← All briefs
InflationRatesGoldAllocation

CPI 3.4 %, the bond market saw it first

Two-year yields started climbing eleven days before the print. What the curve knew, why gold did not flinch, and how the model moved three points from long bonds to gold before the number arrived.

Dear reader,

The Consumer Price Index for August printed at 3.4 % year on year on Wednesday, against a consensus of 3.1 %. Headlines called it a surprise. The two-year Treasury did not. It had been rising for eleven trading days before the release, from 3.62 % to 3.89 %, and it rose another nine basis points on the day. If you only read the inflation number, you were surprised. If you watched the front end of the curve, you were told.

This brief is about that gap: what the bond market was pricing, why the model's inflation-momentum component turned before the print, and why the response — three points from long bonds to gold in every profile — was already half-done by the time the number arrived.

What the curve was saying

The two-year yield is the market's estimate of the average policy rate over the next two years. When it climbs while the ten-year barely moves, the curve is "bear-flattening": the market expects the central bank to stay tighter for longer, but does not expect that tightness to raise long-run growth or inflation. That is precisely what happened from 20 August onward. The 2s10s spread compressed by 22 basis points in two weeks, almost all of it from the front end.

Fig. I US 2-year and 10-year Treasury yields, 60 trading days to 3 September 2026. Dashed line: CPI release. Source: FRED DGS2, DGS10 (daily). Illustrative rendering.

Three things fed that move. Used-car auction prices, which lead the CPI's vehicle component by roughly six weeks, had turned up in July. Regional Fed surveys showed prices-paid indices rising for the third month. And the Cleveland Fed's nowcast, which is public and updated daily, had drifted to 3.3 % by the end of August. None of this was hidden. It was simply spread across sources that most people do not read every day, and the bond market reads them all.

Why the model turned early

Our inflation-momentum component does not use the headline CPI number at all. It uses the three-month annualised change in core CPI relative to its twelve-month trend, plus the two-year breakeven inflation rate from TIPS. The breakeven is the part that matters here: it moved from 2.31 % to 2.49 % across the same eleven days. The component crossed from "cooling" to "re-accelerating" on Friday 29 August — five days before the print — and the model made its first allocation change the following Monday.

The model does not predict inflation prints. It notices when the market has already changed its mind and adjusts the portfolio before the newspaper does.

This is the general shape of how the regime model works, and it is worth restating because it is easy to misunderstand. We are not forecasting. We are describing conditions with a small set of market-based and published inputs that update more often than the official statistics they anticipate. When those inputs move past a threshold, the allocation moves inside its band. The CPI print then confirms or contradicts the move. This week, it confirmed.

Why gold did not flinch

The textbook says higher real yields hurt gold. Real yields did rise this week — the ten-year TIPS yield went from 1.71 % to 1.84 % — and gold fell 0.6 % on the day of the print, then recovered most of it by Friday. It is 2 % below the record it set two weeks ago (the subject of brief #86). That is remarkable resilience, and it has two explanations.

The first is that official-sector buying is not price-sensitive in the way private investors are. Central banks that are diversifying reserves do not stop because real yields rose thirteen basis points. The second is that the market is pricing a central bank that may be forced to tolerate higher inflation rather than one that will fight it aggressively — the bear-flattener, again. In that world gold is a hedge against the credibility of the response, not against the level of rates.

Fig. II Gold (USD/oz) against the 10-year TIPS real yield, inverted, 36 months. The relationship has loosened since 2024. Source: gold reference price PM fix, FRED DFII10. Normalised, illustrative.

What long bonds have to prove

The asset that took the change is long-duration government bonds. The model's bond sleeve is not one thing: it is roughly two-thirds intermediate (5–7 year) and one-third long (20+ year) duration, and the reduction this week came entirely out of the long third. Long bonds are the most exposed asset in the portfolio to a re-acceleration of inflation, because their price is almost entirely the discounted value of fixed coupons twenty years out. If inflation momentum stays "re-accelerating" for two more weeks the model will take another two points from the same place. If the component reverses, it goes back. Either way, the bond sleeve does not fall below its band floor — 20 % in Balanced — because duration remains the only asset that reliably pays in a growth scare, and growth scares are what risk-off regimes are made of.

The liquidity index is the story to watch

Inflation took the headlines this week, but the component that decides the regime is liquidity, and it is close to a line. The three-month impulse of our global broad-money measure is −1.2 %. The Fed's balance sheet shrank another $18 billion; the ECB's reinvestments continue to roll off; the People's Bank of China added liquidity, but not enough to offset. Our rule for the risk-off label is an 8-point move in the meter that holds for two weeks. The meter is at 38, from 44 four weeks ago. One more fortnight below −1.0 % on the liquidity impulse and the label flips from "risk-off (leaning)" to plain "risk-off", which moves the equity sleeve to the bottom of its band in all three profiles.

We are not predicting that. We are telling you where the threshold is so that when you read next week's brief you know what you are looking at.

What this means for the three profiles

In practical terms, very little happened to a Balanced portfolio this week and that is the point. Gold went from 17 to 20 %, long bonds from 11 to 8 %, everything else stayed put. The weekly change cap of five points per asset was not reached. A reader who rebalances quarterly would not act on this at all; a reader working to a 3-point drift threshold would have placed one small set of orders on Tuesday. That is what a regime model is supposed to feel like: slightly boring, most of the time.

Planned for Pro: a historical regime archive showing every inflation-momentum flip since 1986 with the model's response. The archive will be illustrative model output, not a track record.

How our model allocation changed this week

AssetLast weekThis weekΔReason
Equities34 %34 %0Regime unchanged: risk-off (leaning), meter 38 — bottom of the 34–55 % band
Bonds (total)31 %28 %−3Long-duration third reduced on inflation-momentum flip
Gold17 %20 %+3Top of band; hedge against tolerance of higher inflation
Bitcoin6 %6 %0Bottom of band while liquidity contracts
Cash12 %12 %0At floor + regime buffer

Balanced profile shown; Conservative and Growth made the same 3-point move within their own bands. Weights are model output for illustrative profiles, rounded to whole points.

Risk disclaimer. This brief is educational and describes the output of a rules-based model for illustrative risk profiles. It is not investment advice, it does not consider your circumstances, and no return is promised or implied. Model allocations can and do lose money. Past regimes do not predict future ones. Speak to a licensed adviser before acting.
ME
With patience, until Thursday —

Rulebook Wealth Editorial. The brief is written from the week’s published data. Every chart cites a public series and every model figure carries the version that produced it. Corrections are appended with a date, never silently edited. Data sources · Editorial policy

Related

Want to check this brief’s charts yourself? Every series identifier we cite is listed in the data sources, and Pro accounts can export the allocation, bands and caps as JSON or CSV from the portfolio builder.