Mean compounded return, by session
Average across every ticker's full-year compounded return for each strategy. Bars below zero mean the average name lost money running that strategy for the entire year.
Year by year
The final fifteen minutes
Same test, but the "close" leg exits at 19:45 instead of the actual last print at 20:00 — 15 minutes earlier, before the closing auction. Compare against the full-session numbers above.
Exiting 15 minutes early helped the intraday leg in both years (2022: −3.62% → −1.92%; 2023: +10.63% → +11.29%) and hurt the overnight leg in both years (2022: −9.66% → −11.33%; 2023: +4.29% → +3.90%). The last 15 minutes of the session behave like a small, consistent giveback.
Does yesterday's volume matter?
Each ticker-day's total volume is ranked against that same ticker's own year and split into quintiles. Each day's quintile is paired with the overnight return that follows it — close that day (19:45) to next day's open, pooled across all 503 tickers and both years.
The relationship is real but faint: pooled Pearson correlation between volume percentile and next-night return is r ≈ 0.013 (n = 248,164) — volume explains well under 1% of the night-to-night variance. The one consistent pattern is at the extreme: a ticker's top 20%-volume days are followed by mean overnight returns of +0.037% and a 52.1% win rate, both a shade better than every other quintile, in both years independently.
Prices here are not split-adjusted. A handful of overnight figures are artifacts of real corporate actions rather than trading losses or gains — most visibly GOOG, whose close on 2022-07-15 ($2,255.34) meets its open on 2022-07-18 ($113.44) across that stock's 20-for-1 split, registering as a −95% "overnight return" for one night. Sort the table by overnight return to spot similar jumps before drawing conclusions from the tails.
Every ticker, both years
Click a column to sort. Search filters by ticker symbol.