GOODLUCK STRATEGY · FED AGENT
The Room  ·  Dossiers  ·  Meetings  ·  Editor's notes  ·  The Chair  ·  How it works
2026-09-08 · v1 · public
The Chair in depth · 54 dated documents · 168,085 words · four eras

He proposes.
Everyone else responds.

Kevin Warsh was sworn in as the 17th Chairman on 2026-05-22. His card carries the most weight in the simulation and the thinnest evidence for the job he actually holds. Fifteen years of argument, and ninety-eight days of consequence.
54dated documents, all his own words. No paraphrase, no commentary about him.
2FOMC meetings chaired at the time the persona was frozen. The whole behavioural sample.
0dissenting votes he has ever cast, in twelve years on the Board. He complies, publishes, and leaves.
0numeric thresholds anywhere in his stated reaction function. By design, and he says so.
01

Four eras,
one career

The corpus splits cleanly into four periods, and they are not interchangeable. The one that matters most is ninety-eight days long.

Everything here is his own text: 54 dated documents, 168,085 words, as at 2026-08-28. Nothing written about him is in the corpus - not analyst notes, not profiles, not market commentary. The reason is narrow: a persona distilled from what the market thinks he will do would teach the model the consensus, which is the exact inverse of why this machine exists. hard data

Four eras: what each one can settle, and what it cannot Weight follows relevance, not volume. The era with the fewest days in it is the one the simulation leans on hardest. I · Chairman 2026-05-22 to 2026-08-28 32,975 words · 12 documents Settles What he does with the gavel: two meetings chaired, one statement rewritten, one reaction function stated out loud, one prepared keynote. Does not settle How he behaves in a meeting he loses, in a recession, or at a turning point. Two meetings is not a sample of a chairmanship. I-b · Nominee 2026-04-21 27,481 words · 2 documents Settles What he committed to under oath, on the record, in one day: no forward guidance, the interest-rate tool is dominant, four meetings a year is too few. Does not settle What he does when a commitment made in a hearing collides with a decision he has to take. Hearings reward clean answers. II · Private years 2011-08 to 2025-07 51,449 words · 20 documents Settles What he believes with nothing at stake - the ideology, the vocabulary and the arguments he reaches for by instinct. Does not settle What survives contact with the office. He now runs the institution those essays were aimed at. III · Fed Governor 2006-07 to 2011-02 56,180 words · 20 documents Settles How he behaves as one vote among many, including the one time he disagreed with a decision that passed. Does not settle How he behaves as the person who proposes rather than the person who responds. The job is structurally different. One line from the corpus that carries more than its length All eighteen Board speeches plus the November 2010 op-ed and the February 2011 resignation letter - the record of a governor who voted for QE2 and then left. The heaviest weight sits on the thinnest evidence. That is the situation, not a bug we can engineer around.
Figure 28 · Four eras, sized by words - and what each one is actually evidence ofCorpus as at 2026-08-28; the limits column is ours, judgement
Fifteen years of argument tell you what he believes. Two meetings tell you what he does. The simulation has to lean on the second, and the second is almost empty.
02

