Methodology · Sports Cards

How the quant scoring works.

Every PSA-graded card on the watchlist gets a 0-100 composite score that maps to a STRONG BUY → BUY → HOLD → TRIM → SELL verdict. This page explains every input, every weight, and every assumption.

The composite formula

COMPOSITE = 0.30 × MOMENTUM
          + 0.30 × VALUE
          + 0.20 × SCARCITY
          + 0.20 × LIQUIDITY

Each sub-score is normalized to 0-100. The 30/30/20/20 weighting reflects the relative importance of each input for sports cards specifically:

Momentum (30%) — Sports cards move fast on news / performance. Trend matters more here than for vintage paintings.
Value (30%) — Where the live price sits between the buy zone and sell zone. The most actionable input.
Scarcity (20%) — Pop count at the exact grade. Drives long-run institutional bid.
Liquidity (20%) — Recent eBay sold count. Determines if you can actually exit.

MOMENTUM (0-100)

Two inputs combine here:

A hand-curated trend30d in the seed (decimal, e.g. 0.15 = +15% over 30d).
A cross-check vs the live eBay median: if live is +X% above anchor, momentum gets a +X-point bump.

score = 50 + 250 × trend30d + 100 × ((live_median − anchor) / anchor)
score = clip(score, 0, 100)

So a +15% hand-curated trend with a live median +10% above anchor → 50 + 37.5 + 10 = 97.5. A flat card with no live data → 50.

VALUE (0-100)

Where does the live (or anchor) price sit between the buy zone and sell zone?

pct = (t_sell − price) / (t_sell − t_buy)
score = 100 × clip(pct, 0, 1)

At the buy zone: 100 (perfect entry). At the anchor: roughly 50. At the sell zone: 0. Below the buy zone or above the sell zone, the score saturates.

SCARCITY (0-100)

Bucketed by PSA pop count at the exact grade:

Pop range	Score	Tier interpretation
< 100	95	Institutional grail tier
100 - 500	80	Tight float
500 - 1500	60	Standard high-grade scarcity
1500 - 3000	40	Mass-graded
3000 - 6000	25	Commodity-adjacent
≥ 6000	15	Pure commodity

Pop data is seeded manually per card (Cloudflare blocks scraping PSA's pop site). Refresh quarterly from psacard.com/pop.

LIQUIDITY (0-100)

Based on the number of recent eBay sold-comps that match the card's query:

n_samples	Score	Exit ease
≥ 8 / month	95	Very liquid — same-day sale possible
5 - 7	80	Liquid — sells within a week
3 - 4	60	Standard — few weeks
2	40	Thin — patient seller needed
1	20	Very thin — illiquid
0	10	No comps — pricing uncertain

Conviction adjustment

Each card has a conf rating 1-5 reflecting how much I trust the underlying thesis (player career arc, set significance, market structure). After computing the composite:

conf == 5: composite gets +5 points
conf ≤ 2: composite gets −5 points
Otherwise: no adjustment

Verdict ladder

Composite	Verdict	Action
≥ 85	STRONG BUY	Max-conviction entry. Half-Kelly stake.
70 - 84	BUY	Standard entry. Sized normally.
50 - 69	HOLD	Maintain existing positions; don't add.
30 - 49	TRIM	De-risk: cut position 30-50%.
< 30	SELL	Full exit. Capital rotates to higher-composite picks.

Position sizing — half-Kelly with hard caps

Given a bankroll, an assumed edge (your conviction in the EV), the recommended position size:

fraction = 0.5 × (edge / 100)
fraction = min(0.02, max(0, fraction))
stake$ = bankroll × fraction

Two hard caps:

Per-position cap: 2% of bankroll — no single card > 2% regardless of EV.
Sport cap (informal): 50% of bankroll per sport — diversify across BB / FB / BK / etc.

Data refresh cadence

Live eBay sold-comps: weekly Sunday 22:00 UTC via GitHub Action. Median over outlier-rejected samples. Browse-API fallback bypasses Akamai.
PSA pop: quarterly manual update (next: 2026-07-15).
30d trend: hand-curated; updated when major events warrant.
Watchlist additions: ad hoc as new rookies enter the market.
Catalysts calendar: reviewed monthly. 14 events tracked for 2026.

Tax treatment (US)

Sports cards are collectibles under IRC §408(m)(2)(A):

Long-term capital gains (≥ 1 year held) are capped at the 28% rate — higher than the 20% standard LTCG cap.
Short-term gains are taxed at ordinary income rates (up to 37%).
Wash sale rules don't apply to collectibles (unlike securities).
The desk's tax_report.py generates a Schedule D summary and Form 8949 row-by-row from sales.json.

Not tax advice. Consult a CPA before filing.

Caveats and known limits

Anchor prices are point-in-time. They reflect baseline conditions when the watchlist was seeded. Live prices override.
The eBay scraper can be blocked by Akamai from data-center IPs. Set EBAY_OAUTH_TOKEN for the Browse-API fallback.
PSA pop is manually-seeded. It drifts over time as more cards get graded. Refresh quarterly.
Composite is a heuristic, not a guarantee. Even high-composite cards lose. Sample size matters.
Sealed product uses different scoring (trend + risk-adjusted upside) — see sealed_products.py.
Parallels use modified scoring (30/35/35 — momentum/value/print-run-scarcity) since print runs are a stronger signal than pop counts.

References

Source code: github.com/Keyvaniath/bpleone-sports-cards-desk
Public API: /sports-cards/api/
CHANGELOG: repo CHANGELOG.md
Original Pokémon TCG framework that this mirrors: pokemon.bpleone.com