Wine hedonic prices: the Devil is in the details

Transcription

1 Wine hedonic prices: the Devil is in the details Commenté [FB1]: Wine hedonic prices: Is the devil in the details? Jean-Marie Cardebat Larefi University of Bordeaux, INSEEC Benoit Faye Inseec Business School, Bordeaux, Larefi Université de Bordeaux Eric Le Fur Inseec Business School, Bordeaux, Larefi Université de Bordeaux Philippe Masset Ecole Hôtelière de Lausanne Abstract. In this paper, the validity of information provided by the hedonic index on fine wines market was questioned. This is a very current issue indeed, as wealthy households and specific investment funds increasingly include collectibles in their portfolios. However, hedonic analysis implies two successive steps: (i) estimating the implicit prices of a finite number of attributes, and (ii) regressing these implicit prices on buyers characteristics to estimate bid functions while taking account of possible market segmentations. Yet, hedonic literature on collectibles (specifically in auction markets) fails to wholly consider the attributes and the buyers preferences, which are needed to run a complete hedonic analysis. Consequently, an unspecified (goods and agents) heterogeneity generates a price dispersion and biases the information provided by hedonic indices. Without the buyer s information, withheld by auction houses according to their non-disclosure agreements, we were only able to produce some back door solutions to improve market efficiency. First, we proposed an extended hedonic function to better control good heterogeneity by using auction expert comments. Second, we attempted to identify the attributes that provided the most dispersion according to their difference of valuation. A multiplicative heteroscedastic model was used in this way. Overall, this contribution provides a key reference to define the relevant components of wine price hedonic function, and spurs data providers to develop the information on wine products and to specify the origin of the buyers for raising the explanatory power of our hedonic models. Key words: Collectibles; Wine; Hedonic function; Auction prices; Taxonomy; multiplicative heteroscedastic model. JEL classification: B41, C22, G11, Q14 Corresponding author: Benoit FAYE, bfaye@inseec.com 1

2 1. Introduction. Since the 1990s, an increasing number of financial and economic studies have focused on the collectibles market (Burton and Jacobsen 1999). This interest is consistent with the rising weight of alternative assets in the portfolio of high-income households and specific investment funds (Credit Suisse 2014). Art, classic cars, coins, fine wines and stamps are the main components of collectible assets held by investors (EVI The Economist, 2013). A large academic effort has attempted to select and compute indices for each type of collectible to estimate the risks and returns. Repeat sales, hedonic and hybrid indices have been developed to provide market information to buyers. On the other hand, many composite indices have been calculated by professionals as well to evaluate the portfolios of collectible assets. In both cases, academic and practical works, the quality of the indices has been discussed. The market of fine wines is often considered by investors and collectors as the least illiquid amongst all collectibles. However, fine wines do not pay any cash-flows, are traded on a globalized and decentralized market and are present numerous fakes. These specificities generate information asymmetries and induce complexity in calculation of indexes. Two recent papers analyze this phenomenon. First, in the context of financial portfolio diversification, Fogarty and Sadler (2014) try to understand the impact of different index construction approaches (commercial indexes, hedonic and repeat-sales regressions, pooled, hybrid and average adjacent period return models). They conclude that the best results are provided by the pooled and hybrid models. Second, Masset and Weisskopf (2018) study the impact illiquidity on the different indices. Their results suggest that the diversification potential of fine wine is more limited than commonly believed. In our study, because they are often preferred by investors and collectors, we focus only on hedonic indices. Indeed, in capturing the other sources of price variance, hedonic indices are expected to better reveal the time effect on price levels. This approach refers to Rosen s (1974) hedonic analysis in which the goods and the buyer s preferences are supposed to be heterogeneous (but in finite numbers) on a competitive market (see Costanigro et al., 2011 for a synthetic applied presentation). Following Ashenfelter (1989), these two assumptions may be accepted when goods are sold in an English auction process, which is generally the case for fine wines. However, an issue arising in auction sales is the dispersion of prices. A high dispersion in auction prices means that a same item might have different prices during a same auction sale in violation of the law of one price (Cardebat et al., 2017). Which price will then be used for calculating the price index? Each indexes provider has its own answer. The method used will have an impact on the indexes trajectories and therefore on the assets valuation. Thus, we try to understand in which measure an incomplete information of hedonic indexes induces a bias in information obtained by participants in the fine wines market. This identification is crucial in the framework of the portfolio diversification for investors, collectors as well as professional actors (dealers, auction houses, brokers, wine funds ) and also consumers. To highlight this issue, we use a comprehensive database of auction prices with 7261 sales recorded during 120 auctions in six auction houses from October 2012 ( ) to March 2014 ( ). Our total sample contained 62 wines and 133 vintages (from 1816 to 2011). The first methodological stage identifies all the attributes required to describe a bottle of wine, and estimates different hedonic models for comparison purposes, including classical characteristics (used in the literature) and non-classical characteristics (unused in the literature) issued from the auction houses catalogues. Indeed, auction houses have a strict obligation of information to their buyers; therefore, they must produce a large set of comments on each lot 2

