1 
 
The price of wine
†
 
Elroy Dimson
a,b,*
, Peter L. Rousseau
c
, and Christophe Spaenjers
d
 
a
 Cambridge Judge Business School, Trumpington Street, Cambridge CB2 1AG, United Kingdom 
b
 London Business School, Regent’s Park, London NW1 4SA, United Kingdom 
c
 Vanderbilt University, Box 1819 Station B, Nashville, TN 37235, United States 
d
 HEC Paris, 1 rue de la Libération, 78351 Jouy-en-Josas, France  
 
Abstract  
Using historical price records for Bordeaux Premiers Crus, we examine the impact of aging on wine 
prices and the long-term investment performance of fine wine. In line with the predictions of an 
illustrative model, young maturing wines from high-quality vintages provide the highest financial 
returns. Past maturity, famous châteaus deliver growing non-pecuniary benefits to their owners. Using 
an arithmetic repeat-sales regression over 1900–2012, we estimate a real financial return to wine 
investment (net of storage costs) of 4.1%, which exceeds bonds, art, and stamps. Returns to wine and 
equities are positively correlated. Finally, we find evidence of in-sample return predictability. 
 
JEL classification: C43; D44; G11; G12; Q11; Z11. 
Keywords: Wine prices; Alternative investments; Price indexes; Psychic return; Bubbles.
                                                          
†
 We are grateful to Bill Schwert (the Editor), Jianping Mei (referee), and another anonymous referee for helpful feedback 
and suggestions. We also thank Simon Berry and Carol Tyrrell of Berry Bros. & Rudd, David Elswood and Jeff Pilkington 
of Christie’s, Robert Arnott, Orley Ashenfelter, David Ashmore, David Chambers, Géraldine David, Engelbert Dockner, 
Thierry Foucault, Will Goetzmann, Jason Hsu, Boyan Jovanovic, Roman Kräussl, Stefano Lovo, Raphael Markellos, Paul 
Marsh, Pedro Matos, Victor Niederhoffer, Ludovic Phalippou, Raghavendra Rau, Luc Renneboog, Mike Staunton, David 
Thesmar, Viktor Tsyrennikov, Patrick Verwijmeren, Michaël Visser, Mungo Wilson, Josef Zechner, and participants at the 
American Association of Wine Economists 2013 conference, American Economic Association 2013 conference, 
Eurhistock 2013 conference, European Summer Symposium in Financial Markets 2013, Luxembourg Asset Management 
Summit 2014, Maastricht Art Market Symposium 2013, Norwegian Financial Research Conference 2014, Q-Group 2014 
seminar, Research Affiliates Advisory Panel 2014, World Investment Forum 2014, and seminars at Cambridge Judge 
Business School, East Anglia, Gothenburg, HEC Paris, Luxembourg School of Finance, Norwegian School of Economics, 
Saïd Business School, and the Spängler IQAM Research Center for data and comments. For financial support, we are 
grateful to the Newton Center for Endowment Asset Management at Cambridge University (Elroy Dimson) and 
ANR/Investissements d'Avenir under grant ANR-11-IDEX-0003/Labex Ecodec/ANR-11-LABX-0047 (Christophe 
Spaenjers). We have no financial relation with organizations that could benefit from our findings. All errors are ours. 
*
 Corresponding author. Tel.: +44 20 7000 8212; fax: +44 700 607 7390. 
E-mail addresses: ed351@cam.ac.uk (E. Dimson), peter.l.rousseau@vanderbilt.edu (P.L. Rousseau), spaenjers@hec.fr (C. 
Spaenjers). 
2 
 
1. Introduction 
Among wealthy individuals, fine wine is a popular investment. About one-quarter of high net 
worth individuals around the world own a wine collection, which on average represents 2% of their 
wealth (Mitchell, 2012). Many wine funds have sprung up to satisfy the increasing demand to invest in 
fine wines. In light of the long-standing yet rising status of high-end wines as an investment—and given 
the debate on the role of alternative investments in portfolio choice more generally (e.g., Swensen, 2000; 
Ang, Papanikolaou, and Westerfield, 2014)—a study of long-term price trends in this market and a 
comparison with more traditional assets is timely. 
A small literature exists on the returns to storing wine, but the findings are mixed and depend on 
the period being investigated. Based on four years of auction data, Krasker (1979) finds that average 
returns to holding red Bordeaux and California wines are no larger than returns on Treasury bills after 
transaction costs. Jaeger (1981) expands the time frame by four years and finds the opposite. Later 
studies apply more sophisticated methods for constructing prices indexes, but they also work with 15 
years of data or less. Burton and Jacobsen (2001), for example, estimate returns on red Bordeaux wines 
from 1986 to 1996 and find returns to be low and relatively volatile. Examining the subsequent decade, 
Lucey and Devine (2015) find that their Bordeaux and Rhone wine indexes yield returns in excess of 
Treasury bills and with risk below the stock market. Masset and Weisskopf (2010) study a number of 
wines from 1996 to 2009 and conclude that adding wine to an investment portfolio can increase its 
return while lowering risk. Kourtis, Markellos, and Psychoyios (2012) reach similar conclusions in a 
study of wine prices from 2001 to 2010. 
By considering historical prices over many decades, we bring a longer-term perspective to 
studying the price dynamics of fine wine. Our work is in the spirit of recent research on the performance 
of other emotional assets such as art (e.g., Goetzmann, 1993; Mei and Moses, 2002), stamps (Dimson 
and Spaenjers, 2011), or violins (Graddy and Margolis, 2011). It can also be compared with studies of 
3 
 
long-term equity and bond returns (e.g., Schwert, 1990; Siegel, 1992; Jorion and Goetzmann, 1999; 
Dimson, Marsh, and Staunton, 2002) and vintage effects in equities (Jovanovic and Rousseau, 2001).  
Furthermore, we investigate how aging affects wine prices independently of changes in market 
conditions. Identifying the effects of aging requires separating them not only from time effects but also 
from effects related to particular vintages, and this is another dimension upon which our contribution is 
unique. A few studies on cross-sectional variation in wine prices show that older wines tend to 
command higher prices (Di Vittorio and Ginsburgh, 1996; Ashenfelter, 2008) but do not separate effects 
of vintage quality from age.  
One reason that it is interesting to look at the effects of aging on prices and returns is that even 
wines which have lost their gastronomic appeal can be valuable if they provide enjoyment and pride to 
their owners. By estimating life-cycle price patterns, we examine if and when nonfinancial ownership 
dividends codetermine price levels for well-known wines. Considering such non-pecuniary benefits 
along with pure financial returns is relevant from a broader asset pricing perspective. For example, 
nonfinancial utility could also play a role in markets for entrepreneurial investments (Moskowitz and 
Vissing-Jørgensen, 2002), prestigious hedge funds (Statman, Fisher, and Anginer, 2008), socially 
responsible mutual funds (Bollen, 2007; Renneboog, Ter Horst, and Zhang, 2011), social, 
environmental, and ethical portfolios (Dimson, Karakaş, and Li, 2014), and art (Stein, 1977; Mandel, 
2009). Heinkel, Kraus, and Zechner (2001) and Hong and Kacperczyk (2009) show that the non-
pecuniary disadvantages associated with holding particular assets could also affect expected returns. 
We begin by presenting a simple and stylized model of price dynamics that accounts for 
fluctuations in a famous wine’s consumption value and attractiveness as a collectible over its life. The 
model proposes that, in general, a wine’s fundamental value is governed by the maximum of three 
measures: (1) the value of immediate consumption, (2) the present value of consumption at maturity plus 
the nonfinancial ownership dividends received until consumption, and (3) the present value of lifelong 
4 
 
storage (i.e., the value as a collectible). The model ties the values of consumption and ownership 
dividends to financial wealth, which reflects the discretionary nature of luxury goods (Goetzmann and 
Spiegel, 1995; Aït-Sahalia, Parker, and Yogo, 2004). It also implies that, abstracting from changes in 
quality, the price appreciation of wines over time is determined by the growth rate of wealth. Cross-
sectionally, the model delivers different predictions for the price patterns of low-quality and high-quality 
vintages of superstar châteaus over their respective life cycles. The consumption value of a low-quality 
vintage declines quickly after bottling and leads to a fall in prices. This persists until the present value of 
the enjoyment associated with infinite ownership of the wine (storing without the goal of ever drinking) 
exceeds that of consumption, at which point prices start rising with age. Prices of high-quality vintages, 
which improve in quality after bottling, rise strongly until maturity and then stabilize. Eventually, as 
these wines begin to be regarded as collectibles instead of consumption goods, prices advance again. For 
all wines, financial returns reflect the effects of both wealth growth and aging on prices. The expected 
financial return on wine is always below the appropriate discount rate because the nonfinancial 
ownership dividends received while storing a bottle endogenously lower the required capital gain. This 
is especially relevant for wines that are long beyond maturity, as their fundamental values are 
determined by their value as collectibles (i.e., by the future stream of ownership dividends) and not by 
their consumption value.  
We next describe a unique historical database of prices for five long-established Bordeaux wines: 
Haut-Brion, Lafite-Rothschild, Latour, Margaux, and Mouton-Rothschild—the Premiers Crus or First 
Growths. We construct this database using two types of price information: transaction prices realized at 
auctions organized by Christie’s London and retail list prices of the London-based wine dealer Berry 
Bros. & Rudd (BBR). The data are hand-collected from various sources, including archived auction 
catalogues, dealer price lists, and company publications and websites. The database contains 36,271 
5 
 
prices for 9,492 combinations of sale year (e.g., 2007), château (e.g., Latour), vintage year (e.g., 1982), 
and transaction type (dealer or auction) from end-1899 through end-2012.  
We then use this new data set to study the returns to holding wine and the effects of aging on wine 
prices. We measure vintage effects by taking account of both variations in yearly production and a time 
series of weather data. The life-cycle price dynamics implied by the coefficients on age and its 
interactions with vintage quality are generally consistent with our model. High-quality vintages 
appreciate strongly while maturing for a few decades, but then prices stabilize until the wines become 
antiques, after which prices start rising again. For low-quality vintages, prices are almost flat over the 
first few years of the life cycle, but then rise in a near-linear fashion. The observation that prices of 
wines beyond maturity continue to rise with age points to the existence of a nonfinancial payoff from the 
ownership of relatively rare bottles of a well-known château. At the same time, our measurements of the 
financial under-performance of collectible wines compared with maturing wines suggest that the psychic 
return realized by wine collectors is probably small.  
These results are robust to a variety of alternative specifications. Moreover, when examining the 
relation between vintage quality and the number of auction transactions over wines’ life cycles, we find 
that trading volume falls much more quickly for low-quality vintages than for high-quality ones. This 
offers further evidence that wines from low-quality vintages rapidly become rare and can explain why 
they can have positive financial returns even as they become undrinkable. 
To estimate the financial returns to wine investment over the long run, we apply a value-weighted 
arithmetic repeat-sales regression to all price pairs (i.e., combinations of prices for the same château and 
vintage year at different times) in our data. The resulting index picks up both the effects of aging on 
prices and the time series changes in the willingness to pay for wine. We find that inflation-adjusted 
wine values did not increase over the first quarter of the 20th century, experienced a boom and bust 
around the Second World War, and have risen substantially over the last half century. Overall, we find 
6 
 
