Calendar hypotheses · multiplicity · stability
6M Seasonality: Test Calendar Effects Without Forecasting
Twelve months already create twelve possible stories. Add weekdays, turn-of-month windows, holidays, several return horizons, long and short directions, and a few exclusions, and a researcher can inspect hundreds of calendar cells. One unusually strong historical average is then expected somewhere even when no stable seasonal effect exists.
- Hypotheses
- Registered
- Returns
- Roll-clean
- Validation
- Chronological
- Original finding
- None
More cells, more false stories
A Calendar Grid Can Manufacture an Apparent Pattern
Seasonality is a recurring calendar-conditioned distribution, not a chart anecdote and not a mechanism by itself. The family includes every cell, horizon, direction, transformation, subgroup, and manual exclusion inspected. Count it before testing.
Established fact
Calendar labels are observable
Month, weekday, business-day position, holiday status and scheduled event timing can be defined from dated records.
Open hypothesis
A label may condition a distribution
A predeclared calendar state may differ from a matched baseline in return, range, spread, or another named outcome.
Unsupported leap
History does not make a forecast
An in-sample mean, even when positive, does not establish persistence, direction next year, or executable profitability.
NIST’s multiple-comparison guidance explains that repeatedly applying pairwise comparisons does not generally control a larger family. A seasonal report must disclose all tried calendar definitions, not only the cell that survived inspection.
Before opening outcomes
Register the Calendar Hypothesis and Its Economic Clock
A protocol should make the candidate computable without judgment. “The peso is strong in spring” is not a hypothesis. Define the calendar, timezone, eligible dates, anchor, entry and exit timestamps, direction, primary outcome, baseline, mechanism, controls, and failure threshold.
| Registry field | Required specification | Why it matters |
|---|---|---|
| Calendar state | Month, weekday, business-day index, holiday relation, or named interval | Prevents redrawing the window around the outcome |
| Clock | UTC anchor plus named local timezone and date-aware offset | Prevents DST and overnight date drift |
| Universe | Dated 6M contracts, sample dates, exclusions and roll policy | Defines what had a chance to enter the sample |
| Outcome | Signed return, absolute return, range, spread, slippage, or another single primary metric | Stops direction and volatility from being interchanged |
| Baseline | Clock-, horizon-, risk-, and condition-matched non-event dates | Separates a calendar effect from ordinary market structure |
| Mechanism | Predeclared explanation and observable intermediate evidence | Permits a causal story to fail independently of the price result |
| Falsification | Minimum effect, uncertainty, sign, stability and cost requirements | Makes “no seasonal evidence” a valid outcome |
A hypothesis registry is not a findings table. The page offers examples of fields, not a claim that any month, weekday, or holiday window changes 6M behavior.
Years, not rows, carry replication
Construct an Eligible Sample With Honest Independence
A monthly seasonal effect has only one corresponding month per year. Daily bars inside the same month share shocks, policy regimes, and overlapping paths; they do not turn a short history into hundreds of independent annual replications. Report years, episodes, dates, and attrition separately.
Primary sampling unit
Choose the independent unit that matches the claim. For a month-of-year hypothesis, summarize each eligible year-month first. For a turn-of-month event, create one nonoverlapping episode per anchor. Estimate uncertainty with blocks that retain cross-day and within-episode dependence.
- Eligible years
- Reported
- Episodes
- Nonoverlapping
- Attrition
- By reason
- Unit
- Claim-matched
- Freeze the history start. Do not choose a start date because an earlier regime weakens the average.
- Keep missing dates visible. Exchange closure, provider outage, absent contract identity and failed data checks receive distinct reason codes.
- Prevent overlap. One date cannot be treated as an independent observation in several overlapping horizons without dependence-aware inference.
- Reserve later years. Do not use every historical cycle for both discovery and confirmation.
Dated-contract path
Build Returns Without Turning the Roll Into Seasonality
6M is a dated futures contract. Expiry migration recurs on the calendar, so a continuous-series splice can masquerade as a seasonal move. Calculate tradeable paths from dated prices and model any contract change as an actual close-and-reopen transaction.
Select ex ante
Choose the active expiry with information available before the interval begins.
Anchor prices
Use declared trade, quote, or settlement fields and exact timestamps; do not mix price types.
Execute rolls
Charge two sides, spread, slippage, fees, missed fills and any exposure gap.
Reconcile views
Compare actual dated returns with documented adjusted analytical series.
Directional outcome
Signed return
Freeze quote direction and price anchors. Because 6M is quoted in U.S. dollars per Mexican peso, signs cannot be borrowed casually from a pesos-per-dollar cash chart.
Movement outcome
Absolute return or range
Measure magnitude separately from sign. A calendar window with historically larger movement, if established, would still not select long or short.
The CME FX Product Guide and Rulebook Chapter 256 anchor the current instrument and dated-contract context. Use the canonical 6M specification guide for detailed mechanics.
Calendar label versus recurring cause
Control Events, Holidays, Regimes, and Composition
A month or weekday can proxy for scheduled releases, policy meetings, recurring roll activity, holidays, changing liquidity, or one concentrated crisis. Stratify or match these conditions before assigning the effect to the calendar label itself.
