at death for the 106 artists
The right panel of Figure 3 shows that in our sample of artists the number of auction sales increases
after the average artists’ death. This is the case for 80 of the 106 artists (see Table A1). On average,
we find eight more auction sales in the five years after an artist’s death than in the five years before.
Two questions arise. First, does this increase in transactions influence the death effect, and if so,
in what direction? If one maintains that the stock of an artist’s oeuvre determines the price for her
work, the answer should be in the negative. If one believes that the flow supply cannot be neglected,
one would expect the increase in transactions to have a dampening effect on the post mortem prices,
which would, in turn, tend to reduce the measured death effects. We will follow up that issue in
our sensitivity section. The second question that arises is whether the artwork sold before and after
the artists’ death do not differ in observable properties. Table 1 provides descriptive statistics of
the physical properties of the auctioned artwork, the auction houses handling the transactions, the
prices, and compares means on the both sides of the death threshold.
11
Figure 3: Mean (geometric) auction price (left panel) and mean number of auctioned artworks (right panel) against normalized year of sale. Means of observation before death are circles, after death squares. Normalized year of sale starts on the day of artist’s death. Year zero is the first full year after the artist’s day of death. Dashed lines are means before and after death.
Table 1: Descriptive statistics of the 20,635 artworks of the 106 deceased artists
Mean Std. dev. Mean before Mean after Difference
Height (cm) 68.97 50.88 71.02 67.61 -3.40***
Width (cm) 70.04 56.15 72.19 68.63 -3.56***
Oil painting 0.45 0.47 0.44 -0.03***
Work on paper 0.42 0.44 0.41 -0.02***
Print 0.12 0.09 0.15 0.06***
Signature 0.82 0.86 0.79 -0.07***
Christie’s 0.28 0.28 0.28 0.002
Sotheby’s 0.28 0.30 0.26 -0.04***
USA 0.48 0.49 0.47 -0.02***
Price (arithmetic) 120,016 1,175,990 78,931 147,143 68,212*** Price (geometric) 13,686 1.72 11,195 15,625 4,431***
Note: The means and differences in the hedonic characteristics are arithmetic means. The differences of the geometric means are smaller and in most cases insignificant, i.e. we show here the largest differences. For the prices (USD hammer prices), we show both types of means to demonstrate that the difference in prices before and after death are robust to price outliers (when using the geometric mean).
12
In terms of dimension, the artwork is quite heterogeneous. The mean height and width amounting
to about 70 cm are well in line with what most scholars find, but the standard deviation of more
than 50 cm indicates a large variety of formats. Almost one half (45%) of the artwork in our sample
are oil paintings, 42% are (unique) works on paper, and the remaining 12% are prints. 82% of the
artwork are signed. The renowned auction houses Christie’s and Sotheby’s account for more than
one half of all sales in our sample and almost one half of all auction sales were conducted in the
United States, the other half in Europe. The differences in the reported means before and after the
artists’ death are in eight of the nine reported characteristics statistically significant, but quite small
substantively. Artworks sold before and after the artists’ death thus appear to be largely of the same
type—a valuable property when applying regression discontinuity. For our purposes, however, it
is more relevant that the characteristics are smooth around the death threshold for each artist. We
therefore tested for jumps in mean differences in the hedonic characteristics also individually. For
73 artists smoothness could not be rejected, as not more than two characteristic changed with the
death threshold.12 Still for some artists the hammer prices may have changed posthumously because
of changing characteristics of the artwork, which may confound with the estimate of the death
effect. In our baseline approach, we therefore control for hedonic characteristics of artworks and
test the sensitivity of the estimated death effects when ignoring those characteristics.
Turning to the prices, the before-and-after death difference in the mean US dollar prices is
statistically and substantively significant. Depending on which type of mean one uses, the
posthumous mean price increase amounts to at least 40%. This significant price increase is,
however, not to be confused with the average death effect. The prices need to be further corrected
by an art price index to account for changes that are characteristic for that market and, of course,
for the hedonic characteristics of the each individual artwork.
Measures of Artistic Reputation
We use the term artistic reputation to indicate that an artist enjoys an acknowledged presence and
notability in the art community. Measuring artistic reputation is not an easy task, but measuring
artistic reputation at a specific point of time, in our case at the time of the artist’s death, turns out
12 We tested altogether 7 hedonic properties for each artist, namely size, i.e. ln(width*height), oil painting, work on paper, signature, Sotheby’s, Christie’s, and USA. In 72% of these tests, we failed to reject the zero.
13
to be a challenge. Given these difficulties, we settled for employing a variety of measures based on
different types of information, such as online sources, print media, and encyclopedias.
Online encyclopedias offer readily available information. We therefore collected from Wikipedia
for each artist the number of languages in which they have entries.13 To obtain a second online-
based reputation measure, we counted the number of books offered by amazon.com in the
subcategories art history and biographies relating to the artists’ lives and artwork. Given that
amazon is selling both new and used books, this measure includes books in stock as well as many
out of print books.
The standard measure of reputation used in the literature grades eminent persons by using word
counts in entries of well-recognized, high-quality encyclopedias. Galenson (2006), for example,
uses this method to grade artists, and Murray (2003) to grades people of extraordinary
accomplishments in science, philosophy, Western music, visual arts, and literature. Those
measures reflect the so-called “test of time” view of reputation. We use for this purpose word
counts in the digital Oxford Dictionary of Art and Artists (Chilvers 2015).14
The main problem with all of those reputation measures is that they do not measure reputation at
the time of the artist’s death. Moreover, artists who died in the 1980s and 1990s, regardless of their
achievements, may be less present in those fora and, since English is the predominant language of
those electronic media, the measured reputation of artists who have closer relations to the Anglo-
Saxon world may be inflated.
To tackle both of these imperfections, we collected obituaries in general interest newspapers and
magazines published in different countries and created three reputation measures based on a word
counts of obituaries. These measures are, in principle, well suited to measure the reputation of an
artist at the time of death. We nevertheless acknowledge that this measure relies, for practical
reasons, on a rather limited number (five) of print media.15
For our first obituary measure, we only use obituaries published in The New York Times, which is
arguably the most cosmopolitan of the five sources we used. The disadvantage of using a single
13 We could, of course, have collected the number of words of an English Wikipedia entry, or any other language entry. These measures are, however, more volatile, while the number of languages available gives a more stable reputation proxy. 14 We used the electronic version of this dictionary. www.oxfordreference.com/view/10.1093/acref/9780191782763.001.0001/acref-9780191782763 15 We collected obituaries published in The New York Times, the German Der Spiegel, the Spanish El País, the French Le Monde, and the British The Independent.
14
source is that doing so might introduce a national bias; the advantage is that it avoids arbitrary
decisions of how to aggregate various inputs. Our second obituary measure makes use of all
obituaries that we collected by simply taking the average of the normalized word counts of the
obituaries published in all five print media, where normalization refers to dividing the length of an
obituary by the maximum obituary length in the respective outlet.16 The last obituary measure is a
dummy variable, top50, assuming the value one for those artists for whom the aggregated obituary
measure is above the median value.
Table 2 reports descriptive statistics for the six reputation measures. For the average artist,
Wikipedia provides information in about 17 different languages. In this contest, Salvador Dalí wins
with 162 different languages. The average number of amazon art books and biographies is 22 and
the maximum number (Andy Warhol leads the pack) is almost 500 available books. The British
artist Henry Moore, best known for his monumental sculptures, is honored with the longest entry
in the Oxford Dictionary of Art, closely followed by Andy Warhol and Marc Chagall. The longest
obituaries commemorate Salvador Dalí in El País, Marc Chagall in Le Monde, Willem de Kooning
in the NYT, Andy Warhol in Der Spiegel, and the American R.B. Kitaj, a preeminent representative
of British Pop Art, in The Independent.
Table 2: Descriptive statistics of reputation measures
Reputation measure Mean Std. Dev. Min Max Number of Wikipedia languages 16.82 22.71 0 162 Number of book items offered by Amazon 22.27 58.58 0 498 Word count in Oxford Dictionary of Art 136.24 172.38 0 724 Word count in NYT obituaries 688.05 784.13 0 3251 Aggregated obituary measure 0.22 0.23 0 1 Obituary dummy (top 50) 0.5 0.5 0 1
Table 3: Correlations among the reputation measures
Number of Wikipedia languages
Nr. of book items
offered by Amazon
Word count in Oxford Dictionary
of Art
Word count in NYT
obituaries
Aggregated obituary measure
Nr. of book items offered by Amazon 0.8141 1 Word count in Oxford Dictionary of Art 0.7717 0.5711 1 Word count in NYT obituaries 0.5601 0.4035 0.5432 1 Aggregated obituary measure 0.7184 0.5131 0.6559 0.8038 1 Obituary dummy (top 50) 0.5063 0.26 0.5205 0.5644 0.788
Note: Cronbach’s alpha of the five reputation measure amounts to 0.76.
16 One potential concern is that the print media may have changed the length of their obituaries in the observation period. We checked for the average length of obituaries in each outlet in the 1980s, 1990s and 2000s and did not find any striking differences.
15
The correlation between our six reputation measures is reasonably high (mostly between 0.5 and
0.8; see Table 3), a finding that is supported by a high value of Cronbach’s alpha of 0.76 for the
logarithms of our six measures. Among the non-obituary measures, the number of Wikipedia
languages seems to correlate best with the other measures, and among the obituary measures, the
aggregated measure does just as well. Simonton (1984) found similar correlations between
different reputation measures for scientists.17 Those high correlations are also in line with the study
by Graddy (2013) which shows that the ranking of artists by de Piles in 1708 still explains price
differences today – and that after three centuries.
Our preferred measure of reputation at death is the aggregated obituary measure because, being
based on several sources, it measures reputation at the time of the artists’ death in a manner that is
not likely to be unduly biased. Moreover, it is well in line with the alternative measures. We use
the other five measures for sensitivity analyses.
Corrected Auction Prices
For each sale 𝑖 recorded in our auction record we know the hammer price 𝑝 in US dollars, the year
𝑡 in which the sale took place, the artist 𝑚 who created artwork 𝑖, and the usual characteristics
provided by the auction house. To arrive at our dependent variable we correct the hammer price in
two ways. First, we divide the hammer price 𝑝 by an art price index 𝐼 and take the log of that
ratio; ln 𝑝 ln . The art market index 𝐼 is based on a large price dataset covering 20th century
art.18 Many studies of art price formation correct the art prices only for inflation. We do not favor
this approach because it clearly ignores price developments that are specific to the art market.
Second, we correct the resulting real log prices ln 𝑝 for a vector of hedonic characteristics 𝑿 of
item 𝑖 by estimating for each artist 𝑚 the following auxiliary regression:
ln 𝑝 𝛽 𝜷𝑿 𝜀 where 𝑖 1, … ,𝑛 1
17 Simonton finds a reliability index of 0.78 based on 23 distinct measures of scientific reputation in a sample of over 2000 scientists spanning several centuries. 18 We use here an art price index based on a planned extension of the study on art market returns by Renneboog and Spaenjers (2013). We received the index in January 2020. We thank Luc Renneboog for sharing the data.
16
The constant 𝛽 is the artist-specific base price level and 𝜷 is a 𝑘 1 vector of parameters
measuring the influence of the 𝑘 hedonic characteristics. The 𝑘 𝑛 matrix 𝑿 comprises the
following variables: the log of the picture’s size (height width), a signature dummy, medium
dummies (oil and work on paper, while print is the reference group), dummies for the auction
houses Sotheby’s and Christie’s (usually interpreted as proxies for the quality of the artwork), and
a dummy for other European auction houses (the reference group being US based auction houses
other than Sotheby’s and Christie’s). We label the residuals of the hedonic regressions (1) as
corrected prices ln �̂� and use them as our preferred dependent variable in our baseline regressions;
the hammer prices ln 𝑝 and the indexed prices ln𝑝 serve as dependent variables in our sensitivity
tests.
Figure 4: Mean corrected auction prices across all artists against the normalized years of sale. Normalized year of sale starts on the day of artist’s death. Year zero is the first full year after the artist’s day of death. Dashed lines are mean corrected prices before and after death
Since we know the exact day of each auction and the exact day of each artist’s death, we can
compute for each auction sale how many days before or after the artist’s death the sale took place,
thereby create a normalized year of sale which starts on the day of artist’s death. In Figure 4, we
plot for our sample of 106 artists an average of the corrected prices against the normalized year of
17
Figure 5a: Highly reputed artists: Corrected auction prices against normalized years of sale. Number to the right of the artist’s name refers to his or her age at death, the price lines before and after the artist’s death indicate the respective median prices. The online appendix reports in Figure A2 the full set of 106 graphs.
18
Figure 5b: Merely reputed artists: Corrected auction prices against normalized years of sale. Number to the right of the artist’s name refers to his or her age at death, the price lines before and after the artist’s death indicate the respective median prices. The online appendix reports in Figure A2 the full set of 106 graphs.
19
sale. Year zero is the first full year after the artist’s day of death.19 The price averages that we use
for this figure are the means across all 106 artists of the mean corrected price for each individual
artist if at least one of the works of that artist was sold in the respective year. Figure 4 reveals an
economically significant discontinuity in auction prices after death; the price increase amounts to
more than 15%.
The kind of averaging used to generate Figure 4 blurs, of course, individual heterogeneity in death
effects. We therefore depict the corrected prices for eighteen individual artists in Figure 5. The
selection of those eighteen artists is to demonstrate the heterogeneity in death effects across age at
death and reputation. In panel 5a, we showcase highly reputed artists: in the upper row, artists who
died young, in the middle row, artists who lived a full life, and in the bottom row, artists who died
at a very old age. In panel 5b we ordered merely reputed artists (as compared to highly reputed
artists) in the same way. We made sure that each row includes at least one non-US American artist.
We use these 18 artists to present our results on individual death effects; the information about all
106 artists is provided in the Appendix (Table A1 and Figure A2).
A visual inspection of the eighteen panels reveals that the corrected prices jump in the death year
for some artists and remain constant for others. Economically significant price increases appear to
be associated with highly reputed artists who died at a relatively young age. This is of course in
line with our hypothesis derived from the Coasian framework.
4. Estimating individual death effects
To estimate the effect of an artist’s death on the price of her artwork, we introduce the treatment
variable
𝐷𝐸 0 if auction day death day 1 if auction day death day (2)
that indicates whether the creator 𝑚 of an auctioned artwork 𝑖 was alive 𝐷𝐸 0 or dead
𝐷𝐸 1 when the artwork 𝑖 was sold at auction.20 The treatment status DE is thus a
19 Those years are thus artist-specific and should not be confused with calendar years. 20 We had three auction sales, which took place exactly on the day of the artist’s death. We excluded them. This is negligible given the large number of auctions we work with, but allows us to write a sharp inequality in equation (2) without losing a noticeable number of observations.
20
deterministic function of time. We apply the sharp regression discontinuity design (Angrist and
Pischke 2014, ch. 4) and estimate the death effect for each artist individually:
ln �̂� 𝛼 𝛾𝐷𝐸 𝑓 𝑎 𝜀 , where 𝑖 1, … ,𝑛 3
ln �̂� denotes the log auction price (corrected for changes in general art prices across time and for
differences in the hedonic characteristics) of the artworks 𝑖 1, … ,𝑛 created by artist 𝑚. To
identify the treatment effect 𝛾, we control for artist-specific slopes with the help of function 𝑓 𝑎 ,
where 𝑎 , the normalized year of sale at the time of the sale of 𝑖, is our forcing variable. 𝑎 is the
difference in years, elapsed since the artist’s death and the artist’s death age; 𝑎 is therefore negative
as long as the artist is alive, zero in the first full year after death, and positive afterwards.21 Since
we work with a very narrow window around the artist’s death, we restrict ourselves to a linear
function 𝑓 𝑎 .
