Nineteen basis points from the number that reprices every liability

Subject line: Nineteen basis points from the number that reprices every liability

The 30-year Treasury touched 5.33% on Tuesday morning — a fresh 19-year high — before easing back below 5.30%. The Desk puts the probability that the official H.15 series prints 5.50% before year-end at 60%. The interesting part is not the number. It is what separates the four ways of getting to it — a Monte Carlo at 53%, a carry-neutral forward with realised volatility at 65%, a full-sample base rate at 60%, a current-regime base rate at 78% — and the discovery, buried in the model, that how you allocate a fixed amount of uncertainty matters more than how much of it you assume. Uncertainty about which world you are in produces far fewer barrier touches than uncertainty inside the path, at exactly the same terminal dispersion. That structural choice is worth twelve percentage points on this question, and almost nobody states it.

The Trigger

On Tuesday 18 August the 30-year US Treasury yield rose above 5.33% in morning trading — a new 19-year high, on what CNBC attributed to persistent inflation and fiscal concerns — before easing back to trade near 5.29% by mid-morning. It got there in stages, and each stage is documented.

On 12 August, the Treasury's Monthly Statement for July showed a $432.3 billion monthly deficit — the largest for any month since March 2021 — with interest on the federal debt at $117.5 billion for the month, up from $91.9 billion a year earlier, and the fiscal-year deficit near $1.8 trillion with two months still to run. (CNBC, 12 August 2026)

On 13 August the Treasury sold $25bn of new 30-year bonds at a high yield of 5.216% — the highest yield at a 30-year auction since 2001. The bid-to-cover was 2.39, down from 2.44 at the July reopening. Indirect bidders, the category that contains most foreign official demand, took roughly two-thirds of competitive awards (66.8%); primary dealers absorbed 11.6%, more than their trailing-year average. Call the result unremarkable, which is itself the point — the auction cleared, at a price. (TreasuryDirect; CRFB, 14 August 2026)

On Monday 17 August, Treasury published Treasury International Capital data for June. Foreign holdings of Treasury securities fell $72.1bn over the month, from $9.371trn to $9.299trn — the third decline in four months, from a record set in February. Japan's holdings fell 2.3% to $1.116trn. China's fell 3.9% to $633.4bn, a post-2008 low. The United Kingdom's fell 0.9% to $939.9bn. Total net foreign purchases of US long-term securities were still positive on the month ($133.5bn of net inflows, with only $6.8bn into Treasuries); the composition was not.

And the same Monday, at the other end of the world's second-largest government bond market, Japan's 10-year JGB yield touched 2.93% — its highest since September 1996 — after weak Q2 growth data, with super-long JGB yields at or near record highs. (Euronews, 17 August 2026) The largest foreign holder of Treasuries now has a domestic alternative paying more than it has in a generation.

So the tape says two things at once. Foreigners are holding fewer Treasuries, the fiscal deficit is widening again, and the yield needed to clear a 30-year auction is the highest since 2001. And yet — foreigners still showed up for two-thirds of the last auction, and Tuesday's push above 5.33% found buyers within hours. Both halves are true. The gap between them is where this forecast lives.

The Forecast Question

Does the 30-year US Treasury constant-maturity yield print at or above 5.50% on any business day on or before 31 December 2026?

Grading series. Federal Reserve Statistical Release H.15, Selected Interest Rates, "Treasury constant maturities, Nominal, 30-year" — the series carried on FRED as DGS30. The daily H.15 figure is the sole arbiter. No intraday print, no futures-implied yield, no vendor close. Tuesday's 5.33% intraday high, for the avoidance of doubt, resolves nothing — H.15 is an end-of-day constant-maturity series, and the official series stood at 5.25% through 14 August at the time of writing, with the 17 and 18 August observations still to publish.

Window. 18 August 2026 through 31 December 2026 inclusive: 93 business days, net of US federal holidays.

Resolution. On the first business day of January 2027, against the published H.15 series. A single daily observation at or above 5.500 resolves it. Rounding is H.15's own — the release prints to two decimals, so 5.50 resolves and 5.49 does not.

Scenario partition — mutually exclusive, jointly exhaustive, graded on the maximum daily H.15 30-year yield across the window:

	Scenario	Definition
BASE	Absorbed	maximum DGS30 below 5.50%
BREAK	Buyers' strike	maximum DGS30 at or above 5.50%, below 5.75%
TAIL	Disorderly long end	maximum DGS30 at or above 5.75%

Starting level. 5.31% (Monday 17 August market close; Tuesday's session has traded either side of it). H.15 is official through 14 August at 5.25%; the 17 August observation had not yet appeared at the time of writing. The model runs from 5.31% and the 5.25% start is carried as a specification leg — it is worth 9.0 percentage points of probability, the single largest sensitivity in the entire exercise, and readers should know that before they read anything else.

Prior / Base Rate

Strip out every narrative and ask the plainest possible question: starting from any day since 2005, how often has the 30-year risen at least 19 basis points above its starting level at some point in the following 93 business days?

Sample	Observations	P(max rise ≥ 19bp)	P(max rise ≥ 25bp)
2005 – present	5,316	59.6%	52.0%
2015 – present	2,813	61.6%	53.7%
2022 – present	1,061	78.3%	71.0%

A 19bp rise in four and a half months is close to a coin flip across the full sample and a strong favourite in the post-2022 regime. That is the honest prior, and it is high. The 30-year is a volatile instrument: the standard deviation of its 93-business-day change is 42bp in the 2022–present sample, 44bp since 2015, 49bp since 2005. Against a distribution that wide, a 19bp barrier is not far away.

