Going Global: Markups and Product Quality in the Chinese Art Market

This paper presents preliminary findings and is being distributed to economists
and other interested readers solely to stimulate discussion and elicit comments.
The views expressed in this paper are those of the authors and are not necessarily
reflective of views at the Federal Reserve Bank of New York or the Federal
Reserve System. Any errors or omissions are the responsibility of the authors.
Federal Reserve Bank of New York

Staff Reports

Going Global: Markups and Product Quality
in the Chinese Art Market

Jennie Bai, Jia Guo and Benjamin Mandel


Staff Report No. 614

May 2013


Abstract

We analyze two reasons for export prices to be different across markets—namely, quality
differentiation and variable markups—and attempt to parse their relative importance and some of
their underlying drivers. To overcome the substantial measurement issues in this task, we
consider a particular industry as a special case: Chinese fine art. The simplicity of the supply side
of art vis-á-vis marginal cost and the wealth of data on its quality characteristics make it possible
to separately identify the markup and quality components of international relative prices for
Chinese artworks. Through this lens, we trace the process of growth and internationalization of
Chinese art since the year 2000. We find strong support for quality sorting into international
markets at both the level of artist and artwork, as well as substantial markup differences across
destinations. Using a structural model of endogenous quality choice by Feenstra and Romalis
(2012), we argue that much of the international quality premium is driven by per unit distribution
costs (whether physical or informational) rather than destination-specific preferences for quality.

Keywords: international prices, quality differentiation, art auction

Bai, Mandel: Federal Reserve Bank of New York (e-mail: jennie.bai@ny.frb.org,
benjamin.mandel@ny.frb.org). Jia: Graduate School of Business, Columbia University (e-mail:
jguo14@gsb.columbia.edu). For helpful discussions, the authors thank Mary Hoeveler,
Volodomyr Lugovskyy, Phyllis Kao, and Thad Meyerriecks, as well as seminar and conference
participants at the Federal Reserve Bank of New York, Sotheby’s Institute of Art, Empirical
Investigations in Trade and Investment (2013), and Artelligence (2012). The authors also thank
Cullen McAndrews and Skanda Amarnath for excellent research assistance. The views expressed
in this paper are those of the authors and do not necessarily reflect the position of the Federal
Reserve Bank of New York or the Federal Reserve System.

1. Introduction

It is well-documented in the international trade and international macro literatures that exporters
set very different price levels for similar products across destination countries. This observation
has generated intense interest in the drivers of those price differences, with the idea that this
knowledge will abet our understanding of export success (in the case of international trade) and
the international transmission of shocks (in the case of international macro). In this paper, we will
focus on two reasons for export prices to be different across markets, namely quality differentiation
and variable markups, and attempt to parse their relative importance and some of their underlying
drivers.

We build on recent empirical work highlighting the role of quality differentiation in explaining
export price patterns. A representative paper in that line of inquiry is Manova and Zhang (2012),
which documents the firm-level pricing behavior of Chinese goods exporters. They find that firms
that set higher prices in a given destination have higher export revenues, suggesting that product
quality is a dominant determinant of export sales (rather than relatively low quality-adjusted
prices). They also find that for a given firm, prices vary across destinations in a manner which
suggests sending different quality levels to different markets. In a separate body of research, the
role of a firm’s markup over its marginal cost of production has been emphasized as a means for
firms to tailor prices to specific markets. Often the size of the markup is thought of as a function
of destination characteristics such as market size and the level of competition, among other factors.
One prominent example of this type of variable markup in the context of a monopolistically
competitive industry is Melitz and Ottaviano (2008).

The problem we address is that, even at the frontier of data availability (i.e., a census of firm and
product-level prices and sales volume by export destination), it is difficult to make unqualified
statements about the importance of either product quality or markups in explaining international
trade patterns. The first issue is the challenge measuring these components of price. The basic
identification problem is that only two pieces of data are typically observed, export prices and
quantities, which are each influenced by a multitude of factors including the export’s quality level,
marginal cost and markup. This quandary has led researchers to try to estimate a subset of these
factors from the available trade data, however doing so risks ignoring the influence of the omitted
factors on those being measured. The other issue is that, even with a clean measure of product
quality in hand, the underlying determinants of quality choice are also unidentified. An exporter
may choose to ship higher quality varieties to a given destination due to demand factors such as
the importing consumers’ preferences for quality, or supply factors such as the costs of producing
or shipping. Our objective is therefore to take preliminary steps to separately identify quality and
markups across markets and to investigate the determinants of quality sorting.

