The Bank of Japan hiked. The yen fell. Watch the thirty-year.
The Probability Desk · Friday 18 September 2026
A probability-weighted scenario report for long-horizon owners.
Executive summary
At 11:54 Tokyo time this morning the Bank of Japan raised its policy rate to around 1.25 percent, by a seven-to-two vote, effective 24 September. It is the highest Japanese policy rate in 31 years. The yen fell. The dollar rose roughly 1.2 percent against it, to a two-week high near 157.9, after Governor Kazuo Ueda declined to commit to a pace.
A rate rise that weakens the currency is worth a paragraph in a market report. What makes it worth a scenario report is the second thing that did not behave: the long end. Japan's 30-year government bond closed at 4.047 percent on 17 September. In the entire history of the tenor — 6,626 trading days since it was introduced on 2 September 1999 — the 30-year has closed at or above 4 percent on 25 days, and every one of them has fallen in the last four months. The first was 18 May 2026. The all-time high, 4.131 percent, was set on 1 September 2026, seventeen days ago.
Japan's long bond is the marginal price of patient capital. It has been the world's cheapest source of duration for a generation, the funding leg of the largest carry trade in finance, and the anchor against which every other sovereign long bond has looked expensive. It is now none of those things, and the institution that set its price for twenty years is stepping back from the market on a published schedule.
The forecast question: where does the 30-year JGB yield stand on 31 March 2027?
The Desk's weights: A, below 3.50 percent — 5 percent. B, 3.50 to under 4.00 — 20 percent. C, 4.00 to under 4.50 — 40 percent, the base case. D, 4.50 percent or higher — 35 percent.
The probability of a 30-year back below 4 percent by the end of Japan's fiscal year is 25 percent. The probability of 4.50 or higher is 35 percent.
The Desk view in one paragraph. The September hike was not, on the evidence, primarily a tightening event. Core inflation excluding fresh food was 1.7 percent in August, below the 2 percent target for a seventh consecutive month and slower than July. The Bank raised rates anyway, against a stated risk that underlying inflation would deviate above target later. The two dissenters were the board's two newest members. Two further members dissented in the opposite direction, objecting that the Bank's price outlook was too cautious. A board dissenting in both directions widens the distribution of next moves rather than shifting its centre — and the currency market read the split as the slower path within the hour. The long end, meanwhile, is no longer trading on any of this. Over the two and a half years since the Bank ended negative rates, 93 percent of the daily variance in the 30-year JGB is unexplained by the front end and the US long bond combined. What sets its price now is domestic supply and domestic demand — and the two largest domestic buyers are moving in opposite directions. That is the question this report is about, and it will not be settled by a policy rate.
The trigger
What happened. On 18 September 2026 the Policy Board of the Bank of Japan decided, by a 7–2 majority, to encourage the uncollateralized overnight call rate to remain at around 1.25 percent, effective 24 September. The interest rate on the complementary deposit facility goes to 1.25 percent and the basic loan rate to 1.5 percent, both on the same date. (Bank of Japan, Change in the Guideline for Money Market Operations, 18 September 2026.)
Who dissented, and in which direction. This is the part that has been under-read.
Voting against the rise: Asada Toichiro and Sato Ayano. Asada's recorded reason is that with core CPI below 2 percent recently, "it could not necessarily be said that the economic situation was strong." Sato's is that developments "did not appear to have substantially accelerated compared to before."
But the same document records two more dissents, pointing the other way, on the description of the price outlook. Takata Hajime opposed it, considering that CPI inflation "already had generally reached the price stability target." Tamura Naoki opposed it, considering that underlying inflation was already consistent with the target.
Four of nine board members formally dissented from something today, two because the Bank moved too fast and two because its language was too slow. (Takata and Tamura voted for the rate rise; their dissent was on the outlook text.) A committee dissenting in both directions widens the distribution of next moves rather than shifting its centre, and the currency market priced that width immediately.
The appointment channel. Asada and Sato are the board's two newest members, nominated by Prime Minister Sanae Takaichi, and both are identified in market commentary as favourable to loose monetary and fiscal policy. We flag this as the mechanism that matters most over a decade horizon and the one least visible in a headline rate: a government pursuing fiscal expansion is simultaneously changing the composition of the committee that prices its debt. That is not an accusation of impropriety — appointing the board is the government's lawful prerogative in Japan as elsewhere. It is an observation about the joint distribution of fiscal and monetary outcomes, and it is why this report treats the two as one problem.
What the Bank said about the future. The statement's forward guidance is explicit and hawkish in direction if not in pace: "the Bank will continue to raise the policy interest rate and adjust the degree of monetary accommodation, in response to developments in economic activity and prices as well as financial conditions." The named risks are the situation in the Middle East, global AI-related demand, and the exchange rate. The Bank expects core CPI "to accelerate to a level clearly above 2 percent from the second half of fiscal 2026" — though the same passage has that rate declining back toward 2 percent thereafter, with underlying inflation reaching a level "generally consistent with" the target between the second half of fiscal 2026 and fiscal 2027.
What the market did. The yen weakened through 157 and touched 158 per dollar. The dollar rose about 1.2 percent to roughly 157.9. The last published Federal Reserve H.10 observation for the yen before today was 153.71 on 11 September.
The forecast question
Where will the 30-year Japanese government bond yield stand on 31 March 2027?
Resolution source. The Japan Ministry of Finance daily "Interest Rate" table for Japanese government bonds, 30Y column, value published for 31 March 2027 — the last business day of Japan's fiscal year 2026. Sole arbiter. No judgement required. First-published value governs.
Buckets — mutually exclusive, collectively exhaustive on a continuous variable:
| Bucket | Name | Desk weight | |
|---|---|---|---|
| A | Below 3.50% | Absorption | 5% |
| B | 3.50% to under 4.00% | Back below four | 20% |
| C | 4.00% to under 4.50% | The plateau (base case) | 40% |
| D | 4.50% or higher | The term-premium regime | 35% |
Starting level: 4.047 percent, the close on 17 September 2026 — the most recent value the Ministry of Finance had published when this report was built. The 18 September value was not yet posted; we re-requested with the cache defeated to confirm the absence rather than infer it.
