Do Pokémon Sets Gain Value With Age? 100 Sets

Set age is the strongest single property we can measure about how a Pokémon set's price behaves. Across 100 English sets whose recorded value series runs continuously from January 27, 2025 to their latest reading in September 2026, the median set aged between ten and fifteen years returned 190.3% while the median set under three years old returned 22.8%. The rank correlation between a set's age and its move over that window is +0.648. Three qualifiers come before the headline. It is one calendar window. The unit is the set, not the card. And the 100 sets are the ones whose series survives a continuity test, which removes 57 of 157 and removes old sets at roughly twice the rate of young ones, so this is a selected population and the selection runs along the same axis the study is measuring.

We hold a set-value series for 200 sets. A set qualifies here when it carries a recorded release date, a reading inside the first fortnight of the tape (January 27 to February 10, 2025), and a reading on or after September 1, 2026. That is 157 sets. A set value is the sum of the market price of every card we price in that set, so the series moves with card prices, with how many cards we price, and with any change to which printing of a card we quote. On 57 of those 157 the series contains at least one adjacent reading that moves more than 15% in one step, which on a whole-set index is the instrument changing rather than the market moving, so those 57 are excluded and 100 remain.

What those 57 steps are is worth stating precisely, because the temptation is to call them coverage and move on. On September 7, 2026 the value of Jungle fell from $2,833 to $1,100 in one reading, Gym Heroes from $4,534 to $2,012, Fossil from $3,370 to $1,184, Gym Challenge, Team Rocket and Neo Destiny the same way, while Base Set sat flat through the same days. The daily card tape covers that step, and on all six sets the number of priced cards is identical either side of it: 64 Jungle cards before and after, 132 Gym Heroes, 62 Fossil. No card joined and none left. What changed is which printing of each vintage card we quote, and it changed for every constituent on the same day. That is a change in the instrument, and a set whose series carries one cannot be read across it. The excluded sets skew old for the same reason: 48% of the cohort's sets aged ten or more survive the filter against 81% of the younger ones.

Two things follow from the window. The earliest anchor is January 27, 2025, so the youngest set that can qualify is about twenty months old, which means this study holds nothing at all about a set's first year. That ground belongs to The First 90 Days, The Launch Curve and Ten Sets, Measured From the Day They Actually Went On Sale, which restricts itself to the ten sets whose recorded series begins before their own release date. And every set here is measured over the same calendar window, so what is being compared is age at a moment, not a life cycle followed through.

The climb is steady for four buckets and then stops. Sets over twenty years old sit below sets aged fifteen to twenty, on four sets against nine, and neither cell is large enough to argue from. The honest reading of the right-hand end of that table is that we cannot resolve it. The left-hand end is the part with weight behind it: 13 sets under three years old, a median of +22.8%, and the only bucket where a set went backwards more than 25%.

Dispersion matters as much as the median. Every bucket from three years up has 100% or 89% of its sets above water over this window, which says the thing pulling these prices is broad rather than set-specific. The exception inside the fifteen to twenty year row is the BW Trainer Kit at -24.6%, a thirty-card trainer kit rather than an expansion.

A sum over a changing card list is a weak instrument for a change over time, so we built a second one that does not share the weakness. For every English card whose own weekly price series reaches back to February 10, 2025 and forward past September 1, 2026, we froze the constituent list at the window open and computed each set's index two ways: the median card move inside the basket, and the value-weighted move of the basket as a whole. That is 2,494 cards across 41 sets, once one row per TCGplayer product is kept, and 31 of those sets carry a release date. The baskets cannot grow, so nothing in this column is coverage.

The value-weighted column climbs with age without a break, from +37.0% for sets aged three to six to +170.8% for the single set over fifteen. The median-card column does not: the ten to fifteen year band sits below the six to ten year band. Two instruments built from different data agree on the direction and disagree on the smoothness, which is the correct level of confidence to carry out of this. Against the set-value series, the fixed basket also puts a number on the gap between them: on the 14 sets that clear the continuity filter and also carry a basket of ten or more cards, the set-value series reads a median 12.7 percentage points higher than the fixed basket over the same window.

