Bordeaux wines: Actual and predicted prices, proposed by Charles Wyplosz

Charles Wyplosz, Professor of International Economics at the Graduate Institute of International Studies in Geneva proposes:

Bordeaux wines / Actual and predicted prices

Or How are determined the best quality to price picks on Winemega

Are more expensive wines better? In principle, this is how it should be, but is it really worth paying more for better? Except for the happy few, most of us try to find a compromise between price and quality. But prices and quality are not simply related. Using Winemega’s comprehensive data on prices and ratings (2311 bottles over the vintages 1997 to 2005), the following chart (each point represents a wine in a given year) shows that prices tend to become pretty steep as the rating comes close to 100. It also shows many exceptions to the rule “more is better”. In addition, we all know that some particular growths (crus) fetch a higher price.

In order to take account of these various aspects, I submit each year the nearly 300 ratings and prices collected by Winemega to a statistical treatment. The idea is to detect the link between price and rating, taking into account growth. This gives me an average relationship that I next use to answer the following kind of question:
"In the 2005 vintage, what is the average price of a Pomerol rated 89.5?"

Dispersion of wine ratings 1997-2005

For each wine of a given growth and vintage I find what I call a predicted price. I then compare the predicted price with the actual price. If the actual price exceeds the predicted price, it means either that :
- the bottle is overpriced
- that the rating is too low
- or that some people like this wine much more than the average of the professionals whose ratings are tallied by Winemega (remember: Winemega publishes the average ratings, and there never is unanimity). The tables presented on the website display the percent difference between the actual and the predicted price: if it is negative, it means a good buy; if it is positive, it means that the actual price exceeds the predicted price (see our Top Quality to Price 2005 page).

In addition, I also perform the same analysis for all the wines collected in the Winemega data base since the 1997 vintage. In this case, I allow for a separate effect of the vintage. This gives an idea of which vintages are more or less expensive, given the rating and growth of each bottle.

This is not science, but close. In performing my statistical treatment, I must make a number of assumptions that may, or may not, be fully warranted. I have tried lots of things, and found that they make little difference. In the end, I settled for the simplest possible treatment (1), if only to make it as transparent as possible.

(1) For those in the knowing, I run the following linear regression :
Log(price)= a*rating + growth dummies. I obtain an adjusted R2 of about 75%.
Allowing for a non linear effect of rating slightly improves the regression but makes it impossible to measure a constant price elasticity.

Surprises in 2005
The results for the 2005 vintage carry a few surprises:

1. Vintage effect

Looking at the prices, I thought that this was the most expensive year ever. Not so, says the statistical analysis. The table below provides the answer to the following question:
"How much more expensive (in %) are wines in a given year, relative to the cheapest year on record (2003), after taking into accounts the ratings and the growth premia?"

TOP

1997	1998	1999	2000	2001	2002	2003	2004	2005
8.9	0.6	37.9	44.9	56.7	22.0	0.0	4.4	23.8
	*					Base	*
* means not statistically different from zero.

It turns out that 2005 vintage wines are 24% more expensive than 2003 vintage wines, for the same rating and the same growth. This measure captures the vintage reputation effect. This is a lot, but also a lot less than 2001, 2000 and 1999, in that order; in fact, 2005 comes close to 2002.

It may be that the professional tasters have been more restrained in 2005 than in those high-flying years. It may be that producers – on average, mind you – did not realize how good their wines are. At any rate, 2005 is relatively not that expensive, a surprising conclusion. Of course, given the generally higher ratings of 2005 wines, they end up being absolutely expensive.

2. Rating effect
The analysis also allows answering the following question:
"How much, on average, does a rating increase of 1 point (say, from 87 to 88) raises the price
(in %)?"

1997	1998	1999	2000	2001	2002	2003	2004	2005
10%	12%	9%	15%	12%	12%	11%	16%	23%

Interestingly, the otherwise modestly overpriced 2005 vintage is the one where prices are the most sensitive to ratings. The rating premium doubles in percentage terms over the situation in previous good years. Maybe producers have honed their skills at taking advantage of good reviews? Quality is expensive in 2005.

A quick look at the figure above shows that prices become extremely sensitive when the rating approaches 100. This is where we find speculative wines and prestigious wines for which price seems to matter little. This effect is not captured in the above statistical analysis, but it can be done. If I allow for the sensitivity to rise with the rating (2), I find that this is indeed the case. The following table answers the question:
"In the 2005 vintage, how much, on average, does a rating increase of 1 point raises the price (in %) depending on the rating level?"

Rating	80	81	82	83	84	85	86	87	88	89
Price effect	3.3%	0,4%	2.5%	5.4%	8.3%	11.2%	14.1%	17.0%	19.9%	22.8%

Rating	90	91	92	93	94	94	96	97	98	99
Price effect	25.7%	28.5%	31.4%	34.3%	37.2%	40.1%	43.0%	45.9%	48.8%	51.7%

Life is really expensive at the top!

(2) This is done adding a quadratic term: Log(price)= a*rating + b*(rating) 2 + growth dummies

3. Growth (or Region) effect

In the same vintage, for the same rating, some wines are more expensive than others. This may be a notoriety effect or the consequence that wine tasters tend to be more lenient with less prestigious growths. The effect is very significant but varies over time. The table below provides the answer to the following question:
"How much more expensive (in %) are wines from a given region in a given year, relative to the cheapest region in that year, after taking into accounts the ratings?"

	Pessac Leognan /Graves	Margaux	Medoc and Haut Medoc	Pauillac	Pomerol and Lalande de Pomerol	Sauternes	St- Emilion	St- Estèphe	St- Julien	Misc.
1997	75.8	77.4	0	83.5	97.5	44.1	82.3	50.5	92.5	21.8
1998	54.4	63.8	0	82.8	51.2	48.0	72.5	38.6	76.9	23.1
1999	62.6	59.3	0	91.6	104.1	64.1	79.7	72.3	84.6	39.4
2000	43.9	47.6	0	48.4	79.1	35.9	62.5	25.4	56.7	25.6
2001	43.2	46.2	0	48.2	62.5	0.4	44.9	19.9	45.5	3.5
2002	44.9	41.3	0	41.0	77.3	13.8	53.3	14.6	37.5	18.0
2003	63.0	57.9	0	39.7	94.0	50.3	69.4	22.4	54.1	23.1
2004	40.5	37.3	0	34.8	52.9	49.0	40.2	29.5	39.1	17.5
2005	25.6	39.1	0.1	51.5	58.3	0.0	58.6	10.4	34.8	8.6
Misc: Fronsac, Canon-Fronsac, Moulis, Listrac, Montagne Saint-Emilion and Bordeaux Sup.

TOP

Vintage 2005 is the only one where Medoc and Haut-Medoc – they are considered jointly – are not the cheapest. By a non-statistically significant breeze, Sauternes are, after two record years. The surprise comes from Graves and Pessac-Leognan – also considered jointly – and Saint Estèphe.
Note also that the differences between many regions tend to be less large than before. It could be that quality is generally improving.

4. The top-rated wines
While some highly rated wines are in our top 100, most are not. The following table shows the rankings of the 52 wines that are rated 90 and above in Winemega’s survey. This statistical treatment, however, differs from the one used so far. Because prices rise so steeply at the top, I allow for a sensitivity that increases when the ratings become higher, as shown in the table above. The results appear below and top-wine amateurs may find some interesting deals, keeping in mind in otherwise hefty prices of all these wines. (Note that these are estimates that are based on 2005 vintage wines, which are 24% higher on average than those from 2003.)

Classification

Différence (%) between actual and predicted price

Name

Region