3 sold during an auction that can be encoded from their catalogues. The second methodological stage develops a multiplicative heteroscedastic model (Harvey 1976) from a complete hedonic function. This model permitted us to identify which attributes can be seen as a significant source of price dispersion and allows for opening a discussion on the valuation of these attributes by different categories of agents. The results are twofold. First, the findings reveal the significant influence of several unspecified attributes in the classical hedonic index, confirming that these classical hedonic regressions are probably biased and bring misleading information to the agents. Some significant characteristics of wines, such as the state of preservation, liquidity, and specifics, also might concern the valuation of other collectibles. Second, the multiplicative heteroscedastic model identifies the attributes which are source of price dispersion for instance, some speculative vintages or the state of preservation of the capsule and label. Depending on the sign of the coefficients, this model shows the attributes which increase or decrease the dispersion. In other words, it shows for which attributes there is a convergence or a divergence in the agents valuation. Therefore, these results open windows for discussing the role of buyers heterogeneity on auction markets. Indeed, certain attributes such as appellation, ranking, and vintages have explanatory weight so much strong on a diversified perimeter of wines that added attributes do not give the impression to generate variance. While if we focus on a perimeter more homogenous of vintage/wine the effect of these added attributes could be extremely significant on the prices dispersion. This paper contributes to the existing literature at different levels. First, findings suggest that the hedonic indexes for collectible assets could be improved by taking into account all the available information provided by auction houses. For the very first time, this paper proposes a way to encode this information. This method could be used in the future in the hedonic indices construction. Second, the paper participates in the price dispersion literature on auction sales, identifying (previously) unobserved products heterogeneity as a major source of price dispersion. These findings also have professional impacts. Indeed, investors have to highlighted on the previous risks and thus to be able to arbitrage between these risks and information costs. A solution could be to help the auction houses to standardize their comments and thus help the information providers to present unbiased indices. Finally, in the same way of the proposition of Masset and Weisskopf (2018), combine the technical expertise of academics with the large datasets and information held by some professional actors may help to develop more robust indices. The remainder of this paper is set out as follows. The second section presents the data and some descriptive statistics. The third section explains the methodology, presents and interprets the results, and runs some robustness checks. The fourth section discusses our results. The final section concludes the paper. 2. Data. As previously highlighted, high net worth individuals and investors have started to allocate part of their wealth into fine wines. In parallel, answering to this increased demand, wine funds have emerged giving more choice of investment in this asset (Ma sset and Weisskopf, 2015). However, most of indices are built with incomplete information. In order to define bias induced it we use an unsual database composed of 7261 sales were recorded during 120 auctions in six 3

4 auction houses between December 2012 and March 2014 (see Figure 1). Our total sample contained 62 wines from 133 different vintages (from 1816 to 2011). In order to not confound the effect of various characteristics, mixed lots were discarded. A mixed lot is defined as a lot containing different wines or vintages, or qualities of bottles, so that only one bottle lot has been selected. Sale prices are hammer prices in current USD (per liter for taking account of different bottle sizes). The hammer price includes the relevant buyers premium for the auction house, but is exclusive of sales taxes or VAT. A first set of variables is a common core of all variables traditionally found in hedonic analysis of fine wines: wine producer (e.g. Château s name in Bordeaux), vintages, wine color, appellation, region, classification, bottle volume, and expert quality rating (here Parker and Idealwine). Furthermore, the date, the place, and the auction houses were recorded for each transaction. Two variables were added to take into account the financial properties of the object and the declining price effect during each auction sale. The first variable was a measure of liquidity (LIQ) computed by the sum of sold lots of a wine-vintage-size bottle during the period from to in relative frequency, divided by the number of bottles produced (see Faye et al., 2015). The second variable, called the declining price, matched with the sale rank of a bottle when several bottles of the same wine-vintage were sold in the same auction (otherwise 0). Input variables are presented in Table 5 (available in the annex of the article) as well as their descriptive statistics (continuous variables in Table 1 and discrete variables in Figure 1). Table 1 Descriptive statistics of Parker scores and prices by bottle size and per liter Price U/liter Bottle Double Half Marie Nebuchadnezzanazar Salma- Note Statistic (US$) Balthazar (750 ml) Magnum Bottle Imperial Jeroboam Magnum Jeanne Melchior Methuselah Parker No. of obs Min Max 63,521 11,638 47,641 11, , ,804 53, , ,566 28,213 15, st Quart Median , rd Quart ,937 18,955 13, Mean ,721 16,517 11, St. Dev , Skewness Kurtosis Figure 1. Characteristics of the transactions recorded in the database. 4

5 5

6 The crucial contribution of the database was the encoding of the lot description, which was performed by auction house experts and published in auction catalogs. These descriptions indicate the state of the bottle sold (ullage, labels, capsule, cork, wax, glass), its packaging (packing), and the specification of some additional features, which was optional. Auction house experts use similar themes, terminology, and scales 1. The reading of the lot descriptions highlights this convergence of vocabulary and measurements used by the various auction house experts. Two rules were selected for encoding these description lots: exclusivity (the chosen words must have different meanings to constitute independent variables 2 ) and exhaustivity (the selected terms and measurement have to be used by all auction houses, otherwise a single variable merges the synonymous terms). In any auction house, bottles under the age of 10 years and/or in a perfect state are not described by the expert. In other words, no description corresponds to an excellent state of the attributes. 1 To see different charts of auction houses about lot descriptions, see and bottle conditions and descriptions. For Zachy s, Bonhams, and Hard Davis, bottle conditions are attached in each catalogue; for example, see (p. 4) (p. 4) (p. 210) 2 Technically, two terms are considered independents whether or not they are used together in at least one lot description in any auction house. 6

