VLO-XLE 20-session forward excess when VLO-Brent 20d correlation is below -0.3 vs above 0
The setup feels counterintuitive before you look at the data. An oil refiner whose 20-day returns have been moving against crude suddenly, as a group, beats its own sector benchmark by a serious margin over the following month.
VLO's rolling correlation to Brent crude dipped below -0.3 in only 64 of 699 overlapping sessions. In those windows, VLO's mean 20-session excess over XLE was about +7.0%, versus roughly +1% in the 365 sessions where correlation was above zero. The gap holds in the median as well, so it is not driven by a few lucky streaks.
The overlapping forward windows inflate the headline confidence stats, and 64 triggering sessions is a thin base, so read this as a durable tendency rather than a precise forecast. Even with those caveats, the edge is wide enough to warrant a close look at the mechanics. The full methodology and breakdown are below.
Over the past ~3 years, when VLO's rolling 20-day return correlation to Brent crude is below -0.3, does VLO outperform XLE over the next 20 trading sessions by more than when the correlation is above +0.
How this was measured
Aligned VLO and XLE daily closes from minute bars with Brent crude daily values. Daily returns for VLO were computed as close-to-close. Brent returns were lagged one business day to respect the macro data's T+1 publication convention. For each day, a 20-session rolling Pearson correlation between VLO returns and lagged Brent returns was computed using only information available by that close. The next 20-session forward return of VLO minus XLE was measured; days with rolling correlation below -0.3 were compared against days with correlation above 0.0 using means, medians, and a Welch two-sample t-test on the overlapping forward excess returns.
The key numbers
Reading the numbers
Read it simply: when VLO's 20-day correlation to Brent was below -0.3, the next 20 sessions showed VLO beating XLE by 7.0% on average, versus 0.98% when correlation was above 0 — roughly a 6-point gap. But because those forward windows overlap, treat the gap as suggestive, not a proven edge.
The charts
This line tracks VLO's rolling 20-day correlation to Brent from late 2023 through mid-2026. It swings widely, from about -0.56 at its lowest to +0.52 at its highest, and the lower threshold line at -0.3 marks the condition being tested. Only 64 of 699 sessions fell below -0.3, while 365 sat above zero — so the rare deep-negative stretches are exactly the episodes the rest of the report compares.
The box plot compares what actually happened over the next 20 sessions after each regime. The low-correlation group averaged a 7.02% VLO-XLE excess return, with its worst case around -5.5% and best around +18.4%; the high-correlation group averaged only 0.98%, ranging from about -11.1% to +24.7%. So low-correlation days weren't a guaranteed win on every single occurrence, but the typical and average outcomes were clearly better, and VLO beat XLE in 89% of low-correlation sessions versus just 51.5% of high-correlation sessions.
These two bars summarize the whole question directly: after VLO-Brent correlation below -0.3, the mean 20-session VLO-XLE excess return is 7.02%, versus 0.98% after correlation above zero. That 6.04 percentage-point gap is why the low-correlation regime stands out in this study. In plain terms, the average forward outperformance was much larger after the negative-correlation condition, though the report cautions that overlapping windows make the statistical significance look more certain than it truly is.
20-session forward VLO-XLE excess summary
| Regime | N | Mean excess | Median excess | Std excess | % sessions VLO > XLE |
|---|---|---|---|---|---|
| Corr < -0.3 | 64 | 0.0702 | 0.0658 | 0.0578 | 89.06 |
| Corr > 0.0 | 365 | 0.0098 | 0.0029 | 0.0635 | 51.51 |
The takeaway
Short answer: yes, and the gap is not subtle. Across roughly 700 overlapping sessions, the low-correlation state (VLO-Brent 20-day correlation below -0.3) handed VLO a +7.0% average 20-session edge over XLE, versus +0.98% when the correlation was above zero. That 6.0-percentage-point difference shows up in the median too: +6.6% versus +0.29%. The hit rate is arguably the most striking part: VLO finished above XLE in 89% of the low-correlation windows, compared with 51.5% of the positive-correlation windows. The headline p-value (~2.5e-11, about one in 40 billion) makes it look bulletproof, but the 20-day forward windows overlap, so that p-value is too clean and the real degree of certainty is lower. Even granting that, this is not a coin flip: mean, median, and hit rate all say a strongly negative VLO-Brent correlation has been a solid setup for VLO outperforming XLE over the following month. The practical caveat is that only 64 sessions trigger this regime, so treat it as a durable-looking tendency rather than a precise forecast.
The fine print
- 20-day forward returns overlap across consecutive sessions, so the Welch p-value overstates confidence.
- Only 64 of 699 sessions meet the below -0.3 trigger; if those sessions cluster in a few episodes, the effective sample is even smaller.
- Brent returns are lagged one business day for T+1 publication; using same-day Brent instead would shift the correlation regime.
- The -0.3 and 0.0 thresholds are arbitrary, and XLE's own energy exposure means VLO versus XLE is not a fully independent comparison.