Retesting seasonal rules when regimes change

Three checks, not one slogan

A popular calendar or diversification slogan can look tidy across a long sample and still fail in the windows that threaten a portfolio. Editorial reading: do not treat the slogan as one blended story. Split it into three separate checks.

The first check is an average-period test: what happens on typical days. The second is a large-move-day test: what happens on sessions with unusually large swings. The third is a rerun after the idea became common knowledge, which can show publication decay.

Average-period diversification and stress-day correlation

Pairs that show little long-run co-movement, including gold versus crude and stocks from different industries, often move in the same direction on days of very large equity swings. That pattern is stress-day correlation.

When a major industrial average falls 3% in a session, nearly all stocks tend to decline, so ordinary diversification does little to limit damage on that day. Average-period diversification infers risk reduction from low long-run co-movement measured across typical days rather than crash sessions. Editorial reading: the average-period test and the large-move-day test can disagree.

Demand labeled recession-resistant for tobacco, alcohol, and gambling is not guaranteed. Higher costs can cut consumption, and a 2008 credit review downgraded 17 casino companies as revenues weakened with the economy.

The geographic regime hedge

A geographic regime hedge is the assumption that non-domestic markets will offset a domestic decline because they occupy a different country basket. In the first half of 2008 a broad U.S. equity index fell 12.8%, a developed-world index fell 11.7%, and an emerging-markets index fell 12.7%, leaving little geographic offset in that window.

After currency adjustment, that same 2008 first-half window still showed declines in all major world markets except two. Editorial reading: a country-basket story that holds on average-period days can fail the large-move-day test in a shared decline.

Weekday seasonality and publication decay

Weekday seasonality is a calendar pattern in which days of the week differ in how often prices rise or in how much of the sample’s total change they contribute. A weekday test of a U.S. industrial average from 1900 found Monday closes higher only 35.8% of the time, while about half of the total points gained over 108 years still occurred on Mondays.

Extending a mid-1980s weekday study through June 2008 reversed the ranking: Mondays rose 57.6% of the time and Fridays 51.6% over 22 years. Publication decay is the fading or reversal of a documented seasonal ranking once the sample is extended past the original study. Editorial reading: frequency, contribution, and a later-window ranking are three different results. Do not fold them into one weekday slogan.

Outlier-driven seasonality and the calendar abstention rule

The late-spring through autumn window still produced gains more than 60% of the time. The weak seasonal average was largely the product of a few large-loss years rather than a typical six-month outcome. That is outlier-driven seasonality: a seasonal average that looks weak or strong mainly because a few extreme windows dominate the mean, not because most periods share that outcome.

A calendar abstention rule is a seasonal-trading procedure that stays out of the market during a fixed date window instead of holding through that window. A May-exit calendar rule did not beat a continuous hold in tests until 1987, a year that included a 22% single-day October drop, with later concentrated summer-to-autumn declines of about 17% and more than 20% also dominating the average.

Keep the three results apart

Editorial reading: an average-period test can make a calendar abstention rule or a geographic regime hedge look orderly. A large-move-day test can show stress-day correlation and outlier-driven seasonality instead. A rerun after the idea became common knowledge can show publication decay in weekday seasonality. Report the three checks separately so a long-sample mean does not hide the windows that actually threaten a portfolio.