Stats › Seasonality and calendar › Calendar anomalies
Calendar anomalies and what happened to them
For decades researchers kept finding patterns tied to the calendar. Monday returns were reliably bad. The days around the turn of the month were unusually good. The session before a public holiday was better still. None of these had a good explanation and all of them showed up in the data, which made them interesting and slightly embarrassing for the idea that markets are efficient.
The interesting question isn't whether they existed. It's what happened to them after somebody published a paper about them. The weekend effect is the cleanest case anybody has: the same regression, the same index, run on the years before publication and the years after, and the result changes from overwhelming to nothing. Then there's a study that measured the same phenomenon across 97 published predictors and put a number on how much of the decay is caused by people simply reading the paper.
So this page covers the weekend effect, the turn of the month effect and the holiday effect, what the original papers actually claimed as opposed to what gets attributed to them, and the evidence on post publication decay. I put it together because the lesson generalises well beyond calendars: it's really a page about what happens to any edge once it's written down. A fair amount of what's usually quoted here couldn't be traced to a primary source, and where that happened I've said so in the notes and left the figure out rather than passing it along. Sources and dates are at the bottom, and several are free to read.
The weekend effect is the cleanest natural experiment in finance. French measured a reliably negative weekend return over 1953 to 1977 with a coefficient of minus 0.0023 and a t statistic of minus 8.86. Re-estimated over 1978 to 2002, after publication, the coefficient was minus 0.0005 with a t statistic of minus 1.37, no longer distinguishable from any other day. The general result is worse than that one case. Across 97 published return predictors, portfolio returns were 26% lower out of sample and 58% lower after publication, implying about 32% of the decay is caused by people reading the paper.
Members only
The rest of this page is for members
Below this point there are 9 sections, 1 chart and 4 named sources, roughly 2100 words of it. Every figure carries the source it came from and the date the data is from.
You can keep browsing every statistic in the library for free. The intro and the summary are always open.