Methodology

Where the data comes from

All listing data is collected from the official eBay Browse API. We query 7 regional eBay markets and normalize every result into a single dataset:

• US United States
• GB United Kingdom
• DE Germany
• FR France
• IT Italy
• ES Spain
• PL Poland

A full sweep of all 114 models across all 7 markets runs every 8 hours. Each sweep captures the currently active listings; sold listings are recorded as they complete, building the historical record used for the medians below.

How listings are validated

eBay search is noisy: a query for “RTX 3080” returns cooling fans, brackets, laptops, and empty boxes alongside the actual cards. To keep those out of the index, every listing passes through a per-category machine-learning classifier: a TF-IDF vectorizer feeding a LinearSVC model, trained separately for each hardware category (GPU, CPU, RAM, motherboard, and so on). Listings the classifier flags as miscategorized are rejected before they can affect any price.

Surviving listings then pass a per-model price-plausibility check. Each model has an expected price band; listings that fall far outside it (a €20 “RTX 4090”, a €9,000 typo) are dropped as data errors rather than dragging the median around.

Currency normalization

Prices arrive in USD, GBP, EUR, and PLN. Every price is converted to EUR using daily foreign-exchange rates, cached once per day. All figures on the site, and in this dataset, are expressed in EUR unless a local currency is explicitly shown.

How the numbers are computed

• Medians only. Every headline price is a median, not a mean. Medians resist the outliers and price typos that skew averages on a marketplace like eBay.
• Sold vs. asking, always separate. Sold prices (what items actually closed at) and asking prices (what active listings request) are never blended. They are reported as distinct figures because they answer different questions: what a part is worth versus what sellers hope to get.
• Sample size travels with the number. Every median carries the count of listings (n) it was computed from, so a figure backed by 3 sales is never presented with the same confidence as one backed by 300.

In the trailing 90 days this covers roughly 68k+ sold listings worth about €22M+ across 114 models, drawn from 390,000+ listings processed to date.

Known limitations

Honesty about the boundaries of the data:

• eBay only. The index reflects the eBay secondary market. Retail, Amazon, local classifieds, and forum sales are outside its scope, so it is not a universal “street price.”
• Thin markets are flagged, not hidden. Some model/market combinations have few sales. We surface the sample size (n) rather than suppress the figure. A low n means “treat with caution,” not “wrong.”
• Condition granularity. eBay condition labels (Used, For parts, etc.) are coarse and seller-declared. We cannot verify wear, remaining warranty, or whether a card was mined on.
• FX timing. Rates are cached daily, so cross-currency comparisons carry up to a day of exchange-rate lag. Immaterial for trend reading, worth noting for to-the-euro precision.

Citing this data