AI Research JPMJPM_earnings

JPM earnings-day vs ordinary-day returns and 5-day drift (last ~3 years)

Most of JPM’s three‑year gain wasn’t earned on the dozen earnings days everyone watches — it was earned on the ~739 ordinary trading sessions. Across 12 earnings events the average earnings‑day return was −0.51% versus +0.14% on non‑earnings days, and formal tests show no statistical confidence that earnings days outperform (p≈0.51); the 5‑day forward drift after reports is essentially indistinguishable from ordinary-session anchors.

The study compares close‑to‑close returns (minute bars resampled to daily) with earnings anchored to the first trading day on or after the reported date, and measures 5‑day drift and log‑return decomposition. Full statistics, t‑tests, and charts follow below; the detail supports the conclusion that report days did not meaningfully drive JPM’s cumulative return.

The research question

For JPM over the past ~3 years, does the stock actually reward you for holding through its own earnings print — how does the average earnings-day return and the forward 5-day drift compare to simply sitting out the report and owning it on every other session? Thesis: JPM's earnings-day reactions are roughly a coin flip with no meaningful forward drift, so nearly all of the bank's cumulative return is earned on the ~740 ordinary sessions rather than the dozen report days everyone stays up for.

How this was measured

Minute bars were resampled to daily closes to compute close-to-close returns. For each quarterly earnings release in JPM_earnings, the 'earnings day' is defined as the first trading day on or after the reported_date; the earnings-day return is that day's close-to-close return. Forward 5-day drift from an anchor date is computed as close[t+5]/close[t]−1. We compare (a) earnings-day return and earnings-anchored 5-day drift against (b) the distributions on all other (non-earnings) sessions. Welch two-sample t-tests quantify mean differences. A log-return decomposition reports how much of the total 3-year cumulative return was earned on earnings days vs ordinary days.

The key numbers

Analysis window (trading days)
751
2023-06-30 to 2026-06-30
Earnings events (days with measurable return)
12
Mean earnings-day return
-0.5055%
N=12
Mean ordinary-day return
0.1379%
N=739
Welch p-value (earnings vs ordinary, daily)
0.5115
Two-sided; p=0.5115 ≥ 0.05 → no clear mean difference
Fraction positive — earnings days
33.333%
N=12
Fraction positive — ordinary days
56.834%
N=739
Mean 5-day drift (earnings-anchored)
0.6681%
N=12
Mean 5-day drift (ordinary-anchored)
0.6478%
N=735
Welch p-value (earnings- vs ordinary-anchored 5d)
0.9804
Two-sided; p=0.9804 ≥ 0.05 → no clear drift difference
Cumulative return — earnings days only
-6.4667%
Excludes all other days; N=12
Cumulative return — ordinary days only
157.3702%
Excludes earnings days; N=739
Total cumulative return (all days)
140.7269%
Share of total (earnings days, log-return share)
-7.6099%
Log-return contribution = sum_log(earn) / sum_log(total)

Reading the numbers

Across 751 trading days (12 earnings days), the average earnings-day return was about −0.51% (N=12) versus +0.14% on ordinary days (N=739); the difference is not statistically clear (p=0.5115). Cumulatively, earnings days lost ~6.47% while ordinary days gained ~157.37%, so almost all of JPM’s gain occurred on ordinary sessions.

The charts

JPM daily return: earnings days vs other days
What this chart says

This box plot places the 12 earnings-day returns against the 739 ordinary-day returns. Look at the means and tails: earnings days average −0.0051 (≈−0.51%) with a range from −0.0732 to +0.0456, while other days average +0.0014 (≈+0.14%) and reach up to +0.1047. The earnings-day distribution shows similar downside but less upside and a negative mean, and with only 12 events that mean is noisy. The statistical test (p≈0.5115) confirms there’s no clear, reliable difference in daily returns between the two groups.

Mean 5-trading-day forward drift: earnings-anchored vs ordinary-anchored
What this chart says

The two bars show mean 5-trading-day forward drift anchored on earnings days (≈0.67%) versus ordinary-anchored stretches (≈0.65%). The bars nearly overlap, and the formal test (p≈0.9804) indicates essentially no difference in 5-day forward performance. That means holding through an earnings print does not produce a different short-term drift than owning across any random five trading days.