Three historical analogues frame the class of event. 2007, the last time the 30-year traded here: the level marked the end of a tightening cycle, and the long end rallied hard within a year — the analogue for BASE. October 2023, the last term-premium repricing: the 30-year ran from 4.0% to 5.1% in ten weeks on supply and deficit concerns before the November refunding turned it — the analogue for BREAK, and proof that 19bp can go in an afternoon when the driver is term premium rather than policy. The 2022 gilt episode: a long end repricing disorderly enough to force official intervention, transmitted through leveraged liability-driven investors — the analogue for TAIL, and a reminder that the tail mechanism is usually a forced seller, not a view.

Two cautions on the base rate, both of which cut against reading 78.3% as the answer.

First, the windows overlap. 1,061 observations since 2022 contain perhaps eight or nine independent four-month episodes. The apparent precision is an artefact of daily sampling.

Second, the post-2022 sample is drawn almost entirely from a rising-rate regime — the median 93-day change in that sample is +18bp. Conditioning on a period in which yields rose and then asking how often yields rose is close to assuming the answer.

The Evidence-Update Table

The prior is the full-sample base rate: P(touch) ≈ 60%. Each row states what moved, what it is evidence for, and roughly how much it shifted the Desk. The rows sum to approximately zero — the evidence is genuinely two-sided, and the posterior sits on the prior.

Evidence	Verified to	Direction	Weight
30-year touches 5.33% intraday 18 Aug — new 19-year high — then fades below 5.30%	CNBC, 18 August 2026	Mixed — the barrier is 17bp from the morning high, but the fade is the absorption bid showing up in real time	+2pp
July federal deficit $432.3bn, largest month since March 2021; monthly interest cost $117.5bn vs $91.9bn a year ago	Monthly Treasury Statement, 12 August 2026	Toward BREAK — the fiscal arithmetic is deteriorating inside the window	+3pp
Three FOMC dissents for a hike, 29 July, 9–3 vote — the first three-dissent hawkish split since September 2016	Federal Reserve statement, 29 July 2026	Toward BREAK — the policy risk is skewed to tightening, not easing	+4pp
Headline CPI 3.4% y/y in July (from 3.5%), above target for five years; core 2.5% (from 2.6%)	BLS CPI Summary, 12 August 2026	Mixed — elevated, but falling, and core is at a cycle low	+1pp
Foreign Treasury holdings −$72.1bn in June; Japan −2.3%, China −3.9% to a post-2008 low	Treasury TIC data, 17 August 2026	Toward BREAK — the marginal price-insensitive buyer is shrinking	+3pp
Japan's 10-year JGB at 2.93%, highest since 1996; super-long JGBs at records	Euronews, 17 August 2026	Toward TAIL — the repatriation channel is live: Japanese institutions now have a domestic alternative	+2pp
30-year auction cleared at 5.216%, highest since 2001; bid-to-cover 2.39 (from 2.44); indirects ~67% of competitive awards	TreasuryDirect, 13 August 2026	Mixed — a high clearing yield and softer cover, but demand showed up	0pp
Treasury "anticipates maintaining nominal coupon and FRN auction sizes for at least the next several quarters"	Quarterly Refunding Statement, 5 August 2026	Toward BASE — the supply shock is announced, flat and pre-digested	−5pp
Milliman 100 funded ratio 112.1% at 31 July — a 25-year high — on a 41bp discount-rate rise to 6.02%; liabilities fell $52bn in a month when assets returned −1.55%	Milliman Pension Funding Index, 10 August 2026	Toward BASE — surplus plans de-risk into long duration; 5.50% is where that bid gets loud	−7pp
20s30s slope flat at 0bp on H.15; ACM 10-year term premium 0.83% (7 Aug)	Federal Reserve H.15; NY Fed ACM	Toward BASE — no rolldown pressure; the carry-neutral forward is today's level	−2pp
July FOMC minutes due 19 August; FOMC meetings 15–16 Sep (SEP), 27–28 Oct, 8–9 Dec inside the window	FOMC calendar	Volatility, not direction — four scheduled catalysts	0pp, widens tails

Net of the table: the Desk sits at 60%, essentially on the full-sample base rate and 18 points below the post-2022 base rate. The pension-surplus row is doing most of the work in that gap, and it is the row the model cannot see.

The Scenarios

BASE — Absorbed · 40%

The 30-year grinds in a 5.20–5.45% range and never touches the barrier. Core inflation at 2.5% keeps the disinflation narrative alive; the July minutes on 19 August read hawkish-but-patient rather than pre-committing to a hike; the September and October auctions clear at sizes Treasury has already told the market it will hold. Meanwhile every basis point above 5.25% pulls another cohort of US corporate plans through a glide-path trigger — Milliman's index is at a 25-year funding high precisely because discount rates rose — and de-risking flows buy the long end at exactly the moment price-sensitive investors would sell it. Tuesday morning's tape was this scenario in miniature: a new high printed, and the bid appeared. This is the mechanically boring outcome and it is the single most likely one.

Resolving indicator: maximum H.15 DGS30 below 5.50% through 31 December.

BREAK — Buyers' strike · 35%

The barrier goes in a specific, identifiable window: the 19 August minutes plus the September FOMC, or the November refunding. It takes one hawkish surprise — a dissent count rising from three to four, an upside CPI print, an SEP dot shifted to a 2026 hike — and 19bp is a single bad afternoon in a market whose daily standard deviation is 4.5bp and whose fat tails are real; October 2023 covered 19bp in less than a week, twice. Yields touch 5.50–5.74% and then find the pension and insurer bid. Nothing breaks; the number is simply printed.

Resolving indicator: maximum H.15 DGS30 at or above 5.50% and below 5.75%.

