The objection at full strength
Here is the strongest version of the case against this whole framing, and it deserves to be stated without hedging. Monetary policy does not operate on a system near equilibrium. Inflation dynamics, labour markets and credit cycles are non-stationary, reflexive, and shot through with expectations that change the moment the central bank speaks. Fitting an exponential decay to any of that is borrowing the language of a Wheatstone bridge and applying it to something that argues back. Relaxation time, in its original physical sense, describes a displaced system returning towards a fixed equilibrium at a rate set by internal linear response. An economy has no fixed equilibrium. Its "equilibrium" is a moving target that the policy action itself relocates. Importing a term from dielectrics and nuclear spins to describe the transmission of a rate rise smuggles in an authority the subject has not earned. If the physics does not apply, the sampling argument built on it — that under-sampling forecloses information irrecoverably — inherits nothing but the vocabulary.
That is a serious challenge, and an economist reading this should feel it land, not feel it dodged.
What survives the challenge
The strong version of the relaxation argument does not actually need the exponential, or the fixed equilibrium. It needs only this: a perturbation to the system — a rate change, a supply shock, a wage settlement — has a characteristic settling duration, however irregular and however path-dependent that settling turns out to be. Overnight repo markets absorb a rate decision within hours; measured pass-through in retail deposit rates takes weeks; the effect on hiring intentions takes quarters; the effect on wage-setting norms can take years and never fully complete before the next shock arrives. None of that requires linearity to be true. It requires only that these durations differ from each other by orders of magnitude, and that the appropriate frequency of observation differs accordingly. The claim under scrutiny is a ratio — between settling duration and sampling interval — not a differential equation. Reflexivity does not remove the ratio. If anything, a target that moves in response to being observed makes the ratio harder to guess in advance, which is an argument for widening the observation window, not for abandoning the concept of a window altogether.
So the objection is right that the physics does not transfer literally. It is wrong that this weakens the case for continuous intake. An ill-defined settling time is precisely the condition under which you cannot pre-tune a data release calendar to match it. The quarterly cycle assumes a settling duration it never measured.
The clock the committee actually has
A rate-setting committee works from streams with wildly different clocks. Price indices are collected monthly and published with a lag of two to four weeks, then revised — the UK's CPI basket weights are updated annually, seasonal adjustment factors get recalculated in the background, and initial prints for volatile categories are routinely revised by several tenths of a percentage point within the following two releases. Labour flows arrive through household surveys with sampling error large enough that a single month's change in the unemployment rate is frequently statistical noise, confirmed or reversed only two or three releases later. Credit aggregates lag further still, partly because bank balance sheet data is reported monthly but reconciled quarterly. Market-implied expectations are the outlier: continuous, priced every second, and arguably the fastest-relaxing signal in the whole intake pipeline, settling within minutes of a data surprise.
The published transmission lag for a policy rate change — most central bank research puts peak effect on output at twelve to eighteen months and peak effect on inflation nearer eighteen to twenty-four — is itself an estimate built from a sampling grid that is far coarser than the phenomenon it describes. A quarterly national accounts series, revised at least twice after first publication, offers perhaps six to eight genuinely independent observations across the two years over which a single rate move is still settling. That is a return of six to eight points to characterise a curve that determines whether the next decision tightens into a shock that has not yet arrived or one that already has. It is not that the committee lacks data. It is that the data samples the curve too sparsely to distinguish "still relaxing" from "already settled at a new level" from "hit by a second shock before the first one finished."
The failure this produces
The characteristic failure is specific and recurring: policy gets set on data that is revised after the meeting. The Monetary Policy Committee that raises rates in response to a GDP estimate showing resilient growth, only to see that estimate revised down by half a percentage point two months later, has not made an error of judgement. It has sampled a transient once and treated the single sample as the settled state. The economist responsible drafts the policy recommendation from the vintage of data available at cutoff, flags the uncertainty in a footnote, and moves on, because the committee meets on a schedule the data does not keep. The forecast round becomes an exercise in inferring a relaxation curve from too few points and calling the result a central projection.
Two objections that matter here
The first: continuous intake does not deliver a relaxation curve, it delivers volume, and a central bank drowning in high-frequency card-transaction and payroll data can still be systematically wrong about what has settled and why, absent a causal identification strategy. This is granted, and it is the sharper limit on the whole argument. Streaming faster does not resolve the question of whether a fall in core inflation reflects base effects, energy pass-through unwinding, or genuine disinflation in underlying wage growth — that requires identification, not frequency. What continuous intake with provenance does is make that identification problem tractable rather than solved: knowing which index vintage, which seasonal adjustment, which survey wave a number came from is what stops an economist from mistaking a methodology revision for an economic turning point, which happens more often than committees like to admit.
The second objection: some of this is over-sampled already, and more frequency would be waste, not insight. Market-implied expectations already relax within minutes; watching them tick every second buys nothing a five-minute snapshot would not. This is correct, and it is the honest boundary of the claim. The case for continuous intake is not that faster is always better — a monthly labour force survey timed sensibly can outperform a daily scrape prone to sampling noise. The case is narrower and asymmetric: under-sampling a slow-relaxing quantity, like the multi-year drift of wage-setting norms after a large inflation shock, destroys information that cannot be recovered later, while over-sampling a fast one costs storage and analyst attention, which can be trimmed at no informational loss.
The narrower claim
What holds is not that central banks should stream everything at maximal frequency. It is that the committee cannot know in advance which of its inputs are still mid-transient and which have settled, because the true relaxation time of a wage-price dynamic or a credit cycle is not knowable ahead of the fact — and a fixed publication calendar bets, every quarter, that the answer happens to match the calendar's own rhythm. A frozen, cycle-locked data regime forecloses that judgement structurally: it cannot tell a superseded equilibrium from a current one, because it only ever samples once per cycle regardless of what is happening inside the cycle. A regime of continuously updated, provenance-tagged belief — this observation, this vintage, this confidence, revisable as the next print arrives — does not solve identification and does not eliminate revision. It only ensures that when the curve finally does resolve, someone was watching it the whole way, rather than reading its last point and calling it the whole shape.