Threshold probabilities: P(below 4.00) = 25%. P(at or above 4.50) = 35%. P(at or above 5.00) = 11% on the simulation and 0.3% on the base rate — a gap we discuss rather than average away, in the red team below.
Key data
| Variable | Reading | Date | Source |
|---|---|---|---|
| BoJ policy rate | 1.25% (7–2), effective 24 Sep | 18 Sep 2026 | BoJ statement |
| 30Y JGB | 4.047% | 17 Sep 2026 | MoF daily table |
| 30Y all-time high | 4.131% | 1 Sep 2026 | MoF, series from 1999 |
| 10Y JGB | 2.993% | 17 Sep 2026 | MoF |
| 20Y / 25Y / 40Y JGB | 3.824% / 4.075% / 4.036% | 17 Sep 2026 | MoF |
| 1Y JGB | 1.581% | 17 Sep 2026 | MoF |
| 30Y minus 10Y | 105.4bp | 17 Sep 2026 | MoF, derived |
| US 30Y Treasury | 5.29% | 17 Sep 2026 | FRED DGS30 |
| JGB 30Y minus UST 30Y | −124.3bp (was −266.8bp at NIRP exit) | 17 Sep 2026 | derived |
| Core CPI, ex fresh food | 1.7% y/y (from 1.8%) | Aug 2026 | Statistics Bureau |
| Core-core CPI | 1.9% y/y | Aug 2026 | Statistics Bureau |
| USD/JPY | ~157.5–157.9, touched 158 | 18 Sep 2026 | Nikkei, Reuters, Bloomberg |
| BoJ monthly JGB buying | ¥2.3tn Oct–Dec 26 → ¥2.1tn Jan–Mar 27 → ~¥2tn from Apr 27 | plan of 16 Jun 2026 | BoJ |
| BoJ JGB holdings, projected | ~¥480tn end-Mar 2027, −17% vs end-Jun 2024 | 16 Jun 2026 | BoJ |
| FY2026 budget | ¥122.3tn, new bond issuance ¥29.6tn | FY2026 | Cabinet |
| FY2026 debt servicing | ¥31.3tn, +10.8%, on an assumed 3.0% rate | FY2026 | Cabinet |
| GPIF assets | ¥320.4tn; 25.59% domestic bonds | Q1 FY2026 | GPIF |
| GPIF domestic bond buying | ¥5.7tn in Apr–Jun, its largest quarterly domestic-bond purchase in recent years | Q1 FY2026 | GPIF |
| Life insurers, ultra-long JGBs | net sellers ~¥2.1tn | FY2025 | industry reporting |
| Japan's US Treasury holdings | ~$1.2tn, largest foreign holder | Dec 2025 / Jan 2026 | US TIC |
| Japan population 65+ | 29.8% | 2024 (latest World Bank observation) | World Bank |
The prior: what the outside view says
The base-rate leg does not ask what will happen. It asks what has happened to this variable, over this horizon, in classes of period resembling this one. It runs offline from a frozen slice of the Ministry of Finance's own daily curve (13,303 rows, 1974–2026) and it ships with this report as base_rate.py.
The horizon is 130 business days — 17 September 2026 to 31 March 2027, at the Tokyo bond market's own trading density of 244.3 days a year over the last decade.
Four reference classes, each applied to the 4.047 starting level:
| Reference class | n windows | effective n | median change | A | B | C | D |
|---|---|---|---|---|---|---|---|
| 1. Unconditional, 1999–2026 | 6,496 | 121 | +2.0bp | 4.79 | 33.36 | 51.22 | 10.64 |
| 2. After a trailing-year rise ≥50bp | 808 | 17 | +28.5bp | 0.50 | 20.54 | 45.92 | 33.04 |
| 3. Top decile of its own trailing 10-year range | 574 | 14 | +46.7bp | 0.00 | 0.00 | 48.43 | 51.57 |
| 4. Since the end of negative rates (19 Mar 2024) | 482 | 12 | +50.7bp | 0.00 | 0.00 | 39.83 | 60.17 |
| Equal-weight blend | 1.32 | 13.47 | 46.35 | 38.86 |
Three things in that table matter more than the blend.
The conditional classes are tiny. The window counts look reassuring — 482, 574, 808 — but the windows overlap almost completely. A moving-block bootstrap puts the effective sample size at twelve to seventeen independent observations per conditional class. Classes 3 and 4 are, in substance, a dozen draws from a single continuous two-and-a-half-year repricing. They are not four independent views of the world; they are one view, seen four ways.
Classes 3 and 4 return a literal zero for any outcome below 4.00 percent. That is not a finding about the future. It is the arithmetic consequence of the fact that in every 130-day window since the Bank left negative rates, the 30-year rose. A model that inherits that number uncritically has stopped forecasting and started extrapolating.
The unconditional class disagrees sharply with the other three, and it is the only one with a real sample. It puts 38 percent on a 30-year below 4 percent; the conditional classes put between zero and 21 percent. We have not down-weighted the unconditional class for disagreeing with the narrative, and we have not down-weighted the momentum classes for disagreeing with the unconditional. The published base case sits between them.
Analogue check. The nearest live analogue is January 2026, when the 30-year jumped about 30 basis points in four sessions — 3.462 to 3.765 on the Ministry of Finance's own closes, 16 to 20 January — while the 40-year reached 3.901. Capital Economics' contemporaneous account of 26 January put the intraday move above 3.8 percent with the 40-year over 4, after Prime Minister Takaichi pledged to "break free of the spell of excessive fiscal austerity." Both levels have since been exceeded on the close. The episode is the reason this report carries a jump process at all, and it is why the tail is a scenario rather than a rounding error.
