Experts, Reputation and the Price of Wine

Experts, Reputation and the Price of Wine February 13, 2017 Abstract The present paper contributes to the literature on the quality-price relationship in two dimensions. First, we empirically account for both short- and long-run quality effects. An increase in product quality has an immediate positive effect on wine prices but also improves the reputation of a winery, which again contributes to higher wine prices in the following periods. Simulation experiments indicate that this (endogenous) reputation effect is important. The second contribution of our paper relates to the growing literature in economics and marketing science on the effects of expert opinions and test reviews. Experts usually grade a selection of wines only. We find that analysing the selection of wines in a sample-selection framework is important for obtaining unbiased parameter estimates. Keywords: hedonic pricing, wine quality, reputation-effect, endogeneity, sample selection JEL code: C33, L66, Q11 1 Introduction Many markets for vertically differentiated products are characterized by information asymmetries, with producers usually being better informed about important product characteristics than consumers. Consumers may be unable to ascertained quality before actual consumption. In Akerlof s (1970) classical article, it is shown that this kind of uncertainty can hinder the operation of markets to the possible extreme of markets failing to operate despite obvious gains from trade. In the wine market, for example, consumers face a huge variety of different products characterized by a large set of attributes and some of these attributes cannot be evaluated before consumption. Hedonic pricing models for wine, such as Combris et al. (1997) for example, find that sensory attributes do not have an significant impact on market prices. The statistical 1

unimportance of individual sensory variables is interpreted as an indication for the consumers lack of information about wine quality attributes. 1 Wine can be considered an experience good (e.g., Hörner (2002), Ali and Nauges (2007)). Economists have investigated different mechanisms for firms to overcome the experience good problem (the fact that consumers may be unable to ascertained quality before actual consumption). The present paper focuses on two mechanisms that are particularly important: reputation and expert reviews. A number of theoretical studies suggests that producers concern for reputation might limit inefficiencies caused with asymmetric information in a repeat-business setting (Klein and Leffler (1981), Shapiro (1983), Allen (1984), Hörner (2002), Rob and Fishman (2005), Board and Meyer-Ter-Vehn (2013)). In Shapiro s classical paper, individuals make consumption decisions on the basis of firms reputation, which is again determined by product quality delivered in the previous period. The fact that reputation is endogenous implies that product quality may not only have a contemporaneous (direct) effect on market prices, but may trigger a long-run (indirect) effect via improved reputation for quality. Similar dynamic effects of investment in quality via reputation effects are studied in Kreps et al. (1982) and Board and Meyer-Ter-Vehn (2013). Board and Meyer-Ter-Vehn (2013) argue: Once quality is established, it is persistent and generates a stream of reputational dividends until it becomes obsolete (p. 2393). A number of cross-section studies (in the wine market) indeed suggest firm (or product) reputation to have a significant and positive impact on market prices. As noted by Board and Meyer-Ter-Vehn (2013), however, most of these studies are static and focus on quantifying the short-run value of reputation only. Our empirical analysis contributes to this literature by explicitly modelling reputation effects endogenously and thus taking into account both, short- and long-run effects of quality on product prices. We argue that empirical studies using cross-sectional data to investigate the (contemporaneous) relationship between product quality and market prices might produce biased results since long-run (reputation) effects are ignored. Investigating the impact of wine quality on the reputation of wineries empirically is made possible by the fact that our analysis is based on a rich panel of wines (and wineries) over time. This further allows to control for unobserved firm heterogeneity (which is not the case in most other empirical studies investigating this issue). The second contribution of our paper relates to the growing literature in economics and marketing science on the effects of expert opinions and test reviews. A wide range of media provides information on the quality of products based on expert reviews. Empirical studies have been carried out for different markets such as visual and performing arts (Ginsburgh (2003)), movies (Reinstein and Snyder (2005)), books (Ginsburgh (2003) and Sorensen (2007)), cars (Dewenter and Heimeshoff (2014) and Dewenter and Heimeshoff (2015)) and wine (Hilger et al. (2011) and Friberg and Grönqvist (2012)). The present paper addresses the issue of sample-selection for expert reviews. Wine 1 Oczkowski and Doucouliagos (2015) provide an excellent, recent overview of the debate surrounding the main question in the price quality literature: Does quality matter? Section 2 of the present paper provides more details. 2

