One set, bought on the day it released and never traded again: this page asks which slice of it you should have bought. The same cohort of sets, cut ten ways by rarity tier, each tier held from day zero and read against its own launch price — best and safest ranked separately, because they are different questions.
The verdict
Best and safest to holdIllustration rares
One tier wins both questions: Illustration rares paid the most and fell the least.
every era · 36 sets · today's top ten is shown below but never ranked, because it picks its members knowing which ones won
Bulk is printed to demand and the presses run for months after release, so its supply is still arriving while the hype is leaving: bulk (commons, uncommons, rares) reads -14.8% at twenty-four months across 21 sets, with a median fall from launch of 60.0%.
A chase card's supply is fixed by the pull rate rather than by the print run — every extra box prints far more bulk than chase — and the tier holds up accordingly: -36.3% at twenty-four months over 24 sets, 50.7% down at its worst against bulk's own fall.
Buying the dearest ten cards on release day is a different bet from buying the dearest ten now: launch prices carry the hype premium, and the grails tier reads -28.7% at twenty-four months over 24 sets. The ten that are dearest TODAY read -4.6% over the same span, and the gap between those two numbers is the whole cost of guessing wrong on release day — that tier picked its members knowing the answer.
By era · +24 m against launch
Tier
Sword & Shield
Scarlet & Violet
Mega Evolution
Grails (top 10 at launch)
-43.1%n 13
-17.9%n 11
not enough setsn 0
Chase tier
-40.8%n 13
-28.1%n 11
Correlation with the bottom
Correlations are strongest on the EW basket (n = 32, best q = 0.002 on eraOrdinal); that basket's buy list is the one to rank on. Other baskets are shown for comparison.
Tier
Strongest surviving predictor
ρ
q
Sets
Grails (top 10 at launch)
Era, oldest to newest vs when the bottom lands
-0.61
0.027
26
Chase tier
Era, oldest to newest vs when the bottom lands
-0.63
0.021
26
Illustration rares
One set, tier by tier
Method & caveats
Each tier is a basket of one set's cards, assembled on that set's first priced day at or after release and never rebalanced. Membership comes from the rarity printed on the card, through the platform's own normalised rarity ladder — so every tier but today's top ten is known on release day, with no hindsight. A tier needs at least five members in a set before it is built at all; a cell needs 8 sets before it prints a number.
TCGplayer market price (market guide), equal weight across the set's variants. Not a delivered-ask floor and not condition-specific.
Returns are measured against each set's own launch level, so a 2022 set and a 2025 set are compared at the same point in their lives rather than on the same calendar day. Every horizon is a set's OWN age: a set whose record does not reach thirty-six months is absent from that column rather than contributing its latest level under a longer label.
Past about thirty months, age and era are the same variable: only one cohort of sets is that old inside this price history, so the composite there describes those sets rather than "sets at that age". The era mix behind every bin is published beside the curve for exactly this reason.
Most sets enter this study mid-life. Their launch is not on file, so their baseline is where the evidence starts, not where the set started — which is why every set carries the age its record begins at, and why Study A refuses anything that is not a day-zero set.
Before September 2025 most days come from weekly archive rows carried flat across the week, and TCGplayer's own history arrives in one-, three- and seven-day buckets. Short moves inside a bucket are invisible, so early volatility is understated and a trough dated inside that period is dated to about a week, not to a day.
Before about February 2024 the line is a different measurement on a different clock. Those months come from PriceCharting's own MONTHLY figure for an ungraded copy — one point per calendar month, carried flat across it, and a summary of the ungraded eBay market with no condition stated — while everything after is TCGplayer's daily market guide for a near-mint one. The two are never blended, averaged or smoothed into each other: a day the daily guide covers takes the daily guide, and the monthly figure answers only the days nothing else reaches. Inside the spliced era a trough is dated to about a month, not a week and not a day, and a level is comparable across sets rather than exactly comparable with the same set after the seam.
What these words mean here
Grails (top 10 at launch)
— The ten most expensive cards on release day, bought then and held.
Chase tier
— The set's top rarity tier, bought at launch — the cards people open boxes for. Membership is fixed on release day.
Illustration rares
— Every Illustration rares in the set, bought at launch and held.
Ultra rares
— Every Ultra rares (ex, V, VMAX, double rares) in the set, bought at launch and held.
Double rares
— Every Double rares (ex, V) in the set, bought at launch and held.
Holo rares
— Every Holo rares in the set, bought at launch and held.
Bulk (commons, uncommons, rares)
— Every Bulk (commons, uncommons, non-holo rares) in the set, bought at launch and held.
One of each card
One of each card in the set, bought at launch and never rebalanced — the cost of the set.
Questions
After a set launches, which cards should I actually hold?
One tier wins both questions: Illustration rares paid the most and fell the least.
And which is the safest thing to hold?
The same tier: Illustration rares also had the shallowest fall of any tier here, so the best tier to hold and the safest tier to hold are one and the same in this build.
