Three Clocks Run After a Major Earthquake and None of Them Match

A large earthquake produces a headline within minutes and a settled number within years. Almost everything a trader can act on during the first session sits at the wrong end of that gap. The instinct to treat the early figures as a small, noisy version of the final figures is the part that costs money, because the early figures are not estimates of the same quantity at all.

Three separate processes begin at the moment of the shaking, and each one publishes on its own schedule. The hazard agencies publish an impact band inside half an hour and keep revising it. The federal disaster machinery runs on dates written into regulation, and those dates are measured in weeks. Corporate accounting publishes on the quarterly calendar, which means the first audited number a listed company puts on the event may not exist for two or three months. A position taken on the headline is exposed to all three, and can only be checked against the fastest one.

What follows is a description of the clocks, the published values attached to each, and where a same-day position is in fact taking risk. It is not a claim about which direction anything moves.

The first published impact number is a band, and it is built to move

The U.S. Geological Survey runs an automated system called PAGER, for Prompt Assessment of Global Earthquakes for Response. Its documentation states that initial results are generally available within 30 minutes of a significant earthquake, shortly after the location and magnitude have been determined. That is the number most likely to be circulating while the first session is still open.

PAGER does not publish a point estimate of damage. It publishes an alert colour, and the colour is defined by ranges. Green corresponds to zero estimated fatalities and estimated economic losses under one million dollars. Yellow covers one to 99 fatalities, or one to 100 million dollars. Orange covers 100 to 999 fatalities, or 100 million to one billion dollars. Red covers 1,000 or more fatalities, or one billion dollars and above. The two criteria are assessed separately and, in the system's own description, the higher of the two sets the overall alert level.

Two features of that structure matter for anyone reading the alert as a market input. The first is the width of the bands. An orange economic alert spans an order of magnitude on its own, from 100 million to one billion dollars, before any uncertainty around the estimate is considered. The second is that PAGER carries its uncertainty explicitly: the onePAGER product displays histograms showing the percent likelihood that adjacent alert levels occur, and the documentation notes that when the likelihoods of two neighbouring levels are comparable, there is little in the loss model to distinguish between them. An alert sitting on a boundary is a different object from an alert sitting in the middle of its band, and the published product says which is which.

The magnitude itself is also provisional. USGS states that for large earthquakes outside the United States an initial estimate of magnitude and location is released within about 20 minutes, that a first update typically follows within a few hours once the majority of real-time data has arrived, and that a second update follows within days to weeks when the event is reanalysed for the archival catalogue. In several regions with dense local networks, including California, Utah, Alaska and the Pacific Northwest, an initial rapid magnitude can be released without human review, and revisions are possible within minutes. Archived historic magnitudes are sometimes revised again when new calculation methods are developed.

None of this is a defect in the system. Rapid alerting exists so that response agencies can move before certainty arrives, and the documentation is explicit that users should always seek the most current release. The problem is a transplant problem: a number engineered for emergency triage is being priced as though it were a settled damage figure.

One further distinction is worth stating plainly, because it collapses easily in fast markets. The PAGER economic loss estimate is an estimate of total economic loss, not of insured loss. The two are different quantities separated by coverage terms, deductibles, take-up rates and policy limits. A high total-loss alert does not translate into a proportional insured figure, and the relationship between them is not something a colour band contains.

The federal clock has dates written into regulation

The second clock is the slowest one that is publicly documented in advance, which also makes it the most predictable. The process for a U.S. major disaster declaration is set out in 44 CFR Part 206, and the schedule is legible before any given event occurs.

Damage assessment comes first. Under 44 CFR 206.33, assessment teams are composed of at least one federal and one state representative, with local participation where possible, and the resulting Preliminary Damage Assessment supports both the governor's request and FEMA's recommendation to the President. The same section permits the Regional Administrator to waive the joint assessment requirement for incidents of unusual severity and magnitude that do not require field damage assessments.

The governor's request follows. Under 44 CFR 206.36, the request must be submitted within 30 days of the occurrence of the incident in order to be considered, and that period can be extended only if the governor submits a written extension request during the original 30 days. The same section allows an abbreviated request for catastrophes of unusual severity.

The factors weighed are also published. Under 44 CFR 206.48, FEMA uses one dollar per capita of state population as an indicator that a disaster may warrant federal assistance, adjusts that figure annually using the Consumer Price Index for All Urban Consumers, and applies a minimum of one million dollars in public assistance damage regardless of state population. Localised concentration of damage can support a declaration even where the statewide per capita figure is not met, and insurance coverage in force is one of the considerations. On cost sharing, 44 CFR 206.47 sets the federal share of eligible permanent restorative work at 75 percent, and provides that an increase to as much as 90 percent may be recommended where federal obligations under the Stafford Act meet or exceed 100 dollars per capita of state population for disasters declared after 1 January 2001, again indexed to CPI-U.

What the regulation does not fix is elapsed time. The chart below takes every U.S. major disaster declaration in FEMA's public dataset with incident type Earthquake and declaration type DR, from October 2006 through March 2023, and measures the days between the incident begin date and the declaration date.

