XOM volume-confirmed strength on weak Brent: forward 10-session XOM-vs-XLE alpha
The decoupling trade doesn't survive the data. Over the past three years, XOM had 28 days when it closed higher while Brent fell and volume hit the top quintile of its prior 20 sessions. If volume-confirmed strength on a weak crude tape signaled idiosyncratic accumulation, those days should have led to XOM outperforming XLE over the next 10 sessions. They didn't.
The average forward XOM-vs-XLE alpha on those days was -0.28%, with a -0.51% median. XOM beat XLE only about 39% of the time. Both names rose on average, but XLE captured more of the upside. That's slightly worse than picking an ordinary day, where the edge was essentially flat.
The full analysis below breaks down the distribution, tests, and context.
Over the past ~3 years, when XOM's daily volume is in the top quintile of its prior 20 sessions and XOM closes higher on a day Brent crude closes lower, does XOM outperform XLE over the next 10 trading sessions? I expect volume-confirmed strength on a weak crude tape to reflect idiosyncratic accumulation, so XOM decouples from crude beta and leads the energy sector.
How this was measured
Daily bars were built by resampling XOM and XLE minute data to session closes and summing XOM volume; Brent was reindexed onto the equity calendar with forward-fill. A signal day is one where XOM closes higher, Brent closes lower, and XOM volume reaches the 80th percentile of its previous 20 sessions. The 20-session volume window is shifted back one bar so the current day cannot set its own threshold. Forward returns are measured close(t) to close(t+10) for XOM and XLE, and the research metric is XOM minus XLE. The conditional sample is compared against all eligible 10-session alpha observations, against zero via a one-sample t-test, and against non-signal days via Welch's t-test.
The key numbers
Reading the numbers
On the 28 days that fit the setup, XOM trailed XLE by about 0.28% on average over the next 10 sessions, versus a tiny +0.04% edge on all other days, and only about 39% of signals were winners. The t-stat of -0.86 with p=0.396 means that gap is well within random noise, so the volume-confirmed decoupling idea did not show up clearly.
The charts
This side-by-side box plot compares the forward 10-session XOM-minus-XLE return on the 28 signal days against all 742 eligible days. The signal-day box sits lower, with a mean of -0.28% versus +0.04% for the broad group, meaning the typical outcome after the setup was mild underperformance rather than the outperformance you expected. Note how much taller the all-day box is: it spans roughly -6.8% to +5.8%, so its wider spread shows a 28-day sample carries far more noise than the full history.
This histogram shows every one of the 28 signal-day outcomes, from the worst at about -2.9% to the best at about +3.5%. The mass is tilted to the left of zero, which is the visual version of the negative mean and the 39% win rate. The single biggest bar is a loss, and the largest gain is only one observation, so the left-leaning shape is what drives the negative average.
This bar chart breaks the 10-session outcome into its parts: after signal days XOM averaged +0.91% and XLE averaged +1.19%, leaving a -0.28% XOM-minus-XLE gap. The final bar, the all-day alpha of +0.04%, is essentially zero, so the sector and the stock move together almost one-for-one outside the signal. That is the crux for your question: the setup did not produce the idiosyncratic XOM leadership you were looking for; if anything, XOM lagged the sector slightly.
Most recent 20 qualifying XOM signal days
| signal_date | xom_ret | brent_ret | xom_fwd10 | xle_fwd10 | alpha_fwd10 |
|---|---|---|---|---|---|
| 2024-09-09 | 0.0227 | -0.0062 | 0.0189 | 0.035 | -0.0161 |
| 2024-09-23 | 0.0177 | -0.0133 | 0.0687 | 0.0509 | 0.0178 |
| 2024-09-27 | 0.0269 | -0.0196 | 0.066 | 0.0657 | 0.0003 |
| 2024-11-06 | 0.0144 | -0.006 | 0.0024 | 0.0295 | -0.0271 |
| 2024-12-10 | 0.0001 | -0.0012 | -0.0583 | -0.0564 | -0.0019 |
| 2024-12-20 | 0.0053 | -0.0076 | 0.028 | 0.0562 | -0.0281 |
| 2025-02-28 | 0.0081 | -0.0033 | 0.0044 | -0.0142 | 0.0186 |
| 2025-03-10 | 0.024 | -0.0195 | 0.0384 | 0.0677 | -0.0293 |
| 2025-06-10 | 0.0193 | -0.0047 | 0.0105 | 0.0022 | 0.0083 |
| 2025-06-16 | 0.0111 | -0.0125 | -0.0376 | -0.0264 | -0.0113 |
| 2025-07-31 | 0.0085 | -0.0074 | -0.042 | -0.0184 | -0.0236 |
| 2025-08-13 | 0.0123 | -0.0082 | 0.0565 | 0.0419 | 0.0146 |
| 2025-11-04 | 0.0011 | -0.0114 | 0.0521 | 0.0482 | 0.004 |
| 2025-11-06 | 0.0016 | -0.002 | 0.0341 | 0.0051 | 0.029 |
| 2025-12-09 | 0.0204 | -0.0107 | 0.0071 | -0.0183 | 0.0253 |
| 2026-02-27 | 0.029 | -0.0047 | 0.0247 | 0.0294 | -0.0047 |
| 2026-04-08 | 0.0144 | -0.1165 | -0.0446 | -0.0274 | -0.0172 |
| 2026-05-19 | 0.0117 | -0.0179 | -0.0532 | -0.0435 | -0.0097 |
| 2026-06-30 | 0.0042 | -0.0158 | 0.0572 | 0.0626 | -0.0054 |
| 2026-07-28 | 0.0022 | -0.0687 | 0.0227 | 0.045 | -0.0223 |
The takeaway
No — the decoupling thesis didn't show up. On the 28 days in the past three years when XOM closed higher on weak Brent with top-quintile volume, XOM actually gave up ground to XLE over the following 10 sessions: average alpha of -0.28%, with the typical case worse at a -0.51% median. XOM only beat XLE in about 39% of those instances, and both names rose on average — XOM +0.91% versus XLE +1.19% — so the sector captured more of the upside than XOM did. For context, XOM's edge over XLE across all 742 eligible days was essentially flat at +0.04%, which means the signal days were slightly worse than picking a day at random, not better. The statistics confirm there's nothing here to trade: p = 0.40 against zero and p = 0.31 versus non-signal days, both far from meaningful, and the signal's average alpha sits at just the 41st percentile of the everyday distribution. With a mere 28 events, this is inconclusive with a negative tilt rather than a proven anti-signal — but either way, volume-confirmed strength on a weak crude tape has not been a reliable tell for XOM leadership. Practical takeaway: treat this setup as noise; if anything, it has marked slightly better odds for XLE than XOM.
The fine print
- Only 28 signal days — far too few to separate a real effect from randomness, regardless of the direction.
- The 10-session forward windows overlap when signals cluster, so the tests behave as if there were more independent observations than there really are.
- XOM is a large XLE constituent, so XOM-vs-XLE compresses exactly the idiosyncratic move you're trying to isolate.
- Brent's lower close is only known with a T+1 lag, so a live version would have to delay or confirm the signal — and no costs or slippage are modeled.