Six Energy Setups, One Lesson: Sample Size Beats Storytelling

The news tape is quiet this weekend, which suits the research output fine — the platform spent the week stress-testing six energy setups, and the results are a masterclass in how thin a plausible macro story gets once you count the days it actually applies to.

Six ideas, three ways to be wrong

The cleanest example is the simplest. Buy WMB at the close when a 30bp spike in the 10-year yield lines up with a weak tape in the stock: over three years, only 11 sessions met both conditions. On those days WMB averaged +2.41% over the following 20 sessions against +1.59% for XLE, a mean edge of roughly 0.8 points in WMB's favor. It sounds like a finding until you look at the median gap, which was -0.49% — WMB actually underperformed more often than not. A couple of good draws are carrying the average.

The EQT version is nearly identical in shape. When Brent's trailing 20-day return lands in the top quintile — a gain of roughly 6.3% or more — the gas producer averaged -0.58% over the next 20 sessions versus +1.90% for XLE, a 2.5-point lag in exactly the direction the decoupling thesis predicts. Strip out the overlapping forward windows, though, and that gap rests on just 15 readings. The XOP test rested on six episodes. OIH looked like the sturdy one at 135 signal days, with a +1.10% average spread over XLE — and even that study conceded the edge comes from a handful of outsized episodes rather than broad consistency.

Six for six, in the wrong direction

Then there is the one that flipped. The intuition is standard reflation logic: when the 10Y-2Y spread steepens and Brent is bid, small-cap exploration and production should out-lever the wider energy sector. Across the six non-overlapping episodes where the curve steepened at least 15bps over 20 sessions while oil was up, XOP lagged XLE by an average of 2.26% over the following 20 sessions. It underperformed in all six. On ordinary non-signal days the gap was 0.25%.

That unanimity matters more than any average. A six-for-six wrong-way result is a different species of evidence than +0.8 points built on eleven days. It reads like the reflation trade is already embedded in small-cap E&P prices by the time the signal fires — the setup's marginal buyer may be whoever is running the screen.

The strategy that worked and still didn't

Against that backdrop, the XOM sentiment study looks almost cheerful. Buying at the close whenever the stock's daily news sentiment score sat in the bottom quintile produced +15.86% on $100,000 across 31 closed trades, with a 43% win rate, a best trade of +7.96% and a worst of -3.85%. The shape is honest — losers slightly more frequent than winners, winners meaningfully bigger.

And it still trailed SPY buy-and-hold by 69.41 points, because the index returned +85.28% over the same window. That gap is the argument. Positive expectancy in isolation says almost nothing about whether a signal deserves capital; the opportunity cost of not holding the benchmark belongs inside the return calculation, not as a footnote.

The week's strangest result was the TTE question, which could not be answered at all: zero sessions in three years where a 30-day EPS revision was even measurable, because the estimate dataset holds a single current snapshot per quarter rather than a revision history. The Brent-only substitute across 306 sessions came back unusually clean, but the interesting part is the failure mode. That one was the data's fault, not the hypothesis's, and it is the only genuinely fixable item on the list.

The pattern across the set is consistent: the more specific the setup, the fewer the observations, and the more the headline number collapses into a story about two or three days. In energy, where dispersion inside the sector runs wide and everyone reads the same macro inputs, base rates do most of the talking. This week they talked over nearly every clever idea on the board.