MPC vs XLE after refined-margin momentum signals: 20d forward performance of MPC-Brent spread events
The intuition is easy to state: when a refiner outruns crude oil in a rising tape, product cracks are probably widening, and that margin momentum should carry the stock past the oil-beta sector. What's less obvious is whether the tape actually pays for it.
Across 700 eligible sessions, 33 fit the condition — MPC's trailing 20-day return ahead of Brent's by more than 10 percentage points while Brent itself was positive. Over the next 20 sessions MPC did beat XLE, but the average excess was 0.49% and the win rate only 54.5%. The p-value of 0.70 says that edge is indistinguishable from noise.
The magnitude is worth taking seriously if you accept a small sample; overlapping forward windows and a broad peer set argue for caution. The full event study sits below.
Over the past ~3 years, when MPC's trailing 20-day return beats Brent crude's by more than 10 percentage points while Brent crude's 20-day return is positive, does MPC outperform XLE over the next 20 trading sessions? I expect a refiner outrunning crude in a rising-oil tape to be pricing widening product cracks, and that margin momentum should keep MPC ahead of the oil-beta sector.
How this was measured
MPC and XLE minute bars were compressed to daily closes, and Brent crude was aligned from brent_daily_df with a one-business-day lag for point-in-time availability. For each eligible session, I computed trailing 20-session returns for MPC and Brent; a qualifying event required MPC's trailing return to exceed Brent's by more than 10 percentage points while Brent's trailing return was positive. The response was measured as the forward 20-session return of MPC minus XLE, starting at the signal-date close. Event periods were compared with all eligible non-event sessions, using one-sample and Welch t-tests alongside hit rates and medians.
The key numbers
Reading the numbers
Out of 33 times the setup appeared, MPC beat XLE over the following 20 sessions only 54.5% of the time — barely better than a coin flip — and the average edge of +0.49% comes with a p-value of 0.70, meaning a result that large could easily show up by chance alone.
The charts
This box plot lines up the forward 20-day MPC-minus-XLE return on the 33 qualifying event days against the same measure on all 700 eligible days. The thing to notice is that the ordinary-days box sits higher: its mean of +1.71% beats the event group's +0.49%, which is the -1.22% edge gap the report flags. Both boxes have long whiskers reaching roughly -12% to +14% on event days and wider still across the full sample, so the spread of outcomes dwarfs the small difference between the two centers.
These four bars strip the result down to averages: on event days MPC gained 3.17% and XLE gained 2.68% over the next 20 sessions, a gap of 0.49 percentage points. The fourth bar is the one that matters most for the question — on all eligible days the same MPC-minus-XLE gap averaged 1.71%, so the signal days actually produced a smaller relative gain than a randomly chosen day. In plain terms, the refiner-outrunning-crude setup did not translate into a bigger lead over the oil-sector benchmark.
This scatter plots each of the 700 sessions by how far MPC's trailing 20-day return had run ahead of Brent's (x-axis) against the forward 20-day MPC-minus-XLE excess return (y-axis). The entry spread ranges from -53% to +37% with a mean of +2.1%, while forward excess ranges from -16% to +28% around a mean of +1.7%, and the points form a diffuse cloud rather than a clean upward line. Reading it against the question: simply having a larger head start on crude — even well past the +10 percentage point threshold — does not visibly line up with a larger forward advantage over XLE, which is consistent with the weak, statistically inconclusive event results.
Most recent qualifying events
| date | MPC 20d | Brent 20d | spread | fwd MPC 20d | fwd XLE 20d | fwd excess |
|---|---|---|---|---|---|---|
| 2024-04-08 | 0.2164 | 0.0718 | 0.1446 | -0.1635 | -0.0404 | -0.1232 |
| 2024-04-09 | 0.1818 | 0.0807 | 0.1011 | -0.1475 | -0.0458 | -0.1017 |
| 2025-05-14 | 0.3392 | 0.0239 | 0.3154 | 0.0072 | 0.022 | -0.0148 |
| 2025-05-30 | 0.1801 | 0.0087 | 0.1714 | 0.0345 | 0.0498 | -0.0154 |
| 2025-06-02 | 0.1612 | 0.015 | 0.1462 | 0.0629 | 0.042 | 0.0209 |
| 2025-09-10 | 0.1333 | 0.0119 | 0.1214 | 0.0375 | 0.0044 | 0.0331 |
| 2025-09-11 | 0.1583 | 0.0039 | 0.1545 | 0.0403 | -0.0054 | 0.0457 |
| 2025-09-12 | 0.1204 | 0.0067 | 0.1136 | 0.0038 | -0.027 | 0.0308 |
| 2025-09-18 | 0.12 | 0.0148 | 0.1052 | -0.0155 | -0.0448 | 0.0293 |
| 2025-09-19 | 0.1182 | 0.0182 | 0.1001 | -0.0142 | -0.02 | 0.0058 |
| 2026-02-11 | 0.1885 | 0.0858 | 0.1027 | 0.1157 | 0.049 | 0.0667 |
| 2026-02-13 | 0.1382 | 0.0135 | 0.1247 | 0.1365 | 0.0633 | 0.0732 |
| 2026-02-18 | 0.1468 | 0.0418 | 0.1049 | 0.1976 | 0.0675 | 0.1301 |
| 2026-02-26 | 0.1561 | 0.0058 | 0.1503 | 0.2503 | 0.1265 | 0.1237 |
| 2026-02-27 | 0.1392 | 0.0107 | 0.1285 | 0.2632 | 0.1253 | 0.1378 |
| 2026-03-02 | 0.1886 | 0.0045 | 0.1841 | 0.1731 | 0.0957 | 0.0773 |
| 2026-03-03 | 0.1719 | 0.0691 | 0.1028 | 0.179 | 0.0849 | 0.0942 |
| 2026-07-17 | 0.2882 | 0.0091 | 0.2791 | 0.1313 | 0.0715 | 0.0598 |
| 2026-07-20 | 0.3048 | 0.0583 | 0.2466 | 0.1309 | 0.0799 | 0.051 |
| 2026-07-21 | 0.2907 | 0.0963 | 0.1944 | 0.1416 | 0.0858 | 0.0558 |
The takeaway
Across 33 qualifying events, MPC outperformed XLE over the next 20 sessions in 54.5% of cases, with mean excess return +0.49% (MPC +3.17% vs XLE +2.68%). One-sample p=0.7030, so the margin-momentum signal did not hold up at the 5% level. Relative to all eligible days, the event edge was -1.22%. The conditional advantage is meaningful in magnitude if the sample is accepted, but overlapping forward windows and the broad XLE peer set warrant caution.
The fine print
- Brent crude is from brent_daily_df and was lagged one business day to respect T+1 publication timing; same-day settlement prints could shift event alignment slightly.
- Forward returns are close-to-close from the signal date and do not model next-open execution slippage, transaction costs, or ETF expense drag.
- XLE is a broad energy-sector ETF, not a pure refining-margin or oil-beta exposure; relative moves include integrateds, upstream, and non-MPC names.
- Event signals and their 20-session forward windows can overlap and cluster, so the t-tests overstate independence; hit rate and median are more robust than p-values.