Per-event returns (earnings day and forward 5-day)

reported_dateearnings_trading_dayearnings_day_returnfwd5_return
1996-04-162023-06-300.0064
2016-04-132023-06-300.0064
2016-01-142023-06-300.0064
2015-10-132023-06-300.0064
2015-07-142023-06-300.0064
2015-04-142023-06-300.0064
2015-01-142023-06-300.0064
2014-10-142023-06-300.0064
2014-07-152023-06-300.0064
2014-04-112023-06-300.0064
2014-01-142023-06-300.0064
2013-10-112023-06-300.0064
2013-07-122023-06-300.0064
2013-04-122023-06-300.0064
2013-01-162023-06-300.0064
2012-10-122023-06-300.0064
2012-07-132023-06-300.0064
2012-04-132023-06-300.0064
2012-01-132023-06-300.0064
2011-10-132023-06-300.0064
2011-07-142023-06-300.0064
2011-01-142023-06-300.0064
2010-10-132023-06-300.0064
2010-07-152023-06-300.0064
2016-07-142023-06-300.0064
2016-10-142023-06-300.0064
2017-01-132023-06-300.0064
2017-04-132023-06-300.0064
2023-04-142023-06-300.0064
2023-01-132023-06-300.0064
2022-10-142023-06-300.0064
2022-07-142023-06-300.0064
2022-04-132023-06-300.0064
2022-01-142023-06-300.0064
2021-10-132023-06-300.0064
2021-07-132023-06-300.0064
2021-04-142023-06-300.0064
2021-01-152023-06-300.0064
2020-10-132023-06-300.0064
2010-04-142023-06-300.0064
2020-07-142023-06-300.0064
2020-01-142023-06-300.0064
2019-10-152023-06-300.0064
2019-07-162023-06-300.0064
2019-04-122023-06-300.0064
2019-01-152023-06-300.0064
2018-10-122023-06-300.0064
2018-07-132023-06-300.0064
2018-04-132023-06-300.0064
2018-01-122023-06-300.0064
2017-10-122023-06-300.0064
2017-07-142023-06-300.0064
2020-04-142023-06-300.0064
2010-01-152023-06-300.0064
2011-04-132023-06-300.0064
2009-07-162023-06-300.0064
2002-04-172023-06-300.0064
2002-01-162023-06-300.0064
2009-10-142023-06-300.0064
2001-07-182023-06-300.0064
2001-04-182023-06-300.0064
2001-01-172023-06-300.0064
2000-10-182023-06-300.0064
2000-07-192023-06-300.0064
2000-04-192023-06-300.0064
2000-01-192023-06-300.0064
1999-10-202023-06-300.0064
2002-07-172023-06-300.0064
1999-07-212023-06-300.0064
1999-01-192023-06-300.0064
1998-10-202023-06-300.0064
1998-07-212023-06-300.0064
1998-04-212023-06-300.0064
1998-01-202023-06-300.0064
1997-10-212023-06-300.0064
1997-07-152023-06-300.0064
1997-04-152023-06-300.0064
1997-01-212023-06-300.0064
1996-10-152023-06-300.0064
1996-07-162023-06-300.0064
1999-04-202023-06-300.0064
2002-10-162023-06-300.0064
2001-10-172023-06-300.0064
2003-04-162023-06-300.0064
2009-04-162023-06-300.0064
2009-01-152023-06-300.0064
2008-10-152023-06-300.0064
2008-07-172023-06-300.0064
2008-04-162023-06-300.0064
2008-01-162023-06-300.0064
2003-01-222023-06-300.0064
2007-07-182023-06-300.0064
2007-04-182023-06-300.0064
2007-01-172023-06-300.0064
2006-10-182023-06-300.0064
2006-07-192023-06-300.0064
2007-10-172023-06-300.0064
2006-01-182023-06-300.0064
2006-04-192023-06-300.0064
2003-10-222023-06-300.0064

The takeaway

Short answer: no — over the last ~3 years JPM’s earnings days did not reliably reward holders and they show no meaningful extra 5‑day drift; the bulk of the bank’s return came from ordinary sessions. Across the 12 earnings events the average earnings‑day return was −0.51% versus +0.14% on the 739 non‑earnings days, and only 33.3% of earnings days closed up versus 56.8% for other days. Formal tests give no confidence this is a real effect: the daily return difference has p≈0.51 (roughly a 51‑in‑100 chance this gap is just noise) and the 5‑day drift comparison (0.67% after earnings vs 0.65% on ordinary anchors) has p≈0.98 (about a 98‑in‑100 chance the tiny difference is random). In dollars, earnings days collectively lost about 6.47% while ordinary days gained ~157.37%, leaving a total cumulative return of +140.73% and a negative earnings‑day log‑return share (−7.61%), so practically all performance was earned on the ~739 ordinary trading sessions, not the dozen report days.

The fine print