The evidence update
No probability appears in this report without the evidence that moved it. Prior: the unconditional base rate, roughly 5 / 33 / 51 / 11.
| # | Evidence | Dated | Direction | Strength | Effect |
|---|---|---|---|---|---|
| 1 | BoJ hikes to 1.25%, guidance retained: "will continue to raise the policy interest rate" | 18 Sep 2026, BoJ statement | Higher yields | Moderate | Front end up; 30Y beta to front end is only 0.40, so a small direct push. C/D up slightly |
| 2 | 7–2 vote; both dissenters are the newest, government-nominated members | 18 Sep 2026, BoJ voting note | Lower policy path | Strong on policy, weak on the long end | Cuts the policy-path contribution; does not cut D, because a slower BoJ is a weaker yen and a more fiscally permissive regime |
| 3 | Takata and Tamura dissent that the price outlook is too dovish | 18 Sep 2026, BoJ note | Higher yields | Moderate | Widens the policy distribution both ways; raises dispersion, not the centre |
| 4 | Core CPI 1.7% y/y, below target a seventh month | Aug 2026, Statistics Bureau | Lower yields | Moderate | The single best argument for A and B. Raises B |
| 5 | Yen falls ~1.2% to ~157.9 on a hike | 18 Sep 2026 | Ambiguous | Weak, and tested | The yen channel is not in the daily data — see the rejected specification. No probability moved on this |
| 6 | BoJ buying falls to ¥2.1tn/month by Q1 2027; holdings −17% by end-March | plan of 16 Jun 2026 | Higher yields | Strong, and known | Deterministic supply withdrawal across the whole horizon. Principal support for C and D |
| 7 | BoJ pledges "nimble responses… increasing the amount of JGB purchases" if long rates rise rapidly | 16 Jun 2026 | Lower tail | Moderate | A real, stated backstop. This is what caps D short of a disorderly outcome, and is why the >5% threshold is not larger |
| 8 | Life insurers net sellers of ~¥2.1tn ultra-long JGBs in FY2025; FY2026 plans imply no return to net buying | FY2025/26 | Higher yields | Moderate, contested | The traditional marginal buyer of the 30-year has stepped back. Raises D |
| 9 | GPIF buys ¥5.7tn of domestic bonds in Apr–Jun, its largest quarterly domestic-bond purchase in recent years | Q1 FY2026, GPIF | Lower yields | Moderate, contested | A ¥320tn buyer rebalancing into the asset. Raises A and B. Directly contradicts #8. But the purchase was largely mechanical, and post-intervention its domestic-bond rebalancing capacity has turned negative — so it supports A only if it becomes discretionary |
| 10 | FY2026 issuance ¥29.6tn; debt service ¥31.3tn on an assumed 3.0% rate | FY2026 budget | Higher yields | Moderate | Supply, and a government now visibly sensitive to the level of the long end |
| 11 | 30Y minus 10Y is 105.4bp, versus 102.8bp at the end of negative rates — having reached 158.1bp in September 2025 | derived, MoF | Lower tail | Strong, and under-reported | The long-end slope is no wider today than at the end of negative rates, and sits 17bp below its own post-NIRP average of 122bp. A fiscal term-premium shock that had stuck would still be in the slope. It is not. Cuts D |
| 12 | The 30-year is unchanged over the last 21 trading days (−0.7bp) | MoF | Lower yields | Moderate | The momentum the conditional base-rate classes are built on has stalled. Cuts D |
Items 8 and 9 are a genuine, live contradiction about the identity of the marginal buyer, and we have not resolved it in favour of the more dramatic one. It is carried into the model as regime uncertainty, which is why this report's distribution is wide.
Item 11 is the finding we would put in front of a chief investment officer first, because it cuts against the story the headlines are telling.
The scenarios
A — Absorption. Below 3.50 percent. 5 percent.
The marginal buyer returns and is larger than the marginal seller. GPIF's rebalancing into domestic bonds continues at something like the Apr–Jun pace. That purchase was in large part mechanical — an equity rally forced a rebalance into bonds — and the late-July yen intervention has since reversed the arithmetic: on the same rebalancing logic, GPIF's capacity to add domestic bonds has turned negative while its capacity to add foreign bonds has risen. Scenario A therefore needs a discretionary domestic bid, not the one last quarter produced. a 4 percent-plus yen-denominated 30-year with no currency hedging cost is, for a Japanese liability-matching institution, the best asset available anywhere for the first time in a generation, and life insurers re-enter despite completed solvency positioning. Core inflation keeps undershooting, the Bank pauses after September, and the Bank's FY2027 end to tapering is pulled forward in expectation. The 30-year retraces most of a year's repricing.
Trigger: two consecutive quarters of net domestic institutional buying of super-long JGBs, with core CPI at or below 1.6 percent. Tripwire: the 30-year sustaining below 3.80 percent for four weeks, a level it last closed beneath on 14 July and last held for consecutive sessions in late June. Why it might be wrong: nothing in the last 130-day window of any momentum reference class delivers a move of this size in this direction.
B — Back below four. 3.50 to under 4.00. 20 percent.
The most conventional outcome, and the one the published sell-side forecasts imply. Inflation stays below target, the Bank holds through the October and December meetings, the long end grinds back into the range it occupied for most of 2026, and the last four months read in hindsight as an overshoot around a fiscal scare. Oxford Economics' June forecast of 2.8 percent on the 10-year, translated at the stable 104 basis-point spread, lands almost exactly here.
Trigger: no hike at either the 29–30 October or the 17–18 December meeting, with core CPI failing to reach 2 percent by December. Tripwire: the 10-year back below 2.75 percent.
C — The plateau. 4.00 to under 4.50. 40 percent. The base case.
The Bank delivers roughly one more hike, the taper proceeds exactly as published, issuance is heavy, domestic demand is adequate but not enthusiastic, and the 30-year does what it has done for the last month: very little. Yields are high, the curve's shape is stable, and Japan finishes its fiscal year with the highest long-bond yields in the tenor's history and no crisis. This is the outcome in which the repricing is real, structural, and orderly — and the one whose consequences for global portfolios are largest precisely because nothing breaks.
Trigger: one hike across the four remaining meetings; the 30-year holding a 4.00–4.40 range through the January Outlook Report. Tripwire: the 30–10 spread staying within 95–115bp. A plateau with a stable slope is this scenario; a plateau with a steepening slope is scenario D arriving slowly.
D — The term-premium regime. 4.50 percent or higher. 35 percent.
