Methodology

What the price means

The number we chart is TCGplayer Market Price — a weighted average of recent completed sales, not a list price and not an asking price. We record it once per day, per printing, and never modify a recorded value afterwards.

What we deliberately do not show

• The “high” price. Our source publishes one, and it is mostly noise: on the most recent day 51.9% of English records carry a high more than 20× the market price — sellers park listings at absurd prices to avoid re-entering item details. Across our whole history it is 23.7%, and the Japanese marketplace barely does it at all (4.7%), which is one more reason we never put a figure from one market beside a figure from the other. Charting it would be charting noise, so we store it and never display it.
• Graded (PSA/BGS/CGC) prices. A slabbed card is economically a different asset from a raw one. We have no trustworthy free source for graded sales, and an inferred one would be worse than none.
• Individual sold listings. No free, licensed source of sold comps exists for this market. We would rather show one honest aggregate than a scraped feed we cannot stand behind.
• Per-condition prices. Our source publishes prices per printing, not per condition, so a Near Mint and a Lightly Played copy share one series.

How the charts are drawn

• Step lines, not curves. A price holds until it changes; a diagonal between two observations would imply values the card never traded at.
• Gaps stay gaps. When we have no observation, the line breaks rather than bridging the period.
• Two lines, one scale. Where a chart shows the market price and the cheapest listing together, both are the same unit on a single axis. We never draw a second y-scale — two scales let a chart imply any relationship you like.
• Long ranges are aggregated. Views beyond 90 days read weekly or monthly buckets, so a chart is never denser than it is readable.
• Thin data is labelled. A card with under 30 recorded days carries a “limited data” warning instead of a confident-looking trend.

How the price index is built

The index answers “what did the market do” rather than “what did this card do”. It is built from the same daily observations as every other number here — no survey, no estimate, no smoothing, nothing backfilled.

Each weekly step is the geometric mean of price ratios for products priced in both of the two weeks — a matched sample — and the steps are multiplied together to form the line. Matching every step is what makes membership churn harmless: a product first appearing, or finally delisting, produces no ratio and therefore cannot move the level. Only prices can.

• The obvious construction does not measure the market. Averaging daily percent change instead returns anywhere from 50% to 152% a year depending on choices that have nothing to do with prices: whether a price floor is applied, whether flagged observations are dropped, and whether a product absent for months has its entire return counted as a single day’s move. A figure that swings by a factor of three on those choices is not describing the market. Worse, a floor applied continuously lets falling products leave the sample while rising ones never do, so what it mostly measures is who is left. The matched sample removes all three: every step compares the same product against itself, one week apart. A published line annualising past +100% is treated as a regression and fails a test.
• A mean, not a median. A median was tried first and is degenerate on this data: card prices are sticky. Across all 1,605,826 weekly comparisons in our history, 49.1% show no price change at all — 45.9% in the English market and 63.5% in the Japanese one, where prices move far less often — so the median step is exactly 1.000 and the index would read flat forever. Extreme steps are clipped to a halving or doubling before averaging — that touches 0.11% of them and moves the annual figure by under a point, so it is a safety net against a parked price, not the thing producing the answer.
• Sealed and singles are never blended. They are separate markets that move in opposite directions by mechanism — opening boxes adds singles to the supply and removes sealed from it — so a blended line would average away the only signal worth publishing. English and Japanese are likewise kept apart.
• A $5 floor, applied per step rather than standing. Without it a one-cent card moving to two cents reads as +100% and penny rounding dominates. A product must clear $5 in both weeks of a step to count, but falling below never expels it from the index — it simply has no comparison that week, and rejoins when it clears again. The floor changes the answer, which is why it is stated rather than assumed. English singles over the trailing eleven months (2025-09-30 to 2026-08-30) end at 142.4 with no floor, 157.3 at $1, 163.0 at $5 and 155.7 at $20. It also means the index describes the market above $5, not the penny commons below it.
• A week we could not measure truncates the line; it is never bridged. Dropping a thin week and multiplying whatever survived would jump the level across real movement while still drawing an unbroken line. A published line is therefore the longest unbroken run of weeks with enough matched pairs, and each panel states the date it is rebased to 100 — so a shorter window is visible rather than implied.
• The sample size travels with the level. Every panel prints how many products its latest step compared. A whole market needs 30 matched pairs in a week to publish a step for singles and 15 for sealed; a single set needs 12 and 5, because a set holds a handful of sealed SKUs — a box, an ETB, a blister — rather than hundreds of cards. Where there is not enough priced history, the panel says so instead of drawing a line.
• Flagged observations are excluded, so a parked or stale price cannot enter a level. The whole index is rebuilt from scratch each night after the anomaly sweeps rather than extended forward — a price we reject tomorrow must not remain inside a level published today.

Coverage

Daily history begins 8 February 2024 for English cards and 11 December 2024 for Japanese ones — the earliest dates our source’s archive covers. Nothing before that exists in our data, and we do not estimate it.

Card images

Card images are served directly by TCGplayer’s own image CDN. We store the address of each image and your browser requests it from them — we do not copy, re-host or modify the files, and none of them are stored on our servers. Every product on this site links back to its TCGplayer page.

Card artwork is the property of Nintendo, Creatures Inc. and GAME FREAK. Where an image is unavailable we show an empty placeholder rather than substituting a different card’s picture; roughly 5.9% of products across both markets have no image published, mostly recent promotional cards.