The Global Review - Farid Novin

 

Oracle, Prisoner, Navigator

A Bayesian Examination of the Federal Reserve, Forward Guidance, and Credibility under Radical Uncertainty

Farid Novin

Revised October 17, 2026

First version September 28, 2026

Abstract

Commentary on the Federal Reserve often reaches for two images from Greek myth: the Delphic oracle, to mock the confident forecast, and Odysseus bound to the mast, to explain why a central bank might wish to tie its own hands. This essay argues that both images mislead, and for different reasons. The oracle metaphor confuses a fallible, evidence-based, conditional forecast with a claim to transcendent knowledge. The Odysseus metaphor treats a narrow problem of time-inconsistent commitment as though it described all central-bank communication. Drawing on the Federal Reserve’s own documentation, on the academic distinction between Delphic and Odyssean forward guidance, and on episodes from 2008 to 2026, I propose that credibility attaches to the process by which a central bank updates rather than to any particular forecast, and that the apt figure is the navigator. I add a qualification the metaphor invites but does not supply: updating that is late, unexplained, or erratic destroys the very credibility it is meant to protect.

This revised version adds a third and live test. The September consumer price report of October 14 reopened the question whether the Federal Open Market Committee will raise rates again on October 28, six days before the midterm elections. I argue that the choice between hiking and holding matters less for credibility than whether the Committee makes that choice distinguishable from electoral deference. A hold is an action that a mandate-driven committee and a politically cautious one would both take, so by itself it conveys little; only an accompanying account of the reaction function can tell the two apart. A Committee that has abandoned point guidance has fewer instruments with which to supply that account, and the argument of this essay implies that it must therefore use the remaining ones with unusual discipline. The inflation episode of 2021–22, the retreat from forward guidance, and the October decision are treated as the three tests of the thesis.

I. The Category Error: The Fed Is Not an Oracle

A word about vocabulary comes first, because the two myths are already technical terms. Campbell, Evans, Fisher and Justiniano (2012) call forward guidance “Delphic” when it merely forecasts macroeconomic performance and likely policy, and “Odyssean” when it publicly commits the Federal Open Market Committee (FOMC) to a future action, as Odysseus committed himself to his ship by having himself bound to the mast. They are careful to say that they do not use “Delphic” to evoke the oracle’s famously ambiguous utterances. This essay borrows the same two figures but asks a wider question: what does each myth imply about the epistemology and the commitment structure of a modern central bank?

Begin with the oracle. The Delphic tradition claimed access to knowledge that was, by definition, unavailable to ordinary reasoning. Its authority rested not on a reproducible forecasting procedure but on the supposed special status of the institution. To accept the prophecy was to accept the priestess.

A modern central bank operates under an entirely different epistemology. The Federal Reserve builds its projections from observed data, statistical relationships, economic models, financial conditions, surveys and informed judgment. Its projections are fallible hypotheses about the future, not revelations about it, and the institution says so plainly. The projection materials released after the FOMC meeting of September 15–16, 2026 describe each participant’s projections as the most likely outcome given information available at the time of the meeting. The same materials acknowledge that the models used are necessarily imperfect descriptions of the real world, that the path of the economy can be altered by unforeseen developments, and that the confidence intervals built from historical forecast errors may not capture participants’ current assessment of uncertainty and risk (Federal Reserve Board 2026a). The published fan charts are themselves an admission of fallibility: they are constructed from the root-mean-squared errors of forecasters over the previous twenty years, precisely so that readers can see how wrong such forecasts have tended to be (Reifschneider and Tulip 2017).

The distinction matters. A forecast can be wrong without the forecasting institution having behaved irrationally. If a central bank projects inflation of two percent and an unforeseen geopolitical shock delivers five, the projection does not retrospectively become a false prophecy. It was a conditional estimate made with the information then available.

This is why a Bayesian rear-view-mirror image is far closer to the economics of forecasting than any Delphic one. Economic forecasting necessarily begins from information about the past and the present: inflation, employment, wages, productivity, financial conditions, expectations. The forecaster infers from these the probable evolution of the system, ceteris paribus. The difficulty is that the vehicle is moving forward, and the forecaster therefore faces a basic asymmetry. The past is observable; the regime governing the future is not. That asymmetry is the essence of radical uncertainty.

II. The Rear-View Mirror Does Not Mean the Fed Is Blind

The Fed does not, of course, drive by the rear-view mirror alone. Modern forecasting draws on leading indicators, market prices, surveys, structural models, alternative scenarios and probability distributions. Yet even the mirror is imperfect. Orphanides (2001) showed that policymakers act on real-time data that are later revised, so that the picture of the recent past is itself provisional. The road behind is blurred before one even begins to extrapolate it.

It is worth being clear about what kind of uncertainty is at issue. Knight (1921) separated measurable risk from uncertainty that cannot be reduced to a known probability distribution, and Kay and King (2020) have argued forcefully that many of the decisions that matter most are taken in that second world, where the honest question is less “what is the probability?” than “what is going on here?”. A Bayesian framework respects this point rather than evading it. Its priors are disciplined judgments, not pretended frequencies, and its purpose is to state how beliefs should move when evidence arrives, not to claim that the evidence will ever be complete.

A more precise formulation than “driving by the mirror” is therefore available. The central bank drives forward using the rear-view mirror, the windshield, the dashboard instruments and a probabilistic map, but none of these reveals the road beyond the next bend.

The contrast with Delphi can be put in two sentences. The oracle says:

“This is what will happen.”

The Bayesian forecaster says:

“Given what we know now, this is how the probabilities have changed.”

That difference, between announcing a future and revising a belief, is the hinge of the whole argument.