Why are "best" and "safest" ranked separately?
Because they are different questions. Best is the biggest median return at twenty-four months; safest is the shallowest fall and the highest share of sets back above launch. Blending them into one score would hide exactly the trade a buyer is making, so this page ranks three ways and never averages them.
What counts as a chase card here?
The set's top rarity tier as printed on the card: special illustration rares and hyper rares on Scarlet & Violet sets, the secret and alt-art tier on Sword & Shield. Membership is known on release day, from the catalogue, so nothing about this tier needs hindsight.
Why does the typical card fall so much further than the chase cards?
Bulk is printed to demand and keeps arriving for months after release, while a chase card's supply is capped by the pull rate — every extra box prints far more commons than secret rares. Measured here, the typical card takes a median fall of 60.0% from launch.
Is each tier traded, or just bought and held?
evidence through 2026-10-26 · built 2026-10-26
Research classification generated from observable market data with stated confidence and assumptions. Not personalized financial advice.
-28.7%
n 24
+3.4%n 18
-30.5%n 36
-53.5%worst typical -76.3% · n 36
month 13back above launch in 5 m · 20 not yet17 of 30 still near their low at 12 m
6 of 2425% of sets
-75.8%Shining Fates Shiny Vault
Chase tiern 36 in this cohort
16median per set
-33.7%n 33
-38.3%n 30
-36.3%n 24
-28.0%n 18
-39.3%n 36
-50.7%worst typical -72.9% · n 36
month 17back above launch in 6 m · 27 not yet21 of 30 still near their low at 12 m
5 of 2421% of sets
-66.2%Shining Fates
Illustration raresn 22 in this cohort
18median per set
-16.6%n 19
+18.2%n 16
+136.1%n 11
not enough setsn 5
+28.0%n 22
-44.6%worst typical -67.0% · n 22
month 4back above launch in 3 m · 9 not yet1 of 16 still near their low at 12 m
11 of 11100% of sets
+8.4%Surging Sparks
Ultra raresn 36 in this cohort
37median per set
-33.8%n 33
-44.0%n 30
-37.7%n 24
-26.8%n 18
-32.3%n 36
-57.3%worst typical -81.3% · n 36
month 15back above launch in 18 m · 22 not yet21 of 30 still near their low at 12 m
1 of 244% of sets
-71.4%Temporal Forces
Double raresn 23 in this cohort
12median per set
-44.4%n 20
-43.7%n 17
-42.9%n 11
not enough setsn 5
-46.6%n 23
-61.7%worst typical -82.6% · n 23
month 7back above launch in 11 m · 20 not yet3 of 17 still near their low at 12 m
0 of 110% of sets
-70.4%Temporal Forces
Holo raresn 4 in this cohort
3median per set
not enough setsn 4
not enough setsn 4
not enough setsn 4
not enough setsn 4
not enough setsn 4
not enough setsworst typical -73.6% · n 4
not enough setsback above launch in 9 m · 1 not yet0 of 4 still near their low at 12 m
not enough setsn 4
-61.8%Lost Origin
Bulk (commons, uncommons, rares)n 33 in this cohort
86median per set
-30.5%n 30
-16.9%n 27
-14.8%n 21
-30.6%n 14
-34.8%n 33
-60.0%worst typical -75.0% · n 33
month 9back above launch in 4 m · 22 not yet3 of 27 still near their low at 12 m
7 of 2133% of sets
-81.7%Evolving Skies
One of each cardn 36 in this cohort
134median per set
-26.5%n 33
-35.7%n 30
-22.5%n 24
-5.9%n 18
-12.8%n 36
-47.8%worst typical -71.0% · n 36
month 13back above launch in 8 m · 19 not yet20 of 30 still near their low at 12 m
7 of 2429% of sets
-66.2%Shining Fates
The typical cardn 36 in this cohort
134median per set
-24.6%n 33
-23.9%n 30
-24.9%n 24
-22.3%n 18
+4.2%n 36
-45.2%worst typical -62.2% · n 36
month 10back above launch in 10 m · 14 not yet17 of 30 still near their low at 12 m
9 of 2438% of sets
-52.3%Shining Fates
Top 10 todaypicked in hindsightn 36 in this cohort
10median per set
-30.2%n 33
-30.9%n 30
-4.6%n 24
+23.1%n 18
+2.7%n 36
-48.7%worst typical -75.9% · n 36
month 9back above launch in 7 m · 14 not yet16 of 30 still near their low at 12 m
11 of 2446% of sets
-72.1%Shining Fates Shiny Vault
One row per tier, every tier held from day zero and never rebalanced. Each cell carries the sets behind it; a cell with too few sets prints no number at all.
