1. Introduction
Lancaster (Reference Lancaster1966) posits that the utility derived from the consumption of a product is generated by its technical and quality characteristics. Rosen (Reference Rosen1974) further developed this idea in a theoretical model of hedonic price functions derived from the interaction of profit-maximizing firms and utility-maximizing consumers. In this framework, equilibrium prices are functions of the implicit prices of the characteristics that differentiate a range of products, such as beer, spirits, and wine. Furthermore, Rosen implicitly assumed a linear relationship between product attributes and product prices. In consequence, hedonic models that seek to explain the observable variation in wine prices (e.g., Goncalves et al., Reference Goncalves, Rebelo, Lourenco-Gomes and Caldas2021; Rebelo et al., Reference Rebelo, Lourenco-Gomes, Goncalves and Caldas2019; Roma et al., Reference Roma, Di Martino and Perrone2013) and beer prices (e.g., Michis, Reference Michis2024; Smith et al., Reference Smith, McKinney, Caudill and Mixon2016; Thrane et al., Reference Thrane, Lien, Mehmetoglu and Stordal2023) typically include alcohol by volume (ABV) (in contrast to sensory characteristics and reputation, which are subjective attributes) as a control variable. However, little attention is paid to the magnitude of the coefficient and the assumption that the effect is indeed linear. This, in turn, suggests that the notion of “more is always better” may be applicable to the pricing of wine and beer with higher alcohol content.
In this paper, we utilize a novel data set with approximately 900 distinct spirits, ranging from absinthe to vodka, to challenge this assumption. The results demonstrate that inclusion of the squared and cubic terms of ABV in a standard hedonic price model produces some unexpected outcomes. Furthermore, our findings indicate that expert ratings are significantly influenced by the alcohol content of a specific “hard liquor,” with a notable nonlinear effect.
2. Literature survey
Until recently, there has been a paucity of published research employing hedonic models to explain the observable variation in beer prices. The seminal study is that of Smith et al. (Reference Smith, McKinney, Caudill and Mixon2016), in which the authors utilize a data set comprising 400 different beers from 20 countries.Footnote 1 The mean price of a six-pack of beer ranges from $4 to $42, with a standard deviation of $3.3. The ABV ranges from 2.7% to 15%, with a mean of 6.1%, while the consumer rating varies between 50 and 91 points, with a mean of 78 (on a scale with a minimum value of 0 and a maximum of 100). After controlling for bottle size and beer style, the authors' primary finding was that both consumer ratings and ABV had a statistically significant and positive impact on bottle prices. Controlling for other potential confounders, a 1% increase in alcohol content is associated with a 34% increase in bottle prices.Footnote 2
The data set utilized by Michis (Reference Michis2024) includes 675 different beers, with price per unit ranging from 2.60£ to 26.60£, exhibiting a mean of 5.95£ and a standard deviation of 2.90£. The ABV ranges from 2.5% to 17.5% (mean 6.4, standard deviation 2.5), while the average customer evaluation ranges from virtually 0 to 100 (with a mean of 71, standard deviation 30).Footnote 3 As demonstrated in the paper, there is a statistically significant correlation between ABV and both bottle prices and consumer ratings. A 1% increase in alcohol is associated with a 0.7% decrease in consumer ratings, while a 1% increase in consumer ratings is associated with a 0.19% increase in bottle prices.
Thrane (Thrane et al., Reference Thrane, Lien, Mehmetoglu and Stordal2023) employs two different subsamples. The first includes 9,251 beers for which data were available on all independent variables, except customer ratings. The second consists of 3,766 observations for which Parker scale ratings were also available.Footnote 4 The mean price per bottle is 6.90€, with a range of 1.80€ to 51.90€ and a standard deviation of 3.80€. The ABV of the beers in the sample ranged from 4.7% to 20.0%, with an average of 7.6% and a standard deviation of 2.4%. After controlling for country and beer style effects, a 1 percentage point increase in alcohol content was found to be associated with a 9% increase in bottle price. When expert ratings are also taken into account, the effect is slightly reduced to 8%. Additionally, the quality rating exerts a modest, yet positive and statistically significant influence on the price of beer. A 10-point increase in the rating is associated with a 7% increase in price.
