Overnight vs Day Returns

Cumulative return paths

Each leg compounds its own simple returns independently across sessions.

wheel zoom · drag to zoom · double-click reset
applies to the basket close-to-close path
Overnightclose → next open
Dayopen → close
Close to closeclose → close

Return distributions

Empirical CDFs and inverse ECDF quantile functions; endpoints are preserved.

hover for probability and return
Overnightempirical CDF
Dayempirical CDF
Close to closeclose-to-close return
Overnightquantile function
Dayquantile function
Close to closeclose-to-close return

Monthly returns

Compounded basket returns by calendar month for overnight, day, and close-to-close paths. Select a month to inspect daily returns.

Evidence & shape

Per-session means with 95% t-intervals, distribution shape, and where compounded log growth comes from.

Overnight − day, paired by sessionmean difference with 95% interval
TickernMean diff95% intervalt
Log-growth attributionΣ log(1+r) share, overnight vs day
SeriesSkewExcess kurtosis|r−μ|>3σMax drawdownOvernight log share

Over time

Mean per-session return by calendar year, by weekday, and on a rolling 252-session window.

rolling chart: wheel zoom · drag to zoom · double-click reset
Mean return by weekdaybasket · per-session mean
Overnightrolling 252-session mean
Dayrolling 252-session mean

Normal reference

Sample quantiles against a fitted normal — straight lines would mean Gaussian returns. Deviations at the ends are the tails.

hover for quantile pairs
OvernightQ–Q vs normal
DayQ–Q vs normal

Summary

Means and medians are per-session returns, not annualized.

TickerDate rangeOvernight meanOvernight medianOvernight upDay meanDay medianDay upClose to close meanClose to close medianClose to close upOvernight 95% intervalDay 95% intervalClose 95% interval

95% intervals are t-intervals on the per-session mean (mean ± t·σ/√n); they are narrow by construction — with thousands of sessions, tiny means are statistically real without being economically large. Overnight returns span ex-dividend sessions, so dividend drops appear as overnight losses; the decomposition is price-only, not total-return.