VLO vs MPC relative-strength persistence after 10-day outperformance with flat/down Brent
Relative strength only pays when it persists, and this version of the setup didn't. The question under test: after Valero has outrun Marathon Petroleum by more than three percentage points over ten sessions, with Brent crude flat or down, does VLO keep its edge over the following two weeks? Across 19 non-overlapping events in the past 36 months, the answer is no. VLO's median forward spread versus MPC was about -1.2 points, and MPC won 11 of the 19 follow-ups. The effect isn't merely absent — the point estimate points the wrong way.
The full report below details the method, the raw 72 qualifying days that collapse into these 19 deduplicated events, and the t-test and sign-test results. If a widening-crack-spread story is meant to explain VLO's relative strength, this Brent-conditioned screen doesn't capture it. The numbers are worth reading before leaning on the idea again.
Over the past ~3 years, when VLO's 10-day total return outpaces MPC's by more than 3 percentage points while Brent crude's 10-day return is flat to down, does VLO outperform MPC over the next 10 trading sessions? I expect VLO's relative strength to persist as a widening crack spread keeps refining margins elevated before product cracks mean-revert.
How this was measured
Daily closes for VLO and MPC were built from minute bars, joined with Brent's daily crude price, and restricted to the most recent 36 months. For each trading day, the 10-day total return of each refiner was computed, along with the 10-day return spread VLO minus MPC. Brent's 10-day return was computed using a one-business-day lagged Brent series to respect T+1 macro timing. A qualifying day is one where the VLO-MPC 10-day spread exceeds 3 percentage points and Brent's lagged 10-day return is flat to down, defined as <= +0.5%. Forward 10-session returns for both refiners were then measured, and non-overlapping events were selected by requiring at least 11 trading days between anchors. The conditional VLO-MPC forward spread was tested against zero with a one-sample t-test and a binomial sign test.
The key numbers
Reading the numbers
The 19 signal events produced just 8 VLO wins (42%) over the next 10 sessions, with an average VLO-minus-MPC return of about -0.6 percentage points. The test p-value of about 0.42 means this is nowhere near a statistically reliable edge — the data do not show persistence.
The charts
The histogram covers all 19 non-overlapping signal events, with forward VLO-minus-MPC spreads ranging from about -6.5 to +6.4 percentage points. The average lands at -0.6 percentage points, meaning the typical event was followed by MPC doing slightly better than VLO, not the other way around. That is the opposite of what the persistence hypothesis would predict.
This bar chart puts the signal-day average of -0.62 percentage points next to the all-day baseline of about +0.05 percentage points. The event bar is clearly below the baseline, so qualifying days were followed by a worse relative result for VLO than the average day. In short, the trigger condition does not select days when VLO tends to keep outperforming MPC.
Non-overlapping event details
| event_date | vlo10 | mpc10 | spread10 | brent10 | fwd_vlo | fwd_mpc | fwd_spread |
|---|---|---|---|---|---|---|---|
| 2023-12-07 | -0.0155 | -0.0511 | 0.0356 | -0.0982 | 0.0801 | 0.0783 | 0.0018 |
| 2024-03-19 | 0.1523 | 0.1218 | 0.0305 | -0.0013 | 0.0893 | 0.1028 | -0.0135 |
| 2024-04-30 | -0.0486 | -0.1145 | 0.0659 | -0.0264 | -0.0248 | -0.0345 | 0.0098 |
| 2024-05-16 | 0.0103 | -0.0456 | 0.0559 | -0.0427 | -0.0026 | 0.0104 | -0.0129 |
| 2024-07-25 | 0.0708 | 0.0269 | 0.0438 | -0.0409 | -0.0277 | 0.0186 | -0.0463 |
| 2024-10-16 | 0.0017 | -0.0405 | 0.0422 | -0.0215 | -0.0519 | -0.0863 | 0.0344 |
| 2024-11-01 | -0.0617 | -0.1009 | 0.0392 | -0.0151 | 0.0773 | 0.1051 | -0.0278 |
| 2025-03-06 | -0.0944 | -0.1249 | 0.0305 | -0.071 | 0.0484 | 0.0983 | -0.0499 |
| 2025-07-09 | 0.0972 | 0.0618 | 0.0354 | -0.0248 | -0.008 | 0.0165 | -0.0245 |
| 2025-08-12 | -0.0632 | -0.1107 | 0.0475 | -0.0495 | 0.1009 | 0.117 | -0.0161 |
| 2025-09-03 | 0.1209 | 0.0895 | 0.0314 | -0.0013 | 0.0522 | -0.0021 | 0.0543 |
| 2025-10-23 | 0.0679 | 0.0288 | 0.0391 | -0.0762 | 0.0188 | 0.0067 | 0.0121 |
| 2025-11-07 | 0.0267 | -0.0088 | 0.0355 | -0.0439 | -0.0048 | -0.0112 | 0.0063 |
| 2025-12-18 | -0.0712 | -0.1143 | 0.0431 | -0.0493 | 0.1124 | 0.048 | 0.0645 |
| 2026-01-08 | 0.1784 | 0.0925 | 0.0859 | -0.0183 | -0.0234 | -0.0117 | -0.0117 |
| 2026-02-26 | -0.0015 | -0.043 | 0.0416 | -0.0045 | 0.1672 | 0.1658 | 0.0014 |
| 2026-04-15 | -0.0431 | -0.0822 | 0.0391 | -0.0262 | 0.0714 | 0.0806 | -0.0093 |
| 2026-06-26 | 0.0211 | -0.0161 | 0.0372 | -0.2297 | 0.1385 | 0.1644 | -0.026 |
| 2026-08-05 | -0.0251 | -0.07 | 0.0449 | -0.0786 | 0.1441 | 0.2087 | -0.0647 |
The takeaway
No — this version of the relative-strength setup has not been followed by continued VLO outperformance. Across the 19 non-overlapping qualifying events (72 raw matching days deduplicated), VLO's forward 10-day mean was +5.0% versus MPC's +5.7%, making the average VLO-MPC spread about -0.62 points; the median spread was even more negative, about -1.2 points. VLO won only 8 of the 19 follow-ups (42%). With a t-test p-value of about 0.42 and a sign-test p-value of about 0.65, the result is statistically no different from a coin flip, and the point estimate points the wrong direction, not just an empty null. The sample is small at 19 events, so this doesn't prove there's no effect, but it gives no usable support for the persistence narrative. Practically, this screen shouldn't be relied on as a sign that VLO will keep beating MPC over the next 10 sessions — if a crack-spread story is driving relative strength, this crude-price/equity-return setup isn't capturing it.
The fine print
- Only 19 non-overlapping events; with such a small sample the tests are underpowered, so a modest effect can't be ruled out even though the observed sign is negative.
- The screen uses Brent as a lagged crude proxy and VLO/MPC equity returns, not actual product-crack or refinery-margin data.
- 'Flat-to-down Brent' is defined as a 10-day return up to +0.5%; using <=0% or a different lag would change the qualifying dates.
- VLO and MPC share broad refining exposure, so the relative spread can reflect idiosyncratic or plant-level news that isn't controlled for.