The documents
that do the work

Twenty-five of the 54, with what each one establishes. These are the load-bearing ones; the rest corroborate.
DateEraDocumentWhat it establishes
2008-05-21governorThe Federal Funds Rate in Extraordinary TimesThe two priors that never change: avoid the error of false precision about the neutral rate, and treat being too easy for too long as the expensive mistake, because it forces a correction 'at great cost'.
2008-11-06governorThe Promise and Peril of the New Financial ArchitectureA central bank should resist playing a larger role than its statute or economics warrants, and should sometimes accept 'the perils of popularity' and decline to act.
2009-04-06governorThe Panic of 2008Policy should encourage differentiation among firms rather than treat them alike - the intellectual root of his later absolute anti-bailout position.
2010-03-26governorAn Ode to IndependenceAn early, unqualified defence of central bank independence, written before he began arguing that mandate creep is what destroys it.
2010-11-08governorRejecting the RequiemExpanding the balance sheet 'is not a free option': as it grows the Fed becomes 'more of a price maker than a price taker' in the Treasury market and obscures price signals about US indebtedness.
2010-11-08governorThe New Malaise and How to End It (op-ed)Published the same day as the speech above and days after he voted for QE2 - 'the Federal Reserve is not a repair shop for broken fiscal, trade or regulatory policies'. The single clearest instance of his words running ahead of his vote.
2011-02-10governorLetter of ResignationHow he resolves a disagreement he has lost: he did not dissent, he complied, published, and left. Load-bearing for predicting that he will not break a committee to register a view.
2011-12-06privateThe 'Financial Repression' TrapGovernment-induced prices send false signals to users and providers of capital; market-determined prices reveal inconvenient truths sooner.
2014-01-01privateRethinking Macro: Reassessing Micro-FoundationsA long-form statement of his methodological scepticism about the macro toolkit, four years before it becomes a reform programme.
2016-03-08privateChallenging the Groupthink of the GuildNo model should be determinative of policy, 'not even the dynamic stochastic general equilibrium models pushed so prominently by our guild'; institutions lose credibility by overpromising and under-delivering.
2016-05-03privateDecision-Making at the FedHis template for how the Committee should deliberate - the earliest draft of the process changes he made in June 2026.
2016-08-24privateThe Federal Reserve Needs New ThinkingA Jackson Hole-week op-ed calling for institutional reform of the Fed, exactly ten years before he delivered the Jackson Hole keynote as Chairman.
2019-05-03privateTying Down the Anchor: The Task Gets TougherWith ammunition low, credibility is 'at a premium' and will be relied on most when the economy slows - the asymmetry that makes him structurally hawkish even when he sounds calm.
2022-03-01privateReinvigorating Economic GovernanceHis longest single document: a governance blueprint arguing that agencies which stray beyond their statutory core forfeit the independence they claim.
2025-04-25privateCommanding Heights: Central Banks at a CrossroadsHe owns and regrets QE2 in his own words, and states the distinctive position that independence is instrumental rather than sacred - the objective is 'to make the central bank safe for democracy, not to make democracy safe for the central bank'.
2025-07-08privateInflation Is A Choice (full interview transcript)The title thesis in extended form: inflation is an accountability question for the central bank, not a weather event - which is why he will not accept supply-shock excuses as Chairman.
2026-04-21nomineeOpening Statement before the Senate Banking CommitteeThe compressed platform he took into office: 'Inflation is a choice, and the Fed must take responsibility.'
2026-04-21nomineeConfirmation Hearing before the Senate Banking CommitteeThree commitments made under oath and kept: he does not believe in forward guidance; the interest-rate tool is the dominant and fairer one, the balance sheet secondary; and the statutory floor of four meetings a year is 'not enough'.
2026-06-17chairFOMC Statement (12-0)The first statement of his chairmanship: forward guidance and the easing bias deleted, the text cut from 341 words to 131, the named 'Voting for' roster abolished. Unanimous.
2026-06-17chairTranscript of the FOMC Press ConferenceThe 2 percent target is fixed and not up for review until delivered; he submitted no dot of his own; five task forces announced; policy described as unevenly restrictive - tight in housing, not in financial markets.
2026-07-14chairSemiannual Monetary Policy Report to the CongressNo tolerance for persistently elevated inflation, and an explicit refusal to bank on AI: 'We don't know the extent to which the economy will benefit from the AI buildout.' Also the anti-bailout line, 'full stop'.
2026-07-29chairFOMC Statement (9-3)Three hawkish dissents for a 25bp hike - reported as the first time since September 2016 that three policymakers dissented in the same direction. Statement closes: 'The Committee will deliver price stability.'
2026-07-29chairTranscript of the FOMC Press ConferenceThe only explicit statement of his reaction function, and it contains no numbers. Also the two sentences the simulator later built a whole scenario on: markets 'have tightened financial conditions in this intermeeting period', and dissent is 'the design feature'.
2026-08-19chairMinutes of the July 28-29 FOMC meetingThe Committee is split on the same question he is: some participants judged financial conditions not sufficiently restrictive, various participants judged that they had tightened. Several favoured a 25bp increase; many judged tightening would be necessary if inflation did not decline.
2026-08-28chairIn Our Time (Jackson Hole keynote)His hundredth day as Chairman and his first extended prepared text in the job: routine forward guidance retired ('just don't call it forward guidance'), responsibility for '65 months of sustained, elevated inflation' assigned squarely to the central bank, and - against his own simulator's assumption - 'hard pressed to describe broad financial conditions as restrictive'.
The resignation letter of February 2011 is worth more than any three speeches. It is the only document showing how he settles a disagreement he has lost - and the answer is that he complies, publishes, and leaves.
03

