AI Research XLESPYmacro:brent_daily

XLE vs SPY next-10-session performance after a lagged +2% Brent spike on an XLE down day

55
Signal days (lagged Brent > +2% and XLE lower)

A crude spike that energy stocks refuse to follow looks like a classic washout signal. If Brent jumps more than 2% and XLE still closes lower, the theory goes, the cross-asset volatility is flushing out weak hands — and once it clears, energy should reprice the higher crude and lead the broad market.

We tested that over roughly three years of daily data, measuring XLE against SPY over the next 10 sessions after 55 such signals. XLE did lean ahead, averaging about half a point of excess return and beating SPY 60% of the time. But the scatter is wide enough that the result is statistically indistinguishable from noise, and both ETFs rose on average — most of the move was broad market bounce, not energy leadership.

The full breakdown, including the baseline comparison and the outliers dragging the mean, is below.

The research question

Over the past ~3 years, when Brent crude's one-day return is above +2% but XLE closes lower, does XLE outperform SPY over the next 10 trading sessions? I expect a supply-driven crude spike that energy equities cannot hold to mark a broad macro risk flush, and once the cross-asset volatility clears XLE reprices the higher crude level and leads the broad market.

How this was measured

Daily XLE and SPY closes were built from minute bars. The Brent trigger uses brent_daily_df values converted to one-day returns. Because the Brent daily series is T+1, the prior official Brent return is lagged one session and forward-filled onto the US equity trading calendar, so the signal is observable at the XLE close. A signal day requires that lagged Brent return to be above +2% and XLE's same-day close-to-close return to be negative. For each signal day, forward 10-session returns for XLE, SPY, and their difference are measured from close t to close t+10. The analysis compares signal-day excess to zero and to non-signal days.

The key numbers

Signal days (lagged Brent > +2% and XLE lower)
55
2023-08-31 to 2026-08-31
Signal-day mean forward 10d XLE
1.6190%
N=55
Signal-day mean forward 10d SPY
1.1461%
N=55
Signal-day mean forward 10d XLE-SPY excess
0.4729%
Positive means XLE leads SPY
Signal-day median forward 10d excess
1.4709%
Robust to outliers
Signal days where XLE beats SPY over next 10d
60.00%
N=55 signal days
Non-signal baseline mean 10d excess
-0.2128%
N=687 non-signal days
One-sample t-stat (signal excess vs 0)
0.568
Two-sided
One-sample p-value (vs 0)
0.5727
p=0.5727 >= 0.05 -> no statistically clear nonzero mean excess
Welch t-stat (signal vs non-signal excess)
0.806
Unequal variance
Welch p-value (signal vs baseline)
0.4237
p=0.4237 >= 0.05 -> no statistically clear difference from baseline

Reading the numbers

On the 55 days that fit the setup, XLE beat SPY by about 0.47% on average over the following 10 sessions and did so 60% of the time. But that average is small enough relative to the day-to-day noise that it is statistically indistinguishable from zero (p=0.57), so this sample can't confirm the edge either way.

The charts

Average cumulative path after signal: XLE vs SPY and excess
What this chart says

This traces the average path of XLE, SPY and their gap over the 10 sessions after each signal, starting at zero. Both lines grind higher rather than jumping — XLE finishes up about 1.62% and SPY about 1.15%, leaving the excess at roughly +0.47%. The detail worth noting is the excess line: it dips marginally negative early (about -0.1%) before building to a peak near +0.54%, so the outperformance accrues gradually over the window rather than appearing as an immediate repricing.

Forward 10d XLE-SPY excess on signal days (N=55)
What this chart says

This histogram shows how the 55 individual signal outcomes are spread, and the spread is wide — from about -20.3% to +13.6% excess, with an average of +0.47%. The shape matters more than the mean: the median outcome is +1.47%, higher than the average, which tells you a handful of large negative readings are dragging the average down. So the typical signal actually favored XLE more than the headline number suggests, but the losers are big enough to make the result unreliable.

Mean forward 10d return by group
What this chart says

Four bars compare the group averages side by side: XLE +1.62% and SPY +1.15% on signal days, the +0.47% excess, and -0.21% excess on the 687 non-signal days. The signal excess bar is positive while the baseline bar is slightly negative, a gap of roughly 0.7 percentage points in XLE's favor — directionally consistent with the thesis. Two caveats stand out: both XLE and SPY rise on signal days, so these look more like broadly rising markets than macro risk flushes, and the signal-versus-baseline gap is not statistically clear (p=0.42).