TAIL — Disorderly long end · 25%

The bid does not appear. A weak auction is followed by a weaker one; the term premium reprices rather than drifts; the 30-year goes through 5.75% and keeps going. The catalyst is most plausibly foreign, not domestic — a further step down in Japanese or Chinese holdings, a yen or JGB dislocation that forces Japanese life insurers to repatriate into their own record-yielding super-long market, or a fiscal announcement that changes the announced-issuance picture Treasury has just committed to. In this scenario the 30-year is no longer a discount rate; it is a price, and it is being discovered.

Resolving indicator: maximum H.15 DGS30 at or above 5.75%.

Sum: 40 + 35 + 25 = 100. P(touch 5.50%) = 60%.

The aggregation, stated. The Desk's published headline is a weighted blend of the three quantitative estimates this build produced: the Monte Carlo as specified (53.2%, weight 0.4), the carry-neutral-forward reflection estimate (65.0%, weight 0.3), and the full-sample historical base rate (59.6%, weight 0.3) → 58.7%, rounded to the published 5%-granularity 60%. MiroFish was not run for this edition and no multi-agent input is claimed. The BREAK/TAIL split of the 60% follows the Monte Carlo's own conditional split (30.4 : 22.8), rounded to 5%: 35/25. The weights on the three estimates are judgment; the reason the Monte Carlo is not given full weight is structural and is explained in the next section.

The Monte Carlo

What the model does

The 30-year is simulated daily over 93 business days as a decomposition onto two economically linked observable drivers plus an orthogonal residual carrying the term-premium and supply channel:

Δy30(t) = β₂ · Δy2(t) + β_BE · ΔBE30(t) + ε(t) + μ(path)

where BE30 is the 30-year TIPS breakeven (DGS30 − DFII30).

The two drivers are drawn jointly through a t copula — a Gaussian correlation structure with a shared chi-square mixer, which preserves joint tail co-movement instead of diversifying it away — with Student-t marginals whose degrees of freedom come from kurtosis matching. Short-rate expectations and inflation compensation move together in this model, because they move together in the data.

μ(path) is a per-path drift drawn from a three-regime mixture. The contested input — whether the fiscal and supply bid-up of the long end persists — is treated as stochastic rather than assumed.

Seed 20260818. n = 50,000 paths. numpy only, zero other dependencies. The model file mc.py ships with this edition, and its headline block was re-executed and reproduced bit-for-bit on this run's data vintage before publication.

What is fitted, and what is judgment

This distinction was the single most-flagged weakness in the Desk's previous methodology review, and it is stated plainly here.

Fitted to data (FRED daily, 22 Feb 2010 – 14 Aug 2026):

Parameter	Value	Source of the number
β on Δ(2-year)	0.3226	OLS on daily changes, trailing 500 business days
β on Δ(30-year breakeven)	1.2212	same regression
R² of the decomposition	0.518	same regression
σ Δ(2y) / σ Δ(BE30) / σ(residual)	5.20 / 1.74 / 3.12 bp	trailing 500bd sample s.d.
t degrees of freedom (2y / BE30 / residual)	6.8 / 9.8 / 5.3	method of moments on excess kurtosis
Spearman ρ(Δ2y, ΔBE30)	0.432	rank correlation, same window
Carry/rolldown anchor	−0.0bp over the horizon	H.15 20s30s slope, flat at 0bp
Total horizon variance budget	σ = 42bp	realised s.d. of the 93-business-day change in DGS30, 2022–present

Judgment, not fitted — and every one of these is swept in the envelope below:

the three regime drifts (−28bp / +2bp / +38bp over the horizon) and their weights (30/45/25);
the event-day volatility multiplier (1.6×) and which 16 days count as event days;
the choice of a 500-business-day estimation window;
the barrier itself (5.50) and the tail cut (5.75).

The variance budget — and the specification this build rejected

A regime mixture bolted on top of a fully-scaled daily diffusion double-counts dispersion. The realised 93-day change in the 30-year already contains every regime shift the series has lived through; adding a regime drift on top of vol calibrated to the same series counts the same uncertainty twice.

So the total is pinned to the realised 42bp. The regime mixture claims 26bp of that budget; the diffusion takes the residual — the 40bp of fitted daily volatility scaled by κ = 0.759.

The analytic budget is exact only for Gaussian marginals; with Student-t marginals and a shared chi-square mixer the realised dispersion overshoots by roughly 4%. The budget is therefore closed numerically: a deterministic 10,000-path pilot measures the simulated horizon standard deviation and corrects κ until it binds. Simulated horizon s.d.: 42bp, against a 42bp target. A variance budget that is asserted rather than measured is not a budget.

The rejected alternative is worth naming because it is the specification a less careful build would have shipped: fitted daily volatility plus the regime mixture, unbudgeted, implies a 47bp horizon standard deviation against a realised 42bp and returns P(touch) = 94.5%. It is excluded from the envelope. A model that cannot reproduce the dispersion of the series it forecasts is not a conservative model. It is a wrong one.

Results

Outcome	Monte Carlo
BASE — max below 5.50%	46.8%
BREAK — max 5.50% to 5.74%	30.4%
TAIL — max at or above 5.75%	22.8%
P(touch 5.50%)	53.2%

Simulated 30-year quantiles (5 / 25 / 50 / 75 / 95):

Year-end level: 4.64 / 5.04 / 5.32 / 5.61 / 6.03
Horizon maximum: 5.30 / 5.39 / 5.52 / 5.73 / 6.08

P(touch) conditional on regime: disinflation 25.5%, carry-neutral grind 54.0%, supply push 85.3%. The regime assumption is the model's dominant lever, which is why it is stochastic and why it is swept.

