How vintage affects wine scoring (and what Rankquant does about it)

By Ryan Siegal · Founder and Principal

Published 2026-04-23

Why vintage matters so much in wine

Wine is uniquely sensitive to weather in a way almost no other consumer product is. Year over year, the same vineyard can produce wines of dramatically different quality depending on spring frost timing, summer heat, harvest-rain interruption, and hundreds of micro-level factors. The vintage is a shorthand for all of it.

Consider Bordeaux. The 2010 vintage was universally celebrated — professional critics averaged ~94 across classified growths. The 2013 vintage was a cool, damp disaster; critics averaged ~87 across the same producers. That's a 7-point gap between vintages of the same chateaux, independent of their inherent quality.

Why averaging across vintages breaks normalization

If we normalized a 2019 Bordeaux against the average of all Bordeaux of all vintages, the 2019 (an excellent year) would look only slightly above average — because the reference distribution includes the averaging-up effect of decades of great vintages mixed with decades of poor ones. The normalized score would be blunted.

Worse, the 2013 Bordeaux (a poor year) would look much worse than it is — because the reference distribution includes great vintages the 2013 never had a chance to compete with. A "bad" 2013 from a great producer might be a perfectly good wine; our normalized score would punish it.

How Rankquant handles it

Vintage enters as a peer-set dimension. Our hierarchical algorithm normalizes against:

In practice this means: for a 2019 Bordeaux Left Bank Cabernet blend at $50, we compare against other 2018–2020 Bordeaux Left Bank Cabernet blends at $40–$60. That's a defensible peer set of ~40–80 wines in most price tiers, and the z-score tells you genuine relative quality within your actual buying decision.

The edge cases we have to handle

Non-vintage wines — Champagne NV cuvées, many sherries, ports — don't have a vintage to match on. For these the peer set drops to Level 4 (style × grape × region) by definition, and the normalization uses the longer-term score distribution. This is fine because non-vintage wines are released under a house style that's intentionally vintage-invariant.

Library releases — wines re-released from the producer's cellar after extended aging — use their original vintage. A 2009 Château Lafite released in 2025 is still a 2009 Lafite for peer-set purposes.

The open question: vintage charts

Professional publications produce annual vintage charts rating each region's vintage on a simple scale (typically 80–100 per region per year). In theory we could use these to normalize vintage variation explicitly rather than via peer-set matching.

We don't, for two reasons. First, vintage charts themselves have inflation pressure (few charts rate any vintage below 85 for major regions). Second, the peer-set approach already captures vintage variation empirically — the adjustments show up in the observed distribution of scores rather than as a separate correction. Simpler, more transparent, and one fewer parameter to version-bump.

Related: How to read a wine score · Wine Spectator vs Parker · Vintage (glossary)