AI Research TSM

TSM top-decile overnight gap-ups: intraday (open→close) fade vs. baseline over ~3 years

TSM’s biggest pre-market jumps mostly lose their steam during regular hours — but only modestly and not conclusively. We isolated the top‑decile overnight gap-ups over the last ~3 years and compared the session open→close return on those days against the everyday baseline to see whether the pre-market pop is a launchpad or an exit.

On the 74 top‑decile gap days the average open→close return was about −0.3641%, and the cross-sample statistics give a suggestive lean toward intraday selling (p ≈ 0.108) rather than a knockout result. The full minute-bar methodology, distributions, charts and robustness checks are below for the reader who wants the detailed evidence.

The research question

For TSM over the past ~3 years, do overnight gap-ups get bought or faded during the session — after the biggest gap-up opens (top-decile overnight move), is the regular-hours open-to-close return above the everyday baseline or does the day session systematically give it back? Thesis: big gap-ups get sold through the day, so open-to-close returns following top-decile overnight jumps come in below the everyday baseline and often turn negative — the pre-market pop is an exit, not a launchpad.

How this was measured

From minute bars, we isolate US regular trading hours (Mon–Fri, 09:30–16:00 ET). For each session, we take the first RTH minute's open as the session open and the last RTH minute's close as the session close. Overnight gap = open_today / close_yesterday − 1. Day-session return = close_today / open_today − 1. We rank overnight gaps and define 'top-decile gap-ups' as days with gap ≥ the 90th percentile of gap_ret. We then compare the day-session return distribution on those top-decile days against the everyday baseline (all days and the non-event 90%) and quantify the mean gap via a Welch two-sample t-test (top-decile vs other-90%). Window: 2023-07-26 to 2026-06-30 (last ~3 years subject to data availability).

The key numbers

Sessions analyzed
735
2023-07-26 to 2026-06-30
Top-decile threshold (gap_ret ≥)
2.2321%
Overnight open vs prior RTH close
Top-decile count
74
Days with gap ≥ 90th percentile
Mean open→close (top-decile)
-0.3641%
N=74
Mean open→close (all days)
-0.0077%
N=735
Mean open→close (other 90%)
0.0322%
N=661
Edge vs all (top − all)
-0.3564%
Edge vs other 90% (top − non)
-0.3963%
Fraction negative (top-decile)
58.1081%
Share of top-decile days with open→close < 0
Welch t-stat (top vs other 90%)
-1.623
Two-sample unequal-variance
Welch p-value (two-sided)
0.1081
p=0.1081 ≥ 0.05 → no statistically-clear difference

Reading the numbers

735 sessions; top‑decile overnight gap threshold = 0.022320521232497947 (74 days). Top‑decile mean open→close = -0.003640937916782134 vs all = -7.689024649416119e-05; 58.108% of top days close down, but the difference vs the rest is not statistically clear (p=0.1081).

The charts

TSM open→close returns after top-decile gap-ups
What this chart says

This histogram plots the distribution of open→close returns on the 74 top‑decile gap‑up days. The bulk of observations cluster around a slightly negative center (mean = -0.003640937916782134) with a left tail reaching about -0.0702 and a right tail to +0.0365. Look at how many bars sit below zero — 58.1% of these days close down — which shows the pre‑market pop is often given back, though there are still several positive closes.

Mean open→close return: top-decile vs baselines
What this chart says

The bar chart compares mean open→close returns: top‑decile = -0.003640937916782134, all days = -7.689024649416119e-05, other 90% = 0.0003221105516923895. The top‑decile average is a few tenths of a percent below the baselines (edge vs all ≈ -0.003564, edge vs other ≈ -0.003963), so on average big gap‑ups are faded intraday by that magnitude. Note the Welch t‑stat = -1.623 and two‑sided p = 0.1081, which means this difference isn't statistically clear at conventional thresholds.

Overnight gap vs same-day open→close return (all sessions)
What this chart says

The scatter shows every session's overnight gap (x, mean = 0.0026, range -0.1116 to 0.0844) against the same‑day open→close return (y, mean = -0.0001, range -0.0702 to 0.1316). Points are widely dispersed with no obvious upward slope: large positive overnight moves (up to 0.0844) map to both negative and positive intraday returns. That visual spread supports the idea that big premarket jumps do not reliably convert into further intraday gains and are frequently, but not always, faded.

Open→close return summary by group

groupNmeanmedianstdfraction_negative
Top-decile gap-ups74-0.0036-0.0020.02010.5811
Other 90%6610.000300.01850.4977
All days735-0.0001-0.00020.01870.5061

The takeaway

Short answer: there is a modest intraday fade after TSM’s biggest overnight gap-ups, but the evidence is only a suggestive lean, not a knockout. On the 74 top‑decile gap-up days the average open→close return was about −0.36%, versus +0.03% for the other 90% of days, giving a top‑minus‑non-event gap of roughly −0.40%. A slim majority (58.1%) of those big‑gap days finished down intraday compared with about 49.8% on non‑events. Statistically this is not definitive: the Welch test gives p≈0.108 (about an 11% chance this pattern is sampling noise) and there are only 74 event days, so treat the result as a suggestive tendency rather than a reliable rule. Practically, the average fade is small relative to intraday dispersion (std ≈2%), so execution costs or a few outliers could wipe out the edge; use this as a cautionary signal, not a standalone trade trigger.

The fine print