Why the model sits below every other estimate — and why that is the finding

The Monte Carlo returns 53.2%. A driftless diffusion with the identical 42bp terminal standard deviation returns 65.0%. Same dispersion, twelve points apart. This is not an error; it is the most important structural result in the build.

A path driven by a constant per-path drift has a running maximum close to its endpoint. A path driven by diffusion wanders, and its running maximum sits well above its endpoint. Barrier questions are decided by the maximum, not the endpoint — so where you put a fixed variance budget determines the answer.

Putting 26bp of the 42bp budget into a regime mixture is a substantive claim: the uncertainty here is mostly about which world we end up in, not about day-to-day noise inside a known world. That claim is defensible for a long bond over four months, when the dominant question is a regime one. It is also worth twelve percentage points, and a reader who thinks the uncertainty is more diffusive than regime-like should read a number nearer 65%.

Internal validation. Given the drift the 2022–present regime actually realised (+18bp median), the model returns 72.6% against a realised base rate of 78.3% at the same barrier and horizon. The machinery reproduces history to within six points when handed history's drift. The 53.2% headline is the cost of declining to assume the last four years repeat.

Threshold sensitivity

Barrier	P(touch)
5.40%	71.8%
5.45%	61.8%
5.50%	53.2%
5.55%	45.7%
5.60%	38.9%
5.75%	22.8%

Roughly 1.7 percentage points of probability per basis point of barrier in this neighbourhood. A reader who prefers 5.45% as the psychologically relevant level should read 62%, not 53%.

Confidence intervals — the honest one and the fake one

The sampling standard error on 50,000 paths is ±0.22pp. That number describes how stable the arithmetic is. It says nothing whatever about whether the model is right, and reporting it as the uncertainty would be a lie of precision.

The honest band is the specification envelope: seventeen legs, each changing a stated judgment call (two joint legs change three at once).

Specification leg	P(touch 5.50%)
As specified	53.2%
Start at H.15 official 5.25%	44.2%
Estimation window 250bd	53.4%
Estimation window 1,250bd	52.5%
Gaussian marginals (no fat tails)	53.8%
No event-day volatility multiplier	53.4%
Event multiplier 2.2×	52.8%
No rolldown anchor	53.2%
Regimes: dovish tilt 45/40/15	46.9%
Regimes: supply tilt 15/40/45	63.6%
Zero drift, single regime	59.4%
Drift = realised 2022+ median (+18bp)	72.6%
Horizon s.d. from 2015+ (44bp)	55.8%
Horizon s.d. from 2005+ (49bp)	60.1%
JOINT hawkish — supply tilt and 2005 σ and 2.2× events	68.0%
JOINT dovish — dovish tilt and Gaussian and 5.25% start	38.3%
Seeds 20260819 / 20260820	53.7% / 53.8%

The joint legs move three assumptions at once. A one-factor-at-a-time sweep systematically understates model uncertainty, because the legs are not independent — a reader who believes the supply regime dominates probably also believes volatility is closer to its 2005–present level.

Envelope: 38% – 73%. Width 34 percentage points — seventy-seven times the sampling standard error. That is the uncertainty. The ±0.22pp is a footnote.

Reconciliation

	BASE	BREAK	TAIL	P(touch)
Monte Carlo (as specified)	46.8%	30.4%	22.8%	53.2%
The Desk (published)	40%	35%	25%	60%
Delta (Desk − MC)	−6.8pp	+4.6pp	+2.2pp	+6.8pp
Carry-neutral forward + realised vol	—	—	—	65.0%
Base rate, 2005–present	—	—	—	59.6%
Base rate, 2022–present	—	—	—	78.3%

The Desk overrides the headline upward, and does not override the tail at all.

An earlier draft of this edition did the opposite — it accepted the model's headline and pulled 9.3 points out of the tail on a pension-absorption argument. That override has been withdrawn. It was a story about a feedback loop, asserted from an aggregate funded ratio, with no flow data behind it. The absorption argument survives below as a reason the tail is uncomfortable, not as a number.

What the Desk does override is the headline, and upward, by the stated aggregation. The model allocates 26bp of its 42bp variance budget to regime uncertainty and only 33bp to diffusion, which — per the section above — mechanically suppresses barrier touches by roughly twelve points relative to a pure diffusion at the same terminal dispersion. The Desk's judgment is that the true split is somewhere between the two, and 60% is where the model (53%, weight 0.4), the carry-neutral forward (65%, weight 0.3) and the full-sample base rate (60%, weight 0.3) land when blended and rounded to 5%.

The Desk is 18 points below the current-regime base rate of 78.3%, and that gap is the single clearest statement of the Desk's view: the last four years were a rising-rate regime, and we decline to assume it continues. Running the model with that regime's realised drift returns 72.6% — so the disagreement is not with the machinery, it is with the assumption.

Market vs Desk View

There is no listed instrument that prices this barrier. No exchange lists a one-touch on the 30-year constant-maturity yield. The nearest observables are three, and none of them is a market probability:

The carry-neutral forward. The H.15 20s30s slope is flat at 0bp, so rolldown over 4.5 months is approximately zero and the forward 30-year is today's level. Combining that forward with the realised 42bp horizon standard deviation gives, by the reflection principle, P(touch) = 65.0%. This is a statistical barrier probability off a market-implied central case, not an option price, and it should not be described as market-implied.
The full-sample historical base rate: 59.6%.
The current-regime base rate: 78.3%.

