Skip to main content

Conditional co-movement · synchronized clocks · causal restraint

6N Correlations: A Conditional Co-Movement Test

A 20-session window can show 6N and another market moving together while a 120-session window shows the opposite sign. Neither number is automatically wrong. They answer different questions, and neither establishes why the assets moved.

Measure
Returns
Windows
Predeclared
Causality
Not assumed
Original findings
None

Question before coefficient

Specify the Relationship You Mean to Test

“6N is correlated with risk assets” is too broad to test. A registered claim names the second series, return horizon, observation clock, contract-roll policy, sample, conditioning state, estimator, uncertainty method and decision use.

Protocol statement—not a result

Estimate whether synchronized 6N and candidate-market returns exhibit a stable conditional relationship after broad-dollar and rate controls, then test whether that relationship survives later chronological data and executable costs.

Fact
6N is a dated NZD/USD future
Hypothesis
Co-movement may vary by state
Allowed answer
Unstable or unusable
Fact

CME Rulebook Chapter 258 defines the deliverable NZD/USD futures contract. A research series must retain delivery-month identity.

Inference

A positive estimate may be consistent with a shared growth or dollar shock. It does not identify that shock without additional evidence.

Trading application

A relationship becomes usable only if a frozen decision rule improves a relevant benchmark in later data after costs. This page reports no such test.

Comparable transformations

Correlate Returns, Not Convenient Price Levels

Two trending price levels can create a persuasive coefficient even when their changes are unrelated. Convert prices with a declared transformation, use the same horizon, and avoid overlapping observations unless the uncertainty calculation accounts for dependence.

rt = ln(Pt / Pt−1)ρt,w = corr(r6N, rX) over window w
CandidateSeries constructionMain ambiguity
6A futuresDated futures returns with the same roll and clock policy as 6NShared USD quotation versus distinct domestic drivers
Equity futuresReturns from a declared CME expiry and close or synchronized intraday markWhich market session contained the information?
U.S. ratesYield changes, not yield levels; maturity fixed in advanceLevel, slope and policy expectations can move differently
CommoditiesReturn on a named futures contract or official indexExport narrative may be indirect and lagged
Broad USDChange in an external broad-dollar measureContemporaneous control is not a tradable advance signal

RBNZ’s B1 exchange-rate and Trade Weighted Index series supplies official daily NZD reference data and a 17-currency TWI. It is a separate data product from executable CME 6N quotes.

The clock is part of the variable

Synchronize Information Windows Before Estimating Anything

A New Zealand release, a U.S. equity close and a CME trade-date boundary do not share one civil-day clock. Same-date joins can pair a 6N return with information that arrived after it.

01

Preserve source time

Retain exchange timestamps, source timezone, UTC offset and trade date.

02

Define the window

Use close-to-close, fixed UTC bars or event windows chosen before outcomes.

03

Resolve calendars

Version holidays, daylight-saving changes, missing observations and early closes.

04

Join causally

Use only values whose publication or market timestamp precedes the endpoint.

Calendar-date equality is not temporal equality.

When one market is closed or an input is stale, mark the pair ineligible or apply a predeclared carry rule. Never treat a repeated value as a fresh observation without disclosure.

Candidate causes, not labels

Keep Mechanisms Separate From Measured Co-Movement

A mechanism suggests what to condition on and what could falsify the story. It does not authorize a direction forecast. Several channels can act at once, oppose one another, or disappear when policy and positioning change.

Observed 6N–X co-movementOne statistic can be generated by several shared shocks.

Broad-dollar channel

A common USD move can affect both 6N and another USD-quoted market. Test a broad-dollar control and avoid using an index with circular composition as causal proof.

Relative-rate channel

Changes in New Zealand and U.S. rate expectations can alter carry and valuation. Use dated curve changes, not only current policy-rate levels.

Growth and risk channel

A shared growth repricing can move equities, commodities and NZD together. That relationship may fail in a domestically driven shock.

