Conditional co-movement · competing channels · falsification

6C Risk-On and Risk-Off Correlation: Test the Regime

Imagine two equity selloffs of similar size. One begins with a global-demand shock that also pushes crude oil lower. The other begins with Canadian inflation news that raises expected domestic rates while equities weaken. Calling both “risk-off” hides the channel that could make 6C respond differently.

Label
Measured, not assumed
Return clock
Aligned
Controls
USD, oil, rates
Original result
None

No permanent label

Define the Risk State Before Looking at 6C

Fact: 6C is a Canadian-dollar-versus-U.S.-dollar futures price, so both currency sides matter. Mechanism: global growth expectations, commodity demand, relative interest rates, and broad U.S.-dollar demand can transmit a shock into that price. Hypothesis: 6C may co-move with a declared risk factor in some states. “CAD is risk-on” is too broad to serve as either a definition or a finding.

Candidate factorOperational definitionWhat it representsWhat it does not prove
Equity returnReturn on a named broad index over a frozen intervalChange in equity-market valueCause of the move or investor risk appetite by itself
Equity volatilityChange or level in a named options-implied volatility measureMarket pricing of a defined volatility horizonA universal fear gauge for every asset and clock
Credit stressChange in a documented spread or official stress indexFinancing and default-risk conditionsAn intraday signal unless timestamp and frequency match
Cross-asset compositeFrozen weights across equity, volatility, credit, rates, and funding measuresA broader state labelObjectivity if weights were tuned to 6C outcomes
Named shockDates and timestamps from contemporaneous official recordsA specific event familyThat every event in the family has the same mechanism

Choose the primary factor without using future 6C returns. If a composite is necessary, fit its transformation and weights on a training sample, freeze them, and carry the exact construction into validation. Report the components as well as the label; otherwise a single dominant input can masquerade as a diversified measure.

Quote orientation matters

A higher 6C price means a stronger Canadian dollar against the U.S. dollar. Verify the current contract and quotation mechanics in the canonical 6C specification guide. Do not accidentally compare 6C returns with USD/CAD spot returns without reversing the spot orientation.

Comparable observations

Align Returns, Information Sets, and Trading Calendars

Correlating price levels can create a persuasive but spurious chart. Use returns or changes appropriate to the question, define the timestamp at which every input was knowable, and make the market clocks comparable. Daily closes from different timezones can contain different news.

1

Declare the horizon

Choose intraday, daily, or multi-day returns before analysis. A relationship at one horizon does not transfer automatically to another.

2

Select dated contracts

Use actual 6C expiries with an ex-ante roll rule. Keep roll windows visible and never derive a return across an unexplained stitch.

3

Synchronize timestamps

Sample all tradable inputs at common UTC boundaries and set stale-price limits. Do not forward-fill through a closed market as though it were a live quote.

4

Calculate contemporaneously

For event work, use pre-event and post-event prices observable at the same offsets. For daily work, use one common cut.

5

Preserve missingness

Distinguish no trade, market closure, data outage, and a true zero return. Document every exclusion.

Start with Pearson and rank correlation only if their assumptions fit the question, then inspect scatterplots, nonlinear dependence, and tails. Report the coefficient, sample count, uncertainty interval, window, and return construction together. A coefficient without those fields is not reproducible evidence.

Common drivers

Control the Dollar, Oil, and Relative Rates

Equities and 6C can move together because both react to a third variable. The goal is not to “control away” the world until a preferred sign appears. It is to ask whether the risk factor adds information after plausible shared channels are measured.

Name the shock before reading 6CEquity repricing is an observation. The causal candidate must be timestamped and defined independently of the later Canadian-dollar move.

U.S. dollar

The denominator can dominate

The BIS reports that the U.S. dollar remained on one side of most OTC FX transactions in its 2025 survey. Broad dollar demand can therefore coincide with moves across equities, commodities, and 6C. Use a declared broad-dollar measure and disclose any mechanical currency overlap.

Oil

Canada has a real trade channel

Canada is a major energy exporter, but an oil move can reflect demand, supply, geopolitics, inventories, or the dollar. Test crude returns and shock categories separately. The 6C oil study owns the detailed mechanism and lag protocol.

Relative rates

Compare Canada with the United States

Use matched-maturity rate changes or properly constructed expected-policy measures. A Canadian yield alone does not describe the relative return incentive embedded in a CAD/USD price.