To fix ideas, we shall consider a particular industry as a special case: the international market
for Chinese fine art sold at auction, and we will analyze the price of Chinese art sold outside of
China versus that sold inside mainland China. This industry is well-suited to our purposes for
a few reasons. First, quality characteristics of each artwork, such as physical traits, measures of
authenticity, proxies for provenance, and characteristics of the auction transaction, are observed.
This allows us to measure and control for the different quality characteristics of each artwork
and to compare prices across markets on an apples-to-apples basis. Second, isolating the markup
component of price is greatly abetted by the fact that the relationship between marginal cost and
price for artworks can be summarized by observable data. The auction market for fine art is
largely a secondary market, and therefore the concept of the marginal cost of an artwork has to do,
if anything, with an artwork’s former sale price. We attempt to control for that in our empirical
An indicative example of this practice is the measurement of trade quality, where unit value prices are often
used as a proxy, ignoring the influence of productivity and markups. Several authors have suggested alternative measures that use information on inputs or quantities, in addition to prices, to better identify product quality. For example, Brambilla, Lederman, and Porto (2012) utilize the concentration of skilled and unskilled workers within a firm. Hallak and Schott (2011), Khandelwal (2010), and Gervais (2012) use the trade balance, import market share and export quantity, respectively, conditional on price and other controls as a measure of quality. In spite of these advances, though, the influence of variable markups on prices is rarely explicitly taken into account.

An alternative, more familiar, definition of marginal cost is as a production cost, variation in which we do not believe has an appreciable influence on art auction prices. On one hand, the artists of many works sold at auctions are no longer alive, which distorts the relationship between supply factors and output prices. On the other, for contemporary work and, as such, what is being captured in our measures of price differences across markets comes close to being a pure measure of markup differences.

In addition to observing quality and markups, Chinese art is also an industry that has
recently undergone a transition from autarky to exporting. Auction-quality artworks produced by
Chinese artists were until lately only sold at auction houses in mainland China. Beginning in the
late 1990s and early 2000s, however, the share of Chinese artworks sold at auction houses outside
of mainland China began to rise steadily, going from 2.7 percent of the number of sales in the early
2000 to a peak of 9.2 percent in 2007. The relatively well defined limits of the Chinese
art market gives rise to a very manageable scale for empirical analysis, making it at least plausible
to gather information about all auction sales since 2000 to say something about the market as a
whole. That is precisely what we do; we have assembled an almost comprehensive census of
auction transaction prices for Chinese artworks over this period sold anywhere in the world, and
including a detailed list of artwork and transaction characteristics. All in all, the dataset includes
information on over 1 million artworks put up for auction.

With this long list of sales data, we uncover a rich set of facts describing the international pricing
of Chinese art, beginning with the unconditional international price premium and telescoping
into quality adjusted estimates and estimates conditional on certain characteristics of the artwork
and auction transaction. The results can be summarized as follows: (i) internationally sold artworks
and the domestic artworks of internationally selling artists have much higher average prices
than those of artists selling only domestically; (ii) for internationally selling artists, international
works have a higher price than their domestic works; (iii) most of (i) and (ii) is explained by quality
differences between internationally sold artworks and those sold domestically; (iv) after controlling
for quality differences, internationally sold artworks still have a significantly higher price than
porary and classical artists alike, the market structure is such that marginal cost plays a relatively small role in price setting because of the high degree of differentiation among artworks. Our identifying assumption is that the elasticity of prices to production cost is zero, which is a clear difference between art and other industries (where marginal cost is an important determinant of price). For domestic artworks; and (v) the international price premium is most pronounced for auctions taking place in the United States and United Kingdom, for contemporary art, and for artists with the highest number of international sales.

The fact that international sales have higher price and higher quality than domestic sales (i.e.,
results (i), (ii) and (iii)) is closely related to the literature on quality sorting in international trade.
Perhaps the closest antecedent among those papers is Crozet, Head, and Mayer (2012), in which
the quality of French wine is observed to be higher for firms exporting to more destinations, at
higher prices, and for firms selling larger quantities in each destination; these correlations support
trade models in which firms sort into export markets according to their output quality as opposed
to their productivity. Our work also provides a measure of export quality and supports the finding
of quality sorting, but differs in several ways.

First, prices are higher for exports because higher quality outputs have higher markups, not
higher marginal costs, which differs from the common practice of modeling output quality as a
function of input cost.3 Specifying that higher output quality requires higher cost inputs is an intuitive
and reasonable assumption, but it is also done for convenience. That is because extending
models with constant elasticity preferences, such as Melitz (2003), to have quality differentiation
runs into the constraint of constant markups across firms; the only way to introduce price heterogeneity
due to quality is through the marginal cost component of price. Our results demonstrate an
important interaction between markups and quality that is not present in most models of international
trade.