III. Odysseus Answers a Different Question

The Odysseus image concerns not epistemology but commitment. In Book XII of the Odyssey the hero knows that he will be unable to resist the Sirens’ song, and so instructs his crew to bind him to the mast and to ignore any order to release him. The arrangement is deliberately designed to prevent his future self from overriding his present intention. Economists more often use the Latin name, following Elster’s (1979) study of precommitment in Ulysses and the Sirens, and I use the two names interchangeably.

In monetary economics the analogy addresses the problem of time inconsistency identified by Kydland and Prescott (1977). A policy plan that is optimal when announced may cease to be optimal once agents have acted on it, so a policymaker who re-optimizes at every date can end up in a worse equilibrium than one who could bind herself in advance. This point requires a correction to the way the analogy is often stated. The danger is not that tomorrow’s policymakers will behave irrationally. It is that they will behave rationally, and that what is rational for them tomorrow differs from what was best to promise today. Odysseus fears not a fool but a captive of the song.

Seen this way, the classification of Campbell et al. (2012) is exact. Odyssean guidance changes private expectations by committing the FOMC to a future deviation from the policy rule it would otherwise follow; circumstances will tempt the Committee to renege precisely because the rule describes its preferred behavior. The canonical case is the argument of Eggertsson and Woodford (2003) that, at the zero lower bound, optimal policy keeps rates low for longer than the usual rule would require once the shock has passed. Everything else is Delphic in the technical sense: it forecasts.

It follows that most of what the Fed says about the future is not a mast at all. Suppose the FOMC states that, given current conditions, it expects the federal funds rate to remain at a stated level for the foreseeable future. Inflation then accelerates, or an energy shock arrives, or financial conditions change sharply. If the Fed subsequently changes course, that is not a failure of commitment. It may be state-contingent policymaking working as intended. The Fed itself describes the appropriate future policy rate as highly uncertain because it depends on how real activity and inflation evolve (Federal Reserve Board 2026a). Even the Fed’s most explicit efforts to bind itself have been conditional in form. The December 2012 statement, which replaced calendar dates with numerical thresholds, kept the exceptionally low range in place “at least as long as” unemployment stayed above 6½ percent and inflation and inflation expectations behaved, while reserving the right to weigh other information (Federal Reserve Board 2012).

A qualification is owed here, because the literature is not one-sided. Campbell, Fisher, Justiniano and Melosi (2017) find that in the first years after the financial crisis a purely rule-based policy would have done better than the guidance actually given, but that from late 2011, after the introduction of calendar-based communication, Odyssean guidance appears to have raised real activity and moved inflation closer to target. Limited commitment can therefore add value when conventional policy is constrained. My claim is accordingly not that the Fed should never be bound. It is that it should not, in ordinary circumstances, be bound to the mast, and that when it does take a commitment it should be a conditional one that it can loosen in public, with reasons, when the evidence changes. The proper anchor is the mandate and the reaction function, not a forecast.

IV. Credibility Does Not Require Inflexibility

This brings us to what I take to be the strongest part of a Bayesian argument. A Bayesian holds that markets expect the Fed to keep its promises in normal circumstances but understand that extraordinary circumstances can dissolve the premises on which a promise was made. The proposition can be sharpened. Market credibility does not require a central bank to preserve an obsolete forecast. It requires the bank to behave consistently with its stated objectives and to explain why changing information warrants a change in policy.

Consider two hypothetical central banks. Central Bank A announces a policy path and refuses to alter it after a major shock, believing that a change of course would damage its reputation. Central Bank B announces a path conditional on the information then available, receives radically different information, revises its forecast, and changes policy while explaining the revision. Under a naive conception of commitment, A appears the more credible. Under a rational-expectations or Bayesian conception, that conclusion is far from obvious. If economic agents understand that policy responds to incoming information, they should expect Bank B to change course when the state of the economy changes, and should be alarmed by Bank A’s refusal to do so. The credibility of an institution lies not in “we will never change our minds” but in “we will change our policy when the evidence changes, according to a reaction function you can understand.”

The same logic cuts in a direction that the two hypothetical banks do not show. Bank B’s credibility depends on its audience believing that the evidence, and nothing else, moved the policy. If the audience suspects that something other than the evidence is moving policy, including the proximity of an election, then the explanation offered loses its force, because a reaction function that bends to unstated considerations cannot be learned. Section XIII returns to this point, which is the crux of the present decision.

V. Four Episodes

2008–2014: guidance that evolved

The global financial crisis offers the clearest illustration. In December 2008, with the target for the federal funds rate cut to a range of zero to one-quarter percent, the FOMC stated that weak economic conditions were likely to warrant exceptionally low levels of the funds rate “for some time” (Federal Reserve Board 2008). Forward guidance became an instrument in its own right because the conventional instrument had reached its floor.

The important point is what happened next. The language did not stand as a fixed promise. It was recast in 2011 in calendar terms, then in December 2012 in terms of economic thresholds that the Committee described as consistent with its earlier date-based guidance (Federal Reserve Board 2012), and later still, as unemployment approached the numerical threshold, into a broader qualitative assessment of progress toward the mandate. At each stage the wording was adapted to the state of the economy and to what the Committee had learned about how the public read it. This is exactly where the Odysseus analogy strains. Odysseus prevents his future self from changing the decision. The Federal Reserve has always retained the ability to respond to information. Its commitment was conditional, not absolute.

2019: patience

The 2019 shift is instructive because it was small in wording and large in meaning. Through late 2018 the FOMC had signalled that some further gradual increases in the target range would be appropriate. In January 2019 that language was dropped. The Committee said instead that, in light of global economic and financial developments and muted inflation pressures, it would be patient in deciding what future adjustments might be appropriate (Federal Reserve Board 2019). Nothing in the mandate had changed. The Committee had received information, revised its assessment of the balance of risks, and altered its signal accordingly. This is Bayesian updating conducted in public.

