Daily Macro US
Fed September hike (Polymarket)
51.5c
30.5c before Warsh spoke, 52.5c three hours after. A 21-point repricing of the single most important macro question of the year, executed while this letter was already published. CME FedWatch reads about 57%, its first fresh reading in nine sessions.
pos-010 Fed hike in 2026
67.5c, FV 55 -> 67
Our biggest fair-value revision in one session, and we are honest that it is chasing: the Fed chair changed the distribution and we followed. Trim gate above 72c missed by 3.5 cents on Friday.
10Y Treasury
4.73% (Aug 28 SETTLE)
Primary source this time - Treasury's daily par yield curve, not a vendor. Up 6bp on the day; the 2-year up 14bp to 4.34%. Live around 4.72% this morning. Aug 27 confirmed at 4.67 by H.15, so Friday's provisional flag comes off.
pos-017 fair value
61 (FROZEN), model 76.8
The barrier is 7.0bp with 84 sessions left. The Aug 25 standing rule blocks the increase, so the freeze now costs 15.8 points instead of 5.8. Kept, with a pre-committed replacement rule published below.
5.00% consistency check
20.2pp gap - 4th straight failure
Model 41.7 against a live 21.5c market. The sequence is now 14.1 -> 17.0 -> 17.6 -> 20.2, one direction, growing. That is no longer a bad day for the model; it is evidence about the model.
BLS benchmark payroll revision
-79,000 (-0.1%)
RESULT, not preview. Private payrolls -178,000. Against last year's -911k and a 10-year average absolute revision near 0.2%, this is a small and benign number - the scary headline pos-004 was exposed to did not arrive.
We told you on Friday that this letter had moved to 04:06 ET and that the two events which mattered most - Warsh's first Jackson Hole keynote and the BLS preliminary benchmark payroll revision, both 10:00 ET - would be previews here and results somewhere else. Then the largest single-day repricing this book has seen happened at exactly that hour, and we watched none of it. Here is the full accounting, gate by gate, with the hourly evidence. At 14:00 UTC Friday, the hour Warsh spoke, the Polymarket contract for a September Fed hike traded 30.5c. By 17:00 UTC it was 52.5c. Our own pos-010, a hike at any point in 2026, went from 56.5c at 14:00 UTC to 68.5c at 16:00 UTC - twelve points in two hours in a position we hold with $100. Warsh said inflation is still too high, that recent cooler readings do not tell him underlying trends have meaningfully improved, and that the committee may have work to do. The two-year settled 4.34%, up 14 basis points; the ten-year settled 4.73%, up 6. That is a hawkish flattening, and it is the single most important thing that has happened to this book since we opened it. THREE WRITTEN GATES SAT IN THE PATH OF THAT MOVE AND WE COULD NOT SEE ANY OF THEM. We publish what each did. The pos-013 trim armed at 90c did NOT fire: the hourly series peaked at 88.9c on Friday morning, before the speech, and has not been above it since. The pos-010 trim armed above 72c did NOT fire either, but it missed by 3.5 cents - the position printed 68.5c and stopped. That is luck, not process, and we are recording it as luck. The pos-011 efficiency review armed at 40c DID fire, at 42.15c at 04:00 UTC on Saturday, in a weekend market with nobody in it. We conduct that review below, two days late, and we say what it would have concluded on the day. Two other things closed out. The 10-year settle we flagged as PROVISIONAL on Friday is confirmed: H.15 now carries Aug 27 at 4.67, our vendor number was 4.674%, and the flag comes off. And the session-count discrepancy we reported against ourselves has been resolved and both of Friday's numbers were wrong - the house count of 89 and the recount of 84 are replaced by a proper count of remaining bond-market sessions, which gives 84 including today. None of this changes the pos-017 freeze. The barrier is now 7.0 basis points, the model says 76.8, and the standing rule from Aug 25 still blocks any increase because the 5.00% consistency check failed for a fourth consecutive session and got worse again, 20.2 points against 17.6 on Friday. The freeze now understates that position by 15.8 points on our own arithmetic. We are keeping it, and we are pre-committing today to what replaces it if it fails a fifth time, so that the change cannot be timed to our convenience.