The supply arithmetic wins. The Bank's monthly buying falls to ¥2.1tn, life insurers stay out, foreign investors — who net bought ¥13.5tn of ultra-long JGBs in FY2025 and were the segment's largest support, but turned net sellers of ¥81.3bn in April 2026, their first net selling in sixteen months — do not step up to replace the Bank, and the government funds an expansionary programme into a market with no price-insensitive buyer left. The board's dovish drift makes this worse rather than better: a central bank that will not defend the currency with the policy rate leaves the long end to do the work. A weak auction, at the 25-year point where the curve already humps 2.8 basis points above the 30-year, is the most likely proximate trigger.
Trigger: a 30-year or 40-year auction with a tail materially wider than recent averages, followed within a fortnight by a failure to recover. Tripwire: the 30-year closing above 4.35 percent for five consecutive sessions, or the 30–10 spread above 120bp. What would falsify it: the Bank executing its stated "nimble response" — an increase in purchase amounts or a fixed-rate operation. That pledge is in the 16 June plan and it is the reason this scenario stops where it does.
The simulation
Simulation results. A real Monte Carlo was run: 50,000 paths, 130 daily steps, seed 20260918, numpy 2.2.6. The code and both frozen data files ship with this report as mc.py, jgb_slice.csv and us_fred.csv, and reproduce every figure below.
The specification, fitted not assumed. Over the post-NIRP-exit sample (19 March 2024 to 17 September 2026, 587 daily observations), ordinary least squares on daily first differences gives:
Δ(30Y) = 0.263bp/day + 0.4005 × Δ(1Y JGB) + 0.1435 × Δ(US 30Y) + ε, R² = 0.0661, sd(ε) = 3.59bp, AR(1) of ε = 0.016
The R² is the result. Ninety-three percent of the daily variance of Japan's 30-year bond is explained by neither the front end of its own curve nor the US long bond. Before the exit from negative rates, the beta to the front end was 0.70; it is now 0.40. In levels the two remain tightly linked — the correlation of the 1-year and 30-year levels over the same period is 0.963 — so the long end has followed the normalisation trend closely. It is the day-to-day price that has been handed to someone else. That someone is the domestic buyer base, and that is why this report's scenarios are supply-and-demand scenarios rather than policy scenarios.
A specification was tested and rejected, and the test ships in the code. The most popular explanation of the last month — that a weak yen drives the long end — is not in the data. Δ(30Y) on Δ(USD/JPY) alone returns R² = 0.0041. Adding log USD/JPY to the adopted specification moves R² from 0.0649 to 0.0650, and the coefficient comes out negative (−0.065): over this sample a weaker yen is associated, trivially and insignificantly, with a lower 30-year yield. We could not find the yen channel at daily frequency and we are not going to assert it.
The policy leg runs an explicit reaction function over the four scheduled Monetary Policy Meetings before resolution — 29–30 October 2026, 17–18 December 2026, 21–22 January 2027 and 17–18 March 2027, from the Bank's published calendars — with a persistent committee disposition carried across meetings, which is how a 7–2 split propagates. Output: mean 1.48 hikes by 31 March; no hike 17.0 percent, one 35.7, two 31.6, three or more 15.7; mean policy rate 1.62 percent.
That sits deliberately between two anchors that disagree. The 1-year JGB at 1.581 percent stands 33bp above the new policy rate; taken at face value as an average of expected overnight rates, it implies a policy rate near 1.90 percent in a year, roughly 2.6 hikes. Published sell-side terminal views cluster lower — Oxford Economics and BofA at 1.5 percent, Mizuho Securities at 1.75 percent via January and June 2027, and State Street at 1.25 — a level today's decision has already reached, and one the house said would become 1.5 if the yen breached 160. The market is more hawkish than the analysts. The Desk did not resolve that; it split it.
The term-premium leg is a three-regime mixture — Absorption −0.15bp/day, Attrition +0.18, Dislocation +0.55, weighted 0.30 / 0.50 / 0.20 — with a jump process calibrated to the January 2026 episode. Each drift is anchored to something realised: the Dislocation figure is slightly worse than the 30-year's realised +0.49bp/day over the last 130 trading days; the Absorption figure is milder than the −16.6bp the 30-year actually delivered between 1 and 4 September 2026 (4.131 → 3.965).
We declined to extrapolate our own fitted drift. The estimated constant is +0.263bp/day, which over 130 days is +34bp of pure unexplained upward drift. Adopting it as a forecast — which the regression invites — produces 0.1 / 3.7 / 36.9 / 59.3. That run is in the shipped output as counterfactual_adopt_fitted_drift, and the Desk does not publish it. A fitted drift estimated across a one-way repricing is momentum wearing a regression's clothes.
Results.
| A <3.50 | B 3.50–4.00 | C 4.00–4.50 | D ≥4.50 | |
|---|---|---|---|---|
| Simulation, 50,000 paths | 0.77 | 14.60 | 42.13 | 42.50 |
Median 4.418%. P10 3.899, P25 4.138, P75 4.717, P90 5.037, P95 5.236. P(≥4.50) 42.5%, P(≥5.00) 11.2%, P(touching 4.50 at any point) 50.2%.
Dispersion, measured and disclosed. Achieved 130-day standard deviation 44.3bp against an empirical unconditional 34.0bp — the model is 30 percent over-dispersed against history, and we are publishing it that way. The reason is diagnostic rather than technical: switching the idiosyncratic noise term off entirely still leaves 42.2bp of dispersion from the regime mixture alone. Structural uncertainty, not noise, sets the width of this distribution. The historical 34bp was generated inside one policy regime; this report's whole claim is that the regime is the open question. Over-dispersion is the conservative direction. An earlier calibration that matched one leg to history and let the others accumulate on top left the model 61 percent over-dispersed; that error was found and corrected, and the corrected routine is what ships.
Seed stability, 25,000 paths across four seeds: D at 42.1 / 41.8 / 42.2 / 42.7. Sampling error is not the uncertainty that matters here.
Specification envelope, 27 runs across regime weights, drifts and the front-end beta: A [0.1, 4.1] · B [4.2, 33.8] · C [25.8, 46.9] · D [18.3, 69.9]. The structural envelope on D is 51 points wide against a sampling envelope under 1 point — roughly sixty times larger. One parameter dominates it: the term-premium drift, which is a judgement, not an estimate. That is this report's weakest joint and it leads the red team.
