XLE vs SPY after high-volume down-crude capitulation candidates
Volume-confirmed selling in energy looks, at first pass, like it marks something. Across 71 days when XLE closed lower alongside Brent and volume ran in the top quintile of its prior 20 sessions, the ETF beat SPY by 0.88% on average over the next 10 sessions — while the same down-crude setup without the volume filter gave back 0.42%.
That gap is the interesting part: the filter seems to separate forced de-risking from ordinary weakness. The problem is consistency. XLE outperformed in only 53.5% of those cases, and the median edge was 0.46%, so a handful of large winners carry the mean. The spread sits near the edge of significance, and 71 overlapping events is thin ground.
The full breakdown follows.
Over the past ~3 years, when XLE's daily volume is in the top quintile of its prior-20-session distribution and XLE closes lower on a day Brent crude also closes lower, does XLE outperform SPY over the next 10 trading sessions? I expect volume-confirmed selling in the energy ETF on a down-crude day to mark capitulation, so XLE leads SPY once the forced de-risking clears.
How this was measured
Daily XLE and SPY closes were built from minute bars, and daily volume was summed from XLE minute bars. A day was flagged as a confirmed event when XLE closed lower, Brent crude closed lower, and XLE volume exceeded the 80th percentile of the prior 20 sessions, excluding the current session. The outcome was the 10-session-forward differential, XLE 10-session return minus SPY 10-session return. Confirmed events were compared against unconfirmed down days and against all valid days using Welch two-sample tests; a one-sample t-test against zero tested whether the event-day differential itself was detectably positive.
The key numbers
Reading the numbers
After the 71 days where XLE fell on lower crude and heavy volume, XLE beat SPY by 0.88% on average over the next 10 sessions, versus a slightly negative -0.17% average for all days. But a p-value of 0.063 means that gap is only suggestive, not statistically clear, and the 53.5% win rate says it happened slightly more often than a coin flip.
The charts
This box plot lines up the 10-session XLE-minus-SPY gap for three buckets: the 71 volume-confirmed down-crude days, the 155 down-crude days without heavy volume, and all 722 sessions. The confirmed group's average sits at +0.88%, while the unconfirmed down-days average -0.42% and the full sample -0.17%, which is exactly the pattern the question predicted. The catch is the spread: confirmed outcomes run from -10.8% to +12.3%, and the all-days bucket even reaches -20.3%, so the three distributions overlap heavily. That overlap is why a 0.88% edge with a 0.063 p-value reads as suggestive rather than proven.
This histogram spreads out the 71 confirmed-event outcomes so you can see how lopsided the result is. The average is +0.88% but the median is only +0.46%, which means a handful of strong positive outcomes are dragging the mean above the typical event. Combined with the 53.5% win rate, that tells you the edge is not a steady drip of small wins after capitulation, but more like a coin-flip hit rate with occasional big payoffs. That skew, not the headline average, is the reason the evidence lands just short of statistical significance.
