AI Research XLESPYmacro:brent_daily

XLE vs SPY forward 20d performance when XLE is high-Brent-correlated and low-SPY-correlated

100
Qualifying sessions

When XLE's 20-day correlation to Brent jumps while its correlation to SPY falls, is the energy sector actually trading on crude rather than on the broad market — and does that decoupling pay off? That's the regime this study tests across roughly three years of daily data.

On the record, it did. Across 100 qualifying sessions, XLE beat SPY by an average of 3.2 percentage points over the next 20 sessions, against a slight lag on the days that didn't qualify, and it came out ahead 63% of the time.

The caveats matter, though: overlapping windows make those odds look better than they are, and the most recent qualifying stretch runs the other way. The full breakdown, including where the pattern is currently failing, is below.

The research question

Over the past ~3 years, when XLE's rolling 20-day correlation to Brent crude is in the top tercile of its prior 252-session distribution while its rolling 20-day correlation to SPY is in the bottom tercile of its prior 252-session distribution, does XLE outperform SPY over the next 20 trading sessions? I expect supply-shock-driven decoupling from broad equity beta to let XLE trade on crude and lead SPY until the correlation regime reverts.

How this was measured

Built daily closes for XLE and SPY from minute bars, and aligned daily Brent prices to the NYSE trading calendar with forward-fill. Computed 20-session rolling correlations of XLE returns to Brent returns and to SPY returns. For each session, the top-tercile threshold for XLE-Brent correlation and the bottom-tercile threshold for XLE-SPY correlation were estimated from the prior 252 available correlation observations, excluding the current session. A day qualified when the current XLE-Brent correlation exceeded its prior top-tercile threshold AND the current XLE-SPY correlation fell below its prior bottom-tercile threshold. The outcome is the next 20-session simple return of XLE minus the same horizon return of SPY. Conditional outcomes were compared to a one-sample zero test and to all non-qualifying sessions with a Welch two-sample test.

The key numbers

Qualifying sessions
100
Top-tercile XLE-Brent corr AND bottom-tercile XLE-SPY corr; valid window 2024-10-07 to 2026-07-31
Non-qualifying sessions
348
Baseline comparison set
Mean conditional XLE fwd20 return
3.8330%
N=100 qualifying sessions
Mean conditional SPY fwd20 return
0.5974%
N=100 qualifying sessions
Mean conditional XLE-SPY excess
3.2356%
20-session XLE return minus SPY return
Non-qualifying mean excess
-0.4965%
N=348 baseline sessions
Conditional minus baseline excess
3.7321%
Positive means the correlation regime adds edge
Fraction of positive excess
63.000%
N=100 qualifying sessions
One-sample t-stat vs zero
4.205
Tests whether conditional XLE-SPY excess is non-zero
One-sample p-value
0.0001
p=0.0001 < 0.05 -> conditional excess is statistically distinguishable from zero
Welch t-stat vs non-events
4.406
Positive favors conditional outperformance
Welch p-value vs non-events
0.0000
p=0.0000 < 0.05 -> conditional excess differs from baseline

Reading the numbers

On the 100 days when XLE moved with oil but not with the broad market, it beat SPY by an average of 3.24% over the next 20 sessions, versus a 0.50% average shortfall on the other 348 days. The odds that gap is pure luck are about 1 in 10,000.

The charts

XLE 20d rolling correlations to Brent and SPY
What this chart says

This tracks two rolling 20-day correlations side by side. The XLE-Brent line (blue) spends most of the window positive, averaging 0.48 and peaking at 0.87, while the XLE-SPY line (orange) is much weaker on average at 0.14 and repeatedly dives into negative territory, ending the window near -0.43. The stretch to watch is where the two lines pull far apart — oil correlation high, equity correlation below zero — because those are exactly the setups the study isolates, and they are frequent enough to matter rather than one-off events.

XLE-SPY 20d forward excess: qualifying vs non-qualifying days
What this chart says

This box plot lines up the next-20-session XLE-minus-SPY return for the 100 qualifying days against the 348 days that did not qualify. The qualifying box sits higher and averages +3.24%, with outcomes running from -12.55% to +19.69%, while the non-qualifying group averages -0.50%, from -20.45% to +14.60%. The overlap is real — some qualifying days still lost badly — so the takeaway is a shift in the average and not a clean signal, which fits the finding that 63% of qualifying days were positive.

Mean 20d forward XLE-SPY excess by condition
What this chart says

This is the same comparison boiled down to two bars: mean forward excess of +3.24% when the correlation regime is on, versus -0.50% when it is off. The roughly 3.7 percentage point gap between the two bars is the entire claimed edge, and it is the number the significance tests are built on. Read it as the size of the prize conditional on the regime appearing, not as a guarantee for any single occurrence.

Most recent 20 qualifying sessions

datexle_brent_corrxle_spy_corrxle_fwd20spy_fwd20excess
2026-06-020.7719-0.5407-0.087-0.0168-0.0702
2026-06-030.7944-0.5707-0.0915-0.0038-0.0877
2026-06-040.7252-0.5286-0.0872-0.0026-0.0845
2026-06-050.7192-0.2065-0.04040.0178-0.0581
2026-06-090.7717-0.153-0.03910.024-0.0631
2026-06-100.8129-0.2422-0.05770.047-0.1047
2026-06-110.8088-0.38070.00430.0143-0.01
2026-06-120.761-0.40040.00040.0171-0.0166
2026-06-150.7568-0.41090.02320.00360.0197
2026-06-160.7128-0.4390.03730.00040.0369
2026-06-170.7043-0.39530.0646-0.00170.0663
2026-06-180.6722-0.38970.0857-0.00640.0921
2026-07-130.6772-0.66610.05670.03250.0242
2026-07-140.6964-0.66480.06850.02370.0447
2026-07-150.6639-0.58570.08120.02330.0578
2026-07-160.64-0.61430.07020.03860.0316
2026-07-170.6626-0.71710.07150.04510.0264
2026-07-200.6619-0.71140.07990.04110.0388
2026-07-210.6835-0.66460.08580.02540.0604
2026-07-220.673-0.70920.06880.03120.0376

The takeaway

Yes — on the record, this setup did precede XLE beating SPY, and by a decent margin: across the 100 qualifying sessions, XLE outperformed by an average of 3.2 percentage points over the next 20 sessions (XLE +3.8% versus SPY +0.6%), against a slight -0.5% lag on the 348 days that didn't qualify. That's a 3.7-point swing in the direction you hypothesized, with XLE ahead 63% of the time. Taken at face value the separation looks strong — the chance of a gap that size being pure luck is roughly 1 in 17,000, and the comparison against non-qualifying days tells the same story. The catch is that the confidence is softer than those numbers suggest: the 100 qualifying days sit right next to each other and both the correlations and the forward returns are measured over 20 overlapping sessions, so the number of genuinely independent observations is much smaller than 100 and the p-values are flattering. The most recent evidence actively contradicts the pattern — the five latest qualifying days in early June 2026 all show XLE trailing SPY by 6 to 9 points over the following month. Net: this looks like a real, economically sensible regime effect that is currently not working, so treat it as context rather than a live signal.

The fine print