Figures 4 and 5 indicate that the price trend over the normalized year of sale (i.e. our forcing
variable in equation 3) can be different on the two sides of the threshold. We thus consider an
additional specification that can accommodate different time trends before and after the artist’s
death, i.e. a specification with an appropriate interaction term (Angrist and Pischke 2014):
ln �̂� 𝛼 𝛾𝐷𝐸 𝑓 𝑎 𝛿 𝑓 𝑎 𝐷𝐸 𝜀 , where 𝑖 1, … ,𝑛 . 4
Following the recommendation by Lee and Card (2008), we cluster the standard errors on the level
of the running variable 𝑎 in order not to overstate the significance of the death treatment on auction
prices. Table 4 reports the regression results for the eighteen artists already showcased in Figures
5a and 5b. We show the results for three different bandwidths (3, 4, and 5 years) and for the two
specifications (3) and (4), the first assuming the same trend on the two sides of the death threshold,
the second allowing the trends to be different. Our preferred method is to estimate the death effects
by allowing different slopes (specification 4) on the two sides of the threshold (columns 2, 4 and
6) because this strategy is more flexible in accommodating changes in unobservable price
fundamentals that cannot be excluded after an artist’s death. In regression discontinuity regressions
it is generally preferred to work with the smallest possible bandwidths. However, when using small
bandwidths, we are often left with rather few observations, especially for artists whose works are
21 We could have defined 𝑎 on a finer grid, i.e. quarters, months or even days. The auctions are however not evenly spread across the year, with 80% of them taking place from February to April, and from September to November. Thus to have a same corpus of auctions at each value of the forcing variable, we stick with the full years starting from the death day of each artist.
21
not auctioned frequently. It is therefore not surprising that the estimated death effects correlate
between bandwidth 3 and 5 less (0.7) than between bandwidth 4 and 5 (0.87).22 We thus consider
the specification with flexible slopes and a bandwidth of five years (Column 2 in Table 4) to
represent our preferred specification which we discuss in the text.
Among the showcased artists, we estimated the largest death effect for Jean-Michel Basquiat. The
estimate of about 1.55 translates into an impressive death-induced price increase of 370%. In this
sub-sample of eighteen, we estimate for seven artists (five of them highly reputed) statistically
significant positive death effects. In the category of merely reputed artists, we estimate six negative
death effect, however only two of them are statistically significant. For nine artists, we find no
statistically significant death effect.
Turning now to the estimates resulting from our preferred specification of the full sample of 106
artists, Figure 6 illustrates the rough pattern of death effects across duration of life and reputation.
For artists who died relatively young (i.e. before the age of 75), the median of the estimated death
effects amounts to 26% for highly reputed artists, and -7% for merely reputed artists. The
corresponding medians of the artists who lived full lives (i.e. died between 75 and 90) are markedly
smaller at 8% and -20%, and for artists who died at an old age (i.e. after the age of 90) the median
death effect drops slightly to 6% for the highly reputed artists and largely disappears for merely
reputed artists. This pattern agrees, of course, with our theory-based predictions: the death effect
decreases, ceteris paribus, with increasing age at death, disappears for artists who die at a high age,
and is larger for highly reputed artists than for merely reputed ones. The descriptive evidence
illustrated in Figure 6 is also compatible with the prediction that the age-induced decrease in the
death effect will be reputation specific. We will test those hypotheses more thoroughly in Section 6.
The average death effect estimated with our preferred specification (bandwidth 5 years and flexible
trend lines) is positive at 0.07 in our entire sample of 106 artist, i.e. the average death-induced price
increase in our entire sample amounts to 7.3%. The estimates of the individual death effects vary
in precision. If we weight the individual point estimates with the squared roots of the number of
auctions sales in the bandwidth, the average death effects becomes 0.11, which corresponds to an
average price increase of almost 12%. For about half of the artists (about 52%), we estimate
statistically insignificant death effects in all specifications. Among the artists whose death has not
22 We discuss the sensitivity of the estimated death effects in more detail in Section 7.
22
Table 4: Regression discontinuity estimates of the death effect for 18 showcased artists
Bandwidth: 5 years Bandwidth: 4 years Bandwidth: 3 years Artist, age at death a a, DE*a a a, DE*a a a, DE*a (1) (2) (3) (4) (5) (6) Highly reputed artists BASQUIAT, 27 1.588*** 1.548*** 1.650*** 1.589*** 1.592*** 1.588***
(0.137) (0.118) (0.165) (0.135) (0.222) (0.151) WARHOL, 58 0.812*** 0.843*** 0.828* 0.920*** 0.161 0.563**
(0.248) (0.188) (0.381) (0.199) (0.265) (0.148) POLKE, 69 0.486** 0.410** 0.686*** 0.631*** 0.658** 0.667**
(0.186) (0.156) (0.188) (0.165) (0.171) (0.177) DORAZIO, 77 0.308*** 0.272** 0.301** 0.271** 0.336** 0.337**
(0.089) (0.104) (0.087) (0.114) (0.092) (0.099) LEWITT, 79 0.356 0.244 0.295 0.261 0.557** 0.472***
(0.227) (0.141) (0.240) (0.172) (0.209) (0.112) RAUSCHENBERG, 83 0.180 0.188 0.314 0.297 0.331 0.310
(0.398) (0.305) (0.455) (0.353) (0.389) (0.456) VASARELY, 90 0.431*** 0.447*** 0.534*** 0.513*** 0.556*** 0.527***
(0.080) (0.076) (0.069) (0.045) (0.081) (0.055) WYETH, 92 0.236 0.057 0.102 -0.056 0.497 0.131
(0.370) (0.246) (0.476) (0.306) (0.563) (0.217) CHAGALL, 97 0.169 0.204 0.282 0.347** 0.195 0.351**
(0.163) (0.175) (0.185) (0.105) (0.215) (0.113) Merely reputed artists NESBITT, 59 -0.442 -0.428 -0.143 -0.128 -0.225 -0.292
(0.277) (0.268) (0.267) (0.265) (0.393) (0.275) RIZZI, 61 -0.444 -0.265 -1.230* -0.949** -0.882 -1.113***
(0.449) (0.301) (0.627) (0.272) (1.047) (0.088) SCANAVINO, 64 0.074 0.613 0.089 0.511 0.450* 0.259
(0.208) (0.398) (0.230) (0.405) (0.214) (0.295) LORJOU, 77 -0.638* -0.412 -0.414 -0.182 -0.131 0.353
(0.328) (0.332) (0.308) (0.346) (0.317) (0.177) SLOANE, 80 -0.262 -0.595** -0.225 -0.683*** -0.570 -0.986***
(0.389) (0.227) (0.423) (0.184) (0.446) (0.176) GISSON, 82 -0.148 -0.288*** -0.061 -0.300** -0.114* -0.160***
(0.158) (0.060) (0.185) (0.095) (0.046) (0.014) ELLINGER, 90 0.205 0.332** 0.391 0.141 0.552 -0.057
(0.240) (0.105) (0.394) (0.111) (0.428) (0.098) BERMUDEZ, 94 0.610** 0.764*** 0.739** 0.768*** 0.652** 0.648**
(0.254) (0.135) (0.226) (0.139) (0.186) (0.170) ZORNES, 100 -0.177 -0.176 0.070 0.067 -0.043 -0.138 (0.107) (0.113) (0.071) (0.044) (0.074) (0.099)
Notes: Standard errors clustered at the running variable 𝑎 are in parentheses. Odd columns report estimated death effects from Eq. (3), even columns from Eq. (4). We show the full list of results in the appendix (Table A3). * p<0.10, ** p<0.05, *** p<0.01
23
induced statistically significant price changes, we find also superstars such as Salvador Dali and
Marc Chagall. For several artists we estimated statistically significant negative death effects that
are more or less robust across specifications (such as Rizzi and Sloane from our showcase sample).
Larger negative death effects are found for younger or less reputed artists.
Figure 6: Median of the estimated death effects in the sample of 106 artists. Whiskers indicate 90% confidence intervals.
Even though the estimates differ across specifications, the internal coherence of the six types of
estimates is very high with correlations of at least 66% and a Cronbach’s alpha of .95 (see Table
A5). This stability of our estimates across the 106 artists allows us to use one set of estimates (our
preferred specification) when investigating the determinants of the variation in death effects in
Section 6. In Section 7 we present sensitivity tests for all our results using estimates based on
alternative art prices and different estimation methods.
24
5. Anticipated impending death
Regression discontinuity designs go some way towards convincing judicious scholars that causal
effects may be at work -- in our case a causal effect of an artist’s death on the prices of her artwork.
A more telling test of the underlying economic theory would be to show that the public
announcement of an artist’s terminal illness triggers a price increase that anticipates the market
reaction that usually materializes when the artist actually dies. Given informed and rational market
participants, such a precursor effect is what one would expect to observe because in efficient
markets expected changes in asset market fundamentals are priced in at once. The precursor effect
disassociates the observed price change from the event of the artist’s death (which may trigger
price reactions unrelated to the commitment mechanism, such as nostalgia effects) and tie it more
strongly to the unexpected reduction in the size of the artist’s final oeuvre.
We have identified three cases in which the terminal illness of relatively young artists became
publicly known one or two years before the artist’s death.23 That relatively long lapse of time
between the announcement and the artist’s decease allows us to test the hypothesis with our low-
frequency data.
The case of Keith Haring provides perhaps the most striking evidence. Haring rose to fame in the
1980s when he was still in his twenties. After having observed first symptoms in July 1988, he was
diagnosed with AIDS in fall 1988. Shortly afterwards, Haring’s former lover, Juan Dubose, died
of AIDS and those who were close to Haring must have begun to suspect that he himself was
probably infected, too.24 In a Rolling Stone Interview published in August 1989, Haring made his
health status public and established his foundation that provides financial support for AIDS-related
education, prevention, and care. He also used his art in the fight against AIDS. Since in the late
1980s no effective medical treatment against the immune deficiency was available, it was clear that
Haring did not have long to live. At least since Haring’s announcement of his HIV-positive status
23 In our search of artists suffering from some terminal or debilitating illness, we found a few more cases: Buffet, Paik, Scott, Rauschenberg, Dali, and de Kooning. However, none of those artists was still active in their last 5 to 10 years of their lives. It would therefore be exceedingly hard to pinpoint a defensible precursor date. Kippenberger would have been a prime case because he died when he was only 44. He died from liver cancer, but the first signs of his dire state of health only showed a few months before his death in 1997. Looking back on the year 1996, his wife attested him to be in very good shape (even though he was a lifelong heavy drinker). The precursor is thus too close to his death to discriminate. Moreover, Kippenberger was at that time not yet a major artist. He achieved fame only about three years after his death. 24 Gruen (1992), p. 195. In fact, some people suspected that Haring might die of AIDS already before he knew himself with certainty that he was infected. David Galloway, man of letters and academic, writes in his reminiscence “(m)ore than a year before the diagnosis, Newsweek had tracked the artist down in Europe to ask if his protracted stay there was a cover-up for his affliction with AIDS” (https://www.haring.com/!/selected_writing/the-marriage-of-heaven-and-hell).
25
in August 1989, but quite likely already since Dubose’s death in January, the actors in the art
market knew that his life’s work would be smaller than hitherto expected. From a theoretical
perspective, one would therefore expect that the announcement of his health status led to an
immediate increase in the prices of his artwork. This is actually what Haring’s accountant, Margaret
Slabbert, observed at the time.25 Given his reputation and his young age -- Haring was not yet 32
years old when he died -- one would furthermore expect this price effect to be extraordinarily large.
This is exactly what happened: the prices of Haring’s pictures did not rise after his death in 1990,
but already a year before. The scatter diagram in the first panel of Figure 7 illustrates that
announcement effect. The circles represent auction prices fetched before the public knew about his
impending death, the triangles represent prices recorded in the year of the announcement, and the
squares represent prices recorded after Haring’s death. Up to 1988, prices were clearly lower than
afterwards and the jump occurred in 1989 when Haring was still alive, and not after his death in
January 1990.
Figure 7: Reaction of the market on precursors of death for artists of our sample who died prematurely and whose death precursors became publicly known. The circles represent corrected prices (in logs) from auctions before an artist’s death precursor, while the squares represent prices from auctions after the artist’s actual death, the triangles represent prices fetched at auctions that took place between the precursor and death. The number next to the artist’s name is the artist’s age when the precursors of death became publicly known, the number in the bracket is his or her actual age at death.
The first two rows in Table 5 compare the estimated effects of Haring’s announcement of his
impending death (first row) with the price effects of his actual death (second row). The six
25 Gruen (1992), p. 210.
26
specifications (1-6) correspond to the specifications in Table 4. The time windows straddle, of
course, the date of the precursor (beginning of 1989) and the date of Haring’s death (February
1990), as the case may be. Across all six specifications, the estimated precursor effects are
economically large and statistically significant. The estimate of 1.34 estimated with our preferred
specification (bandwidth 5 years and flexible slopes) translates into a price jump of 282%. The
estimated effects of Haring’s actual death are much smaller, with one exception not statistically
significant, and the statistically significant estimate is actually negative.
Table 5: Estimated precursor and actual death effects
Bandwidth: 5 years Bandwidth: 4 years Bandwidth: 3 years
a a, DE*a a a, DE*a a a, DE*a
(1) (2) (3) (4) (5) (6)
HARING, precursor at 30 1.343*** 1.357*** 1.292*** 1.264*** 1.154** 0.774**
(0.287) (0.289) (0.296) (0.287) (0.415) (0.225)
HARING, death at 31 0.577 0.178 0.545 -0.380 0.456 -0.643**
(0.605) (0.508) (0.676) (0.300) (0.668) (0.245)
FRINK, precursor at 60 0.630*** 0.512*** 0.665*** 0.600*** 0.643** 0.523***
(0.170) (0.101) (0.159) (0.086) (0.202) (0.043)
FRINK, death at 62 -0.445*** -0.511*** -0.437** -0.498** -0.096 -0.125**
(0.129) (0.128) (0.161) (0.168) (0.051) (0.038)
IMMENDORFF, precursor at 60 0.638 0.219 0.589 0.189 0.213 -0.078
(0.390) (0.179) (0.408) (0.186) (0.288) (0.174)
IMMENDORFF, death at 61 0.099 -0.140 0.242 -0.042 0.395 0.188*
(0.384) (0.191) (0.387) (0.163) (0.301) (0.093) Notes: Clustered standard errors are in the parentheses. * p<0.10, ** p<0.05, *** p<0.01
Elisabeth Frink, probably best known as one of Britain’s most highly regarded post-war sculptors,
was also a prolific printmaker, which allowed us to include her artwork in our sample. Frink was
not only a prominent player in the international art world but, at least in Britain, also a public figure
who was created a Dame of the British Empire in 1982 and posthumously honored in 1996 with a
stamp labelling her a “20th century woman of achievement.” Her standing in the art world is
illustrated by her being asked to become the first female president of the Royal Academy of Arts,
which she declined, presumably because she feared this office would interfere with her nonstop
artistic production, worldwide exhibition activities, and other public commitments. Early in 1991,
Frink had to undergo an operation for cancer of the oesophagus, followed by a second operation
later that year. Even though she took her schedule up again after the operations, it became clear
that she was deteriorating. Frink died just one week after the installation of her last work, the Risen
Christ in the Liverpool Cathedral, in April 1993. Given her prominence, one can safely assume
27
that her health status became publicly known after her first operation when she was forced to scale
back her activities. We therefore date the precursor of her death to February 1991.
Examining the scatter diagram in Figure 7, the price increase after Frinks precursor in early 1991
is clearly discernible (notice, that in the precursor analysis, the two years 1997 and 1998 shown in
Figure 7 are not part of the 5 years precursor window). The estimates recorded in Table 5
corroborate this impression. All six specifications indicate sizable and statistically significant
precursor-induced price increases ranging from 43% to 67%. By way of contrast, all estimates of
a possible effect of Frink’s actual death are negative.
Jörg Immendorff, born 1945, knew how to feed the mass media to self-promote his art and his
career, first as a political agitator and, later as a quirky egomaniac. In the 1980s, he became one of
the best-known German contemporary artists and was honored in fall 2005 with an exhibition in
Berlin’s New National Gallery with the then German Chancellor, Gerhard Schröder, delivering the
opening address. Immendorff suffered from ALS (amyotrophic lateral sclerosis) and was, in 2005,
already marked by the terminal illness diagnosed already in 1998. When at the end of the year
2000, the lefthander Immendorff was no longer able to use his left hand to paint, he switched to
the right hand. He also began to use assistants to execute his ideas and monitored their every step.
