US Treasury Yield Repricing — Episode 2: The De-Escalation Test

What changed since episode 1

In episode 1 we judged the 5.00% breach partly a war artifact, moved the crux to attribution — durable repricing versus reversible oil premium — and pre-registered the line: a 10-year holding ≥4.80% through an oil retracement would be the cleanest structural-floor evidence yet. That test passed. The breach faded to 4.94–4.96% (2026-09-17→22); Brent fell to $99 on Iran’s Hormuz-reopening proposal (Kyodo, via Trading Economics, 2026-09-22) — and the 10-year held above the line, then broke out on non-oil drivers to 5.11% (FRED DGS10, 2026-09-23) and 5.18% (Treasury par curve, 2026-09-24; FRED lags a day). The 30-year hit 5.47% on the same curve, highest since 2004 (Bloomberg, via Macro Roundup), its ≥5.00% streak at 57 sessions (FRED DGS30). The ambiguity moved: episode 1’s lone soft 20-year generalized into a tenor-band concession stopping short of the 10Y/30Y flagships — untested until mid-October — while a new risk emerged: intervention capping the repricing.

The evidence both ways

Auctions flipped from feast to concession: the 5-year (2026-09-23) cleared at 5.033% with bid-to-cover 2.212 (prior 2.371), indirect 54.3% (prior 61.5%), and a tail among the widest on record; the 7-year (2026-09-24) priced above when-issued at 5.085%, highest since April 1993; the 20-year (2026-09-15) printed 52.1% indirect and a +6.0bp tail — widening tails and retreating indirects at two consecutive coupon auctions, the microstructure regime floor predicts and mean reversion cannot (TreasuryDirect auction records). July TIC (Treasury TIC Table 5) measures the official-sector erosion: Japan −$12.8B on the month, −$106B since April; China −$15.4B; foreign officials −$236B. The BOJ hiked to a 31-year high, JGB 10-year at 3.055% (Reuters, 2026-09-24); hyperscaler USD issuance passed $180B YTD, 41% beyond 15 years (Twenty Four AM, 2026-09-03). Part (b) channels widened: mortgages crossed 7.03% (Freddie Mac PMMS, 2026-09-24); the 20-session daily 10Y/S&P correlation hit −0.679 (window 2026-08-26→09-23, computed from FRED DGS10/SP500).

The counter, at full strength: we saw the most technically explicable auction week of the quarter and called it a regime. The 2Y/5Y/7Y printed 2026-09-22/23/24 into a vertical tape — the 10-year up 15bp-plus on a Houthi spike to Brent $108, dealers short (Trading Economics, 2026-09-22/24) — month-end mechanics, not demand destruction; flagship auctions last ran 2026-09-09/10 at ~79% indirect (TreasuryDirect). The breakout high printed on oil-spike days — as war-contaminated as the print episode 1 flagged; a completed deal could retrace part of the level (no calibrated estimate). Risk assets improved while the 10-year added 18bp: S&P 500 recovered to −1.22% off its high, IG/HY spreads tightened (FRED SP500/BAMLC0A0CM/BAMLH0A0HYM2, 2026-09-24). A part (b) that survives two decisive part-(a) confirmations without moving should itself temper conviction that (a) is structural — real fiscal repricing leaks everywhere; −0.68 is co-movement, not stress. TIC’s own notes flag imprecise custodial attribution — the UK gained $58.4B in the same July table — so Japan’s −$106B may be valuation and rotation. New ceiling mechanism: buybacks tripled to $6B per operation effective 2026-09-09 and visibly failed (Treasury borrowing statement; Bloomberg, 2026-09-01); larger buybacks, a coupon-size cut at the 2026-11-04 refunding, or Fed balance-sheet language on 2026-10-28 could cap the repricing before F1/F2’s “sustained” qualifiers accumulate. Net: the counter is stronger than episode 1’s on level timing and on part (b), and weaker on regime direction — it no longer explains why the floor held when oil fell.

Where we stand

As of 2026-09-25, confidence splits: part (a) moves to high — a thread first — while part (b) is unchanged in verdict, broader in channels, still unconfirmed; the honest composite for the two-part thesis is medium. No numeric probabilities; no calibrated basis exists. Since registration the 10-year’s +46bp is roughly 85% real yield (+33bp DFII10 versus +6bp T10YIE, reals through 2026-09-23 and breakevens through 2026-09-24; the share shifts slightly with those dates); the post-episode-1 leg is all real. The ambiguity moved decisively toward regime floor but did not resolve. Downgrade rule, published: another ~79% flagship indirect with widening elsewhere means tenor-selective weakness and shades part (a) back to medium-high; a sub-65% print or a material flagship tail settles it. Open: whether the concession reaches the flagships; whether the floor survives a completed deal; whether intervention caps first; what breaks in part (b).

What would falsify it now

The ledger carries forward unrevised: no condition, threshold, or window changed since registration; nothing added.

None is close: F1’s distance widened from +70bp to +88bp (5.18% par curve, 2026-09-24; FRED lags a day; zero sessions below 4.30%), F2’s from +56bp to +67bp (30-year 5.47%). F3 stays far: breakevens 67bp under the trigger; the cut leg is dead — sixteen of eighteen dots show another hike, none a cut (Federal Reserve SEP, 2026-09-16). F4: no flight-to-quality — the 10-year rose 22bp in seven sessions amid flat-to-firm equities. F5: no enacted consolidation; only affordability discourse, not legislation. Falsification requires all five firing; neither nearest live risk is a ledger condition: a completed deal unwinding the remaining oil premium (a level, not yet floor, risk), or intervention capping the repricing. Catalysts: August PCE (2026-09-30; BEA schedule, correcting episode 1’s ~09-25 estimate); jobs ~2026-10-02; the flagship 10Y/30Y reopenings ~2026-10-08/09 (pattern-based, unannounced as of this episode); FOMC 2026-10-27/28; the 2026-11-04 refunding (coupon path); the UNGA-track deal and Brent.