The form changed, but the function remains: markets still infer the policy path from public signals. AI-assisted illustration, Miyama Capital.
Drafted August 28, 2026, the day of the speech. Before publication, revised against the official closes of August 28, the Treasury’s buyback operations on August 19 and September 9, and the September 4 jobs revision. The observation start date for every threshold is the publication date.
What He Refused
On August 28, Fed Chair Kevin Warsh delivered his first full read of the economy since taking office. The speech was titled “In Our Time,” and the day marked his hundredth day on the job (source: Federal Reserve official speech text, 8/28/2026). Except where noted, the material and data in this section and the next come from that text; below I refer to him as Warsh or the Chair.
He opened with an outline, then added a line: call it an outline, call it a roadmap, just don’t call it forward guidance.
What followed was an entire section devoted to rejecting forward guidance. He framed it as a crisis-era instrument (his own footnote dates the normalization to December 2008) and argued that in normal times it should be constrained, or it manufactures ambiguity in the name of clarity. Making quasi-commitments on rates across cycles binds decision-makers precisely when they most need the freedom to judge. And when the central bank watches market prices while the market watches central bank signals, each side reads the other’s mirror, and both go blind to new information at the same moment. He called this the hall-of-mirrors problem.
Then he refused the fallback request too — “at least give us a reaction function.” He said he would love for human understanding of the economy to be precise enough to lean on something like a Taylor rule, but the knowledge isn’t there, and the factors that actually matter change over time. Explaining a reaction function through forecasts works better in theory than in practice, better in the lab than in the field.
His closing line is worth keeping on the record: he committed only to a disciplined process of judgment, not to any predetermined policy decision.
What He Gave Instead
Having refused, he did the one thing he could not avoid. He started describing how he reads the economy right now. The figures below are all as cited in the speech text; the underlying sources are the BLS, the BEA, and market data, as of the speech date.
On employment, his read: unemployment at 4.1%, the four-week average of initial claims near a multi-decade low, the labor market broadly consistent with full employment.
On the strength of the economy he offered several numbers. Investment in equipment and intangibles was up about 9% on a four-quarter basis, the highest since 2021; more than half of this year’s capex growth may be attributable to AI-related build-out. S&P 500 constituent earnings were up more than 20% year over year. Credit spreads sat near the low end of their historical range, and banks’ commercial and industrial lending standards were on the easy side. Housing and agriculture were under pressure, but on the whole he found it hard to describe broad financial conditions as restrictive.
Prices looked worse. PCE was up 3.7% over twelve months, 4.1% over six. His own dispersion measure showed that of 199 PCE components, 54% had risen more than 3% over the past twelve months — well below the roughly 77% post-pandemic peak, but still clearly above the 32% that prevailed in the two decades before the pandemic. On a six-month annualized basis the figure was 49%. He said the summer’s PCE and CPI prints came in better than expected but didn’t tell him the underlying trend had genuinely improved. He also flagged the recent rise in goods prices as worth watching.
Three things he made explicit. The target is 2% measured by PCE, a fixed number. CPI and the core versions of both are imperfect but tell the same story. And what he’s trying to capture is the underlying trend after stripping out idiosyncratic factors: trend, not any single point, and not just direction but speed.
Then the standard itself. He said he must have sufficient confidence that underlying inflation is moving clearly, and at a sufficient pace, toward the target; otherwise there is still work to do.
No numerical threshold. No time window. No maintenance condition. By any operational test, it isn’t a reaction function. It is, however, a publicly stated standard for judgment.
The Market’s Answer, Written in the Shape of the Curve
First, the nature of the data. My initial draft used intraday snapshots taken while the speech was ongoing; the final version uses closing values from the Treasury’s constant-maturity yield series throughout. The two differ at the long end: intraday, the 30-year dipped slightly; at the close it turned to a small rise. The inferences below rest only on closing changes and the shape of the curve.
From August 27 to 28, the 2-year rose from 4.20% to 4.34%, up 14 basis points; the 10-year from 4.67% to 4.73%, up 6; the 30-year from 5.19% to 5.22%, up 3. The 2s30s spread compressed from 99 basis points to 88 (source: U.S. Treasury constant-maturity yield series, 8/27/2026 and 8/28/2026). Implied odds of a September hike rose from about 36% before the speech to about 56% after, and to about 60% the next session (source: CME FedWatch, per CNBC 8/28/2026 and 8/31/2026).
