Actionable insights — low correlations, the curve divergence, and the NFP test

1. Sit out month-end — and treat the urge to trade it as the signal to stop

1. Mark the last two or three sessions of each month as a no-new-risk window in advance, so the decision is made when you are calm rather than when the tape is moving.
2. Know what is actually driving price there: index rebalances, pension and target-date rebalancing, fund marks, and options expiry — flows with a deadline, not opinions. They move price for reasons that carry no information about your thesis.
3. Use the urge itself as the trigger. His formulation is a rule about the trader, not the market: "whenever I feel the urge to trade it" — the appetite to act into flow-driven price action is the tell that you are about to mistake noise for signal.
4. Restart the clock only once month-end is behind you, and explicitly discount the preceding sessions when reading charts — do not let a month-end distortion set your reference level for the new month.
5. Keep existing positions; this is a rule about initiating, not about liquidating.

• Month-ends that coincide with quarter-end or a large index reconstitution — the distortion is bigger and lasts longer.
• A "breakout" or "breakdown" printed on the last day of a month: it needs to be re-confirmed in the first few sessions of the next one before it counts.
• Your own rationalisation — the flow explanation is always available after the fact, so the discipline only works if the window was marked beforehand.

2. Read low correlation and wide dispersion as a fragility gauge, not a health gauge

1. Track two cross-sectional measures alongside the index: realised (or implied) correlation across the constituents, and dispersion — the spread between the best and worst performers.
2. Resist the standard interpretation. A low-correlation, high-dispersion tape is normally sold as a "stock-picker's market"; ask instead what that regime lets people do.
3. The answer is the mechanism: low correlation makes concentrated single-name risk look diversified, so risk models permit more of it, and hedges get sized for a world where names do not move together. Dispersion is the accumulated fuel.
4. Name the state, not the date. He calls it "a market vulnerability or state of criticality" — a description of stored energy, with no claim about the spark.
5. Convert the read into a position type rather than a direction: if the failure mode is correlation snapping to one, the thing that pays is index-level convexity (or a correlation/dispersion trade), not a short in any single name.

• Correlation turning up from the lows while the index is still rising — the first sign the crowd is falling into step, and typically the last cheap moment for hedges.
• Implied correlation trading well below realised, which is what makes index vol cheap relative to the single-name vol underneath it.
• The invalidation: dispersion narrowing gently while the index grinds on. A gradual re-convergence discharges the same energy without a Risk Off event — the outcome this framework does not pay for.
• The base rate against acting too early — low correlation can persist for quarters, which is exactly what the next insight is there to handle.

3. Keep the Dornbusch rule between the diagnosis and the position

1. Once the diagnosis is made, immediately separate two questions: is the condition real and when does it decompress. Answer only the first with confidence.
2. Apply the asymmetry explicitly: "it takes far longer for conditions to decompress than you think, but then it always happens faster than you imagined." Both halves matter — the first kills carry-heavy expressions, the second kills reactive ones.
3. Choose the instrument from that asymmetry, not from conviction. A structure that bleeds while you wait is the wrong one; a structure that costs little and pays non-linearly is the right one.
4. Budget the waiting explicitly — how much are you willing to pay per month to stay in the trade, and for how many months — before entering.
5. Never let the delay be read as refutation. A condition that has not decompressed yet is not evidence the diagnosis was wrong; it is the base case of the rule.

• The tell that you have violated it: a hedge you are tempted to take off because it has cost you for months. That is usually the moment it was designed for.
• Rolling costs and time decay accumulating faster than budgeted — the practical reason "early" becomes "wrong."
• The opposite failure — waiting for confirmation. By construction the decompression happens faster than expected, so a confirmed signal is a filled-in price.

4. Use the maturity that refuses to confirm as the tell that a bond move is positioning, not macro

1. Watch the whole curve in futures, not just the 10-year yield — 2yr, 5yr, 10yr and the 30yr bond — because futures is where the leverage and the positioning sit.
2. Ask the confirmation question on every sizeable move: did every maturity participate? A genuine macro repricing (inflation, term premium, growth) drags the whole curve, and the long end usually leads it.
3. Flag the non-confirmation. When the front and belly make new lows in futures and the long bond does not break down, the move lacks a single macro driver — which points at flows: shorts being squeezed, hedges being lifted, a crowded trade unwinding.
4. Pair the observation with the positioning data (COT, dealer and leveraged-fund net positions) rather than treating the price move as self-explanatory — note he leads the sentence with the word positioning.
5. Log it without trading it. "Maybe nothing... maybe something" is the right register for an anomaly with no resolution yet — the entry is the resolution, not the anomaly.

• Resolution up: the 30yr finally breaking down too — the move gets a macro driver and stops being a positioning artefact.
• Resolution down: the front and belly reversing back through their old lows — the classic signature of a squeeze that has run its course.
• Any change in the funding backdrop (auction tails, swap spreads, repo specialness) that would explain the front-end move mechanically rather than macro-economically.
• The corollary for hedging: if the long end has not sold off, the cheapest duration hedge against a risk-off is still available at the long end.

5. Pre-register a dated, bounded forecast with the gauge that would confirm it — then judge the reaction, not the print

1. State the forecast with a window, not a date, and with a shape rather than a magnitude — his is "a soft patch in US data between midyear and mid autumn," made in June.
2. Nominate in advance the single gauge that would confirm or deny it. He uses the Citi Economic Surprise index, which measures data against expectations rather than against its own history — the right instrument when the claim is about forecasters being caught out, not about the level of activity.
3. Identify the dated event inside the window that will test it (here, August non-farm payrolls, due Friday 4 September) and note it before the print.
4. Do not forecast the number. He offers no NFP estimate, no threshold and no recession call — the window and the gauge are the entire claim, which is what makes it falsifiable.
5. Then make the reaction the real signal: "Will a miss on NFP be enough to shake Risk and push some rotation in bonds' direction?" A weak print that fails to bid bonds says far more about the regime than the print itself — it is a live test of whether duration still hedges equity risk.

• The Citi Economic Surprise index turning back up before the window closes — the cleanest single-line falsification of the soft-patch call.
• The payroll reaction itself: a weak print with no bond rally is the more consequential outcome, because it says the risk-off hedge has stopped working — and it would tie straight back into the curve non-confirmation above.
• Revisions rather than headline prints — a soft patch shows up in downward revisions to prior months before it shows up in a headline miss.
• The subtler trap: expectations catching down to the data. The surprise index mean-reverts by construction once forecasters lower the bar, so a rising reading is not automatically a stronger economy.