XLE vs SPY 20-session forward performance after Fed/FOMC headline-spike days with positive Brent 20-session return (N=37)
Thirty-seven sessions over the past three years paired a Fed/FOMC headline spike — intensity in the top quintile of its trailing year — with Brent already up over the prior 20 days. On those days, XLE beat SPY by 3.49% on average over the next 20 sessions, and 8.11% at the median. Across the other 409 valid days, that same excess averaged 0.03%.
That is the shape of an inflation squeeze: policy noise landing while crude is rising, energy cash flows outrunning rate-sensitive broad exposure. Brent was up roughly 21.8% on average at trigger, so the setup itself was real. The edge is a lean, not a proof — significance sits just outside the usual bar, the 20-session windows overlap heavily, and a few January days carry much of the average.
The full breakdown follows.
Over the past ~3 years, when Fed/FOMC headline intensity spikes into the top quintile of its trailing 252-session distribution while Brent crude's 20-day return is positive, does XLE outperform SPY over the next 20 trading sessions? I expect policy-shock headlines hitting while crude is rising to be traded as a supply-driven inflation squeeze, which should favor energy-equity cash flows over broad-market rate-sensitive exposure until the shock fades.
How this was measured
Daily close panels for XLE and SPY were built from minute bars. Fed/FOMC headline intensity was defined as the daily count of SPY and XLE news headlines containing Fed/FOMC/policy keywords, deduplicated by headline and date. A session qualified as high-intensity when the current count exceeded the 80th percentile of the prior 252 trading sessions. The second condition required Brent's 20-trading-session return to be positive, using Brent daily data shifted by one session for publication lag. For each qualifying day, forward 20-session returns were computed from close-to-close prices, and XLE-SPY excess return was analyzed against zero and against all other valid days.
The key numbers
Reading the numbers
Across 37 Fed-high/Brent-positive days, XLE beat SPY over the next 20 sessions by an average of 3.49% and a median of 8.11%; it did so 67.6% of the time vs 48.4% otherwise. But p=0.0758 is above the usual 0.05 cutoff, so the edge is economically interesting but not statistically clear.
The charts
Chart 0 compares the distribution of 20-session XLE-minus-SPY returns for the 37 Fed-high/Brent-positive event days against the 409 other valid days. The event group's mean excess is 3.49% versus 0.03% for other days, and the report's median event excess is 8.11%, so the typical event day is not just a few lucky outliers. But the boxes and whiskers overlap heavily: event returns range from -19.91% to +19.69% while other days range from -22.54% to +18.02%, which is why the p=0.0758 reads as suggestive rather than decisive.
This bar chart distills the same comparison into two averages: 3.49% mean excess for Fed-high/Brent-positive days and 0.03% for all other valid days. The visual gap is large relative to baseline, supporting the idea that policy-shock headlines during rising crude have tended to favor XLE over SPY. The caveat is the sample: only 37 event days, and the p=0.0758 test says that gap is not statistically clear at the usual 5% threshold.
Event-day detail (first 20, chronological)
| event_date | fed_headlines | brent_20d_ret_pct | xle_fwd20_pct | spy_fwd20_pct | excess_fwd20_pct |
|---|---|---|---|---|---|
| 2025-11-18 | 1 | 4.04 | -1.91 | 1.95 | -3.86 |
| 2025-12-05 | 2 | 0.96 | -0.45 | 1.19 | -1.63 |
| 2026-01-09 | 4 | 1.15 | 14.89 | -0.11 | 15 |
| 2026-01-12 | 2 | 3.15 | 15.78 | -0.03 | 15.81 |
| 2026-01-13 | 2 | 5.71 | 17.04 | -0.17 | 17.2 |
| 2026-01-14 | 2 | 8.81 | 12.05 | -1.28 | 13.33 |
| 2026-01-15 | 2 | 11.89 | 14.38 | -1.73 | 16.11 |
| 2026-01-16 | 1 | 10.4 | 12.42 | -1.39 | 13.8 |
| 2026-01-22 | 1 | 8.75 | 12.43 | 0.04 | 12.39 |
| 2026-01-23 | 1 | 5.21 | 11.68 | -0.69 | 12.37 |
| 2026-01-27 | 1 | 6.28 | 10.41 | -0.72 | 11.13 |
| 2026-01-29 | 1 | 12.36 | 10.89 | -1.29 | 12.18 |
| 2026-01-30 | 1 | 13.96 | 11.41 | -0.89 | 12.3 |
| 2026-02-02 | 1 | 17.77 | 12.78 | -2.48 | 15.26 |
| 2026-02-03 | 1 | 9.26 | 8.57 | -0.26 | 8.84 |
| 2026-02-05 | 1 | 14.57 | 8.52 | -0.36 | 8.88 |
| 2026-02-06 | 1 | 14.39 | 4.5 | -2.21 | 6.71 |
| 2026-02-09 | 1 | 11.23 | 3.59 | -2.27 | 5.86 |
| 2026-02-10 | 2 | 9.34 | 6.37 | -3.23 | 9.6 |
| 2026-02-13 | 3 | 1.35 | 6.33 | -1.78 | 8.11 |
Summary of forward 20-session excess return
| group | n | mean_excess_pct | median_excess_pct | pct_positive |
|---|---|---|---|---|
| Fed-high & Brent-positive | 37 | 3.49 | 8.11 | 67.57 |
| All other valid days | 409 | 0.03 | -0.29 | 48.41 |
The takeaway
Directionally yes — when Fed/FOMC headlines spiked and Brent was already rising, XLE did beat SPY over the next 20 sessions, but the evidence reads as a genuine lean rather than a proven edge. The average excess return on those 37 event days was 3.49%, the typical day was even better at 8.11%, and XLE won 68% of the time, versus a near-zero 0.03% average excess on the other 409 days where XLE won only 48% of the time. Brent was up about 21.8% on average at trigger, so the rising-crude half of your supply-squeeze premise was clearly present. The problem is the confidence interval: the excess return is only about a 1-in-13 chance of being pure luck (p=0.076 against zero, p=0.082 against all other days) — suggestive, not conclusive. Two things soften it further: the events cluster in time, so the 20-session windows overlap and the same move gets counted repeatedly, and a small number of January days with 15%+ excess carry a lot of the average while several early events were outright losses. Practical takeaway: the pattern is consistent with your inflation-squeeze thesis and worth tracking, but with 37 observations and significance just outside the standard bar, treat it as a hypothesis to keep monitoring rather than a signal to size up on.
The fine print
- Only 37 event days — thin evidence, and the tests are descriptive rather than confirmatory at this sample size.
- The trailing 80th-percentile headline threshold sits at zero, meaning 'top-quintile spike' effectively collapses to 'any day with at least one Fed headline.'
- Event days cluster (notably January 2026), so overlapping 20-session windows double-count the same moves and make the t-stats look stronger than they are.
- Headline counts come from SPY/XLE news text with same-day timing, and no transaction costs or execution frictions are modeled — this is conditional relative performance, not a tradable net-of-cost signal.