Does a Big PSA Population Lower the Price?

The folk model says population is supply and supply is gravity: a PSA 10 with fifty thousand siblings should be worth less than one with five hundred. Seven weeks ago we tested that on 48 cards and found a rank correlation of -0.46 between registry size and the gem premium. This is the same test on 3,606 cards, and it returns -0.46 again. The model is right about the premium and wrong about the price, and the difference between those two statements is most of what people get wrong about population reports.

Across the 3,606 cards for which we hold a raw ask, a PSA 10 sold median on ten or more verified comps and a trusted PSA registry, registry size and the PSA 10 premium correlate at a Spearman -0.46, while registry size and the PSA 10 sold median itself correlate at +0.11, as of September 22, 2026. Both facts are in the same data and they do not contradict each other. Cards with huge registries are mostly expensive, popular cards, so their slabs sell for more in dollars. What collapses is the ratio between the slab and the raw copy, because when a ten is the ordinary outcome the slab stops certifying anything scarce.

Read the table from the top and the ranking is informative. The strongest single predictor of a thin premium is not how many slabs exist, it is how many TENS exist, at -0.62. Gem rate comes next at -0.55. Registry total is fourth, behind even the raw price of the card. Population reports print the total; the number that moves the premium is the top line of the breakdown.

The premium falls monotonically by a factor of nearly five. The PSA 10 sold median in the last column does not: it dips in the middle and finishes higher than it started. The cards with the deepest registries carry the dearest slabs, because depth is a consequence of demand as much as of supply. This is the half of population giants that its 48-card sample already saw, now on 75 times the population.

Registry size and gem rate move together, weakly, at +0.17. Big registries skew modern and modern cards gem more easily, so some of the population effect is really a gem-rate effect wearing a population coat. Splitting on gem rate instead produces a steeper ladder than splitting on population does, from 24.29x in the hardest quintile to 3.71x in the easiest.

The cleanest way to see the two forces separately is a two-by-two, split at this population's own median registry of 1,591 slabs and its own median gem rate of 48.39%. Both variables survive the other. A small registry with a hard gate carries a median 20.03x. A big registry with an easy gate carries 3.78x. Hold the gate hard and grow the registry and you still lose most of the premium, from 20.03x to 11.12x. Hold the registry small and make the gate easy and you lose about the same, from 20.03x to 10.37x.

Holding the gem rate roughly fixed and sorting by population inside each band, the effect is still there in every band and it is largest in the middle of the gem-rate range. In the 40% to 60% gem band, the smallest third of registries carries a median 14.84x and the largest third carries 4.54x. In the 60% to 80% band it is 13.17x against 3.75x. At the extremes the compression is weaker, because a card that almost never gems keeps a premium whatever its registry does, and a card that almost always gems has no premium left to lose.

Holding price roughly fixed instead, the gem-rate signal strengthens as cards get dearer, running from -0.50 among cards asking under $5 to -0.82 among cards asking $100 or more. At the expensive end of the market, how hard a card is to gem explains most of what its slab is worth over the raw copy.

The deepest registry we can price belongs to a Japanese promo Pikachu, with 290,960 recorded tens out of 332,353 slabs, gemming at 87.55% into a $77.78 PSA 10 sold median against a $34.46 ask. That is a 2.26x premium on a registry the size of a small city. At the other end, Umbreon from Neo Discovery has produced 112 tens from 7,475 slabs, a 1.50% rate, and its PSA 10 sells at a median $75,750.00 against a $752.74 raw ask.

The English Mega Charizard X ex and Charizard V rows are the useful ones. Both sit on registries above 90,000 and both keep premiums above 5x, because both gem at around half. Population did not compress them; the gate held. Compare ニンフィアex at 89.27%, or the Japanese Mega Charizard X ex at 86.79%, where the slab is worth 15% and 19% more than the raw card and nothing else.

This is a cross-section on one day, not a time series, so it cannot say that growing a registry causes a premium to fall on a given card. It says that cards with big registries today carry thin premiums today. The direction of causation is not established here, and both variables are downstream of a third: how desirable the card is. The population growth tax attacks the same question from the time-series side and is the piece to read against this one.

The raw side is an ASK and the graded side is a SOLD median, so every premium here is biased downward wherever asks run above fills. Registry totals are provider data, and the shared-line and continuity rules drop a meaningful number of keys rather than print a figure two cards both claim; the numbers we do not print explains that machinery. And the study is conditional on the card having a sold PSA 10 at all, so the 3,606 are cards whose top grade has a market, which is a filter on both variables.

The headline dies if a recompute on the same tables returns a registry-to-premium correlation that is positive, or if the five population quintiles stop being monotone in the median premium. The secondary claim dies if registry size turns out to correlate negatively with the PSA 10 sold median itself, which would mean population really does push the price down and not just the ratio. And the ranking claim dies if registry total ever out-predicts the PSA 10 count. All three are one join away from being checked.

Universe: rows of `cards` with a TCGplayer product id holding a latest `price_history` ask above zero (day 20718, September 22, 2026), a latest `graded_snapshots` PSA 10 sold median on ten or more verified comps, and a `population_snapshots` reading that survives the shared-line rule, the registry continuity rule and the trusted-gem-count rule described in our registry methodology. No floor on registry size is applied, because a floor would remove exactly the small registries the study is about. Rows are collapsed to one per product id. That is 3,606 cards. Premium is the PSA 10 sold median divided by the raw ask. Correlations are Spearman rank correlations with average ranks on ties, computed on the full 3,606 unless a band is stated. Quintiles are equal-count cuts of that population. Medians are upper-middle on even samples. The prior version of this study, on 48 cards, is population giants; this one supersedes its correlation and quintile figures and reproduces its sign and magnitude. Related: print-run archaeology, the registry is closed, easy grades are cheap grades. Not financial advice.