Today's Market Moves
Fed September 25bp increase
30.5%→51.5%+21.0pp
THE NUMBER OF THE WEEK. Flat at 30.5c through 14:00 UTC Friday, 47.5c an hour later, 52.5c by 17:00 UTC. It faded to 42.5c on Saturday and has rebuilt through the weekend to 51.5c. FedWatch about 57% - so the prediction market is roughly 5.5 points BELOW the futures-implied probability. Below our 10-point entry bar, and a sixth hawkish-Fed expression we will not add.
Fed Rate Hike in 2026 (pos-010)
57.5%→67.5%+10.0pp
Our position. Bid 67 / ask 68 on $8.18M of volume, so this mark is as solid as any in the book. Fair value 55 -> 67, built as P(Sep hike) 53 plus 30% of the remaining 47 across the October and December meetings = 67.1. Trim gate above 72c: the Friday high was 68.5c. Not triggered, by 3.5 cents.
Fed Rate End 2026 = 4.0% (pos-011)
30.65%→39.4%+8.75pp
The one-hike bucket. The whole ladder shifted: 3.75% 33.7 -> 25.15, 4.0% 30.65 -> 39.4, 4.25% 13.45 -> 19.25. Efficiency review armed at 40c FIRED at 42.15c on Saturday at 04:00 UTC. Reviewed below. Note the spread: bid 36.3 / ask 42.5, which is 6.2 points wide on a $25 ticket.
10Y Touches 4.8% Before 2027 (pos-017)
64.5%→67.5%+3.0pp
Bid 65 / ask 70. THE EXIT GATE IS ONE CENT AWAY: the written rule is sell on a 66c BID, and the bid is 65. We are not pre-empting it and we are not moving it. Correction: on Friday we wrote that the gamma feed returns last and mid but not the bid. It returns bestBid and bestAsk, and always did. We were wrong about our own data source.
10Y Touches 5.00% Before 2027 (check leg)
17.5%→21.5%+4.0pp
The diagnostic, not a position. Bid 20 / ask 23. Our barrier model on the same inputs says 41.7. Fourth consecutive failure and the widest yet at 20.2 points. If a model is 20 points rich on the far leg it is not trustworthy on the near leg either, which is the entire argument for the freeze.
Zero Fed Rate Cuts in 2026 (pos-013)
88.7%→87.95%-0.75pp
The only position that went the wrong way, and it makes no sense on the news: a chair signalling hikes should raise the odds of zero cuts, not lower them. We read the 0.75 as weekend noise on a leg where bid 87.9 / ask 88.0 is one tick wide. Fair value 89 -> 90. The 90c trim did not fire; the Friday high was 88.9c.
