Forecast journal — 1 October 2026

1.Calls

Both yield ranges are meant to hold about 3 times in 4.

2.How I reached the probabilities

I started from the framework's working view: B 50%, A 25%, C 25%. Then I updated it on what the data room shows since then:

• A falls 10 points to 15%, because there's no growth trigger and the Fed is hiking, not cutting.
• B gains 5 points. The vote model called B every month from May to August, and September was A1 B2 C0.
• C gains 5 points, for fiscal pressure, global pressure and rising term premium.

We differ by 5 points between B and C. That gap comes down to one thing, inflation expectations, which are still anchored. If the 5y5y crosses 2.6%, I move to your numbers.

3.Three reasons

1. The sell-off is driven by real rates and the Fed, not inflation fear, which is B's signature. Over 12 months the 10-year rose 110bp and the 10-year TIPS yield 111bp (1.80% to 2.91%), while the 10-year breakeven was flat at 2.36%. The 5y5y is 2.36%, inside the B band. The Fed raised its upper bound to 4.00% in September, and 2s30s flattened from 113bp to 70bp. A C regime needs de-anchoring or a bear steepening, and neither is here yet.
2. Fiscal and global pressure is real, which keeps C at 30%.
   • Interest/receipts (BEA) is 21.1%, inside the C band above 20%.
   • The deficit is 5.4% of GDP with unemployment at 4.1%.
   • Term premium is rising: Kim-Wright 0.50 to 1.02, ACM 0.61 to 0.85.
   • JGB yields have jumped: the 10-year from 1.66% to 3.06% and the 30-year to 4.10%, a growing pull on Japanese money.
   • The foreign official share of Treasuries fell from 13.0% to 12.0%.
   • Oil is $96, core PCE is 3.0% and supercore is 3.5%.
3. A has no trigger, and bond demand is still healthy. Unemployment fell from 4.4% to 4.1%, HY spreads are a tight 3.1%, and nominal GDP is growing 6.3%, 1.9pp above the 10-year yield. September's 30-year auction came 6.2bp through, with indirect bidders taking 79% and bid-to-cover at 2.61. Buyers aren't striking.

4.Why these yield ranges, and why yours are too narrow

• Method: I took every 12-month change since 1962, using the subset that started from a 4.5–6.5% level, and applied it to today's yields. I checked the result against current volatility: the 10-year's realised volatility is about 0.7pp a year, the 30-year's about 0.6pp. I then tilted the range upward for the 30% C weight.
• 10-year, 4.25–6.50%: history puts about 75–77% of outcomes in this range.
  • Your 5–7% catches only 52–61%.
  • The upper bound of 7% is fine (only about 4% of outcomes go above it).
  • The floor is the problem: 5% is just 26bp below today, and the 10-year ended a year more than 26bp lower in 34–45% of past cases.
• 30-year, 4.90–6.75%: history puts about 67–81% of outcomes in this range.
  • Your 6–9% catches only 13–30%. The 30-year finished below 6% in 69–87% of comparable years.
  • Reaching 9% needs a rise of 3.4pp. The largest 12-month rise on record is +3.5pp, in 1980.
  • If C is your conviction, express it as 5.25–7.0%, not a higher floor.

5.What would change my mind

• Toward C:
  • 5y5y above 2.6% for two month-ends (the C band starts at 2.75%).
  • The TIPS yield falling while the 10-year rises, the vote model's C rule.
  • 2s30s back above 120bp while the Fed holds or cuts.
  • 30-year auction tails above 3bp twice running, with indirect bidders below 60%.
  • ACM term premium above 1.25.
• Toward A:
  • Unemployment at 4.6% or higher, a 0.5pp rise.
  • HY spreads above 4.5%.
  • Core PCE below 2.5%.
  • The Fed cutting while the 30-year falls (an A vote on policy).