Benchmark	P(touch)	Desk − benchmark
Carry-neutral forward + realised vol	65.0%	−5.0pp
Base rate, 2005–present	59.6%	+0.4pp
Base rate, 2022–present	78.3%	−18.3pp
Monte Carlo, as specified	53.2%	+6.8pp

Around the barrier itself, the observable market context as of this writing: the ACM estimate of the 10-year term premium stood at 0.83% on 7 August (NY Fed) — positive and rising, but far below its pre-2008 norms, which is the quantitative sense in which a further term-premium repricing is the tail's fuel rather than a spent force. Long-maturity investment-grade corporate yields (ICE BofA 15+ year effective yield) reached 6.30% on 17 August (FRED BAMLC8A0C15PYEY); the 30-year fixed mortgage averaged 6.67% on 13 August (FRED MORTGAGE30US). The long end's repricing is already propagating into every private discount rate downstream of it.

What a reader should take from the absence of a listed market: if you want this exposure, you are constructing it from payer swaptions, bond futures options, or outright duration, and you are paying a bid-offer that no probability estimate here accounts for.

The Universal-Owner Portfolio Heatmap

Direction and indicative magnitude of the repricing pressure by asset class, conditional on scenario, over the balance of 2026. Direction is the sign of the expected mark-to-market or discount-rate effect; magnitude bands are qualitative: ▲/▼ = modest, ▲▲/▼▼ = material, ▲▲▲/▼▼▼ = severe. This is a strategic read for owners of diversified, long-duration balance sheets — not a trade sheet.

Asset class	BASE — Absorbed (40%)	BREAK — Buyers' strike (35%)	TAIL — Disorderly (25%)
Long-duration nominal bonds	▲ carry dominates, range-bound	▼▼ mark-to-market losses to 5.75%	▼▼▼ forced-seller dynamics
Inflation-linked bonds	— flat real yields	▼ real yields dragged higher	▼▼ liquidity discount on top
DB pension liabilities (PV)	— stable near 25-year funding high	▼▼ liabilities fall — funded ratios improve further	▼▼▼ liabilities fall fastest — but asset side breaks too
Global public equities	— supported by stable discount rate	▼ multiple compression at the long end	▼▼▼ equity–bond correlation flips positive
Investment-grade credit	▲ 6.3% long IG yield attracts LDI flows	▼ issuance shortens, curve steepens	▼▼ new-issue windows close from the long end in
Private equity / growth	— marks lag, no catalyst	▼ Q4/Q1 discount-rate marks	▼▼▼ refinancing arithmetic breaks on 2021 rate stacks
Infrastructure & core real estate	— bid steady from de-risking flows	▼▼ cap-rate pass-through begins	▼▼▼ lagged marks converge on a repriced curve
Private credit	▲ floating-rate carry intact	— mixed: carry vs. credit stress	▼▼ borrower coverage ratios compress
USD / FX	— dollar index steady near 119	▲ yield-driven dollar bid	▲▲ then reversal risk if foreign selling leads
EM sovereign debt	— floor holds	▼▼ long-end floor rises globally	▼▼▼ index-wide repricing off the US long bond
Gold / reserves	— real-yield headwind	▲ hedging bid	▲▲ disorderly-move hedge premium
Cash / T-bills	▲ 3.63% fed funds floor	▲▲ optionality value rises	▲▲▲ the only dry powder that matters

For an investor who owns a slice of everything, a 5.50% 30-year is not a trade. It is a repricing of the denominator.

Liabilities. The direction is the one most commentary gets backwards: a higher discount rate lowers the present value of a liability. Milliman's July move is the mechanism in plain sight — the discount rate rose 41bp to 6.02%, liabilities fell $52bn on the rate move alone, and the funded ratio went to 112.1% — its highest in more than 25 years — in a month when assets returned −1.55%. For a closed or frozen DB plan, the touch scenario is good news. For an open plan with an active accrual, it is good news that arrives with a higher service cost next year.

Duration positioning. A plan at 112% funded has a decision that a plan at 95% does not: whether to convert a surplus into a locked hedge. The BREAK scenario is the trigger event for that decision across a large cohort simultaneously, and the simultaneity is the risk — a coordinated de-risking bid is also a coordinated withdrawal of the marginal seller. Plans that have not pre-specified their glide-path trigger will find themselves deciding in the same fortnight as everyone else.

Sovereign funds and reserve managers. The TIC series is the mirror of your own behaviour. If Japan, China and the UK are each reducing, the question for a fund with a Treasury-heavy reserve tranche is whether it is being paid enough to be the residual holder. H.15 puts the 30-year TIPS real yield at 3.00% on 14 August against a 5.25% nominal — a 2.25% thirty-year breakeven. Hold that breakeven and a 5.50% nominal is a 3.25% real yield for thirty years, near the top of anything available since the pre-2008 era. That is not a distressed price. It is a generous one, and the discipline is to notice that generosity is what a repricing feels like from the inside.

The 60/40 correlation. The tail scenario is the one that matters most and is priced least. A disorderly long end is the environment in which equities and bonds fall together, because the discount rate on every long-duration asset — growth equity, infrastructure, real estate, private credit marks — rises at once. A universal owner cannot diversify out of this by construction: it owns all of the affected assets. The correct question is not "how do I avoid it" but "what have I written into my liquidity plan for a quarter in which both sleeves are down and capital calls arrive on schedule."

Private-market marks. The pass-through is lagged and partial, which is a reporting advantage and an economic illusion. A 25bp move in the long bond that persists to year-end will show up in Q4 and Q1 valuations of infrastructure and core real estate, in the discount rate applied to unlevered cash flows, and in the refinancing arithmetic of anything levered on a 2021 rate stack.

Second- and Third-Order Effects

Second order.