Screening Table
| # | Market | Expiry | Market Price | Fair Value | Gap (pp) | Direction | Volume | Confidence |
|---|---|---|---|---|---|---|---|---|
| 1 | Fed Rate End 2026 = 4.0% | Dec 31 | 39.4% | 42% | +2.6pp | HOLD $25 YES - the review fired and concluded hold; the entire 2.6pp sits inside a 6.2c spread | $$1.37M on the leg | 3/10 |
| 2 | Zero Fed Rate Cuts in 2026 | Dec 31 | 87.95% | 90% | +2.05pp | HOLD $82 YES - FV 89 -> 90 on Warsh and a benign benchmark revision; 90c trim did not fire and is 2.05c away | $$7.7M | 4/10 |
| 3 | US Unemployment 2026 >= 5.0% | Dec 31 | 10.65% | 8% | +2.65pp | HOLD $100 NO - the benchmark revision landed at -79k and was benign; next real test is Friday's jobs report | $$124K on the leg | 5/10 |
| 4 | Fed Rate Hike in 2026 | Dec 31 | 67.5% | 67% | -0.5pp | HOLD runner $100 YES - FV 55 -> 67 chasing a genuine regime change; essentially fair now, trim above 72c | $$8.18M | 4/10 |
| 5 | 10Y Touches 4.8% Before 2027 | Dec 31 | 67.5% | 61% | -6.5pp | HOLD $25 YES - marked 6.5pp RICH by our own frozen number; exit gate needs a 66c bid and the bid is 65 | $$60K on the leg | 3/10 |
Market vs Fundamentals
Market Price (red) vs Estimated Fair Value (green) — %
Top 5 Opportunities
1
Fed Rate Hike in 2026 - the twelve-point move we published a warning about and then missed — YES
↑ BUY YES-0.5pp
Market price
67.5%
Fair value
67%
Gap: -0.5pp
On Friday we wrote that Warsh began his term with an explicit bias against forward guidance and was therefore unlikely to promise a hike or rule one out. That was wrong, and it was wrong in the most expensive possible direction to be wrong in - he did the opposite, and the position we hold repriced twelve points in the two hours after he spoke. The hourly record: 56.5c at 14:00 UTC, 65.5c at 15:00, 68.5c at 16:00, settling 67.5c. We hold $100 of this at a 55c entry and the mark went from +$4.55 to +$22.73 without us doing anything, which is the point - we did not do anything, because we were not there. Fair value moves 55 to 67 and we want to be precise about what kind of move that is. It is not a model finding something the market missed. It is us updating to a fact the market already has, four days late, because the Federal Reserve chairman said the next move may be up. Analysts who mark to their own conviction after an event like this are usually just marking to the tape with extra steps, and we would rather say so than dress it up. The construction, published so it can be checked: P(hike in September) 53, blending Polymarket's 51.5c with a FedWatch reading near 57; conditional on no September hike, roughly 30% across the October 27-28 and December 8-9 meetings; 0.53 + 0.47 x 0.30 = 0.671. Against a 67.5c market with a one-cent spread and $8.18M traded, that is no edge at all, which is the honest answer.
▵ Bull case
- The Fed chair said in a prepared keynote that inflation is still too high and that recent cooler readings do not show underlying trends improving. That is as close to guidance as a chair who dislikes guidance will get.
- The two-year Treasury settled up 14 basis points against the ten-year's 6. A hawkish flattening is the bond market pricing policy, not term premium - it is the cleanest confirmation available that this was read as a policy signal.
- Cleveland's August core PCE nowcast is unchanged at 3.40% YoY with headline at 3.79%, re-verified at source this morning (stamp 08/28). A committee with 3.4% core and a chair talking about work to do has a live hiking case.
- The BLS benchmark revision came in at -79,000, or -0.1%, against a 10-year average absolute revision near 0.2% and last year's -911k. The labour market did not give the doves the excuse they needed.
▿ Bear case
- At 67.5c with our fair value at 67 there is nothing left to collect. We are holding a position that has done its work and is now priced correctly, which is a capital-efficiency question, not a conviction one.
- Inflation-adjusted consumer spending was FLAT in July after +0.4% in June. Nothing on Friday touched that, and it remains the mechanism by which a hiking Fed becomes a cutting Fed by December.
- Friday's move faded to 42.5c on Saturday on the September leg before rebuilding. Weekend prediction-market liquidity is thin and this whole repricing may be less settled than a single 67.5c print suggests.
- The August jobs report lands Friday Sep 4 with consensus at +60,000 and the unemployment rate expected to tick from 4.1% to 4.2%. July payrolls actually fell 23,000. One weak print and the September hike is not a coin flip any more.
