In Energy, the Relative Bets Worked and the Directional Ones Didn't
The week's backtests all had oil in them, and almost all of them finished the same way: a perfectly reasonable story about crude, a strategy that traded on it, and a scoreboard that read worse than doing nothing. The benchmark across every one of these tests is the same +76.34% from SPY buy-and-hold. That number is the whole argument.
The directional bets kept losing to a nap
Start with the ones that picked a direction. Buying XOM when its trailing 20-day realized beta to Brent crossed whatever threshold the study used returned -8.10% on $100,000 across 36 closed trades, with a 20% win rate — one in five. The best trade made +12.34%, the worst lost only -3.45%, so it wasn't a single blow-up doing the damage; it was the steady grind of being wrong. That strategy finished 84.44 points behind SPY.
Buying TTE when its trailing 20-day total return lagged Brent did better in absolute terms: +35.85%, 26 closed trades, a 54% win rate, best +9.61%, worst -3.75%. It still trailed the benchmark by 40.49 points. And buying XLE when the XLE/SPY ratio closed at a 20-day low and Brent met the study's condition returned +3.60% over 13 trades with a 67% win rate — the highest hit rate of the three, the smallest payoff, and a 72.73-point gap to SPY.
Notice what the win rates do. The 67% strategy and the 20% strategy both lost to the benchmark, just from opposite ends. Hitting often on small edges and hitting rarely on coin flips converge on the same place when the alternative is simply holding the index.
The failures are the interesting part
Two of the negative results are worth more than the positive ones, because they kill stories that sound like they ought to work.
XOM's intraday range on days Brent dropped more than 2% averaged 5.84% of the prior close, against 5.32% when Brent rose more than 2%. The lean points the direction you'd expect — selloffs are more violent — but half a percentage point is not a tradeable asymmetry, and the medians (4.31% on selloff days, no less flat on the other side) suggest the average is carried by a handful of sessions. This is the recurring failure mode of volatility screens: the effect is real, and the effect is too small.
Then the curve trade. After a 15bp-plus steepening in the 2s/10s with Brent's 20-day return positive, FANG trailed XOM over the next 20 sessions by an average of -0.42%, beating XOM on only about 42% of 36 event days. The median spread was -4.29%, which means the typical event was a loss rather than a scratch. A clean macro story — steeper curve, positive momentum, levered producer outperforms — and the result is the mild opposite of the thesis.
The one screen that did work was cross-sectional
The VLO-versus-XLE test is the outlier, and its structure matters more than its return. Across 27 qualifying days, the forward 20-session VLO-minus-XLE spread averaged +7.5%, and VLO was ahead in 81.5% of cases. On the other 421 eligible days, the same spread averaged only +2.7%, so the screen itself added roughly 4.8 points. That looks like a genuine conditional effect rather than plain energy-sector drift.
But read the distribution: mean +7.5%, median +10.1%. The median sits above the mean, which means the losers in the sample are larger than the winners. An 81.5% hit rate with a left tail isn't comfort — it's a setup that pays often and occasionally gets hurt badly.
Across the batch, the pattern is consistent and a little deflating. Directional, macro-triggered energy trades have been a reliable way to underperform a passive benchmark; the only setup showing a real conditional edge was a relative-value spread between two refiners, and even that arrived with a fat left tail. Small samples, one week of reports, no news flow to lean on — so this is a lean, not a law. But the lean is pointing one way.