2020: the prior is overturned

The pandemic was the extraordinary shock. In December 2019 the median FOMC projection for 2020 real GDP growth was 2.0 percent and for the year-end unemployment rate 3.5 percent. By the June 2020 projections the medians were a contraction of 6.5 percent and an unemployment rate of 9.3 percent, and individual participants’ projections for 2020 growth ranged from a fall of 10.0 percent to a fall of 4.2 percent (Federal Reserve Board 2020). Almost every participant also expected the policy rate to stay unchanged through 2022 (Money and Banking 2020). The earlier projection had not been an oracle’s lie, and it would have been absurd for the Committee to defend it. The underlying stochastic process had changed, and the only rational response was to update, at once and visibly.

The episode also shows why forecast credibility and forecast accuracy must not be confused. No central bank could credibly have promised that its December 2019 projections would survive the spring of 2020. What it could credibly promise, and did, was to revise them openly and to explain the revision.

2021–2022: when updating comes late

The fourth episode is the hard one, and an honest version of the argument must confront it. Through 2021 the Committee’s projections and public messaging treated the surge in inflation as largely transitory. On November 30, 2021, with consumer prices up 6.2 percent over the preceding year, the highest reading since 1990 (Scripps News 2021), Chair Powell told the Senate Banking Committee that it was probably time to retire the word, that inflation would persist at least through the middle of 2022, and that policy had to address the range of plausible outcomes rather than only the most likely one (Fox Business 2021; Scripps News 2021). The first increase in the target range followed in March 2022.

It is reasonable to read this episode in two ways, and the argument should not pretend that only one is available. On the first reading, the Fed behaved as a good Bayesian should: it began with a prior that supply disruptions would fade, observed accumulating contrary evidence, revised, and then acted. On the second, the updating was too slow, the prior too stubborn, and the public was left to conclude that the reaction function was less responsive than advertised. Both readings are compatible with the thesis of this essay, but the second imposes a discipline on it. Credibility attaches to the updating mechanism, and a mechanism that updates with a long lag, or that re-labels its own earlier language without a clear account of what changed, damages the asset it is meant to protect. Speed and transparency of revision are part of the process, not extras.

The memory of that lag now frames every decision the Committee takes. An energy-driven rise in headline inflation invites precisely the “look-through” reasoning that proved costly in 2021, and any committee that has absorbed the lesson will ask not only whether the shock is transitory but whether it is spreading into the prices of everything else. That question is at the center of the October debate.

VI. Commitment to an Outcome versus Commitment to a Process

These cases point to a distinction between two kinds of credibility. The first is outcome commitment: “We promise that X will happen.” It is vulnerable to any shock that makes X undesirable or impossible. The second is process commitment: “We promise to evaluate incoming information according to a known framework and to adjust policy consistently with our objectives.” This form is far more compatible with radical uncertainty, and it is arguably the more appropriate form for a modern central bank.

The Fed cannot control every future economic outcome. It can, however, influence expectations by showing that its decisions come from a recognizable procedure. In Bayesian terms, credibility attaches to the updating mechanism, not to the unconditional prediction.

VII. Why a Market Can Understand a Broken Forecast

This resolves an apparent paradox. Suppose the Fed projects two percent inflation and the outcome is four. The forecast was wrong. But suppose the Fed promptly acknowledges the new information, revises its projection and adjusts policy accordingly. Has it necessarily lost credibility? Not necessarily. Markets may reasonably read the revision as evidence that the central bank responds to information instead of clinging to obsolete projections.

The more dangerous situation is the opposite one. The Fed discovers that its forecast has become implausible but declines to revise it, fearing that admitting uncertainty would cost it credibility. That can convert a forecasting error into a credibility error, and the second is much the more expensive of the two. The qualification from 2021–22 applies here as well: the sooner and the more clearly the revision comes, the cheaper the error.

VIII. Forward Guidance as a Bayesian Signal

Forward guidance is not merely a prediction. It is also a signal about the central bank’s reaction function. The Fed explains that forward guidance matters because households and firms build expectations about the future course of policy into current decisions on spending and investment. Two components can therefore be distinguished. The forecast component answers the question “What do we currently expect?” The strategic component answers “How should you interpret our future response to economic developments?”

The second component is the more durable. “We expect inflation to decline” is a forecast. “If inflation remains above target, we will take the steps needed to restore price stability” is a statement about the reaction function. The second can remain credible even after the first has been proved wrong.

There is a further Bayesian subtlety, and the empirical literature supports it. The public does not read Fed announcements only for what they say about policy. It also reads them for what they reveal about what the Fed knows. Campbell et al. (2017) find that puzzling responses of private-sector forecasts to movements in federal funds futures on announcement days are attributable almost entirely to Delphic guidance: forecasters treat a statement about the future rate path as news about the economy itself. Andrade and Ferroni (2021) draw the same distinction for the euro area. A central bank that signals a lower path may therefore be heard as reporting weaker prospects. Guidance thus carries two messages at once, one about policy and one about the state of the world, and the audience updates on both.

This gives a sound theoretical basis for resisting an over-literal reading of the Odysseus analogy. What the audience most needs from the central bank is not a promise about a number but a legible account of how information will move policy.

It is worth noticing that the present Chair has, on occasion, supplied exactly that kind of account while declining to supply a path. In his Jackson Hole address of August 28, Chair Warsh said that the Committee’s task was to see underlying inflation moving to target clearly and at sufficient speed, and that otherwise there was work to do; he also conceded that the summer’s better inflation readings did not persuade him that underlying trends had meaningfully improved (Morningstar 2026). Reuters reported that he went further and pointed to a specific diagnostic, the share of items in the personal consumption expenditures price index that are rising by more than three percent a year (Schneider 2026). Whatever one thinks of that particular gauge, naming a variable of this kind is a statement about the reaction function and not a forecast. It is the sort of disclosure that process credibility requires, and it can be checked against subsequent decisions.

