The Ashenfelter Wine Quality Index [1950-2020]

The Ashenfelter wine quality index created waves of controversy in the wine industry when it was released. In the June 2008 edition of The Economic Journal, Princeton economist Orley Ashenfelter published “Predicting the Quality and Prices of Bordeaux Wine.” You can read the article here. The premise of the article is that weather conditions during the Bordeaux wine season play a substantial role in establishing each vintage’s quality, as indicated by its market price.

This information had been in the market for some time, however. For example, a 1990 New York Times article discusses his finding. His work eventually gained Dr. Ashenfelter visibility because of the conflict it seemingly created with wine experts such as Robert Parker. Mr. Parker is quoted in the NYT article calling Dr. Ashenfelter’s premise “ludicrous and absurd.”

Predicting the Quality and Prices of Bordeaux Wine

Growing grapes to produce wine is an agricultural activity. This means the weather conditions during the growing season impacts the final quality of the product once it’s in the bottle. Ashenfelter’s paper is a regression-based analysis of relating the price of premier Bordeaux wines to the weather conditions (i.e., the Bordeaux wine climate) during a specific vintage.

The 1990 and 1991 prices of London auction wine price data for wines produced in the 1950s to 1980s were secured along with weather data from the Bordeaux-Merignac airport. Regression analysis demonstrates the importance of weather conditions to the quality of the Bordeaux wine, which is assumed to be reflected in its price. The idea that emerges is that the quality of a Bordeaux vintage can be predicted using weather data alone.

The Original Work

The regression analysis in the article demonstrates the price of the premier Bordeaux wines can be modeled using the following equation:
Quality of Wine =
0.240 * Age of Vintage +
0.608 * Average Temperatures during the April to September Growing Season +
-0.00380 * Total August Precipitation +
0.00115 * Total Precipitation during the preceding winter (October to March) +
0.00765 * September Average Temperatures

A key aspect of this equation is the first term. It says the quality of wine increases with age. This comes as no surprise to anyone familiar with evaluating and drinking Bordeaux’s premier wines.

It turns out the data used in the study has been published online and used in a statistics tutorial. Since it’s readily available I’ve accessed it and used it in the remainder of this post.

Bordeaux Wine Price vs. Age
Bordeaux Wine Price vs. Age. This figure recreates Ashenfelter’s Figure 1.

The red line shows the result of linear regression on only the age of the vintage versus it’s age.

The paper only lightly documents the methodology of the analysis. For example, the source of the weather station data is not provided. The data for the Bordeaux-Merignac airport is certainly available if one digs deep enough and is willing to pay for it.

Bordeaux Weather: A Different Source

In this post, I seek to recreate the results of the regression analysis but also reformulate the wine quality index based on what I perceive as a few shortcomings of the methodology. For Bordeaux weather data, I reanalysis data rather than the observed weather station data. Reanalyses, such as the ECMWF ERA5, is a numerical recreation of observed weather data using modern weather forecast modeling technology. The model is run for earlier dates and the solution is forced by the known weather station, radiosonde, satellite, and other sources of observed weather conditions. The ERA5 data is available from 1950 to today.

How valid is it to use the ERA5 data? To answer that, I recreate Figure 2 from the Ashenfelter paper. It shows the relationship between “Summer Temperature” and “Harvest Rain.” Given the regression coefficient descriptions in the paper, I assume this means the April to September temperatures and the September total precipitation.

The graphic below recreates Ashenfelter’s Figure 2. In this case, the temperature and precipitation data have been “centered,” meaning the time-series average of both variables has been removed. Each data point style displays if the original price of the wine that year was above or below normal. We see (note the Y-axis is reversed) that relatively warm summers and lower than normal harvest precipitation generates higher than normal prices.

Ashenfelter Wine Quality Figure 2 Recreation
A recreation of Ashenfelter’s Figure 2 but with ERA5 data.

Using the ERA5 data we can apply Ashenfelter’s equation to look at the quality of more recent Bordeaux vintages. In this case, the first component of the equation relating the age to price is not used. In other words, I’m applying the second through the fifth regression coefficients to a history of ERA5 data to estimate the quality of Bordeaux red wine in each year up to 2020.

Wine Quality Index: Ashenfelter Version
Using Ashenfelter’s regression to create a Bordeaux wine index.

The index time series shows that higher-quality wines have happened more frequently in the past 20 years than in the prior 40 years. Recall the largest coefficient in the regression equation was applied to the growing season temperature term. The index, therefore, is particularly sensitive to changes in temperature.

The recreation of Ashenefelter’s Figure 2 shown above uses data only from 1952 to 1980. We can add the years after 1980 to the graphic.

All years: Ashenfelter Figure 2
Ashenfelter Figure 2 but including all years up to 2020.

In this case, each data point is labeled by the last two digits of the year and color-coded to represent the years greater than 1980 in blue. The majority of blue dots are to the right of the red dots meaning the growing season temperatures were higher than the group of years used to create the original graphic. In other words, the wine-growing season average temperatures have increased over the past 70 years. And, because the regression equation is dominated by the growing season temperature term and the summers are warming the quality index gradually increases.