Governor Warsh vs
Chairman Warsh

On four questions the record before the gavel and the record after it do not agree. We keep the crack open in the card rather than averaging it into a single consistent person.
Governor Warsh and Chairman Warsh, in full Four questions where the record before 22 May 2026 and the record after it do not line up. The card keeps the gap open rather than averaging it away. THEN NOW, AS CHAIRMAN 1 November 2010: he voted with the Committee for QE2, then published his reservations the same day - 'expanding the Fed's balance sheet is not a free option' - and resigned three months later. He never cast a dissenting vote. As Chairman he has chaired two meetings and sat with the majority at both, and he treats dissent as a feature rather than a failure: 'I asked for a good family fight, and I got one. That's the purpose, that's the design feature' (2026-07-29). The governor who would not dissent now runs a committee that does. 2 Out of office, and again as a nominee, he was widely reported to believe an AI productivity boom would permit lower rates; a senator read the claim back to him at the April 2026 hearing as something economists rejected. As Chairman he has held twice and refuses to bank on it. At the hearing itself: 'We can't bank on that.' At Sintra he declined to say whether AI would raise or lower inflation, and in July he told Congress 'We don't know the extent to which the economy will benefit from the AI buildout.' The optimism survived; the policy promise never appeared. 3 As governor his defining institutional argument was about the balance sheet - that an expanding Fed portfolio makes the central bank 'more of a price maker than a price taker' in the Treasury market. As Chairman he ranks the tools the other way round: the interest-rate tool is 'the dominant tool' and 'gets in the cracks - it's fairer', while the balance sheet disproportionately helps those who already own financial assets. Any balance-sheet change has been pre-committed to advance notice: 'we would preview it, explain it, debate it.' 4 From 2008 to 2016 he warned repeatedly against false precision and against models being determinative of policy, and mocked 'breathlessly awaiting trailing data from stale national accounts'. As Chairman he has acted on it: he submitted no dot to the June 2026 SEP, deleted forward guidance from the statement, and told the July press conference that 'the historic problem with data dependence is the data and the dependence'. The scepticism became procedure. This is the single hardest thing to predict about him: which of the two is speaking at any given meeting.
Figure 29 · The fracture between the essayist and the office-holder, kept open on purposechair.json, governor_vs_chair; every clause traces to a dated document
Why we do not smooth this out

A card that resolved the contradiction would predict beautifully and be wrong at exactly the moments that matter - the meetings where the office-holder and the essayist disagree. judgement
Keeping it open costs precision on every ordinary meeting and buys honesty on the rare ones.

04

What he has said
since taking office

Eight positions taken as Chairman, each with the date and his own words. Note how few numbers there are.
DateThe position, and his own words
2026-06-17The 2 percent PCE target is fixed and not up for review until it has been delivered.
I see no reason until we have reestablished our commitment and ability to deliver on the 2 percent inflation objective to revisit that.
2026-06-17Routine forward guidance is gone, and he submitted no projection of his own.
We've dropped forward guidance. ... I can't give any forward guidance about what we're going to do next.
2026-07-29His stated reaction function has no numbers in it: with employment near equilibrium, the direction of underlying inflation decides the direction of policy.
Any central banker, when he or she sees underlying inflation moving higher, he or she is more inclined to tighten policy. Again, when you've achieved the other side of your mandate, and you see underlying inflation falling, he's more inclined to loosen policy. That's my reaction function.
2026-07-29He read the intermeeting rise in market rates as evidence that transmission works - a claim about the Fed's capability, not a reason to wait.
If you look broadly at market prices, they are certainly not saying all clear, but they are working in concert to keep us on our toes, and they have tightened financial conditions in this intermeeting period, and that has provided us some comfort that we've got the ability and capability to deliver.
2026-07-29Dissent is designed in, not suppressed - and he refuses to call the hold a pause.
I asked for a good family fight, and I got one. That's the purpose, that's the design feature.
2026-08-28He does not judge broad financial conditions to be restrictive, and explicitly puts the sectors that are strained outside that judgment. This is a level judgment he has held consistently since June.
Certain sectors-like housing and agriculture-are showing strains. But, on balance, I would be hard pressed to describe broad financial conditions as restrictive.
2026-08-28Responsibility for the overshoot belongs to the central bank, and he counts it in months rather than describing it in adjectives.
There is one signal nobody can miss: The responsibility for 65 months of sustained, elevated inflation sits squarely with the central bank. And that is where it belongs.
2026-04-21On meeting frequency he is on record - before taking office - that the statutory floor of four meetings a year is too few. No target number has ever come from him, and the widely circulated '8 to 4' framing is contradicted by his own words.
I believe the statute requires a minimum of four meetings, but four is not enough. So having more meetings than that is appropriate.
Three dated readings on financial conditions - and the level never moved Two different statements get made about the same thing. One is about the level of conditions. One is about the change since the last meeting. They are not interchangeable. LEVEL 17 June 'It's uneven.' Housing tight, financial markets not: 'I would have a hard time managing to say those words if I were to see what's happening in financial markets.' 28 August 'Certain sectors - like housing and agriculture - are showing strains. But, on balance, I would be hard pressed to describe broad financial conditions as restrictive.' Same sentence shape, ten weeks apart: some sectors strained, financial markets not, net not restrictive. No change of position. CHANGE 29 July, once 'They have tightened financial conditions in this intermeeting period, and that has provided us some comfort that we've got the ability and capability to deliver.' A statement about one six-week window, offered as evidence that transmission works. He has never once said conditions are tight. A model that has him waiting for the market to do his tightening for him has read the change line as if it were the level line.
Figure 30 · One level statement repeated, one change statement made once - and the difference matters in section 08Press conferences of 17 June and 29 July 2026; Jackson Hole keynote of 28 August 2026
The distinction in Figure 30 is the one to keep hold of. Two different statements get made about financial conditions - one about the level, one about the change since the last meeting - and they are not interchangeable. He has made the level statement twice, ten weeks apart, in the same shape both times. He made the change statement once. Reading the second as if it were the first is precisely the error our own model made. hard data
05

