Energy's Chart Signals Keep Losing to the Benchmark

The week's research output has a single, unmistakable center of gravity: energy. Every finding published since Monday touches crude, XLE, or a producer — COP, DVN, EOG, XOP — and read together they form an unusually consistent story about how hard it is to beat the index with a pattern drawn on a chart.

Two backtests, one uncomfortable benchmark

Start with the mechanical strategies. Buying XLE when it crosses above its 20-day simple moving average, run across 19 closed trades, produced a +29.20% return on $100,000 of starting capital and a 58% win rate. That reads like a working system until SPY buy-and-hold is placed beside it: +85.28% over the same window, a trailing gap of 56.08 points. The EOG variant — buying when the day's high-low range lands in the top quintile — is starker still. Fifteen closed trades, an 80% win rate, and +3.81% in total, trailing that same benchmark by 81.47 points.

The detail worth lingering on is the shape of the outcomes. XLE's best single trade made +6.79% and its worst lost -2.20%. EOG's best made +3.75% and its worst lost -4.15%. When the worst trade is larger than the best one, a high win rate is doing less work than it appears to. Frequency of correctness and size of correctness are different quantities, and only the second one compounds.

Where lead-lag keeps breaking down

Three separate findings this week tested the idea that energy assets telegraph each other, and each came back negative or ambiguous. The most interesting is the XLE-to-Brent lead-lag. The one-day beta from XLE into next-day Brent was 0.71 in low-volatility regimes and 0.22 in high-volatility ones — the opposite sign from the front-running-in-stress thesis. The intuition that turbulence makes cross-asset information travel faster simply does not show up in the numbers. Raw correlations tell the same story, at 0.25 in calm periods.

The relative-value setups fare no better. When Brent's 20-day return was negative while COP's was positive, COP lagged XOP by an average of 0.41% over the following 20 sessions, with a median shortfall of 0.69%, coming out ahead on only 44.6% of 121 trigger days. With p = 0.32, that gap is not statistically distinguishable from zero. The DVN setup at least pointed in the predicted direction: after an up-close with a negative final 30 minutes and Brent up 0.5%, DVN trailed XOP by 0.67% on average over the next 10 sessions and was the laggard in 58% of 96 cases. But DVN already lagged XOP by 0.20% unconditionally, which makes the pattern look more like a habit than a signal.

The one lean worth noting

Against that backdrop, the Fed/FOMC finding stands out. On the 37 days when Fed or FOMC headlines spiked while Brent's 20-day return was positive, XLE beat SPY by an average of 3.49% over the following 20 sessions, with a median excess of 8.11% and XLE ahead 68% of the time. On the other 409 days the average excess was 0.03% — effectively flat. That is a narrow, conditional setup on a modest sample, and it reads as a lean rather than a law. But it has a different character from the rest of this week's material: it is not a chart pattern, it is a description of a macro state in which two conditions hold at once.

The through-line here is less about energy being unforecastable than about the gap between signals that describe a market state and signals that describe the shape of a line. The line-based ones mostly trailed a benchmark that simply sat still. The state-based one, on this sample at least, did not.