Coin comps graded and raw, from real sales

Grade is the market, so it is parsed out

A 1921 Morgan in MS63 and the same coin in MS65 are different markets, and mixing them makes a median meaningless. Every row carries the grade and grading service parsed from the listing title, so you can split PCGS from NGC from raw without string matching it yourself.

Verified 12 October 2026, three year window: 1921 Morgan silver dollar returned 199 completed sales on the first page.

curl -G https://api.compsapi.com/v1/sold \
  -H "Authorization: Bearer YOUR_KEY" \
  --data-urlencode "q=charizard psa 10" \
  --data-urlencode "days=1095" \
  --data-urlencode "best_offer=1"

Raw coins are where the offers hide

Slabbed coins mostly trade at or near the asking price. Raw coins and bulk lots are negotiated constantly, and on those a public sold page shows what the seller asked rather than what the buyer paid. We return the accepted figure with a best_offer flag on every row. Across 5,859 matched sales the public number read high 99.0% of the time by a median of 14.9%.

Filtering the noise

Coin searches attract replicas, lots and accessories. A minus word clears them: 1921 Morgan silver dollar -copy -replica -lot -roll. Keep the search short otherwise, because every word has to appear in the seller's title.

Why three years

Bullion linked coins move with metal prices and numismatic premiums move slowly. Ninety days shows you neither. Growth and above cover the full 1,095 days, enough to separate a melt driven move from a collector driven one.