BP vs XLE 10-session forward spread after high BP news sentiment on Brent-down days
A strategy that sounds almost too clean — wait for strong BP-specific headlines on a day crude sells off, then buy the stock against its sector — failed its first real test. Over the past three years, only 20 sessions qualified, and while the average BP-XLE forward spread was positive, the median was a loss of 1.08 percentage points with BP beating the sector just 40% of the time. The split between mean and median is the red flag: a handful of outsized winners are carrying the whole signal.
We tested whether days with top-quintile BP news sentiment and a falling Brent close led to BP decoupling from oil beta and outperforming XLE over the next 10 sessions. Across 702 non-signal comparison days, the signal's extra edge was thin and statistically indistinguishable from luck. In short, the thesis doesn't survive contact with the distribution of outcomes. Full methodology and charts below.
Over the past ~3 years, when BP's daily news-sentiment score is in the top quintile of its prior 20 sessions while Brent crude closes lower on the day, does BP outperform XLE over the next 10 trading sessions? I expect positive BP-specific headlines on a weak crude tape to signal a company-specific re-rating, so BP decouples from commodity beta and leads the sector.
How this was measured
Built daily closes for BP and XLE and aligned them with daily BP news sentiment and Brent crude returns over the most recent 36 months of available data. Daily BP sentiment is the mean of headline ticker_sentiment_score (falling back to overall_sentiment_score where the ticker score is missing). For each day, the prior-20-session 80th percentile of BP sentiment was computed on the previous 20 trading sessions (excluding the current session, requiring at least 10 non-missing sentiment days), and days were flagged when current BP sentiment exceeded that threshold while Brent closed lower than the prior Brent session. The outcome is BP's 10-trading-session forward return minus XLE's 10-session forward return, i.e. BP's sector-relative return. Signal days are compared with all non-signal days using a Welch t-test and a Mann-Whitney U test.
The key numbers
Reading the numbers
Only 20 signal days in ~3 years. The signal-day edge was +0.07% vs -0.16% for other days, a +0.23% gap, but p=0.815 says that gap is easily explained by chance; the median signal-day return was actually -1.08%.
The charts
The signal-day box is built from only 20 days and its whiskers run from roughly -7.5% to +11.3% for BP's 10-day excess over XLE, a much wider spread than the small mean difference would suggest. The signal mean is +0.07% versus -0.16% on non-signal days, and the boxes overlap heavily; with a p-value of 0.815, the tiny edge is not statistically distinguishable from noise. Notice that the median signal-day return is -1.08%, so more than half of those days actually saw BP lag XLE despite the positive average.
This scatter puts the 20 signal days on a BP-versus-XLE 10-day return grid. If the decoupling idea were working, the dots would sit consistently on the BP-beats-XLE side; instead the two averages are nearly identical, BP at +4.03% and XLE at +3.96%. The points range from about -7% to +15% for BP and -6% to +9% for XLE, so individual signal days can look very different even though the group as a whole does not break away from the sector.
Signal vs non-signal summary
| Group | N | Mean excess | Median excess | Std | Win rate |
|---|---|---|---|---|---|
| Signal: high BP news & Brent - | 20 | 0.0007 | -0.0108 | 0.0436 | 0.4 |
| Non-signal days | 702 | -0.0016 | -0.0016 | 0.036 | 0.4687 |
Signal-day event log
| Date | BP sentiment | Brent daily return | BP fwd10 | XLE fwd10 | Excess fwd10 |
|---|---|---|---|---|---|
| 2025-12-29 | 0.272 | -0.0094 | 0.03 | 0.0533 | -0.0233 |
| 2025-12-31 | 0.285 | -0.0152 | 0.0152 | 0.0645 | -0.0493 |
| 2026-01-26 | 0.27 | -0.0067 | 0.0658 | 0.0901 | -0.0243 |
| 2026-02-02 | 0.272 | -0.0627 | -0.0025 | 0.0729 | -0.0754 |
| 2026-02-12 | 0.423 | -0.024 | 0.0594 | 0.0394 | 0.0199 |
| 2026-02-17 | 0.305 | -0.0147 | 0.0508 | 0.0511 | -0.0003 |
| 2026-03-04 | 0.362 | -0.0207 | 0.1546 | 0.042 | 0.1126 |
| 2026-03-19 | 0.485 | -0.0596 | 0.0292 | 0.0041 | 0.0251 |
| 2026-04-14 | 0.322 | -0.0372 | 0.0152 | 0.0357 | -0.0206 |
| 2026-04-15 | 0.388 | -0.0317 | 0.0183 | 0.0593 | -0.041 |
| 2026-05-05 | 0.461 | -0.0317 | -0.0013 | 0.0349 | -0.0362 |
| 2026-05-21 | 0.332 | -0.0284 | -0.0458 | -0.0309 | -0.0149 |
| 2026-05-28 | 0.507 | -0.0169 | 0.0277 | -0.0008 | 0.0285 |
| 2026-06-05 | 0.32 | -0.0171 | -0.0717 | -0.0559 | -0.0158 |
| 2026-06-24 | 0.493 | -0.0476 | 0.0208 | 0.0275 | -0.0067 |
| 2026-07-01 | 0.325 | -0.0173 | 0.138 | 0.0807 | 0.0574 |
| 2026-07-09 | 0.251 | -0.0267 | 0.1391 | 0.0836 | 0.0555 |
| 2026-07-10 | 0.347 | -0.0016 | 0.1235 | 0.0883 | 0.0352 |
| 2026-07-15 | 0.277 | -0.0073 | 0.0505 | 0.0368 | 0.0137 |
| 2026-08-14 | 0.338 | -0.0001 | -0.0108 | 0.0149 | -0.0256 |
The takeaway
Based on the past ~36 months, the answer is no: high BP news sentiment on a down-Brent day has not reliably produced BP outperformance versus XLE over the next 10 sessions. Only 20 days qualified, and although the average BP-XLE spread was +0.07 percentage points on those days versus -0.16 percentage points on ordinary days, the median signal-day outcome was -1.08 percentage points and BP beat XLE just 40% of the time, below the 47% base rate. With p-values near 0.82 and 0.95, that gap is indistinguishable from random noise. The mean-vs-median split is the real tell: a few strong positive days are carrying the small average edge, so this does not look like a consistent commodity-decoupling signal.
The fine print
- Only 20 signal days—too few to establish a firm edge; the p-values are consistent with thin, noisy data.
- Signal-day outcomes are highly dispersed (std ≈ 4.4 percentage points), so a handful of large moves can dominate the average.
- The sentiment score is a daily average of headlines; mixed or conflicting coverage may dilute the company-specific signal.
- Definitions matter: the 20-session/80th-percentile threshold, Brent day alignment, and the XLE benchmark are all judgment calls, and different choices could change the event set.