After the 2005 exhibition, Immendorff had to undergo a tracheoctomy to help him breathe. In his
last year and a half, even though wheelchair-bound, he continued to direct his “factory” assistants.
He finished his last painting, the official portrait of Chancellor Schröder, shortly before dying at
the end of May 2007.
Immendorff’s medical health history admits various possibilities to pinpoint a date at which the art
market participants may have learnt that his death was imminent. ALS is not curable and the life
expectancy of a person diagnosed with ALS is hard to predict. More than half of people who
develop the disease die within 3-5 years, 10% live however for more than 10 years, and 5% over
20 years.26 Immendorff lived for another nine years. Fortunately, we do not have to deal with how
the probability of dying changes as ALS patients continue to live and how those changing
probabilities would have affected the price path. This is because Immendorff never publicized his
affliction. He actually largely refused to acknowledge it himself and instead of following the advice
of his physicians, rather experimented with treatments way outside the bounds of school medicine.
26 Stephen Hawking, for example, the eminent theoretical physicist and author of popular science books, lived for more than 55 years with ALS.
28
When the problems with his left hand became conspicuous in 2000, he went to great lengths to
convince the public that the problem was transitory.27 He informed the public only in May 2003
that he “very likely” suffered from ALS. In 2005 he became completely dependent on care and his
tracheoctomy in November of that year left no doubt that this marked the beginning of the terminal
stage of Immendorff’s life. We use the date of that medical intervention in 2005 as our precursor
and not the vague announcement in 2003 that did not give rise to a price increase, perhaps because
having already survived significantly longer than most patients did, it was difficult to predict his
remaining lifetime.28
The estimates reported in Table 5 reflect the great uncertainty about Immendorff’s odds of survival.
The positive point estimate of the precursor effect from our preferred specification (2) is sizable
but does not reach statistical significance at conventional levels. With one exception (specification
6) the other estimates are also positive and statistically insignificant. The estimated effects of
Immendorff’s actual death are, taken as a whole, smaller and statistically also insignificant, for the
specifications relying on four and five year bandwidths; the two estimates relying on three year
bandwidths are however sizable and larger than those returned for the precursor effects. The last
estimate is even statistically significant at the 1% level. Before interpreting that outlier result as
indicating a death effect, one should consider that a nostalgia effect might have been at work, and,
since nostalgia effects are short-lived, this blimp only shows up when using the three year
bandwidth. It is certainly not farfetched to suppose that the official hanging of former Chancellor
Schröder’s official portrait in the German Chancellery about a month after Immendorff’s death
may have had a pronounced nostalgia effect. That work, as so often in Immendorff’s career, gave
rise to a public stir. The portrait painted in gold with monkeys looking over the Chancellor’s
shoulders and a melting federal eagle at the bottom contrasted strongly with the portraits of all the
former Chancellors and was certainly not to everybody’s taste. Immendorff clearly knew how to
stay in the limelight – even after his death.
To lend further support to our claim that the precursor effects estimated for Haring, Frink, and
Immendorff preempted, at least to a large extent, any death effect, we employed the border-
falsification strategy of the regression discontinuity design (Imbens and Lemieux 2008) proposed
by Becker et al. (2016). According to this strategy, time windows (in the original context:
27 Riegel (2018), p. 234. Our account of Immendorff’s health history draws to a large extent on Riegel’s biography. 28 Even Immendorff’s physician was reluctant to venture a clear prediction (see Riegel, p. 251).
29
geographical bands) that include the actual borderline should not be used. The reason for that
restriction is that placebo tests using fake treatment-groups that include observations of the true
comparison group (or fake-comparison groups that include observations of the true treatment
group) are contaminated by non-germane observations and are thus liable to blur the results.
Figure 8 illustrates the results of our border-falsification placebo tests. To test as closely as possible
to the border, we used in our border-falsification tests for the precursor effect the smallest
bandwidth, i.e. three years, and continue to apply the specification with flexible trends (column 6
in Table 5). Since Haring’s true precursor is dated at the very beginning of the year 1989, the data
used to estimate the true precursor effect are from the period 1986-1991. For the placebo
treatments, we use also bandwidths of three years. To avoid windows including the precursor time
(1989-1990), we test for the existence of an effect in 1986 and 1993. The results clearly show that
those placebo tests do not return statistically positive estimates.
Figure 8: Estimates of the true precursor (bold circles) and two placebo effects for each of the three artists. The circles indicate the estimated effects; the whiskers are the 90% confidence intervals. The estimates of the true precursor effects correspond to the baseline death effects reported in Table 5, column (6). Notice, that we use the exact date of the precursor and delineate the bandwidths, including the placebo bandwidths, accordingly.
We applied an equivalent procedure for Frink and Immendorff. The estimates of the two placebo
treatment effects for Frink are both statistically insignificant. Also for Immendorff, we do not find
any placebo treatment effect. The abortive attempt to find a treatment effect for the year 2001,
30
which happens to be the year in which Immendorff began to work with his right hand, is compatible
with the conclusion that the choice of our precursor was not ill judged, but rather an attempt to find
something that does not exist. Immendorff’s case history may simply not feature a single event that
would mark a threshold after which it was clear to the art world that Immendorff would not survive
a well-defined span of time.
6. Explaining the Death Effect: Reputation and Age at Death
The Coase framework and general asset pricing considerations give rise to two hypotheses that
guided our strategy to explain the estimated heterogeneity in death effects across artists. The first
hypothesis maintains that that the death effect varies, ceteris paribus, negatively with the artist’s
age at death and disappears for artists who die at a very high age. The ceteris paribus clause is
important because the death of young or even middle-aged artists who some insiders believe to be
likely to make it big, may lower the price because death nips those hopes of becoming highly
reputed in the bud and thus frustrates the expectations of early collectors. For those artists one
needs to add the negative frustration effect to the positive effect associated with the death-induced
curtailment of their oeuvre. We thus add a second hypothesis maintaining that the death effect of
artists who die before their time is smaller, perhaps even negative, for merely reputed artists than
for highly reputed ones.
To test the first hypothesis we regress the estimated death effects in our full sample on a polynomial
of the respective artist’s age at death. We have two dependent variables: RD estimates of the death
effect (based on bandwidths of five years) imposing (i) a common trend and (ii) allowing for
different trends on the either side of the death threshold (columns 1 and 2 of Table 6). Table 6
reports the results. Because the precision of the estimates of the death effects vary a great deal, we
use weighted least square regressions that account for the varying number of observations, which
the estimates are based on. The analytical weights used in Tables 6 and 7 are the square roots
𝑛 .29
In the all specifications, the estimate of the age at death variable is negative and statistically
significant, and the estimates of the squared age at death variable are positive and statistically
29 Alternative weightings or no weighting lead to very similar results (Table A8).
31
significant. Figure 9 depicts in the left panel the relationship between the death effect and age at
death resulting from specification (4). The graph has the shape predicted by our first hypothesis:
the death effects are largest for artists who die at a young age and decrease with increasing age at
death.30 Moreover, the Figure shows that our estimates also produce the predicted result that the
death effect disappears for artists who died in their eighties. The slight increase of the curve after
the age of 90 is imposed by the quadratic specification. In columns (5) and (6), we therefore
included a cubic term of the age at death variable. The curve resulting from specification (6)
depicted is depicted in the right panel of Figure 9 and shows a plateau for ages 60 up to 90, and
then declines.31
Table 6: The death effect as a function of age at death
a a, DE*a a a, DE*a a a, DE*a
(1) (2) (3) (4) (5) (6)
age at death -0.012*** -0.013*** -0.066*** -0.061*** -0.198*** -0.253***
(0.003) (0.003) (0.015) (0.016) (0.063) (0.066)
age at death2/100 0.039*** 0.034*** 0.251** 0.343***
(0.011) (0.011) (0.099) (0.104)
age at death3/1000 -0.011** -0.015***
(0.005) (0.005)
Constant 1.086*** 1.171*** 2.876*** 2.751*** 5.313*** 6.300***
(0.224) (0.235) (0.532) (0.570) (1.243) (1.308)
Adj. R-squared 0.139 0.158 0.231 0.219 0.258 0.275 Note: The odd columns use as dependent variable the death effects estimated from regression equation (3), while the even columns use as dependent variable the death effects estimated from regression equation (4), both based on the 5-year bandwidth. We used weighted least squares regressions, where the analytical weights are the square roots of the number of observations the estimate of dependent variable is based on. We replicate estimates of column 6 using different weights or no weights in Table A8. The number of observation is always 106. Conventional standard errors in the parentheses. * p<0.10, ** p<0.05, *** p<0.01
The regression results reported in Table 6 lump together artists who enjoyed at the time of death
different levels of reputation. To test our second hypothesis, claiming that the death effect varies
positively with reputation, whereby age at death moderates the reputation effect, we report in Table
30 The reason why we obtain in our estimates reported in Table 6 a death effect that decreases with age at death, whereas Ursprung and Wiermann (2011) and Etro and Stepanova (2015) obtained a hump-shaped curve can be attributed to the differences in the employed samples. Both Ursprung and Wiermann (2011) and Etro and Stepanova (2015) worked with very large samples which included a large number of artists with little or no reputation to speak of. Our sample, however, only includes artist whose work has been sold many times in a relatively short period, which implies that all of our sample artists enjoyed a substantial reputation. 31 The decline of the curve for high ages at death is imposed again, this time by the cubic specification.
32
7 regressions that include, one by one, our six reputation measures and the interaction of those
variables with age at death.
Figure 9: Death effects predicted by a quadratic function (left) and a cubic function of age at death (right). Shaded areas are the 90% confidence intervals.
The estimated effect of reputation is for all reputation measures positive. Moreover, with the
exception of the measure “Number of Amazon books”, the interaction of reputation with age at
death is always negative, thus lending credibility to our prediction that higher reputation is
associated with higher death effects and that the reputation-induced death effect decreases with
increasing age at death. Even if “reputation” and “reputation*age” are not always individually
significant, the two variables are jointly significant either in the quadratic or the cubic age model.
The economic significance of the estimates reported in Table 7 is substantial. Using, for example,
the reputation measure Number of Wikipedia languages (columns 1 or 2) and an artist who dies at
the age of forty, a 1% increase in the number of available languages is associated with an increases
in the death effect of almost 30%. If, however, the artist dies at the age of 85, that effect is reduced
to about 5%.
33
Table 7: The death effect as a function of age at death and reputation
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
age at death -0.024 -0.208*** -0.061*** -0.259*** -0.040* -0.204*** -0.037** -0.208*** -0.062*** -0.294*** -0.063*** -0.291***
(0.023) (0.066) (0.016) (0.064) (0.021) (0.070) (0.019) (0.069) (0.016) (0.065) (0.016) (0.068)
age at 0.020 0.313*** 0.034*** 0.353*** 0.024* 0.283*** 0.026** 0.294*** 0.037*** 0.412*** 0.037*** 0.407***
death2/100 (0.013) (0.100) (0.011) (0.101) (0.013) (0.106) (0.011) (0.105) (0.011) (0.102) (0.011) (0.108)
age at -0.015*** -0.016*** -0.013** -0.013** -0.019*** -0.018***
death3/1000 (0.005) (0.005) (0.005) (0.005) (0.005) (0.005)
reputation 0.516** 0.484** 0.029 0.042 0.163* 0.134 0.208** 0.165* 1.362 2.229** 0.344 0.857*
(0.226) (0.218) (0.032) (0.031) (0.094) (0.092) (0.087) (0.086) (1.049) (1.017) (0.455) (0.458)
reputation -0.005* -0.005* 0.001 0.001 -0.002 -0.001 -0.002** -0.002* -0.012 -0.022* -0.003 -0.009
*age (0.003) (0.003) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.013) (0.012) (0.006) (0.006)
Constant 0.460 3.955** 2.565*** 6.171*** 1.651* 4.802*** 1.269 4.627*** 2.552*** 6.716*** 2.661*** 6.676***
(1.132) (1.607) (0.586) (1.268) (0.858) (1.533) (0.823) (1.532) (0.574) (1.254) (0.600) (1.296)
Reputation measure
No. of Wikipedia languages (in logs)
No. of Amazon books (in logs)
No. of words in Oxford dict. (in logs)
No. of words in NYT obituary (in logs)
Additive obituary measure (AOM)
Dummy: top 50% reputation (AOM)
Joint F-test reputation [p-value]
[0.003] [0.003] [0.028] [0.016] [0.028] [0.106] [0.022] [0.063] [0.026] [0.003] [0.187] [0.046]
Adj. R-squared 0.289 0.340 0.258 0.319 0.258 0.293 0.261 0.300 0.259 0.341 0.229 0.305
Notes: Weighted least squares regression with squared roots of number of observations as analytical weights. The dependent variable is the death effects estimated from the specification (4) with a 5-year bandwidth (see Table 4, column 2). We used weighted least squares regressions, where the analytical weights are the square roots of the number of observations the estimate of dependent variable is based on. Number of observation is 106. Conventional standard errors in the parentheses. We repeated some of these estimations with death effects based on smaller bandwidths (Table A6) and with death effects measured with alternative methods (Table A7). We replicate estimates of column 10 using different weights or no weights in Table A8. These estimates are similar to those shown in Table 6 and 7. * p<0.10, ** p<0.05, *** p<0.01
34
Figure 10: Predicted death effects for highly reputed and merely reputed artists, based on top (solid) and bottom (dashed) quarter of the additive obituary measure (AOM). Left panel: Column 9 of Table 7 (quadratic age function). Right panel: Column 10 of Table 7 (cubic age function). The shaded areas are 90% confidence intervals.
Our preferred measure of reputation at death is the additive obituary measure (AOM) because it
measures reputation at the time of the artist’s death and, together with the number of Wikipedia
languages, explains the largest share of the variation in death effects (columns 2 and 10). Since our
preferred reputation measure is an index varying between 0 and 1, we illustrate in Figure 10 the
effect of reputation on the death effect by comparing highly reputed artists in the top 25% of the
AOM distribution with the merely reputed artists ones in the bottom 25%. The curve for highly
reputed artists always lies above the curve for merely reputed artists and for very high ages at death
(90+), the two curves converge. Using the quadratic age function, the death effect for the highly
reputed artists becomes zero about 20 years later than for the merely reputed artists. The difference
between the estimated death effects of highly reputed and merely reputed artists becomes even
more pronounced when using the cubic age function.
7. Sensitivity analysis: Alternative estimates of death effects
To further test the sensitivity of our baseline estimates of individual death effects, we subjected our
baseline approach to three kinds of changes. First, we use different art prices to estimate the artist-
specific death effects via regression discontinuity. In a second set of tests, we estimate the death
35
effects using alternative methods. In a third set of tests, finally, we purge our sample of all artists
who have been inactive for a long time before their death and investigate whether post-mortem
increases in auction sales has an undue effect on our baseline estimates of the individual death
effects.
Different Art Prices
In our baseline approach, we use hammer prices corrected for the price development in the art
market at large and for an array of hedonic characteristic of each artwork sold. Alternatives would
be to use only one of those two corrections, or to use the hammer price without any correction. In
Table 8, we show how using those prices affects the regression results of our baseline estimate
reported in Table 7, column 10. For comparison purposes, we reproduce those baseline results
again in Table 8, column 1. In the following three columns, we replace our baseline prices with the
three alternative prices but retain the baseline specification of a five-year bandwidth around the
artists’ death, different time trends before and after death, and the additive obituary measure of
reputation. The regression results reported in Table 8, columns 2-4, show that the baseline point
estimates are not overly sensitive to those changes. Not controlling for the hedonic characteristics
(columns 3 and 4) decreases, however, the explanatory power of the regressions, which affects in
particular the statistical significance of the coefficient estimates of the reputation variable and the
interaction of reputation with age at death.