MiroFish was not run. No multi-agent simulation input is claimed anywhere in this report.
Aggregation
| Leg | Weight | A | B | C | D |
|---|---|---|---|---|---|
| Base rate | 0.35 | 1.32 | 13.47 | 46.35 | 38.86 |
| Expert prior | 0.20 | 8.28 | 49.12 | 38.76 | 3.84 |
| Simulation | 0.45 | 0.77 | 14.60 | 42.13 | 42.50 |
| Aggregate | 2.46 | 21.11 | 42.93 | 33.49 | |
| Published | 5 | 20 | 40 | 35 |
The expert leg is down-weighted to 0.20, and here is why. The resolution variable is the 30-year. Only two dated forecasts of any JGB yield could be verified to a primary document, and both are 10-year calls that the market has already overtaken: Oxford Economics raised its end-2026 10-year forecast to 2.8 percent from 2.5 on 8 June 2026, and the 10-year closed at 2.993 on 17 September; Capital Economics' group chief economist wrote on 26 January 2026 that 10-year yields "could eventually rise to 3%," and eventually took about eight months. A leg resting on two overtaken forecasts of the wrong tenor, translated to the right tenor by a declared +104bp spread adjustment, does not deserve a third of the answer. It is the most dovish leg by a wide margin, and it is the reason bucket B survives at 20 percent.
The rounding is disclosed. The unrounded aggregate puts A at 2.46. The nearest-5 grid that minimises displacement would publish A at zero — and a published zero asserts impossibility, which the Desk is not entitled to on a bucket the ensemble genuinely weights. The rounding rule therefore forbids a 0 on any bucket above 1.0 percent, at a cost of 0.94 percentage points of extra displacement. Rounding took A up (2.46 → 5, widening the rally tail, the conservative direction) and C down (42.93 → 40).
Where the legs disagree. On bucket D: the base rate says 38.9, the simulation 42.5, the expert leg 3.8. That is not noise around a consensus; the published forecasts and the historical record are describing different worlds. The published 35 sits closer to the two quantitative legs because the expert leg's own track record over the last six months is the evidence against it.
Market pricing versus the Desk view
What the market is pricing. The 1-year JGB implies a policy rate near 1.90 percent within a year — more tightening than any published sell-side terminal view. The 30-year sits at 4.047 with the 25-year 2.8 basis points above it: a hump, not a slope, which is what a market clears at when the heaviest supply and the thinnest demand meet at a specific point on the curve rather than at the very end of it.
What looks mispriced to the Desk. The 10-year is the expensive point. It is the tenor the Bank still buys most of, the tenor every forecast is written about, and it sits 105bp through the 30-year — the same spread as at the end of negative rates, and 17bp inside the 122bp average of the two and a half years in between. That is not a spread that never moved; it reached 158bp a year ago. It is a spread that went out and came back, which is a more awkward fact for the term-premium story than a flat line would be. Either way it is where a long-horizon owner learns the most per basis point of attention.
What consensus is missing. Two things. First, that the long-end slope is no wider now than before normalisation began, having widened 55bp and given it all back — which sits awkwardly with nearly every published account of a fiscal term-premium shock. Second, that the daily price of the 30-year stopped responding to monetary policy some time ago, which means the standard analytic reflex — forecast the BoJ, infer the curve — has a correlation of roughly nothing to work with.
What would prove the Desk wrong. A sustained steepening of the 30–10 spread beyond 120bp with the policy rate unchanged would show the term premium moving on its own and would mean bucket D deserved more than 35. A drop back below 3.80 percent on visible domestic buying would show the marginal buyer never left and would mean the momentum classes contaminated the base rate.
Consistency with the Desk's live 10-year call
On 1 September the Desk published a distribution for the 10-year JGB on this same resolution date: below 2.75 percent 20 percent, 2.75 to 3.25 50 percent (the base case), 3.25 to 3.60 20 percent, 3.60 or higher 10 percent — P(at or above 3.25) = 30 percent. That edition named today's Bank of Japan meeting as one of its four interim checkpoints, with the instruction to watch the dissent count, not the outcome. The dissent count was four, in two directions.
Translated at the +104bp spread this report uses in its expert leg, today's 30-year distribution implies a 10-year median near 3.38 percent, with roughly 61 percent above 3.25 and 31 percent above 3.60 — about double and about triple the weights published seventeen days ago. Some of that is seventeen days of news, including today's hike. Most of it is that the two editions do not use the same view of the term structure: the 1 September edition modelled the 10-year directly, and this one models the 30-year and would have to impose a spread to speak to the 10-year at all.
We are logging today's edition as a material update to the 1 September 10-year weights rather than letting two live Desk forecasts resolve against each other on 31 March, and the calibration record carries the cross-reference in both directions. A desk that publishes a public scorecard does not get to run two incompatible forecasts of the same curve into the same grading date and keep whichever one lands.
What it reprices for a universal owner
Analytical, not advisory. The magnitudes are directional bands, not recommendations.
