How the Index is checked, and what that check does not prove

This chapter describes a comparison that is not the one reported further down. It is the method that applies once external reference values are on file, and no part of it runs while that table is empty.

For each external source / vertical pair, we aggregate the Artmetria daily snapshots to a yearly mean, then join with the published external values on the year key. That comparison reports both Pearson (linear) and Spearman (monotonic-rank) correlation coefficients.

Pearson is sensitive to the absolute level alignment ; Spearman is robust to scale and only measures whether the two series move in the same direction. Against a real external series, a strong Spearman with a weak Pearson would be acceptable — that would just mean the two indices use different base levels.

Reference points are transcribed manually from public sources (annual reports, press releases, industry surveys). Each row carries a source_url for traceability.

The reference table (index_external_references) is currently empty. We are in the process of transcribing the most recent yearly values published by:

• Sotheby's Mei Moses — repeat-sale art market index, yearly since 2002, sub-indices by period
• Artprice100& Artprice global — auction-weighted artist index, yearly with sector breakdowns
• The Art Market Report (UBS / Art Basel) — annual auction turnover by segment
• HAGI Top 50 — collectible car index, yearly
• Liv-ex Fine Wine 100 — fine wine index, monthly

Each row will cite the public document it was transcribed from. The audit comparison auto-renders here when the table is populated.

Each vertical's index series is compared against its own three-point centred moving average, and the table below is computed from index_synthetic_values_daily when this page is served — not transcribed.

This is a self-comparison, and it is worth being exact about its limits. Nothing external enters it. A high coefficient here means the series is regular rather than noisy — that its movement is not dominated by day-to-day jitter. It does not mean the series is accurate, and it cannot: an index that is systematically wrong but moves smoothly scores exactly as well as one that is right. Agreement with the art market is what the external comparison above is for, and that comparison has not been made.

What this check does establish is narrower and still worth publishing: that a dated series exists across these markets, over the windows named below, and that it behaves regularly enough to be read. The figures are recomputed on every request, so the page cannot drift away from the series it describes.

• Pearson vs smoothed above 0.95 → the series is regular rather than noisy. This says nothing about whether its level is right
• YoY volatility is expected between 10 and 30 % for art-market indices — higher values are expected on more speculative segments
• High or medium confidence — the fraction of the observed history whose snapshots clear the sample-size threshold

15 of 15 verticals clear the regularity threshold, and 13 of 15 sit inside the expected volatility band. A vertical outside either is not a defect on its own — a speculative segment moves more, and a thin sub-index carries sparse historical lots — but the figures above are what the series actually says, and they are recomputed every time this page is served.

A vertical with fewer than ten recorded snapshots is not listed at all. A coefficient computed on nine points is not a weaker reading of stability; it is a number about almost nothing, and printing it beside a real one would flatten the difference.