Japanese repatriation is reflexive. If JGB yields rise alongside Treasuries — and this Monday the 10-year JGB printed its highest yield since 1996, with the super-long sector at records — the hedged pick-up on US duration for a Japanese life insurer compresses, and the marginal Japanese buyer becomes a marginal Japanese seller. TIC already shows Japan down 2.3% in June. This is the single most likely mechanism by which BREAK becomes TAIL.
Corporate issuance windows close from the long end in. Long-maturity IG yields are already at 6.30%; a 5.50% 30-year Treasury with an investment-grade spread on top puts long corporate issuance meaningfully above 6.5%. Issuers shorten, which steepens the credit curve and shifts refinancing risk forward in time rather than removing it.
The Fed's own balance sheet becomes a talking point. The July statement commits to "maintaining ample reserves" with the balance sheet at $6.76trn (H.4.1, 12 Aug). A disorderly long end invites the question of whether ample-reserves policy extends to duration, and the mere posing of that question is itself a yield-suppressing event.
The interest bill compounds the deficit that drives the yield. July's $117.5bn monthly interest cost is itself a function of the yields this report forecasts; each auction at 5%+ locks the loop tighter for thirty years.

Third order.

Sovereign wealth funds with dollar-linked currency regimes face a widened cost of the peg. For Gulf funds, a higher US long rate raises the domestic cost of capital regardless of local inflation, tightening the hurdle rate on the exact domestic-diversification projects those funds are mandated to make.
Emerging-market issuance reprices off the long end, not the front. A 30-year at 5.50% raises the floor under every sovereign and quasi-sovereign long bond in the index.
The de-risking cohort shrinks the free float. If a large share of the US corporate DB universe converts surplus into locked long-duration portfolios during BREAK, the price-sensitive free float in long Treasuries falls, and the next supply shock lands on a thinner market. The absorption that prevents the tail this year makes the tail more violent next year.

Watch Dashboard

Date	Event	What would move the Desk
19 Aug	July FOMC minutes	Any indication the three dissents had broader sympathy → toward BREAK. Explicit patience language → toward BASE.
20 Aug	30-year TIPS reopening	A weak real-yield auction is a cleaner supply signal than a nominal one
~11 Sep	August CPI	Core above 2.7% y/y → toward BREAK
15–16 Sep	FOMC + Summary of Economic Projections	The dot plot is the single highest-information event in the window. A 2026 median showing a hike → BREAK becomes the base case
~15 Sep	TIC data for July	A second consecutive Japan decline of 2%+ → raise TAIL
Late Sep	30-year reopening, $22bn	Indirect share below 55% of competitive awards → toward BREAK
Mid-Oct	Monthly Treasury Statement, FY2026 close	A full-year deficit materially above $2trn → toward BREAK
27–28 Oct	FOMC	No SEP; statement language on inflation persistence
~4 Nov	Quarterly Refunding Statement	Any deviation from "maintaining nominal coupon sizes for at least the next several quarters" → immediate re-forecast
8–9 Dec	FOMC + SEP	Last scheduled catalyst inside the window
Continuous	20s30s slope on H.15	A steepening from 0bp is the cleanest real-time read on term-premium repricing
Continuous	Japan 10- and 30-year JGB yields	The repatriation channel — 2.93% on the 10-year is the highest since 1996
Continuous	ACM term premium (NY Fed)	0.83% on 7 Aug; a decisive break above 1.25% would mark a genuine regime shift

Red Team — How This Could Be Wrong

The three sharpest counter-arguments first.

1. The estimates are not independent, and their spread is structural. The Monte Carlo (53.2%), the reflection-principle calculation (65.0%) and the full-sample base rate (59.6%) all use the same realised volatility of the same series. They are not three confirmations; they are one volatility estimate run through three different assumptions about where the variance sits. The 12-point gap between the Monte Carlo and the reflection calculation is entirely attributable to that choice, not to any new information. A reader should treat the 53–65% span as a single estimate with a structural ambiguity inside it. What would falsify the call: nothing cleanly — which is precisely the problem; the blend is a judgment about model structure, not about the world.

2. The starting level is doing more work than the model. Running from 5.25% instead of 5.31% moves the answer 9 points. The 5.31% figure is a market close reported by vendors; the H.15 observations for 17 and 18 August were not published at the time of writing. If Monday's official print lands at 5.28%, this brief is meaningfully wrong on its headline before any analysis is tested. What would falsify the call: the H.15 prints for 17–18 August, available within days.

3. The absorption argument is a story, and it has been demoted to one. An earlier draft moved 9.3 points from TAIL to BREAK on a pension-de-risking feedback loop that was asserted, not measured. No flow data was inspected. The Milliman funded ratio establishes that plans are in surplus; it does not establish that they buy long duration at 5.75%, and it says nothing about execution lag, governance delay, or the dispersion of funded status beneath the aggregate. The override has been withdrawn and the Desk now carries the model's tail. The argument remains a reason to watch the tail, not a reason to shrink it. What would falsify the call: Q3 LDI flow data, or a weak auction into a rising funded ratio.

And six more, in descending order of discomfort:

4. The regime weights and the variance split are the model. P(touch) ranges from 25.5% to 85.3% across the three regimes. The 30/45/25 weighting is judgment, the 8bp within-regime drift dispersion has no empirical anchor at all, and the split of the 42bp budget between regime and diffusion is worth twelve points on its own. Everything else in the specification — copula, fat tails, event multipliers, estimation window — moves the answer by two points or less. The elaborate machinery is not what is driving the number; three hand-set judgments are.

5. Absence of evidence on foreign selling. The TIC series measures holdings, which change with valuation as well as flow. A $72.1bn decline in a month when yields rose is partly a price effect, not a sale. This brief has not decomposed it, and treating the full $72.1bn as evidence of foreign withdrawal would overstate the case. The auction data — indirects at roughly two-thirds — is the better flow proxy, and it says demand showed up.