guides usually grade a selection of wines from particular wineries only. We argue that this selection of wines for grading can lead to biased parameter estimates in the qualityprice relationship when the selection process is not explicitly accounted for an issue largely ignored in the existing empirical literature. In the empirical analysis, we find that the estimated effects of product quality on prices are indeed biased if endogenous sample selection is ignored. The remainder of the article is organized as follows. The next section 2 provides an overview of the theoretical and empirical literature on this topic. Section 3 describes the data set and outlines the empirical estimation strategy. Section 4 presents and discusses estimation results and section 5 summarizes and concludes. 2 Literature The theoretical foundations for investigating the relationship between product quality and market prices were developed in Rosen s (1974) hedonic model. Following first empirical applications (Golan and Shalit (1993); Oczkowski (1994); Nerlove (1995); and Combris et al. (1997)), the last two decades have seen a large number of hedonic pricing models in the wine market. The existing empirical literature investigating the quality-price relationship in this market delivers conflicting results, however. 2 A controversial issue in empirical studies on the quality-price relationship relates to the definition and measurement of wine quality. 3 Economists have made different assumptions as to whether and how true quality can be observed and measured. Hedonic analyses assume that both demanders and suppliers have perfect knowledge about the quality attributes of a product, and that these characteristics are objectively measureable (Rosen (1974)). Hedonic price functions may serve to reflect consumers preferences for quality only to the extent that consumers actually have perfect knowledge about the relevant product characteristics. Early hedonic studies (Combris et al. (1997), Combris et al. (2000) as well as Lecocq and Visser (2006)) include sensorial characteristics as explanatory variables in the price equation (in addition to objective attributes, such as vintage, denomination, grape variety, that usually appear on the label). Parameter estimates for some measures of a wine s aroma, body and finish would then indicate the consumers willingness to pay for these quality attributes. In the case of Bordeaux wine, Combris et al. (1997) observe that many of these sensory characteristics of wines do not play a role in the determination of the market price. The authors interpret the statistical unimportance of individual sensory variables as an 2 In the first issue of the Journal of Wine Economics, Lecocq and Visser (2006) summarize this literature and conclude that it has always been animated and controversial (p. 43). More recent surveys are provided in Estrella Orego et al. (2012) and Oczkowski and Doucouliagos (2015). 3 While some wine experts claim that rating wines is an intrinsically subjective process and that the quality of wine is difficult to define, as it is a multi-faceted construct, lacking a uniform and generally accepted definition (Hopfer et al. (2015), p. 8454) Parker s Wine Advocate asserts in every issue that wine is no different from any consumer product. There are specific standards of quality that full-time wine professionals recognize. 3

indication for the consumers lack of information about wine quality attributes. Establishing a reputation for quality is one way for firms to overcome the experience good problem (i.e. that fact that consumers are unable to ascertained quality before actual consumption). In the classical paper of Shapiro (1983), for example, consumers decisions are based on the reputation of the firm, which again is determined by product quality delivered in the past. Here, product quality does not pay off immediately and directly but only indirectly as higher quality feeds into the firm s reputation and future revenue. In a related study by Kreps et al. (1982), reputation is modelled as a Bayesian updating process: based on the observation of past transactions, sellers form a belief about the type of seller they interact with. The dynamic and long-run effects of investment in quality are explicitly stressed in Board and Meyer-Ter-Vehn (2013). The authors adopt a capital-theoretic approach to propose a model of firm reputation. A firm can invest in product quality and the firm s reputation is defined as the market s belief about this quality. They also show that investment in quality pays off with a delay. Most empirical studies, however, investigate the effects of reputation in a static (cross-section) setting. 4 Landon and Smith (1997) and Landon and Smith (1998) estimate several cross-section models for French premium wines. They observe that reputation explains much more variation in the consumers willingness to pay than does current product quality. Oczkowski (2001) also finds significant reputation effects for a cross-section of Australian premium wines. For the Italian premium wine market, Benfratello et al. (2009) suggest that reputation acquired by wines and producers during the years is more important than taste in driving market prices (p. 2206). They conclude that producers should aim at building a well established reputation both at wine and at firm level by promotional activities (e.g. participation to wine exhibitions) which facilitate citations in well-known guides (p. 2207). Empirical studies comparing the effects of reputation and measures of product quality by including both variables in hedonic cross-section price models can be misleading, however, since reputation and product quality are not independent. Good reputation comes from good quality. This interrelationship between reputation and product quality is the main reason why reputation-effects can eliminate the inefficiencies associated with asymmetric information. 5 In addition to any immediate (or short-run) impact of product quality on prices, an improvement in product quality will also have a positive (long-run) effect on reputation which again commands a price premium. Ignoring this endogenous reputation effect implies underestimating the effects of quality on market prices. 6 4 One notable exception is Cabral and Hortacsu (2010) who show that an ebay seller, who receives negative feedback, becomes more likely to receive additional negative feedback and is more likely to exit. 5 This interrelationship between reputation and quality is of key importance in Board and Meyer- Ter-Vehn (2013) and is also clearly expressed in the title of Shapiro s (1983) paper: Premiums for High Quality Products as Returns to Reputation. 6 An empirical analysis of the price-quality relationship should take into account both, short-run 4