Now the composition of the move. In the same series, the 10-year real yield rose from 2.34% to 2.42% and the 30-year real from 2.92% to 2.96% (the 2-year has no matching real tenor); nominal-minus-real breakeven inflation fell 2 basis points at the 10-year and 1 at the 30-year (source: U.S. Treasury constant-maturity nominal and real yield series, same dates). The entire move sat in real rates, with breakevens flat to marginally lower.
The curve flattened, front-end led, real-rate driven.
Consider two channels that can operate at once. One is a policy-path update: if the market extracted a more hawkish standard from the speech, the most direct response should land at the front end, where the next few meetings are priced. The other is policy-function uncertainty: if refusing an explicit reaction function raised the market’s estimation error around the medium-to-long-term path, that should in theory push up the term premium, showing up most directly as a steepening, the long end rising relative to the front.
What August 28 showed is the first channel clearly dominating. The long end did move, but only a fifth as much as the front, and breakevens didn’t rise. This can’t prove the second channel is absent; it only says the second channel didn’t drive that day’s long-end pricing.
So the clearest signal the speech left is that the market did extract usable information from the Chair’s standard and immediately used it to update near-term policy odds. He gave no path; the market backed one out on its own, and once it had, it knew to write the answer at the front end.
Within a day, this went from price to text. The chief economist at a large accounting firm rewrote his standard into a reaction function in a client note: if inflation fails to move toward 2% at a sufficient pace, be ready to tighten (source: Fox Business, citing EY-Parthenon chief economist Gregory Daco, 8/28/2026). He says he won’t give a reaction function; the market’s first response is to estimate one anyway.
Guidance by Commitment, Guidance by Inference
I separate these into two forms.
Guidance by commitment is the practice of the past dozen years. The central bank tells the market roughly where it’s headed, through statement language, the dot plot, or explicit conditional commitments. What the market receives is a conclusion.
Guidance by inference is what’s happening now. The central bank doesn’t tell the market where it’s headed; it only makes public the standard by which it reads the world, and the market backs out the path itself. What the market receives is a partial description of the function. The conclusion is left to it to compute.
The distinction has precedent. The literature has long split forward guidance into a “commitment” type and a “forecast” type — the former binds future action, the latter merely discloses the central bank’s judgment (Campbell et al., 2012, Brookings Papers on Economic Activity). This academic taxonomy is the primary external basis for the framework in this section. As background, a similar distinction has recently been drawn on the market side, separating “committing to a rate path” from “explaining which data matter and how they feed the decision,” treating the latter as the foundation of a central bank’s credibility and accountability; I treat that only as a background reference, not as a load-bearing part of the argument here. What I want to work through is the current form this distinction takes at the Fed.
Pull back, and this isn’t unique to the Fed. Guidance by commitment traced the same arc across major central banks: the Fed moved from date-based commitments in 2011 to conditional guidance with unemployment and inflation thresholds in 2012; the Bank of England hung out a 7% unemployment threshold in 2013 and was soon forced to walk it back as joblessness fell too fast; the ECB introduced forward guidance from 2013; the Bank of Japan tied its guidance to yield curve control until dismantling it in 2024 (source: public records of official policy statements, including the Fed’s August 2011 and December 2012 FOMC statements, the BoE’s August 2013 Inflation Report, the ECB’s July 2013 Governing Council statement, and the BoJ’s March 2024 policy decision). The exit routes differed, but none of them could exit the act of publicly explaining how it reads the economy. That is the structural source of guidance by inference.
So the Chair can kill the first kind. The second is much harder to kill.
The reason isn’t personal; it’s in the structure of the office. As long as the Chair must regularly answer, in public, how he reads inflation, how he reads employment, which risk weighs more, and what would mean policy still has work to do, the market will feed those answers back into its models, back out a policy lean, and reprice.
None of this negates his reform. He said himself that what he wants to change is the “form and function” of the Chair’s forward guidance; he wants market participants to track real economic information and form their own expectations. Guidance by inference asks the market to do exactly that. The real success condition should be defined this way: his every sentence stops being read as a commitment about future policy, and economic data becomes the primary input variable again. The form has already changed. Whether the function changes with it is what the next few sections take up.
A central bank can control how much it says. It cannot control how much signal the market extracts from what it says.