Different Estimation Methods
Even though we are convinced that the regression discontinuity method is much better suited to
estimate artist-specific death effects, for comparison purposes, we estimated those effects also with
the difference-in-difference (DD) method. The DD design is problematic in the art price context
because it requires a contemporaneous control group that is almost impossible to find when dealing
with reputed artists who are by definition incomparable.32
32 In a recent study, Pénasse et al. (2020) also used a DD design to estimate for a sample of 72 artists the average death effect on art prices and turnover. They work with a synthetic control group of living artists that is supposed to be similar to the group of death-treated artists. Unlike their study that focusses on overall price changes, i.e. an average
36
We compare the prices of the artwork of treated, i.e. deceased, artists with the prices of artwork
created by those artists in our sample who are still alive at the time of death of the respective treated
artist. Using the entire panel of artists in our sample, we estimate the equation
ln 𝑝 𝛼 𝜇 𝛾 𝐷𝐸 𝛽𝑋 𝛿 𝑌 𝜃 𝑡 𝜀 . 5
ln 𝑝 is the log hammer price of artwork 𝑖 (created by artist 𝑚 and sold in year 𝑡). We regress
those log prices on matrix 𝑋 containing the 𝑘 hedonic characteristics already used in equation (1).
We also control for changes in general art market prices using the year dummies 𝑌 and artist fixed
effects 𝜇 . To allow for nonparallel trends in auction prices across artists, we include artist-specific
linear time trends 𝜃 𝑡. The variables of interest are the artist-specific death dummy variables
𝐷𝐸 . In line with Angrist and Pischke (2014), we refer to the regression model (5) as our multi-
artist DD regression setup.
Our sample is a pseudo panel in which all artists die in the observation period. Since we only
consider sales in the eleven-year window around each artist’s death, the control-group observations
are therefore taken from an ever changing and towards the end of the observation period shrinking
group of living artists. The results for the 5-year bandwidth are recorded in Table 8, column 5.33
Those results largely repeat the pattern with respect to age at death and reputation, even if less
precisely so than in the other columns. The joint test of the effect of the reputation variables is not
significant here, but the joint test of the age at death variables is (at the 0.02 level). Due to the
deficiencies in our definition of the control group, we find our DD setting not sufficiently
convincing to credit those results with much reliability.
The second alternative method to measure individual death effects is to take differences of
corrected prices before and after death. In particular, we used the difference in median prices. Given
that this approach ignores the trends around the death thresholds, which makes sample size less
death effect, we are interested in precise estimates of individual death effects, which would require an almost perfect match of the deceased and the matched artist. Such matches are unlikely to exist. 33 The N is only 90 in this case. We drop 16 artists who died after 2007 because their control groups would make use of a smaller corpus of auctions: we have no artists who were alive for the entire 5 years between 2007 and 2011.
37
relevant, we use the smallest bandwidth. The death effects measured with that method (column 6,
Table 8) largely replicate the pattern reported in the baseline column.34
In Table A7 of the appendix, we report results of additional regressions using the five alternative
sets of death effects estimates, but assuming age at death and reputation take simpler functional
forms than those reported in Table 8. Those results corroborate the vital role of age and reputation
at death in explaining the size of the death effects. Generally, the results are less precise and explain
far less of the variation in the estimated death effects when prices are not controlled for the hedonic
characteristics; however, they do not contradict the baseline findings.35 In short, this sensitivity
exercise reveals that the correction of the auction prices for hedonic characteristics is a relevant
feature when estimating reliable individual death effects.
A Different Sample and an Additional Independent Variable
So far, we have not taken into account that some of our sample artists have been inactive for a long
time before they died. In those cases, the art market participants must have eventually realized that
the oeuvre of those retired, sometimes even debilitated artists is complete and that their death will
not have any price-relevant consequences. Including the death effects of those retired or even
debilitated artists may therefore distort our results presented in section 6.
To test whether such a distortion is at work, we purged from our sample all 34 artists for whom we
could not find in our dataset any artwork created less than four years before their death. Re-
estimating our baseline regression with that reduced sample, we obtained the results reported in
Table 8, column 7. For that reduced subset, the shape of the age polynomial continues to apply, is
however somewhat less pronounced, and the squared and cubic terms lose statistical significance.
The role of reputation, in particular when interacted with age at death, becomes more prominent.
34 Additional results from regression specifications making use of the DD approach or the median differences of prices before and after-death are reported in in Table A7. 35 Note, that this imprecision cannot not stem from smaller absolute death effects resulting from other prices. Actually, the death effects based on corrected prices are very often smaller than the death effects based on hammer prices, or prices corrected only for general art price changes (see Table A4). Thus, the more plausible explanation is that not accounting for hedonic characteristics overestimates the death effect, which damages the fit of the regression.
38
Table 8: Sensitivity results
baseline
corrected for hedonic characteristics
corrected for general art price changes
hammer prices
diff-in-diff estimates (N=90)
differences in median prices 3 years bandwidth
only active artists (N=73)
controlling for supply growth
(1) (2) (3) (4) (5) (6) (7) (8)
age at death -0.294*** -0.305*** -0.202* -0.234** -0.122 -0.172*** -0.155* -0.277***
(0.065) (0.077) (0.109) (0.118) (0.086) (0.059) (0.089) (0.068)
age at 0.412*** 0.428*** 0.305* 0.354* 0.149 0.219** 0.170 0.382***
death2/100 (0.102) (0.122) (0.172) (0.187) (0.138) (0.093) (0.148) (0.109)
age at -0.019*** -0.019*** -0.015* -0.017* -0.006 -0.009** -0.006 -0.017***
death3/1000 (0.005) (0.006) (0.009) (0.009) (0.007) (0.005) (0.008) (0.005)
reputation 2.229** 2.442** 1.449 1.868 0.241 1.962** 2.723** 2.091**
(1.017) (1.215) (1.710) (1.852) (1.322) (0.918) (1.113) (1.032)
reputation*age -0.022* -0.025* -0.012 -0.018 0.002 -0.020* -0.029** -0.020
(0.012) (0.015) (0.020) (0.022) (0.016) (0.011) (0.014) (0.012)
supply growth -0.014
(0.017)
Constant 6.716*** 7.026*** 4.116* 4.826** 3.339** 4.629*** 4.300** 6.503***
(1.254) (1.499) (2.110) (2.284) (1.646) (1.129) (1.638) (1.283) Joint F-test age at death [p-value] [0.000] [0.000] [0.246] [0.143] [0.023] [0.000] [0.000] [0.000]
Joint F-test reput. [p-value]
[0.003] [0.023] [0.184] [0.254] [0.119] [0.016] [0.012] [0.004]
Adj. R-squared 0.341 0.284 0.028 0.034 0.108 0.342 0.389 0.339
Notes: The dependent variable is the death effects estimated from the specification (4) with a 5-year bandwidth with corrected prices (columns 1, 7 and 8). Alternative prices applied are in columns 2, 3 and 4. Alternative death effect estimations are applied in columns 5 and 6. Weighted least squares regression with squared roots of number of observations as analytical weights. Number of observation is 106, except for column 5 and 7. In column 5 we only consider artists who died before 2007. In column 7 we use a subset of artists that was active until shortly before death. See also Table A7 for further specifications of similar regressions. Conventional standard errors in the parentheses. * p<0.10, ** p<0.05, *** p<0.01
39
When commenting on Figure 3, we already remarked that the number of auction sales typically
increases after the death an artist. Such an increase in turnover can be substantial; Lichtenstein, Lewitt,
Dubuffet, and Appel are cases in point (Table A1). One reason for this increase in sales may be that the
heirs of the deceased artist are in a hurry to divest the estate. It is conceivable that such a surge in flow
supply induces a dampening effect on the post mortem prices, which, in turn, would tend to
reduce the estimated death effects. We thus measured for each artist the post mortem supply surge
defined as the growth rate in the number of auctions in the five years after the artist’s death as compared
to the number of auctions in the five years before. Adding the postmortem supply surges when
explaining the variance in the death effects, we obtained the results recorded in column 8 of
Table 8. The results shows that post mortem supply surges have no systematic influence on the
death effect and including that variable does not change our baseline results.
8. Conclusion
We estimate individual death effects in art prices and explain the observed heterogeneity of
those death effects by referring to the ideas developed by Ronald Coase in his legendary 1972
note on durable goods monopolies. The fine art market fits perfectly the domain of application
that Coase had in mind. Highly reputed artists are monopolists because their artwork is by
definition unique, attributable to a specific producer, novel, and non-substitutable. In the art
market context, the Coasian framework gives rise to two hypotheses that we test with our data.
The first hypothesis maintains that the death effect varies negatively with the deceased artists’
age at death and disappears when the artist dies at a high age. The second hypothesis predicts
that for artists who die before their time, the death effect varies positively with the deceased
artist’s reputation at the time of death.
The main conclusion that we draw from our results is that the basic predictions of asset pricing
theory in general, in particular Coasian dynamics in particular, provide important information
for explaining art price formation. Reputed artists, even though they produce under conditions
of monopolistic competition, are not able to exploit their strong market position to charge
monopoly prices because they lack institutional devices that would allow them to credibly
commit to a restrained production plan. The exceptions are death and terminal or debilitating
illnesses. Our study sheds a new light on this commitment device by identifying and explaining
the price effects of death or anticipated imminent death of individual artists.
40
Empirical studies that measure and explain death effects in the visual art market have hitherto
used panel data to estimate average death effects, i.e. death effects that could not be associated
with any specific individual artist. In this study, we estimate individual death effects for a
sample of artists whose work was sold at auctions sufficiently often to allow estimating artist-
specific death effects with the help of the regression discontinuity method. In our sample of 106
visual artists, we find artists with statistically significant positive and very few with statistically
significant negative death effects. We also find many artists whose death caused no statistically
significant death effects.
Our results strongly support the predictions that we derived from the Coasian framework even
though we are working with a rather small sample. After having estimated the individual death
effects, the first hypothesis maintaining that the death effect varies negatively with the artist’s
age at death is straightforward to test empirically. To test the second hypothesis that associates
the size of the death effect with the artists’ reputation at death, we designed novel measures that
quantify the artists’ reputation at the time of their death. For that purpose, we collected
obituaries of all sampled artists and judged their reputation based on the length of the obituaries
in five well-known general interest newspapers and magazines published in different countries.
Our estimates of the reputation effect turn out to be robust to various versions of our novel
reputation at death measure and to various traditional measures of reputation. Age and
reputation at death explain up to 40% of the variation in individual death effects.
We ran a large number of sensitivity analyses. One set of tests reveals that correcting prices for
the hedonic properties is indispensable when estimating individual death effects with the
regression discontinuity design. The second lesson from the sensitivity exercise concerns
applying the difference-in-differences approach to estimating death effects of individual artists.
As is well known, the difference-in-differences approach relies on a control group that matches
the treatment group very well and imposes an equal trend in prices before and after death. Both
of those implicit assumptions are hardly ever met because first-rate artists are almost by
definition incomparable. We thus came to believe that difference-in-differences is not a suitable
method to estimate death effects of well-known individual artists.
Our study differs from other studies not only by estimating individual death effects instead of
average effects, we also disentangle for the first time the two of the main determinants of the
death effect, i.e. age and artistic reputation at death. So far, empirical estimates of the
relationship between the death effect and age at death has been confounded with influences
arising from artistic reputation, which, of course, to some extent correlates with age. Young
41
artists are less likely to be eminent than older ones, implying that the death of a young artist is
in most cases not followed by a substantial change in the prices of her artwork. If, however, a
young artist is already highly reputed when she dies, the age and reputation effect combine in
letting the price increase. It is therefore not surprising that we found the highest death effects
for the two young superstars Jean-Michel Basquiat and Keith Haring who died very early at the
age of 27 and 31.
Notice, finally, that the fact that an artist is dead does not imply per se that his artwork
commands a higher price. Death only increases the market value of an artist’s work if she died
relatively young and enjoyed at that time already a substantial reputation. If nobody has noticed
her qualities, the price of her artwork does not change at all, and if some early collectors made
a perhaps well-considered wager in the hope of her becoming famous later on, they will have a
lot to regret. What this shows is that one cannot trust wordsmiths with economic matters. In his
story that prompted the preamble, Mark Twain (1893) got it all wrong. Making money from an
artist’s death is not easy. It may even involve murder, at least if you want to believe economist-
turned-mystery-writer Marshall Jevons (2014).
References
Angrist, J. D. and J.-S. Pischke (2014). Mastering’metrics: The path from cause to effect. Princeton University Press.
Ashenfelter, O. and K. Graddy (2006). Art auctions. Handbook of the Economics of Art and Culture 1, 909–945.
Asubel, L. and R. Deneckere (1989). Reputation in bargaining and durable goods monopoly. Econometrica 57(3), 511–531.
Bagnoli, M., S. Salant, and J. Swierzbinski (1989). Durable-goods monopoly with discrete demand. Journal of Political Economy 97(6), 1459–78.
Becker, S. O., Boeckh, K., Hainz, C., & Woessmann, L. (2016). The empire is dead, long live the empire! Long‐run persistence of trust and corruption in the bureaucracy. The Economic Journal 126(590), 40–74.
Board, S. and M. Pycia (2014). Outside Options and the Failure of the Coase Conjecture. American Economic Review 104(2), 656–671.
Bulow, J. (1982). Durable-goods monopolists. Journal of Political Economy 90(2), 314–332.
Canudas-Romo, V. (2010). Three measures of longevity: Time trends and record values. Demography 47(2), 299–312.
42
Cason, T. and T. Sharma (2001). Durable goods, Coasian dynamics, and uncertainty: Theory and experiments. Journal of Political Economy 109(6), 1311–54.
Chilvers, I. (2015). The Oxford dictionary of art and artists (5 ed.). Oxford University Press.
Coase, R. H. (1972). Durability and monopoly. The Journal of Law and Economics 15(1), 143–149.
Ekelund, R. B., J. D. Jackson, and R. D. Tollison (2017). The economics of American art: Issues, artists and market institutions. Oxford University Press.
Etro, F. and E. Stepanova (2015). The market for paintings in Paris between rococo and roman- ticism. Kyklos 68(1), 28–50.
Galenson, D. W. (2006). Artistic capital. Routledge.
Galenson, D. W. and R. Jensen (2001). Young geniuses and old masters: The life cycles of great artists from Masaccio to Jasper Johns (No. w8368). National Bureau of Economic Research.
Galenson, D. W. and B. A. Weinberg (2000). Age and the quality of work: The case of modern American painters. Journal of Political Economy 108(4), 761–777.
Galenson, D. W. and B. A. Weinberg (2001). Creating modern art: The changing careers of painters in France from impressionism to cubism. American Economic Review 91(4), 1063–1071.
Graddy, K. (2013). Taste endures! The rankings of Roger de Piles († 1709) and three centuries of art prices. The Journal of Economic History 73(3), 766–791.
Gruen, J. (1992). Keith Haring: The Authorized Biography. Simon and Schuster.
Gul, F., Sonnenschein, H., and R. Wilson (1986). Foundations of Dynamic Monopoly and the Coase Conjecture. Journal of Economic Theory 39(1), 155–190.
Güth, W., Ockenfels, P., Ritzberger, K. (1995) On durable goods monopolies: An experimental study of intrapersonal price competition and price differentiation over time. Journal of Economic Psychology 16(2), 247–274.
Imbens, G. and Lemieux, T. (2008). Regression discontinuity designs: a guide to practice, Journal of Econometrics 142(2), 615–635.
Itaya, J. and H. W. Ursprung (2016). Price and death: modeling the death effect in art price formation. Research in Economics 70(3), 431–445.
Jevons, M. (2014). The mystery of the invisible hand: A Henry Spearman Mystery. Princeton University Press.
Lee, D. S., and Card, D. (2008). Regression discontinuity inference with specification error. Journal of Econometrics, 142(2), 655–674.
Murfin, J. and R. Pratt (2019). Who finances durable goods and why it matters: Captive finance and the Coase conjecture. Journal of Finance 74(2), 755–793.