Second, the quality sorting results are supportive of the Alchian-Allen (1964) effect, or the
idea that exporters “ship the good apples out.” Alchian and Allen’s conjecture was that any specific
cost applied to an export, such as a flat per unit transport cost or tariff, will shift demand
towards higher-priced, higher-quality varieties by lowering their relative price. Previous studies
supporting this mechanism for quality sorting include Hummels and Skiba (2004), who find that
export costs, taken literally as transportation costs, behave much more like specific costs than ad
valorem costs in their relationship with destination-specific export prices. Lugovskyy and Skiba
(2010) also find a significant portion of transport costs to be specific. On its face, the quality sorting
of Chinese art is consistent with Alchian-Allen, though it probably does not reflect the cost
structure of physical transportation. It could be that there are other specific costs to exporting fine
art, such as those associated with establishing an artist’s brand or acquiring information, which explain
the quality premium on international sales. As alluded to above, quality sorting may also be
driven by demand-side factors such as differential preferences for quality across markets. These
forces have been emphasized by Hallak (2006), Feenstra and Romalis (2012) and Lugovskyy and
Skiba (2010), where the preference for quality is modeled as an increasing function of importer
income. To see which effect (i.e., specific costs a la Alchian-Allen or preferences for quality) is
more prominent in our data, we will adapt the formulation of Feenstra and Romalis (2012) and
derive a structural interpretation of our international quality premium. We find that the quality
premium for Chinese art largely reflects specific costs.

For instance, Kugler and Verhoogen (2012) model endogenous improvement in the quality of labor, capital and output, given firm productivity, and find a correlation between plant size and both output and input prices. Other papers relating output quality to explicit costs include: Baldwin and Harrigan (2011), Johnson (2012), Bastos and Silva (2010), Crozet, Head and Mayer (2012), Fernandes and Paunov (2009), Hallak and Sivadasan (2011), Mandel (2009), Iacovone and Javorcik (2008), Verhoogen (2008) and Sutton (2007).

Third, albeit for only one specific industry, we provide a quantification of the relative contributions
of quality and markups to export prices. Recent studies, such as Manova and Zhang (2012),
Bastos and Silva (2010), Brambilla, Lederman, and Porto (2012) and Kneller and Yu (2008), each
find that average export prices of a given product vary systematically across destinations, in some
instances even within the same exporting firm. The fact that prices are a function of destination
characteristics such as income suggests that exporters are able to price discriminate by adjusting
their quality and/or markups. Building on those results, we show that quality dominates (nonquality-
related) markups in the determination of the international price premium for Chinese art
(result (iii)). We also demonstrate that the quality-adjusted markup differences across markets
are still large in absolute terms (result (iv)). It is noteworthy that this last result highlights the
importance of variable markups and their influence on the distribution of export prices; in standard
heterogeneous firms trade models, the firms with the lowest quality-adjusted price, due to high
productivity, are the ones that export. Here, once marginal cost and quality are held constant,
exporters still have relatively higher prices due to markups.

Fourth, thinking about Chinese artists as firms, we show that these facts are robust to singleand
multi-product producers: export prices are higher both because higher quality artists select
into exporting and because exporting artists export their higher quality artworks. In terms of the
relative contributions of quality and markups, quality is the dominant factor in both the artistlevel
and within-artist international price premium. This is consistent with the predictions of the
multi-product firm model of quality heterogeneity in Bastos and Silva (2010).

Finally, we use the depth of the dataset to try to further analyze quality and markup discrimination
across markets. It turns out that the price and quality premia are quite sensitive to the region
where the auction is located, the medium and period of the artwork, and the degree of internationalization
of the artist (result (v)). These findings are related to the modeling of international trade
prices with variable markups. In a model with linear demand, Melitz and Ottaviano (2008) show
that exporter markups and average productivity are an endogenous function of market size and
the degree of trade integration, with larger and more integrated countries characterized by higher
average productivity and lower average markups. This is the opposite of our result that larger art
markets (i.e., US and UK) and highly internationalized artists have relatively high markups, even
controlling for quality. These results support our earlier argument that there are other forces above
and beyond productivity differences that can drive large wedges in international prices. These
forces might include specific costs such as informational barriers or non-homotheticities in the
demand for quality.

The paper proceeds as follows. The next section provides an overview of the evolution of
the Chinese fine art market over the past decade and details the dataset. Section 3 describes our

empirical specification and results, including a structural interpretation of quality sorting. Section
4 concludes.

2. The Market for Chinese Art

2.1. Data

Our point of departure is a highly detailed list of the auction sales of Chinese artists sold anywhere
in the world since 2000. We obtain these data from www.artron.net, one of the largest online
databases covering auctions of Chinese artworks and antiques. Our dataset contains catalogue information
from 6,978 individual auctions that took place in 424 auction houses in China (including
Hong Kong, Macau and Taiwan) between May 1994 to September 2011, totaling 1,994,178 individual
lots and over RMB 200 billion in sales turnover. In addition, the dataset includes another
165 auctions selling works of Chinese art and antiques held at 16 auction houses outside of greater
China, totaling 39,830 individual lots and over RMB 11 billion in sales turnover.

