Clock windows · market quality · reproducibility

6C Trading Sessions: A Reproducible Liquidity Study

A window labeled “New York morning” does not stay fixed in UTC when North American daylight saving changes, and London does not switch on the same dates. Add Canadian and U.S. releases, regional holidays, and a nearly round-the-clock futures market, and a casual session comparison can move observations between buckets without the researcher noticing.

Window
Predeclared
Clock
Timezone-aware
Outputs
Five distributions
Original result
None

Definition before comparison

A Session Is an Analyst-Defined Clock Window

Fact: CME publishes trading and holiday schedules for its markets. Observation: traders commonly organize the day around Canadian, U.S., European, and Asian business hours. Application: a researcher can declare windows around those purposes. The labels themselves are not exchange-defined states, and they do not establish that one window is liquid, volatile, or directionally favorable.

FieldRequired recordWhy it matters
Local definitionStart, end, named timezone, inclusive boundary ruleExplains the economic purpose of the bin and prevents overlap.
UTC timestampsEvent and market-data time in UTCCreates one ordering across venues and source systems.
DST regimeStandard/daylight flag for every relevant regionCaptures weeks when Toronto, London, and Chicago are temporarily misaligned.
Trade dateCME trade date plus civil calendar dateA Globex trade date is not always the local date on the timestamp.
Boundary policyRule for bars or trades crossing a window edgeStops one volatile bar from being counted twice or assigned after inspection.

Use IANA timezone rules such as America/Toronto, America/Chicago, and Europe/London rather than hard-coded offsets. Save the resolved offset with every observation. If the study uses fixed UTC bins instead, say so plainly and avoid naming them after local business sessions when the match drifts.

Do not turn this study into the personal decision page

This URL measures market conditions across declared windows. The separate 6C execution-window guide is where a reader weighs those measured conditions against strategy horizon, schedule, event tolerance, and order size.

Dated instruments

Build the Dataset Around the Contract Roll

A continuous chart can conceal which expiry produced a quote or trade. Around migration, volume and depth can be divided between the nearby and next contract. A back-adjustment can also alter historical levels. Session comparisons should therefore start from dated 6C contracts and an ex-ante roll policy.

1

Ingest dated symbols

Retain contract month, exchange timestamp, trade date, price, quantity, best quotes, and depth. Reject records whose instrument identity is missing.

2

Audit the feed

Check sequence gaps, duplicate messages, locked or crossed books, stale quotes, impossible prices, and scheduled maintenance boundaries before computing statistics.

3

Declare migration

Choose a rule based on information available at the time, such as prior-day volume. Preserve both expiries around the switch and report sensitivity to a second reasonable rule.

4

Keep execution real

Route hypothetical fills to the actual dated book. Never execute on a back-adjusted price or splice a bar across two contracts.

Contract size, minimum increment, quotation, delivery, termination, and current trading-hours context belong in the canonical 6C contract-mechanics guide. This study should import a verified contract identifier and tick value from that authority rather than reproduce a second specification table.

Five outcomes, not one ranking

Measure Participation, Movement, and Executability Separately

“Active” can mean many trades, large completed volume, frequent quote updates, a wide range, or deep displayed liquidity. Those variables can disagree. A release may create large movement while visible depth retreats and slippage rises. Report each distribution before combining anything into a composite score.

Spread and depth

What was displayed?

Sample the best bid-offer spread and visible quantity at frozen distances from the touch. Report medians and tail percentiles, time at each spread width, and missing-book intervals. Depth is cancellable and does not guarantee a fill.

Trades and volume

What actually transacted?

Count trades, contracts, average trade size, and the distribution of intertrade durations for the selected expiry. Completed volume does not reveal the identities or motives of counterparties.

Realized movement

How far did price travel?

Freeze a range, absolute-return, or realized-variance calculation and preserve the full path. A large high-low range can be directional, reversing, or two-sided; direction is a separate outcome.

Execution

What did the order experience?

Replay declared marketable and passive order rules. Measure arrival-to-fill slippage, fill rate, queue wait, partial fills, adverse selection, commissions, and exchange fees under stated latency and size assumptions.

Primary comparison unit

Use one observation per date-window-contract combination, not every quote update as an independent session. Intraday messages form a dependent path. Summarize the path first, then estimate uncertainty with day-level or longer blocks that respect serial dependence.

