Long-rate regimes from 2026 — a framework, a dashboard and scenarios

1Summary

On 1 October 2026 the US 30-year Treasury yield reached 5.67%, its highest since July 2002, and the 2-year was 4.91%. That is roughly where long rates stood when you arrived in Australia in May 2002 and concluded "this is as low as rates can go". Over the following 18 years they fell almost to zero.

The lesson from 2002 is not that you should have forecast zero. It is that the level of rates told you almost nothing. What mattered was the regime: the slow-moving forces that set the price of long-term money. The same applies now, in reverse. Saying "5.7% is high, so the next big move is down" is the 2002 error turned upside down.

The framework in this note has four layers:

1. Decompose the yield into expected inflation, the expected real short rate and the term premium. The same 5.5% can mean very different things.
2. Score the structural drivers: savings versus investment, fiscal policy, demographics, productivity, globalisation and central-bank credibility.
3. Classify the regime: A (disinflationary, a return toward 2002–2020), B (a higher real-rate equilibrium) or C (fiscal or inflationary, 1960s/70s-like). Overlay short-lived shocks on top.
4. Anticipate the policy reaction: QE, shorter debt issuance and financial repression put a soft cap on yields, but they do so by transferring wealth from savers rather than by making bonds safe.

Working view (judgement, for 3–5 years, to be revised): Regime B 50%, Regime A 25%, Regime C 25%. Within any of these, expect at least one shock episode of 100bp or more in either direction.

Five metrics to watch first:

• the 10-year term premium (ACM)
• the 10-year TIPS real yield
• the 5y5y forward breakeven
• federal interest outlays as a share of revenue
• whether the long end rises while the Fed is cutting

2What the chart actually shows

• Same level, opposite trend. In 2002 the 5–6% reading sat inside a 40-year downtrend of lower highs and lower lows that ran from the 1981 peak of about 15%. In 2026 the same level sits on a rising path from the 2020 low of under 1% on the 10-year. The level matches; the direction of travel does not.
• The post-2020 base has broken upward. The 2023 high near 5% held as resistance for about three years. A close above it on the 30-year is a technical break, but a break only matters if the fundamentals confirm it (see the dashboard).
• The curve is steepening from the long end. With the 2-year at 4.91% and the 30-year at 5.67%, the 2s30s spread is about +76bp. When the curve steepens because long yields rise while short yields hold or fall, the market is pricing term premium (worry about supply, fiscal policy or inflation uncertainty) rather than Fed tightening. That pattern is the main thing separating Regime B/C from a simple cyclical move.
• It is global. CNBC notes borrowing costs rising "around the world… a months-long trend". A rise that is synchronised across countries points to global drivers: the term premium, Japan, and fiscal policy everywhere. A rise confined to the US would point to US-specific fiscal or political risk.

3What the three notes agree on, where they differ, and what they miss

Where they agree

• Anchoring was the 2002 error. "5% is low" was judged against 1970–2000, not against the forces setting the neutral rate (r*).
• The decomposition. Long nominal yield ≈ expected real short rates + expected inflation + term premium.
• The 2002–2020 decline had identifiable drivers. China joined the WTO (Dec 2001), there was an Asian and petrodollar savings glut, ageing populations were in their peak saving years, and later QE and zero rates followed crises (2008, 2020).
• Some of it was visible in 2002. Japan was already showing that an ageing, indebted advanced economy could sit near zero rates for years. Disinflation was under way, and the Fed was worried about deflation by 2003.
• Some of it was not. The size of the 2008 crisis, a decade of QE and the pandemic.
• Many of those drivers have reversed. Deglobalisation, large deficits at full employment, heavy capex (AI, grids, defence, energy), retirees drawing down savings, a rebuilt term premium, and rising Japanese long yields.
• Base case: higher for longer, with the 10-year around 4–5.5% and the 30-year around 5–6.5%.

Where they differ

The reconciliation: a ceiling exists, but it is political rather than economic. Governments and central banks can cap nominal yields, as the UK did in 2022 and Japan did with yield-curve control. They do this by accepting higher inflation or a weaker currency. A nominal cap does not protect your real wealth. That distinction matters a great deal for how you invest.

What all three leave out

• Thresholds. None says what reading of a metric would make you change your mind.
• Probabilities and review dates. Without them you cannot tell later whether the view was any good.
• Base rates. How long regimes typically last and how often they turn.
• The investor's own position. An Australian, AUD-based investor with superannuation, a possible mortgage and home bias faces different questions from a US bond desk.
• Bond arithmetic. How much yield cushion a long bond has against further rises.
• Real-versus-nominal outcomes under a policy cap, as discussed above.

4The analytical framework

Layer 1 — Decompose the yield

The same headline yield can mean very different things:

Practical split: the breakeven (nominal minus TIPS) measures inflation compensation; the TIPS yield measures the real rate; the term premium comes from models (ACM from the NY Fed, Kim-Wright from the Fed Board) and is the "fear and supply" component. Track all three monthly. The question to ask each time is which component moved?