IX. The Lucas Critique and the Reflexive Forecast

One more dimension distinguishes monetary forecasting from forecasting a physical system. Lucas (1976) showed that relationships estimated on historical data can break down when the policy regime changes, because agents adapt their behavior to the regime they expect. The Fed is therefore not forecasting a fixed mechanism. The system is strategic. Markets observe the Fed, the Fed observes markets, households and firms alter their behavior in response to expected policy, and the Fed then observes that altered behavior and revises its own assessment.

Its communication is consequently several things at once: a forecast, a policy signal, an instrument for managing expectations, a description of the reaction function, and an input into the very system being forecast. This is why the Delphi analogy is insufficient and the Odysseus analogy incomplete. A prophecy stands outside history; a central-bank forecast is a move within it.

X. A Better Metaphor: The Navigator

I propose replacing the two metaphors with a third. The Fed is neither Delphi’s oracle nor Odysseus permanently lashed to the mast. It is better understood as a navigator working under incomplete information. The navigator observes the sea, the instruments, the weather, the currents and previous experience. She sets a course and communicates it to the crew. But if the weather turns, holding the announced heading merely because it was announced would demonstrate not competence but vanity. The credible navigator says: “This was our course given what we knew when we set it. Conditions have changed. Here is the evidence. Here is the revised course, and here is why it serves the same destination.”

The metaphor has more to offer than a flattering contrast. Brainard (1967) showed that when the effects of policy are themselves uncertain, the optimal response is often more cautious than certainty would suggest: the navigator moves the rudder by degrees. Greenspan (2004) described modern policymaking as a form of risk management, in which the central bank weighs the distribution of possible outcomes and the costs attached to each and not only the most likely path. The Fed’s current projection materials say almost the same thing: in setting policy, participants consider not only what appears most likely but also the range of alternatives, their likelihood and their potential costs (Federal Reserve Board 2026a). A navigator, in short, steers by the chart of dangers as much as by the destination.

Brainard’s lesson has a direct bearing on October. A navigator who has just turned the wheel does not turn it again at once; she waits to see how the ship answers. That is a respectable argument for patience after a first increase, and I weigh it fairly below. But the same navigator, if she holds her heading for reasons she will not state while the crew suspects that the real reason is a coastline she would rather not approach, loses the crew’s trust in her charts.

XI. The First Live Test: A Committee That Has Dropped Forward Guidance

The argument now meets a live case. At the June 2026 meeting, his first as Chair, Kevin Warsh did not submit a rate projection of his own, said the Committee had dropped forward guidance, and announced a review of Federal Reserve communications, including press conferences, economic projections, transcripts and minutes, to be completed by the end of the year (The Hill 2026). The Committee’s June projections were accordingly compiled from participants other than the Chair (Federal Reserve Board 2026b). Reactions among economists were mixed, with some welcoming the change as overdue and others reading it as a retreat from transparency (Yahoo Finance 2026). In September the Chair again declined to submit a projection (J.P. Morgan Asset Management 2026; Bloomberg 2026).

The September projections came in a period of revision. The median participant now expects the policy rate to end 2026 at 4.1 percent, holding there through 2027; median core PCE inflation for 2026 stands at 3.4 percent, with a return to two percent only in 2029 (FRED Blog 2026; J.P. Morgan Asset Management 2026). The Committee raised its target by a quarter point at the September meeting (Bondsavvy 2026), to a range that previews and later market commentary placed at 3.75 to 4.00 percent (Continuum Economics 2026; 247Wall St 2026), and one market commentary describes the projected path as having moved since March from about half a point of cuts to about half a point of hikes. Whatever one makes of those particulars, the direction of revision is precisely what a Bayesian account would lead one to expect when growth is stronger and inflation higher than previously projected.

How does the framework of this essay read the retreat from guidance? Two answers pull in opposite directions. In favor of the change, Delphic guidance is a forecast published in advance, and a forecast published in advance can become an embarrassment or an unintended commitment when the world moves. In a period of unusually wide uncertainty, refusing to publish a point path lowers the risk that the Committee is held to a stale one. Against it, the strategic component of guidance, the account of how the Committee will respond to data, is exactly what makes a bank legible when its forecasts fail. Withdrawing the forecast component does not remove the need for the strategic one. It raises the burden on statements, press conferences and the consistency of decisions to communicate the reaction function, because the market will still attempt to infer it, and it will infer it from less. One market commentary suggests the remaining dots are better read as a measure of the range of views on the Committee than as a guide to the path of rates (Lord Abbett 2026).

I do not treat this as a settled matter, and the thesis makes a testable claim about it. If the Committee’s reaction function proves legible, so that persistent inflation above target reliably leads to firmer policy and softer data to easier policy, and if decisions are explained in those terms, then process credibility can substitute for point guidance at little cost. If it does not, the likely symptom is greater sensitivity of market pricing to each data release. The October meeting, which falls before the year-end communications review, is the first occasion on which that claim will be tested under pressure.

XII. What the September Data Say, and What They Do Not

The September consumer price report, released on October 14, supplies the evidence on which the October decision will be made. Headline consumer prices rose 0.6 percent on the month, taking the twelve-month rate to 3.7 percent (an unrounded 3.74 percent) from 3.4 percent in August, a little above the consensus forecast of 3.6 percent (investingLive 2026; First Trust Advisors 2026). Energy prices rose by about five percent in the month and accounted for more than half of the headline increase (First Trust Advisors 2026). Core prices, which exclude food and energy, rose 0.33 percent on the month and 2.56 percent over twelve months, against 2.45 percent in August and a forecast near 2.5 percent (investingLive 2026). Both core goods and core services contributed, with goods rising by roughly a quarter to a third of a percent, which reflects tariffs and higher transport costs, and services by a little over a third of a percent. Commentary collected by Reuters noted that core prices have risen at an annualized rate of about 3.4 percent over the past three months, and that the increases were broad, touching automobiles, air fares, education and recreation (Virginia Business 2026).