Ten declarations, and the spread runs from zero days to 113. The median is 47.5 days and the mean is also 47.5. Six of the ten took more than 30 days; five took more than 60. Only two arrived within a week of the incident beginning, and one of those was declared on the same date the incident began, which is what the waiver and abbreviated-request provisions above exist to allow.

That distribution is the useful part. It is not a forecast for the next event, and a sample of ten cannot support one. What it does establish is that a same-day expectation of federal money arriving is not supported by the record, and that the interval between a headline and a declaration has historically been long enough to contain several full reporting cycles.

The accounting clock is the slowest of the three

Corporate disclosure runs on a calendar that has nothing to do with the timing of the event. Under SEC Form 10-Q, General Instruction A.1, a quarterly report is due within 40 days after the end of the fiscal quarter for large accelerated filers and accelerated filers, and within 45 days for all other registrants. That is the first point at which a reviewed figure for the period containing the event exists in a filing.

Form 8-K reports are due within four business days after the triggering event, but the triggers are specific. Item 2.06 requires disclosure once a board or authorised officers conclude that a material impairment charge is required. Item 8.01 is a catch-all for other events of material importance. Nothing in that structure obliges a company to publish a catastrophe loss estimate on any particular day, which is why early figures on a disaster tend to circulate as modelled ranges from third parties rather than as booked amounts from the exposed balance sheets.

How far those early modelled figures move before they settle is not something established here, and it should not be assumed in either direction. The structural point stands on its own: the first number carrying an auditor's involvement and a filing deadline sits at the end of the quarter plus 40 or 45 days, and the first number circulating on the day sits roughly 30 minutes after the shaking. Treating them as the same series is the error.

Separating the exposures before sizing anything

A single headline bundles at least three distinct exposures that resolve on different schedules. Keeping them apart is what makes a position describable in advance.

The practical consequence is about horizon rather than direction. A position whose thesis rests on physical disruption has observable checkpoints within days: whether terminals reopened, whether power was restored, whether closures were lifted. A position whose thesis rests on the size of an insured loss has no checkpoint at all until a filing lands, which means the holding period implied by the thesis is measured in weeks and the position is exposed to everything else that happens in those weeks. A position whose thesis rests on reconstruction spending is on the longest clock of the three, and the record above shows the first gate on that path has historically taken a median of 47.5 days to open.

Sizing that ignores this mismatch produces a familiar failure: a position sized as though the uncertainty resolves today, held across a period where nothing resolves, and closed on an unrelated move. The band width in the alert, not the midpoint, is the honest input to that sizing decision.

What Would Invalidate This

Several conditions break the frame described here, and each is checkable.

• The event is outside the United States. The FEMA process, the 30-day request window, the per capita indicators and the 75 percent cost share are U.S. federal provisions under 44 CFR Part 206. They say nothing about how any other jurisdiction handles disaster relief, and the declaration-lag distribution above does not transfer.
• The exposure is not domiciled where the shaking was. Reinsurance and retrocession move exposure across borders and across balance sheets. Geographic proximity to an event is a poor proxy for who ultimately carries the loss, and that structure is not visible in any of the three clocks.
• The event is small enough that the alert never moves. A green PAGER alert, with zero estimated fatalities and estimated losses under one million dollars, has very little room to be revised upward into something that matters. The revision risk described here is concentrated in the middle bands, where the boundary between yellow and orange, or orange and red, is close.
• The instrument prices the hazard directly. Catastrophe-linked securities and parametric structures can settle on a defined trigger rather than on reported loss. Where a contract specifies its own resolution mechanism, that mechanism governs, and the reporting calendar described above is not the binding constraint.
• A declaration is not the operative variable. Where the relevant spending is state, municipal or private rather than federal, the FEMA timeline is measuring something that does not apply, and the median in the chart is not a proxy for it.

Concrete Framework

1. Write down the alert level and its band before doing anything else. Record the PAGER colour and both defining ranges. If the alert is orange, note that the economic band alone spans 100 million to one billion dollars. Where the onePAGER histogram shows comparable likelihood for an adjacent level, record that too, because it means the model does not distinguish between them.
2. Timestamp the magnitude. Note that a first revision typically arrives within a few hours, and reanalysis within days to weeks. If a thesis depends on the magnitude figure being final, it is not actionable inside the first session.
3. Name which of the three exposures the position is on. Physical disruption, insured loss, or public rebuild spend. A position that cannot be assigned to one of these has not been specified well enough to size.
4. Set the horizon from the exposure, not from the headline. Physical disruption has checkpoints within days. Insured loss has none until the next 10-Q, due 40 days after quarter end for large accelerated and accelerated filers, 45 days for others. Public spend has none until a declaration, which in the 2006 to 2023 earthquake record took a median of 47.5 days and ranged from 0 to 113.
5. Size against the top of the band, not the midpoint. If the position cannot survive the alert being revised to the adjacent level, it is sized to a point estimate that the publishing agency has explicitly declined to make.
6. Set a time stop at the next scheduled information event. Identify the specific date the next real datapoint is due, and treat the absence of resolution by that date as an exit condition rather than a reason to wait longer.
7. Log the revision when it lands. Record the initial alert, the revised alert, the magnitude at each stage, and the declaration date if one occurs. Over several events this builds a personal record of how far initial figures moved in practice, which is the only version of that statistic that will be specific to the instruments being traded.