The second important mechanism for transmitting information about product quality to consumers in the wine market are expert reviews. Wine experts (including winemakers) typically are better at evaluating the quality attributes of wine than novice consumers. A number of empirical studies rely on wine ratings from experts and publicly accessible wine guides, 7 such as scores given by the Australian wine critic James Halliday (Schamel and Anderson (2003)) or the wine magazine Winestate (Schamel and Anderson (2003)), Robert Parker s Wine Advocate (Ali and Nauges (2007), Dubois and Nauges (2010)), the Shield and Meyer guide (Oczkowski (1994)), the Wine Spectator (Landon and Smith (1997) and Landon and Smith (1998), Benfratello et al. (2009)), the Italian Duemila Vini guide (Benfratello et al. (2009)), or from six major Swedish print media (Friberg and Grönqvist (2012)). These empirical studies find that expert grades exert a statistically significant and positive impact on prices but the magnitude of these effects typically is rather small and short-lived. Ali and Nauges (2007), for example, conclude that a one-point increase in [Parker s] grade has almost no effect on the price set by producers (p. 96f). Examining weekly data on the demand for wine in Swedish state liquor stores over a time period of five years, Friberg and Grönqvist (2012) observe that the effects of favourable reviews generates a transitory, but quantitatively important increase in consumer demand. The effect of a positive review peaks in the week after publication and remains significant up to 22 weeks after the review. No significant effects of positive reviews can be found after half a year. The effect of negative reviews always is approximately zero. Although it is plausible to include expert evaluations (i.e. an index that depends on different quality attributes of the product) published in wine guides in the price equation, this approach also has a number of potential drawbacks. A key problem is the selection of wines in the wine guide. When using expert evaluations published in wine guides in hedonic pricing models, the assumption is implicitly made that the selection of wines into these wine guides is random and in particular uncorrelated with the quality of wines. If, however, the selection of wines is correlated with the quality, parameter estimates on the quality-price relationship will be biased. 8 and long-run effects of quality. This would require to explicitly model the effects of current quality on firm reputation. As an alternative, a reduced form approach would include lagged product quality in the price equation (Landon and Smith (1997), Landon and Smith (1998), and Ali and Nauges (2007)). 7 Exceptions are the studies carried out on the basis of data sets obtained from a random selections of wines in France (Combris et al. (1997), Combris et al. (2000) and Lecocq and Visser (2006)) 8 Combris et al. (1997) identify five conditions that are required for using expert reviews in econometric quality-price models and conclude that the widely published and easily accessible wine guides do not in general verify these conditions. Their first condition is that all the wines that are tasted should be included in the sample, regardless of whether the wine is considered good or bad. In wine guides the wines of inferior quality are often deliberately under-represented for commercial reasons. (p. 392).) Note that sample selection effects are also mentioned in one of the first hedonic studies for wine (Nerlove (1995). In contrast to the usual hedonic regression, in which unit variety prices are regressed on a vector of quality attributes, Nerlove estimates a regression on quantity sold on price and quality attributes. This is justified by the assumption that prices and attributes can be taken as exogenous to the Swedish consumer. The author argues: Of course, to the extent that wines imported into Sweden are not a random sample of wines produced world-wide, there is a selectivity bias 5

In the following section, we explicitly investigate these two mechanisms for transmitting information about product quality to consumers: reputation and expert s product reviews. Specific emphasis will be given to endogenous reputation as well as sample selection issues in expert reviews. 3 Data and Estimation Strategy The sample used in the present paper consists of 24,547 wines which are produced between 2004 and 2007 by 488 wineries in Austria. These wineries cover about 35% of the annual production of quality wines. For each wine, numerous characteristics are available, such as its price per standard bottle, type (white, red, rose or sweet) and variety of grape, the year of harvest, age (the time span between harvest and supply to the market) as well as the size of the winery and the diversity of its assortment. Summary statistics can be found in the Appendix. This information is matched with data on the sensory quality of the specific bottle of wine, which is obtained from consulting the most influential guide on Austrian wine the annually published Falstaff-Wine-Guide. Experts grade on a scale from 1 to 100 on color and appearance, aroma and bouquet, as well as flavour and finish. This expert evaluation of wine quality will be denoted q E in the following. The average grade in our sample is 89.05 and only wines on the scale between 82 and 100 are included in the Falstaff-Wine-Guide. It is important to note that producers usually publish price lists and start selling wines of the current vintage between February and June, while the wine guide (including experts grades) is not published until July of the same year. Therefore, producers pricing decisions are determined by their own evaluation of product quality (q W ), which is unobservable and not necessarily corresponds to the experts evaluation in the wine guide (q E ). 9 Data from the Falstaff-Wine-Guide also include information on the reputation of a winery, which is classified on a scale from 0 to 3 between 2004 and 2006 and from 0 to 5 in 2007. To avoid the different scaling of this variable to affect our estimation results, we use relative reputation in the empirical analysis (defined as the level of reputation relative to the maximum level of reputation in that particular year). In order to model long-run effects of quality on prices (i.e. reputation effects), we treat reputation as an endogenous variable. Reputation of a particular producer/winery is determined by the quality of its wines delivered in the past. To investigate this effect empirically, we also collect information on the quality of wine produced in previous present even in the estimates of such confluent relationships (p. 1707). Other problems associated with expert reviews are discussed in Ginsburgh (2003). Oczkowski and Doucouliagos (2015) mention potential problems with the use of expert quality ratings in estimating the quality-price relationship, such as the inconsistency between various expert judges (consensus) and the inconsistency of the same expert judge over time (reliability) when assessing sensory quality. 9 The existing literature differs on who is assumed to observe true wine quality. Ali and Nauges (2007) and Ali et al. (2008) assume that experts make correct assessments about product quality, while Dubois and Nauges (2010) assume that quality is correctly observed by producers only. 6