Two Layers of Estimation Error
So far this is intuitive enough. The genuinely awkward part is below.
Under guidance by commitment, the market only needs to receive a conclusion. Under guidance by inference, it has to estimate two things at once: the mapping from data to policy, and which observable signals the decision-maker treats as trend versus noise, along with how much weight he puts on each.
The August data laid the second one bare.
Align the definitions first. July core CPI was up 2.5% year over year, 0.2% month over month (source: BLS, Consumer Price Index – July 2026, released 8/12/2026; YoY NSA, MoM SA). July core PCE was up 3.3% year over year, headline PCE 3.7% (source: BEA, Personal Income and Outlays, July 2026, released 8/26/2026). On a core basis, PCE runs 0.8 percentage points above CPI. The width and direction of that wedge is itself one reason the market has to re-estimate the signal weights.
He has already answered “which index”: the target is measured by PCE, and both gauges tell the same story. The direction is indeed consistent, both above 2%. What he hasn’t answered is “how to read speed” — and his standard is precisely a standard about speed.
The same price data supports different answers on speed. Core CPI’s twelve-month change slipped from 2.6% to 2.5%, which reads as deceleration; headline PCE’s six-month change of 4.1% sits above the twelve-month 3.7%, which reads as acceleration; his own dispersion measure, 54% on twelve months and 49% on six, reads as marginal improvement. He put these together and concluded “no genuine improvement.” What the market has to learn is how he aggregates these into that conclusion.
Concretely, there are at least four weights the market has to estimate, and one of them dwarfs the rest.
Whether to look through energy is the weight that matters most right now. He quantifies headline PCE, which includes energy, yet says policy targets the underlying trend after stripping idiosyncratic factors; and he flags goods prices as worth watching. With Brent above $100 a barrel, the tension between those two statements is where the real ambiguity sits (source: CNBC market reporting, 9/9/2026).
Then the time window: twelve-month, six-month, three-month momentum, and how much weight each gets. He says look at trend, not a single point, without saying what length of ruler measures it. The dashboard is a problem of its own. The 199-component dispersion isn’t a series the BEA publishes directly; you rebuild it from the components, and most market participants can’t get a real-time version of the indicator the Chair is watching. Then the non-traditional inputs: he lists “money matters” as a principle and asks the Fed to read unfiltered market signals wherever possible, including asset prices, Treasury prices and volumes, the dollar, credit costs, and commodity prices. How any of this enters the decision is written down nowhere.
So the uncertainty of guidance by inference covers not just “what the reaction function looks like” but “which signals the decision-maker uses, and how he weights them.” The weights themselves become latent parameters. The two layers of estimation error stack.
Why This Comes Back to the Long End
Finish that line of reasoning and you walk back to familiar ground.
The mapping function and the signal weights are both estimated, and estimation carries error. Error means the market’s conditional variance around the future short-rate path can rise. And the long-end yield decomposes roughly into expected future short rates plus a term premium, where the term premium is the compensation investors demand for path uncertainty, accumulating along the duration and concentrated at the longest tenors.
Guidance by inference
↓
Market must estimate the policy mapping and the signal weights
↓
Estimation error can raise uncertainty around the future short-rate path
↓
Upward pressure on the term premium
↓
Longest tenors are, in theory, most sensitive to this channelThe conclusion of this chain is “upward pressure,” not “the long end should rise.”
It isn’t the only force acting on the long end. A more hawkish standard, if it raises the market’s confidence that the central bank will ultimately bring inflation down, can also lower forward inflation and the tail risk of a policy that loses control. So the same speech can raise reaction-function uncertainty and raise disinflation credibility at once, and the two channels point in opposite directions at the long end. The August 28 data can only say the former didn’t overwhelm the latter and the other long-end forces; that day’s mix of marginally lower breakevens and higher real yields is consistent with the credibility read, and some reporting even framed the front-end jump as the Chair easing the bond market’s doubts about his credibility (source: Bloomberg, 8/28/2026). But this is still a single-day inference, and a 1-to-2-basis-point breakeven move is close to noise.
Leave a methodological note here. A credibility-shock event day may be usable to separate the two components of the long end, because it acts on the two channels in opposite directions; the official series provides nominal and real together, which makes the separation operable on closing data. A single observation isn’t enough to validate it, so it needs an accumulated sample. I’m putting it on the long-run list of observation methods.
