China's PPI Printed 3.8% on Energy — and the ECB Hikes Into the Same Shock Tomorrow

The split inside the Chinese print is the entire story

Three numbers from the same release, stacked:

• 5.8% — industrial purchasing prices, year on year. What Chinese factories pay for inputs.
• 3.8% — producer prices, year on year. What they charge at the factory gate.
• 1.0% — core consumer prices. What the Chinese household actually faces once you strip food and energy out.

That is a cost shock travelling down a pipe and losing pressure at every joint. Inputs up 5.8%, output prices up 3.8% — a two-point gap that comes out of manufacturer margins — and a consumer who is barely feeling it, because the domestic demand that would let firms pass costs through is not there. Pork down 11.8% is not a story about oil; it is a story about a consumer nobody can raise prices on.

Read the components rather than the headline and the direction is unambiguous. Transport fuel up 8.3%. Non-ferrous metal wires and cables up 19.8%. Food down. This is not an economy heating up. It is an economy importing a price.

Europe’s August print has the same shape, and it is not a coincidence

Eurostat’s flash estimate for August, published 1 September, put euro area annual inflation at 3.3%, up from 2.9% in July. Here is where it came from:

• Energy: 14.3%, up from 10.3% in July.
• Services: 3.0%, down from 3.3%.
• Non-energy industrial goods: 1.2%, from 0.9%.
• Food, alcohol and tobacco: 1.2%, unchanged.
• Core, excluding energy: 2.2%, unchanged from July.

Headline inflation rose four tenths. Energy inflation rose four full percentage points. Services inflation — the component central bankers watch when they want to know whether an inflation is domestic and self-sustaining — went the other way. Core did not move at all.

So two of the world’s three largest economic blocs published August data whose entire acceleration sits in the energy line while the domestically-generated components are flat or falling. Japan, as we wrote last night, gets the same shock in a third form: it sourced 95.9% of its crude from the Middle East in FY2024, so an oil spike arrives there as a terms-of-trade tax on the currency rather than as a price index. One input, three transmission mechanisms, three central banks meeting inside ten days.

The ECB has published research saying this is a supply shock. It hikes anyway at 12:15 UTC tomorrow

This is the part worth sitting with. ECB economists Kristina Barauskaitė Griškevičienė and Claus Brand have written that adverse energy supply factors accounted for around 90% of the increase in energy inflation between January and May 2026, and that unlike 2021–22 the demand and policy-stimulus contributions are minor. That is the central bank’s own research arm describing the current episode as something monetary policy did not cause.

Tomorrow the Governing Council is nonetheless expected to take the deposit rate from 2.25% to 2.50% — a second consecutive move, after 11 June took it from 2.00% to 2.25%. FXStreet reports the hike as effectively a done deal. Market rates imply a peak near 3.00% during 2027, a level Danske’s research team publicly regards as excessive; a separate survey has 91% of economists expecting 2.50% to be the end of it for this year.

We are not going to tell you the ECB is wrong. A central bank that lets a supply shock run into wage-setting ends up with a demand problem too, and a decade of credibility is not a thing you spend to save a quarter of a point. But it does tell you something concrete about tomorrow: the hike is not the event. A 25bp move that is a done deal cannot move a currency. What can move it is whether Lagarde frames August as an energy episode that will unwind on its own, or as something requiring more. That framing arrives at 12:45 UTC, thirty minutes after the decision.

The consensus was wrong this morning in a direction worth remembering

At last night’s slot we flagged that two vendor calendars disagreed on China’s PPI consensus — Investing.com had 3.6%, InvestingLive’s preview had around 3.2% — and said the print would tell us which to trust. It printed 3.8%. Both were low; Investing.com was closer; InvestingLive’s own release page subsequently carried 3.7% as the expected figure, which is closer still.

The useful lesson is not which vendor won. It is why a consensus would be low on a print like this one. A forecast survey closes days before the release. Brent spent the survey window travelling from the low nineties to somewhere between $98 and $100.03 — the readings vary by source and by snapshot, and we are quoting the range rather than pretending to a single figure. When the dominant driver of a price index moves sharply after economists have submitted their numbers, the consensus is not wrong in a random direction. It is stale in a knowable one.

That is worth encoding. If your system trades surprise — the gap between consensus and print — and the release is energy-sensitive, then a large move in oil since the survey closed is a systematic bias in your input, not noise. It will not average out, because it is the same sign every time oil trends into a print.

How to schedule tomorrow, in UTC

Tomorrow stacks three tier-one events into thirty minutes, and we have said before that this is the week’s real risk concentration:

• 06:00 — Germany final CPI for August.
• 12:15 — ECB decision. 2.25% to 2.50% expected.
• 12:30 — US August PPI and weekly jobless claims. Fifteen minutes after the ECB, on a different continent’s currency.
• 12:45 — Lagarde’s press conference and the new staff projections. This is the event.
• 17:00 — US 30-year bond reopening.

Then Friday 12:30 UTC brings US August CPI, expected at +0.4% month on month with core at +0.2%, and the weekly CFTC positioning release.

Today, by contrast, has no European release at all. If your risk budget for the week is a fixed quantity — and it should be — the arithmetic is not subtle. A thirty-minute window containing a rate decision, a foreign inflation print and a central bank press conference is not three events you can size independently. It is one event with three timestamps, and the correlation between them is not something your backtest measured, because windows shaped like that are rare enough that your sample contains a handful.

What this does not tell you

The ECB research quote covers January to May 2026. August is not in that window. We are using it as evidence of how the institution frames this episode, not as a measurement of last month — and it is entirely possible the Governing Council’s view has hardened since.

Core inflation definitions differ across sources and the numbers are not interchangeable. Eurostat’s flash gives core excluding energy at 2.2%, unchanged. At least one commercial vendor reported a different core measure easing from 2.5% to 2.4%. We have used the Eurostat figures throughout because they are the primary source; if you are comparing against another provider’s series, check the definition before you conclude anything moved.

Brent’s Tuesday high is genuinely uncertain. Published figures we could read range from $98 to $100.03, and Trading Economics had it near $98.90 during the European morning today. We have quoted the range and named it as a range. Nobody should be building a level off a number with two dollars of disagreement in it.

We do not know what the ECB staff projections contain, and the guidance is where the whole decision lives. We also cannot tell you whether the energy shock unwinds. If the Middle East supply disruption eases, both of these prints revise away and the central banks reacting to them look early. If it does not, core follows headline eventually and they look late. That is a genuine two-sided outcome and we have no edge on which way it goes — which is precisely why the useful response is a sizing rule for tomorrow’s 12:15–12:45 window rather than a view on the euro.

Related

• FX events calendar — the ECB decision, US PPI and the Lagarde press conference, with times in UTC.
• Signals — how we treat scheduled-release windows and consensus-versus-print surprise.
• Expert advisors — news filters and event-window position sizing.