43
Murray, C. (2003). Human accomplishment: The pursuit of excellence in the arts and sciences, 800 BC to 1950. Harper Collins.
Nava, F. and P. Schiraldi (2019). Differentiated durable goods monopoly: A robust Coase conjecture. American Economic Review 109(5), 1930–68
Pénasse, J., Renneboog, L., & Scheinkman, J. A. (2020). When a Master Dies: Speculation and Asset Float. Review of Financial Studies. Forthcoming.
Reynolds, S. (2000). Durable-Goods Monopoly: Laboratory Market and Bargaining Experiments. RAND Journal of Economics 31(2), 375–394.
Renneboog, L. and Spaenjers, C. (2013). Buying Beauty: On Prices and Returns in the Art Market. Management Science 59(1), 36–53
Riegel, H. (2018). Immendorff: Die Biographie. Riverside Publishing, Zurich.
Simonton, D. (1984). Scientific eminence historical and contemporary: A measurement assessment. Scientometrics 6(3), 169–182.
Stokey, N.(1981). Rational expectations and durable goods pricing, Bell Journal of Economics 12(1), 112–128
Twain, M. (1893). Is he living or is he dead? The Cosmopolitan 15(5), 629-634.
Ursprung, H. (2015). Zum Todeseffekt im Kunstmarkt (the death effect in the art market). In H. Bündge and J. Holten (eds.), Nach dem frühen Tod (after an early death), [bilingual], Staatliche Kunsthalle Baden-Baden, Walther König, Köln,104–113.
Ursprung, H. and C. Wiermann (2011). Reputation, price, and death: An empirical analysis of art price formation. Economic Inquiry 49(3), 697–715.
von der Fehr, N. and K. Kühn (1995). Coase versus pacman: Who eats whom in the durable-goods monopoly?” Journal of Political Economy 103(4), 785–812.
44
Appendix
Table of contents
Content Item number
Descriptive statistics of auctions by artist Table A1
Corrected auction prices against normalized years of sale by artist Figure A2
Full Table 4 of the main text Table A3
Sensitivity death effect estimates Table A4
Correlations between death effects estimates Table A5
Sensitivity estimates of death effects determinants Tables A6-A7
Estimates of death effects determinants with alternative weighting Table A8
45
Table A1: Descriptive statistics of auctions by artist
Age at Death date Nr. of auctions Average hammer price (US $) Reput.
Artist death mm/dd/yyyy before after diff before after diff dummy
BASQUIAT 27 08/12/1988 67 232 165 13,314 71,444 58,131 0.14
HARING 30
(31) 01/01/1989 02/16/1990
56 177 121 4,528 21,428 16,901 0.20
KIPPENBERGER 44 03/07/1997 5 52 47 3,370 41,158 37,788 0.33
JERZY 58 03/12/2001 5 96 91 800 1,443 643 0.00
WARHOL 58 02/22/1987 98 474 376 26,686 110,605 83,919 0.73
NESBITT 59 07/08/1993 49 31 -18 2,701 2,285 -416 0.04
FRINK 60
(62) 02/01/1991 04/18/1993
53 74 21 2,195 3,374 1,179 0.27
IMMENDORFF 60
(61) 11/23/2005 05/28/2007
52 140 88 9,574 40,694 31,120 0.30
RIZZI 61 12/26/2011 12 82 70 1,505 1,800 294 0.04
BEUYS 64 01/23/1986 11 46 35 5,392 22,710 17,319 0.50
BRATBY 64 07/20/1992 28 20 -8 1,528 2,302 774 0.20
SCANAVINO 64 11/28/1986 10 80 70 1,317 6,632 5,315 0.00
VOSTELL 65 04/03/1998 13 23 10 3,394 3,072 -322 0.19
TINGUELY 66 08/30/1991 40 82 42 11,164 8,867 -2,297 0.46
MITCHELL 67 10/30/1992 46 47 1 122,511 102,004 -20,507 0.09
SAURA 67 07/22/1998 99 141 42 27,342 35,435 8,093 0.29
WOLVECAMP 67 10/11/1992 13 66 53 7,796 4,712 -3,084 0.00
ROTH 68 06/05/1998 32 82 50 3,360 5,039 1,679 0.13
LUCEBERT 69 05/10/1994 122 149 27 9,652 6,147 -3,504 0.09
POLKE 69 06/10/2010 265 357 92 119,510 496,366 376,856 0.41
BUFFET 71 10/04/1999 354 330 -24 32,541 27,086 -5,454 0.29
FRANCIS 71 11/04/1994 194 270 76 103,927 40,650 -63,277 0.29
SAINT PHALLE 71 05/21/2002 24 45 21 7,616 7,078 -538 0.20
LICHTENSTEIN 73 09/29/1997 149 521 372 173,149 73,422 -99,727 0.75
PAIK 73 01/29/2006 12 30 18 10,845 7,343 -3,502 0.38
PORTOCARRERO 73 04/27/1985 49 42 -7 2,460 4,812 2,352 0.05
WESSELMANN 73 12/17/2004 350 342 -8 35,787 282,873 247,085 0.45
KITAJ 74 10/21/2007 19 29 10 96,074 69,361 -26,713 0.57
STAMOS 74 02/02/1997 74 57 -17 7,439 7,816 377 0.06
GALL 75 12/09/1987 62 144 82 2,175 5,285 3,109 0.00
GUTTUSO 75 01/18/1987 90 120 30 9,559 18,880 9,321 0.09
NOLAN 75 11/28/1992 14 15 1 2,603 47,951 45,348 0.04
MOTHERWELL 76 07/16/1991 91 64 -27 83,666 44,814 -38,852 0.39
SCOTT 76 12/28/1989 27 22 -5 7,461 15,851 8,391 0.04
ARMAN 77 10/22/2005 129 298 169 13,352 25,773 12,421 0.21
CESAR 77 12/06/1998 29 50 21 3,116 2,920 -196 0.45
DORAZIO 77 05/17/2005 232 348 116 9,363 32,610 23,247 0.18
HELD 77 07/27/2005 15 23 8 12,831 38,850 26,019 0.06
LORJOU 77 01/26/1986 23 68 45 1,547 5,987 4,439 0.00
CHILLIDA 78 08/19/2002 29 107 78 14,759 20,922 6,164 0.34
RIOPELLE 78 03/12/2002 89 125 36 42,304 114,361 72,056 0.11
RIVERS 78 08/14/2002 58 68 10 15,606 36,286 20,680 0.30
BURRI 79 02/13/1995 21 38 17 76,764 194,809 118,045 0.14
DZUBAS 79 12/10/1994 26 27 1 12,296 5,143 -7,153 0.03
GUAYASAMIN 79 03/10/1999 37 28 -9 21,604 25,893 4,289 0.17
HERON 79 03/20/1999 17 38 21 5,729 47,547 41,818 0.37
46
Age at Death date Nr. of auctions Average hammer price (US $) Reput.
Artist death mm/dd/yyyy before after diff before after diff dummy
LEWITT 79 04/08/2007 203 346 143 13,678 25,602 11,924 0.33
MUHL 79 04/04/2008 62 94 32 4,001 4,833 832 0.00
SLOANE 80 03/05/1985 33 57 24 2,294 3,954 1,660 0.04
MANESSIER 81 08/01/1993 68 46 -22 36,383 9,408 -26,976 0.35
SEBIRE 81 12/13/2001 31 22 -9 2,132 2,327 196 0.00
SHEETS 81 03/31/1989 16 37 21 4,294 6,720 2,427 0.01
BACON 82 04/28/1992 11 25 14 2,300,000 710,360 -1,545,486 0.35
GISSON 82 07/28/2003 108 155 47 2,134 2,414 281 0.00
DUBUFFET 83 05/12/1985 162 365 203 29,440 171,219 141,780 0.32
FRANKENTHALER 83 12/27/2011 132 325 193 116,710 132,084 15,374 0.37
LEVIER 83 09/03/2003 49 49 0 1,154 1,402 248 0.00
RAUSCHENBERG 83 05/12/2008 162 119 -43 272,571 585,555 312,984 0.79
TWOMBLY 83 07/05/2011 172 293 121 522,921 2,000,000 1,434,302 0.62
BOHROD 84 04/03/1992 33 29 -4 2,555 3,098 544 0.03
DALI 84 01/23/1989 84 95 11 52,566 87,277 34,711 0.93
STEINBERG 84 05/12/1999 62 45 -17 10,166 12,640 2,474 0.47
APPEL 85 05/03/2006 331 587 256 37,795 67,083 29,289 0.36
DYF 85 09/15/1985 24 66 42 1,600 5,615 4,016 0.00
HARTUNG 85 12/08/1989 225 210 -15 40,694 60,436 19,741 0.15
NOLAND 85 01/05/2010 86 135 49 82,795 123,699 40,904 0.33
OLITSKI 85 02/04/2007 23 63 40 33,849 52,057 18,208 0.21
CARRENO 86 12/20/1999 89 32 -57 53,460 35,329 -18,130 0.00
HAYTER 86 05/04/1988 27 35 8 3,479 9,520 6,041 0.11
VENARD 86 12/30/1999 85 58 -27 1,782 2,675 893 0.00
ZUNIGA 86 08/09/1998 102 47 -55 6,437 6,617 181 0.00
EGGENHOFER 87 03/01/1985 53 6 -47 3,747 5,258 1,512 0.00
FROST 87 09/01/2003 103 278 175 5,608 23,859 18,251 0.03
MARCA-RELLI 87 08/29/2000 23 36 13 9,728 19,436 9,707 0.06
SOYER 87 11/04/1987 100 119 19 6,521 6,265 -256 0.10
FREUD 88 07/20/2011 142 238 96 1,100,000 867,740 -198,415 0.32
LOVELL 88 06/29/1997 16 41 25 4,594 16,507 11,913 0.00
MOORE 88 08/31/1986 92 108 16 26,545 35,222 8,677 0.31
PIPER 88 06/28/1992 47 98 51 8,810 5,799 -3,012 0.19
WIEGHORST 88 04/27/1988 24 40 16 7,341 13,840 6,499 0.04
HAMILTON 89 09/13/2011 126 172 46 29,092 29,898 807 0.25
KINGMAN 89 05/12/2000 44 21 -23 2,286 4,093 1,806 0.04
PASMORE 89 01/23/1998 12 19 7 56,256 14,284 -41,972 0.23
EISENDIECK 90 06/15/1998 40 18 -22 2,255 2,269 15 0.00
ELLINGER 90 03/24/2003 43 60 17 1,565 2,689 1,124 0.00
GRAVES 90 05/05/2001 33 26 -7 11,233 22,229 10,996 0.12
HAMBOURG 90 12/04/1999 115 123 8 7,040 8,071 1,031 0.00
MENKES 90 08/20/1986 11 27 16 4,282 3,637 -645 0.00
VASARELY 90 03/15/1997 184 200 16 9,782 9,903 121 0.43
KLUGE 91 01/09/2003 40 50 10 5,201 4,679 -522 0.00
MASSON 91 10/28/1987 106 124 18 15,385 61,290 45,905 0.50
MATTA 91 11/23/2002 190 282 92 56,459 74,552 18,093 0.28
TAMAYO 91 06/24/1991 114 116 2 117,954 233,698 115,744 0.33
CLAVE 92 09/01/2005 151 155 4 30,579 36,322 5,744 0.16
47
Age at Death date Nr. of auctions Average hammer price (US $) Reput.
Artist death mm/dd/yyyy before after diff before after diff dummy
KOONING 92 03/19/1997 79 127 48 373,222 241,892 -131,330 0.76
MARTIN 92 12/16/2004 44 37 -7 339,187 1,200,000 811,471 0.45
WYETH 92 01/16/2009 68 108 40 407,580 258,067 -149,514 0.51
BERMUDEZ 94 10/30/2008 33 32 -1 21,158 54,804 33,646 0.11
CADMUS 94 12/12/1999 48 52 4 7,047 15,339 8,293 0.09
LE PHO 94 12/12/2001 73 93 20 7,693 19,975 12,282 0.01
CASCELLA 96 08/31/1989 45 54 9 4,310 8,568 4,258 0.00
CHAGALL 97 03/28/1985 139 295 156 91,670 282,460 190,790 1.00
ERTE 98 04/21/1990 67 62 -5 5,234 4,427 -808 0.32
HIRSCHFELD 99 01/20/2003 17 82 65 4,356 6,026 1,670 0.68
ZORNES 100 02/24/2008 92 48 -44 3,317 2,057 -1,260 0.00
BERNSTEIN 111 02/12/2002 22 47 25 3,734 3,742 8 0.05
Notes: Artists are ordered by the age at death (ascending). For the three artists with evident death precursors, HARING, IMMENDORFF, and FRINK, we show both, the date of the precursor and the actual death date (in italics). The death effects of these three artists is based on their precursor date. See more details on the biographical evidence of the precursors in section 5. The reputation measure is the additive obituary variable (AOM), see more details in section 3.
48
Figure A2: Corrected auction prices against normalized year of sale by artist
(next 9 pages)
Notes: In each figure circles are corrected prices (in logs) from auctions before an artist’s death, while the squares indicate prices from auctions after an artist’s death. Prices are placed opposite the normalized year of sale, counted from the death day of each artist. Year zero is the first full year after the artist’s day of death. The dashed lines are median corrected prices (in logs) based on all-before or all-after death corrected prices, respectively. The number next to each artist is artist’s age at death, which determines the ascending order of the figures. Seven observations smaller than -4 and seven ones larger than 4 are not plotted in order to keep the common y-axis range between -4 and 4 for all 106 artists.
