What in the US economy moves the card market?
1. How the test was run (and why most "correlations" would be fake)
- Card side: three monthly series from our ledger. The '86 Fleer Jordan #57 in PSA 8 / 9 / 10 (monthly median sale; PSA 9 is the most liquid, 3,365 sales). A blue-chip basket standing in for Card Ladder's index: Card Ladder blocks automated access to its CL50 values, so I built an equal-weight index from the CL50 members we hold plus the next most-traded basketball blue chips: '86 Fleer Jordan (8/9/10), '96 Topps Chrome Kobe (9/10), '96 Topps Kobe, '92 Topps Shaq 10, '80 Topps Bird/Magic (8/9), '96 Topps Chrome Iverson, '89 Hoops & '90 Fleer Jordan, '97 Topps Chrome Duncan — 16 card+grades. And the Hobby index from the market board (every card, repeat-sales).
- Economy side: 36 FRED series: equities (S&P, Nasdaq), Bitcoin, VIX, M2, the Fed's balance sheet, Fed funds, 2y/10y yields, real rates, credit spreads, CPI / core CPI / PCE prices, consumer sentiment, inflation expectations, savings rate, real income, government transfers (stimulus), unemployment, jobless claims, payrolls, mortgage rates, home prices, housing starts, retail sales, consumer spending, consumer credit, card delinquencies, used-car prices, oil, the dollar.
- Changes, not levels. Everything was compared as 3-month changes (card prices are too noisy month to month). Comparing levels of two trending series produces huge, meaningless correlations, so none are reported.
- Lead/lag 0–12 months, 2,854 tests in all. With that many tests, a correlation needs |r| ≳ 0.48 just to clear a Bonferroni bar, so every candidate also had to pass three more checks: (a) same sign and |r| ≥ 0.25 in both halves (2020-08→2023-08: boom and crash; 2023-09→2026-10: trough and the new boom); (b) hold up on rank correlation and with every 2020 observation removed (the COVID spike dominates ordinary correlations); (c) show a directional asymmetry: stronger when the economy leads cards than when cards lead the economy.
- The honest limit: six years is one full cycle. With a 9-month lead and 3-month windows there are only a handful of independent observations. Treat what follows as the best reading of one cycle, not a law.
2. The leading signal: payroll growth, inverted, ~9 months ahead
Robustness
| Card series | r, all data (Pearson) | rank r, ex-COVID | when cards lead by 9 instead | first half | second half |
|---|---|---|---|---|---|
| '86 Fleer Jordan PSA 9 | −0.84 | −0.72 | −0.21 | −0.89 | −0.60 |
| Blue-chip basket (CL-style) | −0.78 | −0.83 | −0.21 | −0.84 | −0.83 |
| Hobby index (all cards) | −0.61 | −0.83 | −0.71 ⚠ | −0.50 | −0.46 |
The asymmetry column is the tell. For the Jordan and the blue chips, the relationship is strong when payrolls lead and weak when cards lead, which is what a genuine leading signal looks like. For the broad Hobby index it is strong in both directions, so there it is two series riding the same post-COVID arc rather than a lead. This is a blue-chip phenomenon.
What the regimes looked like (2021 onward, 9-month lead, next-3-month card change)
| Payroll growth 9 months earlier (3m, annualized) | Jordan PSA 9 | up | Blue-chip basket | up | Hobby index | up |
|---|---|---|---|---|---|---|
| Weak: under 1%/yr | +14.5% | 95% | +16.9% | 100% | +15.0% | 95% |
| Middling: 1–2%/yr | +3.3% | 70% | −2.3% | 41% | +0.1% | 52% |
| Strong: over 2%/yr | −7.3% | 19% | −8.5% | 16% | −5.9% | 4% |
Average three-month change and the share of windows that rose; 20–24 windows per row.