Each tier at 100 on release day, then the median set of the cohort at each horizon. The axis is logarithmic, so equal percentage moves are equal distances.
not enough sets
n 0
Illustration rares
not measured
+136.1%n 11
not enough setsn 0
Ultra rares
-36.9%n 13
-38.5%n 11
not enough setsn 0
Double rares
not measured
-42.9%n 11
not enough setsn 0
Holo rares
not enough setsn 4
not measured
not measured
Bulk (commons, uncommons, rares)
-25.7%n 10
-9.3%n 11
not enough setsn 0
One of each card
-34.0%n 13
+7.1%n 11
not enough setsn 0
The typical card
-36.8%n 13
+13.2%n 11
not enough setsn 0
Top 10 today
-37.6%n 13
+73.4%n 11
not enough setsn 0
no signal · 30 tests, none survived
Ultra rares
Era, oldest to newest vs when the bottom lands
-0.57
0.042
29
Double rares
What the market was doing at launch vs when the bottom lands
-0.75
0.018
18
Bulk (commons, uncommons, rares)
Era, oldest to newest vs when the bottom lands
-0.82
0.003
29
One of each card
Era, oldest to newest vs when the bottom lands
-0.70
0.006
27
The typical card
Era, oldest to newest vs when the bottom lands
-0.84
0.002
32
Top 10 today
Era, oldest to newest vs when the bottom lands
-0.66
0.006
27
8 of 9 tiers carry a coefficient that survives correction. A coefficient smaller than 0.35 is below what 32 sets can detect at all, so an absent signal here is not evidence of no relationship — only of none this cohort could see.
No move is ever chained across that seam. A link is measured only between two days priced the same way, per card, so on the day a card hands over from the monthly ungraded figure to the daily near-mint guide its level carries across unchanged and no reading — no low, no phase, no rebound — is taken on that day. The link that was refused is published for each set beside its row, with the one-day move a naive chain would have printed: on an equal-weight line that commons dominate, an ungraded floor of about a dollar handing over to a market price of about a dime reads as an 80% fall that never happened.
The default axis is absolute, which is what a buyer pays — and it therefore carries whatever the whole market did. "New sets get cheaper" and "everything got cheaper in 2025" are separated only by the market-relative toggle, and the two readings are published side by side.
Sets are staggered in calendar time, which is what lets a composite separate age from date at all — but the separation is only up to a common trend. A drift shared by every set in the window is absorbed into the age curve under the absolute basis and removed under the peer basis; neither can tell you which one is right.
A set only contributes to an age its own record reaches, so the oldest bins are answered by a shrinking, self-selected subset. Every bin publishes the number of sets behind it, and a bin under the minimum is drawn as a gap rather than as a thinner line.
The y axis is the TCGplayer market guide: a quoted level for the product, not a delivered price, not a condition-specific price, and not what the cheapest listing plus shipping would have cost that day. A delivered-ask floor exists only from September 2025 forward and is deliberately not mixed in.
Weekly. The cohorts behind this page move over months, not hours, so it is rebuilt once a week rather than every night — the owner’s call, and an honest one: a nightly rebuild of a monthly phenomenon publishes noise. Every number here, these answers included, is read from the build below and replaced by the next one. Evidence through 2026-10-26, price history from 2020-12-01 — the cohort and the bottoms behind these tiers are the Bottom Finder study's, and this page is the same evidence cut by rarity.
—
The typical card
— Every card, equally weighted — commons and bulk included. A control, not a portfolio.
Top 10 today
— The ten most expensive cards TODAY, priced back to launch. Picked knowing which ones won.
Held
— Bought once on the set’s first priced day and never rebalanced or traded again.
Median
— The middle set of the cohort — not an average, so one runaway set cannot move it.
Max drawdown
— How far the tier fell below its own launch price at its worst point.
Above launch
— Sets whose tier is worth more than it cost on release day, counted over the cohort.
Bottom
— The age, in months, at which the tier’s low printed — measured after the fact.
Picked in hindsight
— Members chosen knowing which ones went up. Shown for comparison, never ranked.
not enough sets
— Fewer sets have lived through that horizon than a median may honestly be taken over.
Bought once and held. Every tier is a basket assembled on the set's first priced day and never rebalanced, so what you read is what one purchase did — not the result of switching between cards as prices moved.
Why is "today's top 10" flagged?
Because its members were chosen knowing which ones went up. It is the winners' line by construction, it is published because readers ask about it, and it is excluded from both verdicts — a recommendation you could only have followed with hindsight is not one.
Why do some cells say "not enough sets"?
Because fewer than 8 sets in that cohort have lived through that horizon. The count is still printed, so "we looked and found 7" never reads like "we did not look". A median over three sets is a fact about three sets.
What price is all of this measured on?
TCGplayer market price (market guide), equal weight across the set's variants. Not a delivered-ask floor and not condition-specific.
How far back does the evidence go, and how often does this change?
The price history starts 2020-12-01 and this build carries evidence through 2026-10-26, so every reading longer than that span is bounded by it. Weekly. The cohorts behind this page move over months, not hours, so it is rebuilt once a week rather than every night — the owner’s call, and an honest one: a nightly rebuild of a monthly phenomenon publishes noise. Every number here, these answers included, is read from the build below and replaced by the next one.