2
10Y Treasury Touches 4.80% Before 2027 - the freeze now costs 15.8 points and we are keeping it anyway — YES
↑ BUY YES-6.5pp
Market price
67.5%
Fair value
61%
Gap: -6.5pp
The barrier is 7.0 basis points. The ten-year settled 4.73% on Friday against a 4.80% touch level, there are 84 bond-market sessions left in the year, and the arithmetic says 76.8. Our published fair value is 61, frozen since Aug 25, and it is staying there. The rule that freezes it is simple: no increase in this position's fair value until the 5.00% consistency check reproduces. Today it failed for a fourth consecutive session and by the widest margin yet. Same model, same inputs, 27.0bp barrier, 41.7 after the haircut, against a market that trades 21.5c with a bid at 20. The sequence of failures is 14.1, then 17.0, then 17.6, then 20.2 points - one direction, growing. That is no longer a bad day for a model. It is evidence about the model, and the evidence says a fixed-volatility barrier model over-prices far touches by construction and cannot be trusted on the near touch either. So we carry a 61 against a 76.8 arithmetic and a 67.5 market, and the freeze now understates the position by 15.8 points instead of Friday's 5.8. We also note the uncomfortable consequence: at 61 we are marked 6.5 points BELOW the market, so our own published fair value says this position is rich and should be sold, while the written exit gate says sell on a 66c bid and the bid is 65. We are doing neither. We are not selling early because our fair value says so, and we are not moving the gate because it is close. PRE-COMMITMENT, WRITTEN TODAY BEFORE WE KNOW TOMORROW'S PRICES: if the 5.00% check fails a fifth consecutive session, the barrier model is retired from the published fair value and kept only as a diagnostic, and pos-017 fair value is re-anchored to the market mid less a two-point liquidity discount. That rule takes effect at the next run regardless of whether it helps or hurts. We are writing it now precisely so it cannot be timed.
▵ Bull case
- Seven basis points. The ten-year settled 4.73% and needs 4.80% once, at any point in four months, in a market that has moved 6 to 14 basis points in a single session twice in the last week.
- The 4.6% leg on the same ladder has already resolved to 1.00 - the touch has happened at that level. The distribution of realised touches this year has been running ahead of what a fixed-vol model would have predicted at the start.
- The global long end is doing the work: the Bund at its highest since 2011, French yields the highest since 2009, UK 30-year borrowing costs near May peaks that were the highest since 1998. US 4.80% does not need a US-specific story any more.
- Warsh gave the market a reason to price policy tightening, and the two-year moved 14bp on it. Hiking cycles do not usually leave the ten-year 7bp below a round number for four months.
▿ Bear case
- Our own consistency check has failed four consecutive times and is getting worse. The single most likely explanation is that the model is wrong, and the model is the only reason we ever claimed edge here.
- The market is 67.5c with a 65/70 spread - five points wide on $60K of volume. That is a thin leg, and a thin leg is a bad place to insist you know better than the price.
- Our published fair value of 61 says this is 6.5 points rich right now. Taking our own numbers seriously means this position is a sell, and only the wording of the exit gate is keeping it open.
- The mechanism has a date on it: the buyback window expires Nov 4 while the barrier runs to Dec 31, so the last eight weeks of the contract lose the supply-side support that was part of the original thesis.
3
Fed Rate End 2026 = 4.0% - the review fired on a weekend and we conduct it two days late — YES
↑ BUY YES+2.6pp
Market price
39.4%
Fair value
42%
Gap: +2.6pp
This is the gate that actually fired. The efficiency review was armed at 40c; the hourly series shows 42.15c at 04:00 UTC on Saturday, a weekend print in a market with nobody in it, and the leg has since settled back to 39.4c. Friday's letter promised that if a trigger went through while we were away we would say so and act then rather than quietly re-date it. So: it went through, here is the review, and here is its answer. Exiting is worth $1.69. The position is $25 at a 34c entry, which is 73.529 shares. The bid is 36.3c, so a sale realises $26.69 against a $25 stake. Holding to a correct resolution pays $73.53. Giving up $46.84 of upside to book $1.69 and free $26.69 of capital is not an efficiency trade, it is a rounding error with commission risk attached. The review concludes HOLD, and it would have concluded HOLD on Saturday too - the 42.15c print carried a bid that was almost certainly worse than today's 36.3. We are also raising the review threshold to 50c, and we want to flag the shape of that decision rather than bury it: we are moving a trigger upward on the same day it fired, which is exactly the pattern that should make a reader suspicious. The reason is arithmetic, not convenience. The bid-ask on this leg is 36.3 / 42.5, 6.2 points wide on a $25 ticket, so any trigger below roughly 50c produces a review whose answer is structurally always hold. A trigger that cannot produce a trade is decoration. Fair value 31 -> 42, and unlike pos-010 this one is derived rather than judged. The ladder now prices 39.4 / 19.25 / 4.5 across the 4.0%, 4.25% and 4.5%-plus buckets, which sums to 63.15 for any hike, against 67.5 on the standalone hike contract - the two markets on the same event disagree by 4.35 points, which we note against ourselves as a reason not to lean hard on either. Taking our pos-010 fair value of 67 and the ladder's conditional shape gives 67 x 39.4/63.15 = 41.8.
