XOM vs SPY 20-session forward excess after falling 10Y yield plus rising Brent
The setup sounds right on paper: long rates falling while oil rises should squeeze the broad market and favor an integrated major. Three years of data say it mostly doesn't. Of 26 qualifying days, XOM beat SPY over the following 20 sessions by an average of 1.88%, roughly two points better than the -0.15% average everywhere else.
Look closer and the edge thins out. The median signal day was a 0.85% loss to SPY, and XOM came out ahead only 46% of the time — the average is doing the work of a few outsized winners, several of which land in one tight early-2026 cluster.
The statistical tests agree: p of 0.24 against zero, 0.21 against non-signal days. With 26 overlapping observations, that's a lean, not a signal. The full breakdown follows.
Over the past ~3 years, when the 10-year Treasury yield falls by at least 20 basis points over the prior 20 trading days while Brent crude's 20-day return is positive, does XOM outperform SPY over the next 20 trading sessions? I expect rising oil alongside falling long rates to signal a supply-driven inflationary squeeze that favors XOM's integrated cash flows over broad-market rate-sensitive exposure, so XOM should lead SPY until the macro impulse fades.
How this was measured
Daily closes were built from minute bars for XOM and SPY, overlapping on shared trading days. The macro trigger uses the 10-year Treasury yield and Brent crude daily series lagged one trading day to respect T+1 release timing. A signal day is one where the prior 20-session change in 10Y yield is <= -20 basis points and the prior 20-session Brent return is positive. Forward performance is the close-to-close return over the next 20 trading sessions; the excess is XOM minus SPY. Signal-day excess is tested against zero with a one-sample t-test and compared with non-signal days via Welch's t-test.
The key numbers
Reading the numbers
Over 26 qualifying days, XOM beat SPY by 1.88% on average over the following 20 sessions — but the typical (median) day was actually -0.85%, and XOM led only 46% of the time. With t=1.21 and p=0.237, that average gap is well inside what random noise produces, so the edge is not statistically clear.
The charts
This paired bar chart splits average 20-session forward returns into signal days versus everything else. On signal days the XOM bar (2.53%) towers over SPY (0.66%), a gap of roughly 1.9 points — which is the pattern the question is asking about. The catch is in the right-hand pair: on ordinary days XOM (1.65%) and SPY (1.80%) are nearly level, so the signal-day gap comes mostly from SPY's return collapsing (1.80% down to 0.66%) rather than from XOM doing anything special — XOM itself only improves modestly, from 1.65% to 2.53%. That is a broader-market effect on signal days, not evidence of XOM-specific strength.
This histogram shows how the 26 signal-day XOM-minus-SPY excess returns are actually distributed, from a worst case of about -12.3% to a best of about +19.3%. The shape matters more than the +1.88% average: the mass sits near and just below zero, with a thin right tail of a few outsized winners dragging the mean up — consistent with the median of -0.85% and the fact that only 46% of days were positive. In plain terms, the typical signal day had XOM slightly behind SPY, and the headline average is carried by a handful of large up-moves, so a reader should treat the mean as fragile rather than as a repeatable pattern.
Qualifying signal days
| date | 10Y 20d change (bps) | Brent 20d return | XOM fwd20 | SPY fwd20 | XOM-SPY fwd20 |
|---|---|---|---|---|---|
| 2026-03-04 | -23 | 0.2298 | 0.0657 | -0.0442 | 0.1099 |
| 2026-03-03 | -21 | 0.0691 | 0.1082 | -0.0378 | 0.146 |
| 2026-03-02 | -27 | 0.0045 | 0.1131 | -0.0761 | 0.1892 |
| 2026-02-27 | -24 | 0.0107 | 0.1217 | -0.0716 | 0.1933 |
| 2026-02-24 | -21 | 0.0549 | 0.0984 | -0.0414 | 0.1398 |
| 2026-02-19 | -21 | 0.0606 | 0.047 | -0.0354 | 0.0824 |
| 2025-09-18 | -24 | 0.0386 | -0.0321 | -0.004 | -0.0281 |
| 2025-09-17 | -30 | 0.0221 | -0.0312 | 0.0083 | -0.0396 |
| 2025-09-16 | -28 | 0.0086 | -0.0184 | 0.0062 | -0.0247 |
| 2025-09-12 | -23 | 0.0151 | -0.0139 | -0.005 | -0.0089 |
| 2025-09-11 | -25 | 0.0123 | 0.0071 | 0.0245 | -0.0174 |
| 2025-07-02 | -20 | 0.0162 | 0.0087 | 0.0169 | -0.0081 |
| 2025-06-27 | -21 | 0.0434 | 0.0193 | 0.0391 | -0.0197 |
| 2025-06-25 | -21 | 0.0569 | 0.0245 | 0.0468 | -0.0223 |
| 2025-06-24 | -20 | 0.146 | 0.0124 | 0.0456 | -0.0332 |
| 2025-06-23 | -20 | 0.1954 | -0.01 | 0.045 | -0.055 |
| 2025-04-04 | -23 | 0.0205 | -0.0096 | 0.1134 | -0.123 |
| 2024-08-28 | -32 | 0.0284 | -0.0314 | 0.0356 | -0.067 |
| 2024-06-28 | -32 | 0.0675 | 0.0072 | -0.0032 | 0.0104 |
| 2024-06-27 | -22 | 0.0543 | 0.0191 | -0.0051 | 0.0243 |
| 2024-06-26 | -23 | 0.1 | 0.0274 | -0.0085 | 0.0359 |
| 2024-06-25 | -22 | 0.0946 | 0.0066 | -0.0041 | 0.0106 |
| 2024-06-20 | -22 | 0.0352 | 0.0366 | 0.0066 | 0.03 |
| 2024-01-12 | -22 | 0.0823 | 0.0341 | 0.0517 | -0.0176 |
| 2024-01-10 | -21 | 0.0267 | 0.0521 | 0.044 | 0.0081 |
| 2023-12-26 | -57 | 0.0051 | -0.0038 | 0.0239 | -0.0277 |
The takeaway
Three years of data say no — this setup has not reliably handed XOM an edge over SPY. The 26 qualifying days did average a +1.88% XOM-minus-SPY gain over the next 20 sessions, roughly 2 points better than the -0.15% average on the other 635 days, but that headline number is carried by a few big winners rather than broad participation. The median signal day was actually a 0.85% loss to SPY, and XOM only came out ahead 46% of the time — so the typical instance runs mildly against you. Statistically it's a shrug: the one-sample test gives a p of 0.24, about a 1-in-4 chance the positive average is just noise, and the signal-versus-non-signal comparison is no better at p of 0.21. With 26 observations and overlapping 20-session windows — the biggest wins sit in a tight cluster of days in early 2026, meaning the same rally gets counted several times — the real confidence is weaker still than those p-values imply. Treat this as unproven: a faint lean, not a signal you'd size a position around, and certainly not the supply-squeeze thesis showing up cleanly in the data.
The fine print
- Only 26 signal days, and because the 20-day windows overlap during clusters, a handful of early-2026 dates are effectively counted multiple times — the true sample is smaller than it looks.
- The mean is propped up by a few outsized wins; the median and the 46% win rate tell a less flattering story.
- Results depend heavily on the specific choices of 20bp, 20-session lookback and 20-session forward window — nearby thresholds could shift the event count and the measured edge a lot.
- One asset pair over one three-year rate and oil regime; a different window or a different energy name could behave differently.