Up to a point, Professor Kantor


Brian Kantor is chief economist and investment strategist at Investec and something of a closet wine lover if his paper on the value of wine on the Investec website is any indication. Alas, the Professor is no statistician as he makes the most elementary mistake: choosing the wrong model.

bk 615x412 Up to a point, Professor Kantor

When it comes to tasting wine, linearity is a character as much appreciated as it is misunderstood. When it comes to statistics, a linear regression between two variables, let’s say the price of a wine and its score out of 100 in a blind tasting, is the simplest way of predicting one from the other. Except that while price in general is a linear scale – a R100 widget is twice as expensive as a R50 widget, wine scores are far from linear. Prices asked too are typically non-linear as a visit to the Rust en Vrede tasting room will confirm where flagships are an order of magnitude more expensive than entry levels.

Two wines scored 80 and 81 are essentially the same to a judge whereas an 89 point wine and a 90 pointer are quite different. As you approach a perfect score of 100, it’s like traveling at the speed of light. It takes you longer and longer to get there. So the professor’s linear models simply do not apply and his conclusions do not follow.

“A regression equation crafted from the scores realised an R square (or goodness of fit) of 49%. That is to say, the value equation explained only 50% of the observed price.

But it also proves that there is much more than quality and vintage at work on price. These regression results still leave 50% of the price of a representative bottle of South African or Australian wine sold in the US to be explained by other forces: it becomes very difficult to assert a bias in wine prices when more than 50% of the value of a bottle of wine cannot be attributed to its quality and/ or vintage.”

Sure there is “much more than quality and vintage at work on price” but you can’t “prove” that by fitting an inappropriate model.

Linear regressions apply when the data being modelled follow a Gaussian distribution. Wine scores and prices certainly do not. An observation the professor admits when he comments “while the scores range between a practical minimum of 80 and a maximum possible 100, there is no limit to the prices that may be charged from the $10 per bottle minimum that is barely enough to cover packaging and distribution costs. This improves the chances of significant outliers: wines that sell for much more than other wines with similar scores.”

This is the classic Black Swan situation described by Nassim Nicholas Taleb and the professor makes the most elementary mistake of using a linear predictor for non-linear data.

Something we do not do at our weekly blind tastings sponsored by RECM. We simply assume that in an ideal world, the best scoring wine/coffee/olive oil should be the most expensive and so by sorting prices by score we arrive at a value. Wines/coffees/olive oils to invest in are then simply those with best value compared to price. It’s a nonparametric model and scores and prices may be as non-linear as you please.

Linearity certainly does have its place in wine.  From Nederburg cellarmaster Razvan Macici musing on whether “fruit purity and linearity” is the way to express terroir to Winemail describing one of their mail-order offerings as possessing “good concentration of fruit on the palate, elegance and linearity”, ascribing geometric properties to wine, is all the rage and yet another descriptor has been added to winespeak.

Perhaps the first local sighting was at the Old Mutual Trophy Wine Show when Wine magazine interviewed show owner Michael Fridjhon who noted that to “outline the aesthetics that he as chairman wants to see applied during the judging of the show, he came up with the concept of ‘linearity’ or fruit purity.” Like Brian, Michael seems to be a professor at UCT, although perhaps not as well qualified. At least in the formal sense.

In a report on the show, taster Christian Eedes adopted this definition of “linearity” as “fruit purity” and hailed it as an important concept.  “Good Sauvignon Blanc hinges on linearity” was one message from the show but confusingly “it is possible to make a commercially successful Chardonnay where ‘linearity’ plays a very small part in the wine’s appeal.”  For greatness, straightness seems to be the sine qua non: “for Chardonnay or Sauvignon Blanc or any other variety” in order to have an expression of place, which is the holy grail of terroiriste wine making, “it all comes down to linearity.”  As Huey Lewis and the News might have put it: “it’s hip to be linear.”

As an abstract concept, “linearity” is an unfortunate term to introduce into the winespeak lexicon as in mathematics and engineering, linearity is almost an insult.  The interesting problems are all non-linear and to a mathematical winemaker the epithet of linearity would be synonymous with simple.  A far more appealing descriptor would be comparison to a fractal curve that has elegance, symmetry and as much complexity as you like at any scale of observation, the antithesis of a straight line which is after all, simply the shortest distance between two points.

When Steven Spurrier was in the winelands selecting wine for SA Airways, his interpretation of vinous linearity was a little different.  “It’s all Michel Bettane’s fault” he laughed.  Steven traced the concept of linearity back to Michel’s idea of “longueiline” a made-up word in French which Steven translates as “longer than you can taste” i.e. very persistent.  As chairman of the Decanter World Wine Awards, a sort of global Old Mutual, his geometrical insights carry a fair amount of Euclidean weight.

Michel even judged at Old Mutual when I was a panel chair, which puts him on the scene of the crime, and “linearity” is also a philosophical good fit.  Before submerging himself in the wine world on a full-time basis, Michel was a professor of Classics at Fontainebleau.  For a former French academic to invent a new word to describe an idea, Michel is simply following in the footsteps of the most famous French radical philosopher Jacques Derrida, who invented Deconstructionism.  To define Deconstructionism is something of a contradiction in terms (Jacques himself questions whether it can in fact be defined) but it is essentially a subverting of philosophy via a dismantling of language to expose contradictions and false assumptions.

Deconstructionist techniques have been successfully applied in architecture, literature and fine art and longueiline could be the first attempt at deconstructing winespeak, a field overflowing with fuzzy concepts and woolly thinking.  The anatomy of the word is intriguing: it has that certain je ne sais quoi that French terms confer in English with terroir a good example.  Being linguistically close to longueur, it has that Proustian languidness that is well suited to fine wine and being difficult to pronounce, raises the required barriers to appropriation by non-insiders.

Deconstructionism is by no means universally accepted.  Guardian guru Peter Lennon comments “the French excel in fabricated terms of shifty meaning which make it impossible to detect at what point philosophical speculation turns to gibberish.  Deconstruction is a theory which appears to lend itself most readily to babbling obfuscation.”  Which makes longueiline a perfect addition to the vocabulary of winespeak.

A deep philosophical issue lies behind adding the idea of linearity to wine: can a concept exist without a word to describe it and if so, what did we do B.B. (before Bettane)?  Research by the American linguist Benjamin Lee Whorf in the 1930s seems to indicate that language not only influences thought but more strongly, determines it.  So whether in the context of “purity of fruit” or “persistence of flavour”, linearity is a concept no self-respecting wine anorak can ignore, or at least finesse.