▵ Bull case
- The 4.0% bucket IS the one-hike outcome, and Friday moved the probability of exactly one hike more than anything else on the ladder: 3.75% fell 8.5 points and 4.0% gained 8.75.
- Our derived fair value of 41.8 sits above the 39.4 mid, and the two-market inconsistency (63.15 on the ladder versus 67.5 on the standalone contract) points in the same direction - the ladder looks slightly cheap.
- The mechanical link to pos-010 is deliberate discipline. Marking this leg independently would let us tell two different stories about one event, which is the failure mode a book this correlated is most exposed to.
- A single 25bp hike and then a hold is the modal path a chair who says there may be work to do actually delivers, especially with an election-free December meeting and 3.4% core.
▿ Bear case
- The spread is 6.2 points on a leg where our claimed edge is 2.6. On the ask we are 0.5 points RICH. The edge does not survive contact with the order book, and we should say that rather than print a positive gap and move on.
- The review trigger fired on a Saturday print in a market with essentially no depth. Treating a weekend tick as a real level is exactly the sort of thing that produces bad rules.
- Two hikes gets you 4.25%, not 4.0%, and 4.25% has gone from 13.45 to 19.25 in one session. If Warsh means a campaign rather than an adjustment, our bucket is the wrong one.
- We are raising a threshold on the day it fired. Even with a good arithmetic reason, that is the kind of decision that looks different from outside than it does from inside.
4
Zero Fed Rate Cuts in 2026 - the trim did not fire, and here is the hourly proof — YES
↑ BUY YES+2.05pp
Market price
87.95%
Fair value
90%
Gap: +2.05pp
Friday's letter said the 90c trim was armed, sitting 1.3 cents away, and might fire during the reaction to Warsh while we were not watching - and that if it did we would say so and act rather than re-date the trigger. It did not. The hourly series since Friday midnight UTC peaks at 88.9c at 03:00 UTC, before the speech, and the post-Warsh session never traded above 88.75. The low was 87.55 on Saturday afternoon and it is 87.95 now, one tick wide at bid 87.9 / ask 88.0. The gate is not triggered and never was. We are publishing the numbers rather than the assertion because the whole value of arming a rule in advance is that somebody can check it afterwards. The position went slightly the wrong way on a day that should have helped it, which is worth a sentence of honesty rather than a rationalisation: a chair signalling hikes makes zero cuts more likely, not less, and this leg fell 0.75. On a $7.7M market with a one-tick spread we read that as weekend drift rather than information, but we are not going to pretend it confirmed us. Fair value 89 -> 90 on two pieces of genuine news. Warsh is the first, and it is decisive for this contract in a way it is not for the others: the argument for a 2026 cut required a committee that thought inflation was beaten, and the chair said on the record that he does not think that. The second is the benchmark payroll revision, which is the event we had been pointing at for a week. It came in at -79,000, or -0.1%, with private payrolls -178,000. Set against last year's -911k and a 10-year average absolute revision near 0.2%, that is a smaller-than-typical correction. The labour market did not hand the doves anything.
