01 November 2103
Pre-sale information and auction prices for Australian Indigenous artworks*
Lisa Farrella, b and Tim R.L. Fryb
a Department of Strategy and Marketing
The University of Huddersfield, United Kingdom
b School of Economics, Finance and Marketing RMIT University, Australia
Abstract: The catalogue for art auctions conveys information to prospective buyers. We investigate the role that pre-sale information plays in determining auction prices for Australian Indigenous artworks.
Keywords: Pre-sales information, auction prices, Indigenous artwork.
JEL Classification: Z11
Corresponding Author: Prof. Tim R.L. Fry, School of Economics, Finance & Marketing, RMIT University, GPO Box 2476, Melbourne, Victoria 3001, Australia. Email: [email protected]
______
* This work is part of a research project on “Indigenous art at auction” which has been supported by RMIT University funding. We are grateful to Jane Fry for her comments on an earlier version of this paper. The usual disclaimer applies.
1
1. Background This paper examines the role of pre-sale information in explaining the hammer price
for artworks at auction. The focus of our study is works created by Australian
Indigenous artists. Indigenous art is of international relevance with collectors spread
across the globe. Art auction data typically contain many types and styles of artwork,
making the data inherently heterogeneous. We reduce this variability by focusing on
analyzing Indigenous artworks as three categories/groups, namely objects, paintings
and works on paper. Our primary attention is upon information contained in the
auction catalogue but our empirical modeling also controls for the prior information
concerning the auction market for Indigenous artworks as a whole.
The catalogue for an art auction contains information for potential buyers prior to the
auction taking place. Two key pieces of information in such catalogues are a lower
(L) and upper (U) estimate of the likely hammer price for the artwork. In prior
research these two have usually been combined into a single value to give a pre-sale
price estimate. Typically the pre-sale estimate is taken as the midpoint between the
lower and upper estimates1 (M = (L+U)/2). Ashenfelter and Graddy (2006) have
shown that if we view L and U as lower and upper bounds on the true price µ (L = µ –
kσ, U = µ + kσ, k > 0) then M provides a reliable estimate of the sale price.
Moreover, in this framework the range R = U – L is an estimate of the degree of
uncertainty (2kσ) in the auction house pre-sale estimate. Thus L and U convey
information about both the expected hammer price and the degree of (un)certainty
surrounding that estimate.
Ashenfelter, Graddy and Stevens (2002) argue that the ratio of the two pre-sale
estimates (the upper and lower bound) may also convey information. In particular, in
art auctions the convention is that the seller’s reserve price is set to be a proportion –
usually 0.8 – of the lower estimate (L) in the catalogue. In situations when the seller
requires a higher reserve than this implies, Ashenfelter, Graddy and Stevens (2002)
suggest that the value for L will be inflated whilst the value of U will remain
unchanged. In such a case the ratio (S = (U/L)), termed spread, will be compressed.
1 An exception is Bauwens and Ginsburgh (2000) who use L and U separately.
2
Thus, the ratio of the two pre-sales estimates may itself convey information to
potential buyers concerning the seller’s reserve if the ratio on one item is very
different to the ratios for other items at that auction.
In our modeling we are interested to know what role, if any, the midpoint (M), the
range (R) and the spread (S) play in explaining the actual hammer price (P) at auction.
A further piece of information that may play a role concerns the state of the auction
market for Indigenous art prior to the sale. To capture this information we use the
Herfindahl index (H) for the market in the previous year as a measure of how
competitive the market has been. The Herfindahl is calculated using the volume of
Indigenous works offered for sale by auction house.
The plan of this paper is as follows. In the next section we discuss our methodology.
Section three describes the data used and presents some descriptive analysis of the
data. The results and discussion are then presented in section four and section five
presents our conclusions.
2. Methodology
Empirical (regression) modeling is used to understand what, if any, role the pre-sale
information plays in explaining the hammer price for artworks sold at auction. In our
analysis we regress the hammer price (P) of an artwork on pre-sale information from
the auction catalogue – midpoint (M), range (R) and spread (S) – and on the
Indigenous auction market last year (H).