The dataset provides information on each auction, including the name of the auction house,
the time of auction, the ordering of items within an auction, the low and high estimated prices for
each item, whether the item was sold, and how much it was sold for. In addition, the dataset also
provides information on the characteristics of each artwork, including the title, the classification of
the artwork (calligraphy, paintings, jewelry, furniture, etc.), its size and medium, the artist, the time
period in which the work was produced, and any proof of the authenticity of the work provided by
the seller.

In previous studies of art auctions, researchers have identified most of the aforementioned
observable characteristics as important factors in determining art prices. In terms of the characteristics
of the artworks, for example, the literature has found that art prices correlate positively with 1) the size of the work (to a certain extent), 2) the presence of the artist’s signature or some other signs of authenticity, 3) the prestige of the work’s provenance, and 4) the rarity in terms of its medium, style or subject matter. Furthermore, oil paintings often auction for higher prices than other media such as watercolor, presumably because of its superior durability.

In terms of the characteristics of the artist, the literature finds that the price of artworks correlate
positively with 1) the historical significance of the artist, 2) the participation of the artist in major
exhibitions, 3) the prestige of the gallery that represents the artist, 4) the popularity of the subject
matter that the artist specializes in at the time of auction, and 5) whether the artwork was produced
during the best period of the artist’s career. The nationality of the artist and the artistic style that
the artist identifies with are also important drivers of price.

Lastly, the features of the auction itself, including the prestige of the auction house, the location
of the sale, the time of the sale, the ordering of the lots, etc., are important determinants of sales
prices as well. Some studies, such as Ashenfelter and Graddy (2003) and Mei and Moses (2002),
have found that the estimated price range provided by auction houses have an anchoring effect on
the buyers and hence a positive influence on hammer prices.

In this paper, we focus on the works of art in the following classifications: calligraphy, classicalstyle
paintings, and oil paintings. These segments together account for 1,234,338 individual lots,
and over RMB 134 billion in sales, 63.7% of the total sales turnover recorded in the dataset. Among
the 1,234,338 lots, the database does not report sales price for 540,948 lots, which is an indication
that those items went up for auction but were not sold. These observations are removed from our
sample for our baseline specification. In addition, another 10,206 lots are removed because they
do not have information on the artist of the work. Finally, in the resulting 683,184 lots, we focus
on the 679,317 lots that were auctioned on or after January 1, 2000, due to the relative sparsity of
coverage prior to 2000; there are only 644 lots sold per year on average in our sample pre-2000,
compared with an average of 56,610 lots per year afterwards. In addition, the average number of
auction houses that the database covers is only 3 in the years prior to 2000, compared with 77
auction houses afterwards.

2.2. Market Overview - Volume

The past decade has witnessed a boom in the market for Chinese art, which has mirrored the boom
in China’s overall economy. Sales revenue from domestic auctions, defined as sales in mainland
China, went from RMB 96.7 million in 2000 to RMB 32.1 billion in 2010, a compound annual
growth rate of 79 percent. At 2013 values of the renminbi of 6.2 per dollar, sales grew from $15.6
million to $5.2 billion over the same period. As we will show, this increase was driven by both
quantities and prices of Chinese art sold at auction. On the quantity side, the number of works
sold at domestic auctions houses increased from 2,218 in 2000 to 48,480 in 2010, a compound
annual growth rate of 49 percent. The initial low level and blistering subsequent growth of Chinese
art sales indicate that the 2000s represented the birth of this market as a major auction category.
More broadly, China’s domestic market for all types of art underwent dramatic growth over this
period. According to Artprice’s 2011 annual “Art Market Trends” report, Beijing represented 27%
of global auction revenues for the year and finished ahead of New York and London as the leading
marketplace for art in terms of revenue.

We will pay special attention to the international aspect of this market expansion. Figure 1
(drawn from the more detailed data in Table 1) shows the growth of domestic and international sales
over the period 2000 through mid-2011, where international sales are defined as those outside of
mainland China.4 Values for both series were low and flat through 2003, when they each began to
accelerate. Domestic sales shot up to almost RMB 10 billion by 2005 and accelerated dramatically
again in 2008-11. Note that we only observe sales through mid-2011; with data through the rest of
4We treat Hong Kong as part of the international market because Hong Kong’s free-port status results in low taxes, zero tariffs on art, mild regulatory burden, and banking secrecy, which makes it significantly different from the mainland art market.

