Does a Brent 50d-MA trend gate condition VLO's summer seasonal edge vs XLE?
If the thesis is that a summer crack-spread bid only lifts VLO above the broad energy complex when crude is trending up, the data disagrees. Testing roughly three years of 10-day forward relative returns, May–September versus November–March, stratified by whether Brent sat above its own 50-day moving average, produced the opposite of what that logic predicts. The above-trend summer window averaged worse than winter by -2.65 percentage points; the below-trend regime showed a modest positive summer edge.
The catch is thin coverage — just 14 above-trend summer observations — and wide confidence intervals. The formal interaction test lands well within noise territory at p = 0.13, so the apparent flip can't be separated from random variation. The full methodology and statistics are below. The headline is that no reliable trend-gated seasonal edge shows up here.
Over the past ~3 years, does VLO's forward 10-day return relative to XLE during May–September exceed its November–March average only when Brent crude is above its 50-day moving average? I expect the summer-driving crack-spread bid to lift refiners relative to the broader energy sector only when the crude trend confirms strong product demand; below-trend Brent should keep the seasonal edge muted.
How this was measured
VLO and XLE minute bars were resampled to daily closes and joined on common trading days. On each day t, the forward 10-trading-day relative return is computed as VLO's 10-day total return minus XLE's 10-day total return. May through September is the summer-driving period; November through March is the cold-season comparator. Brent's daily close is reindexed onto the VLO/XLE calendar with forward-fill, a 50-day moving average is built, and the trend gate uses the prior trading day's Brent-vs-MA reading to avoid same-day publication lag. The seasonal edge is then measured separately in the two Brent regimes with Welch t-tests, and an OLS interaction term (summer × above-trend) with HAC errors is estimated to formally test whether the Brent trend changes the summer-over-winter gap.
The key numbers
Reading the numbers
The key number is the summer-over-winter edge: -2.65% when Brent is above its 50-day average, +0.33% when below. That is backwards from the hypothesis, and both season comparisons fail to reach significance (p=0.06, p=0.36), so no reliable summer advantage exists.
The charts
The first pair of bars is the tell: when Brent is above its 50-day moving average, summer returns average -0.9% while winter averages +1.7%, meaning the supposed summer edge is actually negative. When Brent is below trend, summer (+1.1%) and winter (+0.8%) are nearly the same. This is the opposite of the expected above-trend-only seasonal bid, and the interaction test (p=0.13) says that gap is not statistically reliable.
Look at the small Summer, Brent>50d MA group: just 14 observations, mean -0.9%, range from -7.8% to +5.6%. Winter above-trend, by contrast, has 37 observations and a mean of +1.7%. The below-trend groups have over 300 observations each and nearly identical means, so the reversal seen in the bar chart rests on a very small cell and should be treated as provisional.
Brent spent much of the sample in the $60-$80 range but spiked to $127.61 and ended around $96.95, while the 50-day average moved from $91.31 down to $87.32. The Brent line and its 50-day average cross back and forth over time, so the above/below-trend regime is not a stable state. That matters because only 14 summer observations fell in the above-trend regime, making the key comparison quite thin.
Forward 10d VLO−XLE relative return by season and Brent trend regime
| Regime | N | Mean | Median | Std |
|---|---|---|---|---|
| Summer, Brent>50d MA | 14 | -0.0091 | -0.0089 | 0.047 |
| Summer, Brent<50d MA | 306 | 0.0114 | 0.0063 | 0.0482 |
| Winter, Brent>50d MA | 37 | 0.0174 | 0.0095 | 0.0271 |
| Winter, Brent<50d MA | 305 | 0.0081 | 0.0035 | 0.041 |
The takeaway
No — the Brent 50-day trend gate doesn't do what the hypothesis predicted. Over the roughly three-year sample, VLO's May–September forward 10-day return versus XLE was actually worse when Brent was above its 50-day moving average: summer averaged -0.91% versus +1.74% in winter, a -2.65 percentage-point seasonal edge. When Brent was below trend, the summer edge was +0.33 percentage points — the opposite sign. The formal interaction test says the difference between those two regimes is well within the range of noise (p = 0.13), and the within-regime seasonal gaps also miss conventional significance (p = 0.06 and p = 0.36). So this is basically inconclusive — more noise than signal — not evidence that a summer crack-spread bid only works with Brent above trend. The above-trend summer bucket has just 14 observations, which makes that negative point estimate fragile. Practical takeaway: there's no reliable seasonal edge here to hang a view on, so the Brent-trend filter is not supported by the data.
The fine print
- Only 14 observations in the above-trend summer cell; that estimate can flip with a handful of days.
- Forward 10-day windows overlap, so Welch p-values overstate effective sample size; the interaction p is more reliable but still approximate.
- Month buckets are rough proxies for driving season, and refiner-specific events like earnings, outages, or M&A can dominate a 3-year sample.
- Brent's close is lagged one trading day; a same-day signal would not have been executable in real time.