The traditional ordinary least squares (OLS) regression model is a model for the
conditional mean of price (P). Alternative specifications exist that model the
conditional quantiles (or percentiles) as depending upon the set of explanatory
variables. One example is a median regression that yields least absolute deviation
estimates. These alternative specifications can be also used to investigate robustness
to the conditional mean parameterization or as an alternative way to investigate the
dependence of the conditional distribution on explanatory variables. Thus we estimate
the model using both OLS and (simultaneous) quantile regression estimation.2 We
2 Estimation was carried out in Stata 13 using the sqreg command and standard errors were computed using 200 bootstrap replications.
3
report robust standard errors (for OLS) and bootstrapped standard errors for quantile
regression.
The model specification for the price of artwork i = 1, …, n is given by3:
𝑃! = 𝛽! + 𝛽!𝑀! + 𝛽!𝑅! + 𝛽!𝑆! + 𝛽!𝐻! + 𝑢!
In this framework we can use hypothesis tests on the individual coefficients to
determine whether the midpoint, range, spread or Herfindahl are significant in
explaining the price.
The intercept 𝛽! may be interpreted as the fixed amount by which the hammer price is
over or under estimated. If the midpoint (M) is a good estimate of the price then
𝛽! = 1. It is possible that M could under or over estimate the true price. Increased
uncertainty as measured through larger values for the range variable (R) may also
reflect higher risk. The idea of “high risk, high return” would then suggest that the
coefficient on R is positive. Alternatively, the auction house and the buyer(s) may
have differing levels of knowledge on risk or different risk preferences. In which case
the sign on R is unknown. Reductions in spread (S) could reflect an increase in seller
reserve. As spread falls, the hammer price may increase through the effects of an
increased seller’s reserve price. Thus we might expect the coefficient on S to be
negative. Increased competitiveness in the market for Indigenous art is reflected in a
lower value of H and is likely to lower the hammer price, so 𝛽! would be positive. If
no “higher order” information matters then the coefficients on R, S and H are all zero.
In this first order information world, a zero intercept and unit slope suggests that the
midpoint is the best estimate of the true sale price.
3. Data and Descriptive Analysis
Our data covers 29,812 artworks by Australian Indigenous artists offered for sale at
auction from 1969 to 2011 as recorded by Australian Art Sales Digest (AASD). We
are concerned with understanding the role of pre-sale information on the hammer
price and thus restrict attention on those artworks that were sold at auction4. Our
3 The regression model is estimated separately for each category of artwork. 4 Ninety eight per cent of all artworks have pre-sale price estimates. The sale rates are 69.3% for objects, 57.8% for paintings and 78.1% for works on paper.
4
analysis looks at the three main classes of Indigenous artwork – objects, paintings and
works on paper5. We begin with some descriptive analysis of our data (tables 1 – 3).
Table 1: Descriptive Statistics for Objects Price Midpoint Price - Midpoint Range Ratio Mean 4,831 4,447 384 1,654 1.508 Median 2,300 2,500 -200 1,000 1.500 Standard deviation 10,132 8,626 3,762 2,912 0.225 Minimum 30 40 -15,000 20 1.091 Maximum 190,000 175,000 40,000 50,000 3.000 Skewness 10.17 11.71 4.06 8.76 1.78 Kurtosis 158.27 208.50 29.84 117.80 7.26 Sample size 757.
Table 2: Descriptive Statistics for Paintings Price Midpoint Price - Midpoint Range Ratio Mean 9,792 10,381 -589 3,450 1.495 Median 3,000 3,500 -450 1,500 1.500 Standard deviation 33,048 33,287 10,932 11,123 0.374 Minimum 20 25 -245,000 10 1.037 Maximum 2,000,000 2,150,000 385,000 700,000 15.000 Skewness 25.07 27.74 7.91 27.25 23.71 Kurtosis 1223.69 1520.58 326.57 1405.20 835.76 Sample size 11,645.
Table 3: Descriptive Statistics for Works on Paper Price Midpoint Price - Midpoint Range Ratio Mean 2,784 2,676 108 864 1.545 Median 475 500 -50 200 1.500 Standard deviation 6,511 5,871 2,235 1,945 0.306 Minimum 15 30 -15,625 10 1.053 Maximum 120,000 70,000 50,000 30,000 4.000 Skewness 5.03 4.06 6.39 5.37 2.01 Kurtosis 43.45 22.87 98.68 45.07 6.69 Sample size 4,744.