As a timing rule, out of sample
Rule: hold cards only when payroll growth nine months earlier was under 1.5%/yr. Scored on non-overlapping quarters, with the threshold chosen before looking at the second half.
| Period | Series | Rule | Buy & hold | Quarters long |
|---|---|---|---|---|
| 2021-10 → 2023-08 (the crash) | Jordan PSA 9 | +10% | −38% | 1 of 7 |
| Blue-chip basket | −9% | −67% | 1 of 8 | |
| Hobby index | −6% | −47% | 1 of 8 | |
| 2023-09 → 2026-10 (the recovery) | Jordan PSA 9 | +123% | +114% | 9 of 13 |
| Blue-chip basket | +130% | +101% | 9 of 13 | |
| Hobby index | +113% | +110% | 9 of 13 |
The value of the signal was in sitting out the 2022–23 crash (strong post-reopening hiring in 2021–22 said "stay out"), and it was back in for most of the 2025–26 boom. In the recovery it only matched buy-and-hold: a leading signal of this kind helps you avoid the fall far more than it helps you time the rise.
Why it would make sense
Weak hiring is what moves the Fed. Softer jobs → rate cuts and easier money → asset prices, with collectibles at the speculative end of that chain and the slowest to react. Strong hiring runs the other way: 2021–22's reopening boom brought the hikes that drained the card market in 2022–23, nine months after the jobs data had already turned hot. The Fed funds rate change itself shows the same sign at the same lead (−0.3 to −0.4 ex-COVID), weaker because policy reacts to jobs with a lag of its own. One clean cycle supports this story; it does not prove it.
3. The concurrent signal: core inflation, inverted
| Core CPI (3m annualized), same quarter | Jordan PSA 9 | up | Blue-chip basket | up | Hobby index | up |
|---|---|---|---|---|---|---|
| Under 2.5% | +17.8% | 84% | +27.6% | 84% | +12.5% | 81% |
| 2.5 – 4% | +8.2% | 73% | +5.3% | 65% | +6.4% | 69% |
| Over 4% | −11.9% | 24% | −15.0% | 17% | −9.0% | 6% |
This is not a lead (it peaks at 0–1 months), but it is the cleanest concurrent mirror of the market and it contradicts the "cards are an inflation hedge" story: in this cycle, accelerating inflation coincided with falling cards, almost without exception.
4. What did not work
- M2 money supply (the usual "liquidity" theory): +0.45 on raw data at a 9-month lead, but rank correlation −0.03, and −0.17 to −0.24 once 2020 is excluded. The entire correlation was the spring-2020 money surge sitting nine months before the 2021 boom. A real-M2 timing rule matched buy-and-hold and nothing more.
- Stocks (S&P, Nasdaq): no lead at any horizon; concurrent correlation only +0.15 to +0.3. Cards did not follow the stock market in this cycle (stocks rose through 2023–24 while cards sat at the trough).
- Bitcoin: moves with blue-chip cards a little (+0.42 to +0.46 concurrent, rank +0.2), consistent with shared risk appetite, but gives no advance notice.
- Fed balance sheet, real income, stimulus payments, consumer sentiment, unemployment claims: impressive raw numbers that collapsed to near zero without the 2020 observations.
- Mortgage rates, 10-year yields, savings rate, home prices, used-car prices, credit delinquencies: weak or unstable across the two halves.
5. Where the signals stand today
Read together: the leading signal says the tailwind that produced 2025–26 remains in place into mid-2027. The concurrent signal says watch core inflation: if it re-accelerates past 4%, that has coincided with falling prices every time in this data, lead or no lead. The first real test of the payroll signal as a top call will come when hiring re-accelerates above ~2%/yr; it has not yet done so.
6. Caveats, plainly
- One cycle. ~60 usable months after removing 2020, perhaps 5–6 independent observations at a 9-month lead. A relationship this strong over one cycle is suggestive, not established.
- Our universe is basketball-heavy (Jordan alone is a large share of volume) and the "Card Ladder" basket is our approximation, not their published index.
- Card Hedge's feed ended Oct 12 2026; since then only our own eBay captures arrive, so the latest card months are thinner.
- The threshold (1.5%/yr) and the 9-month lead were chosen on this data. The out-of-sample half is a fair test of the rule; it is not a fair test of the choice of payrolls among 36 series.
Sources: FRED series PAYEMS, UNRATE, FEDFUNDS, CPILFESL, M2SL, SP500, NASDAQCOM, CBBTCUSD and 28 others (pulled Nov 5 2026); price_ledger.sqlite; market_board Hobby index. Method and code: docs/macro-study/analyze.py (card series in cards.json; FRED CSVs re-downloadable by series id).