Liquidity and positioning

Deleveraging, roll migration or thin books can create transient co-movement that does not represent a durable macro link.

One coefficient is not a regime

Publish Rolling Estimates With Their Uncertainty

Choose a primary window for the decision horizon and a limited sensitivity set. Report the number of paired observations, missing-data policy, confidence interval or block-bootstrap distribution, and every candidate relationship examined.

Primary

Pearson

Measures linear co-movement and is sensitive to outliers. Use only with a documented return and cleaning policy.

Rank check

Spearman

Tests monotonic rank association. It is a robustness check, not an automatic replacement for the primary estimator.

Conditional

Partial model

Residualize against predeclared controls using training data only. Interpret model dependence explicitly.

Timing

Lead–lag grid

Use a narrow preregistered lag set and correct for searching. A best lag selected after inspection is exploratory.

  • Keep signs and magnitudes. Do not reduce estimates to “correlated” or “not correlated.”
  • Show window endpoints. A rolling value must be traceable to the observations it used.
  • Account for dependence. Overlapping returns and rolling windows reduce effective independence.
  • Retain failures. Missing windows and unstable estimates remain visible in the result table.

Prevent mechanical relationships

Control Rolls, Events, Revisions and Search Degrees of Freedom

A correlation can be manufactured by construction choices. Freeze the continuous-series rule, event taxonomy, sample exclusions and candidate list before opening the validation segment.

ConfoundRequired controlFailure signal
Contract migrationMatch dated expiries or use separately documented roll mappingsRelationship appears only at splice dates
Shared USD quoteInclude an external broad-dollar control and non-USD comparisonEffect vanishes after the common factor
Scheduled releasesTag official RBNZ, Stats NZ and U.S. publication timestampsEstimate depends on a few event bars
Data revisionsUse vintage-aware macro data or enforce release-time availabilityRevised values improve historical fit
Multiple comparisonsRegister candidates and report family-aware uncertaintyOnly the best of many undisclosed pairs survives

CME DataMine distinguishes settlements and order-level products; the CFTC Commitments of Traders program publishes position categories and report documentation. Neither is a substitute for a timestamped, licensed research manifest.

Try to break the relationship

Predeclare What Would Make the Correlation Unusable

Robustness is not finding a window that restores the preferred sign. It is exposing the same registered claim to reasonable alternate histories and accepting rejection.

Temporal

Does the sign persist?

  • Early versus late chronological blocks
  • Adjacent non-overlapping horizons
  • Daylight-saving and holiday variants

Structural

Does the channel persist?

  • With broad-dollar and rate controls
  • Event versus non-event periods
  • Roll-proximate versus ordinary dates

Decision

Does it add value?

  • Later sealed sample
  • Risk-matched naive baseline
  • Latency, spread, slippage and fees

Acceptance decision

Correlation Acceptance Record

Complete this record for each candidate. A blank required field or unstable estimate closes the claim; it does not become a qualitative signal.

Decision: accept, limit, or reject

Definition
Pair, returns, clock, rolls, window, estimator and sample hash.
Evidence
Estimate path, uncertainty, controls, all variants and sealed holdout.
Use
Declared decision, baseline, execution delay, costs and invalidation.
Accept
Direction and useful magnitude remain stable under registered checks.
Limit
Relationship is state-specific and the state is observable without hindsight.
Reject
Sign flips, interval is too wide, control removes it, or no net decision value appears.
Research status as of August 16, 2026

No original 6N correlation study, coefficient, confidence interval, regime finding, predictive test or trading result is reported here. This page defines a protocol. Until a sealed study passes the record, the conclusion is not tested.

Sources, methods and editorial disclosure — reviewed August 20, 2026

Sources and methods were reviewed August 20, 2026. This unsponsored article distinguishes contract facts, candidate mechanisms, testable hypotheses, statistical inferences and trading applications. It reports no original market result.