[united893]

YOu're graphing a random walk with a single black swan event (2008 crash). A more honest graph would just have the day/night delta mapped out, not integrated in.

[vitus]

Indeed. I pulled the data and mostly reproduced the results when starting on Jan 1, 1990 (I got 1/20 of a penny for day trading, but "only" somewhere in the $600-650 range for overnight trading, depending on the choice of end date. Close enough to proceed with additional analysis, IMO.

If instead I pick my starting point as Jan 1, 2012, I see $1.15 for day trading vs $2.24 for overnight trading. (Note that there was again another black swan event in Feb-March 2020, before which point day trading was actually doing better than overnight trading.)

Tracking deltas (well, day-over-day multipliers) as you suggest since 1990 shows that the two shapes are qualitatively more similar (although day trading is more volatile):

- 0.1st percentile: day trading: 0.803 vs overnight trading: 0.871

- 1st percentile: day trading: 0.932 vs overnight trading: 0.960

- 99th percentile: day trading: 1.059 vs overnight trading: 1.052

- 99.9th percentile: day trading: 1.189 vs overnight trading: 1.222

[epistemer]

To say it is a random walk is confusing a model of reality with reality.

Totally agree with the other poster though. You can pretty much prove whatever you want in the market depending on the start date and window size.

I would think the driving factor is after market earnings releases, expectations for those releases and how many there randomly happen to be during the window in question.