Following the seminal study by Ginsburg et al. (Reference Ginsburgh, Monzak and Monzak2013), which originated as a conference presentation and was eventually published in 2013, a number of subsequent publications have emerged (early examples are Golan and Shalit, Reference Golan and Shalit1993 as well as Oczkowski, Reference Oczkowski1994) aiming to identify the factors influencing wine prices (for a comprehensive overview of these studies see Outreville and Le Fur, Reference Outreville and Le Fur2020). It is noteworthy that only a limited number of these studies incorporate ABV as a potential determinant.Footnote 5
Roma et al. (Reference Roma, Di Martino and Perrone2013) employ two different data sets, derived from the 2010 editions of two prominent Italian wine guides, Vini d'Italia and Duemila Vini, to ascertain the primary determinants of Sicilian wines. The former sample comprises 558 wines from 94 wineries, while the latter sample includes 402 wines from 59 wineries. The ABV ranges between 11% and 16.5%, with a mean of 13.5 and a standard deviation of 0.8% in the former sample. The values for the second sample are 11 (minimum), 15.5 (maximum), 13.4 (mean), and 0.7 (standard deviation), respectively. The price of a bottle of wine ranges between 4€ and 80€ in the first and between 4€ and 60€ in the second sample. A 1% increase in the ABV of a wine is associated with a 6.5% higher price per bottle in the first sample and a 10.2% higher price per bottle in the second sample. This suggests that consumers of Sicilian wines prefer wines with a higher ABV.
Rebelo et al. (Reference Rebelo, Lourenco-Gomes, Goncalves and Caldas2019) employ two different data sets, one from a specialist retailer (Garrafeira Nacional) comprising detailed information on 1,722 wines and another one from a supermarket chain (Continente) including 725 wines. The wines available from the specialist retailer are priced, on average, at 26.90€ per bottle, with a standard deviation of 53.50€, a minimum price of 1.40€ and a maximum price of 589€. The alcohol content of the wines in question ranges from 9% to 17%, with a mean of 13.1 and a standard deviation of 1.0. The wines available at the supermarket have an average price of 9.60€ per bottle, with a range between 1.40€ and 92.50€ (standard deviation 11.00€). The alcohol content of the wines in question varies between 9% and 15.5%, with an average of 13.1% and a standard deviation of 1.2%. A 1% increase in ABV is associated with a 30% higher price for wines available at the retailer and a 40% higher price for wines sold in the supermarket chain. These effects are strikingly consistent across the price distribution quantiles.
Goncalves et al. (Reference Goncalves, Rebelo, Lourenco-Gomes and Caldas2021) use four data sets, comprising a total of 9,624 bottled wines available at specialized online retailers in four countries: France (Vinatis, 2,094 observations), Italy (XtraWine, 2,803 observations), Australia (Vintage Cellars, 2,063 observations), and Germany (Vinexus, 2,664 observations). The mean price per bottle is 22€ in Germany, 32€ in Italy, 35€ in Australia, and 44€ in France. The coefficient of variation is 96%, 120%, 189%, and 148%, respectively. The average alcohol content of the wines is quite similar, ranging from 13.4% to 13.6% with a minimum of 9 and a maximum of 17%. In the French and Italian samples, the impact of alcohol content on bottle prices increases across the quantiles of the price distribution. In contrast, in the German sample, this impact remains virtually constant, while in the Australian sample, it decreases. In the standard Ordinary Least Squares (OLS)model, a 1% increase in alcohol content is associated with a 2.7% higher bottle price in Australia, an 8.1% higher bottle price in Germany and France, and a 10.8% higher bottle price in Italy.