6.Readings that would halve a probability

1. 5y5y at 2.75% or higher at two consecutive month-ends: halve B (55% to about 28%). Most of the freed weight goes to C.
2. Unemployment at 4.6% or higher and HY spreads at 4.5% or higher in the same month: halve B. A rises to about 35%.
3. 5y5y at 2.10% or lower, ACM term premium at 0.40 or lower, and 30-year auction indirect share at 70% or higher for three auctions: halve C (30% to 15%).

7.Dashboard at time of writing (month-end Sep 2026)

2y 4.89 · 10y 5.26 · 30y 5.59 · 5y5y 2.36 · 10y TIPS 2.91 · ACM TP 0.85 · KW TP 1.02 · fed funds upper 4.00 · core PCE 3.0% y/y · unemployment 4.1% · interest/receipts 21.1% · deficit 5.4% GDP · JGB10 3.06 · WTI 96 · HY OAS 3.12 · 10y realised vol 82bp · votes A1 B2 C0.

Scored on 1 Oct 2027 using journal/forecasts.csv. Past entries are never edited; revisions go in a new dated entry.

Appendix. The watch metrics explained

Added the same day (1 Oct 2026). All history below is computed from the data room unless marked otherwise. "Percentile" means the share of past month-ends below today's reading.

Summary

A1. 5-year, 5-year forward inflation expectation (5y5y)

• What it is. The average inflation rate the bond market expects for the five years that start five years from now (years 6 to 10). Series infl_5y5y, FRED T5YIFR, daily since 2003.
• How it is computed. From nominal and TIPS yields: breakeven = nominal yield minus real yield, at 5 and 10 years. Then 5y5y = ((1 + BE10)^10 / (1 + BE5)^5)^(1/5) − 1.
• What it measures. Long-run inflation credibility. Looking five years out strips out the current oil price and cycle, so it shows whether investors believe the Fed will get inflation back to target. It includes a small inflation-risk premium and a TIPS-liquidity discount.
• The levels. The framework bands are A below 2.0%, B 2.0–2.6% and C at 2.75% or higher. The Fed's 2% target is on PCE, while TIPS pay CPI, which runs about 0.3pp higher, so roughly 2.3% is "on target". At 2.6% the market is pricing persistent overshoot; at 2.75% it is pricing a lasting loss of credibility.
• What would push it up. Fed independence in doubt; the Fed cutting while core inflation is above 3%; a persistent oil or tariff shock that feeds wages; fiscal dominance (deficits financed with the central bank's help); heavy TIPS demand as an inflation hedge.
• History. Median 2.29%, middle half 2.11–2.46%, 95th percentile 2.75%. It was at 2.6% or higher in 13% of months and 2.75% or higher in only 6%, and has not reached 2.6% since July 2014. The 2.75% crossings were almost all in 2010–13, when QE was expected to raise inflation (it didn't). The 2008 extremes (2.97% then 0.48%) reflect a collapse in TIPS market liquidity, not real expectations. Assessment: 2.75% is a genuinely rare reading. Reaching it now, with the Fed hiking rather than doing QE, would be a stronger signal than in 2010–13. The halve-C level of 2.10% is common (25% of months, most recently March 2026), so that trigger leans on its other two conditions.

Month-end readings, 2003 to Sep 2026 (%). Bold = 2.60 or higher (the "change my mind toward C" level); italics = 2.10 or lower (the halve-C level). Same data with months as rows and years as columns, for a spreadsheet: data/tables/infl_5y5y_month_end_months_rows.csv.

Reading the grid: the bold clusters are 2010–13 (QE-era reflation hopes) and the italic run is 2015–20 (low-inflation years). Since 2021 every month-end has sat between 2.06 and 2.48, inside the B band.