Three features of the report deserve to be kept apart, because a Bayesian reading assigns them different evidential weight.

The first is the composition of the surprise. Much of the headline rise is energy, and energy is the channel through which the Persian Gulf conflict reaches American prices. A supply shock of this kind lowers real incomes and, if the central bank tightens in response, lowers output further; the standard counsel is to look through it provided inflation expectations remain anchored. The evidence on wages favors that counsel. Real hourly earnings have been slightly negative this year, and the September report marked the sixth consecutive month in which twelve-month consumer price inflation exceeded annual wage growth (Virginia Business 2026). Nothing in the labor-cost data resembles a wage-price spiral.

The second is the breadth of the core. A core increase of a third of a percent in a month, repeated across goods and services, is more probable if underlying inflation is persistent than if it is the echo of a single shock, and the three-month annualized rate sharpens that point. The likelihood has moved toward the persistent hypothesis, but modestly. A single core print of this size is what each hypothesis would produce about one month in three or four, and the Chair himself has urged the Committee not to build on individual data points. The proper posterior shift is real and small.

The third is the measurement gap. Core consumer price inflation of 2.6 percent sits well below the core PCE reading of 3.0 percent for August and below the Committee’s own median projection of 3.4 percent for the end of 2026, because the two indices weight their components differently (Virginia Business 2026). The comparison is unusual, since core CPI normally runs above core PCE, and it implies that the Committee’s preferred measure is the hotter of the two. Any argument that the core CPI figure is benign therefore understates the problem that the Committee itself sees. A message that points to 2.6 percent while the Committee’s own projections say 3.4 would be a reaction-function error of the very kind this essay warns against.

The real economy complicates the picture further. Activity remains resilient, supported by the continuing boom in investment related to artificial intelligence. Retail sales for September are reported to have risen 0.8 percent, well above expectations, with August revised up to 1.3 percent, and payroll employment continues to grow at a steady pace. A strong economy lowers the cost of tightening and weakens the case that a hike would tip activity into contraction. It also means that the demand side of the economy is not obviously helping to bring inflation down. Finally, September was reported as the sixty-sixth consecutive month in which both headline and core inflation have exceeded the Federal Reserve’s two percent target (First Trust Advisors 2026). That fact bears on the anchoring assumption on which look-through reasoning depends. A committee asking markets to treat the energy shock as temporary is asking them to believe in an anchor that has been above its mooring for more than five years.

XIII. The October Test: Hold, Hike, and the Problem of Observational Equivalence

The Committee meets on October 27 and 28. Market pricing has swung widely: an October increase was priced at roughly one in five before the inflation report, and at somewhere between forty and fifty percent afterward, depending on the source and the hour, with about 36 basis points of tightening priced for the remainder of the year (investingLive 2026; 247Wall St 2026). Earlier in the autumn, remarks by Vice Chair Jefferson and by President Williams, read alongside the Chair’s own, had been taken to signal that there was no immediate rush to tighten, even if further increases remained possible. The September report has partly undone that reassurance, and the question is now what a rational Committee should do and what its audience will infer from whatever it does.

The case for patience

Three considerations support holding. The first is the Chair’s declared approach, which attends to underlying inflation across many measures and discourages fine-tuning in response to single readings. The second is Brainard’s caution: the Committee raised rates in September, monetary policy acts with a lag, and a navigator who has only just moved the rudder has some reason to watch before moving it again. The third is the composition of the shock, discussed above, in which energy dominates and wages are not responding. A Committee that holds in October and states plainly that it will tighten in December if the core continues to run hot, if the breadth of price increases widens, or if inflation expectations drift, would be behaving as the navigator should. Several forecasters appear to expect exactly that sequence, a hold in October and a further increase in December (247Wall St 2026).

The case against

Three considerations also support a hike. The first is consistency. The September increase followed a core reading of a third of a percent. The September report delivers a core reading of a third of a percent again, with a broader base, a hotter three-month trend and a hotter headline. A Committee that raised after the first print and holds after the second must explain what changed, and the answer cannot simply be that the first hike needs time to work, since that was known in September. The second is the Chair’s own marker. Having said at Jackson Hole that there would be work to do if underlying inflation did not move clearly toward target at sufficient speed, he has made a conditional statement that the data can now test, and Reuters observed at the time that the onus was on him to deliver (Schneider 2026). The third is the term structure: a failure to act when the data warrant it risks a rise in long-term yields and a drift in inflation expectations, the outcome the Committee’s critics have warned of since the summer.

The midterm overlay and the problem of observational equivalence

A further consideration is one that the Committee will not state and the market cannot ignore. The meeting falls six days before the midterm elections of November 3. It is argued, with some force, that an October hike on the eve of an election would be politically untenable, and some commentators expected the Committee to wait until December for that reason (247Wall St 2026). I would put the contrary argument as follows. If the Committee holds, and the market attributes the decision to electoral caution, the cost to credibility would be very large, and far larger than the cost of a hike that proves slightly premature. The point can be made precise in the language of Bayesian games.