Official events
Mexico and U.S. releases
Timestamp Banxico, INEGI, Federal Reserve, BLS, and BEA publications from official records. Calendar timing does not imply response direction.
Trading calendar
Holidays and shortened sessions
Version CME schedules and local holidays. A shorter or thin session changes exposure time and execution opportunity.
Market state
Volatility and liquidity
Use past-only state definitions, sufficient observations, and separate tails. Do not invent a regime boundary after seeing the seasonal outcome.
Contract state
Roll and expiry proximity
Report outgoing and incoming participation, days to termination, and the rule that selected the analyzed contract.
Mechanism evidence should be observable and distinct from the return. For example, an event-timing hypothesis should show that the named releases actually cluster in the candidate window and that the price effect remains after event-matched comparison. A plausible story without those steps stays a hypothesis.
Publish distributions, not slogans
Estimate Effect Size, Tails, and the Selection Penalty
Report the full conditional and matched-baseline distributions: sample counts, medians, means if justified, quantiles, dispersion, drawdowns for trading applications, and dependence-aware intervals. A p-value alone does not show whether the effect is economically meaningful or stable.
| Requirement | Publish | Fail when |
|---|---|---|
| Family disclosure | Every month, weekday, horizon, direction, filter and exclusion tried | Only the selected cell is shown |
| Effect size | Candidate-minus-baseline difference in natural and risk units | Significance has no practical magnitude |
| Uncertainty | Year- or episode-blocked interval and sample support | Rows are treated as independent despite shared paths |
| Tails | Worst episodes, quantiles, drawdown and concentration by year | A few years create the mean |
| Costs | Spread, slippage, fees, rolls, misses and stress | Executable result disappears under reasonable friction |
NIST’s stationarity discussion notes that stable mean, variance, and autocorrelation structure are assumptions in many time-series methods and that seasonality is typically modeled explicitly. In markets, changing regimes make stability an empirical question rather than a default.
Later cycles decide
Validate Out of Sample and Across Reasonable Definitions
Develop the hypothesis on early years, make one controlled selection on later validation years, then freeze code and open a final untouched period once. A seasonal claim should also survive adjacent windows and a second defensible roll rule without being retuned.
Shift the start and end by a small declared amount. A one-day optimum surrounded by reversals is fragile.
Require direction and useful magnitude to persist rather than averaging a structural break.
Repeat with a second ex-ante contract-selection rule and dated execution.
Show whether the calendar label adds information beyond scheduled events.
Delay entry, widen costs, and preserve unfilled orders for any trading application.
Predefine expiry, drawdown, drift, and revalidation limits; do not wait for a story after decay.
The CFTC advisory on commodity trading systems explains that simulated results can benefit from hindsight and do not reproduce actual financial risk. Even a completed seasonal backtest remains hypothetical evidence.
Ending matrix
Falsification Matrix for 6M Calendar Claims
Fill the matrix before examining the final holdout. A blank required cell means the claim is not accepted.
| Candidate family | Must beat | Must survive | Reject when |
|---|---|---|---|
| Month of year | All other months or a predeclared matched baseline | Later years, adjacent month boundaries, event and roll controls | Sign reverses, interval is uninformative, or few years dominate |
| Weekday | Clock-matched other weekdays | Holiday removal, event matching, DST-safe clocks | Effect is an event or shortened-session proxy |
| Turn of month | Same-length ordinary business-day windows | Nonoverlap, alternate business-day definitions, later cycles | Window is redrawn or relies on one horizon |
| Holiday relation | Comparable nonholiday sessions with equal exposure time | Named holiday definitions, schedule versions, liquidity stress | Missing or early-close data create the result |
| Trading application | No-trade and a simple risk-matched rule | Dated fills, costs, latency, rolls and prospective limits | Gross pattern lacks net, executable value |
No original study is reported. No original 6M seasonal sample, month ranking, weekday effect, holiday effect, return estimate, significance test, or trading result is reported here. The table is a protocol. Until a registered study passes it on sealed later data, the evidence state is untested.
Sources, methods and editorial disclosure — reviewed August 13, 2026
- CME Group Mexican Peso futures contract page, 2026 FX Product Guide, and CME Rulebook Chapter 256 for current product, quotation and dated-contract context.
- CME DataMine historical-data catalog for available official historical data categories. No market dataset was purchased or downloaded for this article.
- NIST/SEMATECH guidance on multiple comparisons for comparison-family control.
- NIST/SEMATECH guidance on stationarity and time-series analysis for temporal dependence and structural-stability context.
- Banco de México 2026 publication calendar, INEGI 2026 statistical release calendar, Federal Reserve FOMC calendars, BLS release calendar, and BEA release schedule for event-control design.
- CME Group trading-hours and holiday notices for contemporaneous session and holiday controls.
- CFTC advisory on commodity trading systems and hypothetical performance for simulation and hindsight limits.
Sources and methods were reviewed August 13, 2026. This page is unsponsored editorial analysis and presents a seasonality-testing protocol, not original 6M findings or forecasts.