WMB vs XLE after Brent 20d weakness plus WMB estimate revision >= +1%
Midstream's pitch is that fee-based cash flows don't much care where crude goes. Williams' next-quarter estimates kept climbing through stretches of falling Brent — three separate times in roughly three years — and each time WMB beat XLE over the following 20 sessions. The average gap was 2.96 percentage points, against 1.21 points across the full sample.
The catch is sample size. Three episodes, and a p-value of 0.061 that misses the usual 5% bar — the shortfall is about how rarely the setup fired, not how large the effect was. The full breakdown, including each episode's returns and how the benchmark was built, is below.
Over the past ~3 years, when Brent crude's trailing 20-day return is negative while WMB's next-quarter consensus EPS estimate is revised upward by at least 1% over the prior 30 days, does WMB outperform XLE over the next 20 trading sessions? I expect midstream's fee-based cash flows to keep estimates climbing even as crude falls, so WMB decouples from oil-beta energy equities and leads the sector until the tape stabilizes.
How this was measured
Daily WMB and XLE closes were built from minute bars and joined with Brent daily prices. For each session, Brent's trailing 20-session return was measured from the prior session's close (T+1 availability) and WMB's next-quarter EPS revision was read as the 30-day percent change in the next-forward quarterly consensus row. A joint signal fired when Brent's trailing 20-session return was negative and WMB's next-quarter revision was at least +1%. Consecutive eligible days were collapsed to episode starts, and forward 20-session WMB and XLE returns were measured close-to-close from each start. The benchmark is the full-sample WMB minus XLE 20-session forward gap.
The key numbers
Reading the numbers
Across three event-start episodes, WMB beat XLE by 2.96% on average over the next 20 sessions, and every episode favored WMB. But with only three observations, the p-value of 0.0611 misses the usual 5% bar, so the edge is suggestive rather than statistically proven.
The charts
This bar chart contrasts the event-start episodes with all days. The WMB event bar is at 2.90% while the XLE event bar is just below zero at -0.06%, a gap of about 2.96 percentage points; on all days, WMB is 2.54% and XLE is 1.32%, a narrower gap closer to the 1.21% full-sample baseline. That is the pattern the question is asking about: when Brent is weak and WMB estimates are revised up, WMB did lead XLE, but the event bars rest on only three episode starts.
This histogram shows the three episode-start WMB-minus-XLE forward 20-session returns. The key detail is that all three observations sit above zero, with the smallest at 1.45%, the largest at 3.96%, and the mean at 2.96%. A clean sweep in the same direction matches the 100% episode hit rate, but three bars are a very small sample, which is why the result is directionally supportive rather than statistically clear.
The scatter plots each event start's WMB-minus-XLE forward 20-session gap. All three dots sit above zero, ranging from 1.45% to 3.96% around a 2.96% mean, so every episode-start anchor ended with WMB ahead of XLE. The visual caution is the sample size: three dots cannot show whether the edge is stable over time or whether one episode carries the result.
WMB vs XLE 20-session forward return summary
| group | n | mean_wmb_fwd | mean_xle_fwd | mean_diff | median_diff |
|---|---|---|---|---|---|
| Event-start episodes | 3 | 0.029 | -0.0006 | 0.0296 | 0.0346 |
| All analysis days | 690 | 0.0254 | 0.0132 | 0.0121 | 0.0167 |
Last 20 event-start episodes
| date | brent_20d_ret | wmb_rev_30d | wmb_fwd20 | xle_fwd20 | wmb_minus_xle |
|---|---|---|---|---|---|
| 2024-07-22 | -0.0085 | 0.0244 | 0.0056 | -0.0089 | 0.0145 |
| 2024-08-29 | -0.0166 | 0.0244 | 0.0017 | -0.033 | 0.0346 |
| 2024-09-04 | -0.0135 | 0.0244 | 0.0798 | 0.0402 | 0.0396 |
The takeaway
Yes, directionally — WMB beat XLE every time the setup fired, but with only three distinct episodes this is a lean, not a proven pattern. In each of the three event starts, WMB's next 20 sessions beat XLE, with a mean outperformance of 2.96% and a median of 3.46% — including the widest case at 3.96 percentage points. Compare that with the benchmark: across all 690 analysis days, WMB beat XLE by only 1.21% on average over 20 sessions, so the event edge is roughly 1.7 points above normal. The statistical test lands at p = 0.061 — about a 6-in-100 chance of seeing this by luck — just missing the usual 5% bar, and that weakness is entirely about sample size, not effect size. Three episodes over roughly three years simply isn't enough to call it a real signal; treat it as a sensible idea that the data so far happens to support. Practically, the thesis is plausible and worth watching, but don't size it like a confirmed edge until more episodes accumulate.
The fine print
- Only 3 event-start episodes over ~3 years — the hit rate of 3-for-3 is encouraging but far too thin to be conclusive.
- The estimate-revision signal is a per-quarter snapshot carried across dates rather than true daily revision history, so event timing is approximate.
- Forward windows overlap across the closely spaced 2024 starts (late July through early September), so the p-value is not exact.
- The largest episode (+7.98% for WMB) carries the average; results are close-to-close with no spread or slippage modeled.
- Tied to WMB-specific corporate events over a single window — may not extend to other midstream names.