7 Traditionally, the lot description starts with the possible bottle numbering, then the ullage (the wine level in the bottle), the state of the labels (main label, back label, neck label, slip label, import label), the cork, and the capsule (or wax capsule). Finally, the lot description concludes with an expert s subjective assessment of the bottle sold and possibly some specifics of the lot (cf. infra). The ullage is a crucial datum. If it is normal that the wine level drops with natural evaporation, an excessive evaporation signals a problem caused by the easing of the cork. Ullage is differently assessed according to the bottle shape (identified by a dummy variable, p). It is measured by a level-rating system based on the shoulder of the bottle for the Bordeaux bottles and on the foil for sparkling bottles. The other bottle shapes are measured in centimeters numbered from below the cork. A level up to 3.5 cm below the cork is normal for wines under 20 years, 5 cm for wines between 20 and 40 years, and up to 5 7 cm beyond 40 years. For bottles of Bordeaux, normal levels are in the neck for wines under 10 years, very high shoulder for wines from 10 to 20 years, high shoulder for wines from 20 to 40 years, and mid shoulder for wines 40 years or more. The cork state is both information on the preservation of wine and an aesthetic criteria of the bottle. The cork level is precisely described: depressed or crowned, and protruding or raised (according to auction houses). For each term, the expert appreciates the alteration degree on a scale from 0 to 6. Furthermore, cork integrity may be altered by seepage. Three modalities describe the degree of seepage (0 = no sign of seepage; 1 = a slight sign of seepage; 2 = a sign of seepage). Note that because the state of the cork is being measured by scale variables, their sum may be an overall assessment (SUM CORK on a scale of 0 18). The capsule is the part of the bottle covering the cork, and it may be in metal (capsule) or wax (wax capsule). Besides its aesthetic character, its preservation function is crucial to maintain the quality of the wine. The auction house experts share a large vocabulary to describe the capsule state and, for each name, a scale rates the intensity of each type of degradation (from 0 excellent status to 6 bad status ). Three other variables are used to describe the absence of the capsule (MISSCAP), its disbonding (LOOSECAP), and finally its length (short capsule or not). These three previous variables are considered as dummies. A short capsule is generally affixed to the bottle at the end of the bottle reconditioning process. Variables describing the wax capsule are fewer and different from those of the (metal) capsule. All variables are considered as scale variables except for three dummies that describe the complete or partial disappearance of the wax (TOTMISSINWX) and possibly its reconditioning (REWAX). It should be noted that when the main label state does not indicate the winery and the vintage of a bottle, the experts may decide to cut the capsule for reading this information on the cork. However, we chose not to distinguish accidental and voluntary cuts, as the bottle would have deteriorated for the buyer. As in the case of the cork, we created an overall assessment of the capsule (SUMCAPS) and the wax capsule (SUMWX) varying from 0 to 90 and from 0 to 42, respectively. Each sum was the sum of the different scale variables. Finally and arbitrarily, the maximum score was affected in the case of the complete or partial disappearance of the capsule. The main label is the crucial aesthetic component of the bottle, even if its status does not impact the quality of the wine. Auction house experts share a very extensive and specific vocabulary to describe the main label state, and scale variables assess the degradation of each characteristic. Furthermore, several dummies specify the existence of other types of labels on the bottles: neck, 7

8 back, slip, and strip. Even if they are fewer in number and less precise 3 than those of the main labels, the characteristics describing these secondary labels are very similar. Multinomial variables are added for depicting the content of secondary variables (import countries, name of importer, vintages, etc.). Except for the strip and the import labels, overall assessments of labels (SUMAINLAB, SUMNECKLAB, SUMBACKLAB, SUMSLIPLAB) are implemented by summing up the scale variables or by dedicating the maximum score when the label is detached. Several studies outlined the effect of packing type (without packing, carton, original cardboard box, original gift box, wooden case, original wooden case, and strapped original wooden case) on price lot. The packaging is not only a kind of added value, but also protects the bottles and may be perceived as a guarantee against counterfeit products. Further, each packing type was encoded by a dummy. Two added dummies specify two types of original wooden case degradations: missing lid and damaged case. We finally note that experts sometimes decide to open a banded original wood case to inspect bottles inside. In this case, a tag that reads banded prior to inspection discloses the way in which the case has been opened attesting the prior conservation of bottles. Ullage, labels, capsule, cork, and packing are the basis of the lot description process. Three additional components complete this basis, concerning reconditioned lot, released bottles, and rare lot specificities. Each of them may have a heavy influence on the price. First, the reconditionment issue appears when the holder of a damaged fine wine bottle requires the dry material (capsule, cork, labels) to be reconditioned. This recondition process leads to a loss of authenticity; however, as mentioned previously, this depends on the place where the reconditionment is made. A bottle reconditioned in its original winery may be considered as authentic, while an unknown reconditionment location suggests a possible counterfeit. Furthermore, the number and the type of dry material reconditioned define the degree of unauthenticity. Finally, historicity is not authenticity and certain reconditionments are very old, so this historicity may be valued by collectors. Hence, auction house experts have to precisely describe the dry material concerned as well as where and when the bottle was reconditioned. The information for each of these aspects has been encoded in the database. Second, auction house experts need to precisely state whether the bottle comes from a release ; namely, the placing on the market by the winery of a bottle after the year of its vintage. Since wine conservation by the winery is a guarantee of quality, these bottles may be more expensive. Note that sometimes experts are not sure of the release character of a bottle and are only able to specify a believed late or recent release. Again, this issue was encoded in the database. Third, experts notify rare specificities concerning certain bottles, which can have positive or negative effect on the sale price. A first dummy identified the existence of these specificities in a sold lot. The first specificity concerned the main label. On the one hand, the main label may be silkscreened by a contemporary artist that appeals to collectors. On the other hand, the main label may be altered by certain inscriptions: autographed, pen marked, printed, stamped, or stickered. Some inscriptions can value the bottle. For instance, a mythic winemaker autograph may increase the price of the bottle, whereas any pen mark may decrease the price. In the same way, a tag like interdiction d importer aux USA may be valued by an American collector, 3 For instance, the rather general term damaged is not used in the main label description; however, it is for the secondary labels. Note also that this term is used with other descriptive terms in at least a description lot, which means it must be regarded as an independent variable. 8

