Brent takes $108, the ECB hikes into it, and every long end in the world makes a new high

The United States — PPI, the auction, and Oracle

August PPI was the trigger. Headline +0.4% m/m was in line but the annual rate accelerated to +5.4% against 5.3% consensus; core was +0.2% m/m, below the 0.3% estimate, so the beat was entirely in the pipeline, not in the underlying. That nuance did not survive contact with a tape already watching Brent. Initial claims were 206k against 205k with continuing claims 1,774k — no labour-market deterioration, which is exactly the condition that lets a hike narrative run. Existing home sales fell 2% to a 3.98m annual rate, the lowest since June 2025.

Equities fell for a fourth straight session: S&P 500 7,591.70 (−44.66, −0.58%), Nasdaq Composite 26,081.72 (−0.65%), Dow 52,064.10 (−0.60%), Russell 2000 2,890.95 (−1.04%) — small caps down for a third day. Semis were the weak spot with SMH −2.5%, Intel −5.7%, Micron ≈−5%, Nvidia −2%; the commodity-equity complex was worse, with Freeport −7.3% and Southern Copper −7.1% on a day copper made a record, which tells you the equity risk premium is beating the underlying commodity. Gainers were defensive or idiosyncratic: Apple +3.6%, Reddit +4.7%, Charter +4.5%, Elevance +4.1%. Cooper Companies was the worst S&P name at −14%. The S&P closed below the 200-day EMA (7,663), the 200-day SMA (7,693) and Schwab's named 7,600–7,620 shelf, with RSI at 32.7 and the McClellan Oscillator below −72, a five-month low.

The $22bn 30-year auction at 13:00 ET was the session's standout and its paradox. It cleared at 5.308% against 5.335% when-issued — a 2.7bp stop-through versus a six-auction average tail of +0.13bp, with bid-to-cover 2.61× (avg 2.38×), indirects 79.5% (avg 66.4%) and dealers taking only 2.2% (avg 11.5%). Foreign and indirect demand absorbed essentially the entire issue. The long bond then closed 4–6bp above the 13:00 stop. Three auctions, three stop-throughs, and yields at 2007 highs: the market will take the paper, it just wants to be paid.

Oracle reported after the close and beat everything that was asked of it — revenue $19.35bn, +30% y/y against $19.14bn consensus, adjusted EPS $1.92 against $1.74, total cloud $11.6bn (+62%), OCI $7.4bn (+121%), RPO $664bn against $630.6bn expected and up $209bn year on year, FY27 guidance raised to ≥$90bn revenue and $8.10 adjusted EPS. The shares closed −5.2% and traded +7% after hours. The number the credit desk cares about is different: quarterly capex of $28.5bn against $8.5bn a year ago, total debt of $125–129.5bn, S&P at BBB− (one notch above high yield) and a 5-year CDS that has exceeded 198bp, its widest ever, through 2008 levels. Management pre-emptively said the new AI contracts would not force a near-term change to its capital-raising strategy — issuers only say that when the bond desk is the binding constraint. Roughly half the RPO is tied to a single counterparty.

Europe — the ECB delivers, and France pays for it

The ECB raised all three rates 25bp, taking the deposit facility to 2.50%, MRO to 2.65% and the marginal lending rate to 2.90%, effective 16 September. Lagarde described the decision as a unanimous "no-brainer," framed the shock as "predominantly a supply shock," said the Council "did not debate the future rate path," and declined either to validate or push back on market pricing — "markets do what they have to do." The statement's operative sentence was that "inflation is set to remain well above target for an extended period." QT, PEPP and the TPI were unchanged. The staff projections carried the message: 2026 HICP 3.0%, 2027 revised up to 2.5% from 2.3%, 2028 2.1%, with 2027 core at 2.6% — core above headline two years out. Reuters and Bloomberg then both carried sources stories saying officials expect more tightening with October "in play," contingent on data and on Iran. Post-decision pricing is genuinely unsettled: October hike odds were quoted anywhere from 29% to 61% on the day, December is described as close to fully priced, and roughly 85–88bp of further tightening to end-2027 implies a terminal near 3.35–3.40%.

