Energy Backtests Are Winning Trades and Losing to the Benchmark
The news tape is silent today, which leaves the research feed as the loudest thing on the desk. And the research feed has spent the last three sessions chewing through one energy setup after another, arriving at an uncomfortable shared conclusion: these strategies win a lot of trades and still lose the race.
The win rate is not the edge
The cleanest-looking result of the batch is a crude-conditioned entry in VLO: buy at the close when Brent crude closes up more than 1.5% on the day. It returned +53.48% on $100,000 across 45 closed trades with a 76% win rate. That is a real hit rate — three out of four trades green — with a best trade of +6.14% against a worst of -7.86%. It also finished 22.85 points behind SPY buy-and-hold, which returned +76.34% over the same window.
A 76% win rate bought roughly two-thirds of what simply holding the index delivered, plus 45 entry decisions and the drawdown that came with them. Then there is DVN, bought when it closed above its prior 20-day high alongside a Brent condition: -5.69% across 22 trades, a 36% win rate, and a gap of 82.03 points to the benchmark. The tell is inside that record — DVN's best single trade was +13.67%, the fattest winner in the whole group, sitting inside the only strategy that lost money. That is the classic shape of a setup that pays large occasionally and bleeds in between, and the worst trade was -11.81%. PSX's version, triggered when its 10-day total return clears 4 percentage points, managed +5.85% with a 54% win rate and trailed by 70.48 points.
Where the signal actually leans
The more interesting work from this week is not about entries at all. XOM's daily beta to Brent comes in at 0.26 on down days versus 0.16 on up days — directionally the asymmetry people assume, but a 0.10 gap the platform itself declines to call statistically clean. What is clean is the linkage itself: the up-day beta of 0.16 carries a t-stat of 3.34 and p = 0.0008. So the oil sensitivity is not in question, only the asymmetry. That distinction matters, because a lot of oil-equity folklore is built on the asymmetry rather than the linkage.
The LNG-versus-XLE rotation study lands in the same bucket. On high-volatility Brent days, LNG averaged +3.20% over the following 20 sessions against XLE's +1.64%, a 1.56% spread in LNG's favor. LNG also won 56.6% of those event days versus 36.7% on non-event days, with the event-versus-non-event contrast at +1.92% and p = 0.009. That is a strong-enough result to take seriously and weak enough that the write-up calls it a lean, not a proven edge. Both of these studies condition on regime — volatility state, directional day type — and both produce tighter statistics than the candle-pattern entries above.
Seven trades is not a track record
One finding deserves a flag on sample size alone. The VLO study built on a compressed 10-day average true range, buying when that range sits in the bottom quintile, returned +20.26% across seven closed trades with a 57% win rate, a best trade of +9.88% and a worst of -3.28%. It trailed SPY by 56.07 points. Seven observations is a hypothesis wearing a backtest's clothing; the soft worst trade means the sample has barely been stress-tested. This looks like the kind of result that should be filed as "worth watching" rather than "worth trusting."
The pattern across the batch seems clear. Regime-conditioned research — XOM's t-stat, the LNG/XLE p-value — is where the statistically defensible signal keeps surfacing. Discrete entry rules produce flattering hit rates and equity curves that lag a benchmark that has simply been generous. In a tape where buy-and-hold returned +76.34%, doing something frequently is itself a cost, and 76% win rates do not pay it back. The energy complex may well hold a rotation story this autumn; the data leans toward finding it in volatility regimes rather than in a single green candle in Brent.