49
50
51
52
53
54
55
56
57
58
Table A3: Full Table 4 of the main text: Death effects based on the regression discontinuity with corrected prices
Bandwidth: 5 years Bandwidth: 4 years Bandwidth: 3 years
a a, DE*a a a, DE*a a a, DE*a Artist (1) (2) (3) (4) (5) (6) BASQUIAT 1.59*** (0.14) 1.55*** (0.12) 1.65*** (0.17) 1.59*** (0.14) 1.59*** (0.22) 1.59*** (0.15)
HARING 1.34*** (0.29) 1.36*** (0.29) 1.29*** (0.30) 1.26*** (0.29) 1.15** (0.42) 0.77** (0.23)
KIPPENBERGER -0.17 (0.32) 0.25 (0.32) -0.28 (0.24) -0.28 (0.19) -0.58* (0.26) -0.39 (0.21)
JERZY -0.03 (0.20) -0.26 (0.50) -0.03 (0.20) -0.26 (0.50) -0.03 (0.20) -0.26 (0.50)
WARHOL 0.81*** (0.25) 0.84*** (0.19) 0.83* (0.38) 0.92*** (0.20) 0.16 (0.27) 0.56** (0.15)
NESBITT -0.44 (0.28) -0.43 (0.27) -0.14 (0.27) -0.13 (0.26) -0.23 (0.39) -0.29 (0.28)
FRINK 0.63*** (0.17) 0.51*** (0.10) 0.67*** (0.16) 0.60*** (0.09) 0.64** (0.20) 0.52*** (0.04)
IMMENDORFF 0.64 (0.39) 0.22 (0.18) 0.59 (0.41) 0.19 (0.19) 0.21 (0.29) -0.08 (0.17)
RIZZI -0.44 (0.45) -0.27 (0.30) -1.23* (0.63) -0.95** (0.27) -0.88 (1.05) -1.11*** (0.09)
BEUYS 0.08 (0.70) 0.53 (0.68) 0.61 (0.65) 0.79 (0.69) 0.25 (1.01) 1.10* (0.44)
BRATBY 0.48** (0.16) 0.46** (0.14) 0.29* (0.13) 0.30* (0.13) 0.28 (0.21) 0.36* (0.17)
SCANAVINO 0.07 (0.21) 0.61 (0.40) 0.09 (0.23) 0.51 (0.41) 0.45* (0.21) 0.26 (0.29)
VOSTELL 0.30 (0.41) 0.60** (0.26) 0.16 (0.42) 0.50* (0.24) 0.78 (0.47) 0.86* (0.36)
TINGUELY 0.16 (0.35) 0.02 (0.34) 0.19 (0.36) 0.02 (0.42) 0.19 (0.47) -0.10 (0.57)
MITCHELL -0.01 (0.34) -0.02 (0.34) 0.16 (0.37) 0.21 (0.31) 0.45 (0.34) 0.50 (0.33)
SAURA 0.22* (0.10) 0.21 (0.14) 0.17 (0.14) 0.10 (0.12) 0.04 (0.12) 0.01 (0.12)
WOLVECAMP 0.16 (0.16) -0.22 (0.19) 0.16 (0.16) -0.19 (0.19) 0.16 (0.17) 0.35* (0.14)
ROTH -0.16 (0.30) -0.03 (0.45) 0.09 (0.37) -0.12 (0.48) 0.42 (0.43) 0.03 (0.61)
LUCEBERT -0.33 (0.22) -0.38** (0.13) -0.27 (0.19) -0.29 (0.15) -0.37 (0.30) -0.49** (0.16)
POLKE 0.49** (0.19) 0.41** (0.16) 0.69*** (0.19) 0.63*** (0.16) 0.66** (0.17) 0.67** (0.18)
BUFFET 0.11** (0.05) 0.11** (0.04) 0.14** (0.05) 0.11* (0.06) 0.19** (0.05) 0.17** (0.05)
FRANCIS -0.05 (0.12) -0.05 (0.14) -0.14 (0.12) -0.15 (0.12) -0.21 (0.12) -0.25** (0.09)
SAINT PHALLE 0.59 (0.39) 0.42 (0.37) 0.40 (0.40) 0.85*** (0.18) 0.03 (0.42) 0.61*** (0.13)
LICHTENSTEIN 0.29 (0.16) 0.17 (0.13) 0.23 (0.20) 0.01 (0.06) 0.16 (0.17) -0.01 (0.07)
PAIK -0.43 (1.22) -0.68 (0.91) -0.54 (1.48) -0.97 (1.10) -1.18 (1.38) -1.26 (1.24)
PORTOCARRERO -0.15 (0.30) -0.07 (0.15) 0.08 (0.24) 0.08 (0.13) 0.38* (0.17) 0.23 (0.13)
WESSELMANN 0.03 (0.11) 0.03 (0.11) -0.10 (0.13) -0.12 (0.11) -0.16 (0.17) -0.17 (0.15)
KITAJ 0.26 (0.91) -0.11 (0.54) 0.82 (0.77) 0.41 (0.69) 0.83 (0.98) -0.14 (0.30)
STAMOS 0.30 (0.30) 0.30* (0.15) 0.38 (0.34) 0.48** (0.17) 0.27 (0.42) 0.45* (0.20)
GALL -0.32 (0.21) -0.16 (0.16) -0.39 (0.25) -0.19 (0.17) -0.35 (0.23) -0.15 (0.18)
GUTTUSO -0.23 (0.17) -0.27** (0.11) -0.10 (0.20) -0.22*** (0.03) -0.01 (0.18) -0.16*** (0.01)
NOLAN 0.69 (0.75) 0.98** (0.37) -0.09 (0.89) 1.00*** (0.15) -0.45 (0.85) 0.96*** (0.12)
MOTHERWELL 0.07 (0.17) 0.04 (0.16) 0.12 (0.15) 0.10 (0.16) 0.19 (0.17) 0.08 (0.22)
SCOTT 0.74*** (0.13) 0.69*** (0.15) 0.79*** (0.18) 0.85*** (0.13) 0.24 (0.21) 0.31** (0.10)
ARMAN 0.09 (0.15) -0.11 (0.08) 0.05 (0.14) -0.09 (0.11) 0.02 (0.17) -0.19* (0.08)
CESAR 0.14 (0.51) 0.08 (0.56) 0.35 (0.63) 0.25 (0.65) 0.42 (0.99) 0.01 (0.58)
DORAZIO 0.31*** (0.09) 0.27** (0.10) 0.30** (0.09) 0.27** (0.11) 0.34** (0.09) 0.34** (0.10)
HELD -0.02 (0.58) 0.02 (0.50) 0.36 (0.79) 0.64 (0.51) 0.72 (1.03) 0.97 (0.60)
LORJOU -0.64* (0.33) -0.41 (0.33) -0.41 (0.31) -0.18 (0.35) -0.13 (0.32) 0.35 (0.18)
CHILLIDA 0.36 (0.21) 0.19 (0.16) 0.34 (0.26) -0.05 (0.19) 0.43 (0.25) 0.07 (0.18)
RIOPELLE 0.55*** (0.12) 0.53*** (0.16) 0.66*** (0.13) 0.72*** (0.09) 0.74** (0.19) 0.87*** (0.06)
RIVERS 0.27 (0.34) 0.08 (0.24) 0.14 (0.44) -0.15 (0.27) 0.52* (0.24) 0.34* (0.14)
BURRI -0.40 (0.31) -0.73*** (0.20) -0.24 (0.18) -0.40 (0.29) -0.15 (0.24) -0.36 (0.30)
DZUBAS -0.09 (0.44) -0.20 (0.43) 0.11 (0.50) -0.02 (0.57) 0.46 (0.58) 0.53 (0.48)
GUAYASAMIN -0.34 (0.24) -0.32* (0.15) -0.48** (0.17) -0.45** (0.13) -0.36** (0.12) -0.36* (0.15)
HERON 0.20 (0.23) 0.20 (0.15) -0.19 (0.17) 0.00 (0.09) 0.00 (0.15) 0.26** (0.07)
LEWITT 0.36 (0.23) 0.24 (0.14) 0.30 (0.24) 0.26 (0.17) 0.56** (0.21) 0.47*** (0.11)
MUHL -0.11 (0.15) -0.28 (0.24) -0.16 (0.19) -0.36 (0.29) 0.06 (0.15) 0.00 (0.23)
SLOANE -0.26 (0.39) -0.59** (0.23) -0.23 (0.42) -0.68*** (0.18) -0.57 (0.45) -0.99*** (0.18)
MANESSIER 0.21 (0.34) 0.20 (0.31) 0.34 (0.37) 0.34 (0.39) 0.22 (0.63) 0.09 (0.28)
SEBIRE -0.25 (0.26) -0.25 (0.23) -0.28 (0.26) -0.28 (0.25) -0.23 (0.30) -0.24 (0.26)
SHEETS -0.07 (0.36) -0.25 (0.29) -0.37 (0.27) -0.26 (0.24) -0.56 (0.40) -0.19 (0.18)
BACON -0.41** (0.16) -0.29 (0.16) -0.91*** (0.11) -0.75*** (0.08) -0.78** (0.20) -0.83*** (0.10)
GISSON -0.15 (0.16) -0.29*** (0.06) -0.06 (0.19) -0.30** (0.10) -0.11* (0.05) -0.16*** (0.01)
59
Bandwidth: 5 years Bandwidth: 4 years Bandwidth: 3 years
a a, DE*a a a, DE*a a a, DE*a Artist (1) (2) (3) (4) (5) (6) DUBUFFET -0.12 (0.16) -0.09 (0.15) 0.04 (0.17) 0.02 (0.18) -0.10 (0.22) -0.02 (0.15)
FRANKENTHALER 0.07 (0.19) 0.05 (0.15) 0.03 (0.19) -0.03 (0.18) 0.43 (0.27) 0.40 (0.20)
LEVIER -0.11 (0.11) -0.11 (0.12) -0.21 (0.15) -0.19 (0.13) -0.07 (0.20) 0.01 (0.04)
RAUSCHENBERG 0.18 (0.40) 0.19 (0.31) 0.31 (0.45) 0.30 (0.35) 0.33 (0.39) 0.31 (0.46)
TWOMBLY 0.84** (0.36) 0.82* (0.41) 0.59 (0.38) 0.46 (0.34) 0.30 (0.52) -0.05 (0.11)
BOHROD 0.29 (0.21) 0.28 (0.22) 0.20 (0.19) 0.17 (0.22) -0.10 (0.22) -0.23** (0.07)
DALI 0.17 (0.33) 0.00 (0.20) 0.20 (0.36) -0.01 (0.21) 0.11 (0.39) -0.10 (0.17)
STEINBERG 0.06 (0.13) 0.08 (0.12) -0.17 (0.10) -0.17 (0.10) 0.08 (0.24) 0.33 (0.25)
APPEL 0.23* (0.12) 0.12 (0.08) 0.12 (0.11) 0.11 (0.13) 0.24 (0.17) -0.02 (0.05)
DYF 0.25* (0.13) 0.44** (0.15) 0.07 (0.14) 0.03 (0.22) 0.18 (0.15) 0.45*** (0.08)
HARTUNG 0.47 (0.49) 0.21* (0.10) 0.39 (0.49) 0.15** (0.04) 0.30 (0.37) 0.10* (0.05)
NOLAND -0.07 (0.21) 0.05 (0.22) -0.03 (0.27) 0.21 (0.26) 0.22 (0.18) 0.25 (0.26)
OLITSKI -0.16 (0.55) -0.34 (0.48) -0.91 (0.53) -1.02** (0.38) -0.61 (0.73) -1.07*** (0.16)
CARRENO 0.51 (0.29) 0.52* (0.27) 0.41 (0.33) 0.42 (0.34) 0.45 (0.45) 0.44 (0.49)
HAYTER 0.35* (0.17) 0.26** (0.10) 0.22** (0.09) 0.24 (0.14) 0.06 (0.11) -0.01 (0.04)
VENARD -0.13 (0.22) -0.14 (0.23) -0.19 (0.27) -0.19 (0.28) -0.18 (0.36) -0.17 (0.35)
ZUNIGA -0.07 (0.10) -0.06 (0.06) -0.11 (0.09) -0.11 (0.09) -0.06 (0.13) -0.08 (0.06)
EGGENHOFER 0.18 (0.59) 0.56** (0.21) 0.18 (0.59) 0.56** (0.21) 0.42 (0.46) 0.55 (0.30)
FROST 0.10 (0.36) 0.36 (0.28) 0.53** (0.19) 0.51** (0.20) 0.79** (0.27) 0.41** (0.12)
MARCA-RELLI -0.37 (0.23) -0.26 (0.19) -0.10 (0.20) 0.13 (0.10) 0.10** (0.03) 0.08 (0.04)
SOYER -0.06 (0.14) -0.07 (0.16) -0.06 (0.16) -0.03 (0.16) 0.14 (0.09) 0.18* (0.08)
FREUD 0.10 (0.20) -0.02 (0.09) 0.12 (0.21) -0.05 (0.07) 0.13 (0.25) -0.27* (0.12)
LOVELL -0.60 (0.59) -0.58 (0.61) -0.54 (0.66) -0.53 (0.66) 0.18 (0.66) 0.29 (0.60)
MOORE 0.24 (0.30) -0.01 (0.22) 0.37** (0.15) 0.27 (0.25) 0.42** (0.16) 0.61** (0.22)
PIPER -0.31 (0.20) -0.35* (0.19) -0.03 (0.24) -0.21 (0.21) -0.26 (0.28) -0.48** (0.15)
WIEGHORST -0.34* (0.17) -0.20 (0.18) -0.33 (0.23) -0.29 (0.18) -0.05 (0.23) -0.59** (0.22)
HAMILTON 0.30 (0.40) 0.10 (0.37) 0.25 (0.40) -0.02 (0.45) 0.01 (0.29) -0.08 (0.44)
KINGMAN 0.38 (0.35) 0.61** (0.24) 0.21 (0.32) 0.32 (0.22) -0.16 (0.35) 0.15 (0.19)
PASMORE -1.44* (0.67) -1.39* (0.68) -1.78** (0.55) -1.88** (0.59) -2.11*** (0.50) -2.06** (0.55)
EISENDIECK -0.20 (0.13) -0.19 (0.15) -0.27* (0.13) -0.30*** (0.05) -0.35* (0.14) -0.29*** (0.06)
ELLINGER 0.21 (0.24) 0.33** (0.10) 0.39 (0.39) 0.14 (0.11) 0.55 (0.43) -0.06 (0.10)
GRAVES -0.90* (0.45) -0.55 (0.41) -0.76 (0.40) -0.55 (0.44) -1.35*** (0.26) -1.44** (0.38)
HAMBOURG -0.05 (0.22) 0.01 (0.17) -0.01 (0.29) 0.14 (0.14) 0.09 (0.28) 0.16 (0.18)
MENKES -0.66 (0.49) -0.87* (0.37) -0.61 (0.59) -0.99** (0.31) -0.66 (0.79) -1.21*** (0.00)
VASARELY 0.43*** (0.08) 0.45*** (0.08) 0.53*** (0.07) 0.51*** (0.04) 0.56*** (0.08) 0.53*** (0.05)
KLUGE -0.11 (0.20) -0.11 (0.20) -0.14 (0.22) -0.14 (0.24) 0.37 (0.19) 0.38 (0.20)
MASSON 0.60*** (0.18) 0.55** (0.20) 0.57** (0.23) 0.57** (0.20) 0.44 (0.28) 0.56*** (0.05)
MATTA -0.16 (0.34) -0.37 (0.22) 0.00 (0.34) -0.23 (0.24) -0.06 (0.22) -0.22 (0.27)
TAMAYO 0.52** (0.16) 0.49*** (0.15) 0.39** (0.13) 0.43*** (0.12) 0.15 (0.10) 0.22*** (0.00)
CLAVE -0.05 (0.23) -0.13 (0.18) 0.12 (0.18) 0.04 (0.17) 0.09 (0.16) 0.09 (0.15)
KOONING 0.03 (0.20) 0.10 (0.18) 0.08 (0.23) 0.23 (0.20) 0.55** (0.18) 0.72** (0.26)
MARTIN 0.61 (0.35) 0.61* (0.29) 0.52 (0.48) 0.64* (0.28) 0.71 (0.42) 0.80** (0.26)
WYETH 0.24 (0.37) 0.06 (0.25) 0.10 (0.48) -0.06 (0.31) 0.50 (0.56) 0.13 (0.22)
BERMUDEZ 0.61** (0.25) 0.76*** (0.13) 0.74** (0.23) 0.77*** (0.14) 0.65** (0.19) 0.65** (0.17)
CADMUS 0.39 (0.25) 0.25 (0.17) 0.43 (0.25) 0.39** (0.16) 0.36 (0.37) 0.26 (0.16)
LE PHO 0.65* (0.32) 0.76** (0.24) 0.22 (0.32) 0.32 (0.19) 0.07 (0.39) 0.29 (0.23)
CASCELLA 0.42 (0.40) 0.57* (0.28) 0.54 (0.48) 0.74** (0.22) 0.49 (0.57) 0.79*** (0.16)
CHAGALL 0.17 (0.16) 0.20 (0.18) 0.28 (0.18) 0.35** (0.10) 0.19 (0.21) 0.35** (0.11)
ERTE -0.14 (0.35) -0.08 (0.10) 0.18 (0.35) -0.05 (0.14) -0.02 (0.16) -0.11 (0.07)
HIRSCHFELD 0.37 (0.27) -0.10 (0.07) 0.45 (0.31) -0.06 (0.06) 0.52 (0.31) -0.03 (0.10)
ZORNES -0.18 (0.11) -0.18 (0.11) 0.07 (0.07) 0.07 (0.04) -0.04 (0.07) -0.14 (0.10)
BERNSTEIN -0.63 (0.36) -1.54*** (0.22) -0.33 (0.41) -1.06*** (0.16) -0.20 (0.31) -1.04*** (0.23)
Notes: This table is the full Table 4 of the main text. Odd columns report estimated death effects from Eq. (3), even columns from Eq. (4). Artists are ordered by the age at death (ascending). Standard errors clustered at the running variable 𝑎 are in the parentheses. * p<0.10, ** p<0.05, *** p<0.01
60
Table A4: Sensitivity death effect estimates
baseline
hammer prices corrected only
for hedonic characteristics
hammer prices corrected only for general art price
changes hammer prices diff-in-diffs differences in medians
Artist (1) (2) (3) (4) (5) (6)
BASQUIAT 1.55*** (0.12) 1.67*** (0.20) 0.34 (0.23) 0.53* (0.26) 0.99*** (0.18) 1.404***
HARING 1.36*** (0.29) 1.51*** (0.37) 1.20*** (0.22) 1.37*** (0.31) 1.09*** (0.28) 1.621***
KIPPENBERGER 0.25 (0.32) 0.18 (0.34) -0.32 (0.89) -0.38 (0.91) -0.34 (0.45) 0.545*
JERZY -0.26 (0.50) -0.30 (0.40) 0.36 (0.36) 0.25 (0.33) 0.27 (0.28) 0.051
WARHOL 0.84*** (0.19) 1.34*** (0.28) 0.09 (0.33) 0.62 (0.40) 0.55* (0.29) 1.005***
NESBITT -0.43 (0.27) -0.49 (0.30) -0.75* (0.38) -0.66 (0.38) -0.45 (0.49) 0.041
FRINK 0.51*** (0.10) -0.17 (0.13) 0.59*** (0.15) -0.13 (0.22) 0.63*** (0.11) 0.735***