It might be necessary to reformulate Ashenfelter’s regression.

Revisiting Ashenfelter’s Wine Quality Index

Unfortunately, the paper does not provide much detail regarding analytical methodology. But we can infer quite a bit of information from the graphs and the regression results shown. For example, it appears the data was not normalized before being used in the regression analysis. Normalization of the data removes the average value of each variable’s time series and divides it by its standard deviation. The result is a data set in which the average of each time series is zero and the values represent the deviation in standard deviations.

The regression equation as presented in the paper makes it difficult to compare the relative importance of one term relative to another. In other words, is the summer average temperatures more or less important than the harvest total precipitation? We don’t know because the data is normalized.

Normalizing the data and calculating the regression again allows us to compare the relative importance of the different factors. We’ll recalculate the regression results but we want to isolate the impact of weather on wine quality. To achieve that goal we’ll first correct the original wine pricing data for its time dependency.

Correcting the Price of Wine for Time

The first graph of this post shows the relationship between the age of the vintage and the price of the wine. Ashenfelter regression documents that the slope of the red line is that the price changes by 0.0354 GBP for each year of vintage age. We can use this information to control for the impact of age on the price of wine.

Corrected Bordeaux Wine Price for Age
Corrected Bordeaux Wine Price for Age

Correcting the wine price data for age removes the age dependency. What remains in the price data is theoretically the impact of weather conditions on the quality of the red wine.

Updated Regression

The regression results below are calculated using the time-correct wine auction pricing data but with the ERA5 weather data for Bordeaux.

Weather vs Bordeaux Wine Price Regression
Weather vs Bordeaux Wine Price Regression Results

Because the data was normalized before being used in the model, we can compare the relative importance of the regression coefficients to predict the quality of Bordeaux white wine. The coefficients have been rounded to facilitate the presentation. The most important factor determining the price of Bordeaux wine is growing season temperature, followed by winter precipitation, then September temperatures, followed by August precipitation.

The results are interesting. Above normal precipitation during the winter and in August are negative factors for price. Growing season and September temperatures are positive factors for price. The negative coefficient on winter precipitation is opposite to the finding of Ashenfelter in the original paper.

The original Ashenfelter paper suggests a slightly positive impact of increased wintertime precipitation. This result suggests winter precipitation does not help to increase wine prices.

Bordeaux Wine Quality: A New Formula

Using the wine price regression results with the ERA5 weather data for Bordeaux yields a new formula to estimate Bordeaux wine quality. That formula is:

Wine Quality ~
0.15 * Winter Precipitation +
0.45 * Growing Season Temperature +
0.03 * August Precipitation +
0.17 * September Temperatures

Bordeaux Wine Quality Index: 1950 – 2020

Using this new formula we create a new wine quality index that is independent of the age of the vintage and includes the relative importance of the weather factors. It is shown below. Note the new formula suggests the August precipitation has almost no importance relative to the other factors.

New Weather Wine Quality Index for Bordeaux Red Wine
New Weather Wine Quality Index for Bordeaux Red Wine

The new index still suggests the more recent vintages are of higher quality than prior vintages. This is certainly because the average growing season temperatures have increased since the 1950s and 1960s. Comparing this graph to the index based on Ashenfelter’s regression results, however, we see less impact from temperatures in recent years because the index values are not quite as extreme.

The new Bordeaux red wine quality index suggests 2003 is the best year on record and that 2020 and 2018 are not far behind. Does that hold true? The Wine Cellar Insider describes the 2003 vintage as a “scorched, hot, dry growing season.” Bordeaux experienced extreme temperatures in 2003 and the revised index is certainly picking up that signal in the growing season average temperature.

The Buyer, however, quotes Guy Seddon as saying “Bordeaux 2020 is good. It’s really good. But coming hot on the heels of two strong vintages, we’re going to need some convincing.” So, at least there is an indication from Guy Seddon the index is working and accurately assess wine quality in recent years.

The index also ranks 1990 as a decent year. The Wine Spectator gives the 1990 Chateau Latour (which I’ve had the pleasure of enjoying) 100 of 100 points and listed it as the #1 wine released in 1993. Further evidence that while the index contains useful information temperature trends may make the index less useful.

Temperatures in the Bordeaux region are going to increase for the rest of our lifetimes. At some point, the increasing temperatures will become damaging to Bordeaux’s red wine quality. That factor is not represented in the regression equation.

Finally, it might be helpful to rank the years of the index.

The Bordeaux Wine Quality Index: Sorted by Value

Top 10 Bordeaux Vintages

The top 10 Bordeaux vintages as predicted by the wine weather quality index is shown below.

Top 10 Bordeaux Vintages: Wine Quality Index
Top 10 Bordeaux Vintages based on the Wine Quality Index

The Ashenfelter wine quality index creates waves of controversy when it was first introduced. This post reevaluated the index while correcting for a few issues in an effort to improve it. Once reconstituted, we can calculate the top 10 Bordeaux vintages.