6. The barrier is arbitrary and round. 5.50% is a human number. The threshold table exists precisely so a reader can substitute their own; at 5.45% the answer is 62%, at 5.60% it is 39%. Nothing about 5.50 is economically privileged.

7. The disinflation case is under-weighted. Core CPI is 2.5% y/y — a cycle low — and headline fell to 3.4% from 3.5%. If the energy shock continues to fade (the July energy index fell 1.5% on the month) and the labour market cools, the front end rallies, the 2-year drags the 30-year with it at a beta of 0.32, and the barrier never comes into view. The Desk gives the disinflation regime 30% weight. A reasonable person could give it 45% — that leg is in the envelope at 46.9%.

8. Volatility does not cluster in this model. A supply-push path carries the same daily volatility as a disinflation path. Real repricing episodes do not behave that way — volatility rises with the drift. Since barrier-touching is driven by path volatility, a model without clustering under-counts touches in exactly the regime where they matter most.

9. Nothing here contains a fiscal surprise. Treasury has committed to flat coupon sizes for "at least the next several quarters." Every model in this brief takes that at face value — even as the July deficit prints at a five-year monthly high. A supplemental funding requirement announced in October is outside the distribution entirely.

Methodology Box

Question type: barrier / one-touch, binary, single official series.
Grading: Federal Reserve H.15 Selected Interest Rates, 30-year Treasury constant maturity nominal (FRED DGS30). Resolved on the first business day of January 2027.
Window: 18 Aug – 31 Dec 2026, 93 business days net of federal holidays.
Ensemble and aggregation: Monte Carlo as specified (weight 0.4) + carry-neutral forward with realised volatility (0.3) + full-sample base rate (0.3) → 58.7%, published at 5% granularity as 60%. BREAK/TAIL split follows the Monte Carlo's conditional ratio. MiroFish not run; no multi-agent simulation input is claimed.
Simulation: 50,000 paths, seed 20260818, numpy 2.2.6, no other dependencies. Daily decomposition onto Δ(2-year) and Δ(30-year breakeven) with an orthogonal residual; t copula with Student-t marginals; per-path three-regime stochastic drift; total horizon variance pinned to the realised 93-day change standard deviation (42bp, 2022–present) via a scale factor κ = 0.759 on the fitted diffusion, closed numerically on a deterministic 10,000-path pilot because the analytic budget is exact only under Gaussian marginals. Simulated horizon standard deviation: 42bp, matching the target. The headline block was re-executed on publication day and reproduced exactly.
Estimation window: trailing 500 business days to 14 August 2026; full data 22 Feb 2010 – 14 Aug 2026.
Data: FRED series DGS30, DGS20, DGS2, DFII30, T10YIE, THREEFYTP10, BAMLC8A0C15PYEY, MORTGAGE30US, DTWEXBGS, WALCL, retrieved 18 August 2026 via the St. Louis Fed API.
Uncertainty reported: the 17-leg specification envelope (38–73%), including two joint legs that move three assumptions simultaneously. The Monte Carlo sampling standard error (±0.22pp) is reported but explicitly demoted; it measures arithmetic stability, not model correctness.
Rejected specification, disclosed: unbudgeted variance (fitted daily volatility plus regime mixture), P = 94.5%, excluded because it implies a 47bp horizon standard deviation against a realised 42bp.
Known simplifications, disclosed: the t copula uses a single degrees-of-freedom parameter for the driver pair, min(ν₂, ν_BE) = 6.8, so fitted per-leg kurtosis is not preserved exactly on the breakeven margin; the residual carries a separate mixer, so the three daily shocks are not jointly t-distributed. The within-regime drift dispersion (8bp) has no empirical anchor. The event-day multiplier redistributes volatility across the horizon rather than adding to it — an unnormalised multiplier was found to re-inject 3.5 percentage points of unbudgeted probability and was corrected before publication.
Not modelled: absorption feedback from pension de-risking; volatility clustering conditional on regime (a supply-push path carries the same daily volatility as a disinflation path, which is not how repricing episodes behave); bid-offer or financing cost of expressing the view; fiscal surprise outside announced issuance; the decomposition of TIC holdings changes into price and flow.
Reproducibility: mc.py is deterministic under the stated seed and numpy version, and prints the reconciliation line, the threshold grid, the full envelope and the base-rate table on every run. It is published with this edition.

Probabilities are the UAO Probability Desk's, weighted across base-rate, market-benchmark, and Monte Carlo simulation inputs. Methodology available on request.

Editorial scenario analysis only. Not investment, actuarial, or geopolitical advice. This report is for informational and research purposes only and does not constitute investment, legal, tax, or financial advice. It is a probabilistic forecast and it will sometimes be wrong; that is what a probability of 60% means.

Source Ledger

All links verified to resolve on 18 August 2026 unless annotated. Confidence: H = primary/official, M = reputable secondary.

