Seasonality Effects

A seasonality effect is any pattern where the calendar seems to predict returns: stocks that do better in January, the market that supposedly limps through summer, the extra pop around the turn of the month, the Monday slump. These stories are irresistible because they're simple and testable, and a handful of them have survived decades of scrutiny. Many more are pure coincidence dressed up as a rule.

The honest starting point is Market Efficiency (The EMH): if a calendar edge were reliable and free to trade, arbitrageurs would buy the day before and sell into it until the pattern flattened. So a surviving seasonal effect usually needs a reason it can't be easily arbitraged — a real, recurring cash flow, a tax deadline, or a structural flow of money at fixed dates.

The classic patterns

• Turn-of-the-month. A disproportionate share of the market's gains has historically landed in the last day and first three days of each month, when pension contributions and index reinvestment hit.
• January / small-cap effect. Small stocks have tended to jump in early January, plausibly a rebound from December tax-loss selling.
• Sell in May ("Halloween indicator"). Bouman and Jacobsen found returns from November through April dwarfed May through October across many countries.
• Day-of-week and holidays. Mondays historically weak, the day before holidays historically strong.

Worked example: is the turn-of-the-month real?

Suppose over 40 years the market averaged 0.04% per trading day. You isolate the four "turn" days each month (last day plus first three) and find they averaged 0.16% per day, while the other ~17 trading days averaged 0.01%.

The turn days are four times the daily average, and there are 12 turns a year, so roughly $4 \times 12 = 48$ turn-days. Their total contribution is about

$$48 \times 0.16\% = 7.7\% \text{ per year},$$

against a full-year return near 10%. In other words, almost the entire annual gain clustered into one-fifth of the trading days, and the other four-fifths were roughly flat. That is the real, replicated finding — and it points to a mechanism (monthly retirement and index inflows) rather than magic, which is why it has held up.

Why most "seasonality" is a mirage

Here is the trap. There are 12 months, 5 weekdays, 21-odd trading days a month, plus holidays — hundreds of calendar slices. Test enough of them and, by pure chance, several will show "significant" returns. This is textbook multiple testing: at the 5% level you expect 1 in 20 random slices to look significant even when nothing is there (p-values and Multiple Testing). Sullivan, Timmermann and White showed that once you account for how many calendar rules were tried across the literature, most lose their statistical shine.

Even the effects that were once real tend to fade. The January small-cap effect has weakened markedly since it was published, the standard fate of a tradable inefficiency once everyone knows about it.

Seasonality is best understood as one slice of a broader habit — pulling apart a return series into recurring components (see Seasonality and Decomposition) and time-of-day patterns like Overnight vs Intraday Returns. Some of that structure is genuine and even tradable; the discipline is separating the few real calendars from the many that your search invented.