Layer 2 — Score the structural drivers

For each driver, record a score of −1 (pushing yields down), 0 (neutral) or +1 (pushing them up), with a one-line reason and a date.

The total score is not a forecast; it is a map of pressure. What matters is how it changes between reviews.

Layer 3 — Classify the regime

Shock overlay (weeks to months, can happen in any regime):

• a fiscal stress episode such as UK gilts in 2022: a disorderly spike followed by an official backstop
• a flight to safety (growth or credit shock): a 100–200bp rally that does not by itself change the regime

Do not mistake an overlay for a regime change. Wait for confirmation from Layer 2.

Layer 4 — Anticipate the policy reaction

When yields climb high enough to hurt, governments respond. Watch for:

• Treasury shifting issuance toward bills (shorter weighted average maturity)
• QT ending or QE restarting
• rules that force banks or pension funds to hold government debt
• pressure on the Fed to cut in spite of inflation
• yield-curve control

Each of these lowers the nominal yield but raises the inflation risk. In Regime C, a capped yield and a falling real value of bonds are the same outcome. Assets that hold real value and currency diversification matter more than the headline yield.

5Signposts dashboard

Review monthly; record the reading and the regime it points to. Thresholds are indicative starting points for you to calibrate, not established rules. Readings taken from the chart are dated 1 Oct 2026; others are marked "check" and should be filled in from source on the review date.

How to read it: count how many metrics point to each regime this month and last month. A regime call needs at least 3 of the 5 headline metrics (rows 2, 3, 4, 6, 9) to agree for at least 3 consecutive months. Anything less is noise or an overlay.

6Scenario builder

Probabilities are personal judgement as of 1 Oct 2026 for a 3–5 year horizon. Revise them at each quarterly review and keep the history.

Scenario B — "Expensive money, no crisis" (50%)

• Story: Deficits stay around 6% of GDP, the AI/grid/defence capex cycle continues, inflation settles around 2.5% and the Fed stays credible. Capital is simply priced higher than in the 2010s.
• Path: The 10-year trades 4.25–5.75% and the 30-year 4.75–6.25%, with a positive term premium. Rallies on growth scares fade.
• Confirming signs (12–36 months): TIPS above 1.75%, 5y5y at 2.0–2.6%, term premium stable or positive, productivity improving, auctions orderly.
• Assets: Long bonds earn a fair real return but are not a trade for a large rally. Ladders and holding to maturity work. Cash is decent. Equity valuations are capped by a higher discount rate, so earnings growth matters more than multiple expansion.

Scenario A — "2002 again: disinflation returns" (25%)

• Story: Recession or a credit event (commercial property, private credit, an AI capex bust), or AI-driven disinflation arrives faster than expected. The Fed cuts hard, foreign safe-asset demand returns and the term premium compresses.
• Path: The 10-year falls to 3–4% (lower in a deep slump) and the curve bull-steepens.
• Confirming signs: 5y5y below 2%, unemployment rising quickly, HY OAS above 500bp, capex guidance cut, core PCE below 2.25%.
• Assets: Long duration is the big winner (see the arithmetic below). Equities fall first, then recover. Cash yields fall quickly, which is the reinvestment risk in waiting.

Scenario C — "Fiscal dominance / 1970s-lite" (25%)

• Story: Deficits persist and interest costs compound. Political pressure overrides the Fed, energy or tariff shocks hit, and foreign buyers step back. Either long yields break above 6%, or policy caps them and inflation does the adjustment.
• Path: Either a 30-year at 6.5%+ with high volatility, or a capped nominal yield with a falling real yield. Both are bad for nominal bondholders.
• Confirming signs: 5y5y above 2.75%, term premium above 1.25%, auction tails, a shift to bill issuance, interest above 20% of revenue, the long end rising while the Fed cuts.
• Assets: TIPS and inflation-linked ACGBs, short duration, real assets (resources, infrastructure with pricing power), currency diversification. Avoid long nominal bonds and long-duration growth valuations.

Shock overlays (could happen in any scenario)

• Gilt-style spike: 50–100bp up in weeks, then an official backstop. This is an opportunity to add duration in B, but not in C.
• Flight to safety: 100–200bp down in a growth or market crash. This is an opportunity to trim duration in B and C, but not in A.

7From view to decisions

Separate the view from the position

In 2002 the costly part was not the view; it was how much rode on one forecast. Decide the view (regime probabilities) and the position (size, duration, timing) as two separate steps.

Bond arithmetic: how much cushion does the yield give you?

A rough 12-month breakeven is yield ÷ modified duration: the rise in yields that would wipe out a year's income.

