Energy Relative-Value Edges Keep Losing to Plain S&P Exposure

The headline feed came up empty today, which leaves the research queue as the only narrator. It happens to be an energy-heavy stretch — a refiner, a pipeline operator, a shale gas name, a couple of integrateds — and read end to end, this week's findings tell one story worth sitting with: the difference between beating your sector and beating the market, and how often a clever-looking filter quietly subtracts value.

A 59% win rate still trails badly

Three mechanical long strategies closed out in the queue. Buying ET at the close when its daily news-sentiment score sits in the bottom quintile returned +28.61% on $100,000 across 94 closed trades, a 59% win rate. Buying TTE when its trailing 20-day total return lags Brent crude returned +35.85% across 26 trades, 54% winners, best single trade +9.61%, worst -3.75%. Buying EQT on a compressed 10-day average true range returned -4.27% across just seven trades, a 29% win rate.

Over the same windows, SPY buy-and-hold returned +76.34%. So the ET strategy trailed by 47.72 points, TTE by 40.49, EQT by 80.61. The interesting number isn't the return, it's the shape. The ET system won more often than it lost and still got lapped, because its best trade made +3.95% while its worst gave back -10.43%. A frequent small winner with an occasional large loser is a distribution that looks healthy in a win-rate column and grim in a terminal wealth column. This looks less like a broken signal than a sizing and exit problem, but either way the benchmark doesn't care.

Edges that survive the error bars, barely

The relative-value work was more careful. On hot CPI readings with crude sliding, XOM averaged +2.36% over the next 20 sessions versus XLE's +1.93% across 97 such days — a +0.43% spread, XOM ahead 57% of the time, t = 2.10, p = 0.038. That's a roughly 4-in-100 chance of noise, which reads as respectable, but the note's own caution is the right one: a 0.43-point monthly edge is thin enough to disappear into frictions. Statistically real and economically useful are different claims.

The WMB setup — Brent 20-day weakness plus a WMB estimate revision of at least +1% — is the opposite problem. WMB beat XLE in all three times the setup fired, mean outperformance 2.96%, median 3.46%, widest case 3.96 percentage points. Three episodes. The research itself labels that a lean, not a proven pattern, and that honesty is the most valuable thing in the batch.

The filter that did no work

The cleanest lesson is VLO against XLE. Screening for top-quintile retail-sales growth plus a positive 20-day Brent move produced 49 days on which VLO averaged +1.26% while XLE lost 1.81% — a 3.07-point edge, with VLO ahead on roughly 90% of those days. Impressive until you see the baseline: across all 421 eligible days, that same VLO-minus-XLE spread averaged 3.32%. The screen didn't find the edge. It found slightly worse days than doing nothing, and then took credit for the sector's structural spread.

That's the theme running through everything here. Relative-value research isn't mostly about finding a condition that precedes outperformance; it's about proving the condition beats the unconditional baseline, and about whether any of it clears a simple index. On a day with no headlines to blame or credit, the queue is a reminder that the boring comparison is usually the one doing the grading.