The reaction was a violent bear flattening in core Europe and a rotation into the periphery. Bund 2y +18.2bp to 3.23%, 10y +6.4bp to 3.50% — the highest since 2011, 30y +0.4bp. OAT 10y +11bp to 4.44%, a November-2008 high, taking OAT–Bund to ~94bp, the widest since 2012, with the French 30y at levels last seen in 2003. BTP 10y fell 1.3bp to 4.27%, compressing BTP–Bund ~8bp to ~77bp and putting Italy through France by roughly 17bp. Equities split cleanly on bank weight: Stoxx 600 637.73 (−0.5%), DAX −0.84%, CAC 40 −0.49%, FTSE 100 −0.57%, but FTSE MIB +0.58% and IBEX +0.26% — the only two majors green — with SocGen +1.6%, Deutsche Bank +1.3% and Santander +0.7% intraday. Energy rose 0.7% (BP +1.2%, Shell +0.7%); mining was the worst sector at −3.9% with Antofagasta and Anglo American off more than 5%. Associated British Foods fell 8–9%, its worst day since January, on Primark and a demerger plan. Italian industrial production beat at +0.7% m/m; Swedish monthly GDP contracted 0.8%.

Asia and Australia — Japan the only gainer

Japan was the region's outlier, with the Nikkei 225 +0.20% to 65,270.95 and TOPIX +0.2% to 4,055 — the sole major Asian market up, and the sole major global index up on the week (+0.38%). JGBs bear-steepened in sympathy with everything else: 10y +3.2bp to 2.92%, 20y +5.3bp to 3.76%, 30y +5.2bp to 4.01%, with the 2y anchored at 1.83%. Board member Masu said the Bank "will keep raising" to cap the price trend at 2%, six days before the decision. Elsewhere the tape was heavy: Hang Seng ~24,911 (−1.44%), Shanghai 3,934.36 (−0.43%), Shenzhen Component −0.77%, CSI 300 −0.54%, with Tencent −1.5%, Meituan −2.0% and Xiaomi −1.8%. Chinese August vehicle sales fell 5.1% y/y from −0.3%. KOSPI 7,034 (−0.25%) faded Wednesday's record but only marginally, with Samsung Electronics flat and SK Hynix +0.16% — the semis complex held while the rest gave way. TAIEX 46,940.49 (−0.51%). India bucked it, the Sensex +0.19% to 74,902.59 and Nifty +0.20%, snapping a three-day losing run; the arithmetic reconciles this edition, which it did not last.

The ASX 200 closed 8,819.40, −92.0 points, −1.03% on 836.9m shares, a six-week low with an intraday trough of 8,742.60 that was −1.89%. Breadth was brutal — 761 decliners against 333 advancers across the market, 162 of the 200 down — and the ASX 200 VIX rose 11.84% to a three-month high of 13.14. Materials were the worst sector and the banks fell 1.4–1.8%; telcos were the only green sector. The session's tell was BHP −2.9% to $62.72 on the day LME copper printed a record. Best performers were Megaport +4.53%, Mercury NZ +4.50% and AP Eagers +3.75%; worst IperionX −8.04%, Nine Entertainment −6.5% to a record low, Westgold −5.83% and Austal −5.79%.

The week in review — four sessions that repriced the cycle

This was a four-session week (Monday 7 September was Labor Day) and it changed the regime rather than extending it. Brent +13.1% from $96.28 to $108.90 — path 96.28 → 97.00 → 97.92 → 101.21 → 108.90 — is the engine; everything else is transmission. UST 2y +19.4bp and 10y +18.6bp left 2s10s essentially unchanged at ~40bp: a parallel bear shift, which is a repricing of the policy path, not a term-premium event, whatever the 30-year's absolute level suggests. Equities fell everywhere except Tokyo: S&P −1.64%, Nasdaq −1.60%, Dow −2.53%, Russell 2000 −2.81%, Stoxx 600 −1.46%, ASX 200 −2.07%, Hang Seng −2.85%, Nikkei +0.38%. The dollar fell 0.30% while the US 2-year rose 19bp — a meaningful breakdown in the rates–FX correlation, and the clearest evidence yet that the market is reading the hike as a growth-negative supply-shock response rather than a carry positive. USD/JPY −1.73% into a live BoJ was the week's largest FX move. Gold fell 0.52% on a futures basis during a geopolitical oil shock — the real-rate channel beat the haven bid, which is the single most instructive cross-asset fact of the week. VIX rose only 2.02 points to 16.55 across four consecutive down days and a 13% oil move; skew stayed at the first percentile and put/call at the 16th. Credit did not move at all through Tuesday, and there is no post-shock observation.