years (from 1999 until 2003). From observations on the quality ratings of all wines of a particular winery produced at time t k (with k = 1,..., 4), we compute the average quality of wines produced by this winery at t k. 10 This measure of average quality of wines in previous time periods is expected to have a positive impact on the reputation of the winery. We further expect to find, that the impact of average wine quality on reputation becomes weaker as k increases. Note that the Fallstaff-wine guide does not grade all wines produced by the 488 wineries in our sample. Quality grades are available for a subset of 9.914 bottles of wine only (about 40% of all wines in our data set). The number of wines graded per year also differs across wineries: for most wineries, six wines per vintage are graded (see Figure 1). While five (four) wines are graded for 20% (12%) of all wineries, only 14% of all wineries have more than six wines graded in the wine guide. Typically, these are wineries with a high reputation: reputation averages 0.68 for this group of wineries, whereas the average reputation is 0.20 for wineries where six or less wines are graded. Since not all wines produced by a winery can be graded by the wine guide, each winery has to decide which wines to submit for evaluation. 11 Clearly, this decision will be based on the wine growers subjective evaluation of the quality of his/her wine (q W ), which is unobservable to the econometrician. If the wine grower s subjective evaluation of a particular wine is very positive, the probability of submission to the wine guide will be high and at the same time, the wine grower will demand a high price. In the case where the experts evaluation (q E ) is identical to that of the wine grower (q E = q W ), i.e. the experts evaluation also is very positive, we should expect to see a positive correlation between the observable quality ratings (q E ) and wine prices. If, however, experts and wine growers disagree on the quality of a particular wine, two types of errors can occur. In the first case, wine growers overestimate the quality of their wine (i.e. q W > q E ) and the price of the wine (determined on the basis of the optimistic evaluation of the wine grower) will be higher than expected given the experts evaluation (q E ). In the second case, where q W < q E, wine prices will be lower than expected on the basis of the observed rating of the experts. If both cases occur with the same probability, the quality evaluation of wine growers and wine experts should be equal on average (E[q W ] = E[q E ] and the relationship between prices and experts evaluation will be unbiased. However, if the wine growers submission of wines to the wine guide is based on their subjective evaluation of wine quality (the producer only sends the best wines to the wine guide and the probability of submission S increases with q E ), we expect to find the first case with q W > q E to be overrepresented in our data set. This would imply that (a) our measure of product quality (q E ) although probably 10 We also experiment with using the highest rating of a wine from a particular winery at time t k. Pennerstorfer and Weiss (2013) show that the form of aggregation of individual quality components (average quality vs. maximum quality) can be crucial in influencing firms incentive to invest in high quality products. 11 Note that the wine guide is not obliged to publish the results of all wines received by the wineries; rejections to publish reviews of particular wines are very rare events, however. So, basically, the producer decides which wines to select for grading rather than the wine guides experts. 7

percent Figure 1: Number of Graded Wines per Winery (in percent) 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 exogenous in the entire population of wines is endogenous in the selected sample and (b) the relationship between experts grading (q E ) and wine prices is overestimated when sample selection is not taken into account. Since the experts evaluations will also influence the reputation of the winery, we expect to find a similar sample selection bias with respect to the impact of reputation. The econometric model to be estimated is: P rob (S t = 1 Z) = Φ (Zγ) (1) E [ p t ( q E t, r t, X ), S t = 1 ] = α 1 q E t + α 2 r t + Xα + ρ 1 λ (Zγ) (2) E [ r t ( q E t k, Y ), S t = 1 ] = β 1 q E t k + Y β + ρ 2 λ (Zγ) (3) In equation (1), S t indicates selection/grading in the wine guide of wine i from winery (producer) w at time t (subscripts i and w are ignored to simplify notation) with S t = 1 if a particular wine is graded in the guide and S t = 0 otherwise. Z is a vector of explanatory variables, γ is a vector of unknown parameters, and Φ is the cumulative distribution function of the standard normal distribution. Estimation of a probit-model yields results that can be used to predict the probability of selection and grading in the wine guide for each individual wine. In a second stage, we correct for self-selection in the price and reputation equation (equations (2) and (3)) by incorporating a transformation of these predicted individual probabilities as an additional explanatory variable. The conditional expectation of prices given the wine is graded is determined by the experts quality grading (qt E ), reputation (r t ) and a vector of additional explanatory variables (X). λ is the inverse 8

Mills ratio evaluated at Zγ. Similarly, the conditional expectation of reputation of a winery given the wine of this winery is graded is determined by previous quality gradings (qt k E ), another vector of explanatory variables (Y ) and the inverse Mills ratio λ (Zγ). α 1, α 2, β 1, ρ 1 and ρ 2 (α and β) are (vectors of) unknown parameters. In estimating equations (1), (2) and (3), we include winery fixed effects (µ w ) to control for unobserved and time-invariant heterogeneity between wineries like differences in costs, soil conditions, consumer preferences or managerial skills. We further include time fixed-effects (a dummy variable ω t ) to control for unobserved differences in the quality of wine between different years. 4 Results Selection of Wines for Grading Table 1 reports results of a panel-probit model estimated for explaining the variation in the binary variable S iwt. The econometric model correctly classifies 74% of all observations; 72% of wines that are actually submitted for grading are predicted correctly. The number of times the particular wine has been graded before (between 1999 and 2002) has a significant and positive impact on the probability of the wine to be selected for grading in a particular year in the sample period (2004-2007). An increase in the average quality of wines from a particular winery in the past tends to increase the probability of a wine to be considered for grading. Conditioned on the number of wines produced, this indicates that an increase in average quality tends to increase the likelihood of submission and grading in the following years. However, parameter estimates for the average quality of wines are significantly different from zero at the 10%-level in the previous year only, but insignificant in the years before. The larger the number of different wines produced, the smaller is the probability of a single wine to be considered for evaluation. This negative effect of diversity of production on selection probability can be explained by the limited space available in the wine guide; only the winery s best wines will be submitted for grading and only a very small share of wineries will succeed in having more than six wines graded. A longer time span between harvest and selling (age of wine) is associated with a higher probability of a wine to be selected for evaluation. This might be explained by the fact that these wines can be expected to receive higher grades. Winery Reputation and Wine Prices The results of a two-stage least squares (2 SLS) estimation on prices and reputation are summarized in Table 2. Columns [1] and [2] include fixed effects for wineries, vintage, variety of grapes and types of wines but ignores the sample selection problem. The parameter estimates for the quality of a particular wine are positive and significantly different from zero at the 1 %-level in the price equation. An increase in wine quality by one index-point increases prices by 10.5%. This quantitatively important effect on 9

