XOM daily-return correlation with SPY in Brent-down vs Brent-up months
The expected split never quite showed up. Over the past three years, XOM’s daily-return correlation with SPY did run higher in months when Brent crude fell — 0.077 average versus 0.004 in up-oil months — but that gap is nowhere near statistically meaningful. With a p-value of 0.54, the difference is indistinguishable from noise.
That matters because the thesis is intuitively appealing: oil down should make XOM trade less like a commodity and more like the broad market. The direction of the numbers agrees, but the evidence is too thin to support it. Thirty-six months, split 17 down and 19 up, simply isn’t enough to separate a real effect from random variation.
The full breakdown below walks through the methodology, the monthly correlation series, and the test results in detail.
For XOM over the past ~3 years, does its daily-return correlation with SPY increase in months when Brent crude declines? Thesis: XOM behaves like a commodity play during oil rallies but reverts to broad-market beta when crude sells off, so its correlation to SPY is higher in down-oil months.
How this was measured
Resampled XOM and SPY minute bars to daily close returns; computed monthly within-month Pearson correlation using all trading days with ≥5 days. Brent crude monthly returns were derived from last-day-of-month prices. Months were classified as 'Brent down' (negative monthly return) and 'Brent up' (non-negative). The two groups' mean correlations were compared via a Welch two-sample t-test. The thesis posits higher XOM–SPY correlation during oil declines (commodity-beta convergence) and lower during rallies (commodity-led divergence).
The key numbers
Reading the numbers
The average XOM-SPY correlation was 0.077 in months when Brent fell vs 0.004 when Brent rose — 17 vs 19 months. That difference looks supportive, but the Welch p-value of 0.54 means it is nowhere near statistically clear.
The charts
Each dot is a month, with Brent's monthly return on the horizontal axis and the XOM-SPY daily-return correlation on the vertical axis. The dots are scattered in a broad cloud rather than showing a clear trend: months with falling oil (left of zero) and rising oil (right of zero) both produce very high and very low correlations, from about -0.68 to 0.83. That flat, messy pattern is the first clue that the relationship in the thesis is weak at the monthly level.
The two box plots compare the spread of monthly correlations in the 17 Brent-down months and the 19 Brent-up months. The Brent-down months have a higher center, with a mean around 0.077, versus about 0.004 for up months, but the boxes and whiskers overlap heavily — both regimes contain very negative and very positive correlations. So while the middle of the oil-decline distribution sits higher, the two groups are not cleanly separated.
This bar chart makes the headline comparison: the average correlation in Brent-down months is 0.0765, far above the 0.0036 average in Brent-up months. But the accompanying two-sample Welch test, with a p-value of 0.544, says a gap this size could easily arise by chance, so the taller bar is not convincing evidence that XOM tracks SPY more closely when crude sells off.
Correlation summary by Brent regime
| Regime | N months | Mean corr | Std corr |
|---|---|---|---|
| Overall | 36 | 0.038 | 0.3434 |
| Brent down (oil decline) | 17 | 0.0765 | 0.42 |
| Brent up (oil rally) | 19 | 0.0036 | 0.2645 |
The takeaway
Over the past roughly three years, there is no clear evidence that XOM's daily-return correlation with SPY rises in months when Brent crude falls. The numbers point in the thesis's direction — the average correlation was about 0.077 in down-oil months versus just 0.004 in up-oil months, a gap of 0.073 — but it is far from conclusive. With only 36 months split 17 down versus 19 up, the statistical test gives p = 0.54, meaning a difference this size would show up by chance roughly half the time even if oil had no real effect on the correlation. So the pattern is suggestive at most, not a real signal. The practical takeaway is that the commodity-versus-beta story may exist, but these data are too noisy and the sample too thin to hang much on it.
The fine print
- Only 36 months of data, with 17 down-oil and 19 up-oil months — small sample makes the comparison coarse.
- Correlations are computed on overlapping daily returns within each month, so one volatile week can swing a month's estimate.
- The oil regime is defined by the full month's Brent return, so this captures a simultaneous relationship, not a predictive one.
- A few outlier months could be doing most of the work since the down-oil correlation standard deviation is high (0.42).