The evidence in what follows comes in three layers, and they’re worth keeping apart. The event-day evidence uses the single case of July 29. The monthly correction fixes a component misallocation from a prior piece. And the buyback contamination is how the Treasury, from September on, stops this stretch of the long end from being clean.
Start with the event-day evidence. The case that really drove guidance by inference into the long end came a month earlier. At the July 29 post-FOMC press conference, he referred several times to the marked rise in Treasury yields, in a tone the market read as welcoming its own tightening of financial conditions (source: Yahoo Finance and PBS reporting on the press conference, 8/27/2026 and 8/28/2026). That day’s closing data: the 2-year fell 4 basis points, the 30-year rose 11, and 2s30s widened 15; implied odds of a September hike fell from about 79% before the meeting to about 60% after the press conference (source: U.S. Treasury constant-maturity yield series, 7/28/2026 and 7/29/2026; CME FedWatch, per Schwab 7/29/2026). That time too he gave no path, only a stance on the level of the long end, and the market traded it as guidance for the long end: the front end marked down its need for hikes, the long end took on the tightening job.
Set the two communication days side by side. July 29 was a steepening, front down and long up; August 28 was a flattening, front up and long barely moving. Same Chair, two curve shapes. What the market is inferring is which segment of the curve he wants the tightening to come from. This says more about where guidance by inference lands than any single-day basis-point count: whichever segment of the curve the Chair speaks to is where the guidance effect lands.
Now the monthly correction. The July 29 baseline was 30-year nominal 5.20%, real 2.98%, breakeven 2.22%, all on the same constant-maturity basis. On August 17 the 30-year nominal closed at 5.31% and real at 3.06%, the high of this cycle, with the nominal the highest since 2007 (source: U.S. Treasury constant-maturity nominal and real yield series, 7/29/2026 and 8/17/2026; “highest since 2007” per CNBC, 8/20/2026). The August 20 30-year TIPS reopening auction stopped at a real yield of 2.973%, the highest for that tenor since October 2001 (source: U.S. Treasury auction results, 8/20/2026); that’s on an auction basis, not directly subtractable from constant-maturity, so it stands only as corroboration of the level. The 30-year breakeven spent all of August between 2.21% and 2.28% (source: Treasury nominal-minus-real constant-maturity series, August 2026 sessions).
At the monthly scale, real yields parked at multi-year highs while breakevens didn’t budge. That combination is unfriendly to the narrative that a credibility discount on inflation drives the long end. My earlier piece on the long end’s tolerance band wrote the real-yield and breakeven layers as roughly half and half; that was a misallocation. At the monthly scale, the repricing of real yields, growth, and supply is the main driver, and the credibility discount is a tail risk. I’m booking that correction here.
The Third Mirror
Now the buyback contamination. The next question in long-end pricing isn’t inside the Fed.
His hall-of-mirrors argument is that the Fed watches the market and the market watches the Fed, and both go blind together. His prescribed fix is for the Fed to read unfiltered market prices as much as possible, including Treasury prices and volumes. But from August 19, this same stretch of prices gained an active shaper. The Treasury announced it would at least double the size of its liquidity-support buybacks in the 10-to-30-year sector, running from September 9 through November 4; on the announcement day the 30-year closed down 9 basis points, the 2-year unchanged, real yields likewise down 9, breakevens unchanged, before recovering 8 basis points within two sessions to 5.27%, near the pre-announcement 5.28%. The first operation on September 9 was scaled up to $6 billion, and the 10-year closed at 4.83%, which market reporting called the highest since November 2023 (source: U.S. Treasury announcement, 8/19/2026; U.S. Treasury constant-maturity series, 8/18/2026 through 8/24/2026 and 9/9/2026; CNBC market reporting, 9/9/2026).
I’m not evaluating these operations. What I’m describing is only the structure: the Fed takes long-end prices as an input while the Treasury is simultaneously rewriting that input; the Chair reiterated that the short rate is the primary tool and that unconventional policy should be used sparingly, which the market read as the Fed’s balance-sheet duration continuing to shorten (source: CNBC citing Gavekal Research, 8/31/2026), the opposite direction from the Treasury’s long-end buybacks. Some market analysis also argues that the buybacks deviate from the “regular and predictable” convention of debt issuance and could, over the long run, push risk premia higher instead (source: CNBC citing a JPMorgan analyst client memo, 8/20/2026).