Table 1: Regression Results Explaining the Selection of Wines for Evaluation Method Probit Dependent variable Select Variables Coeff. Std. Dev Sign. Number of times wine received grade 0.3513 (0.0132) *** between 1999 and 2002 Average quality (at t 1) 0.0408 (0.0222) * Average quality (at t 2) 0.0122 (0.0190) Average quality (at t 3) -0.0242 (0.0167) Average quality (at t 4) 0.0095 (0.0176) Number of different wines of winery (in year t) -0.0406 (0.0105) *** Age: 2 Years 0.4609 (0.0686) *** Age: 3 Years 0.8468 (0.1569) *** Constant -4.5457 (4.4980) Winery effects Yes (346) Vintage effects Yes (3) Variety of the grape Yes (31) Type of wine Yes (6) Number of observations 16,758 Log-likelihood -8,722 Pseudo-R 2 0.241 χ 2 -test statistic on instruments (df) [p-value] 768.4 (10) [0.0000] Notes: Standard errors are reported in parentheses and are clustered at the winery level. *** significant at 1 %, ** significant at 5%, * significant at 10 % level. 10

prices is in contrast to some earlier empirical studies who find that the magnitude of effects of expert s grades are very small. Despite the difficulties in objectively and consistently assessing wine quality using expert ratings, these ratings seem to correlate with producers quality evaluations and to provide valuable information for consumers in the present context. Consistent with the existing empirical studies, winery reputation also has a significant and positive effect on product prices. An increase in reputation from 0 to 1 commands a substantial price premium; according to column [1] of Table 2, prices increase by 67.7%. If the time span between the harvest of the grapes and selling to consumers (age of wine) is two years instead of one year (the reference category) prices increase by 38.3%. A time span of three or more years is associated with a price increase of 76.8%. The endogeneous variable in column [2] is the reputation of a winery. Statistical tests reject the assumption that winery reputation is exogeneous; the explanatory variables used significantly contribute to the explanatory power of the model. A Hansen-J-test of overidentifying restrictions is not rejected at the 10%-significance level, indicating that the explanatory variables for winery reputation are valid (i.e. uncorrelated with the error term). An increase in current and past wine quality is associated with a higher reputation of the winery. Column [2] further suggests that the effect of average quality on reputation decreases over time (with the exception that the effect of average quality at t 3 is slightly larger than at t 2). The parameter estimate for the average quality in t 4 is still positive but not significantly different from zero at the 10%-level. Wine age has a significant and negative impact on reputation. Columns [3] and [4] explicitly consider sample selection effects by including the inverse Mills ratio (IM R(λ)) computed from the selection equation. In the model estimated on wine prices (column [3]), the parameter estimate for IM R is significantly different from zero at the 10% level. 12 Accounting for sample selection reduces the impact of an increase in one quality point on prices to 7.1% while an increase in reputation from 0 to 1 is associated with an increase in prices by 38.8%. The impact of quality therefore drops by roughly one third, and the influence of reputation by nearly one half. The parameter estimate for IMR is not significantly different from zero in the auxiliary regression on reputation. Consequently, correcting for the sample selection hardly affects these regression results. 12 Since the IMR is a so-called generated regressor rather than an observed variable the standard errors and the test statistics of the structural equation should be corrected (Wooldridge, 2001). However, adjusting the variance of the parameter estimates of the structural equation because of the two-step estimation is cumbersome (p. 564). We circumvent this problem by using bootstrapped standard errors. As the standard deviations of the parameter estimates are hardly affected when using bootstrapped standard errors we report these results in an appendix, which is available upon request. The results support Wooldridge (2002) supposition that [i]n many cases, the adjustments do not lead to important differences (p. 562). 11