IMMENDORFF 0.22 (0.18) 0.29 (0.19) 0.07 (0.46) 0.16 (0.51) 1.27*** (0.30) 0.954***
RIZZI -0.27 (0.30) -0.06 (0.27) -0.57* (0.28) -0.54* (0.25) 0.64 (0.76) 0.232
BEUYS 0.53 (0.68) 0.79 (0.57) 0.32 (0.94) 0.68 (0.92) 0.67 (0.67) 0.269
BRATBY 0.46** (0.14) 0.22 (0.16) 0.42** (0.15) 0.18 (0.18) 0.52*** (0.16) 0.329***
SCANAVINO 0.61 (0.40) 1.07* (0.53) 0.19 (0.36) 0.64 (0.39) -0.02 (0.26) 0.882
VOSTELL 0.60** (0.26) 0.60* (0.26) 1.23** (0.48) 1.21** (0.46) 0.58* (0.33) 0.263
TINGUELY 0.02 (0.34) -0.50 (0.48) 0.21 (0.50) -0.34 (0.64) 0.10 (0.20) -0.098
MITCHELL -0.02 (0.34) -0.32 (0.55) -0.02 (0.52) -0.38 (0.72) -0.02 (0.27) 0.295**
SAURA 0.21 (0.14) 0.18 (0.10) -0.42 (0.24) -0.45* (0.23) 0.23*** (0.08) 0.359***
WOLVECAMP -0.22 (0.19) -0.51 (0.43) -0.89*** (0.21) -1.35** (0.45) -0.25 (0.20) -0.096
ROTH -0.03 (0.45) -0.06 (0.38) -0.34 (0.71) -0.38 (0.64) -0.08 (0.20) -0.148
LUCEBERT -0.38** (0.13) -0.11 (0.15) -0.48** (0.17) -0.19 (0.19) -0.10 (0.18) 0.02
POLKE 0.41** (0.16) 0.37 (0.23) 1.15* (0.59) 1.14 (0.64) 0.58*** (0.22) 0.291*
BUFFET 0.11** (0.04) 0.16*** (0.05) 0.06 (0.10) 0.11 (0.08) 0.14* (0.07) 0.045
FRANCIS -0.05 (0.14) 0.27 (0.20) -0.11 (0.18) 0.18 (0.24) 0.20 (0.13) 0.036
SAINT PHALLE 0.42 (0.37) 0.45 (0.42) 0.31 (0.92) 0.36 (0.94) 0.33 (0.37) 0.231
LICHTENSTEIN 0.17 (0.13) 0.07 (0.13) 0.51* (0.24) 0.42 (0.25) 0.18 (0.18) 0.310***
PAIK -0.68 (0.91) -0.52 (1.01) -0.42 (0.98) -0.25 (1.07) -0.73 (1.00) 0.35
PORTOCARRERO -0.07 (0.15) -0.01 (0.17) 0.08 (0.14) 0.23 (0.15) -0.25 (0.34) -0.193*
WESSELMANN 0.03 (0.11) 0.15 (0.17) 0.34 (0.31) 0.49 (0.36) 0.26 (0.17) 0.642***
KITAJ -0.11 (0.54) -0.07 (0.58) -0.36 (0.61) -0.50 (0.66) 0.09 (0.59) 0.273
STAMOS 0.30* (0.15) 0.20 (0.12) 0.27 (0.25) 0.16 (0.24) 0.32 (0.33) 0.416*
GALL -0.16 (0.16) 0.18 (0.19) -0.22 (0.14) 0.12 (0.22) -0.34 (0.25) 0.147
GUTTUSO -0.27** (0.11) 0.27** (0.11) -0.46 (0.61) 0.09 (0.68) -0.21 (0.24) -0.243
NOLAN 0.98** (0.37) 0.68 (0.51) 1.49 (0.87) 0.97 (0.88) 2.82*** (0.53) 0.893
MOTHERWELL 0.04 (0.16) -0.58* (0.31) -0.52 (0.35) -1.21** (0.51) -0.23 (0.16) 0.034
SCOTT 0.69*** (0.15) 0.21 (0.20) 0.90** (0.31) 0.38 (0.35) -0.27 (0.54) 0.932
ARMAN -0.11 (0.08) 0.01 (0.13) 0.05 (0.16) 0.18 (0.20) -0.10 (0.12) 0.129
CESAR 0.08 (0.56) 0.13 (0.53) -0.68 (0.46) -0.63 (0.46) 0.14 (0.42) -0.003
DORAZIO 0.27** (0.10) 0.41** (0.14) 0.16 (0.20) 0.31 (0.23) 0.38*** (0.07) 0.599***
HELD 0.02 (0.50) 0.32 (0.54) 0.49 (0.48) 0.75 (0.48) -0.23 (0.63) 0.137
LORJOU -0.41 (0.33) -0.08 (0.32) -0.40 (0.23) 0.01 (0.16) -0.48 (0.39) -0.169
CHILLIDA 0.19 (0.16) 0.27 (0.16) 0.40 (0.41) 0.48 (0.41) 0.22 (0.26) 0.366**
RIOPELLE 0.53*** (0.16) 0.64*** (0.18) 0.63*** (0.18) 0.69*** (0.18) 0.18 (0.22) 0.600***
RIVERS 0.08 (0.24) 0.17 (0.25) 0.63* (0.34) 0.71* (0.34) 0.51 (0.35) -0.253
BURRI -0.73*** (0.20) -0.29 (0.19) -0.42 (0.90) 0.04 (0.78) 0.41 (0.43) -0.166
DZUBAS -0.20 (0.43) 0.16 (0.38) -0.88* (0.41) -0.43 (0.40) -0.15 (0.40) -0.222
GUAYASAMIN -0.32* (0.15) -0.29* (0.14) -0.46 (0.35) -0.44 (0.32) -0.24 (0.21) -0.221
61
baseline
hammer prices corrected only
for hedonic characteristics
hammer prices corrected only for general art price
changes hammer prices diff-in-diffs differences in medians
Artist (1) (2) (3) (4) (5) (6)
HERON 0.20 (0.15) 0.28* (0.14) 1.71** (0.58) 1.76** (0.59) 0.03 (0.29) 0.334**
LEWITT 0.24 (0.14) 0.22 (0.17) 0.30* (0.14) 0.26 (0.17) 0.31** (0.14) 0.510***
MUHL -0.28 (0.24) -0.36 (0.21) 0.35* (0.18) 0.19 (0.17) 0.10 (0.13) 0.101
SLOANE -0.59** (0.23) -0.31 (0.38) -0.62** (0.27) -0.31 (0.35) -0.25 (0.26) -0.058
MANESSIER 0.20 (0.31) 0.20 (0.30) -0.03 (0.34) -0.02 (0.32) -0.00 (0.24) -0.064
SEBIRE -0.25 (0.23) -0.21 (0.22) -0.50 (0.42) -0.44 (0.40) -0.24 (0.20) -0.132
SHEETS -0.25 (0.29) -0.32 (0.29) -0.53* (0.27) -0.58* (0.26) -0.51 (0.33) -0.043
BACON -0.29 (0.16) -1.25*** (0.10) -1.87*** (0.56) -3.04*** (0.62) 0.73* (0.39) -0.032
GISSON -0.29*** (0.06) -0.15** (0.05) -0.24 (0.17) -0.12 (0.14) -0.18 (0.12) -0.293*
DUBUFFET -0.09 (0.15) 0.21 (0.13) -0.40*** (0.10) -0.08 (0.11) -0.05 (0.13) 0.300***
FRANKENTHALER 0.05 (0.15) 0.04 (0.14) -0.40 (0.40) -0.35 (0.41) -0.13 (0.21) 0.171**
LEVIER -0.11 (0.12) -0.04 (0.09) -0.01 (0.10) 0.09 (0.09) -0.03 (0.16) -0.082
RAUSCHENBERG 0.19 (0.31) -0.05 (0.29) -0.70 (0.44) -0.97** (0.40) 0.21 (0.25) 0.282
TWOMBLY 0.82* (0.41) 0.84* (0.42) 1.76*** (0.31) 1.84*** (0.30) 0.73*** (0.20) 0.771***
BOHROD 0.28 (0.22) -0.01 (0.20) 0.34 (0.29) 0.01 (0.24) 0.50 (0.34) 0.476
DALI 0.00 (0.20) 0.15 (0.21) -0.17 (0.12) -0.10 (0.13) 0.01 (0.26) -0.015
STEINBERG 0.08 (0.12) 0.14 (0.11) -0.05 (0.17) 0.01 (0.15) -0.03 (0.15) 0.084
APPEL 0.12 (0.08) 0.17 (0.11) 0.16* (0.08) 0.21* (0.11) 0.09 (0.13) 0.423***
DYF 0.44** (0.15) 0.75*** (0.18) 0.60** (0.20) 0.99*** (0.22) 0.17 (0.23) 0.286**
HARTUNG 0.21* (0.10) -0.08 (0.15) 0.22 (0.13) -0.07 (0.19) 0.39 (0.34) 0.650***
NOLAND 0.05 (0.22) -0.04 (0.22) 0.31 (0.21) 0.26 (0.27) -0.26 (0.29) 0.268
OLITSKI -0.34 (0.48) -0.39 (0.46) -0.74 (0.44) -0.77 (0.43) -0.25 (0.27) 0.314
CARRENO 0.52* (0.27) 0.62** (0.21) 0.22 (0.36) 0.30 (0.32) 0.55** (0.26) 0.398
HAYTER 0.26** (0.10) 0.56** (0.23) 0.17 (0.29) 0.49 (0.38) 0.57 (0.41) 0.462**
VENARD -0.14 (0.23) -0.10 (0.19) -0.15 (0.34) -0.09 (0.33) 0.09 (0.15) 0.238*
ZUNIGA -0.06 (0.06) -0.03 (0.07) -0.19** (0.07) -0.17* (0.09) -0.15*** (0.06) 0.016
EGGENHOFER 0.56** (0.21) 1.06*** (0.09) 0.50 (0.59) 0.89 (0.59) 0.57 (0.41) 0.469
FROST 0.36 (0.28) 0.42 (0.35) 0.27 (0.28) 0.41 (0.31) 0.37 (0.40) 0.516***
MARCA-RELLI -0.26 (0.19) -0.18 (0.20) -0.97 (0.78) -0.93 (0.80) -0.64 (0.42) -0.039
SOYER -0.07 (0.16) 0.27* (0.12) -0.11 (0.33) 0.23 (0.23) -0.00 (0.23) -0.101
FREUD -0.02 (0.09) -0.14 (0.08) -0.60 (0.36) -0.61 (0.41) -0.45* (0.25) 0.290*
LOVELL -0.58 (0.61) -0.61 (0.63) -1.06* (0.48) -1.12* (0.51) 0.87* (0.48) 0.281
MOORE -0.01 (0.22) 0.40* (0.20) -0.25 (0.27) 0.22 (0.25) 0.07 (0.30) -0.356*
PIPER -0.35* (0.19) -0.68*** (0.18) 0.10 (0.23) -0.27** (0.11) -0.52** (0.25) 0.049
WIEGHORST -0.20 (0.18) 0.10 (0.27) -0.85 (0.54) -0.62 (0.65) -0.58* (0.31) -0.473
HAMILTON 0.10 (0.37) 0.15 (0.36) 0.35 (0.20) 0.39* (0.18) 0.58 (0.43) 0.029
KINGMAN 0.61** (0.24) 0.63* (0.29) 0.16 (0.37) 0.15 (0.42) 0.59 (0.48) 0.667
PASMORE -1.39* (0.68) -1.48* (0.66) -1.63*** (0.48) -1.69*** (0.45) -1.31** (0.55) -0.799
EISENDIECK -0.19 (0.15) -0.23 (0.17) -0.17 (0.26) -0.18 (0.32) -0.09 (0.16) 0.037
ELLINGER 0.33** (0.10) 0.45*** (0.12) 0.53* (0.26) 0.65** (0.23) 0.25 (0.19) 0.003
GRAVES -0.55 (0.41) -0.57 (0.43) -0.69 (0.48) -0.74 (0.49) -0.93* (0.51) -0.311
HAMBOURG 0.01 (0.17) 0.03 (0.14) -0.28 (0.24) -0.26 (0.20) -0.06 (0.15) 0.131
MENKES -0.87* (0.37) -0.83* (0.35) -1.37*** (0.37) -0.88* (0.45) -0.93 (0.72) 0.116
VASARELY 0.45*** (0.08) 0.33*** (0.09) 0.87*** (0.10) 0.77*** (0.11) 0.20 (0.13) 0.339***
KLUGE -0.11 (0.20) 0.01 (0.20) -0.18 (0.20) -0.10 (0.19) -0.35 (0.24) -0.219*
MASSON 0.55** (0.20) 0.94** (0.36) 0.47* (0.22) 0.87** (0.33) 0.73*** (0.15) 0.750***
62
baseline
hammer prices corrected only
for hedonic characteristics
hammer prices corrected only for general art price
changes hammer prices diff-in-diffs differences in medians
Artist (1) (2) (3) (4) (5) (6)
MATTA -0.37 (0.22) -0.26 (0.22) -0.19* (0.10) -0.10 (0.11) -0.24 (0.23) -0.082
TAMAYO 0.49*** (0.15) -0.28* (0.15) 0.38 (0.37) -0.38 (0.25) 0.59** (0.24) 0.651***
CLAVE -0.13 (0.18) -0.11 (0.22) 0.24 (0.44) 0.43 (0.49) -0.06 (0.19) -0.06
KOONING 0.10 (0.18) 0.02 (0.17) -0.34 (0.46) -0.45 (0.47) -0.16 (0.26) 0.053
MARTIN 0.61* (0.29) 0.72** (0.29) 1.07 (0.63) 1.21 (0.66) 0.58** (0.26) 0.479***
WYETH 0.06 (0.25) -0.27 (0.18) -0.15 (0.79) -0.50 (0.79) -0.16 (0.63) -0.43
BERMUDEZ 0.76*** (0.13) 0.53** (0.16) 0.69 (0.59) 0.43 (0.61) 0.72*** (0.25) 0.611***
CADMUS 0.25 (0.17) 0.28 (0.21) 0.72*** (0.21) 0.75*** (0.19) -0.76* (0.44) 0.305**
LE PHO 0.76** (0.24) 0.88** (0.29) 0.46 (0.33) 0.50 (0.36) -0.05 (0.53) 0.596***
CASCELLA 0.57* (0.28) 0.43* (0.23) 0.98** (0.33) 0.85** (0.29) 0.43 (0.47) 0.490***
CHAGALL 0.20 (0.18) 0.43** (0.14) 0.30 (0.23) 0.52** (0.23) 0.12 (0.21) 0.347***
ERTE -0.08 (0.10) -0.62*** (0.10) -0.12* (0.07) -0.66*** (0.13) 0.16 (0.43) 0.102
HIRSCHFELD -0.10 (0.07) 0.04 (0.04) -0.04 (0.11) 0.08 (0.10) 1.81*** (0.46) -0.093
ZORNES -0.18 (0.11) -0.21 (0.14) 0.31 (0.22) 0.10 (0.23) 0.26 (0.31) -0.268**
BERNSTEIN -1.54*** (0.22) -1.48*** (0.19) -1.91*** (0.22) -1.86*** (0.20) -0.60 (0.46) -0.470*
Notes: Baseline (column 1) and sensitivity estimates (columns 2 to 6) of death effects. The baseline estimates in column 1 are based on Eq. (4) with corrected prices. We repeat here the column 2 of Table A3 for comparison ease with the alternative estimates. In column 2 we list death effect estimates based on Eq. (4), with hammer prices corrected only for hedonic characteristics, in column 3 with hammer prices corrected only for general art price changes and column 4 based solely on hammer prices. In column 5 we list death effect estimates stemming from difference in differences estimation (Eq. 5). Finally, in column 6, we report death effects as plane differences in median prices before and after death. Standard errors clustered at the running variable a (column 1 to 4) and at artist*year*death dummy (column 5) are in the parentheses. In the last column the significance stars are based on nonparametric K-sample test on the equality of medians. In columns 1 to 5 we consider prices from the 5 years bandwidth, in the last column we only rely on prices from the 3 years bandwidth. * p<0.10, ** p<0.05, *** p<0.01
63
Table A5: Correlations between baseline death effect estimates: Baseline (RD, corrected prices, 5-year bandwidth, different trends) vis-á-vis alternatives
correlation
RD, corrected prices, 5-years bandwidth, equal trends 0.91
RD, corrected prices, 4-years bandwidth, different trends 0.90
RD, corrected prices, 4-years bandwidth, equal trends 0.78
RD, corrected prices, 3-years bandwidth, different trends 0.77
RD, corrected prices, 3-years bandwidth, equal trends 0.66
RD, hammer prices corrected only for hedonic characteristic, 5-years bandwidth, different trends 0.86
RD, hammer prices corrected only for general art price changes, 5-years bandwidth, different trends 0.76
RD, hammer prices, 5-years bandwidth, different trends 0.67
Diff-in-diffs á la Eq. 5, 5-years bandwidth 0.61
Difference in medians before and after death from 3-years bandwidth 0.78
Notes: List of correlations of baseline death effects with sensitivity estimates of death effects discussed in Sections 6 and 7 of the main text and fully reported in Tables A3 and A4. The contextual difference vis-á-vis the baseline death effects is highlighted bold. The Cronbach's alpha amounts to 0.95.