9 whereas a médaille d or tag may not. A multinomial variable (INKSPECITYP) identifies the inscription type and another multinomial variable specifies its content. The second specificity concerns the shape and the composition of the glass bottle. For instance, some hand-blown glass bottles are made by master glassmakers implying particular shapes and volumes. Obviously, these bottle singularities may be very attractive for collectors. Furthermore, in the same winery, bottle shape and glass may change over time according to design selection and output constraints (e.g. green or blue glass during wartime). Consequently, a multinomial variable identifies all specificities of the shape and color of glass bottles. However, some modalities of these multinomial variables were sufficiently frequent to be included in the hedonic model, so these modalities were changed in the dummy variable. Lastly, auction house experts provide a subjective judgment concerning the equilibrium between the age of the bottle and the state of its attributes. These judgments tackle the general bottle appearance, the color and clarity of the wine in the bottle, and, in particular for white sweet wines, an amber color scale is supplied. Three scale variables measure these judgments. All these variables (presented in the annex, Table 6) provide a large informational spectrum on wine bottle attributes. Since we have completed the vector attributes, our methodological strategy will be tackled in the next section. 3. Methodology and results. The methodological strategy is twofold. First, we will assess the impact of a set of product heterogeneity variables (i.e. the encoded auction catalog comments describing each bottle and liquidity) on the price of wine. The underlying hypothesis is that these traditionally unconsidered variables play a role as price determinants. In this case, we expected that additional variables would be significantly different from 0. Therefore, in prior research work the price dispersion might have been due to the non-observation of these variables. The first equation is: P i = α + β 1 Hedonic i + β 2 Auction i + β 3 Declining i + β 4 Liquidity i + β 5 Catalog i + ε i (1) All quantitative variables are expressed in log form. P i is the price of a chateau-vintage pair i; Hedonic i is a set of variables gathering the traditional hedonic variables (chateau fixed effect, vintage, age, presence and value of experts grades, exceptional vintage, date of the sale, type of bottle size); Auction i is a set of variables revealing which auction houses and in which place a chateau-vintage pair i has been sold; Declining i is a variable controlling for the declining price effect (a price determinant during an auction following the literature on auction markets); Liquidity i measures the trading frequency of chateau-vintage pair i; Catalog i is a set of variables gathering all the encoded descriptions of the chateau-vintage pair i. ε i and δ i are the residual of respectively equation (1) and (2). A complete description of these variables is noted in the data section. Therefore, we had to run an equation where the dependent variable was now the price dispersion and not the price level. The underlying hypothesis was that the presence of heterogeneous agents on the auction market would imply a higher price dispersion (cf. Table 1). 9

10 Second, to test this hypothesis, we followed the procedure provided by Harvey (1976). The method consists of a two-step regression. In the first stage, we applied ordinary least squares (OLS) regression to equation (1) to obtain ε it. In a second stage, OLS regression of log(ε it 2 ) was realized as an estimation of Var(ε it ) (see Harvey 1976, p. 462). Var(ε it ) corresponds to the second moment of the price distribution and thus to the price dispersion. Equation (2) is then: log(ε it 2 ) = α + β 1 Hedonic i + β 2 Auction i + β 3 Declining i + β 4 Liquidity i + β 5 Catalog i + δ i (2) Then, if the first equation attempts to better control the product heterogeneity, the second tries to identify the attributes by which the buyer valuations may be the most diverse. These two steps highlight the risk generated by current hedonic indices. Concerning the outcomes, for readability concerns, we used a parsimonious presentation where only significant variables were exhibited in the tables of results. The châteaux and the date of sale dummy variables also were removed, although most of them were significant. Furthermore, all the coefficients are normalized in order to ensure the comparability between products characteristics impacts on price (eq.1) or on price dispersion (eq.2). The complete results are available from the authors upon request. Table 2 displays the hedonic regression results issued from equation (1) in the cases of the classical specification (model 1) and for a complete specification of attributes (model 2). This complete hedonic model (model, 2) exhibits a high explanatory power of (R² of 0.81) and a significant impact of the product heterogeneity (model 2). The catalog variables, as well as the traditional variables (parker grade, age, great vintage, bottle size, etc.), have the expected sign in most cases. In particular, the location of the sale, defined by cities as well as auction houses, played a major role on the wine price. Hong Kong appeared as the most expensive place, which is consistent with Masset et al. (2016) describing the Hong Kong prime. Four auction houses exhibited higher prices: Acker Merall, Sotheby s, Christie s, and Zachy s. The liquidity appeared as an important variable while the declining price was not significant. Regarding the catalog variables, unsurprisingly, the preservation problems negatively influenced the price of wine. 10