12 Table 2: Regression Results Explaining the Effects of Quality and Reputation on Prices Method 2SLS 2SLS Model [1] [2] [3] [4] Dependent variable ln(price) Reputation ln(price) Reputation Variables Coeff. Std. Dev. Sign. Coeff. Std. Dev. Sign. Coeff. Std. Dev. Sign. Coeff. Std. Dev. Sign. Quality 0.1049 (0.0035) *** 0.0058 (0.0009) *** 0.0705 (0.0150) *** Reputation 0.6770 (0.1918) *** 0.3884 (0.1643) ** Average quality (at t 1) 0.0338 (0.0067) *** 0.0330 (0.0069) *** Average quality (at t 2) 0.0134 (0.0054) ** 0.0125 (0.0055) ** Average quality (at t 3) 0.0209 (0.0053) *** 0.0205 (0.0053) *** Average quality (at t 4) 0.0061 (0.0042) 0.0063 (0.0043) Wine age: 2 Years 0.3830 (0.0171) *** -0.0098 (0.0031) *** 0.4293 (0.0279) *** -0.0026 (0.0066) Wine age: 3 Years 0.7678 (0.0399) *** -0.0249 (0.0099) ** 0.8561 (0.0610) *** -0.0077 (0.0118) IM R(λ) -0.0330 (0.0193) * 0.0005 (0.0126) Constant -7.6367 (0.3198) *** -6.7727 (1.2902) *** -4.4488 (1.3506) *** -6.0788 (1.3173) *** Winery effects Yes (353) Yes (353) Yes (346) Yes (346) Vintage effects Yes (3) Yes (3) Yes (3) Yes (3) Variety of the grape Yes (31) Yes (31) Yes (31) Yes (31) Type of wine Yes (3) Yes (3) Yes (3) Yes (3) Type of sweet wine Yes (3) Yes (3) Yes (3) Yes (3) Instrumented variables Reputation Reputation Quality Number of observations 7,403 7,403 7,358 7,358 R 2 0.823 0.940 0.822 0.940 Hansen-J χ 2 -test 5.685 df = (3) p = 0.1280 10.991 df = (8) p = 0.2022 F -test on exogeneity 50.419 df = (1, 353) p = 0.0000 24.842 df = (2, 346) p = 0.0000 F -test on instruments 8.768 df = (4, 353) p = 0.0000 4.254 df = (10, 346) p = 0.0000 Notes: Standard errors are reported in parentheses and are clustered at the winery level. *** significant at 1 %, ** significant at 5 %, * significant at 10 % level. Winery level fixed effects have been partialled out to calculate the Hansen-J-test statistic. The F -tests on endogeneity of instrumented variables (H0: instrumented variables are exogenous) are based on a robust regression tests developed by Wooldridge (1995). The F -test on instruments tests the hypothesis that the coefficients on the instruments in the regression on reputation are jointly zero. The regression on reputation includes all exogenous variables of the regression on prices and (in model [1]) on quality.

Summarizing the results of the estimation models, we conclude that an increase in wine quality has (a) an immediate and positive effect on wine prices as well as (b) a significant and positive (long-run) effect on the reputation of the winery, which allows wineries to raise prices even further. Figure 2 illustrates the size of short- and long-run price effects of quality improvements. We apply a bootstrap simulation technique to account for the uncertainty of the estimated parameter values. Each parameter is drawn randomly from a normal distribution with the mean and the standard deviation obtained from the regressions in columns [3] and [4] of Table 2. Figure 2 (a) shows the price effects if a winery increases the quality of each wine in a particular year (t) by one point. Wine prices in year t increase by 7.1% on average. In the year after the (non-recurring) quality improvement wineries are still expected to charge 1.3% higher prices for the new wines. This additional (lagged) effect of quality is due to the fact that higher quality of a particular wine improves the reputation of the winery which again has a positive effect on wine prices. In the following years (t + 2 and t + 3), the growth of prices is still positive but below 1%. Figure 2 (b) shows the cumulative effect of a one-time quality increase by one point. The largest price-effect is observed in the period of the quality increase (7.1%), but sizeable price increases can be observed in the years after the quality improvement. The cumulative price effect amounts to 9.9% after four years. Due to this (endogenous) reputation effect of quality, the long run price effects are about 40% larger than the contemporaneous price increase. 5 Summary and Conclusions The present paper investigates the impact of product quality on wine prices. We extend the existing literature on this topic in two dimensions. First, in addition to the direct effect of quality on price, we also explicitly model indirect reputation effects. And secondly, we control for selection effects in the quality-price relationship. The empirical analysis is based on 24,547 observations (bottles of wine) from 488 wineries in Austria in the period 2004 until 2007. Note that only a subset of all wines of a particular winery can be graded in the Fallstaff-Wine-Guide. The fact that submission for grading as well as the decision on the price of a particular wine depends on producer s quality evaluation (which is unobservable and can deviate from the experts evaluation) introduces a bias in the estimated quality-price relationship. Explicitly modelling the producers /wineries selection of wines for grading in a sample-selection framework suggests that selection effects are important in the price equation. Consistent with the existing literature, we find that expert quality ratings exert a positive and significant impact on the price of wine. However, this positive effect of product quality on the price of wine becomes weaker once the selection of wines is explicitly taken into account. In addition to this direct effect of quality on price, we also observe a significant indirect effect: An increase in product quality improves the reputation of a winery in 13