ECB — the read-out. The decision was the easy part; the projections were the message. Revising 2027 headline HICP up to 2.5% from 2.3%, with 2027 core at 2.6% — above headline — is a statement that the Council no longer believes this is a pure energy pass-through. Lagarde refused a path ("we are not pre-committing," "can't anticipate what will be the next move," the Council "did not debate the future rate path") and declined to engage with market pricing at all. She also volunteered that she had been surprised by economic resilience — consumption, public investment, a services recovery, AI-driven global demand — and that the near-term growth outlook "has improved." A central bank that is surprised to the upside on growth and has just revised inflation up two years out is not done. Reuters and Bloomberg both carried sources stories within hours putting October in play. The market has not settled: October hike odds were quoted at 29% by one source and ~61% by another on the same day, and a December that is "fully priced" sits awkwardly with a 29% October, which argues the higher figure is the more internally consistent. We are publishing the range rather than picking.

Fed — six days out and in blackout. The distribution has moved a long way in four sessions: 71.0% for 16 September against 60.4% on Wednesday and roughly 58% a week ago, with December now modally two hikes at 46.3% and a cumulative 94.5% chance of at least one by then. The cross-checks disagree by a lot — a CME-derived read around 65%, Polymarket at 54% — and a 17-point futures-to-prediction-market basis six days before a meeting is unusual enough to note. Statement and SEP at 14:00 ET Wednesday 16 September (04:00 AEST Thursday), press conference 14:30 ET. Blackout runs to 17 September. The July meeting already produced three dissents for immediate tightening, so the hawkish minority does not need converting; it needs company. Tonight's CPI is the only remaining input.

BoJ — and a date that keeps being got wrong. The MPM runs 17–18 September and the decision is announced on Friday 18 September; third-party calendars that print "17 September" are quoting the meeting start. Board member Masu said on Thursday that the Bank "will continue to raise the benchmark interest rate" to cap the price trend at 2%, and Ueda has said hikes are on the table at every meeting "including this month's." Pricing is unsatisfactory — the tracker we can source is stamped 8 September at 63.4% and therefore predates this week's move, while one market read circulating in Sydney is as high as 97%. The yen is near its strongest in seven months. Tokyo is then shut for three days, 21–23 September, immediately after the decision and the week before the RBA — size any yen position for a gap, not a drift.

RBA — the Board goes in nearly blind. Two OIS trackers put 29 September at 70% and 72%; a 75% and a 77% are circulating and are not used. Deputy Governor Hauser has said the debate at the next meeting "would focus on whether to raise interest rates." The scheduling is the story: on the ABS forward calendar there is no mid-September labour force print, the August Labour Force release lands on 29 September — the decision day itself — and the monthly CPI indicator lands on 30 September, the day after. The Board will therefore set policy with essentially no new top-tier domestic data, which means oil, the AUD and global yields decide the final pricing rather than the domestic dataflow. Bank calls remain split 1–3: NAB for September; Westpac (flipped), CBA and ANZ for November. The market is well ahead of three of the four majors.

Closed today — and neither closure is comfortable

V002, US 10-year tactical long — CLOSED WRONG. Opened 7 September at 4.78% on a positioning squeeze thesis: leveraged funds short 2.06m contracts, a 4.81% double-top, and Treasury buybacks starting 9 September. Wednesday's edition published a 4.85% stop and noted it was 1.3bp away. Thursday's close of 4.970% took it out by 12bp; the loss is 19bp from entry. What makes this instructive rather than merely bad is that the supporting thesis was correct in every particular. The auctions did go well — three stop-throughs in three, Thursday's 30-year clearing 2.7bp through with a 2.61× cover and dealers taking 2.2%. The buybacks did start. And the yield rose anyway, every single session. The error was treating strong auction demand as evidence about direction when it is only evidence about clearing. A market that absorbs $22bn at 2.7bp through and then sells off 6bp is not short of buyers; it is repricing the policy path, and no amount of demand data speaks to that. This is the second time in five editions a view has been built on the right fact pointed at the wrong question.