11 Table 2. Hedonic model including the bottle description (7261 observations) Model 1 (With classical attributes) Model 2 (Augmented hedonic) Model 1 (continued) Model 2 (continued) SCORE AND AGE LIQUIDITY 0.330*** PARKER score 0.055*** 0.001** CATALOG VARIABLES AGE 0.286*** 0.334*** CAPSULE CRACKED *** AGE^ *** 0.136* CAPSUME MISSING *** BOTTLE SIZE CAPSULE OXYDED *** BOTTLE *** *** CAPSULE TARNISHED ** DOUBLE MAGNUM *** *** CAPSULE TORN *** HALF BOTTLE *** *** CORK SIGN OF SEEPAGE *** IMPERIAL ** * BACK LABEL NICKED 0.004*** JEROBOAM *** *** BACK LABEL DAMP STAINED 0.004** MAGNUM *** *** MAIN LABEL WATER STAINED *** MARIE JEANNE *** *** MAIN LABEL GLUE STAINED *** METHUSELAH ** ** MAIN LABEL LOOSE ** VINTAGE MAIN LABEL NO *** VINTAGE *** 0.061*** MAIN LABEL TORN ** VINTAGE *** 0.017*** SLIP LABEL LOOSE *** VINTAGE *** 0.073*** ULLAGE BELOW SHOULDER 0.011*** VINTAGE *** *** ULLAGE CENTIMETERS BELOW CORK *** VINTAGE *** 0.107*** ULLAGE EXCELLENT LEVEL 0.078*** VINTAGE *** 0.030** ULLAGE HIGH TO MID SHOULDER *** VINTAGE *** 0.046*** ULLAGE INTO THE NECK 0.018*** VINTAGE *** 0.102*** ULLAGE LOW SHOULDER ** VINTAGE *** 0.051*** ULLAGE MID SHOULDER LEVEL *** VINTAGE *** 0.065*** ULLAGE LEVEL RATING SYSTEM *** VINTAGE *** 0.044*** PACKING BANDED CASES 0.016*** PLACES PACKING BORDEAUX BOTTLE 0.212** CHICAGO 3.070*** PACKING DAMAGE CASE 0.007** ALL PLACE 0.039*** PACKING ORIGINAL GIFT BOX ** HONG KONG 0.148*** RECONDITIONNED LOT 0.037** INTERNET * RECONDITIONNED BY CHATEAU 0.030** LONDON 0.058*** RECONDITIONNED NEW CORK *** NEW YORK 0.062** SPECIFICITIES INSCRIPTION TYPE 0.025* PARIS 0.023*** SPECIFICITIES BOTTLED BY ** SAN FRANCISCO 0.022* SPECIFICITIES 0.078*** AUCTION HOUSES WAX CAPSULE CHIPPED 0.016* ACKER MERALL 3.084*** WAX CAPSULE CROWNED 0.003* CHRISTIE S 3.583*** WAX CAPSULE RECONDITIONNED ** SOTHEBY S 3.703*** WAX CAPSULE SCUFFED 0.006*** ZACHY S 3.792*** Adjusted R² Notes: *** Significance at 1%; ** at 5%. All coefficients are normalized. 11

12 The main exception comes from the wax capsule, where the variables described damage as having a (weak) positive impact on the price. Indeed, this fact is not so abnormal when we consider that a missing part of a wax capsule allows for checking the state of the cork. This information was more crucial for buyers than the state of the wax capsule. Specificities and a reconditioned bottle or packing also had an ambiguous effect on the price. Precisely, as expected (see Cardebat et al., 2017) and even damages, banded original wooden cases had a positive impact on the price. In contrast, gift packing had a negative impact, probably because it did not offer the same guarantee against fakes. Reconditioned bottles also increased the risk of counterfeit products; therefore, buyers prefer the guarantee of reconditioning by the winery. Today, almost all reconditioned bottles sold in auction houses post this guarantee. This fact explains why reconditioned and reconditioned by the château have a positive impact on price while reconditioned cork (non-guaranteed by the winery) has a negative influence. Concerning specificities, if the bottler is not the winery the effect on the price is negative, while an inscription on the bottle (generally the sign of a reputed collector or winemaker) has a positive impact. The level of wine in the bottle has the expected result in all cases: the higher the level the higher the price. First, a better understanding of product heterogeneity allowed an increase in the explanation of price variance (from 0.77 to 0.81). This variation of the adjusted R-square may be considered weak, but at this explanation level this variation is crucial. This product heterogeneity can explain marginal differences in the price (price dispersion) but, except in extreme cases, cannot be seen as the main determinant of the price. The main determinants still remain the brand, the vintage, etc. Second, with hedonic specification of attributes being complete, the unexplained part of variance (more than 18% in model 2) could arise from agent heterogeneity. However, through lack of information about the nature of buyers, we could only attempt to identify the attributes for which buyers valuations were in disagreement, that is, the attributes which increase the price dispersion. Table 3 shows the results for the price dispersion equation (2). The aim was to establish which variables increased or decreased the price dispersion. Table 3 first posts variables with significant coefficients implying a high impact on price dispersion. The interpretation of the coefficients sign might be based on convergence and divergence between different categories of agents. If agents opinions are convergent (divergent) that is, they make a similar (dissimilar) evaluation of the considered product attribute then the coefficient is negative (positive) meaning that the dispersion is decreasing (increasing). A few variables had a significant impact on the price dispersion. Except the very speculative status of Hermitage La Chapelle, the only positive and significant impact came from the age variable and three relatively recent vintages (1989, 2000, and 2009) 4. These vintages are considered to be extremely sought after by investors and consumers due to their outstanding quality. The purchasing aims of the different categories of agents also seem different regarding the age of wines. This might explain the positive and highly significant impact of the age of the wine on the price dispersion. The higher the age, the greater the price dispersion. Moreover, bottle size had a decreasing impact on price dispersion: the higher the bottle size, the lower the price dispersion, meaning that there is a large consensus about this attribute. A consensus also exists concerning the presence of an expert s score, which reduce the risk about wine quality. Finally, we observed a strong agreement of buyers concerning bottle damage that reduces the was traded only from ; therefore, it is at the end of the studied period, while 2009 was traded during almost the whole studied period. 12