an annualized real return of 5.3% between 1900 and 2012, but adjustment for the storage and insurance 
costs incurred by wine investors lowers the estimated return to 4.1%. 
Including the benefit of cash dividends, equities have been a better investment than wine over the 
past century, and accounting for differences in transaction costs likely would lower the relative 
performance of wine investments even further, especially over short horizons. Yet returns on wine have 
exceeded those on government bonds and Treasury bills, as well as art and investment-quality stamps. 
As our model suggests, substantial positive correlation exists between the equity and wine markets. The 
strong price appreciation over the last half century also coincides with dramatic growth in the number of 
high net worth households worldwide with access to the market for Bordeaux wines. In recent years, for 
example, Chinese consumption of Bordeaux wines increased dramatically (Financial Times, 2014; 
Muhammad, Leister, McPhail, and Chen, 2014). To the extent that such globalization of the market was 
unforeseeable, historical returns could have been higher than anticipated ex ante. 
Next, we report long-term evidence that is consistent with transitory bubbles in the price of 
collectibles such as wine. Periods that are identified in-sample as ones when financial markets were 
overvalued have a spillover impact on the prices of investments of passion, reinforcing these assets’ 
mean-reverting behavior. Moreover, future returns on a collectible (e.g., wine) can be predicted to be 
lower after periods of outperformance relative to other collectibles (e.g., art and stamps). 
We conclude by observing that the annualized return on First Growths that we report is best 
considered an upper bound on the long-term investment performance of wine more generally, as the 
relative popularity of the First Growths could have risen over our time frame. For the period 1972–2012, 
we find slightly lower returns for the sweet white wine Yquem and for a selection of ports.  
The paper proceeds as follows. Section 2 provides an illustrative model of wine prices. Section 3 
describes our data. Section 4 examines the impact of aging on prices, and Section 5 studies the long-
term investment performance of wine. Section 6 concludes. 
7 
 
 
2. An illustrative model of wine prices 
How can the price of a well-known wine be expected to change over time? How do returns differ 
between low-quality vintages that decline in quality quickly and high-quality ones that spend several 
decades maturing? And how can the impact of aging on prices be disentangled from time effects? In this 
section, we present a simple model that suggests answers to these questions. Crucially, our model 
accounts for both changes in a wine’s consumption value and its attractiveness as a collectible over the 
life cycle. 
Suppose that a representative collector-investor has wealth W0 in period 0, with wealth growing at 
a constant rate z so that Wt = W0×(1+z)
t
. The value of consuming a j-year-old bottle of wine at time t can 
be defined as a function of the wine’s drinkability ci,j and the investor’s wealth, i.e., Ci,j,t ≡ ci,j×Wt, where 
i represents the quality type of the vintage. The dependence of consumption value on wealth reflects the 
discretionary nature of luxury consumption (Aït-Sahalia, Parker, and Yogo, 2004). We assume two 
quality types. Wines of low-quality vintages, without aging potential, deteriorate over time so that cL,j = 
a
j
×cL,0 for each age j > 0, where a < 1 is the rate of deterioration. By contrast, wines of high-quality 
vintages improve monotonically by maturing until age M, i.e., cH,j = b
j
×cH,0 for each age j ≤ M, with b > 
1. After maturity, high-quality wines’ drinkability stays constant, i.e., cH,j = cH,M for each age j > M.
1
   
Just like an artwork or a precious diamond, an unopened bottle of a famous château can be a 
source of enjoyment. We capture this nonfinancial utility with the parameter di,j (with dH,0 > dL,0), which 
grows with age, reflecting the higher enjoyment of owning older and rarer bottles. To the extent that 
nonfinancial dividends reflect the rarity of a wine, their growth rate is linked to the cumulative aggregate 
                                                          
1
 Ratings and tasting notes by experts reflect the differences in life cycles between vintages of different qualities. For 
example, in 2012, Robert Parker’s website (http://www.erobertparker.com) labeled the 90-point 1997 vintage of Margaux 
“late” and the 88-point 1993 vintage “old,” while the 98-point 1928 and the 100-point 1900 vintages were considered 
“mature.” 
8 
 
consumption over the life cycle. However, even though we can expect to see a peak in consumption at 
maturity, when the value of drinking exceeds that of storing, we keep our model simple by assuming 
that the ownership dividend parameter increases with age at a constant rate g.
2
  
The equivalent monetary value of the psychic ownership dividend for a bottle of quality type i and 
age j in period t is defined as Di,j,t ≡ di,j ×Wt. Our setup resembles the model in Goetzmann and Spiegel 
(1995) in which art values depend on collectors’ wealth. Under the assumptions outlined above, the 
nonfinancial dividend grows at the rate k ≡ (1+g)×(1+z)–1, which is assumed to be smaller than the 
appropriate discount rate r. In the model, tastes and the growth rate of wealth do not vary over time, and 
this discount rate can be set equal to the risk-free rate. However, if wealth is risky, the positive 
correlation between shocks to wealth and wine prices implies a required return close to that of the 
market portfolio. Uncertainty about future tastes and storage costs can further drive up the expected rate 
of return to wine investment, while uncertainty about future availability can depress it by increasing the 
convenience yield. The magnitude of the relevant discount rate is ultimately an empirical question. 
In our setup, at each time t, the price of a j-year-old bottle of the low-quality type should be the 
maximum of two values, namely the value of immediate consumption and the present value of all future 
ownership dividends received conditional on never consuming: 
          (        
          
   
)          (1) 
For the high-quality type, as long as the wine has not reached maturity, the price is the maximum of 
three measures: (1) the value of immediate consumption, (2) the present value of consumption at 
                                                          
2
 We assume that individuals deciding whether to drink or to store do not consider the marginal effect of their personal 
consumption on the attractiveness of the remaining bottles, though we note that Jovanovic (2013) provides an equilibrium 
model that endogenizes this decision. Various frictions can explain why not all bottles are consumed at maturity. For 
example, wine can be forgotten in large cellars, collectors with more wine than they can consume could be reluctant to sell, 
or a private-value component could make the utility from lifelong ownership exceed the consumption value, even at 
maturity (e.g., a collector wanting a complete collection of vintages). 
9 
 
maturity plus the present value of all ownership dividends received until consumption, and (3) the 
present value of infinite storage: 
          (        
         
(   )   
 
          
   
 (  (
   
   
)
   
)  
          
   
)  if       (2) 
If the high-quality wine is at or beyond maturity, the price is the maximum of the value of consumption 
and the present value of all future ownership dividends: 
          (        
          
   
)                (3) 
Fig. 1 shows an example of the resulting (log) price dynamics for a low-quality and a high-quality 
vintage of a famous château. We set j and t equal to zero in the first period. For the low-quality vintage 
in Panel A, the price decreases initially due to the decline in consumption value, until the value as a 
collectible (i.e., the present value of ownership dividends) exceeds the value of consumption. After this, 
the price grows at a constant rate k. In Panel B, we show the dynamics for a high-quality vintage that 
grows in drinkability for 40 years and does not change in consumption value thereafter. If the growth in 
consumption value prior to maturity exceeds the discount rate r, the price increases at a rate that 
approaches r as the wine nears maturity. After maturity, wine prices grow slowly with wealth until the 
present value of ownership dividends takes over, after which the price grows at a rate equal to k. Our 
simple model thus predicts very different price patterns for bad and good vintages. By comparing the 
two panels in Fig. 1, we can also see that the cross-sectional premium for vintage quality could be 
smaller for very old wines. 
[Insert Figure 1 near here] 
Eqs. (1) and (3) imply that the nonfinancial dividend yield D/P on collectible wines (i.e., those 
wines for which the price is determined by the value of lifelong storage) equals r minus k. This suggests 
that the psychic return on wines substantially beyond maturity can be approximated by their under-
10 
 
performance relative to young high-quality wines that are still maturing. As such, our model closely 
relates to studies that attribute the under-performance of art relative to financial assets with the same risk 
profile to the “viewing pleasure” (Stein, 1977) or “conspicuous consumption utility dividend” (Mandel, 
2009) associated with art ownership.  
In Fig. 1, the price dynamics over age and over time are one and the same. Nevertheless, 
decomposing the returns into time and age effects is straightforward. Abstracting from aging-induced 
variation in quality (i.e., holding j constant), wine values grow with wealth over time, as the (future) 
consumption value and the (future) ownership dividends all rise at the constant rate z. More formally,  
        
      
                 (4) 
By contrast, abstracting from the effects of time—and thus changes in wealth—on valuations (i.e., 
keeping t constant) delivers cross-sectional life-cycle patterns similar to those illustrated in Fig. 1, 
although the relative price differences between two consecutive age groups are lower than before: 
        
      
 
          
      
 
 
   
           (5) 
Identifying the life-cycle price patterns of vintages of different qualities is the first goal of our empirical 
analysis. Afterward, we turn to estimating the total returns realized by wine investors since the 
beginning of the 20th century. 
 
11 
 
3. Data  
We now introduce the wines that we study and describe how we collect price data from Christie’s and 
Berry Bros. & Rudd. We also show some descriptive statistics. 
3.1. Selection of wines 
We study transactions for five red Bordeaux wines: Haut-Brion, Lafite-Rothschild, Latour, 
Margaux, and Mouton-Rothschild. The Bordeaux region has long been among the world’s leading wine 
areas, and the production of fine wines developed quickly after the introduction of bottles and corks in 
the late 17th century (Simpson, 2011). The important châteaus already had established reputations in the 
18th century, a time when most other wine was still sold under the name of the shipper, not the grower. 
In 1855, wine brokers compiled a classification of wines for the Universal Exhibition of that year based 
on historical prices and labeled Haut-Brion, Lafite-Rothschild, Latour, and Margaux as the four red 
Premiers Crus or First Growths. By the end of the 19th century, this 1855 classification was well known. 
Mouton-Rothschild was classified as the first of the Second Growths but was widely believed to have 
the quality of a First Growth and traded at similar prices. The château was formally upgraded to the top 
category in 1973.  
Not much time variation exists in the perceived quality of these wines. One reason for the relative 
stability in rankings is the importance of natural conditions, such as climate and soil, to the potential 
quality of a wine. Today, the Premiers Crus remain among the most highly appreciated and frequently 
traded wines in the world, and total annual production is about 70 thousand cases, or 840,000 bottles, on 
average, though production varies slightly across vintage years. The smallest producer among the First 
Growths is Haut-Brion, with eight to ten thousand cases per year. The largest ones are Lafite and 
Mouton, which typically produce 15 to 20 thousand cases per year (Leve, 2014). 
12 
 