These are approximate, ignore roll-down, and include convexity only roughly. The point stands: at these yields the payoff of long bonds is asymmetric in your favour if you can hold, and the cushion against further rises is thin. Above 5.5% you are being paid to wait in A and B. In C you are being slowly eaten by inflation, even if the nominal price is capped.

Decision rules to consider

• Survivability first. Size any duration bet so that a 30-year at 7% (C) or a 10-year at 3% (A, if you are short duration) is uncomfortable but not damaging.
• Scale in by trigger, not by forecast. For example, add a third of the intended duration at the current level, a third if the 30-year passes 6% and the 5y5y is still below 2.6% (an overlay, not C), and the last third only when the dashboard confirms B or A.
• Barbell or ladder rather than all-in. A ladder of 1–10 year bonds plus a TIPS sleeve plus some long duration hedges A and C at the same time.
• Use inflation-linked bonds for the C tail. TIPS or inflation-linked ACGBs at real yields near 2% are a cheap hedge against the scenario that hurts most.
• Run a pre-mortem. Before acting, write the paragraph "It is 2031 and this position lost money because…" for each regime.

An Australian investor's view

• ACGB vs UST. Australian 10-year yields broadly follow US yields, but RBA policy, iron ore and the China cycle drive the spread. Track ACGB 10y − UST 10y.
• Currency. Unhedged US bonds add AUD/USD exposure. The AUD tends to fall in global risk-off episodes (Scenario A), which helps unhedged holdings. In Scenario C a weaker USD hurts them. Decide the hedge ratio deliberately.
• Superannuation. Check the actual duration and inflation-linked share in your super fund's "balanced" or "conservative" option. That is often your largest bond exposure, and you may not know it.
• Mortgage and cash. Most Australian mortgages are variable or short-fixed, so they track the RBA, not US long rates. Offset accounts and cash earn at the short end. In Scenario A those yields fall quickly; locking some term protects against that.

8How to think about this more concretely

These are practical steps that turn "what do I think rates will do?" into a process you can check later.

1. Keep a one-page forecast journal. Date it. Write regime probabilities (A/B/C), a 12-month range for the 10-year and 30-year, and the three reasons. Re-score quarterly without looking at the old page first, then compare. After two years you will have evidence on your own calibration.
2. Write "what would change my mind" before you need it. For each regime, list the two or three dashboard readings that would make you cut its probability by half. This is the antidote to anchoring, which was the 2002 error.
3. Use base rates. US long-rate regimes have been long: rising from about 1946 to 1981 (about 35 years) and falling from 1981 to 2020 (about 39 years). Turning points were recognised only years afterwards, and within each regime counter-trend moves of 100–200bp were common. Expect to be unsure in real time, and judge yourself on the process, not on calling the turn.
4. Back-test the framework on 2002. Rebuild the dashboard with data from May 2002: 10-year TIPS around 3%, Japan's 10-year around 1.3%, China just inside the WTO, core inflation falling, foreign official buying rising. Would the framework have flagged Regime A? This is the most direct way to learn from your own experience.
5. Steelman the other side. Each quarter, write the strongest case for the regime you rate lowest. If you cannot write a convincing one, your probabilities are probably too confident.
6. Read primary sources, not commentary. Useful ones:
   • Treasury quarterly refunding statements (issuance mix)
   • the NY Fed ACM term-premium page
   • CBO's budget and long-term outlook
   • the BIS annual economic report
   • the Fed's Summary of Economic Projections (the long-run dot as the market's r* proxy)
   • Bank of Japan JGB purchase plans
7. Review on a fixed schedule. First weekday of each month for the dashboard (30 minutes); quarterly for the probabilities and the journal. Do not react to single days like today's.
8. Automate the dashboard (optional). Most rows are on FRED. A small script or spreadsheet pulling the codes in section 5 can produce the monthly table and a chart of the A/B/C vote count. That makes the discipline cheap enough to keep up.
9. Get the facts right. Several figures in the source notes (below) should be checked against primary data before relying on them.

9Caveats and figures to verify

• To verify: "US debt-to-GDP above 120% (~$39T)", "interest costs above $1 trillion", and "deficits about 6% of GDP" come from AI-generated notes and are plausible but unchecked. Confirm with Treasury's Debt to the Penny, CBO and FRED before relying on them.
• Taken from the screenshot (1 Oct 2026): 30-year 5.6702% (+3bp, highest since July 2002) and 2-year 4.91% (+2bp). The 10-year level is read off the chart (about 5%) and should be confirmed.
• Indicative only: dashboard thresholds and scenario probabilities are judgement for you to calibrate, not established rules.
• Not financial advice. This is an analytical framework for personal decision-making. Consider professional advice for portfolio, tax and superannuation decisions.

Sources (in ../01-source-notes/): claude_treasury-yields-2002-vs-2026.md, gemini_interest_rates_macro_analysis.md, openaa_interest-rate-regimes-2002-vs-2026.md, 10year_30year_US_treasury_yield.png (CNBC, 1 Oct 2026).