Suppose the market entertains two hypotheses about the type of Committee it faces. The first is a mandate-driven committee, whose decisions follow its reaction function whatever the electoral calendar. The second is a politically cautious committee, which adjusts its decisions at the margin to avoid acting near an election or against the preferences of the executive. The market does not observe the type and holds a prior about it. That prior is not neutral. Presidential pressure on the Federal Reserve is well documented, from the Nixon tapes and Arthur Burns (Abrams 2006) to the present administration’s public demands for lower rates, and the Chair’s first months were marked by criticism that he was all talk and no action (Rabobank 2026). The prior therefore assigns the second hypothesis more weight than a committee confident of its reputation would wish.

Now consider the actions. A hike is costly for the politically cautious type and, if the data warrant, natural for the mandate-driven type. It is therefore a costly signal in the sense of Spence (1973): only a committee that is willing to bear the political cost will send it, and the market learns something from the fact that it was sent. A hold, by contrast, is an action that both types would take. A mandate-driven committee holds because the data on balance support patience, and a politically cautious committee holds because of the election. The two are observationally equivalent. A hold is a pooling action, and a pooling action conveys nothing about type. The market’s posterior after a bare hold will be close to its prior, which is to say that the doubts will survive intact, and, since the prior leans toward the political reading, the doubts will be strong.

This is where the retreat from forward guidance bites. A committee that publishes a path can separate itself from the cautious type by showing that the path conforms to its stated rule. A committee that publishes no path must separate itself by other means: by the explanation in the statement, by the Chair’s press conference, by the minutes, and by the content of the conditional commitments it makes about December. In the terms of this essay, the hold is not the credibility event. The accompanying account of the reaction function is. If that account is specific, naming the indicators, the thresholds and the pace at which further tightening would follow, the hold becomes distinguishable from deference, and a modest cost in time buys the Committee its credibility. If the account is vague, the hold will be read as political, the Committee will have paid the credibility cost of a political decision without the benefit of the decision’s being right, and the market will conclude that the reaction function bends in the weeks before an election. That inference would persist beyond November 3. It would color every later meeting at which the data and the political calendar diverge.

The counter-argument deserves a fair statement. The Federal Reserve has long held that it does not take elections into account, and a policy of never acting near an election would itself be a form of electoral timing, only in the opposite direction. The record offers a precedent for acting: in November 2022 the Committee raised rates by 75 basis points six days before the midterms, as press accounts of the present decision have recalled (247Wall St 2026). A hike on October 28 would be consistent with that record. But a hike chosen in order to demonstrate independence is also a decision influenced by the election, in the same way that a hold chosen to avoid it is. The cure for both is the same: decide on the evidence, state the evidence, and make the statement specific enough to be falsified.

Scenarios

I assess three outcomes, with probabilities that represent my own current judgment and that should be revised as the Committee’s communications and the data develop. The first is a hold accompanied by a specific, conditional statement of the reaction function, to which I assign roughly forty percent. In this case the Committee names what would trigger a December increase and presents the hold as a decision about timing, not direction. I judge this the modal outcome because it fits the Chair’s approach and the pre-report signals, and because it keeps the political risk low. It is also the outcome the thesis of this essay most favors, provided the specificity is real.

The second is a hold accompanied by vague or ambiguous communication, to which I assign roughly twenty-five percent. The Chair has no published path with which to supplement the statement, the communications review is unfinished, and the Committee’s earlier stumbles in this area are fresh. This is the dangerous outcome for credibility. Its likely symptoms are a steeper long end of the curve, higher inflation compensation at long horizons, a larger market reaction to each subsequent data release, and a persistent commentary that the Committee flinched before the election.

The third is a hike of a quarter point, to which I assign roughly thirty-five percent, a little below where the market stood in the days after the report. The case for it is the consistency and credibility argument above; the case against is the Chair’s stated aversion to responding to individual readings and the unusual proximity of the election. If the hike comes, it should be explained as the application of the same reaction function as in September, not as a demonstration of independence, because a demonstration is itself a response to the political question and would invite the same suspicion in reverse.

The ranking of these outcomes by their effect on credibility is not the same as ranking them by their effect on inflation. A clear hold may do more for the Committee’s reputation than an unexplained hike, and an explained hike may do more than either. What is common to the best versions of all three is the discipline of saying, in advance and in specific terms, what evidence would change the decision.

XIV. The Necessary Qualification: Flexibility Must Be Rule-Like

It would be too strong to claim that markets will forgive any deviation from prior guidance whenever the Fed invokes exceptional circumstances. If a central bank repeatedly changes its guidance without a coherent explanation, markets will come to regard its communications as unreliable. Flexibility must itself be rule-like. The Fed needs to make clear what information can change its assessment, which objectives remain invariant, which indicators matter, how risks are weighed, and why the new policy follows from the new information. The first of these is the easiest to state and the hardest to honour: a reaction function that cannot be described in advance cannot be seen to have been followed afterwards.

The reputational literature supplies the underlying logic. Barro and Gordon (1983) showed that a central bank’s incentive to deviate from its stated policy is restrained only by the reputational cost of being found to have deviated, and Rogoff (1985) showed that society may gain by delegating policy to an institution seen to give heavy weight to price stability. Both results depend on the institution’s type being believed. The more often the market suspects that the type is political, the less the reputation restrains and the less the delegation delivers. The present Committee is spending that reputation on the Chair’s approach to communication, and it should spend it knowingly.

The Fed’s institutional machinery is already moving in this direction. The projection materials separate the central projection from uncertainty, from historical forecast errors, and from participants’ own judgments on whether risks are balanced (Federal Reserve Board 2026a). Staff research is pushing further. Herbst, Konzem and Scofield (2026) document 1,265 alternative scenarios presented to the FOMC between 1968 and 2020 and find that the most accurate of them often anticipated major developments even when they missed the magnitudes. Adrian, Giannone, Luciani and West (2026) propose a Scenario Synthesis that places narrative scenarios and predictive distributions within a single Bayesian framework and assigns probabilities to scenarios consistently with a reference distribution. These are staff working papers and do not represent the views of the Board or the Committee, but they show the institution’s own analysts converging on the position argued here: under deep uncertainty, what should be communicated is a disciplined way of weighing possibilities, not a single prediction. A Committee without a dot plot could do worse than to publish, with each decision, its own short account of the scenarios it weighed and the evidence that would move it from one to another.