Center
Median
Tails
Quantiles
Uncertainty
Blocked
Costs
Observed plus stress

Calendar overlays

Separate Ordinary Windows From Scheduled Information and Holidays

Canada and the United States release market-moving information on published schedules. The Bank of Canada sets policy on fixed announcement dates; Statistics Canada publishes a release calendar; the Federal Reserve, BLS, and BEA publish U.S. schedules. These sources identify timing. They do not establish the direction or magnitude of a 6C response.

ControlPredeclared treatmentAlternative explanation it addresses
Bank of CanadaTag decision time, press conference, Monetary Policy Report, and minutes since each itemPolicy information, not the clock, drives the distribution.
Canadian releasesTag actual, contemporaneous expectation if licensed, revision, and exact release timestampEmployment, inflation, trade, or GDP news dominates an otherwise ordinary window.
U.S. releases and FedUse official calendars and isolate event buffers rather than deleting only the release barThe USD denominator or relative-rate repricing moves 6C.
Canadian/U.S. holidaysIdentify full and partial holidays separately; compare with matched normal weekdaysRegional participation or exchange hours explain thin conditions.
CME schedule changesIngest the contemporaneous holiday notice and maintenance intervalA shortened session or closure creates artificial zero activity.

Publish at least three views: all eligible dates, ordinary dates excluding frozen event buffers, and named event families with adequate sample counts. If the apparent session difference vanishes outside a few announcements, the result is event-specific. That is useful evidence, but it is not a generic session effect.

Conditional distributions

Ask Whether the Ranking Survives Regime Splits

A full-sample average can blend calm and stressed years, policy-tightening and easing periods, low- and high-volatility weeks, and very different contract participation. Predeclare a limited set of economically motivated splits and require enough independent dates in every cell.

Longer-run volatility

Classify the state using past-only information, then compare each clock window within the same state.

Policy phase

Date Bank of Canada and Federal Reserve information from official records; avoid inventing a phase boundary after seeing 6C.

Oil state

Use a frozen oil-return or volatility definition only if the hypothesis was declared. The oil relationship study owns the deeper test.

Day of week

Report stability rather than assuming every weekday has the same participants and event calendar.

Roll proximity

Compare normal dates with the declared migration interval and show nearby/next-contract results.

Order size

Replay more than one frozen size. A book that supports one contract may not support a larger order at the same cost.

Every split increases the number of questions asked. List the entire analysis family, publish negative cells, and reserve an untouched chronological holdout. A ranking that appears only after many unplanned partitions is exploratory, not validated.

Interpretation boundary

A Better Distribution Does Not Select a Trade

A window with a tighter median spread can still have poor tail behavior on the day an order matters. A high-range window does not supply direction. Displayed depth is not committed liquidity, and historical fills do not reproduce a trader’s live latency or queue priority. These distinctions are limits on the claim, not footnotes.

Research status

No original result is reported for 6C session quality. This page produced no market-data sample, session ranking, spread estimate, depth statistic, volume profile, range comparison, slippage result, or trading-performance record. It therefore does not identify a best or worst session. The proposed metrics and splits are a falsifiable study design, not findings.

If the confidence intervals overlap materially, tail losses reverse the median ranking, or the result changes under adjacent clock boundaries, conclude that the windows are not reliably separated for that metric. “No defensible ranking” is a valid outcome.

Ending deliverable

Reproducible 6C Session-Study Protocol

Freeze this record before opening the final holdout. Publish it with the results so another analyst can reconstruct every date-window observation.

  1. Question: name one primary outcome and the exact clock windows being compared.
  2. Time: store UTC, local timezone, resolved offset, DST regime, civil date, and CME trade date.
  3. Universe: list dated contracts, data fields, feed provenance, known outages, and the ex-ante roll rule.
  4. Quality: define sequence-gap, stale-book, crossed-book, missing-quote, and outlier rejection thresholds.
  5. Measures: freeze spread, depth, volume, range, and slippage calculations plus order size and latency.
  6. Controls: timestamp official events, holiday schedules, maintenance, roll proximity, and regime labels.
  7. Inference: use day-aware uncertainty, correct for the declared comparison family, and report sample attrition.
  8. Falsification: reject a universal ranking if it fails adjacent boundaries, later years, alternate roll rules, or cost stress.
Sources, method and editorial disclosure

Sources and methods were reviewed August 13, 2026. This article presents an original research protocol, not original market results. It is unsponsored editorial analysis.