XLE vs SPY around top-quintile Iran/Hormuz/tanker/sanctions headline spikes
A headline spike on Iran or Hormuz sounds like a clean setup: buy energy on the fear, sell it once the tankers keep moving. On 36 top-quintile spike days over the past three years, XLE beat SPY by just 0.33% on average — and won only 56% of the time, roughly a coin flip.
The fade is messier still. The mean 10-session excess return is -0.45%, but the median is +1.31%, meaning the typical spike is followed by XLE outperforming, with a few large losers dragging the average down. Neither window clears statistical significance, and XLE shows no persistent drift against SPY across the full 751-session sample.
So the pattern is plausible storytelling, not a measurable edge at this sample size. The full breakdown of means, medians, hit rates, and t-tests is below.
Over the past ~3 years, when news headline intensity for Iran/Hormuz/tanker/sanctions keywords spikes into the top quintile of its prior 20-session distribution, does XLE outperform SPY on spike day and then underperform over the next 10 sessions? I expect a Hormuz risk premium to trigger immediate energy-sector buying that decays once the headline passes without confirmed supply disruption, so XLE should lag broad equities as the premium unwinds.
How this was measured
XLE and SPY minute bars were resampled to a common daily close panel, and daily excess returns were computed as XLE return minus SPY return. Headline intensity was built by combining XLE_news and SPY_news, de-duplicating on date/source/title, and counting each day's articles whose title or summary contained any of: Iran, Hormuz, tanker, sanctions. A spike day was defined as a day with positive intensity at or above the 80th percentile of the prior 20 sessions, excluding the current session. For each spike day with a full 10-session forward window, we measured the same-day XLE−SPY excess return and the cumulative XLE−SPY excess return from the spike-day close to the t+10 close. Means, medians, hit rates, and one-sample t-tests against zero are reported.
The key numbers
Reading the numbers
Across 36 headline-spike days, XLE beat SPY by about 0.33% on the day itself, then gave back roughly 0.45% over the following 10 sessions. Both moves are small and statistically fuzzy, so read the fade as a pattern worth watching, not a proven edge.
The charts
This line traces the average XLE-minus-SPY gap session by session after a headline spike, and it starts slightly negative right away rather than climbing. The eye should go to the trough: the cumulative shortfall deepens to about -1.24% at its worst before recovering to -0.45% by t+10, so the damage is not a straight drift but a mid-window sag that partially heals. That shape fits the idea of an energy premium unwinding, though the whole path sits inside noise-level territory.
This bar chart puts the two headline windows side by side: a small positive bar of +0.33% on the spike day, and a negative bar of -0.45% cumulatively over the next 10 sessions. The sign flip is the story the question asked about, and it points the way a decaying Hormuz risk premium would. Just remember both bars are small relative to typical daily sector swings, so the contrast is suggestive rather than decisive.
Event-level XLE−SPY excess returns
| event_date | intensity | spike_day_er | fwd10_er |
|---|---|---|---|
| 2025-11-26 | 1 | -0.0008 | 0.0186 |
| 2026-01-09 | 1 | -0.0076 | 0.0561 |
| 2026-01-14 | 1 | 0.0297 | 0.0458 |
| 2026-02-03 | 1 | 0.0455 | 0.064 |
| 2026-02-09 | 2 | 0.0022 | 0.0355 |
| 2026-02-19 | 1 | 0.0088 | 0.026 |
| 2026-03-02 | 1 | 0.0152 | 0.039 |
| 2026-03-05 | 1 | 0.0119 | 0.0844 |
| 2026-03-10 | 2 | -0.0059 | 0.1226 |
| 2026-03-11 | 2 | 0.0365 | 0.0896 |
| 2026-03-12 | 2 | 0.0189 | 0.0979 |
| 2026-03-18 | 2 | 0.0135 | 0.0162 |
| 2026-03-24 | 2 | 0.0157 | -0.066 |
| 2026-03-26 | 2 | 0.0271 | -0.1264 |
| 2026-03-27 | 3 | 0.0418 | -0.1751 |
| 2026-03-30 | 2 | -0.0066 | -0.2029 |
| 2026-04-02 | 5 | 0.0088 | -0.1566 |
| 2026-04-14 | 4 | -0.034 | 0.01 |
| 2026-04-15 | 2 | -0.0106 | 0.0406 |
| 2026-04-17 | 2 | -0.0407 | 0.0579 |
| 2026-04-24 | 2 | -0.0084 | -0.0552 |
| 2026-04-28 | 4 | 0.0218 | -0.0385 |
| 2026-05-04 | 2 | 0.0117 | -0.0094 |
| 2026-05-06 | 2 | -0.045 | 0.0404 |
| 2026-05-20 | 3 | -0.0296 | -0.0421 |
| 2026-06-04 | 2 | -0.0083 | -0.0757 |
| 2026-06-09 | 2 | -0.0096 | -0.0694 |
| 2026-06-10 | 1 | 0.0374 | -0.0874 |
| 2026-06-12 | 2 | -0.0005 | -0.0586 |
| 2026-06-24 | 1 | -0.0212 | 0.0082 |
| 2026-06-26 | 1 | 0.0045 | 0.0266 |
| 2026-07-08 | 1 | 0.0166 | 0.0636 |
| 2026-07-20 | 1 | 0.0053 | -0.0089 |
| 2026-07-27 | 1 | -0.0235 | -0.0143 |
| 2026-07-29 | 2 | 0.0216 | -0.0117 |
| 2026-08-04 | 1 | -0.0237 | 0.0929 |
The takeaway
Short answer: no — the Hormuz risk-premium pattern you described doesn't show up as a reliable trade in this data. On the 36 top-quintile headline-spike days, XLE beat SPY by an average of just +0.33%, and it only won 56% of the time, which is about what you'd get flipping a coin. The follow-through is where it gets messier: the average 10-session excess return is -0.45%, which looks like the fade you predicted, but the median for those same 10 sessions is +1.31% — meaning the typical spike is followed by XLE *outperforming*, and the negative average is being dragged down by a handful of outsized losers. Neither window is close to statistically meaningful (p = 0.39 on the spike day, p = 0.73 over the next 10 sessions), so with 36 events and this much scatter, you can't distinguish any of it from noise. For context, XLE's unconditional daily excess return versus SPY across the full 751-session sample is roughly zero, so there's no persistent drift doing the work here either. The practical takeaway: the immediate energy bump and the subsequent decay are plausible storytelling but not a measurable edge at this sample size — and the fact that the mean and median disagree in the forward window means you shouldn't trade the average without knowing which few events are producing it.
The fine print
- Only 36 spike days, and mean versus median disagree sharply over the forward window — the 'fade' is a few big losers, not the typical outcome.
- Forward 10-session windows overlap when spikes cluster, so the t-tests overstate precision; the true confidence is even weaker than p = 0.73 suggests.
- Keywords like 'sanctions' and 'tanker' aren't topic-filtered, so many spikes aren't actual Hormuz supply-risk headlines.
- The top-quintile bar is relative to the prior 20 sessions — in quiet stretches a single headline qualifies, which dilutes the signal.