3.2. Data collection  
We compile a long-run price history for the five wines listed above, starting in 1899. Two other 
criteria guide the data collection. First, we focus on vintages since 1855. The compilation of the 
classification in that year makes it a natural starting point. The second half of the 19th century also saw 
the introduction of estate bottling for high-quality wines, along with their distinctive labels and corks. 
Second, we gather prices only for standard-size bottles, for a number of reasons: They make up a very 
large majority of all transactions historically, nonstandard bottles such as magnums are more likely to be 
valued for their uniqueness, and the aging process is affected by the size of the bottle, so excluding 
nonstandard bottle types simplifies the analysis. 
 We collect two types of historical price data: prices realized at auctions in the London sales rooms 
of Christie’s (and W. & T. Restell, an auction house bought by Christie’s in the 1960s) and retail list 
prices at Berry Bros. & Rudd, a London dealer of wines and spirits. At least until the First World War, 
“an important quantity” (Simpson, 2011) to “nearly all” (Penning-Rowsell, 1975) of the best Bordeaux 
wines were sold to British buyers. By considering only one auction house and one dealer over the long 
term, we mitigate concerns that our findings are affected by temporal changes in the nature of the price 
data. Moreover, these first-tier sellers have had a high reputation for a long time, which reduces worries 
about fakes or errors in item descriptions. 
3.3. Auction prices: Christie’s  
Christie’s is one of the world’s two leading auction houses. Its first sale was held in December 
1766 in London and consisted of “the property of a Noble Personage deceas’d.” The auction included 
not only furniture, jewelry, and firearms, but also some “fine claret” (lots 30–34)—claret is the British 
name for red Bordeaux wine—and “fine old madeira” (lots 35–38). Christie’s held its first session 
dedicated solely to wine in 1769. In the early decades, detailed descriptions of the bottles being 
13 
 
auctioned were often lacking. For a long time, wine was sold anonymously or under the name of the 
merchant who had imported it (Penning-Rowsell, 1972). It was not until 1788 that a Christie’s catalogue 
(mis)named the Bordeaux châteaus “Lafete” and “Margeau” (Penning-Rowsell, 1973). Vintage quality 
became increasingly relevant only in the early 19th century, when Christie’s catalogues regularly started 
to include information on château and vintage year. In 1941, the Christie’s premises on King Street were 
destroyed by a firebomb, forcing the firm to move (Sheppard, 1960). There were occasional wine sales 
at the temporary offices, but these stopped altogether in 1945 and did not resume when Christie’s 
returned to its original location in 1953. In 1966 the auction house renewed its wine business and 
acquired W. & T. Restell, the only other wine auctioneer in London at the time (Broadbent, 1985). Wine 
auctions remain an important part of Christie’s activities in London today, with wine sales conducted on 
a near-monthly basis.  
The long tradition of auctioning wines makes Christie’s a unique source of long-term information. 
Nevertheless, building a database of wine prices is challenging due to the lack of a database that covers 
the firm’s entire history and the fact that Christie’s did not hold wine auctions continuously. We thus 
need to draw upon a number of different documents and sources.  
For the period 1899–1971, we use data from archived catalogues containing the results of sales at 
Christie’s (before 1945 and 1966–1971) and Restell (1941–1965). Fig. 2 shows an excerpt from an 
auction catalogue, annotated by the auctioneer, of a sale from 1935. For 1972–1979, we obtain price data 
at London auctions from the annual Christie’s Wine Review, which is a publication that lists prices paid, 
generally at Christie’s, over the previous calendar year. If more than a single lot of a particular wine was 
sold, the Wine Review reports the lowest and highest price and sometimes more price points. If no sale 
took place for a given château-vintage pair, the Wine Review repeats older price information. We 
eliminate these duplicates.  
[Insert Figure 2 near here] 
14 
 
For the years 1980–1984 and 1988, we collect data from the Christie’s Vintage Wine Price Index 
books, which succeeded the Wine Review. We obtain data from auctions at Christie’s London for 1985–
1987 and 1989–1998 from David Ashmore at Liquid Assets. Finally, we collect data on London wine 
sales over the period 1999–2012 from the Christie’s website (http://www.christies.com). Throughout our 
analysis, we focus on homogeneous lots and do not consider mixed lots of wines from different châteaus 
or vintages. Over the period 1999–2012 (for which we have the most detailed data), a lot contained 
slightly less than ten bottles on average, with older (younger) vintages typically containing fewer (more) 
bottles per lot.
3
 
We make three remarks about the auction data. First, the UK government has historically taxed 
sales of alcohol through excise duties. The payment of the duty (and value added tax), however, can be 
postponed by keeping the wine in bond. Duty is paid only when the bottle is removed from a bonded 
warehouse for delivery to a private address. Thereafter, the wine can be traded without additional taxes. 
We assume that all prices are duty-paid, and, thus, do not try to correct price levels for transactions in 
bond or for sales from foreign cellars. For high-end wines such as the ones considered here, excise duty 
is in any case relatively unimportant. At the end of 2012, the duty stood at 1.80 British pounds (GBP) 
per bottle.  
Second, Christie’s London introduced a buyer’s premium in its wine auctions in the fall of 1986. 
This additional fee, payable by the winning bidder, was initially 10% of the hammer price of the lot and 
gradually increased to 15% by 2012. When necessary, we transform the observed prices so that they are 
inclusive of the premium. Because buyers take the premium into account as they bid, it can be 
considered a transaction cost imposed on the seller (Ashenfelter and Graddy, 2005; Marks, 2009). 
                                                          
3
 We can provide an indication of the volume of relevant sales at Christie’s London as follows. In 2011, it sold 519 unmixed 
lots of standard-size bottles of First Growth wines. The lots represented 4,884 bottles and about 2.8 million British pounds 
(GBP) of sales (including buyer’s premium). The worldwide total of all wine sales by the big auction houses amounted to 
about 250 million GBP in the same year (Bloomberg News, 2012). 
15 
 
Therefore, the evolution of hammer prices exclusive of buyer’s premium would underestimate the 
growth in the willingness to pay for wine. 
 Third, London’s other major auction house, Sotheby’s, started organizing wine sales in the 1970s, 
and we use its prices to validate our long-term database. We already know from Ashenfelter (1989) that 
in the mid-1980s total buyer’s prices were similar at Christie’s London and Sotheby’s London. Here we 
examine London wine auction results for both Christie’s and Sotheby’s over a longer time frame, using 
the 1985–1998 data source described earlier. Keeping the combination of château, vintage, and 
transaction year fixed, Sotheby’s mean prices are a statistically insignificant 0.9% below Christie’s. We 
also collect recent prices from the Sotheby’s website (http://www.sothebys.com) and examine the 12 
château-vintage combinations that traded at both houses in 1985 and in 2012. The annualized price 
appreciation differs between the two houses by less than 0.3%. We are reassured that our findings are 
not sensitive to the choice of auctioneer. 
3.4. Dealer prices: Berry Bros. & Rudd 
In 1698, a small grocery store was founded at 3 St. James’s Street in London, not far from where 
Christie’s is located today. By the early 19th century, the shop had come into the hands of George Berry, 
son of a wine merchant, who transformed it into a wine business, and the Berry family has been active in 
the company ever since. Price lists show that French and German wines, spirits (e.g., brandy, whiskey, 
gin), and fortified wines (e.g., port, sherry, madeira) were the backbone of the business in the early 
1900s. Hugh Rudd, also from a family of wine merchants, joined the company in 1914. Today, BBR 
also has an online wine shop and brokering service.  
As is the case for Christie’s, the history of BBR offers a long-run perspective on the evolution of 
wine prices. Since the early 20th century, BBR has generally issued price lists in the spring and fall of 
each year, though it recently reduced the frequency to once per year. For the period 1905–1978, we can 
16 
 
collect data on the five Bordeaux wines that we study from a set of 11 bound volumes of price lists. We 
use loose copies of the relevant price lists for a number of years not included in the bound volumes and 
for the period since 1978. All documents were consulted at the London headquarters of BBR. Fig. 3 
reproduces two pages from the May 1909 price list.  
[Insert Figure 3 near here] 
Each list typically includes from a handful to a few dozen prices useful for our study. During the 
late 1980s and the 1990s, the lists do not always include prices for relevant wines, which are mentioned 
to be available on request. In the early 2000s, BBR introduced separate Blue Lists with prices for “the 
finest reserve wines and wines for laying down”. Prices from these alternative lists also enter our 
database. In recent years, the BBR website (http://www.bbr.com) has largely assumed the role once 
played by the printed price lists. We, therefore, complete our database with November 2012 prices taken 
from the BBR website.  
A few further comments on the dealer price data are in order. From the 1920s until the 1960s, the 
lists often include both credit and cash prices, and we work with the latter. Our prices are also duty-paid 
and inclusive of value added tax (which we add when necessary) and thus reflect the total cost to 
domestic buyers who take physical possession of the wine. Whenever possible, we use prices per bottle 
instead of per case. We also ignore quantity discounts because we lack detailed information on them for 
each period, and we do not take into account other discounts such as for imperfect quality or seasonal 
promotions offered by BBR. For these reasons, prices in the retail lists are likely an upper bound on the 
true underlying values, just like catalogue prices in other collectibles markets. At the same time, we are 
mainly interested in quantifying the trends in prices, and a systematic upward bias in all prices would 
not affect these trends.  
17 
 
3.5. Construction of final database and descriptive statistics 
In total, we hand-collect 36,271 prices from the various Christie’s and BBR sources. If we know 
that an auction took place or that a dealer issued a price list in the first half of the year, we assign the 
accompanying price points to the previous year-end. In all other cases, we date the price to the end of 
the year. Next, so as not to overweigh certain periods or transaction types, we average prices per bottle 
by quartet of year-end (e.g., end-2007), château (e.g., Latour), vintage (e.g., 1982), and transaction type 
(dealer or auction). Our final database contains price information for standard-size bottles on 9,492 such 
combinations (our units of observation from now on) since end-1899.
4
  
Table 1 presents some descriptive statistics for our data set. Panel A of Table 1 shows the number 
of observations per transaction type and per château for each decade since the 1900s (where 1899 is 
added to the first period). The growth of the database in the 1970s reflects the increasing availability of 
auction sources. Panel B gives more information on the distribution of the averaged prices in GBP for 
each decade, in nominal and real (year 1899) terms. Until the early 20th century, no wine sold for more 
than one pound per standard-size bottle. For the most recent years (2010–2012), the average price level 
per bottle is 758.46 GBP, with prices ranging from 80.50 to 8,510 GBP. We observe much stronger 
increases in both nominal and deflated prices in the second half of the 20th century than in the first half.  
[Insert Table 1 near here] 
                                                          
4
 If we assume that all wines could be sold as of the first year after the vintage (which is not always the case), there would be 
114,570 combinations of year-end (1899–2012), château, vintage year (1855–2011), and transaction type for which prices 
could in theory be observed. Our database thus covers 8.3% of the population of (potential) values. 
18 
 