Federal Reserve, H.15 Selected Interest Rates (Daily) — 30-year constant maturity 5.25% for 14 August (latest official); 20-year 5.25%; 2-year 4.17%; 30-year TIPS 3.00%. H — the grading series. federalreserve.gov/releases/h15
FRED, series DGS30 — grading series, retrieved via the St. Louis Fed API on 18 August 2026; latest official observation 2026-08-14 at 5.25%. H fred.stlouisfed.org/series/DGS30
CNBC, 18 August 2026 — "30-year Treasury yield tops 5.33%, new 19-year high on inflation, spending concerns"; yield subsequently ~5.29%. M — intraday levels are vendor quotes, not the grading series. cnbc.com
CNBC, 18 August 2026 — "The 30-year Treasury yield just hit a 19-year high. Three things could drive it even higher." M cnbc.com
CNBC, 17 August 2026 — 30-year Treasury yield tops 5.31%, highest in 19 years. M cnbc.com
US Treasury, Treasury International Capital data for June, released 17 August 2026 — foreign holdings of Treasuries $9.299trn, down $72.1bn from $9.371trn; Japan $1.116trn (−2.3%), UK $939.9bn (−0.9%), China $633.4bn (−3.9%); net foreign inflows $133.5bn, of which $6.8bn into Treasuries. H home.treasury.gov TIC system
TIC, Major Foreign Holders of Treasury Securities (Table 5) — country-level holdings series. H ticdata.treasury.gov
Traders Union, 18 August 2026 — corroboration of the June TIC release figures. M tradersunion.com
TreasuryDirect, auction results, 30-year bond CUSIP 912810UW6, auctioned 13 August 2026 — $25bn offered, high yield 5.216%, bid-to-cover 2.39 (2.44 at the 9 July reopening), indirects 66.8% of competitive awards, directs 21.6%, primary dealers 11.6%, 0.4bp tail. H treasurydirect.gov
Committee for a Responsible Federal Budget, 14 August 2026 — "Treasury Auction Yield Hits Highest in 25 Years." M crfb.org
US Treasury, Quarterly Refunding Statement, 5 August 2026 — $125bn refunding; Treasury "anticipates maintaining nominal coupon and FRN auction sizes for at least the next several quarters." H home.treasury.gov/news/press-releases/sb0590
US Treasury, Monthly Treasury Statement for July 2026, released 12 August 2026 — monthly deficit $432.3bn, largest since March 2021; receipts $334bn, outlays $766bn; monthly interest on the public debt $117.5bn (vs $91.9bn in July 2025); FY26 deficit to date ≈$1.8trn; customs receipts negative a third consecutive month on tariff refunds. H fiscaldata.treasury.gov
CNBC, 12 August 2026 — "U.S. budget deficit surged in July to highest level since March 2021." M cnbc.com
Federal Reserve, FOMC statement, 29 July 2026 — target range maintained at 3½–3¾%; approved 9–3; Hammack, Kashkari and Logan dissented in favour of a ¼-point increase. H federalreserve.gov
CNBC, 29 July 2026 — Fed holds rates; first time since September 2016 that three policymakers dissented in the same direction. M cnbc.com
Federal Reserve, FOMC meeting calendar — July minutes due 19 August; remaining 2026 meetings 15–16 September (SEP), 27–28 October, 8–9 December (SEP). H federalreserve.gov/monetarypolicy/fomccalendars.htm
Bureau of Labor Statistics, Consumer Price Index Summary, July 2026, released 12 August 2026 — headline +0.1% m/m, 3.4% y/y (from 3.5%); core +0.2% m/m, 2.5% y/y (from 2.6%); energy −1.5% m/m, +14.7% y/y; shelter roughly two-thirds of the monthly rise. H bls.gov (returns 403 to automated retrieval from this environment — a bot wall, not a broken link; figures cross-checked against contemporaneous reporting)
CNBC, 12 August 2026 — CPI inflation report July 2026 corroboration. M cnbc.com
Milliman, Pension Funding Index, August 2026 edition (data to 31 July), published 10 August 2026 — Milliman 100 funded ratio 112.1%, highest in more than 25 years; discount rate up 41bp to 6.02%; liabilities −$52bn on the rate move; assets $1.296trn after a −1.55% monthly return. H milliman.com
Milliman, Pension Funding Index, June 2026 edition — funded ratio 109.6% at 31 May; discount rate 5.62%. H milliman.com
Milliman, Pension Funding Index, July 2026 edition — funded ratio at 30 June, the base for July's rise. H milliman.com
Euronews, 17 August 2026 — Japan's 10-year JGB yield touches 2.93%, highest since September 1996, after disappointing Q2 growth data. M euronews.com
Yahoo Finance / Reuters, August 2026 — "Japan's super-long bond yields soar to records as market frets about demand." M finance.yahoo.com
Federal Reserve Bank of New York, Term Premia estimates (ACM) — 10-year term premium 0.83% at 7 August 2026 (FRED THREEFYTP10). H newyorkfed.org
FRED, series BAMLC8A0C15PYEY — ICE BofA 15+ Year US Corporate effective yield 6.30% at 17 August 2026. H fred.stlouisfed.org
FRED, series MORTGAGE30US — 30-year fixed mortgage average 6.67% at 13 August 2026. H fred.stlouisfed.org
FRED, series DTWEXBGS — broad dollar index 118.90 at 14 August 2026. H fred.stlouisfed.org
FRED, series WALCL — Federal Reserve total assets $6.760trn at 12 August 2026 (H.4.1). H fred.stlouisfed.org
FRED, series DFII30 — 30-year TIPS constant maturity 3.00% at 14 August 2026. H fred.stlouisfed.org
FRED, series T10YIE — 10-year breakeven 2.28% at 17 August 2026. H fred.stlouisfed.org
TreasuryDirect, Announcements, Data & Results — upcoming auction schedule, including the 30-year TIPS reopening announced for 20 August 2026. H treasurydirect.gov

(FRED web pages are not reachable from this environment's network egress; the St. Louis Fed API returned every cited series normally on 18 August 2026 — annotated, not broken. Reuters primary URLs are not used per licensing constraints; Reuters reporting is cited via licensed mirrors.)

THE PROBABILITY DESK · UNIVERSAL ASSET OWNERS Monte Carlo n=50,000, seed 20260818. Model source published with the edition.

Continue the briefing. Read the daily brief · watch the daily video briefing · listen to The Universal Owner · view the chart of the day.

Produced and edited by the UAO editorial desk. Not investment advice.