For the mechanism in this piece, this carries a concrete consequence: any threshold that tests the “Fed communication mechanism” using term-premium model estimates will be contaminated by the buybacks between September 9 and November 4. Call three below is adjusted accordingly.
Where I Could Be Wrong
Direct observation, high confidence: the speech text itself, the official closing changes at each tenor, the inflation and employment figures cited in the speech, and the content of the Treasury’s buyback announcement.
Inference, medium confidence: the guidance-by-inference framework, and reading the July 29 press conference as a long-end case. The framework is consistent with both the speech text and that day’s price reaction, but consistent isn’t identified; the second rests on official price data, with the tone of the press conference described via market reporting.
Inference, low confidence: three points, all of them important.
The first is the quantitative transmission from inference error to the term premium. The mechanism holds logically, but I can’t isolate the actual magnitude, and during the buyback operation it can’t be isolated at all.
The second is attributing that day’s relative resilience in the long end to the credibility channel. It’s a candidate explanation, the sample size is one, and the breakeven move is close to noise.
The third matters most. The large reaction that day is consistent with at least three states. One: the regime shift hasn’t taken hold, and the market still treats the Chair as the primary signal source. Two: it’s the transition phase of the regime shift, the market is learning his standard, so each utterance carries unusually high information content — and the high information content is precisely because the framework isn’t yet understood. Three: the first two aren’t mutually exclusive. The market can gradually put economic data back at the center while still weighting heavily how the Chair reads that same data, as long as the mapping from data to policy remains unclear. If it’s the third, what we end up seeing isn’t the Chair’s words losing influence. Their role shifts from “giving the answer” to “helping the market calibrate its decoder.”
These three states look observationally equivalent in a single day’s price. To tell them apart you can’t look at the level; you have to look at the time derivative. That’s also why the next section’s thresholds are written as ratios, and why I’ve demoted that one to a supporting indicator.
The Calls, and the Thresholds That Score Them
⚠️ On the nature of what follows: these thresholds are operational definitions for research monitoring. They bind my own prediction record, and I will score myself against them. For the reader, they are not trading rules and do not constitute any basis for entry or exit. The observation window runs from this memo’s publication date through year-end 2026.
Call one (form). The low-guidance form of policy communication holds unchanged through year-end 2026.
Termination condition: if, before year-end 2026, the Chair himself or an FOMC statement again provides a conditional policy path with specific numerical thresholds and a time window, the form is deemed terminated. A dot plot published on the committee’s usual schedule does not constitute termination. In June the Chair didn’t submit his own dot and said he would conduct a year-end review of the press conference, the dot plot, and the meeting format (source: CNBC, 6/17/2026). Whether he again declines to submit on September 16, and the outcome of the year-end review, are the two observation points for this call. This kind of termination is a switch in institutional state, not a wrong call. Once it happens, I’ll score call one as terminated, and the score is independent of where inflation data goes.
Call two (mechanism). The guidance effect hasn’t disappeared, and it currently lands mainly on the shape of the curve via the Chair’s communication days: on the front end in August, on the long end in July.
Primary evidence: two series. First, the path of the implied September-hike odds. About 79% before the July 29 meeting, about 60% after the press conference; 42% after the August 12 CPI; about 36% on August 27; about 56% after the August 28 speech, about 60% on August 31; 58% after the September 4 jobs report (source: CME FedWatch, per Schwab 7/29/2026 and CNBC 8/12, 8/28, 8/31, 9/4/2026). In the series I have, two repricings of roughly 20 percentage points ran in opposite directions, both on the Chair’s communication days. Second, the curve. Across the two communication days since July 29, the absolute 2s30s moves were 15 and 11 basis points; across the four major data days (the August 7 and September 4 payrolls, the August 12 CPI, the August 26 PCE), the absolute 2s30s moves were 3, 4, 2, and 1, with absolute 2-year moves of 6, 3, 2, and 2; on the Treasury buyback announcement day the absolute 2s30s move was 9 (source: U.S. Treasury constant-maturity yield series, each day’s difference from the prior session).