V007, Japan banks versus exporters — CLOSED, SCRATCH, on the deadline set in writing. Wednesday's edition committed: "if Thursday's close cannot be marked, this view is closed as a scratch on Friday 11 September." It cannot be marked. A fourth consecutive session of attempts returned Japanese single-stock prices that were stale, mutually contradictory or undated — one provider served five bank and exporter names carrying four different dates spanning 21 August to 11 September, and its Advantest quote (−2.84%) directly contradicted its own same-page commentary (+3.3%). So the view closes unscored. For what it is worth, the sector proxy went against it: TOPIX Electric Appliances +0.76% versus TOPIX Banks +0.27% on Thursday, exporters outperforming by 49bp, with the bank leg itself inferred from an undated quote page. That is not a mark and it is not being treated as one. The governance point stands: a view that cannot be scored is worse than a view that was wrong, and the fix is a data source with an explicit per-name as-of date before any Japanese relative-value risk is taken again.

Opened today

V025, OAT–Bund widener, entry ≈94bp, Low conviction, 1–3 months. Theme 5 resolved into a position. On a day the ECB hiked and revised 2027 inflation up, the OAT sold off 11bp to a November-2008 high while the BTP rallied 1.3bp, putting Italy through France by roughly 17bp — the second consecutive edition in which the periphery has outperformed the semi-core into a hawkish event. Conviction is Low deliberately: 94bp is already the widest since 2012, so this is buying a breakout rather than a dislocation, and French political risk is not the driver — the debt-service path under a 3.35–3.40% terminal is. A compression inside 80bp closes it.

Friday scorecard

Twenty views open. Five closed to date: one right, two wrong, two scratch. Of the three decided views, one paid — a 1-in-3 hit rate on a sample far too small to mean anything, but pointed in a direction that does not flatter. This week alone: V013 Brent closed right at the level it named; V002 closed wrong at its published stop; V007 closed unscored after four sessions of unmarkable data; and V024, the successor to the winner, has cost roughly 8.2% of foregone upside by declining to re-own Brent above $100 four days before it printed $108.90.

That last one deserves more than a line. V013 was closed correctly and V024 was the wrong shape. The reasoning on Thursday was that cheap calls bought at $96.52 had done their work and re-owning the same risk at $100 with expensive implied vol was a different and worse trade. That reasoning is still defensible in isolation — but it produced a book with no meaningful exposure to the single largest cross-asset move of the week, in the one theme this note has been right about since edition one. The lesson is not "never take profit." It is that an expression-based exit should be replaced by a cheaper expression of the same thesis, not by an absence of one. The refined-product leg named in V024 — the diesel crack, LNG exporters — was the correct answer and was never actually put on.

What is working: V006 ASX (−3.1% from entry, the best view on the book), V017 US credit (CCC monotonically wider on no supply through an oil shock — the cleanest expression we have), V005 S&P (−1.64%), V010 USD/JPY (−1.6%, and the week's largest G10 move), V016 iron ore (third session sub-$100), V019 ETH (seventh rejection), V004 ACGB flattener (39bp from 43bp), V023 AUD/NZD (markable at last, and above entry on both marks). Against us: V009 AUD/USD, which turned overnight and is now ~0.5% below entry; V003 5s30s, marked for the first time and bear-flattened against us on the week; V014 gold, ~0.9% underwater with the invalidation seven dollars away; V021 Korea, where the KOSPI leg is still above entry. Awaiting: V001, V008 (two of three triggers fired), V011, V012, V018 (now inside its buy zone), V022 ($260/t against a $300/t trigger), V024, V025.

Portfolio-level read

The shape is unchanged — low gross, long convexity, short the yen carry, underweight the credit tail, underweight Australia — but the week has revealed a pattern worth naming. Every view that expressed a structural mechanism worked, and every view that expressed a tactical level did not. The credit underweight, the ASX underweight, the yen short and the iron-ore fade are all mechanism trades and all paid. The 10-year long, the gold entry level and the Brent exit level were all level trades, and all three either lost money or cost upside. Into an event window containing a CPI print tonight, a CLARITY vote Tuesday, an FOMC Wednesday, a BoE Thursday, a BoJ Friday and a $6.2trn expiry the same day, that is an argument for carrying less tactical risk, not more. Reduce gross into 12–18 September, own the wing while skew is at the first percentile, and let the mechanism trades run.

These are analytical framings for a professional reader, expressed in the vernacular of a macro desk. They are not personalised investment advice, do not consider any individual's objectives or circumstances, and carry no position sizing. The "what changes the view" column is the accountability mechanism; every view is logged and scored in the project's views ledger.