The Calendar Effects Are Real and Still Useless

Scope: measured on the ETFs listed (SPY, QQQ, IWM, DIA, GLD, SLV). Findings apply to the instruments measured and do not automatically transfer to other markets.

Thursday has the widest median session in both SPY and QQQ — and knowing that tells you nothing about next Thursday. The median is 1.018% against Monday's 0.873% on SPY, an ordering that repeats independently on QQQ. But the middle half of Thursdays runs 0.701% to 1.419%, and the middle half of Mondays runs 0.632% to 1.305%. The two overlap almost completely.

The ordering is real#

Two instruments, measured separately, put the days in nearly the same order. That is not what noise usually does.

2025-02-10 to 2026-08-07, regular session only (09:30–16:00 ET). Range = (session high − session low) ÷ session open, per day, then the median across days.

Note what this already kills: Friday is not the wild one. It ranks third on SPY and second on QQQ. The folk version of this — that Friday carries the week's volatility — does not survive contact with the data on either instrument.

And it is still useless#

Here is the same SPY data with the distribution left in rather than collapsed to a median.

Same window and definition. n = 72–78 sessions per weekday.

A quiet Thursday (0.701%) is calmer than a typical Monday. A busy Monday (1.305%) is wider than a typical Thursday. The gap between the weekday medians is 0.16 percentage points; the spread within a single weekday is 0.7. The between-day signal is roughly a fifth the size of the within-day noise.

So both statements are true at once: the ordering is probably real, and it is worth nothing on any particular day. The mistake is not believing in the effect — it is believing an effect that survives averaging will survive being applied to tomorrow.

Month-end: not measurable here, and worth saying so#

The last trading day of the month has a median SPY range of 0.962% against 0.921% for every other day. QQQ shows 1.432% against 1.278%.

Both point the same way, and neither is a result: there are 19 month-end sessions in this window against 356 ordinary ones. A difference this small on a sample this size is indistinguishable from chance, and no amount of restating it changes that. The honest label is undetermined — not "no effect".

That is itself the practical point. A calendar effect that occurs twelve times a year needs years of data to test. The rarer the pattern, the longer you must wait to learn whether it was ever there — and the more confidently it tends to get asserted in the meantime.

What to take from this#

Averages compress. A weekday median is built from ~75 sessions that ranged from calm to violent, and the number that comes out the other end describes none of them. Before acting on any statistic of this shape, ask for the spread — if nobody publishes the spread, the ordering is all you have, and an ordering is not a forecast.

What this does not say#

It does not say weekday effects are absent — the cross-instrument agreement suggests something is there. It says the effect is too small relative to daily variation to guide a position.

It also measures range, not direction. A wide day is not a profitable day; these numbers say nothing about which way price went.

And this is one 18-month window. A different regime could reorder the days entirely, which is precisely the kind of claim that needs re-testing rather than citing.

Related reading#

• Best time of day to buy an ETF — where inside the session that range gets built.
• The overnight gap is mostly over before you see it
• Key levels on the chart