4. Aging and prices 
In Section 4, we examine the effect of aging on the price of wines of different quality, by estimating 
hedonic regressions. We also discuss how auction trading volume changes over the life cycle.  
4.1. Methodology  
Hedonic models relate transaction prices to their value-determining characteristics (Rosen, 1974) 
and are commonly used to study price formation in markets for infrequently traded assets such as real 
estate (e.g., Campbell, Giglio, and Pathak, 2011) and art (e.g., Renneboog and Spaenjers, 2013). 
Ashenfelter, Ashmore, and Lalonde (1995), Di Vittorio and Ginsburgh (1996), and Combris, Lecocq, 
and Visser (1997) were among the first to estimate hedonic regressions for Bordeaux wines.  
Determinants of price levels that we would like to control for in our analysis of age effects (on 
different quality levels) are the identity of the château, the transaction type, vintage attributes, and the 
year of sale. Vintage attributes are often summarized by the year in which the grapes were harvested. 
However, vintage-year dummy variables are unsuitable for our regressions because they would 
introduce multicollinearity with age and year of sale. The superficially appealing approach of including 
dummy variables for all three dimensions must therefore be ruled out.
5
 Instead, we estimate vintage 
attributes in a more direct way. We assume that the vintage attributes that impact prices are the annual 
levels of production (as a larger supply can be expected to relate to lower prices) and each year’s 
weather quality. 
With respect to production, we use an anonymous First Growth château’s historical yields 
(measured in hectoliters per hectare) from Chevet, Lecocq, and Visser (2011). Next, we use information 
on the weather in each vintage year as a proxy for quality. Ashenfelter, Ashmore, and Lalonde (1995) 
                                                          
5
 A similar problem is faced by studies that aim to disentangle age, cohort, and year effects in household portfolio choice 
(e.g., Ameriks and Zeldes, 2004; Malmendier and Nagel, 2011). Our work also relates to research that disentangles age 
from time effects in values of durable corporate assets such as aircraft (e.g., Staunton, 1992). 
19 
 
and Ashenfelter (2008) show how weather data predict the quality and prices of Bordeaux wines. Using 
daily data from a weather station in Bordeaux (Météo Climat, 2012), we measure the average 
temperature between April and August (the growing season) and total rainfall in August and September 
(the harvest season) for each vintage year between 1873 and 2011.
6
 We then sort all vintages into deciles 
according to both measures and assign a score of one to ten to each vintage year for each measure, 
where higher temperatures and less rainfall are associated with higher scores. Our weather quality 
variable sums the two scores. It is thus close to 20 for a warm growing and dry harvest season.   
Advances in technology over our sample period could have made the weather a less important 
determinant of vintage quality. Later in this section, we report the robustness of our results to using an 
alternative price-based quality measure. We do not use vintage charts or expert ratings as quality 
measures for two reasons. First, they are not exogenous, as today’s scores are typically determined by 
tastings from recent years and are susceptible to updating over time. Resolution of uncertainty about 
quality could thus correlate with both returns and changes in scores (e.g., Jones and Storchmann, 2001). 
Second, no rating system covers all vintage years considered here. For example, Robert Parker Online 
has only very selective coverage for the first half of our time period, with better years more heavily 
represented. Nonetheless, the correlations (across all in-sample combinations of château and vintage 
year) of our weather quality variable with either end-2012 Parker scores or red Bordeaux vintage chart 
scores from an international retailer of rare wines (SoDivin, 2012) are relatively high (0.36 and 0.63, 
respectively) and statistically significant at the 0.01 level. 
Formally, our regression model is  
  (        )                    
       
 (       )               (6) 
                                                          
6
 Weather data for Bordeaux are available since 1880. For 1873–1879 and later years with missing data (e.g., 1915–1920 and 
1940–1945), we impute values using linear regression models relating monthly data on temperature and rainfall in 
Bordeaux to data for Paris and Marseille (and Nantes, when possible) and month fixed effects over the period 1880–2011.  
20 
 
where Pi,j,l,t is the price of a wine from château i and vintage year j at sale location l (i.e., the transaction 
type: dealer or auction house) in year t. The different α denote fixed effects, Yj measures the production 
yield in year j, Wj picks up the quality of the weather in the same year, and X is a polynomial age 
function. In our baseline model, X is a third-degree polynomial. We include interactions of our newly 
created weather quality variable with the age polynomial to test our hypothesis that the relation between 
age and price levels depends crucially on the quality of the vintage. 
4.2. Results 
We estimate Eq. (6) using ordinary least squares with the price level in real GBP as the dependent 
variable and cluster standard errors by sale year. The sixth column in Table 2 summarizes the results. 
The R-squared statistic is 0.74, which is higher than the explanatory power of the models in the first five 
columns of Table 2 that combine subsets of variables in different ways.  
[Insert Table 2 near here] 
Before we study the effect of aging on prices, we review the results for the other variables 
included in our benchmark hedonic model. The coefficient on the transaction type dummy indicates that 
dealer prices exceed auction prices on average. Clients could value the condition (and certainty about 
authenticity and provenance) of bottles sold by BBR, but the list prices should in any case be considered 
an upper bound on true transaction prices. Next, Mouton-Rothschild carries a premium relative to the 
other four châteaus in our sample. Haut-Brion, which is omitted from the regression due to 
multicollinearity, is the least expensive, ceteris paribus. The results also indicate strongly significant 
relations between our proxies for vintage effects and prices, with the coefficients relating lower 
production and better weather to higher prices as expected. Finally, the year-of-sale dummies control for 
changes in wine values over time, independent of aging effects. We provide a detailed description of 
historical price trends, focusing on the total net returns to investors, in Section 5. 
21 
 
We now turn to the relation between age and price levels. An F-test shows that the coefficients on 
the interaction terms between age and quality are jointly statistically significant. This suggests that 
wines of low- and high-quality vintages exhibit different life-cycle patterns. Fig. 4 shows the life-cycle 
price patterns implied by the coefficients on the weather quality variable, the age polynomial, and the 
interaction terms, for otherwise identical wines of the lowest and highest weather quality categories. We 
rescale the predicted price of the lowest-quality wine at age zero to unity and show results up to an age 
of one hundred years. (Less than 3% of our observations are for wines older than a century.) Fig. 4 also 
shows the confidence intervals around the predicted price levels. 
[Insert Figure 4 near here] 
For the lowest-quality vintages, prices increase little over the first few years of the life cycle. The 
geometric average price appreciation over the first two decades implied by our results is only 0.6%. 
Afterward, prices rise faster and in a near-linear fashion. In contrast, the highest-quality wines 
appreciate strongly while they are maturing after the vintage. We estimate a geometric average price 
appreciation of 4.0% over the first 20 years. Prices stabilize once high-quality wines fully mature, and 
their increase in value between age 40 and 80 is very limited. As the wines become antiques, we begin 
to observe new price increases. The first row of Table 3 summarizes the cross-sectional price changes 
over the life cycle for the worst-quality type (until an age of 60 years) and the best-quality wines as 
predicted by our regression results.  
[Insert Table 3 near here] 
The patterns in Fig. 4 are generally in line with the model presented in Section 2 and illustrated in 
Fig. 1. The highest returns are realized on young high-quality wines that are maturing. Given that we are 
using an exogenous measure of quality, this result is not driven by resolution of uncertainty about the 
wine’s quality over its life cycle. The observation that the prices of wines substantially beyond their 
optimal consumption point—low-quality wines older than a few decades or high-quality wines 
22 
 
approaching a century—go up with age is consistent with the theorized existence of a growing 
nonfinancial payoff to owning such wines. Finally, given that consumption quality slowly becomes 
irrelevant for all wines, it is not surprising that the relative difference in price levels between high-
quality and low-quality vintages is smaller for very old wines than for young wines.  
According to our model, the return difference between high-quality wines approaching maturity 
and wines long beyond maturity (i.e., collectible wines whose prices are determined by the nonfinancial 
dividends they provide) should offer an indication of the psychic return to holding the latter type of 
wines. The results in Table 3 suggest that the nonfinancial return realized by collectors is probably 
relatively small. For example, the financial outperformance of pre-maturity high-quality vintages (ages 
zero to 20 years) relative to post-maturity low-quality vintages (ages 40 to 60 years) is only about 1.3% 
per year.
 Under our model’s assumptions, this estimate would imply that a collectible wine worth 100 
GBP provides a nonfinancial dividend of 1.3 GBP to its owner over the course of a year.  
4.3. Robustness checks 
We now check the sensitivity of our results to a number of robustness checks. First, we split the 
sample in the period before 1980 and the years since 1980, which gives two subsamples with a nearly 
equal number of observations. Second, we repeat our hedonic regression model using auction 
transactions only, to mitigate concerns that the documented patterns are due to risk sharing between 
wine producers and dealers, leading to overpriced bad vintages and underpriced good ones. Third, we 
replace the third-order age polynomial with a fourth-order polynomial, allowing for more flexibility in 
the estimation of the age effects. Fourth, we construct an expanded measure of weather quality. 
Ashenfelter (2008) shows that, in addition to temperature in the growing season and the rainfall during 
the harvest season, rainfall during the winter preceding the vintage can affect wine quality to some 
extent. All else equal, more winter rain is associated with slightly better wines in the next vintage. We 
therefore create a score from one to ten based on the decile to which each vintage year belongs when 
23 
 
sorted on cumulative rainfall in the period from October to March and add half of this score to the 
previously constructed weather quality variable. Fifth, we construct an entirely new exogenous measure 
of vintage quality as follows. Bordeaux producers have historically sold forward contracts on wine not 
yet bottled through the system of en primeur transactions. Short-term changes in the so-called opening 
price set by the château reflect variation in vintage quality (Ali and Nauges, 2007). We use data from 
Dovaz (1999) on the opening price of Latour (1855−1997) and create a relative vintage quality measure 
by dividing the opening price in a year by the average opening price in the two surrounding years. We 
exclude vintages with multi-year contracts.   
Table 3 shows the predicted geometric average price appreciation over different segments of the 
life cycle of the lowest-quality and highest-quality vintages for these alternative specifications based on 
the coefficients for the age and age-quality variables. It also compares the results with our benchmark 
model. In each specification the general patterns are similar to the benchmark results: Maturing wines 
show the highest returns to aging, but wines long beyond maturity also increase in value with age.
7
  
4.4. Vintage quality and trading over the life cycle 
We argue that wines from poorer vintages lose whatever consumption value they have more 
quickly. As they are more likely to be consumed when still young, we should expect that the trading 
volume of these wines goes down more rapidly than that of better vintages. We test this hypothesis using 
data on the number of sold auction lots per combination of vintage year and sale year between 1985 and 
2012 (for which we have volume data at Christie’s) in our database. We compute the average annual 
auction volume for each age both for relatively low-quality wines (weather quality < 8) and for 
                                                          
7
 Results are also similar when extending the benchmark model with an interaction of the dealer dummy and a linear time 
trend (to account for the possibility that the dealer premium has changed over time) or when including pre-1973 and post-
1973 Mouton dummies (to allow for a potential change in demand following the upgrade to First Growth). We also check 
whether a positive relation exists between the reputation of the artist designing the Mouton label and transaction prices, but 
we do not find evidence of such an effect.  
24 
 
relatively high-quality wines (weather quality > 12). We then average over five-year age groups (starting 
with ages 5–9, because wines rarely sell at a younger age). Fig. 5 shows that the result is in line with our 
expectations. Wines from worse vintages become rarities more quickly—after two decades or so—
which is consistent with the price patterns reported earlier. These results also indicate that the high 
returns to young high-quality wines are not driven by decreases in availability. Wines from high-quality 
vintages remain in relatively high supply for about four decades, corresponding to the time it takes to 
fully mature. 
[Insert Figure 5 near here] 
 