13 price dispersion. The main conclusion from Table 3 was that the age of the wine was the crucial variable explaining the price dispersion. Table 3 Estimation of the multiplicative heteroscedastic model (eq.2). Normalized coefficients. VARIABLES MODEL 1 (With classical attributes) MODEL 2 (augmented) NOTE IDEAL ** PARKER * AGE 0.191*** 0.192*** AGE^ *** *** VINTAGES VINTAGE ** *** VINTAGE ** *** VINTAGE *** VINTAGE * VINTAGE ** ** VINTAGE ** *** VINTAGE *** 0.038*** VINTAGE ** 0.027*** WINES HERMITAGE LA CHAPELLE 0.053*** 0.072*** ECHEZEAUX (DRC) * * ROMANEE CONTI (DRC) ** ** GRANDS ECHEZEAUX (DRC) ** * KRUG CLOS DU MESNIL ** BOTTLE SIZE BOTTLE *** * DOUBLE MAGNUM *** HALF BOTTLE *** IMPERIAL *** * JEROBOAM *** MAGNUM *** MARIE JEANNE *** MELCHIOR *** * METHUSELAH *** SALMANAZAR *** ** NEBUCHADNEZZAR *** * PLACES CHICAGO *** ** AUCTION HOUSES ACKER MERALL *** ** BONHAMS *** ** CHRISTIES *** ** SOTHEBY S ** ** ZACHY S *** ** STATE OF PRESERVATION CAPSULE CRACKED ** SLIP LABEL STAINED * BACK LABEL BIN SOILED *** BACK LABEL MISSING *** BACK LABEL NICKED *** MAIN LABEL TATTERED * NECK LAB WRINKLED *** SLIP LABEL LOOSE *** GLUED OVER MAIN LABEL ** SLIP LABEL SCUFFED ** PACK BG BOTTLE ** RECONDITIONNED BY CHATEAU ** 13

14 WAX CROWNED *** WAX NICKED *** R-squared: Adjusted R-squared: These results are in line with the existence of dissimilar types of agents/buyers on the fine-wine auction market who evaluate differently some of the wine attributes. This hypothesis would also be helpful to justify the 19% of unexplained variance of the complete hedonic model. However, the question of what these categories of agents are is still open. 4. Discussion. Our results indicate that there is a strong influence on the wine price of different attributes such as the specificities, the state of preservation, the location of the bottle sale and the liquidity. These results are very interesting and crucial in the framework of the portfolio diversification notably for investors, collectors as well as for consumers who wish to take pleasure drinking a great wine bought to the good rate. In order to discuss to this crucial point we retain the taxonomic process which is defined as the choice of a rule for grouping non-identical objects deemed equivalent in order to structure the environment (Bruner 1957). Economic theory of auctions and hedonic practice considers that objects with the same attributes (or belonging to the same taxonomic category according to the rule model) are identically valued. In addition, a taxonomic settlement is not only a structure of objects but also a structure of prices. Basically, three types of actors are discussed in the literature: consumers, collectors, and investors. Their goals, expectations, and the literature related to in the wine field are presented in the following table. Table 4 Literature on the existence of basic market segments. Agents Goals Basic level Main information needed Literature Consumers Quality and reputation Medium (Tirole, 1996) Reputation rating Combris et al. (1997), Cardebat and Figuet (2004, 2009), Lecocq and Visser (2006), Landon and Smith (1998), Delmastro (2005), Bombrun and Summer (2003), Schamel and Anderson (2003), Oczkowski and Doucouliagos (2014) Collectors Scarcity and singularity Low (Belk, 1988) Rarity (age), specificities, state of preservation Belk (1988, 1995), Apostolou (2011), O Guinn (1991), Grayson and Schulman (2000), Grayson and Martinec (2004) Investors Liquidity and return High (Baumol, 1986) Trade frequency, financial properties (transaction and holding costs, liquidity) Krasker (1979), Jaeger (1981) Dimson et al. (2015), Masset, Weisskopf, and M. Cossutta, (2015) Our results are partially in line with this taxonomy. First, the age is a characteristic inducing a higher price dispersion. The presence of collectors on the fine-wine market would justify such a price dispersion. Collectors are looking for old (and rare) vintages. Investors and consumers are not interested because of the illiquidity of such wines (investors) or because of the dubious 14