| Asset class | Direction if C (plateau) | Direction if D (≥4.50) | Magnitude band | Why |
|---|---|---|---|---|
| Global long-duration sovereign | Mildly negative | Negative | Moderate–large | Japan has been the marginal supplier of cheap duration; at 4%+ it competes for it |
| US Treasuries, long end | Neutral–negative | Negative | Moderate | Japan is the largest foreign holder, ~$1.2tn. The gap to the US 30-year has narrowed by more than half since 2024, from −267bp to −124bp |
| Euro sovereigns | Neutral–negative | Negative | Moderate | Same hedged-yield arithmetic, smaller stock |
| Global investment-grade credit | Neutral | Negative | Moderate | Japanese lifers and banks have been structural buyers; a domestic alternative at 4% is new |
| Private credit and direct lending | Neutral | Negative | Moderate–large | The asset class was sized against a zero risk-free rate for Japanese capital. It no longer is |
| Infrastructure and core real assets | Neutral–negative | Negative | Large | Long-dated, discount-rate-sensitive, and held by exactly the institutions repricing their liabilities |
| Real estate, core | Negative | Negative | Large | Cap rates versus a domestic 4% alternative |
| Japanese equities | Mixed | Negative | Moderate | Banks and insurers benefit; leveraged domestics and long-duration growth do not |
| Japanese banks and insurers | Positive | Mixed | Moderate | Reinvestment yield improves; mark-to-market on legacy holdings does not |
| Global equities, long-duration growth | Mildly negative | Negative | Moderate | The global discount rate has a Tokyo component again |
| Yen | Mildly stronger | Ambiguous | Wide | A higher long end can mean repatriation (yen-positive) or loss of confidence (yen-negative). We do not claim to know which |
| Gold and reserves | Neutral | Positive | Moderate | A reserve-currency issuer with a contested long end raises diversification demand |
| Japanese pension and insurance liabilities | Positive | Positive | Large | Higher discount rates reduce the present value of liabilities. For a Japanese plan sponsor, this is the best news in thirty years |
That last row is the one most often dropped from the coverage, and it is why a rising JGB yield is not a Japanese disaster. The sign is the opposite of the intuition: for an owner with long-dated yen liabilities, the repricing improves the funded position. The pain sits with the holders of the old, low-coupon stock — and with everyone outside Japan who has been quietly borrowing the world's cheapest duration.
Second- and third-order effects
- The carry trade's funding leg reprices. A 1.25 percent policy rate and a 4 percent 30-year change the arithmetic of yen-funded positions everywhere, not by making them impossible but by making them cost something.
- Repatriation is a stock effect, not a flow event. Japanese institutions hold roughly $1.2tn of US Treasuries alone. They do not need to sell for this to matter; they need only to stop reinvesting. The transmission is slow, large, and nearly invisible in monthly data.
- Global term premia acquire a Tokyo component. For twenty years the price-insensitive JGB buyer suppressed a global anchor. The Bank's holdings fall to about ¥480tn by end-March 2027, 17 percent below their peak, on a published schedule.
- The government's cost of carry is now a live political variable. Debt servicing is ¥31.3tn in FY2026, up 10.8 percent, against an assumed 3.0 percent rate that the 30-year has already passed. Every further basis point is a line in a future budget.
- Appointments become a market variable. If the composition of the Policy Board is understood to track the fiscal stance of the government, then board vacancies price bonds. Today's 7–2 is the first clean data point.
- The hedged-yield inversion reverses. Years of a wide BoJ–Fed differential made hedged foreign bonds uneconomic for Japanese buyers. Domestic 4 percent paper with no hedging cost is the first genuinely competitive liability-matching asset in a generation.
- Solvency regimes amplify. With Japan's economic-solvency framework largely bedded in, insurers no longer need super-long duration to pass a test — which removes a structural, price-insensitive bid precisely when supply is heaviest.
- Demography is the slow current under all of it. Close to thirty percent of Japan's population is 65 or over. A decumulating population draws down savings; the domestic buyer base for JGBs shrinks by arithmetic, regardless of yield.
Watch dashboard
| # | Indicator | Current | Threshold that changes the model |
|---|---|---|---|
| 1 | 30Y JGB, MoF daily | 4.047% (17 Sep) | >4.35% for 5 sessions → D live. <3.80% for 4 weeks → A live |
| 2 | 30Y minus 10Y spread | 105.4bp | >120bp → +5 to D. <95bp → +5 to B |
| 3 | 25Y minus 30Y hump | +2.8bp (25Y above) | Hump widening → auction stress at the pressure point |
| 4 | BoJ Summary of Opinions, Sept MPM | due 1 Oct 2026 | Hawkish dissent language → policy leg up |
| 5 | BoJ Minutes, Sept MPM | due 5 Nov 2026 | Full Asada/Sato reasoning; board-drift evidence |
| 6 | MPM, 29–30 Oct 2026 (Outlook Report) | — | Hike → +5 to C/D. Hold with dovish Outlook → +5 to B |
| 7 | MPM, 17–18 Dec 2026 | — | Second consecutive hold → B live |
| 8 | MPM, 21–22 Jan 2027 (Outlook Report) | — | The decisive meeting inside the horizon |
| 9 | Core CPI ex fresh food | 1.7% y/y (Aug) | ≥2.2% → +5 to D. ≤1.4% → +5 to B |
| 10 | 30Y and 40Y auction tails | 3 Sep 30Y cleared 4.080% (from 3.937%); tail 0.28 vs 0.21 in August, bid-to-cover 3.79 vs 3.86 — the wires split on whether that was weak | Tail above ~0.35, or two consecutive auctions wider than the trailing average → D trigger |
| 11 | Life insurer net super-long JGB flows | net sellers ~¥2.1tn FY2025 | Two quarters of net buying → A live |
| 12 | GPIF quarterly domestic bond purchases | ¥5.7tn Apr–Jun | Repeat ≥¥4tn → +5 to A/B |
| 13 | Foreign net flows in ultra-long JGBs | +¥13.5tn FY2025; −¥81.3bn in April 2026, the first net selling in 16 months | Two further months of net selling → +5 to D |
| 14 | BoJ "nimble response" — any unscheduled purchase increase or fixed-rate operation | not triggered | Triggered → caps D; the single cleanest tripwire in the model |
| 15 | USD/JPY | ~157.5–158 | >165 → political pressure channel, ambiguous sign. Intervention → regime change |
| 16 | Japan's US Treasury holdings, TIC | ~$1.2tn | Three consecutive monthly declines → repatriation confirmed |
Indicator 14 is the one we would watch above all others. It is observable, it is binary, and the Bank has already told the market it exists.
Red team: how this could be wrong
1. The term-premium drift is a judgement dressed as a parameter, and it dominates everything. The specification envelope moves bucket D from 18.3 to 69.9 percent — a 51-point range — and almost all of that is the three-regime drift mixture. The sampling envelope is under one point. We chose −0.15 / +0.18 / +0.55 bp/day with weights 0.30 / 0.50 / 0.20, anchored to realised moves, but they are choices. A reader who disagrees with those three numbers disagrees with this report's answer, and is entitled to. This is the weakest joint and everything else is stronger than it.
