2-Year Yield × Russell 3000 — The Divergence, Back-Tested
The empirical backbone for the Macro Divergence trade: is the rate–equity decoupling a tradeable rubber band, or a trend artifact of two trending series? Rolling correlation, a divergence-Z oscillator & the reversion base rate · Data 2016–2026 · Updated August 31, 2026 · 6:49 PM ET
Corr ρ · 3-yr (levels)
reproduces the study's −0.76
Corr · changes (full)
+0.10
the honest day-to-day co-movement
Divergence Z · now
+1.95σ
richly decoupled
Decouplings ≥+1.5σ
since 2020 · reversion base rate below
📄 Source blueprint (PDF) 🟢 The live trade
The source study rests its whole case on one number — a −0.76 correlation between the 2-year Treasury yield and the Russell 3000 (IWV) — and asserts that because it "fractured" since March, it must snap back. That is the one claim the PDF never tests. This page does. We reproduce the correlation, separate the honest signal from the trend artifact, and measure what actually happened after prior decouplings of this magnitude.
First, the number is fragile. The −0.76 is a levels correlation — IWV's price against the yield level — and two trending series (equities up, yields down through 2023–24) will correlate strongly without any tradeable relationship between them. The honest measure, the correlation of daily changes (returns vs yield moves), is just +0.10 over the full sample — far weaker. A "fracture" in a trend co-movement is not, by itself, a coiled spring.
1 · Rolling 1-year correlation — levels vs changes
The levels correlation (purple) swings hard — deeply negative when both series trend, flipping positive in decoupling regimes like the present. The changes correlation (grey) hugs zero throughout: day-to-day, IWV and the 2-year simply do not move together reliably in either direction. The study's headline lives entirely in the volatile purple line.
2 · The divergence-Z oscillator — how far equity sits above the rate-implied level
A rolling regression of log(IWV) on the 2-year yield gives a "rate-implied" equity level; the residual, z-scored, is the divergence. Z > 0 = equity rich versus what rates imply (the current, red-shaded state at +1.95σ); Z < 0 = cheap. Gold markers flag each prior cross above +1.5σ — the same decoupling the trade is fading.
3 · The reversion base rate — what happened after a ≥+1.5σ decoupling (n=21)
HorizonIWV median% downall-days baseΔ 2-yeargap narrowedread
+21d+0.9%33%+1.8%+2 bps67%below baseline (reverts)
+42d+0.8%40%+3.4%+2 bps70%below baseline (reverts)
+63d+1.3%47%+4.6%+8 bps74%below baseline (reverts)
+90d-0.3%50%+6.4%+12 bps67%below baseline (reverts)
After equity decoupled richly above rates, forward IWV returns are compared to the all-days baseline. % down = how often the equity leg fell (the trade's thesis); Δ 2-year = the median forward move in the yield (the other way the gap can close); gap narrowed = how often the divergence Z shrank toward zero. This is the number a position-sizer needs — and it is measured, not asserted.
The honest read. The rubber-band framing is half right. The divergence Z does mean-revert — the gap narrowed by 63 sessions in 74% of the 21 prior cases — and it runs almost entirely through the equity side. After a rich decoupling, IWV lagged the all-days baseline by roughly 3 points at 63 sessions (median +1.3% vs +4.6% for any day, down 47% of the time against a market that is usually higher), while the 2-year barely moved (+8 bps median). The gap closed because equity gave back its excess — not because yields normalized. That reverses the source's emphasis. The bearish-IWV leg (Leg 1) is the empirically supported half of the structure; the long-SHY / yield-normalization leg (Leg 2) has little historical backing — the 2-year did not reliably fall after prior decouplings of this size. The market-neutral wrapper still earns its keep (some of the equity drag is just beta), but the edge lives in the equity leg — and the 4.50% weekly-close yield stop is exactly the right rail, precisely because the rate side is the unreliable one. This is one pair, in-sample, n=21: a base rate, not a guarantee.
Prior decoupling events & their forward paths
Cross >+1.5σZIWV +21d+42d+63d+90d
July 30, 2026+1.60σ+3.6%
June 17, 2026+1.81σ+0.5%+4.0%
May 15, 2026+1.58σ+2.0%+1.5%+5.6%
January 28, 2025+1.52σ-3.8%-8.4%-8.6%-1.2%
December 24, 2024+1.52σ+0.9%-2.9%-7.6%-6.7%
November 18, 2024+1.57σ-0.5%+3.4%+1.8%-4.7%
October 16, 2024+1.53σ+0.9%+4.1%+4.3%+2.1%
June 27, 2024+1.50σ+0.3%+3.2%+5.5%+6.6%
May 15, 2024+1.67σ+1.5%+5.1%+4.1%+8.0%
December 1, 2023+1.59σ+2.7%+7.8%+10.8%+11.7%
August 30, 2023+1.57σ-4.9%-7.4%+1.3%+6.6%
July 12, 2023+1.50σ-0.2%+0.2%-2.3%+0.6%
April 13, 2022+1.57σ-9.9%-16.1%-13.4%-6.9%
March 2, 2022+1.52σ+3.0%-5.7%-7.2%-13.4%
Methodology. Daily IWV (adjusted) and the U.S. Treasury 2-year par yield, 2016–2026 (2,662 aligned sessions). Rolling window 252 sessions for the correlation; the divergence regression fits log(IWV) on the 2-year yield over a 756-session (3-year) window — long enough to hold the durable negative relationship rather than adapt to the decoupling — with the residual z-scored over 252 sessions. Events = the residual Z crossing up through +1.5σ, deduplicated to ≥21 sessions apart. Forward returns are close-to-close; the all-days baseline is the median IWV forward return over every session. In-sample, single-pair, no multiplicity correction — descriptive, not inferential. The 2-year is the Treasury par yield (the study cited TVC:US02Y, a hair higher).