▵ Bull case
- The chair said inflation is still too high and that cooler recent readings do not show underlying trends improving. A committee cutting in 2026 now has to overrule its own chair in public.
- The benchmark payroll revision at -79,000 is smaller than the 10-year average absolute revision near 0.2% and a fraction of last year's -911k. The single scheduled event that could have flipped this thesis passed harmlessly.
- Cleveland's August core PCE nowcast is unchanged at 3.40% YoY, re-verified at source this morning with the stamp at 08/28. No committee cuts into 3.4% core with a hawkish chair.
- GDPNow had Q3 at 4.61% as of Aug 26 - carried, because the model does not refresh until tomorrow, and flagged as carried. Growth like that removes the emergency case for cutting entirely.
▿ Bear case
- Inflation-adjusted consumer spending was FLAT in July after +0.4% in June. This is still the only real mechanism by which this position loses: not inflation falling, but the economy weakening enough that the committee cuts into it.
- July payrolls fell 23,000 and August consensus is +60,000 with the unemployment rate expected to tick to 4.2% from 4.1%. Two soft prints in three months is the start of a pattern, not noise.
- At 87.95c there are 12.05 points of upside against 87.95 of downside. Even a correct thesis is a poor risk-reward at this price, and our own trim rule exists because of that.
- The leg fell on the most bullish news it could have received. We are calling that noise, but if it happens twice the interpretation changes.
5
US Unemployment 2026 >= 5.0% - the event we had been pointing at for a week landed, and it was nothing — NO
↑ BUY YES+2.65pp
Market price
10.65%
Fair value
8%
Gap: +2.65pp
For a week this letter described the BLS preliminary benchmark payroll revision as this position's only real scheduled event, and warned about a specific risk: a frightening headline moving the price without moving the probability, because a benchmark revision restates establishment-survey payroll LEVELS while this contract settles on the household-survey unemployment RATE. Different surveys. The result was -79,000, or -0.1%, with private payrolls revised down 178,000. That is smaller than the 10-year average absolute benchmark revision of roughly 0.2% and a fraction of last year's -911,000. There was no frightening headline. The YES leg went from 11.2c to 10.65c across the weekend, which is the market agreeing. So the distinction we spent a week explaining turned out not to matter, because the number was too small to test it. We are writing that down rather than claiming the call: we were right about the mechanism and the size of the revision made the mechanism irrelevant. Fair value on the YES stays at 8, unchanged. The next real test is Friday. Consensus for the August employment report is +60,000 payrolls with the unemployment rate expected to tick up to 4.2% from 4.1%, and July payrolls actually fell 23,000. Getting from 4.1% to 5.0% by December needs roughly 1.4 million additional unemployed in four months. A 0.1 point tick is not that, but a run of them starts to be.
▵ Bull case
- The benchmark revision at -79,000 is a small number by historical standards and revised the payroll level, not the unemployment rate this contract settles on.
- Initial claims printed 203,000 against about 208,000 expected with the insured unemployment rate steady at 1.2%. That is not a labour market shedding 1.4 million people in four months.
- GDPNow had Q3 at 4.61% as of Aug 26 (carried, refreshes tomorrow). Unemployment does not rise nine tenths of a point against growth like that.
- A Fed that is discussing hikes rather than cuts is a Fed that does not currently see labour-market collapse in its own data.
▿ Bear case
- Consensus has the unemployment rate ticking to 4.2% on Friday, and July payrolls fell 23,000. The direction of travel is the wrong one for a NO holder, even if the level is nowhere near.
- Private payrolls were revised down 178,000, more than twice the headline revision. The private-sector picture is weaker than the -79,000 suggests.
- A hiking Fed raises the tail risk of a policy-induced break later in the year, which is exactly the scenario that could move the unemployment rate fast in Q4.
- At 10.65c the remaining collection is $10.65 of price against a position already worth $119.13. The edge is real but small and the capital is tied up until December.