2. The 11 percent probability of a 30-year above 5 percent is the least supported number here. The base rate says 0.3 percent. The entire gap is the Dislocation regime plus the jump process. A 95bp rise in 130 days would be roughly a 2.8-standard-deviation move against the unconditional history of this tenor, and Japan has a central bank that has explicitly promised to intervene against exactly that. We publish it as a threshold rather than a bucket precisely because we do not want it read as a forecast.
3. The most important single fact in this report argues against its own tail. The 30-year-minus-10-year spread is 105.4bp today and was 102.8bp when negative rates ended. In between it reached 158.1bp — on 8 September 2025 — and averaged 122.2bp. So the term premium did move; it simply is not there now. If Japan were in a fiscal term-premium shock today, the slope would show it, and the slope has round-tripped instead. Capital Economics' January analysis makes the stronger version of this case: on their high-frequency estimates Japan's general government deficit last year was nearer 0.5 percent of GDP than the 2.5 percent the IMF and others assume, and net debt could fall toward 100 percent of GDP — so the ultra-long sell-off, in their reading, reflects reflation and a higher neutral real rate rather than fiscal stress. If they are right, this is a repricing to a new equilibrium and it is largely done; bucket B deserves more than 20 and bucket D much less than 35. We have given that view 20 percent of the ensemble and it is the most respectable case against us.
4. The conditional base-rate classes may be contaminated beyond repair. Classes 3 and 4 have effective sample sizes of 14 and 12 and return exactly zero for any outcome below 4 percent. We used them anyway, at 35 percent weight across four equally weighted classes. A reasonable critic would drop them entirely and run the unconditional class alone — which would move the answer toward 5 / 33 / 51 / 11 and cut D by more than two-thirds.
5. We could not verify a single dated 30-year JGB forecast from a named institution. Not one. The expert leg of a three-leg ensemble is carried by two 10-year forecasts, both overtaken, translated by a spread assumption. We disclosed it and down-weighted it, but disclosure is not the same as having the input.
6. The horizon contains four central bank meetings and one fiscal year-end. March 31 is a Japanese fiscal-year boundary, a date on which domestic balance-sheet behaviour is not representative of the rest of the year. A single-day resolution on that date carries window-dressing risk that a monthly average would not. We chose it for unambiguous resolvability and we accept the cost.
7. The 18 September close is not in our data. The Ministry of Finance had not published it when this report was built; we anchored on 17 September and verified the absence rather than assuming it. If the 18th moved materially on the decision, every level in this report starts one day stale, in a direction we cannot sign.
Methodology
Probabilities are the UAO Probability Desk's, weighted across three inputs by a written rule: a base rate from four reference classes on the Ministry of Finance's own daily JGB curve (weight 0.35), published and dated expert priors (0.20, down-weighted for the logged reason above), and a 50,000-path Monte Carlo (0.45). Weights sum to 100 and are rounded to the nearest 5 percent, with a rule forbidding a published zero on any bucket the ensemble weights above 1 percent. Every figure in this report traces to a named, dated source or to code that ships with it: base_rate.py, mc.py, aggregate.py, make_chart.py, and the frozen data slices jgb_slice.csv and us_fred.csv. MiroFish was not run and no multi-agent simulation input is claimed. The forecast is logged to the public calibration record and will be graded against the Ministry of Finance's published 31 March 2027 value.
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.
Source ledger
| # | Source | Date | Data point used | Conf. | Moves model |
|---|---|---|---|---|---|
| 1 | Bank of Japan, Change in the Guideline for Money Market Operations (k260918a.pdf) | 18 Sep 2026 | 1.25%, 7–2 vote, effective 24 Sep, all four dissents, forward guidance | H | Yes |
| 2 | Bank of Japan, Plan for the Outright Purchases of JGBs (k260616b.pdf) | 16 Jun 2026 | Taper schedule to ¥2tn; ¥480tn end-Mar 2027; "nimble responses" pledge; 7–1 vote, Tamura dissent | H | Yes |
| 3 | Bank of Japan, Scheduled Dates of Monetary Policy Meetings in 2026 | 31 Jul 2025 | Oct 29–30, Dec 17–18 meeting dates | H | Yes |
| 4 | Bank of Japan, Scheduled Dates of Monetary Policy Meetings in 2027 | 31 Jul 2026 | Jan 21–22, Mar 17–18 meeting dates | H | Yes |
| 5 | Japan Ministry of Finance, Interest Rate (JGB), daily, historical file | to 31 Aug 2026 | Full curve 1974–2026; base rate; all levels | H | Yes |
| 6 | Japan Ministry of Finance, Interest Rate (JGB), current month | to 17 Sep 2026 | September curve; 4.047% anchor; absence of the 18th | H | Yes |
| 7 | FRED, DGS30 / DGS10 (US Treasury constant maturity) | 17 Sep 2026 | US 30Y 5.29%, 10Y 4.94%; estimation of β₂ | H | Yes |
| 8 | FRED, DEXJPUS | 11 Sep 2026 | 153.71; the rejected yen specification | H | Yes (as a rejection) |
| 9 | Oxford Economics, Yasuko Koido, "Why we expect JGB yields to stay high" | 8 Jun 2026 | End-2026 10Y forecast raised to 2.8% from 2.5% | H | Yes |
| 10 | Oxford Economics, Yasuko Koido, "The BoJ hiked a policy rate and will end the tapering…" | 16 Jun 2026 | Terminal 1.5%, hikes ~every six months; taper ends FY2027 | H | Yes |