To the best of our knowledge, Fanasch and Frick (Reference Fanasch and Frick2020) were the first to include not only the linear term but also the squared term of ABV in their estimation of a hedonic price model using a large sample of 55,500 different wines from Germany. The results demonstrate that the implicit assumption that ABV has a positive and statistically significant linear effect on bottle prices is not supported (Figure U2, not included in the original publication). In particular, wines with an alcohol content of less than 9.5% sell at a higher price than those with a content of 9–10%. Subsequently, prices increase exponentially. Upon re-estimation of the data set, including a cubic term, it was found that the latter coefficient failed to reach a conventional level of significance. This suggests that the documented effect in Figure U2 is not only statistically significant but also economically highly relevant. In that data set, the range of ABV is from 5.5% to 20%, with a mean of 12.1 and a standard deviation of 1.8. The former is lower than in the studies summarized above, while the latter is higher, indicating that the wines in the aforementioned data set are more heterogeneous (see Figure U1 in the Appendix).
3. Data and findings
In this paper, we extend the analysis of wine and beer price determinants by using data from the 2023 edition of the “Austrian Spirits Trophy.” The results of this study were recently published in the gourmet magazine Falstaff. The data set comprises 906 different spirits from across the globe that received an award, spanning a range of categories from absinthe to vodka.Footnote 6 For 872 of the spirits in question, the retail price is available in intervals ranging from 1€ to 10€, from 11€ to 20€, from 21€ to 50€, from 51€ to 100€, and above 100€. The lowest alcohol content in the sample is observed in an Italian aperitivo, with a value of 11%, while the highest alcohol content, at 68%, is observed in an absinthe from Austria. The mean alcohol content is 37.8, with a standard deviation of 9.7 (Figure 1).
Figure 1. Kernel density estimation of alcohol by volume in spirits data.
Our models are of the following general form:
(1) BP/JG = α0 + α1 ABV + α2 ∑ STP + α3 ∑ CD + α4 BS + ε
(2) BP/JG = α0 + α1 ABV + α2 ABV2 + α3 ∑ STP + α4 ∑ CD + α5 BS + ε
(3) BP/JG = α0 + α1 ABV + α2 ABV2 + α3 ABV3 + α4 ∑ STP + α5 ∑ CD + α6 BS + ε
whereBP: bottle price (in categorized form)
JG: jury grade (>80 points on a 100 points scale)
ABV: alcohol by volume (in %)
STP: spirit type dummies (n = 36; Absinthe … Vodka)
CD: country dummies (n = 39; Australia … Venezuela)
BS: bottle size (in Liters)
As expected, the linear term for ABV has a statistically significant and positive effect on bottle prices (Model 1), as illustrated in Table 1. In the second model, the linear term is negative and statistically significant, while the squared term is positive and significant, indicating a U-shaped pattern (Figure 2). Finally, in the third model, the coefficients of the linear and cubic terms are negative and significant, while the coefficient of the squared term remains negative and retains its statistical significance (Figure 3). Model 4 also incorporates an assessment of the spirits by an expert jury, with only those spirits that receive a score of 80 points or above being eligible for an award. The inclusion of this variable results in a slight reduction in the magnitude of the three alcohol coefficients, yet their statistical significance remains unaltered.
Figure 2. Predictive margins from the OLS model of bottle price with linear and squared term.
Figure 3. Predictive margins from the OLS model of bottle price with linear, squared, and cubic term.
Table 1. OLS estimation results
Standard errors in parentheses.
* p < 0.10, **p < 0.05, ***p < 0.01.
These findings indicate that the implicit assumption underlying the majority of the available evidence that ABV has a strictly positive impact on bottle prices, particularly of beer and wine, may be inappropriate. Furthermore, this assumption is not aligned with the concept of decreasing marginal returns, which is a fundamental tenet in most economic analyses.
It is noteworthy that other studies have included additional terms in their models, such as the squared and cubic terms of bottle age in the analysis of auction prices of fine wines (Dimson et al., Reference Dimson, Rousseau and Spaenjers2015) and the squared and cubic terms of winery reputation in the estimation of bottle prices to account for superstar effects (Castriota et al., Reference Castriota, Corsi, Frumento and Ruggeri2022).
