The two-week daylight saving trap

[Shadow Trader]

Each spring and autumn, the US and Europe change their clocks on different dates. In between, New York and London are five hours apart on some days and four on others. For roughly two to three weeks, the whole structure of the day shifts.

The New York open arrives an hour earlier relative to London. The overlap grows or shrinks. A data release that always hit during the London afternoon now lands during the lunchtime lull, or the other way around.

Strategies tied to clock times in server time suffer. A rule like "trade only from 13:00 to 16:00 server time" captures a different part of the market for those two weeks. Backtests rarely handle this correctly either, since many use fixed offsets.

Australia and New Zealand shift in the opposite direction, because it's spring there when it's autumn here. That adds another layer to Asian session timing, with the Sydney open moving by an hour relative to Tokyo.

A calendar reminder on the two shift weekends takes care of most of the damage. Check where the key session events are now, and update anything keyed to hour numbers.

Which session rule has been broken by a clock change at some point?

[Amanhasnoname]

The rule that broke on me was a simple one: no new trades after 18:00 server time. In normal weeks that lined up nicely with the point where New York activity fades out. During the spring gap between the US and European changes it suddenly cut me off while things were still active, and in the autumn it did the reverse and kept me trading into the dead part of the afternoon.

I didn't spot it for two seasons, which is a little embarrassing. What finally showed it up was a review where I sorted results by week of the year, and the same few weeks in March and late October kept coming out noticeably worse than the weeks around them.

I now write all my session rules relative to an event rather than a clock number, like an hour after the US cash open or the London close. Backtests got the same treatment, and I rebuilt the time filter so it reads the real session times for each date instead of a fixed offset.