A2. 10-year TIPS real yield falling while the 10-year rises

• What it is. The vote model's C rule: over three months, the 10-year TIPS real yield (tips10y, FRED DFII10) falls by more than 10bp while the 10-year nominal yield (ust10y, DGS10) rises by more than 10bp.
• How it is computed. From the month-end change over 3 months in each series. When both conditions hold, the whole rise in the nominal yield comes from a rise in breakeven inflation.
• What it measures. Whether higher yields reflect inflation fear rather than a higher real cost of money. Rising nominal yields with falling real yields is the classic signature of financial repression or fiscal dominance: savers are paid less in real terms even as headline yields climb.
• The levels. The ±10bp thresholds filter out noise, since daily moves of 5–10bp are routine.
• What would trigger it. The Fed cutting, or capping yields, while inflation expectations rise; QE or yield-curve control; an oil or tariff shock with a dovish Fed; a flight from cash into inflation hedges.
• History. Only 7 of 285 months since 2003 qualified: March and May 2009, January and November 2010 (QE1 and QE2), and December 2020 to May 2021 (pandemic reflation with Fed buying). Each episode was a Fed-assisted reflation, the mild version of C. Today is the opposite: over the last three months real yields rose 71bp, carrying most of the 10-year's 82bp rise. Assessment: the signal is rare and specific, so even one qualifying month would matter.

A3. 2s30s curve slope (above 120bp with the Fed on hold or cutting)

• What it is. The 30-year Treasury yield minus the 2-year, in basis points. Series slope_2s30s, derived from FRED DGS30 and DGS2.
• How it is computed. 100 × (DGS30 − DGS2), daily, and taken at month-end for the watch rule. The Fed condition uses the 3-month change in the effective fed funds rate (DFF), which must be 10bp or less.
• What it measures. The extra yield demanded to lock money up for 30 years rather than 2. The 2-year tracks expected Fed policy; the 30-year adds term premium (compensation for inflation, supply and fiscal risk). A curve steepening because the 30-year rises while the Fed is not hiking (a "bear steepener") is the market demanding more for duration, which is the C signature. A curve steepening because the 2-year falls in a cutting cycle (a "bull steepener") is normal and points to A.
• The levels. 120bp sits a little above the long-run median of about 105bp. It is meant as a "clearly positive term premium" marker.
• What would push it up. Larger long-dated issuance; weaker foreign or pension demand for duration; fiscal scares; the Fed cutting to protect growth while inflation stays high; quantitative tightening.
• History. Median 105bp, middle half 38–212bp, range −248bp (1980) to +400bp (2011). Above 120bp in 45% of months, mostly in easing cycles (1991–94, 2001–05, 2008–17, 2020–21). Restricted to the C version (above 120bp, the Fed not hiking, and the 30-year up more than 10bp over three months), it happened in 9% of months, most recently December 2025 and January 2026. It is 70bp now because the Fed's hike lifted the 2-year. Assessment: above 120bp on its own is too common to be a C signal. Read the trigger only with the 30-year-rising condition. That version has already flashed briefly this cycle, which is one reason C is at 30% rather than 20%.

A4. 30-year auction tail and indirect bidders

• What it is. Two results from each 30-year bond auction, from the TreasuryDirect auction file. The tail (auc30_tail_proxy) is how far the auction cleared above the market yield. The indirect share (auc30_indirect_share) is the share of competitive bids won by indirect bidders, mainly foreign central banks and asset managers bidding through dealers.
• How it is computed. True tail = auction high yield minus the when-issued yield at the 1pm deadline. The when-issued yield is paid data, so the proxy uses the auction high yield minus that day's closing 30-year yield (DGS30), which is noisier. A positive tail means a weak auction and a negative one (a "stop-through") means a strong one. Indirect share = indirect bidder accepted / total competitive accepted.
• What it measures. Real-money demand for long Treasuries at the margin. It is the most direct test of the "buyers' strike" part of the C story.
• The levels. A true tail above 2–3bp is considered poor in market commentary. Because the proxy is noisy, it takes two in a row, together with indirect bidding below 60%, to treat it as signal. Indirect bidding at 70% or more for three auctions marks unusually strong demand (the halve-C condition).
• What would push the tail up and indirect share down. Larger auction sizes; reserve managers diversifying away from Treasuries; higher Japanese yields drawing money home; a weaker dollar; fiscal or political shocks around the auction; dealers having to absorb more.
• History. Tail proxy: median +0.5bp, 5th–95th percentile −4.9 to +5.0bp. Above 3bp in 14% of auctions, but twice running only 5 times since 1998 (2008, 2009, 2012, 2019, 2024). June 2026 had a single 7.0bp tail with indirect bidding at 60%, so half of the trigger has already appeared. Indirect share: it rose structurally from about 35% in 2010–13 to 63–69% since 2016, so below 60% now means a real fall back. It was below 60% in 22% of auctions since 2015. At 70% or more three times running it has happened once (September 2022). September 2026's 79% is near the record of 80% (October 2024). Assessment: the C trigger is rare and meaningful. The halve-C condition is demanding, having been met once in 18 years, so it would take an exceptional run of demand to fire.