Supporting indicator, not an identification test: take the Chair’s future communication days, both FOMC press-conference days and non-FOMC public remarks, tagging each; compute the “communication-day absolute 2-year move” and the “communication-day absolute 2s30s move,” each divided by the median of the corresponding moves on CPI, PCE, and payroll release days over the same period. The scoring rule has three bands: if both ratios drop below 1.0 and hold there for two consecutive instances, call two is scored as failed (the guidance effect no longer dominates communication-day pricing); if they stay above 1.5, it’s scored as holding and stays under observation; between 1.0 and 1.5 is a gray band — no score, keep accumulating sample. Separately, I track the ratio of the sum of communication-day absolute 2-year moves to the sum on data days over the window, to net out the “fewer remarks, heavier each” confound.
Identification limit: a speech with no new content naturally draws a small reaction, which is a different thing from “the framework is understood”; conversely, as long as the Chair keeps re-weighting already-published data in his speeches, his remarks will move the market even if the regime succeeds. So the ratios above can’t identify, on their own, whether the regime has succeeded. A success definition closer to the Chair’s own is this: the surprise magnitude on data-release days increasingly explains market reactions, while his speeches mainly confirm, correct, or re-weight known data. That definition needs more sample to quantify.
Sample rule: September 5 to 17 is the pre-meeting blackout period, so the earliest non-FOMC sample is September 18; the September 30 PCE and the BEA annual revision land on the same day, and that data day is tagged separately. If there are fewer than three non-FOMC public remarks before year-end, I tag “insufficient sample” and roll over to Q1 2027, treating it as neither validated nor failed.
Call three (transmission, hypothesis). Inference error accumulates along the duration and puts upward pressure on the term premium; credibility, inflation risk, and growth expectations can act in the opposite direction at the same time.
Status: hypothesis, not scored this round. The plan was to baseline against the New York Fed’s ACM 10-year term-premium estimate for the week of August 28, cross-checked against the Board’s Kim-Wright model (FRED THREEFYTP10). At publication I can’t confirm the week’s values or the confirmation dates for either baseline, so call three doesn’t enter this round’s scoring and stays a research hypothesis, pending a traceable baseline (with source, period, and confirmation date). On top of that, the Treasury buybacks act directly on the 10-to-30-year sector, so estimates from September 9 to November 4 are contaminated by the operation and can’t be called clean to begin with.
Mechanism-downgrade condition (effective once the baseline is confirmed): provided the low-guidance form holds unchanged and the ACM baseline above has a traceable source, if the ACM monthly average from November 5 to December 31 falls 25 basis points or more below that baseline, call three is operationally downgraded. Whether that amounts to causal refutation still requires ruling out the concurrent influence of inflation, supply, and growth. If there’s no clean enough observation window before year-end, or the baseline can never be confirmed, it rolls over to Q1 2027.
Call four (credibility, watch item). If the Chair keeps using a hawkish standard while the FOMC’s actual actions keep coming in below the force the market infers from that standard, the market may gradually estimate the “standard” and the “actual reaction function” separately, forming a credibility discount.
How I’ll watch it: compare the repricing of policy odds after the Chair’s major communication days with the consistency of the subsequent FOMC decisions, statements, and dot plots. The first observation point is September 16, with the market pricing hike odds somewhere between five and six in ten going in. If a sequence of “hawkish remarks, market marks up hike odds, FOMC action clearly weaker than that inference” shows up several times, I’ll add the gap between remarks and action as a new state variable. The sample is too small to set a numerical threshold for now.
Timestamps and Watch List
Threshold observation runs from the publication date. A snapshot of the key variables follows; all dates are Eastern time.