For each type of artwork, all variables in the data are highly variable, positively
skewed and heavy tailed. We also see that on average the hammer price and midpoint
5 The “text fields” in our database reveal that almost all Works on Paper are actually (watercolour) paintings.
5
are different, with the hammer price exceeding midpoint for objects and works on
paper but midpoint exceeding the hammer price for paintings One variable that is
remarkably similar across all classes is the spread (or ratio) variable. The average in
all classes is close to 1.5 suggesting that the upper bound (U) is, on average, some
50% higher than the lower bound (L). However, there is a wide range of values for
the ratio suggesting that there is some movement away from the 50% mark up in
practice.
If the midpoint is a good estimate of the hammer price then the difference between
them should appear random and centered around zero. Histograms for the three
classes of art show that the distribution of the price difference is unimodal but
positively skewed, which results from a few works — particularly in the category of
paintings — having hammer prices much larger than their midpoint (figures 1 – 3).
Figure 1
7
Using the AASD data we also derive a value for the Herfindahl index for the market
for Indigenous artworks. The index is derived using the volume of art offered for
auction by each auction house each year. Figure 4 shows the value of the index over
time.
The index shows that since the late 1980s the market for Indigenous artworks has
become much more competitive. This is primarily due to a rapid expansion in the
number of auction houses offering Indigenous artwork for sale.
4. Results and Discussion
We now turn to the results of our regression modelling (tables 4 – 6).
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
1969
19
71
1973
19
75
1977
19
79
1981
19
83
1985
19
87
1989
19
91
1993
19
95
1997
19
99
2001
20
03
2005
20
07
2009
20
11
Figure 4: Herfindahl Index over time
8
Table 4: Regression Results for Objects Constant Midpoint Range Spread Herfindahl R2 O.L.S. 408.0123 0.8566 0.7274 -681.5942 2368.272 0.8714 (1250.056) (0.2875) (0.9124) (895.446) (1924.611) Percentiles
10% -313.5245 0.7895 -0.3291 62.1138 421.3875 0.4745 (243.6287) (0.0731) (0.1800) (149.8452) (402.0111) 25% -169.5268 0.8328 -0.1963 -32.6413 552.2031 0.5349 (303.9005) (0.0820) (0.2166) (184.6789) (361.5536) 50% -0.4857 0.9501 -0.2758 -27.3475 148.6625 0.5689 (486.1514) (0.1877) (0.4636) (275.8669) (439.6199) 75% 306.2492 0.8738 1.1121 -225.4167 164.7635 0.6256 (429.0587) (0.2130) (0.6129) (248.3850) (638.7488) 90% 911.0540 0.9829 1.9386 -486.6893 201.8290 0.6944 (756.9173) (0.3304) (0.9894) (409.5594) (2177.276) Sample size 757. Standard errors are in parentheses.
Table 5: Regression Results for Paintings Constant Midpoint Range Spread Herfindahl R2 O.L.S. 545.863 0.8784 0.1880 -1162.693 6446.001 0.8949 (1408.379) (0.0902) (0.2711) (907.657) (1302.175) Percentiles
10% 200.5308 0.6409 0.0014 -294.9670 397.0334 0.5439 (212.7534) (0.0486) (0.1540) (164.6029) (103.7913) 25% 198.4446 0.7186 0.0450 -264.5125 452.0947 0.6372 (164.5397) (0.0255) (0.0700) (107.7302) (107.4625) 50% -108.8832 0.8737 -0.1305 -6.2422 394.9654 0.7044 (69.2752) (0.0247) (0.0719) (38.7772) (95.3468) 75% 41.9553 0.8331 0.5828 -117.5168 778.3444 0.7528 (101.9833) (0.0328) (0.1317) (47.9524) (163.0120) 90% 45.7143 0.9499 1.2775 -191.1334 1687.1040 0.8054 (238.5128) (0.1255) (0.3409) (122.7440) (431.1193) Sample size 11,645. Standard errors are in parentheses.