A5. ACM 10-year term premium

• What it is. The extra yield investors demand to hold a 10-year bond rather than rolling short-term bills for ten years. Series acm_tp10, the New York Fed's Adrian-Crump-Moench estimate, daily since 1961. The data room cross-checks it against the Kim-Wright estimate (kw_tp10).
• How it is computed. A statistical model of the whole yield curve (five principal components of yields) estimates the expected path of short rates. Term premium = actual 10-year yield − the model's expected average short rate over 10 years. It is an estimate, and different models can disagree by 0.5pp or more.
• What it measures. Compensation for duration risk: uncertainty about inflation, about supply, and about who will buy the bonds. It is the purest gauge of the "fiscal and supply" channel, separate from Fed expectations.
• The levels. The framework bands are A below 0, B 0.25–1.0 and C at 1.25 or higher. 0.40 marks a premium back near the post-2010 norm.
• What would push it up. Higher deficits and issuance; more inflation uncertainty; quantitative tightening; less price-insensitive demand from foreign official buyers or the Fed; higher global yields (JGBs); higher bond volatility.
• History. Median since 1961 is 1.37 and the middle half is 0.36–2.28, so 1.25 is a below-average level by long history. It was above 1.25 in 53% of months, mostly before 2014 (1980s average 3.4). It was last above 1.25 in February 2014; from 2010 the premium was compressed by QE and low inflation volatility (2010s average 0.44, 2020s −0.20). Today's 0.85 is only the 39th percentile of all history but the 82nd since 2010. Assessment: the C threshold is a return to the pre-2014 normal, not an extreme. That is exactly the framework's point: Regime C can come about through normalisation alone. The halve-C level of 0.40 was seen as recently as March 2025.

A6. Unemployment rate (4.6% or higher, a 0.5pp rise)

• What it is. The share of the labour force without a job and looking for one (U-3). Series unemployment, BLS via FRED UNRATE, monthly since 1948, seasonally adjusted.
• How it is computed. From the monthly household survey of about 60,000 households: unemployed / (employed + unemployed).
• What it measures. Labour-market slack, and the main input to whether the Fed eases. A rise of 0.5pp from the low is the Sahm-rule recession signal.
• The levels. 4.6% is 0.5pp above the current 4.1% (also the 12-month low), so it tests the Sahm-type rise rather than a level.
• What would push it up. Tighter financial conditions from high real yields; a credit event; weak hiring from tariffs or an AI-related investment pullback; fiscal tightening; an oil shock that hits demand.
• History. Median 5.5%; at 4.6% or higher in 70% of months since 1948, so the level is unremarkable and the signal is the rise. A Sahm-rule trigger (3-month average up 0.5pp from its 12-month low) occurred in 15 episodes since 1949. Nearly all coincided with recessions; the near-misses were 2003 and August 2024. Assessment: a 0.5pp rise is one of the most reliable recession signals available. If it fires, A becomes a live scenario quickly, which is why it halves B and raises A to about 35%.

