Does Momentum Work in Pokémon Cards? 28,288 Observations Say Yes, With a Catch

Most Pokémon investing advice is anecdote. We hold recorded daily price curves for roughly 7,000 cards, so we did the thing anecdote cannot: a point-in-time backtest. For each card at each month-boundary in its history, we computed its trailing 90-day return using only data available at that date, then measured what it actually did over the next 90 days, scored against the average card in the same window. That is 28,288 card-month observations across 3,298 cards. The question it answers is the one every buyer asks without a way to check the answer: does a card that has been going up keep going up?

The pattern is monotonic through the middle and it is strong: a card up 25 to 50 percent over the prior quarter beats the average card by 4.7 points over the next quarter and wins nearly half the time, against a base rate far below that. The mirror image is the flat card, and it is brutal: the -3 to +3 percent bucket is the worst of the momentum buckets, lagging its cohort by 4.0 points with a 23 percent win rate. (Two other cuts of the same data are worse still, the lowest-volatility and the priciest cards, both near -5.3; more on the first below.) A tape that is not moving is not resting; it is losing.

The top bucket is where naive momentum gets you killed, and only in one place. Split the up-more-than-50-percent cards by era and the story divides cleanly: older cards that have run keep running, while modern-era cards (Sword & Shield and Scarlet & Violet) already up more than 50 percent reverse hard, losing 10.6 points of alpha over the next 90 days across 461 observations with a 24.5 percent win rate. New product prints; a modern parabola is buying into supply that has not finished arriving. A vintage run has no such overhang. This single result is why our pick engine's overextension cap is now era-conditional rather than a flat rule, and it is the mechanism behind the modern crashes covered in Launch, Mania, Markdown.

Two cards can both be up 40 percent over 90 days and be completely different bets. One climbed in steady monthly steps; the other was flat for eleven weeks and jumped in one. We tested the difference. Within the 15-to-50-percent momentum band, cards whose three consecutive monthly medians each rose (a stair) beat their cohort by 4.8 points at a 50.4 percent win rate; the same-momentum cards without that shape managed 3.7 points and 44 percent. Measured from concentration instead: moves where no single 30-day window held more than half the run returned +6.5 points, versus +2.2 for moves where one window held more than 80 percent. Smooth beats lumpy. This is the compounder shape Base Set Charizard traced on its way up, and the reason our engine now scores it as its own signal.

Population: recorded per-card daily price curves with at least 20 points (7,053 cards). Eval dates every 30 days require 104 days of prior history and 90 days of forward history, giving 28,288 card-month observations across 3,298 cards, June 2025 to June 2026. Price at a date is the last recorded point within 13 days; every signal is computed strictly from points on or before the eval date, every forward return strictly after, so there is no lookahead. Alpha is the forward return minus the equal-weight mean of all cards evaluated in the same 30-day window (14 cohorts), which nets out the sample's valuable-set skew in level. Win rate is the share of observations with positive alpha. No signal is reported without at least 200 observations and a consistent sign across the 30/60/90-day windows. The deep curves cover bought sets and skew toward the SM/SWSH eras; see the boundary note.