Synchronized returns · common-factor controls · stability tests

6S Correlations: Test Co-Movement, Don’t Rank It

By Kyle ParrottUpdated August 20, 2026

A 6S correlation can be positive in one window and negative in the next without either estimate being a calculation error. Policy regimes change, the dollar becomes a stronger or weaker common factor, event composition shifts and one outlier can dominate a short sample. A permanent “top correlations” list hides exactly the instability a trader needs to see.

Build synchronized returns Try to break the relationship

Input: Returns, not levels
Estimate: Rolling + uncertain
Drivers: Controls required
Permanent ranking: Rejected premise

Comparable changes on one clock

Synchronize Returns Before Estimating Co-Movement

Correlating trending price levels can generate an impressive statistic with little decision meaning. Define completed returns or yield changes over identical causal windows, preserve dated contracts and use only source values available by each endpoint.

r_t,h = ln(P_t / P_t-h)ρ_t,w = corr(r^6S, r^X) over window w

Identify

Name instrument, venue, expiry, quote direction and source for every series.

Normalize

Map source timestamps to UTC while preserving local and exchange dates.

Transform

Compute declared returns or yield changes from completed past data.

Pair

Join only intervals with valid, non-stale observations on both sides.

Quote direction matters.

6S is a CHF/USD future: a higher futures price means more U.S. dollars per Swiss franc. A USD/CHF spot series runs in the inverse quote direction. Transforming both to consistent economic returns is mandatory before interpreting a sign.

Candidate set before coefficients

Define Equities, Gold, Rates, JPY and USD Precisely

These are research families, not ranked drivers. Each can share a mechanism with CHF in some states and diverge in others. Register one primary series and a limited sensitivity choice per family.

Family	Required construction	Competing explanations
Equities	Named cash index or dated equity future; synchronized return	Risk appetite, growth news, USD funding and session mismatch
Gold	Named spot source or dated gold future with its own roll rule	Real yields, dollar exposure, inflation, liquidity and haven narratives
Rates	Yield changes or dated rate-futures returns at fixed maturities	Policy path, term premium, growth and inflation components
JPY	Consistently oriented JPY/USD return from a declared venue	Shared haven/funding state versus distinct policy and regional shocks
Broad USD	External dollar-index change with publication timing retained	Mechanical CHF/USD quotation and common-dollar influence

The Federal Reserve H.10 summary defines broad, advanced-economy and emerging-economy dollar indexes. The BIS effective exchange-rate portal documents nominal and real trade-weighted indexes. Those official series are external controls or comparison measures; neither is a real-time 6S execution feed.

A coefficient needs a history and interval

Publish Rolling Estimates With Uncertainty

Choose one primary lookback matched to the decision horizon and a small registered sensitivity set. Each point should expose its window endpoints, paired observation count, missing-data treatment and uncertainty.

Primary

Pearson

Linear co-movement on declared returns. Outlier sensitivity and distribution assumptions remain visible.

Robustness

Spearman

Rank association for monotonic co-movement. It tests a different quantity and is not an automatic substitute.

Tail

Conditional estimates

Predeclared stress or volatility states with enough independent observations and wider uncertainty.

Timing

Lead-lag grid

A narrow registered lag family that enforces availability and execution delay.

Report magnitude and sign. Do not collapse estimates into “correlated” or “uncorrelated.”
Show effective sample size. Rolling and overlapping returns create dependence.
Keep unstable windows. Sign flips and wide intervals are the result, not chart clutter.
Separate description from prediction. Same-window correlation cannot establish a tradable lead.

Ask what generated the statistic

Control Common Factors Without Erasing the Question

A partial relationship is model-dependent, but it can reveal whether a pair merely shares a broad-dollar, rate or event shock. Select controls from a causal diagram before results and fit all transformations inside the training period.

Observed 6S–X correlationSeveral common causes may produce the same coefficient.

Broad-dollar factor

Test whether common USD repricing explains both series, while avoiding an index that mechanically embeds the dependent return as causal proof.

Relative-rate factor

Use declared curve changes and matched horizons rather than current policy-rate levels.

Scheduled events

Tag FOMC, SNB and major data releases; report event and non-event windows separately.

Liquidity state

Account for session, holidays, quote staleness, contract rolls and stress where valid fields exist.

A control is not a tradable signal.

A contemporaneous broad-dollar or yield input may explain a historical relationship while arriving too late to support a 6S order. Prediction requires a separate causal-timing and cost test.

Five families can become hundreds of trials

Count Every Pair, Window, Lag and Regime

Correlation research expands rapidly when analysts try many instruments, transformations, lookbacks, lags and sample filters. Freeze the family before the final sample and use family-aware uncertainty or a discovery/validation split.

Search dimension	Registry field	Disclosure
Instrument	Every equity, gold, rate, JPY and dollar series tried	Report all candidates, not only the strongest
Transformation	Return type, horizon, winsorization and scaling	Separate primary from sensitivity results
Lookback	Every rolling window and minimum sample count	Show stability rather than a chosen snapshot
Lag	All leads, lags, processing delay and outcome horizons	Distinguish contemporaneous from predictive
Regime	Every event, volatility, policy and liquidity split	Require real-time identifiability

Primary question—not a ranking

Which, if any, preregistered candidate maintains a useful relationship with 6S across later chronological data after common-factor controls, multiple-testing adjustment and conservative implementation?

Candidates: Declared family
Validation: Chronological
Allowed result: None stable

Stability earns use

Try to Break Every Candidate Relationship

A correlation becomes unusable when its sign, magnitude or timing depends on an undisclosed choice. Robustness means accepting failure under reasonable alternate histories, not searching until the preferred relationship returns.

Temporal

Does it survive time?

Early versus late chronological blocks
Adjacent non-overlapping horizons
Leave-one-event-period-out estimates

Structural

Does it survive controls?

Broad-dollar and rate factors
Event versus non-event periods
Roll, holiday and liquidity states

Decision

Does it add value?

Later sealed sample
Naive and no-signal baselines
Latency, spread, slippage and fees

Correlation acceptance record

Definition: Pair, clocks, returns, rolls, window, estimator, lags and sample hash.
Evidence: Full rolling path, uncertainty, controls, all variants and sealed holdout.
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 remains.

Research status as of August 21, 2026

No original 6S correlation coefficient, ranking, confidence interval, regime result, predictive test or trading performance is reported here. No original result is reported. Equities, gold, rates, JPY and USD measures are candidates for a protocol—not established permanent drivers.

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

CME Rulebook Chapter 254: Swiss Franc/U.S. Dollar futures for 6S contract identity.
CME Swiss Franc futures contract page for current product context; exact mechanics are centralized in the canonical 6S contract guide.
Federal Reserve H.10 dollar-index summary for official dollar-index definitions and revision notes.
Federal Reserve H.15 selected interest rates for official U.S. Treasury yield-series definitions.
BIS effective exchange rates data portal for official trade-weighted index definitions, coverage and methodology links.
CME DataMine historical-data catalog for exchange dataset distinctions. No data was purchased or analyzed for this article.
NIST/SEMATECH time-series analysis guidance for dependence and time-order context.

Sources and methods were reviewed August 21, 2026. This unsponsored article separates source definitions, candidate mechanisms, empirical estimates and trading use. It reports no original market result and no permanent correlation ranking.