Signal-day events (most recent 40)

signal_datebrent_lag1_retbrent_same_retxle_day_retxle_fwd10spy_fwd10excess_fwd10
2024-08-010.0269-0.0002-0.0269-0.00320.0231-0.0263
2024-08-130.0245-0.0066-0.0050.01290.0348-0.0219
2024-10-080.0305-0.0434-0.0266-0.00840.0181-0.0266
2024-10-230.0314-0.012-0.00560.04080.02070.0202
2024-12-040.0273-0.0016-0.023-0.0893-0.0314-0.0579
2025-04-100.0376-0.0287-0.07390.07030.05560.0147
2025-04-210.02040.0205-0.02830.01450.0906-0.0761
2025-04-230.0204-0.0097-0.0136-0.01260.0455-0.058
2025-05-140.0253-0.0152-0.013-0.03280.0034-0.0361
2025-05-290.0218-0.017-0.00360.05660.02050.0361
2025-06-180.0486-0.0041-0.011-0.0050.0501-0.0552
2025-07-140.0239-0.0153-0.01440.0010.0212-0.0202
2025-07-180.0237-0.0036-0.0066-0.0035-0.01060.0072
2025-07-300.0330.0105-0.0131-0.0220.009-0.0311
2025-08-150.0282-0.012-0.00140.05810.00250.0555
2025-10-240.0649-0.0078-0.00940.0171-0.00730.0244
2025-11-170.0211-0.0046-0.02150.0040.0257-0.0217
2025-11-190.0269-0.0167-0.01650.02510.02210.003
2025-11-250.0327-0.013-0.00170.04030.01450.0258
2025-12-240.02380-0.00290.05360.00590.0477
2025-12-260.02380-0.00360.04910.00570.0433
2026-01-090.0370.0279-0.00030.0539-0.00220.0561
2026-01-120.02790.0045-0.00790.0720.00470.0673
2026-01-260.0412-0.0067-0.00250.09010.00110.0891
2026-03-030.0830.0782-0.01040.0373-0.0140.0513
2026-03-040.0782-0.0207-0.00490.042-0.03790.0799
2026-03-090.0807-0.0145-0.01680.0748-0.02480.0996
2026-03-130.12530.0083-0.00020.0932-0.04270.1359
2026-03-180.07270.0895-0.00040.0096-0.00660.0162
2026-03-300.07130.0034-0.0105-0.10130.1016-0.2029
2026-04-010.0395-0.0563-0.035-0.03560.0726-0.1082
2026-04-140.0354-0.0372-0.02210.03570.02570.01
2026-05-130.0496-0.0098-0.0005-0.01270.0155-0.0282
2026-06-040.0325-0.0266-0.0028-0.0832-0.0076-0.0757
2026-06-300.0204-0.0158-0.00770.06260.01120.0514
2026-07-090.0658-0.0267-0.01580.0836-0.01690.1005
2026-07-140.09790.0254-0.00050.0225-0.01260.0351
2026-07-150.0254-0.0073-0.00740.0368-0.02950.0663
2026-08-030.0548-0.083-0.02360.06620.01880.0473
2026-08-070.0346-0.0226-0.0120.1047-0.00830.113

Forward 10-session return summary

seriesNmeanmedianstdpositive_share
Signal XLE550.01620.01710.04680.6182
Signal SPY550.01150.01120.02990.6909
Signal XLE-SPY excess550.00470.01470.06180.6
Non-signal XLE-SPY excess687-0.0021-0.00430.04550.4731

The takeaway

Short answer: the setup doesn't hold up as an edge — XLE does lean toward beating SPY after these signals, but the tilt is small enough that it's indistinguishable from noise. On the 55 qualifying days (lagged Brent up more than 2%, XLE down), XLE averaged +0.47% of excess return over SPY across the following 10 sessions, and it finished ahead 60% of the time, versus a non-signal baseline of -0.21%. That looks encouraging until you price in the scatter: the median excess was +1.47% — bigger than the average, meaning a handful of bad signals are dragging the mean down — and the spread of outcomes was wide, with one August 2024 event showing XLE trailing SPY by 5.8%. Statistically it's a coin flip: the chance of seeing a +0.47% average excess this large purely by luck is about 57 in 100 (p = 0.57), and comparing signal days against ordinary days gives almost the same verdict (p = 0.42). Note too that both ETFs rose on average after these signals (+1.6% XLE, +1.1% SPY), so most of what you're capturing is broad market bounce, not energy-specific leadership. Practical takeaway: the story is plausible but this sample can't confirm it — treat the signal as a hypothesis that needs more events or a cleaner filter (e.g., excluding cases where the crude spike was already reversing), not as a tradeable rule.

The fine print