5. The long-term investment performance of wine 
We now evaluate wine’s investment performance over the 113 years spanned by our sample, a 
performance that reflects changes in market conditions in addition to the price increases related to aging 
discussed in Section 4. We also compare wine with other assets, examine the relation between wealth 
creation and wine prices, and study whether wine returns are predictable. Finally, we explore the impact 
of success bias on our results. 
5.1. Methodology  
To build a returns index, we apply a repeat-sales regression to the 8,582 price pairs in our database 
for which the château and vintage (and transaction type) are identical.
8
 The standard repeat-sales 
regression (e.g., Bailey, Muth, and Nourse, 1963) estimates the return on an underlying portfolio of 
assets by relating the log returns implied by the individual price pairs to the periods over which the 
assets are held. An advantage of the methodology is that it explicitly controls for the uniqueness of each 
                                                          
8
 One example of a price pair is the following: The average transaction price of a bottle of Margaux of vintage year 1945 at 
auction was 133.51 GBP in 1985 and 168.48 GBP in 1986. For many château-vintage combinations, we observe prices for 
long series of consecutive years.  
25 
 
combination of château and vintage year. There is thus no need to estimate the average premia 
associated with, for example, different categories of weather quality. However, an issue with the 
standard repeat-sales model is that, just like the log-linear hedonic model of which it is a special case, it 
estimates an equal-weighted index based on the geometric (instead of arithmetic) average of prices in 
each period. As this is undesirable, we follow the variant proposed by Shiller (1991), which works with 
absolute prices instead of log returns. The result is a value-weighted arithmetic repeat-sales index that 
more accurately tracks the investment performance of wines held by collectors.  
5.2. Results 
The line plot in Fig. 6 shows the price index, in real GBP, that we obtain with the arithmetic 
repeat-sales technique. The index is set to unity at the start of 1900. We geometrically interpolate the 
index values for the years 1912, 1916, 1917, and 1947, for which we have no transactions that enter the 
estimation. The index shows that, despite the positive average effect of aging on the consumption value 
and attractiveness of wines, wine prices did not increase in real terms over the first quarter of the 20th 
century. Fig. 6 also shows that the value of wines boomed during the Second World War. Prices 
increased by more than 600% between 1940 and 1945. Many factors probably played a role: The war 
upset the trade in high-end French wines, with the port of Bordeaux and many châteaus occupied by 
Nazi Germany (Kladstrup and Kladstrup, 2001); the UK government prohibited sales of wines and 
spirits by unlicensed auction houses; Christie’s had to limit its sales activities after its main offices were 
bombed; and many wine bottles were sold through Red Cross charity auctions that are not included in 
our data but could have pushed up price levels. The boom was followed by sharp decreases in wine 
prices in the years after the end of the war. In the second half of our time frame, wine prices grew 
strongly, although the increases were punctuated by declines in real price levels of more than 20% in 
1973–1975, 1980, 1990–1992, 2003, and 2011–2012. Over the complete 1900–2012 period, we find a 
geometric average annual real return of 5.3%. For completeness, the bars in Fig. 6 show the nominal 
26 
 
return for each year on the left-hand axis. The geometric average annual nominal return over our time 
frame is 9.4%. 
 [Insert Figure 6 near here] 
The index values are relatively precisely estimated. The standard errors on the last index values 
imply a 95% confidence interval around the annualized return estimate that is about 1 percentage point 
wide. Moreover, for the time frames that overlap with earlier research on wine returns, i.e., 1969–1976 
(Jaeger, 1981), 1986–1996 (Burton and Jacobsen, 2001), and 1996–2009 (Masset and Weisskopf, 2010), 
the estimated trends are broadly in line with those reported by others.
9
 
5.3. Accounting for storage and insurance costs  
The condition of a bottle of wine is determined by factors such as temperature and humidity. A 
poor storage environment can cause bottles to start leaking or evaporating, wines to become oxidized, 
and labels and packaging to experience damage. Such wines are less likely to be included in our 
database. Auction houses would nowadays typically not even sell bottles that have a questionable 
provenance, even if they appear to be in a decent condition. The reported returns in Section 5.2 therefore 
have been realized only by investors who stored their wines properly. Today, storing wine bottles does 
not need to cost much. Robinson (2010) mentions a cost of 10 to 20 GBP per dozen bottles per annum at 
professional storage providers. However, relative to the average price of wine, storage was more 
expensive before the increases in wine values of the last decades. For example, the BBR price list of 
March 1940 shows that a case of wine could then be stored at a price of 1.6 shilling, or 0.075 GBP, per 
year—a cost equivalent to 0.94% of the average end-1939 price for a dozen bottles. Repeating this 
exercise in 1950, 1960, 1970, 1980, 1990, and 2000 delivers estimates of 0.83%, 0.88%, 0.39%, 0.13%, 
                                                          
9
 As an additional robustness check we run a hedonic regression model with dummies for the different châteaus, transaction 
types, and vintage years as controls. The resulting return series is highly correlated (0.84) with the one presented here, and 
the geometric average annual return estimate is similar (4.6%).  
27 
 
0.24%, and 0.23% respectively. We use these estimates of transaction costs at the start of each decade to 
correct our annual returns. We use the 1940 cost estimate for the years prior to 1940, and for 2010–2012 
we use the 2000 estimate.  
In addition, the reported returns clearly include only those bottles that were unaffected by or 
insured against fire, flood, accidental breakage, theft, and other hazards. Wine storage contracts often do 
not include insurance or do not insure against all risks at full market value. Meltzer (2005) notes that, 
although the exact policy and cost is a function of the insurer, premiums are fairly uniform. A typical 
wine insurance contract costs close to 0.5% of the market value of the collection per year.  
In Fig. 7, we show our deflated price index after accounting for these storage and insurance 
expenses. (The observation that storage and insurance costs are low relative to historical returns suggests 
that returns are not driven by a negative convenience yield.) We estimate an annualized real return, net 
of storage and insurance costs, of 4.1% between 1900 and 2012. This corresponds to an annualized 
nominal return of 8.2%.  
[Insert Figure 7 near here] 
5.4. Comparison with other assets  
Fig. 7 also shows returns to a number of other financial assets and collectibles. Data on British 
equities, government bonds, and bills are from Dimson, Marsh, and Staunton (2013). For the art market 
we use an index for Great Britain from Goetzmann, Renneboog, and Spaenjers (2011), updated until 
end-2012 using data from Artprice (2013). Data for British stamp prices are from Dimson and Spaenjers 
(2011), updated using returns on the Stanley Gibbons GB30 index. Table 4 shows summary statistics for 
the different distributions of returns in both nominal and real terms.  
[Insert Table 4 near here] 
28 
 
Annualized real returns over the period 1900–2012 are 5.2%, 2.4%, and 2.8% for equities, art, and 
stamps, respectively. Wine has thus under-performed equities over this time, although the difference in 
cumulative appreciation has narrowed since the start of the 21st century. When comparing average 
returns on wine with those on financial assets, however, it is important to bear in mind that transaction 
costs could lower the relative performance of wine investments more than trading costs depress the 
returns on investment in financial assets, especially over short holding periods. For example, the buyer’s 
premium at Christie’s London was 15% at the end of 2012, while the commission paid by the seller can 
be as large as 10%. So a seller could receive only about 75% of the amount that the winning bidder pays 
out. These estimates could still underestimate true costs, as purchasers and sellers of wine can incur 
expenses related to transportation, handling, and administration when moving the wine from one storage 
facility to another. 
At the same time, Table 4 and Fig. 7 show that wine has outperformed not only government 
bonds, but also art and stamps, even when ignoring the insurance and storage costs associated with 
investments in those types of collectibles. This might not be surprising as a majority of the wines that 
trade are relatively young and high-quality, which give the highest financial returns. Moreover, even as a 
collectible, age can be a more important determinant for the attractiveness of wine than for art or stamps, 
making wine an asset with higher capital gains but a lower nonfinancial dividend yield.
10
 
                                                          
10
 Wine also provides British investors with an unusual tax advantage, as it is regarded for tax purposes as a wasting asset. 
Wasting assets have a predictable life span of no more than 50 years and can be sold without liability to capital gains tax 
(HMRC, 1999). (Port and other fortified wines are not considered as wasting assets, as they have a very long storage life.) 
Even if the investor sells a bottle of wine following the 50-year cutoff, it could still be tax-free if the price does not exceed 
6,000 GBP. These advantages were unforeseen before 1965 (when the capital gains tax was introduced) and do not apply to 
the other assets depicted in Fig. 7. They could therefore have provided a modest boost to returns. 
29 
 
Finally, Table 4 shows that our wine index is volatile, although the standard deviation falls to a 
level similar to equities when we remove the boom and bust caused by the Second World War.
11
 
5.5. Wealth and wine prices  
In the model presented in Section 2, time effects in wine prices reflect growth in the wealth of 
wine investors, as drinking and collecting wine are both forms of discretionary luxury consumption. To 
examine whether wine prices respond to wealth shocks, we run a regression (without constant) of the 
real wine returns (before costs) against the returns on equities, which results in a market model beta of 
0.44. The (aggregated) slope coefficient increases to 0.73 when taking into account non-synchronicity in 
returns by adding a lagged and a leading equity return to the regression, following Dimson (1979).
12
 
These results point to a strong relation between the creation of financial wealth and wine prices.
13
 
Our simple representative-collector model does not consider changes in the population and 
attributes of wine buyers. In practice, however, economic globalization, coupled with strong increases in 
wealth outside the UK, probably contributed to the elevated returns exhibited by our index since the 
1960s. As high-end wines are in fixed supply, in wide demand, and easily transportable, growth in the 
number of high-income households with access to the market can raise prices even if each individual’s 
reservation price does not change (Gyourko, Mayer, and Sinai, 2013). The crowding out of English 
middle-class families and Oxbridge colleges in the market for high-end Bordeaux wines by new and 
                                                          
11
 The standard deviation of the real returns on wine falls to 20.3% when we exclude the years from 1941 to 1948, which is 
close to the 19.8% observed for equities. Even so, standard deviations can still overestimate the true volatility of returns to 
holding wine before the 1970s due to the relatively low number of observations upon which our return estimates are based 
for the first half of our sample (Bocart and Hafner, 2015). However, the use of dealer price lists and the aggregation of 
prices over one-year periods can lead to artificial smoothing in the return series. 
12
 Non-synchronicity between equity returns and wine returns can arise for several reasons. For example, while equity returns 
can be measured exactly at year-end, wine returns are estimated based on (infrequently observed) auction and dealer prices 
both before and after the turn of the year. Excluding the period 1941–1948, the aggregated market model beta equals 0.57. 
13
 In additional unreported analysis, we find a weak positive correlation of wine returns with GDP growth and a weak 
negative correlation with changes in the USD/GBP exchange rate. 
30 
 
wealthy groups of wine enthusiasts from around the world is consistent with this superstar mechanism.  
The globalization of the wine market—not only at the high end—becomes clear from long-term 
trade statistics compiled by Anderson and Nelgen (2011). While worldwide beverage wine consumption 
has not changed much between 1961–1964 and 2005–2009 (around 22,500 megaliters per year), the 
volume of international trade has more than tripled over the same period (growing from 2,602 to 8,375 
megaliters per year). Moreover, the share of imports in Europe within worldwide imports has dropped 
from 90% to 75% (even as exports to the UK have risen strongly).  
In recent years, the growth in China’s demand has been especially striking. In US dollar terms, 
wine imports into China grew at an annualized rate of 61.5% in the 1990s and 37.1% in the 2000s, 
making it the country with the strongest import growth rate over this time frame. China (including Hong 
Kong) became the world’s largest importer of Bordeaux wines in 2010 (Anson, 2011), claiming a 
position historically held by the UK. In 2012, Christie’s sold 909 (unmixed) lots of First Growth 
Bordeaux in Hong Kong, about 80% more than in London.
14
 Such fundamental changes in Chinese 
demand patterns were almost certainly unforeseen (Muhammad, Leister, McPhail, and Chen, 2014). 
Historical returns can, therefore, have been higher than wine buyers anticipated ex ante—and also higher 
than should be expected going forward.  
5.6. Predictability of the returns to wine and other collectibles 
It seems plausible that collectibles such as wine exhibit bubble-like behavior, especially given the 
difficulties with pinning down fundamental values and the impossibility of short selling.
15
 We explore 
the power of two simple measures, both based on asset prices, that can be used to predict five-year real 
                                                          