64
Table A6: Sensitivity estimates of death effects determinants. The death effect as a function of age at death and reputation (RD, corrected prices, but different time windows and trends)
de5a age at death -0.013*** -0.061*** -0.253*** -0.062*** -0.269*** -0.062*** -0.294***
(0.003) (0.016) (0.066) (0.016) (0.064) (0.016) (0.065) age at death2/100 0.034*** 0.343*** 0.035*** 0.367*** 0.037*** 0.412***
(0.011) (0.104) (0.011) (0.100) (0.011) (0.102) age at death3/1000 -0.015*** -0.017*** -0.019***
(0.005) (0.005) (0.005) reputation 0.387** 0.425*** 1.362 2.229**
(0.150) (0.143) (1.049) (1.017) reputation*age -0.012 -0.022*
(0.013) (0.012) Constant 1.171*** 2.751*** 6.300*** 2.687*** 6.500*** 2.552*** 6.716***
(0.235) (0.570) (1.308) (0.556) (1.262) (0.574) (1.254) Adjusted R2 0.158 0.219 0.275 0.260 0.326 0.259 0.341 de5 age at death -0.012*** -0.066*** -0.198*** -0.067*** -0.216*** -0.067*** -0.240***
(0.003) (0.015) (0.063) (0.014) (0.059) (0.014) (0.060) age at death2/100 0.039*** 0.251** 0.039*** 0.279*** 0.042*** 0.321***
(0.011) (0.099) (0.010) (0.093) (0.010) (0.095) age at death3/1000 -0.011** -0.012** -0.014***
(0.005) (0.005) (0.005) reputation 0.477*** 0.505*** 1.536 2.181**
(0.136) (0.133) (0.952) (0.943) reputation*age -0.013 -0.020*
(0.011) (0.011) Constant 1.086*** 2.876*** 5.313*** 2.797*** 5.550*** 2.651*** 5.751***
(0.224) (0.532) (1.243) (0.505) (1.170) (0.521) (1.163) Adjusted R2 0.139 0.231 0.258 0.307 0.344 0.309 0.358 de4a age at death -0.012*** -0.067*** -0.196*** -0.068*** -0.212*** -0.068*** -0.234***
(0.003) (0.017) (0.073) (0.017) (0.071) (0.017) (0.072) age at death2/100 0.039*** 0.247** 0.040*** 0.273** 0.042*** 0.311***
(0.012) (0.114) (0.012) (0.112) (0.012) (0.115) age at death3/1000 -0.010* -0.012** -0.013**
(0.006) (0.006) (0.006) reputation 0.377** 0.406** 1.366 1.965*
(0.161) (0.159) (1.118) (1.123) reputation*age -0.012 -0.019
(0.013) (0.013) Constant 1.078*** 2.893*** 5.267*** 2.829*** 5.481*** 2.697*** 5.666***
(0.252) (0.605) (1.430) (0.593) (1.395) (0.612) (1.395) Adjusted R2 0.117 0.192 0.210 0.226 0.251 0.224 0.258 de4 age at death -0.011*** -0.075*** -0.156** -0.076*** -0.176*** -0.077*** -0.201***
(0.003) (0.016) (0.067) (0.015) (0.064) (0.015) (0.065) age at death2/100 0.046*** 0.176* 0.047*** 0.207** 0.050*** 0.251**
(0.011) (0.105) (0.011) (0.100) (0.011) (0.102) age at death3/1000 -0.007 -0.008 -0.010*
(0.005) (0.005) (0.005)
65
reputation 0.478*** 0.497*** 1.851* 2.300**
(0.143) (0.142) (0.990) (1.002) reputation*age -0.017 -0.022*
(0.012) (0.012) Constant 1.026*** 3.151*** 4.636*** 3.070*** 4.898*** 2.887*** 5.112***
(0.237) (0.553) (1.318) (0.528) (1.253) (0.542) (1.245) Adjusted R2 0.107 0.230 0.234 0.299 0.310 0.306 0.325 de3a age at death -0.010*** -0.054*** -0.224*** -0.054*** -0.236*** -0.054*** -0.239***
(0.003) (0.019) (0.079) (0.019) (0.079) (0.019) (0.081) age at death2/100 0.032** 0.307** 0.032** 0.325** 0.031** 0.330**
(0.013) (0.125) (0.013) (0.125) (0.014) (0.129) age at death3/1000 -0.014** -0.015** -0.015**
(0.006) (0.006) (0.006) reputation 0.270 0.305* -0.172 0.510
(0.182) (0.179) (1.264) (1.271) reputation*age 0.005 -0.002
(0.015) (0.015) Constant 0.885*** 2.337*** 5.469*** 2.293*** 5.619*** 2.351*** 5.644***
(0.273) (0.666) (1.560) (0.663) (1.548) (0.685) (1.563) Adjusted R2 0.064 0.104 0.136 0.114 0.152 0.106 0.144 de3 age at death -0.008*** -0.068*** -0.214*** -0.069*** -0.226*** -0.068*** -0.221***
(0.003) (0.016) (0.069) (0.016) (0.068) (0.016) (0.070) age at death2/100 0.043*** 0.278** 0.043*** 0.297*** 0.041*** 0.289**
(0.011) (0.108) (0.011) (0.107) (0.012) (0.111) age at death3/1000 -0.012** -0.013** -0.012**
(0.005) (0.005) (0.005) reputation 0.287* 0.317** -0.560 0.005
(0.156) (0.154) (1.084) (1.092) reputation*age 0.010 0.004
(0.013) (0.013) Constant 0.815*** 2.787*** 5.458*** 2.741*** 5.614*** 2.852*** 5.576***
(0.245) (0.576) (1.349) (0.570) (1.330) (0.588) (1.343) Adjusted R2 0.054 0.159 0.189 0.178 0.214 0.175 0.207
Notes: Each panel contains in each column a separate regression where dependent variable are death effects estimated from a regression discontinuity design with corrected prices allowing either equal slope (de#) of the running variable a or different slopes on the two sides of the death threshold (de#a), where # is the indicated bandwidth. Each dependent variable is estimated against age at death and reputation in seven different specifications. Reputation is measured with the additive obituary measure. Weighted least squares are applied with the squared roots of number of observations in the respective bandwidths as analytical weights. The number of observations is 106. Highlighted columns are those already reported in Tables 6 or 7 of the main text. Conventional standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
66
Table A7: Sensitivity estimates of death effects determinants: The death effect as a function of age at death and reputation with sensitivity death effect estimates
hammer prices corrected only for hedonic characteristics
age at death -0.015*** -0.062*** -0.262*** -0.063*** -0.276*** -0.063*** -0.305***
(0.003) (0.019) (0.077) (0.018) (0.076) (0.018) (0.077)
age at death2/100 0.034** 0.355*** 0.034*** 0.376*** 0.037*** 0.428***
(0.013) (0.122) (0.013) (0.120) (0.013) (0.122)
age at death3/1000 -0.016*** -0.017*** -0.019***
(0.006) (0.006) (0.006)
reputation 0.335* 0.374** 1.539 2.442**
(0.176) (0.171) (1.235) (1.215)
reputation*age -0.015 -0.025*
(0.015) (0.015)
Constant 1.351*** 2.907*** 6.602*** 2.852*** 6.777*** 2.686*** 7.026***
(0.270) (0.662) (1.531) (0.655) (1.506) (0.676) (1.499)
Adjusted R2 0.156 0.199 0.243 0.219 0.270 0.218 0.284
hammer prices corrected only for general art price changes
age at death -0.006 -0.018 -0.172 -0.019 -0.187* -0.019 -0.202*
(0.004) (0.025) (0.107) (0.025) (0.106) (0.025) (0.109)
age at death2/100 0.008 0.255 0.009 0.279* 0.010 0.305*
(0.018) (0.168) (0.018) (0.167) (0.018) (0.172)
age at death3/1000 -0.012 -0.014 -0.015*
(0.008) (0.008) (0.009)
reputation 0.387 0.418* 0.766 1.449
(0.239) (0.238) (1.680) (1.710)
reputation*age -0.005 -0.012
(0.020) (0.020)
Constant 0.576 0.954 3.796* 0.890 3.992* 0.838 4.116*
(0.354) (0.892) (2.112) (0.886) (2.093) (0.919) (2.110)
Adjusted R2 0.011 0.003 0.014 0.019 0.034 0.009 0.028
hammer prices
age at death -0.008* -0.021 -0.199* -0.022 -0.213* -0.022 -0.234**
(0.005) (0.027) (0.115) (0.027) (0.115) (0.027) (0.118)
age at death2/100 0.009 0.296 0.009 0.317* 0.011 0.354*
(0.019) (0.181) (0.019) (0.181) (0.020) (0.187)
age at death3/1000 -0.014 -0.015* -0.017*
(0.009) (0.009) (0.009)
reputation 0.341 0.376 1.075 1.868
(0.259) (0.258) (1.823) (1.852)
reputation*age -0.009 -0.018
(0.022) (0.022)
Constant 0.763** 1.177 4.470* 1.121 4.647** 1.020 4.826**
(0.382) (0.965) (2.279) (0.962) (2.270) (0.998) (2.284)
67
Adjusted R2 0.019 0.012 0.026 0.019 0.037 0.011 0.034
diff-in-diffs
age at death -0.009** -0.046** -0.097 -0.045** -0.124 -0.045** -0.122
(0.004) (0.019) (0.085) (0.019) (0.084) (0.019) (0.086)
age at death2/100 0.027* 0.109 0.026* 0.153 0.025* 0.149
(0.014) (0.134) (0.013) (0.133) (0.014) (0.138)
age at death3/1000 -0.004 -0.006 -0.006
(0.007) (0.007) (0.007)
reputation 0.378* 0.408** -0.041 0.241
(0.192) (0.194) (1.283) (1.322)
reputation*age 0.005 0.002
(0.015) (0.016)
Constant 0.841*** 2.045*** 2.977* 1.930*** 3.360** 1.983*** 3.339**
(0.281) (0.672) (1.650) (0.663) (1.628) (0.686) (1.646)
Adjusted R-squared 0.061 0.090 0.084 0.119 0.119 0.110 0.108
differences in medians age at death -0.015*** -0.057*** -0.137** -0.057*** -0.149** -0.058*** -0.173***
(0.003) (0.014) (0.058) (0.013) (0.057) (0.013) (0.058)
age at death2/100 0.030*** 0.161* 0.030*** 0.178* 0.034*** 0.220**
(0.010) (0.092) (0.010) (0.091) (0.010) (0.092)
age at death3/1000 -0.007 -0.007 -0.009**
(0.005) (0.005) (0.005)
reputation 0.278** 0.296** 1.554* 1.979**
(0.131) (0.131) (0.904) (0.915)
reputation*age -0.015 -0.020*
(0.011) (0.011)
Constant 1.415*** 2.797*** 4.284*** 2.751*** 4.432*** 2.585*** 4.637***
(0.202) (0.486) (1.150) (0.478) (1.129) (0.490) (1.121)
Adjusted R2 0.237 0.295 0.302 0.318 0.329 0.325 0.345
Notes: Each panel contains in each column a separate regression where dependent variable are estimated death effects from five alternative approaches discussed in Section 7 of the main text. The type of death effects used is indicated top left of each panel. Each dependent variable is estimated against age at death and reputation in seven different specifications. Reputation variable is the additive obituary measure. Weighted least squares procedure is applied with the squared roots of number of observations in the respective bandwidths as analytical weights. In diff-in-diffs panel we only consider artists who died before 2007 (N=90). Otherwise the number of observations is 106. Highlighted columns are reported in Table 8, of the main text. Conventional standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
68
Table A8: Estimates of death effects determinants with alternative weighting
weights 𝑛 no weights precision precision squared
(1) (2) (3) (4)
age at death age at death -0.253*** -0.280*** -0.263*** -0.223***
(0.066) (0.073) (0.063) (0.058) age at death2/100 0.343*** 0.387*** 0.360*** 0.292***
(0.104) (0.111) (0.098) (0.092) age at death3/1000 -0.015*** -0.018*** -0.016*** -0.013***
(0.005) (0.005) (0.005) (0.005) Constant 6.300*** 6.751*** 6.457*** 5.771***
(1.308) (1.506) (1.265) (1.142) Adj. R-squared 0.275 0.214 0.275 0.313
age at death and reputation age at death -0.294*** -0.296*** -0.295*** -0.275***
(0.065) (0.072) (0.063) (0.058) age at death2/100 0.412*** 0.416*** 0.414*** 0.379***
(0.102) (0.111) (0.098) (0.092) age at death3/1000 -0.019*** -0.019*** -0.019*** -0.017***
(0.005) (0.005) (0.005) (0.005) reputation 2.229** 1.592 1.993* 2.368**
(1.017) (1.287) (1.150) (1.086) reputation*age -0.022* -0.015 -0.019 -0.023*
(0.012) (0.015) (0.013) (0.012) Constant 6.716*** 6.797*** 6.732*** 6.362***
(1.254) (1.483) (1.238) (1.114)
Joint F-test reputation [p-value] [0.003] [0.078] [0.029] [0.007]
Adjusted R2 0.341 0.238 0.311 0.366
Notes: Each panel contains in each column a separate regression where dependent variable are death effects estimated from a regression discontinuity design with corrected prices allowing for different slopes on the two sides of the death threshold (baseline death effects). Reputation is measured with the additive obituary measure. Except for column 2 weighted least squares are applied with the indicated weights as analytical weights. Column 2 reports unweighted OLS results. The number of observations is 106. Highlighted columns are those already reported in Table 6 (column 6) and Table 7 (column 10) of the main text. Conventional standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01