Read the last column against the fourth. The sets released before 2020 are sitting at or within a few percent of the highest value their own recorded series has ever printed. The four released since 2023 are between 13.5% and 31.0% below theirs. That is the same finding as the age gradient seen from the other side, and it is the part of this study that a reader can check on any set page in a few seconds.

Three observations would break the finding. First, a second calendar window in which the age ordering inverts: if the next twelve months leave sets under three years old ahead of sets over ten, then what we measured was a 2025 to 2026 rotation into older cardboard and not an age property. Second, a fixed-basket index that stops agreeing with the set-value series on direction, which would mean the gradient lives in our coverage rather than in prices. Third, a set-by-set test that controls for price level: if the age effect disappears once you compare sets whose median card costs the same, then age is standing in for cheapness and should be described that way. That third test has been run here rather than left open. On the 58 sets in the clean cohort that also carry a median card price from the daily tape, the price level does correlate with the move, at +0.532, but the partial rank correlation between age and move with the price level held fixed is +0.507, barely below the raw age correlation on those same 58 sets. The age effect survives the control, which is a weaker claim than saying age causes anything and a stronger one than the raw correlation alone.

It cannot show a life cycle. Every set is measured over one shared calendar window, so a 2012 set and a 2024 set are compared at two different points of their own lives on the same twenty months of market. It cannot speak to the first year of a set at all, because the cohort's youngest member is twenty months old. Prices here are asks from the raw market, not sold medians, and a set index built from asks can hold a level that nothing is transacting at. The continuity filter removes 57 sets and it removes old ones at roughly twice the rate of young ones, so the cohort is selected on the same axis the study measures. The excluded sets are not quietly weak performers: measured on their own contaminated series they read far higher than the survivors, which is exactly why they cannot be used. What is left is the subset of old sets whose quoting did not change, and nothing here rules out that those are the old sets that were already being watched. Companion work published alongside this piece tests set age as one factor in a card-level cross section (What Predicts a Pokémon Card Return) and tests sealed box prices against years since a set left print (Do Sealed Pokémon Boxes Go Up in Value); a card-level or sealed result that points the other way is not a contradiction of this one, because the unit of analysis and the window are different, and any comparison should say which it is using.

Population: the 200 sets holding an owned `sethist2:` value series, joined to release dates in `report:spread:setsmeta` on a normalized set key. A set enters the cohort with a release date, a reading between 2025-01-27 and 2025-02-10, and a reading on or after 2026-09-01: 157 sets. It enters the continuous cohort when no two adjacent readings in its post-release window differ by more than 15%: 100 sets. Move is the last reading divided by the anchor reading minus one; every set in the cohort anchors on 2025-01-27, and windows run 600 to 603 days across the 100 continuous sets because they are not all read on the same day at the close. All-time highs and lows are windowed to readings dated on or after the recorded release date, never the first reading in the series. Medians take the upper middle value on even samples. Rank correlations average tied ranks. The fixed-basket index uses `ppthist:` per-card weekly series joined through `cards.tcg_id`, English only, constituents frozen at the first reading on or after 2025-02-10, requiring the card's series to reach past 2026-09-01 and the set to hold at least ten such cards: 2,494 cards in 41 sets, 31 dated. 240 English catalog rows share a TCGplayer product id with another row (trainer-kit halves and Trainer Gallery cards that sit in both the subset and its parent); one row per product id is kept throughout, preferring the catalog-native id over the synthetic `enx-` one and the more specific set row over its parent. Rank correlation is Spearman on set age in years against the move. The partial correlation quoted under falsifiability is the standard first-order partial of the three rank correlations among age, median card price and move, on the 58 continuous sets that also carry at least ten priced cards. Card prices come from the daily `price_history` tape as of 2026-09-22, which is the same figure the card pages publish; the per-card `ppthist:` series is used only for the fixed basket, because several hundred of those series have stopped updating. Prices are TCGplayer market asks for the flagship printing. Nothing here is investment advice.