Skip to main content

Event timestamps · matched controls · rejection record

6S During FOMC Weeks: A Contamination-Aware Event Study

Calling every Monday-through-Friday interval containing a Federal Reserve decision an “FOMC week” creates a contamination problem. Swiss data, SNB communication, U.S. releases, holidays, contract rolls and unrelated shocks can enter the same window. A valid 6S study must timestamp those influences before it asks whether pre-decision, decision or post-decision behavior differs from ordinary weeks.

Windows
Pre / decision / post
Comparison
Matched controls
Contaminants
Recorded, not erased
Original results
None reported

The window is a data object

Map Every Release to a Causal Timestamp

The Federal Reserve meeting calendar supplies decision dates plus links to statements, minutes and projection materials. It does not supply a ready-made 6S event window. Preserve the official release timestamp, the exchange timestamp, the source timezone and the historical UTC offset.

SegmentFrozen boundaryInformation allowedPrimary threat
Prior snapshotDeclared instant before the decision week beginsOnly information available by that instantBackfilling a later expectation
Pre-decisionStart and end in UTC, not weekday proseScheduled and unscheduled news tagged as it arrivesCalling anticipation an FOMC effect
DecisionStatement timestamp through a fixed post-release horizonStatement, implementation note and contemporaneous market recordsMixing initial repricing with the press conference
Post-decisionFixed endpoint after all declared communication windowsOnly later information known by each observationAttributing a new shock to “follow-through”
Minutes are a separate event.

The Federal Reserve says minutes of regularly scheduled meetings are released three weeks after the policy decision. They belong to their own timestamped event record, not retroactively inside decision-week information.

Surprise requires a frozen expectation

Separate the Prior From the Decision Package

A target-rate change alone is not a surprise measure. The test must retain the market prior and describe which part of the announcement differed from it. If no licensed, timestamped expectations series is available, label the surprise field unavailable rather than reconstructing it from the subsequent 6S move.

Observed 6S returnCHF/USD futures can reflect several concurrent changes.

Policy surprise

Change in the expected U.S. policy path, measured from a declared pre-event market series.

Communication surprise

Statement, projections or press-conference information that changes the curve after the initial release.

Broad-dollar shock

Common USD repricing that may explain 6S alongside other dollar pairs.

Swiss or global shock

SNB, Swiss, European or risk information arriving inside the same measured window.

Ordinary weeks need comparable states

Match Control Weeks Without Looking at Outcomes

A raw FOMC-versus-non-FOMC average can confound policy communication with seasonality, volatility state, holidays or roll migration. Match using variables observable before the candidate week; never choose controls because their later 6S paths look similar.

Calendar

Same weekday structure

Match holiday, early-close and month-end status. Apply historical timezone rules.

Contract

Same lifecycle state

Match days to expiry, roll participation and the exact dated contract or documented splice rule.

Information

Comparable event load

Tag major scheduled U.S., Swiss and European releases using their official historical timestamps.

Market

Past-only volatility

Match a volatility or liquidity state estimated strictly before the window begins.

Protocol question—not a finding

After matching these states, do preregistered 6S volatility, range or market-quality measures differ across pre-decision, decision and post-decision segments relative to control weeks?

Unit
Dated FOMC meeting
Comparison
Matched non-event weeks
Allowed result
No detectable difference

Do not hide the path in one average

Measure Direction, Volatility and Executability Separately

A positive average return does not describe dispersion; a wide range does not prove tradeable liquidity; high volume does not establish favorable fills. Each claim needs its own field set and uncertainty estimate.

QuestionPrimary measureRequired dataDo not infer
DirectionCompleted 6S return over a frozen segmentDated contract, price field and exact endpointsA permanent directional bias
VariationAbsolute return, realized variation or declared range measureSequenced intraday observations and cleaning logExecutable opportunity
LiquiditySpread, displayed depth, trade size and quote-update intensityTimestamped quote or order-book recordsYour queue position
ImplementationArrival-price slippage, fill rate and implementation shortfallOwn order and acknowledgement timestampsCounterfactual fills for unsubmitted orders

CME DataMine lists settlements, Market by Order and PCAP as different historical products. That catalog clarifies that a settlement file cannot stand in for intraday book evidence. No CME dataset was purchased or analyzed for this article.

Make preferred stories expensive

Stress the Event Definition Before Believing It

Register a primary design and a limited sensitivity set before opening the final sample. Every tried window, filter and outcome remains in the audit record so one attractive specification cannot impersonate a stable pattern.

Window stress

Shift the boundaries

  • Adjacent non-overlapping pre and post segments
  • Statement-only versus statement-plus-press-conference
  • UTC and exchange-trade-date representations

Sample stress

Remove fragile dependence

  • One meeting at a time
  • Scheduled-projection versus other meetings
  • Holiday, roll and major-data exclusions

Inference stress

Respect repeated search

  • Meeting-level resampling
  • Family-aware uncertainty across outcomes
  • Chronological holdout never used for tuning
  • Keep all meetings. Crisis episodes and quiet decisions cannot be removed because they weaken the average.
  • Preserve revisions. Historical event fields must reflect what was knowable at the time.
  • Report dependence. Overlapping windows and repeated measures from one meeting are not independent observations.
  • Publish nulls. An interval too wide for the decision is an unusable result, not a hidden near miss.

Close the claim explicitly

FOMC-Week Rejection Record

Complete one record for each registered claim. The page does not assume that a measured difference becomes a trading rule.

Accept, limit, or reject the study claim

Registration
Sample, timestamps, prior field, windows, controls, measures, exclusions and code hash.
Coverage
Eligible meetings, missing records, contaminated windows and control-match quality.
Estimate
Effect size, uncertainty, all variants and meeting-level influence.
Accept
Useful magnitude and sign survive the registered holdout and robustness set.
Limit
Difference exists only in a state identifiable before the window and remains useful after costs.
Reject
Controls erase it, sign changes, uncertainty is too wide, data are incomplete, or later evidence fails.
Research status as of August 21, 2026

No original 6S FOMC-week dataset, event estimate, coefficient, confidence interval, predictive result or trading performance is reported here. This article specifies a study protocol. Until a sealed implementation passes the record, the conclusion is not tested.

Sources, methods and editorial disclosure — reviewed August 21, 2026

Sources and methods were reviewed August 21, 2026. This unsponsored article distinguishes official event facts, proposed measurements, hypotheses, statistical inference and trading application. It reports no original market result.