XV. The Central Proposition

The central thesis can be stated as follows. A credible central bank does not promise to be right about the future. It promises to decide rationally when the future turns out to differ from what was expected.

This resolves the apparent contradiction between forecasting and flexibility. The Fed’s forecast can be wrong without its communication being dishonest. Its guidance can change without its institutional credibility collapsing. An extraordinary shock can invalidate a previous policy path without showing that the original judgment was irrational. The real test of credibility is therefore not “Did the Fed do exactly what it said it would do?” but “Did the Fed respond to new information in a manner consistent with its mandate, its stated reaction function, and a transparent explanation of why its earlier assessment had changed, and did it do so in time?”

The final clause is the price of the argument and should not be dropped. Credibility resting on process must be earned by conduct: by revising promptly, by explaining plainly, and by keeping the reaction function stable enough to be learned. The October decision adds a corollary. Credibility resting on process must also be protected from the suspicion that something other than the process is at work. A decision that cannot be distinguished from electoral deference will be read as deference, and a Committee without published guidance has only the quality of its explanations with which to prove otherwise.

Delphi represents impossible certainty. Odysseus represents commitment bought at the price of flexibility, a price worth paying occasionally at the lower bound and rarely otherwise. Bayesian navigation represents rational policy under radical uncertainty. The Federal Reserve’s projections belong to the third category. Its forecasts are empirical rather than oracular; its commitments are conditional rather than absolute; and its credibility should derive less from never changing course than from showing, each time the economic regime shifts, that the process of updating remains intelligible, disciplined and consistent with its mandate. On October 28 the Committee will choose a heading. What it says about why it chose it will matter more than the heading itself.

References

Sources marked “staff research” are working papers that do not represent the views of the Board of Governors or the FOMC. Web addresses were checked on October 17, 2026. Market-commentary sources are cited for reported market pricing and for press accounts of events and are not treated as authorities on economic theory.

Abrams, Burton A. 2006. “How Richard Nixon Pressured Arthur Burns: Evidence from the Nixon Tapes.” Journal of Economic Perspectives 20 (4): 177–188.

Adrian, Tobias, Domenico Giannone, Matteo Luciani, and Mike West. 2026. “Risks and Uncertainty in Monetary Policy.” Finance and Economics Discussion Series 2026-061. Washington: Board of Governors of the Federal Reserve System (staff research). https://doi.org/10.17016/FEDS.2026.061

Andrade, Philippe, and Filippo Ferroni. 2021. “Delphic and Odyssean Monetary Policy Shocks: Evidence from the Euro Area.” Journal of Monetary Economics 117: 816–832.

Barro, Robert J., and David B. Gordon. 1983. “Rules, Discretion and Reputation in a Model of Monetary Policy.” Journal of Monetary Economics 12 (1): 101–121.

Bloomberg. 2026. “Why Kevin Warsh Again Skipped the Fed’s Dot Plot.” September 16. https://www.bloomberg.com/news/articles/2026-09-16/why-kevin-warsh-again-skipped-the-fed-s-dot-plot

Bondsavvy. 2026. “September 2026 Fed Dot Plot Sees Low 4% Fed Funds in 2027.” https://www.bondsavvy.com/fixed-income-investments-blog/fed-dot-plot

Brainard, William C. 1967. “Uncertainty and the Effectiveness of Policy.” American Economic Review 57 (2): 411–425.

Campbell, Jeffrey R., Charles L. Evans, Jonas D. M. Fisher, and Alejandro Justiniano. 2012. “Macroeconomic Effects of Federal Reserve Forward Guidance.” Brookings Papers on Economic Activity, Spring: 1–80. https://www.brookings.edu/articles/macroeconomic-effects-of-fomc-forward-guidance/

Campbell, Jeffrey R., Jonas D. M. Fisher, Alejandro Justiniano, and Leonardo Melosi. 2017. “Forward Guidance and Macroeconomic Outcomes Since the Financial Crisis.” NBER Macroeconomics Annual 2016, vol. 31. https://conference.nber.org/confer/2016/Macro16/Campbell_Fisher_Justiniano_Melosi.pdf

Continuum Economics. 2026. “FOMC Preview for September 16: Recent Data Back Warsh Tightening Hints.” https://continuumeconomics.com/a/25db91a4/fomc-preview-for-september-16-recent-data-back-warsh-tightening-hints

Eggertsson, Gauti B., and Michael Woodford. 2003. “The Zero Bound on Interest Rates and Optimal Monetary Policy.” Brookings Papers on Economic Activity 2003 (1): 139–211.

Elster, Jon. 1979. Ulysses and the Sirens: Studies in Rationality and Irrationality. Cambridge: Cambridge University Press.