14
 Moreover, a hedonic regression on the transaction data for the same year shows that prices were about 10% higher in Hong 
Kong than in London (controlling for château and vintage effects). We thank Roman Kräussl for providing us with the 
Hong Kong auction data. 
15
 LeRoy (2004) argues that both rational and irrational bubbles can exist on collectibles. Renneboog and Spaenjers (2013) 
and Pénasse, Renneboog, and Spaenjers (2014) show a relation between sentiment measures and prices in the art market. 
31 
 
returns, starting at year-end t, on wine, art, and stamps. Our first variable is the difference between the 
annualized return, measured over the five years prior to t, on a collectible (e.g., wine) and the average 
annualized return, measured over the same period, on the other two collectibles considered (e.g., art and 
stamps). If similar macroeconomic fundamentals, such as wealth creation and globalization, drive long-
term price trends for all collectibles, then deviations from the average price path can provide evidence of 
a temporary fad. The second return predictor that we propose is an equity valuation ratio at time t. Given 
the tight relation between wealth and luxury consumption demand, overvaluation in financial markets 
could spill over into collectibles markets. In this paper, we use data on ten-year cyclically adjusted price-
dividend ratios for the UK equity market from Dimson, Marsh, and Staunton (2013), which we de-trend 
over the sample period. 
Panel A of Table 5 shows the in-sample correlations of the five-year returns to wine, art, and 
stamps with these two prior measures. We see negative correlations between –0.13 and –0.28. The first 
column of Panel B shows the result of a regression of collectible returns against the two predictors and 
collectible fixed effects. Because our regressions are based on overlapping observations, we report 
standard errors that are adjusted using Newey-West (with four lags). We see a strongly negative relation 
between our measures of overvaluation and future real returns. In the second and third columns, we split 
the sample period in half. In both subperiods, we find evidence of a negative correlation between recent 
relative performance and future returns. Moreover, while the coefficients on the equity valuation 
variables lose statistical significance at traditional levels (having p-values of 0.10 and 0.15), they are of 
the same order of magnitude as before.  
[Insert Table 5 near here] 
The results in Table 5 show that there exists some in-sample predictability of returns in 
collectibles markets. We leave it to future research to test out-of-sample predictability and to evaluate 
the potential profitability of trading strategies based on forecasts generated by price prediction models. 
32 
 
5.7. Exploring the impact of success bias  
In this paper, we estimate the returns to the best red Bordeaux wines. Although these wines were 
well known and highly appreciated before the start of our time frame, a few other types of wine have 
historically been popular as well, even for the purpose of investment. For example, the Christie’s Wine 
Review 1972 noted that vintage port “has been the wine, par excellence, for the English [to] lay down—
to invest in—and to drink.” So if today’s professionally managed wine portfolios invest over 80% of 
their funds in only eight red Bordeaux wines (Miles, 2009), it is likely that such an overwhelming focus 
on claret would have seemed unnatural to wine buyers a century, or even a few decades, ago. The 
reported returns on red First Growths should therefore probably be considered an upper bound on the 
long-term investment performance of wine more generally. 
To get a better sense of the importance of this success bias, we perform two checks. First, for the 
years since end-1971, we were able to collect auction and dealer prices for Château d’Yquem, a Superior 
First Growth sweet white wine from Bordeaux, from the same sources as before. The data allow us to 
estimate 40 years of returns for this very different type of wine. A value-weighted arithmetic repeat-
sales regression generates a geometric average return estimate of 4.8% between start-1972 and end-
2012. This compares with 6.9% for the (red) wine price index presented earlier. Second, for each vintage 
port that was included in the 1972 Wine Review, we check whether we can find a transaction at 
Christie’s London during the last three years of our time frame (2010–2012). We find five such 
instances: Croft 1945, Fonseca 1966, Warre 1955, and Taylor 1945 and 1963. The geometric average 
real returns implied by these price pairs range from 4.6% to 7.2%.  
This suggests that the returns on other types of wine have been somewhat lower over the last few 
decades. Yet the differences are not dramatically large, and even d’Yquem has performed as well as 
government bonds over the last four decades (net of storage and insurance costs). 
 
33 
 
6. Conclusion  
We first study how financial returns change over a wine’s life cycle. A simple model of wine 
prices delivers predictions of how prices can be expected to change differently over the life cycle for 
low-quality and high-quality vintages. We then construct a database containing prices for 9,492 
combinations of Bordeaux First Growth château, vintage year, year-end, and transaction type since end-
1899, and we use these data to estimate age effects in wine prices. We parameterize vintage effects by 
using measures of annual production yield and weather quality. The life-cycle price patterns implied by 
the results are generally consistent with our model. High-quality wines appreciate strongly while they 
mature, but their prices stabilize after a few decades. Prices start rising again when the wines become 
antiques. By contrast, wines from low-quality vintages appreciate little during the first years after 
bottling but show a near-linear price appreciation starting around the time that trading volume dries up. 
Our results are consistent with the existence of a nonfinancial payoff—increasing with age—from the 
ownership of rare wine, even if the nonfinancial return is probably small relative to the capital gains 
realized by wine collectors.  
Next, we estimate the long-term returns in the market for high-end wines. We apply a value-
weighted arithmetic repeat-sales regression to the price pairs in our database to construct a price index in 
real GBP terms. We find a geometric average real return of 5.3% between 1900 and 2012. Taking into 
account storage and insurance costs lowers this estimate to 4.1%. Over our time frame, wine has been 
outperformed by equities, and transaction costs could further reduce the relative attractiveness of wine. 
However, the performance of wine has been better than that of art and stamps. Moreover, we find 
evidence of positive correlation between wealth creation and wine prices. Consistent with this 
observation, we show that periods that are identified in-sample as ones when financial markets were 
overvalued have a spillover impact on the prices of collectibles. Future returns on a collectible (e.g., 
wine) also are predicted to be lower when the recent performance compared with other collectibles (e.g., 
34 
 
art and stamps) has been better. Finally, the historical returns on wine, impacted by the strong growth in 
the number of high net worth individuals worldwide, might have been higher than anticipated ex ante. 
 
  
35 
 
References 
Aït-Sahalia, Y., Parker J., Yogo, M., 2004. Luxury goods and the equity premium. Journal of Finance 59, 2959–3004. 
Ali, H., Nauges, C., 2007. The pricing of experience goods: the example of en primeur wine. American Journal of 
Agricultural Economics 89, 91–103. 
Ameriks, J., Zeldes, S., 2004. How do household portfolio shares vary with age? Unpublished working paper, Columbia 
Business School, New York, NY. http://www.ssrn.com/id=292017.  
Anderson, K., Nelgen, S., 2011. Global Wine Markets, 1961 to 2009: A Statistical Compendium. University of Adelaide 
Press, Adelaide, Australia. 
Ang, A., Papanikolaou, D., Westerfield, M., 2014. Portfolio choice with illiquid assets. Management Science 60, 2737–2761. 
Anson, J., 2011. Bordeaux and China. Wine Business International. May. 
Artprice, 2013. Artprice Global Indices. http://imgpublic.artprice.com/pdf/agi.xls. 
Ashenfelter, O., 1989. How auctions work for wine and art. Journal of Economic Perspectives 3, 23–36. 
Ashenfelter, O., 2008. Predicting the quality and prices of Bordeaux wine. Economic Journal 118, F174–F184.  
Ashenfelter, O., Ashmore, D., Lalonde, R., 1995. Bordeaux wine vintage quality and the weather. Chance 8, 7–14. 
Ashenfelter, O., Graddy, K., 2005. Anatomy of the rise and fall of a price-fixing conspiracy: auctions at Sotheby’s and 
Christie’s. Journal of Competition Law & Economics 1, 3–20. 
Bailey, M., Muth, R., Nourse, H., 1963. A regression method for real estate price index construction. Journal of the American 
Statistical Association 58, 933–942. 
Bloomberg News, 2012. Wine auctions drop 19%: Chinese demand cools for Bordeaux. December 18. 
Bocart, F., Hafner, C., 2015. Volatility of price indices for heterogeneous goods with applications to the fine art market. 
Journal of Applied Econometrics 30, 291–312. 
Bollen, N., 2007. Mutual fund attributes and investor behavior. Journal of Financial and Quantitative Analysis 42, 683–708. 
Broadbent, M., 1985. My life and hard wines. Speech at the Empire Club of Canada, Toronto, Canada. February 28. 
Burton, B., Jacobsen, J., 2001. The rate of return on investment in wine. Economic Inquiry 39, 337–350. 
Campbell, J., Giglio, S., Pathak, P., 2011. Forced sales and house prices. American Economic Review 101, 2108–2131. 
Chevet, J.-M., Lecocq, S., Visser, M., 2011. Climate, grapevine phenology, wine production, and prices: Pauillac (1800–
2009). American Economic Review 101, 142–146. 
Combris, P., Lecocq, S., Visser, M., 1997. Estimation of a hedonic price equation for Bordeaux wine: does quality matter? 
Economic Journal 107, 390–402. 
Di Vittorio, A., Ginsburgh, V., 1996. Des enchères comme révélateurs du classement des vins: les grands crus du Haut-
Médoc. Journal de la Société Statistique de Paris 137, 19–49. 
Dimson, E., 1979. Risk measurement when shares are subject to infrequent trading. Journal of Financial Economics 7, 197–
226. 
36 
 
