The Reflation Reflex in Energy Keeps Failing the Platform's Backtests

The headline feed is quiet today, which turns out to be a decent excuse to read the research log instead. And the research log, over the past three sessions, has produced an unusually tight cluster of energy-sector findings — and they all lean the same way. The reflation reflex, the one that says rising crude plus a steeper curve means buy the levered upstream names, keeps failing in the platform's own data. Not marginally. Usually in the opposite direction.

The reflex that keeps flipping

The cleanest test: the 10Y-2Y spread steepening by at least 15 bps over 20 sessions while Brent rose. Six non-overlapping episodes qualified. Over the following 20 sessions, XOP lagged XLE by an average of 2.26%, and XOP underperformed in all six — not a close call in any of them. On ordinary non-signal days the same gap ran just 0.25%. Six for six is a striking sign pattern, and it deserves the caveat attached to it: six episodes is not a sample anyone should lean hard on.

The same shape shows up elsewhere. Across eight positive quarterly earnings surprises at XOM with complete data, XOM averaged +1.16% over the next 20 sessions while XLE averaged +2.52% — a shortfall of roughly 1.35 percentage points, median -0.94%, with XOM beating the sector ETF in only three of the eight. And when the 2-year yield rose at least 30 bps over 20 days while Brent's 20-day return was positive — 40 sessions — FANG beat XOM by an average of 1.36% over the next 20 sessions rather than lagging it. Both names were weak, but XOM was weaker: FANG came in near flat at -0.04% while XOM lost 1.39%. The presumed winner keeps finishing second.

The small-sample versions of the same instinct

The FANG gap-fade study is the honest small version. Across nine sessions where Brent closed up at least 1% and FANG gapped up at least 2%, open-to-close averaged -0.32%, against roughly -0.01% for the other 729 regular-session days. Two-thirds of trigger days closed below the open. The direction matches the intuition exactly, and the research itself flags that nine observations is too thin to call an edge. That restraint is the right call, but the gap between -0.32% and a flat tape is still the kind of thing worth remembering.

Where the sample is big enough to say something

The capitulation study is the one with real sample behind it: top-quintile first-30-minute XLE volume on days Brent fell at least 1%. Fifty-five days met both conditions. XLE averaged +3.27% over the following 20 sessions against SPY's +2.69% — a spread of +0.58%, statistically indistinguishable from zero at p around 0.60, and essentially no different from the +0.37% spread logged on 441 ordinary days. The capitulation-then-bounce story isn't wrong so much as indistinguishable from what XLE does on any random Tuesday.

And the study that actually cuts against the mood is the volatility one. When XLE's 20-day realized volatility sat in the bottom quintile of its spread to SPY while Brent's 20-day return was positive — 17 qualifying days — XLE went on to lag SPY by an average of 3.39%, median -1.26%, ahead in only 29% of cases. Roughly one in three. Calm energy plus rising crude has been, in this panel, a setup for underperformance.

What the cluster is saying

The temptation with a run like this is to narrate a single culprit — refiners versus producers, integrated versus shale, some structural shift in how crude flows into earnings. The data here doesn't license that. What it does support is a narrower and more useful point: energy is behaving less like one macro block and more like a collection of separate bets wearing a sector's costume. Buying the sector because the curve steepened and oil rose is treating a dispersion story as a beta story.

The most instructive number in the whole stack isn't any of the signal returns — it's the 0.25% gap on ordinary days. Signal days move the XOP-versus-XLE spread from a quarter of a point to more than two points, which means the signal is detecting something real. It's just that the something real points the other way. Where statistical tests exist, they say noise; where the sign pattern is cleanest, the sample is smallest. That combination argues for holding all of it loosely — and for being far more suspicious of the reflation reflex than the reflex deserves.