To develop fine wine market tools we have to decide what wine data to look at. The decisions we make on data will trickle down and affect our tool development.
We begin by manually defining four groups of fine Bordeaux wines based on qualitative factors important to wine investors. Our reason for coming up with an initial clustering is to ensure that we include a good range of investment wines in our analysis.
The table below shows: the wines we pre-selected, the manually assigned group number, the first available time stamp for the first available vintage on Liv-ex, the years for which ex-Château prices are missing and then finally the decision of whether to include the pre-selected wine in this analysis. All price data is courtesy of Liv-ex.
To track the development of wine prices over time, we are interested in release prices and monthly market price data for vintages going back as far as possible. Most of the supplied data had a first time stamp of 31/07/1983, but only a few wines (e.g. Lafite Rothschild, Mouton Rothschild, Margaux, Léoville-Las Cases) actually had prices for the 1982 vintage going back to this date.
The ex-Château price was taken as the “release” price in favour of the London release price for two reasons. 1.) The ex-Château price is the first available indicator for the price-level of a wine, any later prices necessarily reference it. 2.) More pragmatically, more ex-Château prices than London release prices were available to analyse. Even then, many wines were missing ex-Château prices for more than three of the years and were therefore excluded in this analysis.
We can only speculate on reasons for the missing data (reader comments welcome). La Mission Haut-Brion was acquired by the Dillon family in 1983, could that be related to the missing 1982 value? Pontet Canet started adapting to biodynamic practices in 2004 and 2007 was the last time chemicals were used in their vineyards. Biodynamic certification was obtained in 2010. Perhaps the gap in ex-Château prices from 2007 to 2010 is related to Pontet Canet working towards certification during this time?
Clusters based on price dynamics
We now continue our journey with eighteen wines across five manually defined groups, which reflect a good range of investment grades. Our next step is to choose representative individual wines from this list. To do this, we extract key summary statistics for each wine across its price history and use these as features for a hierarchical clustering.
The Cluster Dendrogram illustrates that good separation is achieved by either three or four clusters. (Height in the dendrogram can be interpreted as a measure of separation. The aim is to pick clusters, indicated by red boxes below, that are comparatively stable to vertical shifts.)
We choose to separate the wines into four rather than three clusters. This highlights the special nature of Lafite-Rothschild as the “blue-chip” investment wine par excellence. The translation from wine-number to wine-name is given in the following table.
Time for some more choices: we pick from each of the four clusters, choosing to proceed with Margaux, Clerc-Milon, La Conseillante and Lafite Rothschild as representative wines. We do not claim that these wines are representative of Bordeaux wine categories in general, but simply that they share common features in their price dynamics.
In the next post, we start developing our first tool: the Channel Spread.
MK @ WineQuant
Credits: Data from Liv-ex.
Value & Risk in Fine Wine by WineQuant is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Based on a work at https://winequantblog.wordpress.com/.