Figure 2: Short- and Long-Run Price Effects of Quality 10% 8% (a) additional effect 14% 12% (b) cumulative effect 6% 10% 4% 8% 2% 6% 0% -2% t t + 1 t + 2 t + 3 t + 4 4% t t + 1 t + 2 t + 3 t + 4 Notes: The results are based on the point estimates and on the standard deviation of the respective parameter estimates of model [2]. The solid line denotes the median of the price effects and the dotted lines denote the 2.5% and the 97.5% percentiles. the following periods, which again contributes to higher wine prices in the next years. Simulation experiments indicate that this (endogenous) reputation effect of quality is important; the long-run price effects of a one-time quality increase are about 40% larger when reputation effects are taken into account compared to a situation where reputation effects are ignored. This result also has managerial implications. Form the observation that reputation is more important than taste in determining market prices (Benfratello et al. (2009)), researchers suggest that producers should primarily focus at building an established reputation for their product by pursuing promotional activities. While we do not question the importance of promotional activities, our results suggest that it is the sustained sensory quality of a wine over time that leads to a wineries high reputation. Our analysis thus provides strong empirical support for theoretical models focusing on the interrelationship between quality and reputation ((Board and Meyer-Ter-Vehn, 2013)) in experience goods markets. We further provide a clear answer to the ongoing debate in the wine price and quality literature on the question: Does quality matter? Our answer is affirmative; quality matters and is particularly important if the long-run consequences of improvements in quality, i.e. the positive effects on reputation, are explicitly taken into account. References Ali, H. H., Lecocq, S. and Visser, M. (2008). The Impact of Gurus: Parker Grades and En Primeur wine Prices. The Economic Journal 118: 158 173. 14

Ali, H. H. and Nauges, C. (2007). The pricing of experience goods: The example of en primeur wine. American Journal of Agricultural Economics 89: 91 103. Allen, F. (1984). Reputation and product quality. The RAND Journal of Economics 15: 311 327. Benfratello, L., Piacenza, M. and Sacchetto, S. (2009). Taste or reputation: what drives market prices in the wine industry? Estimation of a hedonic model for Italian premium wines. Applied Economics 41: 2197 2209. Board, S. and Meyer-Ter-Vehn, M. (2013). Reputation for quality. Econometrica 81: 2381 2462. Cabral, L. and Hortacsu, A. (2010). The dynamics of seller reputation: Evidence from ebay. The Journal of Industrial Economics 58: 54 78. Combris, P., Lecocq, S. and Michael, V. (1997). Estimation of a hedonic price equation for bordeaux wine: does quality matter? Economic Journal 107: 390 402. Combris, P., Lecocq, S. and Michael, V. (2000). Estimation of a hedonic price equation for burgundy wine. Applied Economics 32: 961 967. Dewenter, R. and Heimeshoff, U. (2014). Media bias and advertising: Evidence from a german car magazine. Review of Economics 65: 77 94. Dewenter, R. and Heimeshoff, U. (2015). Do expert reviews really drive demand? evidence from a german car margazine. Applied Economics Letters 22: 1150 1153. Dubois, P. and Nauges, C. (2010). Identifying the effect of unobserved quality and expert reviews in the pricing of experience goods: empirical application on Bordeaux wine. International Journal of Industrial Organization 28: 205 212. Estrella Orego, M., Defrancesco, E. and Gennari, A. (2012). The wine hedonic price models in the old and new world : State of the art. Revista de la Facultad de Ciencias Agrarias, UNCuyo 44: 205 220. Friberg, R. and Grönqvist, E. (2012). Do expert reviews affect the demand for wine? American Economic Journal: Applied Economics 4: 193 211. Ginsburgh, V. (2003). Awards, success and aesthetic quality in the arts. The Journal of Economic Perspectives 17: 99 111. Golan, A. and Shalit, H. (1993). Wine quality differentials in hedonic grape pricing. Journal of Agricultural Economics 44: 311 321. Hilger, J., Rafert, G. and Villas-Boas, S. (2011). Expert opinion and the demand for experience goods: An experiemental approach in the retail wine market. The Review of Economics and Statistics 93: 1289 1296. 15

Hopfer, H., Nelson, J., Ebeler, S. E. and Heymann, H. (2015). Correlating wine quality indicators to chemical and sensory measurements. Molecules 20: 8453 8483. Hörner, J. (2002). Reputation and competition. The American Economic Review 92: 644 663. Klein, B. and Leffler, K. B. (1981). The role of market forces in assuring contractual performance. Journal of Political Economy 89: 615 641. Kreps, D., Milgrom, P., J., R. and Wilson, R. (1982). Rational cooperation in the finitely repeated prisoners dilemma. Journal of Economic Theory 27: 245 252. Landon, S. and Smith, C. E. (1997). The use of quality and reputation indicators by consumers: the case of Bordeaux wine. Journal of Consumer Policy 20: 289 323. Landon, S. and Smith, C. E. (1998). Quality expectations, reputation, and price. Southern Economic Journal 64: 628 647. Lecocq, S. and Visser, M. (2006). What determines wine prices: Objective vs. sensory characteristic. Journal of Wine Economics 1: 42 56. Nerlove, M. (1995). Hedonic price functions and the measurement of preferences: The case of swedish wine consumers. European Economic Review 39: 1697 1716. Oczkowski, E. (1994). A Hedonic Price Function for Australian Premium Table Wine. Australian Journal of Agricultural and Resource Economics 38: 93 110. Oczkowski, E. (2001). Hedonic wine price functions and measurement error. Economic Record 77: 374 382. Oczkowski, E. and Doucouliagos, H. (2015). Wine Prices and Quality Ratings: A Metaregression Analysis. American Journal of Agricultural Economics 97: 103 21. Pennerstorfer, D. and Weiss, C. (2013). Product quality in the agri-food chain: Do cooperatives offer high quality wine? European Review of Agricultural Economics 40: 143 162. Reinstein, D. A. and Snyder, C. M. (2005). The influence of expert reviews on consumer demand for experience goods: A case study of movie critics. The Journal of Industrial Economics 53: 27 51. Rob, R. and Fishman, A. (2005). Is Bigger Better? Customer Base Expansion through Word-of-Mouth Reputation. Journal of Political Economy 113: 1146 1162. Rosen, S. (1974). Hedonic prices and implicit markets: product differentiation in pure competition. The Journal of Political Economy 82: 34 55. 16