Dimson, E., Karakaş, O., Li, X., 2014. Active ownership. Unpublished working paper, London Business School, London, 
UK. http://www.ssrn.com/id=2154724. 
Dimson, E., Marsh, P., Staunton, M., 2002. Triumph of the Optimists. Princeton University Press, Princeton, NJ. 
Dimson, E., Marsh, P., Staunton, M., 2013. Global Investment Returns Yearbook. Credit Suisse Research Institute, Zurich, 
Switzerland. 
Dimson, E., Spaenjers, C., 2011. Ex post: the investment performance of collectible stamps. Journal of Financial Economics 
100, 443–458. 
Dovaz, M., 1999. Château Latour. Editions Assouline, New York, NY. 
Financial Times, 2014. China overtakes France as biggest consumer of red wine. Daneshkhu, S., January 29. 
Goetzmann, W., 1993. Accounting for taste: art and the financial markets over three centuries. American Economic Review 
83, 1370–1376 
Goetzmann, W., Renneboog, L., Spaenjers, C., 2011. Art and money. American Economic Review 101, 222–226. 
Goetzmann, W., Spiegel, M., 1995. Private value components, and the winner’s curse in an art index. European Economic 
Review 39, 549–555. 
Graddy, K., Margolis, P., 2011. Fiddling with value: violins as an investment? Economic Inquiry 49, 1083–1097. 
Gyourko, J., Mayer, C., Sinai, T., 2013. Superstar cities. American Economic Journal: Economic Policy 5, 167–199. 
Heinkel, R., Kraus, A., Zechner, J., 2001. The effect of green investment on corporate behavior. Journal of Financial and 
Quantitative Analysis 36, 431–449. 
HMRC, 1999. Wasting assets: wines and spirits. Inland Revenue Tax Bulletin 42.  
Hong, H., Kacperczyk, M., 2009. The price of sin: the effects of social norms on markets. Journal of Financial Economics 93, 
15–36. 
Jaeger, E., 1981. To save or savor: the rate of return to storing wine. Journal of Political Economy 89, 584–592. 
Jones, G., Storchmann, K.-H., 2001. Wine market prices and investment under uncertainty: an econometric model for 
Bordeaux Crus Classés. Agricultural Economics 26, 115–133. 
Jorion, P., Goetzmann, W., 1999. Global stock markets in the twentieth century. Journal of Finance 54, 953–980. 
Jovanovic, B., 2013. Bubbles in prices of exhaustible resources. International Economic Review 54, 1–34. 
Jovanovic, B., Rousseau, P., 2001. Vintage organization capital. Unpublished working paper, National Bureau of Economic 
Research, Cambridge, MA. http://www.ssrn.com/id=263428. 
Kladstrup, D., Kladstrup, P., 2001. Wine and War: The French, the Nazis, and the Battle for France’s Greatest Treasure. 
Broadway Books, New York, NY. 
Kourtis, A., Markellos, R., Psychoyios, D., 2012. Wine price risk management: international diversification and derivative 
instruments. International Review of Financial Analysis 22, 30–37.  
Krasker, W., 1979. The rate of return to storing wine. Journal of Political Economy 87, 1363–1367. 
37 
 
LeRoy, S., 2004. Rational exuberance. Journal of Economic Literature 42, 783–804. 
Leve, J., 2014. Winery profiles. http://www.thewinecellarinsider.com/bordeaux-wine-producer-profiles.   
Lucey, B., Devine, L., 2015. Was wine a premier cru investment? Research in International Business and Finance 34, 33–51.  
Malmendier, U., Nagel, S., 2011. Depression babies: do macroeconomic experiences affect risk taking? Quarterly Journal of 
Economics 126, 373–416. 
Mandel, B., 2009. Art as an investment or conspicuous consumption good. American Economic Review 99, 1653–1663. 
Marks, D., 2009. Who pays brokers’ commissions? Evidence from fine wine auctions. Oxford Economic Papers 61, 761–
775. 
Masset, P., Weisskopf, J.-P., 2010. Raise your glass: wine investment and the financial crisis. Unpublished working paper, 
University of Fribourg, Fribourg, Switzerland. http://www.ssrn.com/id=1457906. 
Mei, J., Moses, M., 2002. Art as an investment and the underperformance of masterpieces. American Economic Review 92, 
1656–1668. 
Meltzer, P., 2005. Insuring your wine collection. http://www.winespectator.com/webfeature/show/id/Insuring-Your-Wine-
Collection_3536.  
Météo Climat, 2012. Météo Climat. http://meteo-climat-bzh.dyndns.org. 
Miles, J., 2009. History and development of the fine wine investment market. Speech at the Hong Kong International Wine 
and Spirits Fair Trade Conference, Hong Kong. November 4. 
Mitchell, R., 2012. Profit or pleasure? Exploring the motivations behind treasure trends. Barclays Bank Wealth Insights 15, 
6–54.  
Moskowitz, T., Vissing-Jørgensen, A., 2002. The returns to entrepreneurial investment: a private equity premium puzzle? 
American Economic Review 92, 745–778. 
Muhammad, A., Leister, A., McPhail, L., Chen, W., 2014. The evolution of foreign wine demand in China. Australian 
Journal of Agricultural and Resource Economics 58, 392–408. 
Pénasse, J., Renneboog, L., Spaenjers, C., 2014. Sentiment and art prices. Economics Letters 122, 432–434. 
Penning-Rowsell, E., 1972. Christie’s wine auctions in the 18th century. In: Broadbent, M. (Ed.), Christie’s Wine Review. 
Christie’s Wine Department, London, UK, pp. 11–14.  
Penning-Rowsell, E., 1973. Christie’s wine auctions 1801–1850. In: Broadbent, M. (Ed.), Christie’s Wine Review. Christie’s 
Wine Department, London, UK, pp. 29–33.  
Penning-Rowsell, E., 1975. Christie’s wine sales 1901–1914. In: In: Broadbent, M. (Ed.), Christie’s Wine Review. Christie’s 
Wine Department, London, UK, pp. 33–36.  
Renneboog, L., Spaenjers, C., 2013. Buying beauty: on prices and returns in the art market. Management Science 59, 36–53. 
Renneboog, L., Ter Horst, J., Zhang, C., 2011. Is ethical money financially smart? Nonfinancial attributes and money flows 
of socially responsible investment funds. Journal of Financial Intermediation 20, 562–588. 
Robinson, J., 2010. Professional wine storage in the UK. http://www.jancisrobinson.com/articles/a200809111.html.  
38 
 
Rosen, S., 1974. Hedonic prices and implicit markets: product differentiation in pure competition. Journal of Political 
Economy 82, 34–55. 
Schwert, G., 1990. Indexes of U.S. stock prices from 1802 to 1987. Journal of Business 63, 399–426. 
Sheppard, F., 1960. King Street. In: Survey of London, Volumes 29 and 30: St. James Westminster, Part 1. 
http://www.british-history.ac.uk/report.aspx?compid=40575.  
Shiller, R., 1991. Arithmetic repeat sales price estimators. Journal of Housing Economics 1, 110–126.  
Siegel, J., 1992. The equity premium: stock and bond returns since 1802. Financial Analysts Journal 48, 28–38+46. 
Simpson, J., 2011. Creating Wine: The Emergence of a World Industry, 1840–1914. Princeton University Press, Princeton, 
NJ. 
SoDivin, 2012. Grand cru vintage chart, from 1900 to 2003. http://www.sodivin.us/purchase-wine/Achat-de-vins/Grand-cru-
vintage-chart-from-1900-to-2003_11_638.html.  
Statman, M., Fisher, K., Anginer, D., 2008. Affect in a behavioral asset-pricing model. Financial Analysts Journal 64, 20–29. 
Staunton, M., 1992. Pricing of airline assets and their valuation by securities markets. Ph.D. dissertation. London Business 
School, London, UK. 
Stein, J., 1977. The monetary appreciation of paintings. Journal of Political Economy 85, 1021–1036. 
Swensen, D., 2000. Pioneering Portfolio Management: An Unconventional Approach to Institutional Investment. Free Press, 
New York, NY.  
39 
 
Table 1 
Descriptive statistics 
The table shows descriptive statistics for our database. Historical prices of the five red Bordeaux First Growths were 
collected from Christie’s London (an auction house) and Berry Bros. & Rudd (a dealer). The unit of observation is the 
average price (in British pounds) per bottle per quartet of year-end, château, vintage, and transaction type. Panel A shows the 
number of observations per transaction type and per château for each decade since end-1899. Panel B shows summary 
statistics for the distributions of nominal and real prices (deflated to end-1899 prices) for each decade since end-1899. 
Inflation data are from Dimson, Marsh, and Staunton (2013). 
 
Panel A: Number of observations per transaction type and per château  
Period 
Transaction type 
 
Château 
 
Total 
Auction Dealer   Haut-Brion Lafite Latour Margaux Mouton   
1899–1909 179 47 
 
12 88 36 48 42 
 
226 
1910–1919 37 69 
 
10 40 10 24 22 
 
106 
1920–1929 54 86 
 
9 52 25 45 9 
 
140 
1930–1939 62 141 
 
24 38 47 74 20 
 
203 
1940–1949 88 15 
 
18 24 21 24 16 
 
103 
1950–1959 47 131 
 
24 60 42 36 16 
 
178 
1960–1969 163 96 
 
47 44 91 45 32 
 
259 
1970–1979 1,568 89 
 
251 383 442 304 277 
 
1,657 
1980–1989 1,839 96 
 
341 436 419 344 395 
 
1,935 
1990–1999 1,956 198 
 
360 477 452 361 504 
 
2,154 
2000–2009 1,830 254 
 
320 448 475 347 494 
 
2,084 
2010–2012 409 38 
 
63 125 91 76 92 
 
447 
           1899–2012 8,232 1,260 
 
1,479 2,215 2,151 1,728 1,919 
 
9,492 
 
Panel B: Summary statistics of prices per bottle (in British pounds) 
 
Nominal 
 
Real 
Period Mean 
Standard 
deviation Minimum Maximum   Mean 
Standard 
deviation Minimum Maximum 
1899–1909 0.25 0.17 0.07 0.88 
 
0.23 0.16 0.06 0.86 
1910–1919 0.36 0.20 0.07 1.00 
 
0.26 0.15 0.06 0.87 
1920–1929 0.51 0.24 0.06 1.50 
 
0.24 0.12 0.03 0.73 
1930–1939 0.53 0.28 0.13 1.33 
 
0.30 0.16 0.07 0.75 
1940–1949 1.88 1.09 0.35 4.38 
 
0.80 0.47 0.15 1.85 
1950–1959 1.51 0.66 0.42 3.50 
 
0.45 0.22 0.15 1.19 
1960–1969 4.31 9.65 1.25 150.00 
 
0.82 1.86 0.24 29.09 
1970–1979 29.96 58.30 1.33 1,050.00 
 
2.22 3.83 0.16 53.75 
1980–1989 90.84 157.45 3.33 2,090.00 
 
2.96 4.97 0.13 62.56 
1990–1999 174.05 358.15 11.00 7,150.00 
 
3.55 6.97 0.25 135.10 
2000–2009 300.34 542.81 26.21 8,250.00 
 
4.86 8.95 0.39 132.03 
2010–2012 758.46 1,161.84 80.50 8,510.00 
 
10.04 15.63 1.04 110.42 
 
 
  