The year, 2011 would have been, quite literally, off of the chart. International sales grew steadily
through 2007 before being dented by the financial crisis in 2008-9. All in all, international sales
went from RMB 83 million in 2000 to RMB 3.5 billion in 2010, a compound annual growth rate
of 46 percent.

Table 1 drills down further into the geographic composition of Chinese art sales. The majority
of artworks by Chinese artists are auctioned in mainland China, which accounts for over 89% of the
total sales turnover and 97% of the number of works sold in our sample. In particular, the largest
domestic auction houses are concentrated in Beijing, accounting for over 71% of the total sales
and 62% of the total works sold. The largest foreign-operated auction houses, namely Sotheby’s
and Christie’s, auction the majority of their portfolio of Chinese artworks in Hong Kong, which
accounts for 9% of overall sales and 2% in terms of the number of works sold.

One can draw distinctions between the nature of domestic and international transactions along
several dimensions. First, the type of artist sold internationally tends to be quite different from
that which has sold only domestically, as illustrated in Table 2. We note that it is only a relatively
small subset of artists, about 14 percent, who are sold internationally, and that those artists account
for a disproportionate share of overall sales volume. We define domestic-only artists as those who
never had their works auction outside mainland China in our sample, whereas international artists
are those with at least one. As show in Panel A of Table 2, there are a total of 41,812 individual
artists in our sample, of which 5,881 are international. However, the median number of works sold
is much higher for international artists, at 7, relative to domestic-only artists, at 1. Figure 2 shows
the cumulative number of domestic-only versus international artists in each year. The majority of
Chinese artists never sell outside of mainland China, accounting for 58% to 87% of all artists in a
given year. Furthermore, 29% to 43% of the international artists in a given year are those who have
auctioned internationally only once. Finally, Table 3 compares the works of international artists
sold at auctions in mainland China versus their works sold in international markets. It is evident
that the majority of works by international artists are auctioned in mainland China, totaling over
RMB 101 billion in sales, or 85% of their total turnover.

A second dimension of the international nature of art transactions reflects the composition of
auction houses, acting as intermediaries in art transactions, that have sold Chinese art. Panel C
of Table 1 illustrates that mainland intermediaries grew very quickly in the 2000s, going from 6
to 128. Moreover, this growth was relatively consistent among Beijing, Shanghai and other locations.
These observations are somewhat surprising in light of the fact that revenues are highly
concentrated in China’s two flagship auction houses, China Guardian and Poly International Auction.
In contrast, sales internationally have taken place at a much smaller number of auction
houses, concentrated primarily in Hong Kong.

To be sure, Guardian and Poly have been a key role in the rise of the Chinese art market. China
Guardian started business in Beijing in 1993 and operates on a private business model similar to
Sotheby’s and Christie’s. It leads the mainland auction houses with $901.8 million, about 8 percent
of the world’s auction sales revenue in 2011. Poly was founded in 2005 and also saw its share of
global auction revenue grow to 8% in 2011. Both auction houses now rank third and fourth in the
global auction market, next to Sotheby’s and Christie’s whose combined global market share has
fallen from 73% to 47% in just ten years, according to “Art Market Trends 2011”. As a measure of
the growth in the Chinese market, even smaller auction houses, such as Beijing Hanhai and Beijing
Council, ranked among the top ten auction houses in the world in terms of revenue in 2011.

2.3. Market Overview - Prices

We will now turn our attention to art price differences across markets, which is the main focus of
the paper. The top panel of Figure 3 shows the average price level of internationally sold artworks
relative to those of artists who have only sold domestically. In addition, since we observe all of the
works of all artists, we can define a third category of artwork as the domestic sales of artists who
have sold internationally at some point during the sample. The international sales have a price level substantially higher than domestic sales; the average price of works sold in mainland China
is RMB 173,170, compared with an average of RMB 531,667 of works sold in the international
markets. The artworks of domestic-only artists have the lowest average price level, while the
domestic sales of internationally-sold artists carve an intermediate path in between international
and domestic-only. The bottom panel indexes each series to its level in 2000 to gauge the relative
growth rates of prices. International sales, in spite of their high levels and almost doubling through
2011, grew the slowest while domestic sales, led by the domestic sales of international artists, grew
the fastest. In summary, the unconditional average price of international sales was substantially
higher than domestic sales during the 2000s, but domestic sales were catching up.

One additional characterization of art prices in each market is their variance, shown in panel
E of Table 1. We note that there is a very high degree of price dispersion in all markets, with
the overall standard deviation of prices equivalent to a 184 percent deviation from their geometric
mean. At the beginning of the sample, the variance of international sales is below the variance
of domestic sales, though this changes over time as increasing price dispersion in Hong Kong
pulls the international standard deviation above the domestic one. These trends are germane
to the evaluation of art quality in the following sections and are suggestive of evolving quality
composition across markets.

