Seasonality

Definition

Seasonality claims assert that returns depend on the calendar — month-of-year, turn-of-month, day-of-week, holiday effects, and commodity crop/heating cycles. It is the KB’s designated small-sample minefield: a century of data contains only ~100 independent observations of any annual pattern, calendar partitions are legion (making data mining nearly free), and several famous effects faded after publication. The entry’s job is to hold seasonal claims to a higher evidentiary bar, not to deny the physical cases.

How it works / structure

• The claim taxonomy: STATISTICAL seasonals (equity month-of-year patterns — the Halloween/sell-in-May effect documented across countries by Bouman-Jacobsen 2002; January small-cap lore; turn-of-month) vs PHYSICAL seasonals (agricultural crop cycles, natural-gas heating demand — mechanisms exist, instrument-agricultural-futures) vs STRUCTURAL calendar events (options expiration, index rebalances — real flows on known dates, event-opex, event-index-rebalance, covered as events not seasons).
• The counter-evidence: Sullivan-Timmermann-White (2001) showed that mining the space of calendar rules produces “significant” effects by construction — after multiple- testing adjustment, calendar effects in their study lost significance; several classic anomalies (January effect) attenuated sharply after publication.
• Engine discipline (executable): any seasonal parameter carries its instance count (a 30-year monthly pattern is n=30), its multiple-testing context (how many calendar partitions were searched), and a mechanism tag (physical / flow / none) — “none” caps position size at exploratory.

When it applies

Physical commodity cycles (the defensible core); structural flow dates (as events); statistical seasonals only as secondary tilts with the discipline above. The platform never gates a primary thesis on a statistical seasonal alone.

Risk profile & failure modes

• Small-n false confidence: annual patterns cannot reach large samples in one lifetime; the standard error never shrinks the way daily-signal errors do.
• Publication decay: documented seasonals attract capital that erodes them (lens-quantitative decay applies with extra force to famous calendar lore).
• Mechanism-free drift: seasonal stories attached post-hoc to mined patterns (bias-recency dressed in a calendar); the mechanism tag exists to force the distinction.
• Regime override: any macro event steamrolls a seasonal tilt; seasonal positions carry full market risk while waiting for a small average effect.

Evidence & limits

Bouman-Jacobsen (2002) is the strongest peer-reviewed statistical seasonal (documented across 36 of 37 countries in their sample; debated and partially attenuated since); Sullivan-Timmermann-White (2001) is the discipline. Physical commodity seasonality has mechanisms and exchange-documented patterns with the same small-n caveat on any specific trade. Net platform stance: mechanism-backed seasonals are usable with size honesty; mined calendar lore is folklore.

Falsifiable-thesis examples

Illustrations only, not signals:

• “The index’s November-April return will exceed its May- October return in at least 6 of the next 10 years (Halloween-effect persistence)” — falsified by the decade’s tally.
• “Natural gas front-month IV will rise into November in at least 7 of 10 replay years (heating-season mechanism)” — falsified by the IV seasonal record.

Cross-references

• The discipline source: lens-quantitative (multiple testing, decay)
• The defensible core: instrument-agricultural-futures, instrument-energy-futures, strategy-futures-calendar-spread
• Structural calendar flows (events, not seasons): event-opex, event-index-rebalance
• The bias behind mined patterns: bias-recency