AI Research EQTXLEmacro:brent_daily

EQT decoupling from Brent while rising: next-20d EQT vs XLE performance

700
Analysis days

A natural gas producer decoupling from crude while its own stock rises sounds like precisely the kind of name you'd want against the energy sector. That's the setup examined here: EQT's trailing 20-day correlation to Brent below zero, its trailing 20-day return above zero, and what EQT does versus XLE over the following month.

On the day-level numbers it looks real. EQT outgained XLE by 2.16% on average across the 72 qualifying days, against -0.44% on all others, a 2.60% edge that the daily test puts at p=0.010.

Then the independence check breaks it. Those 72 days collapse into just 19 distinct episodes, and using only the first day of each flips the average spread to -0.69% with a p-value near 0.89. The full breakdown, including win rates and both tests, is in the analysis below.

The research question

Over the past ~3 years, when EQT's rolling 20-day return correlation to Brent crude is negative while its 20-day return is positive, does EQT outperform XLE over the next 20 trading days? I expect a natural-gas producer that is decoupling from crude and rising on its own supply/demand to keep leading the energy sector as gas-market strength gets re-rated independently of oil beta.

How this was measured

Daily closes for EQT and XLE were resampled from minute bars, and Brent daily values were aligned to the same trading calendar via forward-fill. We computed daily returns, then a trailing 20-day Pearson correlation between EQT and Brent returns and a trailing 20-day EQT return. The trigger condition was: trailing 20-day EQT-Brent correlation < 0 AND trailing 20-day EQT return > 0. For each eligible day we measured the forward 20-trading-day spread as EQT forward return minus XLE forward return. We compared condition days to all other valid days using Welch's t-test, and also report a partially de-overlapped version using the first day of each contiguous condition episode. Eligibility required a full trailing 20-day history and a full 20-day forward window.

The key numbers

Analysis days
700
2023-09-29 to 2026-07-31
Condition days
72
Trailing 20d EQT-Brent correlation < 0 and trailing 20d EQT return > 0
Condition prevalence
10.286%
Share of eligible days hitting the condition
Condition mean EQT-XLE fwd 20d
2.1597%
N=72 condition days; positive means EQT leads XLE over next 20d
Non-condition mean EQT-XLE fwd 20d
-0.4394%
N=628 non-condition days baseline
Condition edge vs baseline
2.5991%
Positive = condition days outperform non-condition days
Condition mean EQT fwd 20d
0.5433%
EQT alone over next 20 trading days
Condition mean XLE fwd 20d
-1.6164%
XLE alone over next 20 trading days
Condition % EQT beats XLE
55.56%
Fraction of condition days with positive EQT-XLE forward spread
Welch t-stat (daily condition vs baseline)
2.617
Positive favors condition days
Welch p-value
0.0105
p=0.0105 < 0.05 -> condition-day spread differs from baseline
Distinct condition episodes
19
First day of each contiguous run of condition days
Episode-start mean EQT-XLE fwd 20d
-0.6924%
N=19; partially reduces overlapping-window inflation
Episode-start % EQT beats XLE
47.37%
Share of episode starts with positive forward spread
Episode Welch p-value
0.8911
p=0.8911 >= 0.05 -> no statistically-clear episode effect

Reading the numbers

On the 72 days (10.3% of the sample) when EQT had been falling with oil but rising on its own, it beat XLE by 2.16% over the next 20 trading days on average, versus a 0.44% loss on all other days — a 2.60% edge with a p-value of 0.010. But those 72 days cluster into only 19 separate stretches, and counting each stretch once instead of each day flips the edge to -0.69% with p=0.891, so treat the he

The charts

Forward 20-day EQT-minus-XLE return: condition vs non-condition days
What this chart says

This side-by-side box plot shows the spread between EQT's and XLE's next-20-day returns for the 72 condition days versus the 628 days that did not meet the condition. The thing to look at is the shift in the middle of each box: the condition group's average is +2.16% while the other group's is -0.44%, which is the entire claimed edge. Just as important is how much the two boxes overlap — condition-day outcomes run from -17.6% to +15.1% and non-condition days from -16.8% to +22.5% — so the improvement is in the center of the distribution, not in every observation, which is why a handful of stretches can drive the result.

Condition-day forward 20d EQT-minus-XLE return distribution
What this chart says

This histogram breaks the 72 condition days into buckets by their forward 20-day EQT-minus-XLE return. The bulk of the mass sits close to zero and the average of +2.16% is pulled up by a right tail reaching +15.1%, with the worst single day at -17.6% on the left. Only 55.6% of these days actually saw EQT beat XLE, so the positive mean comes from winning bigger than it loses rather than from winning more often — a detail worth holding onto if you are sizing any position off this signal.

EQT trailing 20d Brent correlation and trailing 20d EQT return
What this chart says

This line chart tracks two rolling 20-day measures over the sample: EQT's return correlation to Brent and EQT's own return, with 153 plotted values per line. The correlation line averages +0.16 and usually sits in positive territory, topping out near +0.64 and only dipping as low as -0.40, so the decoupling episodes your thesis depends on are the minority state rather than the norm. The question is whether those downward dips in the correlation line, when they line up with the return line above zero, mark the start of EQT's independent leadership over the following month — which is exactly the pairing the first two charts test.

Condition episodes (first day of each contiguous run; max 20 shown)

dateroll_correqt_roll20eqt_fwd20xle_fwd20outperform
2023-10-30-0.0540.0603-0.0392-0.0022-0.037
2024-03-05-0.0130.1168-0.03370.1335-0.1672
2024-03-08-0.0190.08510.00820.1114-0.1031
2024-04-10-0.2020.08020.071-0.03570.1067
2024-04-23-0.0240.09440.0921-0.04180.1338
2024-04-25-0.0370.1579-0.0041-0.05050.0464
2024-10-02-0.040.12030.0538-0.02660.0804
2024-10-04-0.1110.1516-0.0276-0.05620.0286
2024-11-12-0.030.20960.042-0.03770.0797
2025-01-21-0.0010.21020.0201-0.01490.0349
2025-03-21-0.020.0525-0.08-0.10420.0242
2025-03-25-0.0450.1047-0.0986-0.11250.0139
2025-11-10-0.0010.0939-0.00750.0099-0.0174
2025-11-13-0.0760.1525-0.09090.0051-0.096
2025-11-25-0.0480.0823-0.04790.0073-0.0552
2026-01-21-0.3760.02330.110.1317-0.0217
2026-03-19-0.0540.0719-0.126-0.0661-0.0598
2026-05-06-0.1230.0036-0.03560.0239-0.0595
2026-07-23-0.0130.04050.00930.0728-0.0634

The takeaway

At first glance, yes: when EQT's 20-day Brent correlation was negative and EQT's own 20-day return was positive, EQT beat XLE by an average of +2.16% over the next 20 trading days, versus -0.44% on all other days. That +2.60% edge looks meaningful, and the day-level Welch test gives p=0.010, which normally reads as only about a 1-in-100 chance of being pure luck. The win rate was modest, though: EQT topped XLE on 55.6% of the 72 condition days, and EQT's own average forward return was +0.54% while XLE fell -1.62%. The problem is independence: those 72 days cluster into just 19 distinct episodes. When you use only the first day of each episode, the average spread flips to -0.69%, the win rate drops to 47.4%, and the p-value jumps to 0.891—essentially a coin flip. So this is not a real, tradable signal; it's a suggestive daily overlap that doesn't survive a stricter event-level test. Treat the decoupling-plus-rising setup as an interesting narrative, not a reliable edge for EQT versus XLE over the next month.

The fine print