Checked against
a text he had not written yet

On 28 August the card met a document that did not exist when it was built. It matched on five things and failed on the one the whole prediction rested on.
Jackson Hole, 28 August: the scorecard in one view The persona was distilled on 1 August. The keynote did not exist until 28 August. Five framings matched; one mechanism did not. 5 HITS Counted to the same month of overshoot - 65 - on its own arithmetic Employment at equilibrium: model 4.2%, he said 4.1%. Same judgement Direction of underlying inflation carries the decision, not the level of the print No numeric threshold, no date, no sequence language - and he gave none Responsibility for the overshoot sits with the central bank 1 MISS - AND IT IS THE ONE THE HOLD RESTED ON Model: the market has already delivered the tightening, so I can wait Him: hard pressed to describe broad financial conditions as restrictive Then it got worse, and better Three days later the cause was overturned. The sentence was his own, from 29 July. The model had taken a hawkish argument - the market proves our transmission works - and used it as a dovish one, and had promoted a six-week change into a standing judgement about the level. Leans more hawkish, not less. One speech is not one vote, and two checks are not a score. The first score is 16 September.
Figure 31 · The compact version of the check - the full account is in section 08 of the primerchecks.json, dated 2026-08-28 and re-diagnosed 2026-08-31
What the Chairman's record does not establish - handed to the model as a gag order This list is read back to the simulated Chairman with an instruction: you may not claim any of it. Without it the model will invent a threshold, and it will sound exactly like the parts that are true. 1 No numeric threshold or trigger of any kind - no inflation rate, unemployment rate, wage or expectations level at which he would hike or cut, and no terminal rate. He has explicitly refused to give one and submitted no dot. 2 No forecast of his own for growth, inflation, unemployment or the funds rate; he does not participate in the SEP. 3 No stated view on how much further tightening, if any, he thinks is needed, or on the timing of any move. 4 The transmission-channel question put to him on 2026-07-29 is unanswered: he denies relying on labour-market cooling and did not name a substitute mechanism. 5 No target size, composition or pace for the balance sheet - only 'smaller' and 'shorter', and an explicit refusal to fix a number. 6 The task-force outputs do not exist yet, so his positions on the communications framework, the future of the SEP, the inflation framework and the data regime are aspirations, not decisions. 7 No documented position as Chairman on immigration, tariffs or fiscal specifics, or on Governor Cook's tenure. 8 A live conflict the documents make explicit rather than resolve: out of office and as a nominee he argued an AI productivity boom and a smaller balance sheet could permit lower rates; as Chairman he has held twice, refused to bank on AI disinflation, and is being pushed by dissenters to hike. The tension is the finding - do not resolve it by inventing a position. Every one of these is a question a language model would answer fluently and wrongly if the card did not forbid it.
Figure 32 · The band of the persona card that does the most workmembers.json, Warsh entry; hard data - each item is an absence in the corpus
The card was right that he never gives a numeric threshold - on the same day, in the same document, that it was wrong about why he holds. Being right about the shape of someone's reasoning is not the same as being right about their conclusion.

Sources and limits

Corpus · 54 dated documents, 168,085 words, as at 2026-08-28: Board speeches, newspaper op-eds, think-tank essays, conference remarks, the confirmation hearing transcript, two FOMC press-conference transcripts, semiannual testimony, the July 2026 minutes, the Jackson Hole keynote, and the February 2011 resignation letter.
Not in the corpus · anything written about him rather than by him. Analyst notes, profiles and market commentary are used for scoring and comparison and are cited where they appear, but they are never distilled into the card.
The frozen persona · built 1 August 2026 from 53 documents. The 54th, the Jackson Hole keynote, is the text it was checked against. The card has not been rebuilt since, because rebuilding it while a prediction is unresolved would mean rebuilding the exam paper from the answers.
How to read the confidence tags · hard data already happened, public source; anchored extrapolation derived from real data by a rule we state; judgement our opinion, the most likely of the three to be wrong.
Limits · two chaired meetings is the entire behavioural sample for the job he holds. Everything about how he behaves under stress, in a downturn, or at a turning point is extrapolation from a different role. Back to the primer · the other eighteen · the Room.
Fed Agent · goodluckstrategy.com · data as of 8 September 2026 · deepseek-v4-pro · one full committee run ≈ ¥1.5, calibrated · Never scored against a real meeting — first scoring point 16 September 2026 · research tool, not investment advice