{
  "url": "https://www.moonstonehq.com/articles/we-were-wrong-about-momentum",
  "slug": "we-were-wrong-about-momentum",
  "title": "Momentum Does Not Persist in Pokemon Cards: 65,685 Observations Say So",
  "summary": "On 65,685 observations across 52,041 cards and a 729-day span, a card's recent strength tells you nothing usable about its next move. The effect that shows up on a smaller sample does not survive the bigger one, which is the ordinary fate of momentum findings and the reason we re-run them at scale before acting on anything. The full method, the sample and the counter-result are below.",
  "section": "Quant research",
  "published": "2026-09-07",
  "readMinutes": 8,
  "publisher": "moonstone",
  "findings": [
    {
      "value": "65,685",
      "measures": "point-in-time observations",
      "basis": "up from 28,288"
    },
    {
      "value": "52,041",
      "measures": "cards measured",
      "basis": "up from 3,298"
    },
    {
      "value": "+7.36 pts",
      "measures": "forward 90d vs cohort, tapes down more than 10%",
      "basis": "the best bucket; +22.60% mean and +7.41% median in absolute terms"
    },
    {
      "value": "-2.00 pts",
      "measures": "forward 90d vs cohort, flat tapes",
      "basis": "the worst bucket; +0.56% median in absolute terms"
    }
  ],
  "tables": [
    {
      "title": "Forward 90-day return against the same-window cohort, by trailing 90-day return, 65,685 observations",
      "columns": [
        "Trailing 90d",
        "Observations",
        "Forward 90d vs cohort",
        "Beat the cohort"
      ],
      "rows": [
        {
          "Trailing 90d": "down more than 10%",
          "Observations": "7,349",
          "Forward 90d vs cohort": "+7.36%",
          "Beat the cohort": "40.5%"
        },
        {
          "Trailing 90d": "down 3 to 10%",
          "Observations": "6,890",
          "Forward 90d vs cohort": "-3.67%",
          "Beat the cohort": "31.6%"
        },
        {
          "Trailing 90d": "flat, within 3%",
          "Observations": "15,935",
          "Forward 90d vs cohort": "-2.00%",
          "Beat the cohort": "26.5%"
        },
        {
          "Trailing 90d": "up 3 to 10%",
          "Observations": "10,394",
          "Forward 90d vs cohort": "-1.18%",
          "Beat the cohort": "34.6%"
        },
        {
          "Trailing 90d": "up 10 to 25%",
          "Observations": "12,905",
          "Forward 90d vs cohort": "+0.23%",
          "Beat the cohort": "39.7%"
        },
        {
          "Trailing 90d": "up 25 to 50%",
          "Observations": "7,832",
          "Forward 90d vs cohort": "+0.65%",
          "Beat the cohort": "42.4%"
        },
        {
          "Trailing 90d": "up more than 50%",
          "Observations": "4,379",
          "Forward 90d vs cohort": "+1.66%",
          "Beat the cohort": "40.4%"
        }
      ]
    }
  ],
  "charts": [],
  "sections": [
    "The curve is a U, and the bottom of it is where most cards sit",
    "The first study measured a survivor pool and called it the market",
    "The number that decides whether any of this is tradeable",
    "Three things we removed or reversed today"
  ],
  "method": [
    "Point-in-time study over our recorded daily price curves unioned with the canonical daily price accretion, 52,041 cards, span 729 days, as of 2026-09-08. For each card, evaluation points are taken every 30 days where 104 days of prior history and 90 days of forward outcome both exist, giving 65,685 observations. The trailing signal uses only data dated at or before the evaluation day. Forward return is measured from the evaluation day to a window centred 90 days later. The bucket table reports the mean EXCESS return against the mean of the same evaluation window, and its final column the share of the bucket whose excess return was positive; both are relative measures whose population value is zero by construction. Absolute forward returns and the share clearing the 24.14 percent round-trip cost are quoted separately and are the only figures comparable to a cost. The cost floor itself is (1 + 0.08) / (1 - 0.13) - 1, the ask spread and the exit fee the pick engine ships, and the code is src/server/pick-signals.ts. The independent cross-check described above computes returns relative to the mean and median of the same evaluation window and bootstraps confidence intervals clustered by card, so that many observations from one card cannot masquerade as independent evidence. The prior study, retracted here, is left published at its original URL with a link to this piece rather than edited, because a research record that quietly rewrites itself is not a record."
  ],
  "terms": "Figures are moonstone's own recorded measurements, dated as published. Quote them with the population stated beside them. The article at `url` is the human form of this file."
}