Taking a step back from the data, we acknowledge that there are several shortcomings to analyzing
art markets using auction prices. First, industry reports suggest that auctions account for
less than 50% of the artworks transacted in the market, with the rest taking place in galleries and
via dealers. However, since the dealer market is highly segmented and not transparent, it is dif-
ficult to obtain comparable data. Also, as Goetzmann (1993) argues, auction data have inherent
survivorship bias as only works that do not fall out of fashion or are acquired by museums and major
private collectors can appear on the auction market. An analogous selection bias is that auction
houses (especially the larger ones) may select only the works of the highest calibre. Second, other
factors that we do not observe in our auction dataset may be influencing the price of artworks.

For example, the characteristics of buyers (e.g., motivation, valuation, art historical knowledge,
information set) can be significant drivers of prices. Pommerehne and Feld (1997) have argued
that public museums often purchase artworks at above-average prices because they tend to target
works whose calibre and historical significance are often not in question. As a result, such
works have lower risk and require a higher premium. Finally, many external forces are at play.
Renneboog and Spaenjers (2012) identify economic growth, disposable income (inequality) and
lagged equity returns as important determinants of art prices. The regulatory structure of a market
are also important. As pointed out by Plattner (1996), the tax benefits associated with donations
to cultural institutionsintheUSmayplayaroleintheselectionofartworksboughtatauction. Wedescribe
several other factors in the following section. We describe several ‘other factors’ in the following
section.

2.4. Other Factors

We conclude our description of the market for Chinese art by outlining some other concerns that
may drive differences in auction price between mainland and international transactions. First of
all, we don’t expect physical shipping costs to substantially alter the relative price of international
transactions, as anecdotes suggest that most artworks auctioned outside China are not exported
from China directly before the sale.5 Exceptions are primarily modern and contemporary artworks,
however for that group of artworks, shipping costs are typically a de minimus share of the final
auction price.

Second, the buyer’s premium (i.e., the fee charged by the auction house) is not very different
in mainland and international auction houses. Take Christie’s as an example, where the buyer’s
premium is 25% for an auction price up to $75,000, 20% for an auction price between $75,000 and

5 Artworks dating prior to the 19th century and which are auctioned in the international market had likely been taken out of mainland China before the foundation of the People’s Republic of China in 1949.

$1.5 million, and 12% for a price larger than $1.5 million.

6  By way of comparison, at the top two Chinese auction houses, Poly and Guardian, the buyer’s premium ranges between 10-25%.


3. Empirical Results

In this section, we document the average price and quality of internationally selling artworks and
internationally selling artists (including their domestic sales) relative to those of artists that only
sold in mainland China. The framework used is the hedonic regression. The marginal values of
each artwork characteristic and transaction characteristic are estimated, including physical characteristics,
auction house estimates and proxies for authenticity or provenance, among others. Whether the artwork was sold domestically (i.e., at auction in mainland China) or internationally (i.e., outside of mainland China) enters as an additional characteristic in the hedonic regression, and will be our gauge of the relative markup on international transactions. Further, we will consider two different measures of this international price premium.

In the first, we do not control for the influence of all of an artwork’s quality characteristics described above; this will be our measure of the unconditional international price premium. In the second, we do control for those characteristics. The difference between the international price premium that does not control for
quality and that which does is our estimate of how quality influences international prices (i.e., it
is an international quality premium). Finally, we also assess the effect of being an internationally
sold artist on the price and quality premia of that artist’s domestic sales.

http://www.christies.com/features/guides/buying-guide/related-information/buyers-premium

Poly’s website: http://en.polypm.com.cn/english/pmgz.php. China Guardian’s website (in Chinese):
http://www.cguardian.com/tabid/68/Default.aspx

3.1. Baseline

In our baseline model, the log price of a given artwork for sale, ln pi
ct , is a function of whether the work is sold internationally, Intl Intl, whether it is the domestic work of an internationally selling artist, Dom Intl, a vector of quality characteristics Z, and a full set of semi-annual fixed effects:  (Omitted)


The average difference in price between international and domestic artists is estimated by the
coefficients b0 and b1: b0 is the price of internationally sold artworks relative to those of artists
only selling domestically, while b1 is the price of domestically sold artworks of internationally
selling artists relative to those of artists only selling domestically. The quality characteristics
contained in the vector Z include: the type of artwork, the period it was produced, its size, among
others listed in the Appendix A.

