Gold–Risk Asset Decoupling — Pre-registration

The thesis

Gold and equities co-move because they share a reaction function: both price market-implied rate expectations and real-yield dynamics — cut expectations lift both, hike expectations drag both. Our claim is conditional: once elevated inflationary pressure — driven primarily by energy and commodity supply shocks — becomes entrenched (persistent, feeding core and services inflation, de-anchoring expectations, forcing sustained restriction), that shared function breaks down: gold responds to inflation persistence directly, as a store of value, while equities stay dominated by the discount-rate and earnings channels of tightening.

Gold’s post-decoupling direction is unspecified; the claim is the correlation breakdown itself — conditional and regime-switching, not that gold always or never co-moves. Two parts, scored independently: (a) entrenchment, ACTIVE under the composite below, and (b) collapse of gold–equity correlation while (a) is active. Decorrelation without entrenchment does not confirm — it falsifies (condition F2). Confirmation requires positive causal evidence: the active condition, an observed break, a plausible transmission chain — not a liquidity artifact.

What would prove it wrong

The instrument, fixed for the thread’s life: the 60-day rolling Pearson correlation of daily simple returns of GLD against SPY, from public daily ETF prices. Entrenchment is a three-leg composite, ACTIVE only when all legs hold simultaneously: (i) core PCE YoY above 3.0% for 6+ consecutive months (FRED PCEPILFE); (ii) Brent sustained above $75/bbl over the same window (FRED DCOILBRENTEU); (iii) second-round effects — at least one of services inflation accelerating, wage growth above 4% YoY, or the 5-year breakeven above 3.0% (FRED T5YIE). Episodes report it ACTIVE, PARTIAL, or INACTIVE.

Falsification requires all of F1–F4 to fail. F5 is component-level: it adjudicates the structural-demand leg, not the thesis; a verdict flip mandates reassessing F3’s weight, and its firing strengthens rather than weakens the macro-asset reading. Failure to falsify is not confirmation; the thesis merely survives. One adjustment is registered: the correlation leg alone would trip F1 under PARTIAL entrenchment, so F1 requires the full composite ACTIVE — hence F1 is partial at registration: the +0.40 leg is already live (79 sessions), while the entrenchment condition has not fired.

Where things stand

As of registration the instrument reads coupled and elevated: the 60-day GLD/SPY correlation is +0.487, its 79th consecutive session above +0.40 (last break 2026-04-29), while core PCE has held above 3.0% YoY for six months and Brent above $75 (computed from public daily GLD/SPY data, 2026-08-22; FRED PCEPILFE, DCOILBRENTEU).

The evidence for: gold sits about 13% below its 2026-01-29 peak while SPY hit a new high 2026-08-13, and the 30-day correlation (+0.263) sits well below the 90-day (+0.538) — coupling weakest in the freshest window (same computation). Official demand looks price-insensitive: record net Q2 purchases of 289t (+62% y/y) accumulated through a roughly 16% intra-quarter gold decline (WGC Gold Demand Trends, 2026-07-30); Poland bought 82t in 2026 toward a 700t target (NBP communications, July 2026); the PBoC made its 21st straight monthly addition in July (Caixin Global, 2026-08-08).

Against: entrenchment is PARTIAL, not ACTIVE — leg (iii) fails all three sub-tests (core CPI 2.5% YoY and decelerating, June 2.6% to July 2.5%, BLS; average hourly earnings 3.2% and falling, BLS; 5-year breakevens 2.34%, FRED T5YIE). The dollar’s COFER share rose from 56.42% to 57.13% in Q1 largely on valuation (IMF COFER, Q1 2026 brief) — one of F5’s two required flat-to-rising quarters. The structural bid is decelerating: Q1 was revised from 244t to 57t, leaving H1 2026 at 346t — the weakest first half since 2022, roughly 692t annualized, below F5’s own 700t floor (WGC, 2026-07-30).

The counter-case, at full strength, has three moves. First, the premise may be a fiction: the 60-day statistic has swung from −0.596 to +0.804 over two years around a +0.157 mean, 16% of last-year sessions negative (same computation) — a noisy, mean-reverting variable, not a regime; if decorrelation is constant and entrenchment rare, the conjunction will look confirmed by chance. Second, the March 2026 Iran-strike shock — genuine supply-shock stagflation — was the cleanest natural test available, and gold stayed coupled: the 60-day floor only +0.117, never negative; later breaks are more parsimoniously explained by positioning or liquidity artifacts, and whether the January–April correction was positional or fundamental remains open. Third, entrenchment itself is retreating (core CPI, wages, breakevens all falling): the thesis may run out of preconditions before its window opens. Registered response, not resolution: the +0.40 leg is a persistent 79-session state under genuinely elevated inflation legs, not a random draw, and every correlation reading is conditional on entrenchment status. As registered, the counter-case outweighs the thesis — hence low confidence.

The backdrop sharpens the test: the Fed held July 28–29 under Chair Warsh with hawkish rhetoric (Federal Reserve, 2026-07-29; effective rate 3.63%), and federal funds futures imply a hold at the September 15–16 meeting (CME FedWatch, through 2026-08-21) — not the thesis’s rate-cut regime; F1-type coupling faces a hold-or-hike test. Catalysts ahead, as events not outcomes: July core PCE (BEA, due 2026-08-26) extends the streak or opens F2; August CPI (BLS, 2026-09-11) re-tests leg (iii); the FOMC decision (2026-09-16) tests whether the elevated state survives; WGC Q3 tonnage, the COFER Q2 brief, and the September jobs report (late October and ~2026-10-02, estimated) settle F4–F5; Iran/Hormuz developments drive Brent leg (ii).

How to read this series

Each episode carries the full ledger forward, met conditions included; amendments appear only as dated revised entries; the series closes with a sunset ledger and a terminal verdict. Survival is not confirmation: falsification needs all of F1–F4 to fail, while confirmation needs the causal conjunction — decorrelation while entrenchment is ACTIVE. The cadence is irregular, keyed to the release calendar; the next assessment is planned by 2026-09-30.