Schamel, G. and Anderson, K. (2003). Wine quality and varietal, regional and winery reputations: hedonic prices for Australia and New Zealand. Economic Record 79: 357 369. Shapiro, C. (1983). Premiums for high quality products as returns to reputations. The Quarterly Journal of Economics 98: 659 679. Sorensen, A. T. (2007). Bestseller lists and product variety. The Journal of Industrial Economics 55: 715 738. Wooldridge, J. (2002). Introductory econometrics: A modern approach. Mason, OH: South Western Educational Publishing. Wooldridge, J. M. (1995). Score diagnostics for linear models estimated by two stage least squares. In Maddala, G. S., Phillips, P. C. B. and Srinivasan, T. N. (eds), Advances in Econometrics and Quantitative Economics. Oxford: Blackwell, 66 87. Wooldridge, J. M. (2001). Econometric analysis of cross section and panel data. Cambridge, MA: MIT press. Appendix 17

Table 3: Descriptive Statistics of Variables used in the Empirical Analysis Variable # of Obs. Mean Std. Dev. Min Max Price 24,547 9.80 6.44 2.20 100.00 Quality 9,914 89.05 2.05 83 98 Reputation 22,210 0.29 0.31 0 1 Select 24,547 0.40 0 1 Average Quality (at t-1) 21,705 88.87 1.31 85.40 94.00 Average Quality (at t-2) 20,792 88.94 1.43 85.33 94.38 Average Quality (at t-3) 19,762 88.88 1.56 84.50 95.00 Average Quality (at t-4) 18,370 88.70 1.69 84.00 95.00 Time between Harvest and Bottling: 1 Year 24,547 0.75 0 1 Time between Harvest and Bottling: 2 Years 24,547 0.24 0 1 Time between Harvest and Bottling: 3 Years 24,547 0.01 0 1 Number of times Wine received Grade between 1999 and 2002 24,547 0.98 1.35 0 4 Number of different Wines of Winery (in Year t) 24,547 15.38 5.12 1 38 Harvest 2002 24,547 0.05 0 1 Harvest 2003 24,547 0.23 0 1 Harvest 2004 24,547 0.25 0 1 Harvest 2005 24,547 0.26 0 1 Harvest 2006 24,547 0.20 Type of Wine: White wine 24,547 0.55 0 1 Red wine 24,547 0.33 0 1 Sweet wine ( Süßwein ) 24,547 0.08 0 1 Rosé wine 24,547 0.03 0 1 Type of Sweet Wine: Spaetlese 24,547 0.02 0 1 Beerenauslese 24,547 0.02 0 1 Trockenbeerenauslese 24,547 0.03 0 1 Eiswein 24,547 0.02 0 1 Variety of Grape: Blauburger 24,547 0.01 0 1 Blaufränkisch 24,547 0.05 0 1 Blauer Portugieser 24,547 0.01 0 1 Blauer Wildbacher 24,547 0.00 0 1 Chardonnay 24,547 0.08 0 1 Cabernet Sauvignon 24,547 0.02 0 1 Cuvee Rot 24,547 0.09 0 1 Cuvee Weiss 24,547 0.04 0 1 Frühroter Veltliner 24,547 0.00 0 1 Gemischter Satz 24,547 0.01 0 1 Gelber Muskateller 24,547 0.02 0 1 Grüner Veltliner 24,547 0.15 0 1 Merlot 24,547 0.01 0 1 Muskat Ottonel 24,547 0.01 0 1 Müller Thurgau 24,547 0.01 0 1 Neuburger 24,547 0.01 0 1 Pinot Gris / Grauburgunder 24,547 0.01 0 1 Pinot Noir / Blauburgunder 24,547 0.03 0 1 Rotgipfler 24,547 0.01 0 1 Riesling 24,547 0.09 0 1 Rose 24,547 0.02 0 1 Roter Veltliner 24,547 0.01 0 1 Sämling 88 / Scheurebe 24,547 0.01 0 1 Sauvignon Blanc 24,547 0.05 0 1 Schilcher 24,547 0.01 0 1 Sankt Laurent 24,547 0.02 0 1 Sortenvielfalt Weiss 24,547 0.01 0 1 Syrah 24,547 0.01 0 1 Traminer 24,547 0.02 0 1 Weissburgunder / Pinot Blanc 24,547 0.05 0 1 Welschriesling 24,547 0.04 0 1 Zierfandler 24,547 0.01 0 1 Zweigelt 24,547 0.10 0 1 Note: 0.00 denotes that the value is rounded and not exactly zero. 18