An important additional quality characteristic is the pre-sale auction house estimate of the artworks
value which is a range composed of a low estimate and a high estimate, and which we
include as part of Z. The pre-sale range of estimates captures a wide array of value-determining
characteristics which are otherwise difficult to measure, including the significance of the artist and
artwork, the value from being sold at a particular auction house and even geographic characteristics.
It is also well-documented the low estimate tends to be closely related to the seller’s reserve
price below which the item fails to sell at auction. We exploit this fact in two ways. First, in our
baseline, the low estimate is a control variable which additionally captures variation in prices due
to the seller’s reserve. To the extent that the seller’s reserve price is related to the past sale price of
an artwork, the low estimate serves as a control for the artwork’s cost basis. Second, since models
of international trade featuring imperfect competition stress the role of the firms as price setters, in
the following section we will examine the robustness of the results using the low estimate as the
dependent variable.

Given the inclusiveness of the pre-auction estimate variable in terms of its informational content,
we do not include artist dummy variables in our baseline specification. We will shall introduce
those later on as a robustness check to control for unobserved artist characteristics. We
note, however, the fact that auction house estimates may reflect the value of being an internationally
sold artist, and hence might be collinear with b0 and b1. Such a correlation (and the implied
inclusive definition of quality) would imply that b0 and b1 are conservative estimates of the true
underlying international price premium. We do not include fixed effects for auction houses given
their collinearity with the geographically defined international dummy variables. And finally, the
semi-annual fixed effects control for general trends in the prices of Chinese fine art across all
markets.

Our baseline results are shown in Table 4. The first column shows the unconditional price
premium commanded by internationally selling artists, that is, without controlling on the quality
characteristics of each artwork. It is evident that internationally selling artists have much higher
prices than domestically selling artists and, further, that this result is driven by both the international
and domestic sales of the internationally selling artists. International sales have the rather
large price premium of 155 percent while the domestic sales of internationally selling artists have
a price premium of 100 percent. It is immediate that prices of international sales are about 50
percent higher than the domestic sales of internationally selling artists.

The second column reports results from a similar regression, though this time the quality controls
(including auction house pre-sale estimates) are added to the right-hand side of the equation.
We can interpret the price premium for international artworks in this regression as a ‘qualityadjusted’
premium since it takes into account the composition of works sold in each market. It is
evident that the premium for both internationally sold works and the domestic works of internationally
selling artists goes down considerably when quality differences are accounted for, as the
estimates drop to 28.5 percent and 14.8 percent, respectively. That said, apples-to-apples price differences
of almost 30 percent still indicate a substantial amount of international price dispersion.

We take this to mean that the markup charged in the international market is substantially higher
than in the domestic market. Moreover, the markup charged for internationally selling artist in
the domestic market is larger than for the rest of the domestic market. There could be various
reasons for this latter observation, such as some type of anchoring or signaling mechanism orthogonal
to measured quality characteristics which drives up the markups of the domestic artworks of
international artists.

One can interpret the difference between the unconditional estimates in the first column and
the quality-adjusted estimates in the second column as the influence of quality composition on
international art prices; this difference is shown in the third column entitled ‘international quality
premium.’ Reading across the first row, the results indicate that internationally selling artworks
are 155 percent more expensive than those of domestic artists, with 127 percent of this premium
due to the fact that internationally sold artworks have higher-valued quality characteristics. We
also observe that most (over 85 percent) of the premium for the domestic works of internationally
selling artists is due to quality differences, as implied by the estimates in the second row.

Overall, the results in Table 4 suggest a high degree of quality sorting into export markets:
(i) artists with higher quality artworks select into selling internationally, and (ii) the higher quality
works of those artists are the ones that sell internationally. The first of these statements is supported
by the fact that the international quality premium is positive for both international artworks and
domestic works of international artists. The fact that all types of sales of international artists
have higher quality implies that, on average, the quality of internationally selling artists is higher
than that of domestically selling artists. The second statement is illustrated by the relatively large
premium for international works. In summary, a sufficient condition for these two statements to
be true is:  International quality premium (intl.) > International quality premium (dom.) > which holds in our baseline in Table 4, as well as in the balance of robustness checks presented in the next section.
The pattern of quality and markup sorting also holds over time during this period, as shown
in Figure 4 on an annual basis. However, the magnitude of the quality premium and the qualityadjusted
premium have distinct dynamics. The quality premium was very high at the beginning of
the period, reaching a peak of 412 percent in 2001, before falling steadily to less than 100 percent
in 2007 and ticking up thereafter. In contrast, the quality-adjusted price premium grew steadily
through 2007, going from 9 percent to 45 percent before falling back to 13 percent in 2010.