9
Table 6: Regression Results for Works on Paper Constant Midpoint Range Spread Herfindahl R2 O.L.S. 9.9350 1.0065 0.1157 -82.7372 444.844 0.8839 (220.971) (0.0829) (0.2793) (127.016) (132.514) Percentiles
10% -109.3168 0.7609 -0.1269 19.4899 42.1361 0.5572 (21.3728) (0.0308) (0.0912) (13.8591) (14.0717) 25% -51.7364 0.8966 -0.2904 10.7756 34.7271 0.6410 (20.7919) (0.0232) (0.0514) (11.6528) (8.2583) 50% -48.3295 1.0011 -0.3026 28.7919 34.0195 0.7092 (27.3557) (0.0420) (0.1373) (19.8452) (14.5580) 75% 5.3560 1.1215 0.2704 -18.6893 79.3325 0.7947 (47.1146) (0.0841) (0.2574) (34.3464) (25.4561) 90% 82.0653 1.3089 0.7220 -52.0422 93.4877 0.8495 (60.4107) (0.1152) (0.3350) (41.7924) (70.0857) Sample size 4,744. Standard errors are in parentheses.
Turning first to the OLS results we see that for all three classes of artworks the
constant and coefficients on range and spread are not statistically significant. In all
three cases the coefficient on the midpoint is significant and the hypothesis that it
equals one is accepted. This suggests that only the midpoint (M) obtained from the
catalogue explains the hammer price and that it is a good estimator of the price. We
also see that the Herfindahl index, reflecting overall market information (that is,
competitiveness of the market), is significant and positive both for paintings and for
works on paper. Thus, an increase in competition in the indigenous auction market
lowers the hammer price for paintings and works on paper. However, for objects that
tend to be more heterogeneous in nature than paintings and works on paper
competition has no significant effect on price.
Results of a least absolute deviation (median or 50%) regression are very similar to
the OLS results, with two exceptions. First, for paintings, the coefficient on the
midpoint is not equal to one and the estimate indicates price tends to be less than the
midpoint. This may indicate that potential buyers of works on paper are more risk
averse than potential buyers of objects and paintings.
Looking at the results across the conditional percentiles it is clear that the constant
and spread are never statistically significant in the estimation. The coefficients on the
midpoint increase in magnitude and indicate that, for objects and paintings, the
10
midpoint becomes a more reliable estimate of the hammer price as we move up the
percentiles. For works on paper, above the 50th percentile the midpoint increasingly
overestimates price.
For paintings and works on paper, the range increases in magnitude and for works on
paper the range also changes sign from negative to positive across the percentiles.
One potential explanation for this may be that at the lower percentiles of the
conditional distribution buyers are risk averse but at the higher conditional percentiles
high risk is associated with high return (and price). The results on the market
information (Herfindahl) are consistent with the OLS results. Thus, an increase in
competition lowers the hammer price and the reduction is larger for the higher
percentiles (except for objects where the Herfindahl has no significant effect).
5. Conclusions
This paper has investigated the role that pre-sale information plays in explaining the
hammer price for three classes of Indigenous artworks (0objects, paintings and works
on paper). The auction catalogue contains two important pieces of information – a
lower and an upper estimate of the hammer price. These pieces of information can be
combined to form a single price estimate (the midpoint M), an estimate of uncertainty
in the estimation of price (the range, R) and the spread of estimates (the ratio or
spread, S). The results suggest that the midpoint M is significant in explaining price
and is a reliable price estimate. The spread does not play a role and the results on the
range (uncertainty) are mixed. We also investigate the potential role of information
concerning the Indigenous art auction market as a whole. We find that this is
significant in explaining the hammer price of both paintings and works on paper.
11
References Ashenfelter, O., Graddy, K., (2006). “Art Auctions”, in: Ginsburgh, V., Thorsby, D.
(Eds.), Handbook of the Economics of Art and Culture, Volume 1. Elsevier B.V., Amsterdam, 909–945.
Ashenfelter, O., Graddy, K., Stevens, M. (2002). “A study of sale rates and prices in
impressionist and contemporary art auctions”. Unpublished manuscript. Australian Art Sales Digest, http://www.aasd.com.au/ Bauwens, L., Ginsburgh, V., (2000). “Art Experts and Auctions: Are pre-sale
estimates unbiased and fully informative?”, Recherches Economiques de Louvain, 66(2), 131–144.