AI Research XLESPYmacro:brent_daily

XLE vs SPY after high-volume down-crude capitulation candidates

71
Confirmed event days

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.

The research question

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

Confirmed event days
71
XLE down, Brent down, XLE volume above prior-20-session 80th pct
Unconfirmed down days
155
XLE down and Brent down, but volume not in top prior-20-session quintile
All valid days
722
2023-09-29 to 2026-08-17
Confirmed mean XLE-SPY 10d differential
0.8837%
N=71 events
Confirmed median XLE-SPY 10d differential
0.4643%
Robust to outlier events
Confirmed win rate (XLE > SPY)
53.52%
Fraction of event days with positive 10d XLE-SPY differential
Confirmed mean XLE 10d forward return
2.0805%
Raw XLE 10-session return after event
Confirmed mean SPY 10d forward return
1.1968%
Raw SPY 10-session return after event
Unconditional mean XLE-SPY 10d differential
-0.1705%
N=722 all valid days
Unconfirmed down-day mean XLE-SPY 10d differential
-0.4210%
N=155 XLE-down/Brent-down days without volume confirmation
One-sample t-stat (events vs zero)
1.548
Positive favors XLE outperformance over SPY
One-sample p-value (greater)
0.0630
One-sided; p=0.0630 >= 0.05 -> no statistically-clear positive differential
Welch t-stat: confirmed vs unconfirmed
1.972
Positive favors volume-confirmed down days
Welch p-value: confirmed vs unconfirmed
0.0509
Two-sided; p=0.0509 >= 0.05 -> no statistically-clear difference from unconfirmed down days
Welch p-value: confirmed vs all days
0.0811
Two-sided; p=0.0811 >= 0.05 -> no statistically-clear difference from all-day baseline

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

XLE-SPY 10d forward differential by group
What this chart says

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.

Confirmed event-day XLE-SPY 10d differential
What this chart says

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

GroupNMeanMedianWin rateStd
Confirmed high-vol710.00880.00460.53520.0481
Unconfirmed down-day155-0.0042-0.00920.44520.0416
All valid days722-0.0017-0.00360.47920.0472

Confirmed event days, most recent first

dateXLE daily retBrent daily retXLE fwd10SPY fwd10XLE-SPY fwd10
2026-08-04-0.0049-0.02730.0858-0.00710.0929
2026-07-27-0.0232-0.08460.03110.0454-0.0143
2026-05-28-0.007-0.0169-0.0008-0.02310.0223
2026-05-27-0.0135-0.05490.0195-0.03870.0582
2026-05-21-0.0099-0.0284-0.0309-0.0148-0.0162
2026-05-20-0.024-0.0498-0.02270.0194-0.0421
2026-05-07-0.0097-0.01810.04750.01740.0301
2026-05-06-0.036-0.09440.04770.00730.0404
2026-04-17-0.0286-0.15430.07090.0130.0579
2026-04-01-0.035-0.0563-0.03560.0726-0.1082
2026-03-10-0.002-0.04780.0953-0.02730.1226
2026-03-09-0.0168-0.01450.0748-0.02480.0996
2026-03-04-0.0049-0.02070.042-0.03790.0799
2026-01-07-0.0083-0.01640.0828-0.00120.084
2026-01-06-0.0292-0.01430.0722-0.00750.0797
2025-12-16-0.0279-0.02630.02770.01010.0176
2025-11-20-0.0101-0.00220.0370.0482-0.0113
2025-11-19-0.0165-0.01670.02510.02210.003
2025-11-17-0.0215-0.00460.0040.0257-0.0217
2025-10-10-0.0274-0.04190.03150.0385-0.0071
2025-09-30-0.0127-0.007-0.0324-0.0031-0.0292
2025-09-05-0.0191-0.02240.01540.0299-0.0144
2025-09-03-0.0217-0.00480.01740.0281-0.0107
2025-08-06-0.0071-0.01690.01330.00660.0068
2025-08-01-0.0163-0.0392-0.00350.0357-0.0392

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