3.2. Robustness

To gauge the robustness of the international price and quality premia in the baseline model, there
are several alternate specifications that we explore. First, we return to the notion that auction sales
prices may systematically differ from the reserve price of the seller. This could be introducing bias
into the measures of the international premium due to the use of auction transaction prices in the
baseline regression; to the extent that transaction prices deviate from reserve prices either more or
less in international transactions than domestic ones, the international premium would reflect that
difference in prices. Table 5 shows the resulting estimates of (1) where we have substituted the
transaction price on the left-hand side with the low estimate (a proxy for the seller’s reserve price).
Since the low estimate was previously used as a quality control variable, the specification in Table
5 also adds artist fixed effects for the top 300 artists by physical sales volume into the qualityadjusted
international premium specification. The results are broadly in line with our baseline,
though the unconditional price premium falls somewhat for both international sales and domestic
sales of international artists and the quality-adjusted estimates rise. Importantly, as before, the
quality premium indicates both across artist and within artist sorting into international sales.
In Table 6, we address the possibility that there is selection bias in our baseline sample of
artworks which only included those that consummated a sale at auction (and did not include lots
put up for auction that failed to sell). This concern is not a trivial one as roughly one third of
our 1.1 million observation sample does not have a transaction price, indicating that these works
did not end up being sold. Building on our specification using the low estimate as the dependent
variable, we proceed to run the regression over all lots brought up for auction. The estimates
shown in Table 6 are minimally different from those in Table 5, indicating that even if there are
systematic pricing differences between sold and unsold artworks at auction, they do not manifest
themselves across geographic auction locations.

In Table 7, we consider whether the churning of artists in the sample had an appreciable in-
fluence on the international premia. Given the rapid growth in the number of domestic artists
documented above (see Figure 2) there may be concern that the composition of artists is changing
in such a way that amplifies the price differences across markets. For example, if the new domestic
entrants are of relatively low price and quality, it would create the appearance of relatively high
quality international and incumbent domestic artists. We test this possibility by reverting to auction
transaction prices on the left-hand side and holding the composition of artists in the year 2000
constant over subsequent periods. Table 7 shows that doings so, despite dropping roughly half of
the sample, has little effect on the international price and quality premia of internationally sold
artworks. The domestic-international premia decline somewhat but are qualitatively similar to the
baseline.

Next we examine the international price and quality premia in particular destination markets.
To do so, we divide the international cohort into two groups: Honk Kong/Macau/Taiwan and
US/UK/Other, which roughly divides the world into the broader Asian market and the rest of the
world, respectively. Table 8 shows regression estimates for the US/UK/Other group of auction locations.
It is immediately apparent that the premia for international works, both unconditional and
quality-adjusted, is much larger than for the overall sample. Artworks sold in the US/UK/Other
fetched over three times more than works sold in mainland China. Again, the majority of this

difference (i.e., 283 percent of the 331 percent premium) is accounted for by the higher quality of
the international sales, and even controlling for quality there is a large price premium of about 50
percent. Interestingly, the domestic sales of international artists have virtually the same price and
quality premia as the overall sample, indicating that there is nothing in particular about selling in
US/UK/Other that allows internationally selling artists to charge more in mainland China.

Tables 9 and 10 duplicate the analysis for the different subgroups of medium and period, respectively.
Among the media, both the international price premium and quality premium are higher
for oil paintings than for classical-style paintings and calligraphy, though the quality-adjusted premium
is roughly the same for each medium. That is, while quality sorting seems to be stronger
for oil paintings, the quality-adjusted markup is about the same for all media. Much the same pattern
holds for contemporary versus non-contemporary art, which is in part by construction since
contemporary art is less likely to be classical-style or calligraphy.

Finally, Table 11 decomposes the price and quality premia by degree of internationalization,
which is proxied by the number of international sales for a given artist. As shown in Figure 2,
the number of artists with very few international sales is not trivial; by 2011, over half of the
cumulative number of international sales were by artists who only sold a single painting abroad
over the period. This distinction turns out to be quite meaningful for the international premia, as
the top quartile of artists by number of international sales have substantially higher relative price
and quality than the bottom quartile. For the most internationalized artists, the selection of high
quality artists into international markets is also the strongest, as evidenced by the relatively close
estimates of the quality premium of international versus domestic works of international artists.
In other words, the artists with the highest quality works sell more works abroad, but at a similar
quality and price level to their domestic sales.

3.3. A Structural Model of Quality Sorting
Having shown evidence for quality and markup differentials inside versus outside of China, in this
section we attempt to rationalize the results through the lens of theories that give rise to quality
sorting. To uncover the drivers of the quality premium, we shall follow the structural approach of
Feenstra and Romalis (2012) and adapt their model of endogenous quality choice. Their framework
features specific transport costs and non-homothetic preferences for quality, a structure that
allows for the identification of the quality component of average export prices in the trade data.
Given that we observe relative trade quality from our empirical estimates, we shall invert their
model equations to identify the parameters governing specific costs and preferences for quality.

Tables, figures and formulae omitted.