LNG vs XLE after positive Brent with LNG-Brent correlation in trailing bottom quintile
A positive crude tape plus an LNG that has stopped tracking oil sounds like the setup for company-specific repricing. It isn't. When Brent's 20-day return was positive and LNG's correlation to crude sat in the bottom fifth of its trailing year, LNG went on to lag XLE over the next 20 sessions, averaging -0.36% relative performance across 57 qualifying days.
The median outcome was worse: a 2.12% LNG shortfall. Both names rose on these crude-up tapes, so oil beta lifted the sector and LNG simply didn't outrun it. The hit rate, 40.4%, was slightly below the 42.3% on non-qualifying days, leaving the so-called edge with the wrong sign.
The full study below breaks down the signal, the forward returns, and why the decoupling thesis fails in this sample.
Over the past ~3 years, when Brent crude's 20-day return is positive while LNG's 20-day rolling correlation to Brent crude is in the bottom quintile of its trailing 252-session distribution, does LNG outperform XLE over the next 20 trading sessions? I expect a positive crude tape with LNG decoupled from oil to signal company-specific contract and capacity repricing that the oil-beta sector underreprices, so LNG leads XLE until the correlation regime normalizes.
How this was measured
Daily closes were resampled from LNG_df and XLE_df minute bars, and Brent crude was aligned to that US-session calendar with a one-session lag relative to the oil price series to respect T+1 macro publication timing. A 20-session Brent return was computed from the lagged price path, and a 20-session rolling correlation between LNG daily returns and Brent daily returns was formed. The correlation's current value was then compared with the 20th percentile of its own trailing 252 prior values. A qualifying day required both the Brent 20-session return to be positive and the current LNG-Brent correlation to be at or below that trailing bottom-quintile threshold. The forward outcome is the 20-session close-to-close LNG return minus the XLE return, compared against all other valid signal days via Welch's t-test.
The key numbers
Reading the numbers
Only 57 of 731 sessions met the setup, and when they did, LNG beat XLE over the following 20 sessions just 40% of the time, with an average shortfall of 0.36 points. Days that did not meet the setup were nearly identical (42% win rate, -0.04 points), and the p-value of 0.72 says that gap is well within normal randomness.
The charts
This tracks LNG's 20-day correlation to Brent (jagged line, swinging from -0.71 to +0.74) against the trailing bottom-quintile cutoff drawn as the lower line. The key detail is that the cutoff itself drifts, sitting near -0.18 at the start of the window and around -0.13 at the end, so 'decoupled' meant something slightly different in 2023 than it does recently. The setup requires the correlation line to dip at or below that moving cutoff while crude's own 20-day return is positive, a combination that occurred on just 57 of 731 sessions. That scarcity is the first clue that the sample may be too thin to separate signal from noise.
Two side-by-side distributions of LNG-minus-XLE forward 20-session returns: the 57 qualifying signals and the 381 other valid days. The qualifying group's average is -0.36% versus -0.04% for the rest, but the two boxes sit almost on top of each other, and the qualifying range runs from -8.75% to +15.43% — an enormous spread relative to the mean. Read it this way: the typical outcome for the decoupling setup is a loss relative to XLE, but the dispersion is so wide that a handful of large winners could flip the average either direction. That fragility is why the mean, not the tail, should be the focus.
This strips the comparison down to two bars: average forward outperformance for qualifying signals (-0.36%) and for other valid days (-0.04%). The bars are nearly the same height and both sit just below zero, which is the visual version of the report's conclusion: the crude-up, low-correlation setup did not produce LNG leadership over XLE. The 0.32-point gap between them is smaller than the week-to-week wobble in either group, so there is no reliable edge to act on here.
Forward-return summary by signal group
| Group | N | Mean LNG−XLE | Median LNG−XLE | Win rate | Mean LNG | Mean XLE |
|---|---|---|---|---|---|---|
| Qualifying signal | 57 | -0.0036 | -0.0212 | 0.4035 | 0.03 | 0.0336 |
| Other valid days | 381 | -0.0004 | -0.0165 | 0.4226 | 0.0136 | 0.014 |
| Difference | -0.0032 | -0.0191 | 0.0164 | 0.0196 |
The takeaway
No — the setup doesn't work. When Brent's 20-day return was positive and LNG's correlation to crude sat in the bottom fifth of its trailing year, LNG went on to lag XLE over the next 20 sessions rather than lead it, averaging -0.36% relative performance across 57 qualifying days. The hit rate was 40.4%, actually a touch worse than the 42.3% on the 381 days that didn't qualify, so the so-called edge is -0.32 percentage points — the wrong sign. Strip out the averaging and it looks worse, not better: the middle qualifying outcome was a 2.12% LNG shortfall. The decomposition explains why the thesis misfires — both names rose on these crude-up tapes (LNG +3.00%, XLE +3.36%), meaning oil beta lifted the whole sector and LNG simply didn't outrun it. Statistically this isn't close to a signal: p = 0.72 says a gap this size shows up from pure chance roughly seven times in ten, and with only 57 observations across overlapping windows the honest read is that this is indistinguishable from a coin flip leaning slightly the wrong way. Practically, a positive crude tape plus a decoupled LNG is not a green light to favor the company over the oil-beta basket — if anything the data nudges the other direction, and the contract-repricing story the hypothesis leans on doesn't show up in the forward prices.
The fine print
- Just 57 qualifying sessions, and they're tail-conditional by construction — a thin base for any conclusion.
- The 20-day forward windows overlap heavily across consecutive signals, so the effective number of independent observations is well below 57.
- The bottom-quintile correlation threshold is recalculated daily and regime-dependent; 'decoupled' in one oil regime is not the same as in another.
- Returns are gross — no costs, slippage, or modeling of how rarely the setup actually triggers.