Federal Reserve Board. 2008. “FOMC Statement.” December 16. https://www.federalreserve.gov/newsevents/pressreleases/monetary20081216b.htm

Federal Reserve Board. 2012. “FOMC Statement.” December 12. https://www.federalreserve.gov/newsevents/pressreleases/monetary20121212a.htm

Federal Reserve Board. 2019. “FOMC Statement.” January 30. https://www.federalreserve.gov/newsevents/pressreleases/monetary20190130a.htm

Federal Reserve Board. 2020. “FOMC Projections Materials, June 10, 2020.” https://www.federalreserve.gov/monetarypolicy/fomcprojtabl20200610.htm

Federal Reserve Board. 2026a. “FOMC Projections Materials, September 16, 2026.” https://www.federalreserve.gov/monetarypolicy/fomcprojtabl20260916.htm

Federal Reserve Board. 2026b. “Summary of Economic Projections.” Monetary Policy Report, July 10, 2026 (June 16–17 meeting). https://www.federalreserve.gov/monetarypolicy/2026-07-mpr-part3.htm

Federal Reserve Bank of St. Louis. 2026. “FOMC Summary of Economic Projections, September 2026.” FRED Blog. https://fredblog.stlouisfed.org/2026/09/fomc-summary-of-economic-projections-september-2026/

First Trust Advisors. 2026. “The Consumer Price Index (CPI) Rose 0.6% in September.” October 14. https://www.ftportfolios.com/Commentary/EconomicResearch/2026/10/14/the-consumer-price-index-cpi-rose-0.6percent-in-september

Fox Business. 2021. “Powell admits Fed got it wrong on inflation, says they should stop calling it ‘transitory.’” November 30. https://foxbusiness.com/politics/powell-fed-wrong-inflation-not-transitory

Greenspan, Alan. 2004. “Risk and Uncertainty in Monetary Policy.” American Economic Review 94 (2): 33–40.

Herbst, Edward, Scott Konzem, and Cristina Scofield. 2026. “Alternative Scenarios at the Federal Reserve from 1968 to 2020: Data, Interpretation, and Evaluation.” Finance and Economics Discussion Series, May (staff research). https://www.federalreserve.gov/econres/feds/alternative-scenarios-at-the-federal-reserve-from-1968-to-2020-data-interpretation-and-evaluation.htm

investingLive. 2026. “US September CPI 3.7% vs 3.6% expected.” October 14. https://investinglive.com/news/us-september-cpi-vs-3-6-expected-0/

J.P. Morgan Asset Management. 2026. “FOMC Statement: September 2026.” https://am.jpmorgan.com/us/en/asset-management/adv/insights/portfolio-insights/fixed-income/fixed-income-perspectives/fomc-statement-september-2026/

Kay, John, and Mervyn King. 2020. Radical Uncertainty: Decision-Making Beyond the Numbers. London: The Bodley Head.

Knight, Frank H. 1921. Risk, Uncertainty and Profit. Boston: Houghton Mifflin.

Kydland, Finn E., and Edward C. Prescott. 1977. “Rules Rather than Discretion: The Inconsistency of Optimal Plans.” Journal of Political Economy 85 (3): 473–491.

Lord Abbett. 2026. “September Fed Meeting: From Signals to Action.” https://www.lordabbett.com/en-us/financial-advisor/insights/markets-and-economy/2026/september-fed-meeting-from-signals-to-action.html

Lucas, Robert E., Jr. 1976. “Econometric Policy Evaluation: A Critique.” Carnegie-Rochester Conference Series on Public Policy 1: 19–46.

Money and Banking. 2020. “The Fed’s Crystal Ball: Looking Beyond the COVID-19 Recession.” June 15. https://www.moneyandbanking.com/commentary/2020/6/15/the-feds-crystal-ball-looking-beyond-the-covid-19-recession

Morningstar. 2026. “Warsh Sounds Hawkish, but Will There Be a September US Rate Hike?” https://www.morningstar.com.au/markets/warsh-sounds-hawkish-will-there-be-september-us-rate-hike

Orphanides, Athanasios. 2001. “Monetary Policy Rules Based on Real-Time Data.” American Economic Review 91 (4): 964–985.

Rabobank. 2026. “Federal Reserve: Warsh’s Hawkish Shift Reshapes Rate Outlook.” August 31. https://www.fxstreet.com/news/federal-reserve-warshs-hawkish-shift-reshapes-rate-outlook-rabobank-202608311210

Reifschneider, David, and Peter Tulip. 2017. “Gauging the Uncertainty of the Economic Outlook Using Historical Forecasting Errors: The Federal Reserve’s Approach.” Finance and Economics Discussion Series 2017-020 (staff research). https://www.federalreserve.gov/econresdata/feds/2017/files/2017020pap.pdf

Rogoff, Kenneth. 1985. “The Optimal Degree of Commitment to an Intermediate Monetary Target.” Quarterly Journal of Economics 100 (4): 1169–1189.

Schneider, Howard. 2026. Reuters, September 2. https://935thelloyd.com/?p=1002809 (syndicated copy).

Scripps News. 2021. “Jerome Powell, Janet Yellen ditch ‘transitory’ term to describe inflation.” https://san.com/cc/fed-chair-jerome-powell-ditches-transitory-term-to-describe-inflation

Spence, Michael. 1973. “Job Market Signaling.” Quarterly Journal of Economics 87 (3): 355–374.

The Hill. 2026. “Federal Reserve shifts away from forward guidance under new chief Kevin Warsh.” June 17. https://thehill.com/business/5929155-warsh-ends-fed-forecasts/

Virginia Business. 2026. “Hot September CPI Sends Rate-Hike Bets Soaring.” October 14. https://virginiabusiness.com/us-consumer-price-index-september-rate-hike-bets/

Yahoo Finance. 2026. “No Dot Plot, No Forward Guidance: Kevin Warsh’s First Fed Meeting Draws Mixed Reactions From Economists.” June. https://finance.yahoo.com/economy/policy/articles/no-dot-plot-no-forward-113128646.html

247Wall St. 2026. “Traders Give Warsh a 42% Chance of Doing Something This Month That Has Only Happened Once in the Past 36 Years.” October 15. https://247wallst.com/investing/2026/10/15/traders-give-warsh-a-42-chance-of-doing-something-this-month-that-has-only-happened-once-in-the-past-36-years/