15 quality of an old wine (consumers). The collector would value positively the age of a wine while it would be the opposite for investors and consumers. Second, some of the most speculative vintages (2000 and 2009) are significantly associated with a higher price dispersion. The presence of investors on the fine-wine market might explain this situation. These vintages are very sought after and offer interesting return due to their increasing trend price. Furthermore, they are recent and then liquid, which is a fundamental characteristics for the investors. Consumers and collectors are less sensitive to this aspect. Their willingness to pay for these expensive vintages could then be lower than investors, explaining the price dispersion. Nevertheless, the liquidity variable is not significant. Third, the experts ratings are not significant in these results, while we expected that consumers would be more focused than other agents on this characteristic regarding the taxonomy analysis. Globally, the agents heterogeneity seems to be confirmed as a source of price dispersion following this basic taxonomy analysis. These agents heterogeneity issue is very new in collectibles (and specifically wine) literature. From an academic viewpoint, these findings provide a guideline for the relevance of variables included in hedonic function. It might help academics in the implementation of hedonic indices, improving the information provided to the agents. The outcomes also highlight the need to better consider the buyer heterogeneity in price modelling. However, auction houses do not deliver this information, and getting information on buyers still remains a challenge. From a professional viewpoint, these results outline that providers of market price indices could increase their relevance by taking into account all the dimensions of the product heterogeneity. The valuation process of collectible portfolios should be improved for all actors on the fine wine market: information providers, buyers and auction houses. 5. Conclusion. In this paper, we have tried to give a better understanding of consequences on the different categories of buyers on the fine wine market when using incomplete information contained in hedonic indexes. To do that, we built an extensive and comprehensive database (based on auction house comments) which was able to master a good heterogeneity of wine. Furthermore, capturing the impact of buyer heterogeneity on price dispersion without information on them implies a back-door approach. A hedonic function is estimated with an exhaustive vector of attributes drawn from an extended database of wine auction sales from to This analysis allowed us to evaluate the explanatory power of each category of attributes. We found two major results. First, we showed the significant influence on the wine price of the state of preservation, the specificities, the traded frequency (liquidity) and the location of the sale of a bottle (also taking into account all the traditional hedonic variables). The product heterogeneity (traditionally unobserved variables) was confirmed as significant determinants of the price and could explain the price difference between two bottles of the same châteauvintage couple. In particular, this hedonic regression would justify the rejection of the law of one price on the fine wine market. Second, we presented an explanation of the price dispersion and identified the wine attributes which represent the major sources of price dispersion. The interpretation is based on the agents heterogeneity. In particular, the goal and expectations of the agents seemed different regarding the age of the wine. This opens a room for a categorization of the buyers on this auction market and maybe for a taxonomic perspective of the hedonic approach. 15

16 References Apostolou, M. (2011) Why men collect things? A case study of fossilised dinosaur eggs, Journal of Economic Psychology 32(3), Ashenfelter, O. (1989) How auctions work for wine and art, The Journal of Economic Perspectives, Baumol, W. J. (1986) Productivity growth, convergence, and welfare: what the long-run data show, The American Economic Review Belk, R. (1988) Possessions and self. John Wiley & Sons, Ltd. Belk, R. W. (1995) Collecting in a consumer society. Psychology Press. Bombrun, H., and D. A. Sumner, (2003) What determines the price of wine, AIC Issues Brief, 18, 1-6. Bruner, J. S. (1957) On perceptual readiness, Psychological review 64(2), 123. Burton, B. J., and J. P. Jacobsen (1999) Measuring returns on investments in collectibles, The Journal of Economic Perspectives Cardebat, J. M., B. Faye, E. Le Fur, and K. Storchmann (2017) The Law of One Price in the Fine Wines Market, Journal of Wine Economics, 12(3) Cardebat, J. M., and J.M. Figuet (2004) What explains Bordeaux wine prices? Applied Economics Letters 11(5), Cardebat, J. M., and J.M. Figuet (2009) Estimation of a hedonic price equation for Alsace, Beaujolais and Provence wines, Applied Economics Letters 16(9), Combris, P., S. Lecocq, and M. Visser (1997) «Estimation of a hedonic price equation for Bordeaux wine: does quality matter? The Economic Journal, Costanigro, M., J. J. McCluskey, J. L. Lusk, J. Roosen, and J.F. Shogren (2011) Hedonic price analysis in food markets. The Oxford Handbook of the Economics of Food Consumption and Policy, Credit Suisse international (2014) Credit Suisse annual report, London. Delmastro, M. (2005) An investigation into the quality of wine: evidence from Piedmont, Journal of Wine Research 16(1), Dimson, E., Rousseau, P. L., & Spaenjers, C. (2015). The price of wine. Journal of Financial Economics, 118(2), EVI (2013), the economist valuables index, Fruit of passion, Aug 17 th, New York, Faye, B., E. Le Fur, S. Prat (2015) Dynamics of fine wine and asset prices: evidence from short-and long-run co-movements, Applied Economics 47(29), Fogarty, J. J., and R. Sadler (2014) To save or savor: A review of approaches for measuring wine as an investment, Journal of Wine Economics, 9(3), Grayson, K., and D. Shulman (2000) Indexicality and the verification function of irreplaceable possessions: A semiotic analysis, Journal of Consumer Research 27(1), Grayson, K., and R. Martinec (2004) Consumer perceptions of iconicity and indexicality and their influence on assessments of authentic market offerings, Journal of consumer research 31(2), Harvey, A. C. (1976) Estimating regression models with multiplicative heteroscedasticity, Econometrica: Journal of the Econometric Society, Jaeger, E. (1981) To save or savor: the rate of return to storing wine, The Journal of Political Economy, Krasker, W. S. (1979) The rate of return to storing wines, The Journal of Political Economy Landon, S., and C.E. Smith (1998) Quality expectations, reputation, and price, Southern Economic Journal,