A7. High-yield credit spread (above 4.5%)

• What it is. The extra yield on US junk bonds over Treasuries. Series hy_oas, the ICE BofA US High Yield Index option-adjusted spread, FRED BAMLH0A0HYM2.
• How it is computed. The index's yield over Treasuries of matching maturity, adjusted for embedded call options.
• What it measures. Default and liquidity risk priced by credit markets. It widens before and during recessions and financial stress, and is a timely read on whether high real yields are starting to break something.
• The levels. 4.5% is about 1.4pp above today and roughly the long-run median.
• What would push it up. Defaults among refinancing-heavy borrowers at higher rates; recession; equity sell-offs; bank or private-credit stress; an oil-price collapse (energy issuers).
• History. Data room limitation: FRED serves only a rolling three-year window for ICE indices, so the stored history starts October 2023 and accumulates from now on. In that window the range is 2.66–4.61% (the high was the April 2025 tariff shock), with a median of 3.02%. From published ICE history (not in the data room): the long-run median since 1997 is about 4.5–5%; lows near 2.4% (mid-2007); stress peaks near 8.9% (February 2016), 10.9% (March 2020) and about 21% (December 2008). Assessment: today's 3.1% is close to historical tights, so investors are not pricing stress. Moving above 4.5% would be a move to average, but combined with a 0.5pp unemployment rise it has reliably meant recession.

A8. Core PCE inflation (below 2.5%)

• What it is. The year-on-year change in the PCE price index excluding food and energy, the Fed's preferred inflation gauge. Series core_pce, BEA via FRED PCEPILFE, monthly since 1959.
• How it is computed. 100 × (index this month / index 12 months earlier − 1). The data room stores the index and computes the change.
• What it measures. Underlying inflation trend, and how close the Fed is to its 2% target. It runs about 0.3–0.5pp below core CPI because of different weights (less housing) and coverage.
• The levels. 2.5% is halfway between today's 3.0% and the 2% target. Below it, the Fed can cut without losing credibility, which opens the A path.
• What would push it down (toward A). Housing disinflation; a weaker labour market; productivity gains (AI); a stronger dollar; tariffs dropping out of the year-on-year comparison. Pushing it up instead: services and wage stickiness (supercore is 3.5%), and oil feeding through.
• History. Median 2.45%; below 2.5% in 50% of months since 1959 but 81% since 1995. It was last below 2.5% in March 2021 and has stayed between 2.7% and 3.2% for the past year. Today's 3.0% is the 89th percentile since 1995. Assessment: below 2.5% was normal for a quarter of a century and is now 5½ years in the past. Getting back there within a year would need a clear demand slowdown.

A9. Policy vs 30-year vote (the Fed cutting while the 30-year falls)

• What it is. The vote model's policy rule, which compares the 3-month change in the effective fed funds rate (fed_funds_eff, DFF) with the 3-month change in the 30-year yield (ust30y, DGS30).
• How it is computed. It votes A if fed funds fall more than 10bp and the 30-year also falls more than 10bp (cuts trusted by the long end). It votes C if fed funds fall more than 10bp but the 30-year rises more than 10bp (cuts not trusted), and B if fed funds move 10bp or less. Otherwise it casts no vote, as now: the Fed hiked 25bp and the 30-year rose 68bp.
• What it measures. Whether the long end believes Fed easing is non-inflationary. It is the market's verdict on Fed credibility in real time.
• What would trigger an A vote. A growth scare or financial stress prompting cuts, with inflation falling; a flight to quality into long bonds.
• History. Since 2003 the rule voted A in 24 of 285 months: 2007–08 (before and during the financial crisis), 2019 (insurance cuts), March–June 2020 (pandemic), September 2024, and October–November 2025. It voted C in 14 months, notably October 2024 to January 2025 (the Fed cut 100bp while the 30-year rose) and again in December 2025 and January 2026. Assessment: A votes come in short bursts tied to easing cycles. With the Fed now hiking, an A vote within a year would require a sharp turn, which is consistent with A at 15%.