In the second set of models, the dependent variable is the evaluation provided by experts. After controlling for country of origin, spirit type, and bottle size, it was found that ABV has a negative and statistically significant association, indicating that experts tend to dislike particularly strong spirits (Model 1 in Table 2). Including the square of ABV results in an increase in the magnitude of the coefficient of the linear term and a reduction in its statistical significance. Nevertheless, the squared term does not reach a conventional level of statistical significance (Model 2). Ultimately, Model 3 demonstrates that the inclusion of the cubic term has rendered the squared term statistically significant, thereby yielding the predictive margins displayed in Figure 4.
Figure 4. Predictive margins from the OLS model of jury grade with linear, squared, and cubic term.
Table 2. Impact of alcohol by volume on jury grade
Standard errors in parentheses.
* p < 0.10, **p < 0.05, ***p < 0.01.
4. Summary and implications
The implicit assumption underlying the majority of hedonic prize models that the influence of different product characteristics is strictly linear has yet to be subject to empirical testing. In this study, we utilize a novel data set comprising nearly 900 spirits from the 2023 Austrian Spirits Trophy to challenge the common assumption in previous research that ABV has a strictly linear positive effect on prices.
The results of our analysis indicate that the relationship between ABV and bottle prices is nonlinear, with notable effects observed for linear, squared, and cubic terms of alcohol content. This indicates that the pricing dynamic may be more complex than previously assumed. Furthermore, expert evaluations of spirits also demonstrate a nonlinear relationship with ABV, indicating that higher alcohol content does not always correlate with higher quality ratings. The findings indicate that spirits with excessively high alcohol levels are associated with lower expert evaluations and potentially reduced sales. This implies that producers should exercise caution in selecting the alcohol content of their spirits, as excessive alcohol levels are associated with lower expert evaluations.
These results contribute to and enhance the existing literature on hedonic pricing models for alcoholic beverages. While previous studies on wine and beer (e.g., Goncalves et al., Reference Goncalves, Rebelo, Lourenco-Gomes and Caldas2021; Michis, Reference Michis2024; Rebelo et al., Reference Rebelo, Lourenco-Gomes, Goncalves and Caldas2019; Roma et al., Reference Roma, Di Martino and Perrone2013; Smith et al., Reference Smith, McKinney, Caudill and Mixon2016; Thrane et al., Reference Thrane, Lien, Mehmetoglu and Stordal2023) have generally found positive linear relationships between ABV and price, our analysis reveals a more nuanced picture for spirits. Our work builds upon and expands the approach of Fanasch and Frick (Reference Fanasch and Frick2020), who were among the first to include nonlinear terms for ABV in wine pricing models. By applying this methodology to spirits and incorporating expert ratings, we provide a more comprehensive understanding of the influence of alcohol content on pricing and perceived quality across different types of alcoholic beverages.
Further research could extend this approach to other categories of alcoholic beverages or investigate additional nonlinear effects of other product characteristics. Moreover, cross-cultural studies could ascertain whether the observed relationships persist across different markets with presumably different consumers. In conclusion, this study underscores the significance of challenging implicit assumptions in hedonic pricing models and illustrates the value of considering nonlinear effects in analyzing the determinants of alcoholic beverage prices and quality perceptions.
Acknowledgments
We thank two anonymous referees and the editor for their helpful comments and suggestions. Any remaining errors and omissions are, of course, our own.
Appendix
Table A1 in the Appendix displays the results of ordered probit estimations showing that our findings are robust to different estimation techniques.
Table A1. Ordered probit estimation results.
Standard errors in parentheses.
* p < 0.10, **p < 0.05, ***p < 0.01.
Figure A1. Kernel density estimation of alcohol by volume in wine data (not included in the initial publication by Fanasch and Frick, Reference Fanasch and Frick2020).
Figure A2. Predictive margins from random effects model of alcohol by volume with linear and squared term (not included in the initial publication by Fanasch and Frick, Reference Fanasch and Frick2020).
Open access
