Methodology — Synth Market Observatory

1. Data collection

Our pipeline runs approximately every hour and visits four public marketplaces:

Source	Market	Type
Hispasonic.com	Spain	Second-hand classifieds
Soundsmarket.com	Spain	Second-hand classifieds
Audiofanzine.com	Europe (multi)	Second-hand classifieds
Noiz.gr	Greece	Second-hand classifieds

For each listing we record: title, asking price (converted to EUR where necessary), city, country, listing date, and instrument condition. We do not store seller identities.

eBay listings are shown in real time via the Browse API but are not stored in our database, in compliance with the eBay API License Agreement.

2. 72-hour embargo

Every listing is held for 72 hours before it can appear in any public output. This gives sellers time to modify or remove their listing before it becomes part of the dataset, and reduces the risk that we publish ephemeral or accidental posts.

3. Product normalisation

Listing titles vary enormously across sources and sellers (e.g. "Korg minilogue xd", "KORG MINILOGUE XD Module", "minilogue xd poly"). We use a machine learning classifier trained on a curated catalogue of synthesizer models to match each listing to a canonical product name (e.g. "Korg Minilogue XD").

The classifier uses TF-IDF features and a nearest-neighbour approach. Listings that do not meet the confidence threshold are left unclassified and excluded from aggregate statistics.

The product catalogue currently covers approximately 3,000 canonical models from major synthesizer manufacturers.

4. Fair Market Price

For each product with sufficient data, we calculate three price statistics over a rolling 90-day window:

Statistic	Meaning
P25 — Low end	25th percentile — one quarter of listings are below this price. A deal worth investigating.
P50 — Fair price	Median — half of listings are above and half below. The most typical asking price.
P75 — High end	75th percentile — three quarters of listings are below this. Still within normal market range.

We require a minimum of 5 classified listings per product to publish a Fair Market Price. Outliers (prices more than 3× the median or below 20% of the median) are excluded from the calculation.

Fair Market Prices are statistical summaries of asking prices, not professional valuations. Actual transaction prices may differ. Condition, included accessories, and geographic factors are not weighted in the current model.

5. Known limitations

Asking price ≠ sale price: we observe listed prices, not confirmed transactions.
Classification errors: the ML classifier is not perfect. Unusual titles or bundles may be misclassified.
Geographic bias: Spanish and Greek marketplaces are over-represented. Coverage of Northern and Eastern Europe is limited.
Condition is not weighted: "mint" and "fair condition" listings contribute equally to Fair Market Price calculations.
Audiofanzine coverage: only listings with an explicit synth category in the URL are captured; uncategorised listings are missed.

6. Open dataset

Aggregate statistics are published as open data under CC BY 4.0. Individual listing data (titles, prices, URLs) from third-party marketplaces is not included in the public dataset. See the Terms of Use for the dataset licence.