XLE-SPY 10d forward differential summary
| Group | N | Mean | Median | Win rate | Std |
|---|---|---|---|---|---|
| Confirmed high-vol | 71 | 0.0088 | 0.0046 | 0.5352 | 0.0481 |
| Unconfirmed down-day | 155 | -0.0042 | -0.0092 | 0.4452 | 0.0416 |
| All valid days | 722 | -0.0017 | -0.0036 | 0.4792 | 0.0472 |
Confirmed event days, most recent first
| date | XLE daily ret | Brent daily ret | XLE fwd10 | SPY fwd10 | XLE-SPY fwd10 |
|---|---|---|---|---|---|
| 2026-08-04 | -0.0049 | -0.0273 | 0.0858 | -0.0071 | 0.0929 |
| 2026-07-27 | -0.0232 | -0.0846 | 0.0311 | 0.0454 | -0.0143 |
| 2026-05-28 | -0.007 | -0.0169 | -0.0008 | -0.0231 | 0.0223 |
| 2026-05-27 | -0.0135 | -0.0549 | 0.0195 | -0.0387 | 0.0582 |
| 2026-05-21 | -0.0099 | -0.0284 | -0.0309 | -0.0148 | -0.0162 |
| 2026-05-20 | -0.024 | -0.0498 | -0.0227 | 0.0194 | -0.0421 |
| 2026-05-07 | -0.0097 | -0.0181 | 0.0475 | 0.0174 | 0.0301 |
| 2026-05-06 | -0.036 | -0.0944 | 0.0477 | 0.0073 | 0.0404 |
| 2026-04-17 | -0.0286 | -0.1543 | 0.0709 | 0.013 | 0.0579 |
| 2026-04-01 | -0.035 | -0.0563 | -0.0356 | 0.0726 | -0.1082 |
| 2026-03-10 | -0.002 | -0.0478 | 0.0953 | -0.0273 | 0.1226 |
| 2026-03-09 | -0.0168 | -0.0145 | 0.0748 | -0.0248 | 0.0996 |
| 2026-03-04 | -0.0049 | -0.0207 | 0.042 | -0.0379 | 0.0799 |
| 2026-01-07 | -0.0083 | -0.0164 | 0.0828 | -0.0012 | 0.084 |
| 2026-01-06 | -0.0292 | -0.0143 | 0.0722 | -0.0075 | 0.0797 |
| 2025-12-16 | -0.0279 | -0.0263 | 0.0277 | 0.0101 | 0.0176 |
| 2025-11-20 | -0.0101 | -0.0022 | 0.037 | 0.0482 | -0.0113 |
| 2025-11-19 | -0.0165 | -0.0167 | 0.0251 | 0.0221 | 0.003 |
| 2025-11-17 | -0.0215 | -0.0046 | 0.004 | 0.0257 | -0.0217 |
| 2025-10-10 | -0.0274 | -0.0419 | 0.0315 | 0.0385 | -0.0071 |
| 2025-09-30 | -0.0127 | -0.007 | -0.0324 | -0.0031 | -0.0292 |
| 2025-09-05 | -0.0191 | -0.0224 | 0.0154 | 0.0299 | -0.0144 |
| 2025-09-03 | -0.0217 | -0.0048 | 0.0174 | 0.0281 | -0.0107 |
| 2025-08-06 | -0.0071 | -0.0169 | 0.0133 | 0.0066 | 0.0068 |
| 2025-08-01 | -0.0163 | -0.0392 | -0.0035 | 0.0357 | -0.0392 |
The takeaway
The capitulation thesis points the right way here, but the evidence is a lean, not a proven edge. Across 71 volume-confirmed down-crude days, XLE beat SPY by +0.88% on average over the following 10 sessions, while the same setup without the volume confirmation actually gave back -0.42% and all days averaged -0.17% — so the volume filter does appear to be separating something real from ordinary weakness. The catch is consistency: XLE outperformed in only 53.5% of cases, which is barely better than a coin flip, and the typical (median) edge was just +0.46%, meaning a few large winners are carrying the average. Statistically there's roughly a 6-in-100 chance (p = 0.063) you'd see a spread this big if nothing were going on, and the gap against unconfirmed down days sits right on the line at p = 0.051. Treat this as a suggestive pattern worth watching, not a confirmed signal — 71 events with overlapping windows is thin ground for a firm conclusion.
The fine print
- The 10-session windows overlap whenever events cluster, so the tests treat correlated outcomes as independent — true confidence is weaker than p = 0.063 implies.
- Top-quintile volume is one arbitrary cutoff; a 75th or 90th percentile threshold would change the event count and could change the verdict.
- Brent's daily close is joined by calendar date with forward-fill, so if the source stamps it at T+1 the 'down crude' flag can be off by a session.
- This covers one energy regime over roughly three years, with no transaction costs or slippage modeled.