17 Lecocq, S., and M. Visser (2006) What determines wine prices: Objective vs. sensory characteristics, Journal of Wine Economics 1(01), Masset, P., & Weisskopf, J. P. (2015). Wine funds: an alternative turning sour?. The Journal of Alternative Investments, 17(4), Masset, P., J. P. Weisskopf, and M. Cossutta, (2015) Wine tasters, ratings, and en primeur prices, Journal of Wine Economics 10(01), Masset, P., J. P. Weisskopf, B. Faye, and E. Le Fur, (2016) Red obsession: the ascent of fine wine in China, Emerging Markets Review 29, O Guinn, T. C. (1991) Touching greatness: the central midwest Barry Manilow fan club. Highways and buyways: Naturalistic research from the consumer behavior odyssey, Oczkowski, E. and H. Doucouliagos, H. (2014) Wine prices and quality ratings: A metaregression analysis, American Journal of Agricultural Economics aau057. Rosen, S. (1974) Hedonic prices and implicit markets: product differentiation in pure competition, Journal of political economy 82(1), Schamel, G., and K. Anderson (2003) Wine quality and varietal, regional and winery reputations: hedonic prices for Australia and New Zealand, Economic Record 79(246), Tirole, J. (1996) A theory of collective reputations (with applications to the persistence of corruption and to firm quality) The Review of Economic Studies

18 Annex Table 5 List of classical variables Categories Variables Types Wine colors Red, sparking, white wine, white sweet Dummies Wine names Chambertin Armand Rousseau, Chambertin Clos de Beze Armand Rousseau, Château Angelus, Château Ausone, Château Cheval Blanc, Château Cos d Estournel, Château Ducru Beaucaillou, Château Figeac, Château Gruaud Larose, Château Haut Brion, Château Haut Brion Blanc, Château L`Evangile, Château La Mission Haut Brion, Château Lafite Rothschild, Château Lafleur, Château Latour, Château Le Pin, Château Leoville Las Cases, Château Lynch Bages, Château Margaux, Château Montrose, Château Mouton Rothschild, Château Palmer, Château Pavie, Château Petrus, Château Pichon Longueville Comtesse de Lalande, Château Trotanoy, Château d'yquem, Châteauneuf du Pape Hommage à Jacques Perrin, Château de Beaucastel, Corton Charlemagne Coche Dury, Dom Perignon, Dominus, Echezeaux Domaine de la Romanee Conti, Echezeaux Henri Jayer, Echezeaux Henri Jayer, Grands Echezeaux Domaine de la Romanée Conti, Harlan Estate, Hermitage La Chapelle Paul Jaboulet Dummies Aîné, Krug Clos du Mesnil, Krug Collection, La Tache Domaine de la Romanee Conti, Louis Roederer Cristal, Masseto Tenuta dell'ornellaia, Montrachet Domaine Ramonet, Montrachet Domaine de la Romanee Conti, Montrachet Domaine des Comtes Lafon, Musigny Cuvée Vieilles Vignes Comte Georges de Vogüé, Musigny Domaine Georges Roumier, Musigny Domaine Comte Georges de Vogüé, Musigny Vieilles Vignes Comte Georges de Vogüé, Musigny Vieilles Vignes Grand Cru Comte Georges de Vogüé, Opus One, Ornellaia Tenuta dell'ornellaia, Penfolds Grange, Richebourg Domaine de la Romanee Conti, Romanee Conti Domaine de la Romanee Conti, Romanee St Vivant Domaine de la Romanee Conti, Sassicaia Tenuta San Guido, Screaming Eagle Cabernet Sauvignon, Vega Sicilia 'Unico', Vosne Romanee Cros Parantoux H. Jayer, Vosne Romanee H. Jayer Wine regions Australia, Bordeaux, Burgundy, Champagne, Tuscany, Napa, Rhône, Spain Dummies Vintages 133 vintages from 1816 to 2011 Dummies Auction companies Acker Merall, Bonhams, Christie s, Hart Davis, Sotheby s, Zachy s. Dummies Auction places Amsterdam, Chicago, E-auction, Geneva, Hong Kong, London, New York, Paris, San Francisco Dummies Dates (by month) of sales From to Dummies Price per bottle (US$) Hammer price in (current) US$ per bottle for any size. Continuous Price per liter (US$) Hammer price in (current) US$ per wine liter Continuous Parker scores Scoring publication by Parker Dummy Parker Score (from 50 to 100) Continuous Bottle sizes Balthazar, Bottle (750 ml), Magnum, Double Magnum, Half Bottle, Imperial, Jeroboam, Marie Jeanne, Melchior, Methuselah, Nebuchadnezzar, Salmanazar Dummies Declining price Rank of the lot sold in an auction when a wine vintage appears several times per auction, 0 otherwise Continuous Table 6 18