| 11 | Capital Economics, Neil Shearing, "Ignore the Truss comparisons" | 26 Jan 2026 | 30Y +30bp to >3.8% and 40Y >4% in Jan; deficit ~0.5% of GDP estimate; 10Y "eventually" 3%; r* ~1% | H | Yes |
| 12 | Capital Economics, "BoJ undermines yen rally, again" (Capital Daily) | 18 Sep 2026 | Same-day house read on the decision's yen effect | M | Contextual |
| 13 | Capital Economics, "Fiscal loosening need not spell trouble" (Japan Economics Weekly) | 18 Sep 2026 | Same-day house view on the fiscal channel | M | Contextual |
| 14 | Statistics Bureau of Japan, Consumer Price Index, August 2026 release (stat.go.jp / e-Stat) | Aug 2026 | Core CPI 1.7% y/y, core-core 1.9%, headline 1.9%; 7th month below target | H | Yes |
| 15 | Nikkei Asia, "Yen touches 158 to the dollar…" | 18 Sep 2026 | Yen level and market interpretation | M | Contextual |
| 16 | Bloomberg, "Yen Drops Against Dollar After BOJ Raises Rates as Expected" | 18 Sep 2026 | Yen reaction | M | Contextual |
| 17 | Bloomberg, "BOJ Hikes Rates at Fastest Pace Since 1990…" | 18 Sep 2026 | Pace characterisation (attributed, not adopted) | M | No |
| 18 | Reuters via Investing.com, "BOJ Governor Ueda's comments at news conference" | 18 Sep 2026 | Ueda press-conference tone | M | Contextual |
| 19 | FXStreet, "Japanese Yen stays weak near two-week low on dovish BoJ hike" | 18 Sep 2026 | USD +1.2% to 157.897 | M | Contextual |
| 20 | XTB, "Chart of the Day: Yen in a Trap" | 18 Sep 2026 | Identifies Asada and Sato as the newest members, Takaichi-nominated | M | Yes |
| 21 | CNBC, "Bank of Japan raises interest rates to 31-year high" | 18 Sep 2026 | 31-year-high framing; JGB context | M | Contextual |
| 22 | Cabinet of Japan FY2026 budget, as reported (Kyodo/Tribune/Malay Mail) | FY2026 | ¥122.3tn budget; ¥29.6tn issuance; ¥31.3tn debt service at an assumed 3.0% | M | Yes |
| 23 | Reporting on the Takaichi stimulus package | 2026 | ¥21.3tn package; three pillars | M | Contextual |
| 24 | GPIF quarterly results, as reported | Q1 FY2026 | ¥320.4tn AUM; 25.59% domestic bonds; ¥5.7tn domestic bond purchases | M | Yes |
| 25 | Chief Investment Officer, "Foreign, Domestic Equities Propel Japan GPIF" | 2026 | GPIF return and allocation context | M | Contextual |
| 26 | Aviva Investors, "Bond Voyage — Dancing to a new tune" | Feb 2026 | Lifers as anchors for 20/40-year demand; J-ICS solvency effects | M | Yes |
| 27 | Reporting on life insurer FY2026 investment plans | 2026 | Net sellers ~¥2.1tn ultra-long JGBs FY2025; no return to net buying expected | M | Yes |
| 28 | Bloomberg, "Japan's Super-Long Bonds See First Foreign Outflow Since 2024", reporting MoF/JSDA flow data | ~20 May 2026 (April 2026 flow) | −¥81.3bn in April 2026, first net selling in 16 months; FY2025 foreign net purchases +¥13.5tn | M | Yes |
| 29 | Japan MoF, Auction Result of 30-Year JGBs | 3 Sep 2026 | Cleared 4.080%, from 3.937% previously | H | Yes |
| 30 | Vantage Markets, "Japan 30-Year Bond Yield Nears Record Ahead of Auction" | 3 Sep 2026 | Record framing; Takaichi budget-request context | L | Contextual |
| 31 | Bloomberg, "Japan's 10-Year Bond Yield Hits 3% for First Time Since 1996" | 1 Sep 2026 | 10Y 3% milestone — independently re-derived from MoF (last ≥3.0% before 2026 was 6 Sep 1996) | H | Yes |
| 32 | US Treasury TIC, Major Foreign Holders | Dec 2025 / Jan 2026 | Japan ~$1.2tn, largest foreign holder | H | Yes |
| 33 | Congressional Research Service RS22331, Foreign Holdings of Federal Debt | 2026 | Japan 12.8% of foreign-held US federal debt | H | Contextual |
| 34 | World Bank, NY.GDP.MKTP.CD, Japan | 2025 | GDP $4.435tn | H | Contextual |
| 35 | World Bank, SP.POP.65UP.TO.ZS, Japan | 2025 | 30.0% of population aged 65+ | H | Yes |
| 36 | Oxford Economics, "The BoJ will likely hike higher and faster" | 2026 | Hike-path revision framing | M | Contextual |
| 37 | BofA forecast of BoJ terminal rate | 2026 | Terminal 1.5% | M | Yes |
| 38 | State Street (Loo), BoJ base case, as reported in 18 Sep 2026 decision coverage | 18 Sep 2026 | Terminal 1.25%, rising to 1.5% if USD/JPY breaches 160 | M | Yes |
| 39 | Mizuho Securities, Global Economy and Finance Weekly Watch | 14 Sep 2026 | +25bp at Jan 2027 and Jun 2027 → 1.75% terminal | M | Yes |
| 40 | Berenberg, "Reforms in Japan — Risks for the Rest of the World" | 2026 | Cumulative ~$1.6tn Japanese outflow into global bonds | L | Contextual |
| 41 | BNN Bloomberg, "How Japan's bond rout is turning the tide of global capital" | 2 Sep 2026 | Repatriation framing | L | Contextual |
| 42 | Société Générale, "Any respite for Japan's bonds is likely to be short-term" | 2026 | House directional view on JGBs | L | Contextual |
| 43 | Euronews, "Bank of Japan hikes rates to 31-year high to battle inflation" | 18 Sep 2026 | Decision confirmation, second source | M | Confirmatory |
| 44 | Seoul Economic Daily, "Bank of Japan Raises Key Rate to 1.25%" | 18 Sep 2026 | Decision confirmation, third source | L | Confirmatory |
| 45 | GuruFocus report of the BoJ decision | 18 Sep 2026 | 1.00% → 1.25%. This aggregator alone dates the level to April 1995; the Desk publishes "31 years" and does not adopt the month | L | Confirmatory only |
Confidence: H = primary document or official statistical series; M = established institution or major wire; L = secondary aggregator, used for direction only and never as a sole source for a figure.
The Probability Desk is the afternoon intelligence property of Universal Asset Owners. Probabilities are the Desk's own, weighted across base-rate, expert-prior and simulation inputs, and every published scenario is logged and graded when its horizon resolves.
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.