The level-correlation audit
Which of the published FRED correlations survive when the series are differenced instead of correlated on levels?
Several published correlations collapse on changes; the home-prices-to-confidence figure goes from -0.89 on levels to -0.04 on changes (p = 0.66). Supersedes the FRED table in the 2025 validation log.
Bears on: The Static State of Change · Live dashboard
Audit · 2026-09-11
docs/research/FRED_LEVELS_VS_CHANGES_2026-09-11.mdThe level-correlation audit: what survives when you difference the series
Date: 2026-09-11
Status: COMPLETE — supersedes the FRED table in COSMIC_CLAIMS_VALIDATION_LOG_2025-12-15.md
Data: FRED (St. Louis Fed) via api.stlouisfed.org, monthly averages, same series IDs and
windows as src/app/api/research/validate-claims/route.ts
Replication: scripts/research/fred_levels_vs_changes.py — public series, no proprietary data.
Why this was run
Dispatch #13 (The Dam Release, Sep 12) teaches the reader a rule:
Compare the raw levels of two series and you get a correlation near 0.9, which tells you nothing at all: both have trended upward for decades, so they would agree even if neither had ever touched the other.
The dashboard was breaking that rule the day before the dispatch taught it. validate-claims
correlates raw monthly levels — there is no differencing, no detrending, no transformation
anywhere in the route — and the seven results were published on the live dashboard under the
heading VALIDATED CORRELATIONS. The validation log's FRED table states no transformation for
any pair, which is how this survived a year: nobody reading it could tell what had been measured.
A physics or finance professor pulling the same series would find this in an afternoon. The NMSU faculty emails point at that dashboard this week.
Result
Levels reproduce the archived figures to within 0.02 across all seven pairs, which confirms the archive was measuring levels. On changes, four of the seven collapse.
| Pair | r (levels) | r (changes) | ex-COVID | YoY | Verdict on changes |
|---|---|---|---|---|---|
| GDP ↔ Consumption | +0.995 | +0.931 | +0.653 | +0.903 | ✅ Survives — but see note |
| Fed Funds → 2Y Treasury | +0.964 | +0.476 | +0.413 | +0.519 | ✅ Survives |
| Consumer Credit → Employment | +0.940 | +0.250 |
Civilizational Science
Four preregistered tests of this thesis have come back unsupported, and all four are public. The dispatches carry the same standard out to energy, food and money. Free, and it stays free.
No charge, and no paywall on the dispatches. Unsubscribe in one click.