40 
 
Table 2 
Hedonic regression results 
The table presents estimates from our hedonic regressions using ordinary least squares. The dependent variable is the 
average price (in real British pounds) per bottle per quartet of year-end, château, vintage, and transaction type. Yield data (for 
an anonymous First Growth) come from Chevet, Lecocq, and Visser (2011). The weather quality variable is based on 
temperature during the growing season and rainfall during the harvest season of each vintage year. All age variables are 
converted from years to centuries: age has been divided by one hundred, age
2
 by ten thousand, etc. Robust standard errors are 
clustered at the year level and reported in parentheses beneath the coefficients. *, **, and *** denote significance at the 0.10, 
0.05, and 0.01 level, respectively.  
  (1) (2) (3) (4) (5) (6) 
Dealer 0.07 0.23*** 0.50*** 0.26*** 0.50*** 0.49*** 
 
(0.07) (0.07) (0.06) (0.06) (0.05) (0.05) 
Lafite 0.43*** 
  
0.31*** 0.16*** 0.16*** 
 
(0.04) 
  
(0.04) (0.03) (0.03) 
Latour 0.26*** 
  
0.22*** 0.17*** 0.17*** 
 
(0.03) 
  
(0.02) (0.02) (0.02) 
Margaux 0.11*** 
  
0.08*** 0.04** 0.04** 
 
(0.02) 
  
(0.02) (0.02) (0.02) 
Mouton 0.36*** 
  
0.37*** 0.31*** 0.31*** 
 
(0.04) 
  
(0.04) (0.04) (0.04) 
Ln(yield) 
 
-0.71*** 
 
-0.70*** -0.21*** -0.21*** 
  
(0.04) 
 
(0.04) (0.02) (0.02) 
Weather quality 
 
0.08*** 
 
0.08*** 0.07*** 0.04*** 
  
(0.00) 
 
(0.00) (0.00) (0.01) 
Age 
  
3.01*** 
 
2.90*** -0.80 
   
(0.46) 
 
(0.38) (0.96) 
Age
2
  
  
-1.58* 
 
-2.25*** 5.87** 
   
(0.82) 
 
(0.75) (2.44) 
Age
3
  
  
0.89** 
 
1.28*** -2.96* 
   
(0.43) 
 
(0.43) (1.56) 
Age × weather quality 
     
0.32*** 
      
(0.09) 
Age
2
 × weather quality 
     
-0.69*** 
      
(0.23) 
Age
3
 × weather quality 
     
0.36** 
      
(0.15) 
Year dummies? Yes Yes Yes Yes Yes Yes 
       F-test age variables 
  
439.36*** 
 
248.91*** 60.37*** 
F-test age-quality variables 
     
8.05*** 
N 9,492 9,224 9,492 9,224 9,224 9,224 
R-squared 0.45 0.61 0.69 0.62 0.73 0.74 
  
41 
 
Table 3 
Aging and wine prices 
The table shows estimates of the geometric average price appreciation over different segments of the life cycle for our 
benchmark model (reported in the last column of Table 2) and for a number of alternative specifications of the hedonic 
model. ** and *** denote significance at the 0.05 and 0.01 level, respectively. 
  
F-test 
age-
quality 
variables 
  Estimates of annualized price appreciation over life cycle 
  
Lowest-quality vintages 
 
Highest-quality vintages 
    
Age 
0–20 
Age 
20–40 
Age 
40–60   
Age 
0–20 
Age 
20–40 
Age 
40–60 
Age 
60–80 
Age 
80–100 
Benchmark model 8.05***   0.6% 1.9% 2.7%   4.0% 2.0% 0.9% 0.9% 1.7% 
            Before 1980 29.05*** 
 
-0.8% 2.3% 3.8% 
 
5.9% 1.3% -0.5% 0.5% 4.3% 
Since 1980 3.37** 
 
1.4% 1.7% 2.0% 
 
3.1% 2.1% 1.4% 1.1% 1.3% 
Auction data only 6.62*** 
 
1.1% 2.1% 2.6% 
 
3.6% 1.8% 0.9% 0.9% 1.8% 
Fourth-order age polynomial 8.37*** 
 
0.8% 1.9% 2.7% 
 
3.0% 2.4% 1.1% 0.4% 1.7% 
Add data on winter rainfall 4.75*** 
 
1.1% 2.1% 2.7% 
 
3.4% 1.9% 1.2% 1.2% 1.8% 
Price-based quality measure 34.86***   1.0% 2.5% 3.3%   4.5% 1.8% 0.4% 0.3% 1.5% 
 
  
42 
 
Table 4 
Wine versus other assets 1900–2012 
The table shows the distribution of returns (in nominal and real British pounds) for wine and other assets over the 
period 1900–2012. A deflated wine price index is estimated by applying a value-weighted arithmetic repeat-sales regression 
to the price pairs in our database. The resulting return estimates are corrected for storage and insurance costs. Data on British 
equities, government bonds, bills, and inflation are from Dimson, Marsh, and Staunton (2013). Art return data are from 
Goetzmann, Renneboog, and Spaenjers (2011), updated until end-2012 using data from Artprice (2013). Stamp return data 
are from Dimson and Spaenjers (2011), updated using the Stanley Gibbons GB30 index. 
  Mean returns per annum   Dispersion of annual returns 
  Geometric Arithmetic   
Standard 
deviation Lowest   Highest   
Nominal returns 
        Wine 8.2% 10.9% 
 
26.9% -34.9% 1949 144.3% 1942 
Equities 9.4% 11.2% 
 
21.6% -48.8% 1974 145.6% 1975 
Bonds 5.5% 6.1% 
 
11.9% -17.4% 1974 53.1% 1982 
Bills 4.9% 5.0% 
 
3.8% 0.3% 2012 17.2% 1980 
Art 6.4% 7.2% 
 
13.2% -31.2% 1930 46.6% 1968 
Stamps 6.9% 7.6% 
 
13.5% -8.8% 1982 83.2% 1979 
Inflation 3.9% 4.2% 
 
6.5% -26.0% 1921 24.9% 1975 
         Real returns 
        Wine 4.1% 6.7% 
 
26.3% -37.1% 1949 145.6% 1942 
Equities 5.2% 7.1% 
 
19.8% -57.1% 1974 96.7% 1975 
Bonds 1.5% 2.4% 
 
13.7% -30.7% 1974 59.4% 1921 
Bills 0.9% 1.1% 
 
6.3% -15.7% 1915 43.0% 1921 
Art  2.4% 3.1% 
 
12.4% -29.7% 1915 38.4% 1968 
Stamps 2.8% 3.5%   12.3% -19.2% 1915 56.3% 1979 
 
  
43 
 
Table 5 
Predictability of the returns on collectibles 
Panel A shows the correlations of the five-year returns on wine, art, and stamps with two return predictors. Panel B 
shows the results of a set of regressions with the five-year return on a collectible as the dependent variable. Standard errors 
adjusted using the Newey-West procedure (with four lags) are reported in parentheses below the coefficients. Returns to wine 
are estimated in this paper. Returns to art are from Goetzmann, Renneboog, and Spaenjers (2011), updated through the end of 
2012 using data from Artprice (2013). Returns to stamps are from Dimson and Spaenjers (2011), updated using the Stanley 
Gibbons GB30 index. Cyclically adjusted price-dividend ratios are taken from Dimson, Marsh, and Staunton (2013) and 
subsequently trend-adjusted. 
Panel A: Correlations  
 
Five-year real returns 
  Wine Art Stamps 
Prior performance relative to other collectibles -0.28 -0.28 -0.13 
Prior cyclically adjusted price-dividend ratio of equities -0.20 -0.20 -0.16 
 
Panel B: Regressions 
 
Sample 
  All Until 1955 After 1955 
Prior performance relative to other collectibles -0.22*** -0.25*** -0.19* 
 
(0.08) (0.09) (0.11) 
Prior cyclically adjusted price-dividend ratio of equities -0.30** -0.35 -0.24 
 
(0.14) (0.21) (0.17) 
Collectible fixed effects? Yes Yes Yes 
    N 312 156 156 
R-squared 0.13 0.15 0.14 
 
 
  
44 
 
Panel A: Low-quality vintage 
 
Panel B: High-quality vintage 
 
Fig. 1. An illustrative model of wine prices. The figure shows examples of life-cycle log price patterns implied by the model 
described in Section 2 for a low-quality vintage that starts deteriorating in drinkability immediately after the vintage (in Panel 
A) and a high-quality vintage for which drinkability first improves for 40 years and then remains constant (in Panel B). The 
following parameter values were used: r = 10%, g = 5%, z = 2%, a = 0.96, b = 1.12, and dL,1 = 0.10 × dH,1 = 0.01 × cL,0 = 0.01 
× 0.50 × cH,0. 
  
45 
 
 
Fig. 2. Christie’s auction catalogue (© Christie’s). The figure shows an excerpt (lots 114–121) from the annotated catalogue 
of the wine auction that took place at Christie’s London on December 9, 1935. The left page is from the original pre-sale 
catalogue and contains some handwritten notes of the auctioneer on the number of bottles in each lot and on commission bids 
submitted prior to the sale. The right page was added by the auctioneer and shows the price paid for each lot (in pounds, 
shillings, and pence), the equivalent price (in shillings) per dozen bottles, and the name of the buyer.  
  
46 
 
 
Fig. 3. Berry Bros. & Rudd price list (© Berry Bros. & Rudd). The figure shows two pages from the May 1909 price list. 
Prices are quoted in shillings per dozen bottles. 
  
47 
 
 
Fig. 4. Aging and wine prices. The figure shows the predicted life-cycle price patterns for the lowest-quality and highest-
quality vintages as implied by the coefficients on the weather quality variable, age variables, and the age-quality interaction 
terms in the benchmark model (reported in the last column of Table 2). Age is expressed in years. The price level for the 
lowest-quality category at age zero is set to unity. The dotted lines denote 95% confidence intervals. 
  
48 
 
 
Fig. 5. Aging and wine transactions at auction. The figure shows life-cycle average annual auction volume patterns for low-
quality and high-quality vintages, based on the number of sold lots at Christie’s London per combination of vintage year and 
sale year between 1985 and 2012. Age is expressed in years. We compute the average annual auction volume for each age 
both for relatively low-quality wines (weather quality < 8) and for relatively high-quality wines (weather quality > 12). We 
then average over five-year age groups (starting with ages 5–9). 
  
  
49 
 
 
Fig. 6. Deflated wine price index and nominal returns. The line in the figure shows a wine price index in real British pounds 
(against the right-hand axis), which is estimated by applying a value-weighted arithmetic repeat-sales regression to the price 
pairs in our database over the period 1900–2012. The index is set to unity at the start of 1900. Index levels in this graph are 
not adjusted for storage or insurance costs. The bars show the estimated nominal return for each year (against the left axis).  
  
50 
 
 
Fig. 7. Wine versus other assets 1900–2012. The figure shows price indexes in real British pounds for wine and other assets 
over the period 1900–2012. A deflated wine price index is estimated by applying a value-weighted repeat-sales regression to 
the price pairs in our database. The resulting return estimates from Fig. 6 are corrected here for storage and insurance costs. 
Returns on British equities, government bonds, and bills include reinvested income. These returns and inflation are from 
Dimson, Marsh, and Staunton (2013). Returns to art are from Goetzmann, Renneboog, and Spaenjers (2011), updated 
through the end of 2012 using data from Artprice (2013). Returns to stamps are from Dimson and Spaenjers (2011), updated 
using the Stanley Gibbons GB30 index. All indexes are set to unity at the start of 1900.