Modelling prices and the reputation of individual named wines

ABSTRACT

This article develops empirical models to assess the relation between the reputation of an individual named wine and its price. Unrestricted and polynomial distributed lag models are used to assess the impact of past expert quality ratings on the prices of Australian premium wines. Results point to the practical unimportance of current wine quality scores impacting prices and suggest that quality score lag effects up to six years may be important. The largest individual lagged impact of quality on price is estimated to occur at approximately two years, and prices are estimated to increase by more than 10% over six years for a one-point quality score increase. A procedure for identifying potential wine price bargains based on a comparison of price predictions from estimated wine reputation and current quality measures is illustrated. The implications of the findings for wine producers are also discussed.

KEYWORDS:

• Wine prices
• individual wine reputation
• wine quality

JEL CLASSIFICATION:

• C21
• Q11

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ To facilitate accurate analysis, collated data should cover at least two years of current releases to ensure that results are not unduly influenced by a current release’s tasting or price effects of a single year. It appears that the use of more previous editions for new releases from the Halliday guide is unnecessary for analysis. We have sufficient observations to undertake a thorough and statistically meaningful analysis with two recent wine guide releases. In part, information is simply duplicated further if more previous guides are employed.

² Given the pooled analysis of 2015 and 2016 prices, some consideration was given to using real prices for analysis. Using real prices made no substantial differences to results. In general, wine prices have been relatively stable (changes of 1% p.a.) in the Australian context, see Oczkowski (2016a).

³ In choosing the optimal p and q values for the polynomial lag model, we employ the general model selection statistics, such as AIC and BIC, as recommended by Judge et al. (1985) and Gujarati and Porter (2009). Relatedly, Giles and Smith (1977) demonstrate how the minimum error variance rule for choosing the best model also applies to the polynomial distributed lag model.

⁴ The summation of beta estimates for shorter unrestricted lag length models is: 0.0444 (q = 0), 0.0674 (q = 1), 0.0818 (q = 2), 0.0901 (q = 3) and 0.0958 (q = 4). This pattern illustrates how at q = 5 and q = 6 the summation of marginal price impacts reasonably stabilizes implying that longer lags may be unnecessary for capturing any additional important price effects.

⁵ The price predictions are based on the anti-logs of the ln(Price) and employ a small sample correction bias of exp$\left({\hat{\sigma }}^2/2\right)$.

⁶ Only very minor differences to the predictions for the q = 6 p = 4 emerge. The average reputation measure is 92.40 and 64% of wines have quality exceeding reputation scores. The mean mispricing values are: 5.87 (positive values) and −3.96 (negative values) for the reputation model and 2.355 (positive values) and −1.81 (negative values) for the current quality model. The correlation between the q = 6 p = 3 and q = 6 p = 4 model reputation scores is 0.999 and the correlations between the corresponding predicted prices are 0.999 or higher.