Variable: 2-year
Aug 28 close: 4.34%, +14 bp on the day
Sep 9 close: 4.43%
What to watch: Where guidance by inference landed in August
Variable: 10-year
Aug 28 close: 4.73%, +6 bp on the day
Sep 9 close: 4.83%, reported highest since Nov 2023
What to watch: Whether the belly starts to take it on
Variable: 30-year
Aug 28 close: 5.22%, +3 bp on the day; 5.31% on Aug 17
Sep 9 close: 5.28%
What to watch: Not treated as mechanism evidence during the buyback
Variable: 2s30s
Aug 28 close: 88 bp, −11 bp on the day
Sep 9 close: 85 bp
What to watch: Communication-day vs data-day move ratio
Variable: 30-year real (constant-maturity)
Aug 28 close: 2.96%; 3.06% on Aug 17
Sep 9 close: — (Sep 9 value not obtained; excluded from thresholds)
What to watch: Not to be mixed with the auction basis
Variable: 30-year breakeven
Aug 28 close: 2.26%; Aug range 2.21–2.28%
Sep 9 close: — (Sep 9 value not obtained; excluded from thresholds)
What to watch: A break above 2.50% is a credibility-discount warning
Variable: ACM / Kim-Wright 10-year term premium
Aug 28 close: Baseline not obtained; not scored this round
Sep 9 close: Not scored, Sep 9 to Nov 4
What to watch: Call three; folded into tracking once the baseline is confirmed
Variable: September hike implied odds
Aug 28 close: 36% → 56%
Sep 9 close: 58% (Sep 4)
What to watch: Primary evidence for call two; the answer key arrives September 16
Variable: Treasury buybacks
Aug 28 close: Doubling announced Aug 19
Sep 9 close: First operation Sep 9, $6 billion
What to watch: The third mirror
Variable: Event calendar
Aug 28 close: —
Sep 9 close: CPI Sep 11; FOMC Sep 15–16; blackout ends Sep 18; PCE and BEA annual revision Sep 30; buybacks end Nov 4
What to watch: Scoring points for calls one through four
(Sources: U.S. Treasury constant-maturity nominal and real yield series; CME FedWatch; U.S. Treasury announcements; CNBC daily reporting. Items marked “not obtained / excluded” are kept out of the scoring thresholds because there was no traceable source, period, or confirmation date at publication.)
Three Ways to Read This, and What Each Costs
This section does a research-level scenario classification, and that’s all it’s for; it doesn’t assign anyone to a position. On the open question of whether guidance by inference lasts, there are three internally consistent ways to read it, each with its own cost and its own precondition. The judgment call rests with the reader.
The first reads the status quo as a transition phase, assuming the market learns the framework within a few months and the anomalous communication-day volatility settles on its own. The cost: if the mechanism is in fact durable, the long end’s uncertainty compensation gets systematically underpriced, and the underpriced part falls exactly at the longest tenors. This posture only stands if you’re insensitive to daily volatility and observe in units of years.
The second reads the Chair’s communication days as a new event class, assuming communication days and data days now differ in the nature of their risk enough to warrant a separate tag in the analytical framework; the two opposite-direction curve moves in July and August are the best evidence this posture has so far. Its cost is higher monitoring overhead, and because the timing of non-FOMC remarks isn’t fully predictable, there’s limited room to prepare in advance. Its precondition is that you already have to understand rate volatility at daily resolution.
The third suspends judgment on net direction and tracks the relative strength of the two channels instead, assuming the credibility channel and the uncertainty channel keep offsetting each other, and that during the Treasury’s buybacks the long end simply isn’t a clean object to observe. Its cost is that it needs a second layer of observation tools, and the conclusion may stay unresolved for a long time, during which you have to tolerate not concluding. Its precondition is research resources and a judgment horizon measured in quarters.
Not choosing any of these is also a reading. Its cost just gets booked somewhere else.
Signed, September 10, 2026. The September 16 FOMC is the first observation point for calls one and four; after the blackout ends on September 18, call two starts accumulating sample; call three has no clean window until the buybacks end on November 4 and a traceable term-premium baseline is in hand.
Disclaimer
This article reflects my personal investment philosophy. It is not investment advice. Make your own informed decisions.
Miyama Capital manages proprietary capital only and does not solicit external investors.
Legal Disclaimer
This memo represents the author’s personal views on macroeconomic conditions, interest rate environments, and asset allocation as of the date of writing. It does not constitute a solicitation, recommendation, or guarantee regarding the purchase or sale of any security, fund, bond, or other financial instrument. Investing involves risk; bond prices, interest rates, foreign exchange rates, and economic/policy conditions may materially affect asset values. Scenarios and instruments discussed may become inapplicable as market conditions change. Readers who make investment decisions based on this memo do so at their own risk, and the author accepts no liability for any gains or losses arising from the use or citation of this material.
Additional note for this article: The monitoring thresholds in this memo are operational definitions used solely for the author’s own prediction record. They are not trading rules and do not constitute any basis for entry or exit, and they reflect only the publicly available information at the time of writing.
Kuan H. Wang Founder & CIO, Miyama Capital