Official events

Separate scheduled repricing

Tag Bank of Canada, Federal Reserve, Statistics Canada, BLS, and BEA releases from contemporaneous calendars. A handful of announcement windows can drive a full-sample relationship.

Primary and controlled estimates

Publish the raw conditional co-movement first. Then estimate a predeclared model in which 6C returns are related to the risk factor and controls using only information available at the same timestamp. Treat residual correlation as conditional association, not causal proof.

Primary
Raw return relation
Secondary
Controlled relation
Direction
Not assumed
Causality
Not established

Stability, not one coefficient

Roll the Window and Predeclare Regime Tests

A full-history coefficient compresses changing policy, commodity, volatility, and liquidity conditions into one number. Use rolling estimates to describe stability, but avoid choosing the window length after seeing the prettiest series. Then test a small set of declared regimes on untouched dates.

Risk direction

Compare up and down factor moves, not just one pooled slope. Require enough observations in both tails.

Shock size

Test central observations and frozen tail thresholds. A tail-only relationship should be labeled tail-specific.

Oil direction

Separate periods when crude confirms or contradicts the risk label.

Rate channel

Split by widening or narrowing Canadian-U.S. rate differences using past-only classifications.

Event versus ordinary

Estimate scheduled-event windows independently from ordinary trading intervals.

Volatility state

Use a frozen state definition and test whether signs and uncertainty remain stable.

Correct for the complete family of factors, horizons, windows, and splits. Keep a chronological development period, a validation period, and a final sealed holdout. Re-estimation after the holdout is opened converts that holdout into development data and requires a new untouched period.

Rival explanations

A Risk Label Can Be an After-the-Fact Story

For every apparent relationship, record at least one alternative explanation that would generate the same picture. This protects against assigning motive to a correlation chart.

Apparent observationCompeting explanationDiscriminating evidence
6C rises with equitiesBroad U.S.-dollar weakness moves both, rather than risk appetite causing CAD demandCondition on a declared broad-dollar return and inspect event timing.
6C falls in an equity selloffOil and Canadian terms-of-trade expectations weaken at the same timeSeparate oil-demand shocks from other selloffs and compare controlled estimates.
6C resists a selloffCanadian rate expectations rise relative to U.S. expectationsMeasure matched-maturity relative-rate changes around the same timestamp.
Relationship strengthens near the closeNon-synchronous sampling or closing-auction timing creates alignment artifactsRecalculate at common tradable timestamps and adjacent cutoffs.
Lead-lag appears intradayOne feed is slower, stale, or timestamped differentlyAudit source clocks, message latency, and stale-quote rejection before causality tests.

The CFTC’s Traders in Financial Futures report can provide a weekly, aggregated positioning context. Its categories do not reveal why a participant held a position, and weekly snapshots cannot identify the initiator of an intraday 6C move. Use that data as a slow conditioning variable, not a transaction-level explanation.

Conditional monitor

Falsify the Relationship Before Using It

No original result is reported for 6C correlation. No coefficient, lead-lag, regime effect, hedge ratio, forecast, backtest, or performance result is claimed. The decision tree below states what a future estimate would have to survive.

Reject

The relation does not survive

  • The sign flips under adjacent return horizons or timestamp repair removes the lead-lag.
  • Common U.S.-dollar, oil, or relative-rate controls absorb the apparent relation.
  • A later sealed holdout fails or realistic costs erase the proposed application.

Narrow

The evidence supports less

  • The estimate depends on one crisis, one tail, one horizon, or one volatility state.
  • Uncertainty spans economically different outcomes or a regime has too few independent observations.
  • State exactly where the relation appeared; do not promote a conditional result into a universal risk label.

Continue conditionally

Keep every gate visible

  • Verify data completeness, dated contracts, roll rule, calendar, and timezone alignment before each update.
  • Display raw and controlled estimates, uncertainty, sample size, exclusions, and the complete test family.
  • Require a separate costed decision rule with observable invalidation, while permitting the terminal conclusion “no stable conditional relationship.”
Sources, method and editorial disclosure

Sources and methods were reviewed August 13, 2026. This article supplies an original testing and falsification framework but reports no original empirical result. It is unsponsored editorial analysis.