XOP-XLE weekly return spread → next-week Brent return (N=155 weeks)
After 155 weeks of data, the XOP-XLE spread does not do what the front-running thesis says it should. If equity traders in exploration and production names were consistently anticipating moves in crude, a wider spread this week should point to higher Brent next week. It doesn’t. The correlation is 0.069, the sign call lands 48.4% of the time, and the regression explains half a percent of the variance. Statistically, this is indistinguishable from noise.
The idea was worth testing: E&P stocks are more leveraged to oil than integrated majors, so their relative performance could in theory reveal information before the commodity reacts. The data says otherwise. At best there is a faint, statistically meaningless lean in the predicted direction; as a signal it has no practical edge.
Below is the full breakdown — correlation, sign accuracy, tertile spreads, and regression output — so you can see exactly why this thesis fails.
For XOP over the past ~3 years, does the weekly return spread between XOP and XLE predict the next-week direction of Brent crude? Thesis: E&P outperformance vs integrated majors acts as a leading signal for crude, because equity traders front-run commodity moves, so a wider spread this week precedes a same-direction Brent move next week.
How this was measured
Resampled XOP, XLE, and daily Brent (brent_daily_df['value']) to Friday-close weeks. Weekly return = (close / prev close) - 1. Spread = XOP return minus XLE return (capturing E&P outperformance relative to integrated majors). The dependent variable is next-week Brent return (shifted forward one week — no look-ahead). We examine Pearson correlation, sign-hit rate (does spread direction predict oil direction?), tertile analysis of spread magnitude, and linear regression (Brent_next ~ alpha + beta * spread). A positive slope and high sign accuracy would support the thesis that equity traders front-run commodity moves.
The key numbers
Reading the numbers
The headline result is that the spread has no reliable predictive edge: the correlation with next-week Brent is only 0.069 (p≈0.40), and same-direction sign accuracy is 48.4%, slightly worse than a coin flip. The spread explains less than 0.5% of next-week Brent's movement, so the leading-signal thesis is not supported by these 155 weeks.
The charts
Each dot is one week pairing the XOP-minus-XLE spread with Brent's return the following week. The cloud is scattered with no clear diagonal tilt; if a wide spread reliably led Brent higher, points should climb from left to right. The correlation of 0.069 is essentially zero in practical terms, and the p-value of 0.40 means this pattern is easily explained by chance. In plain terms, knowing the spread this week tells you almost nothing about next week's Brent direction.
Grouping the 155 weeks into three spread buckets shows no simple staircase: the low-spread bucket and the high-spread bucket both have positive average next-week Brent returns (0.42% and 0.64%), while the middle bucket is negative (-0.47%). That up-down-up pattern is not what the thesis predicts, and all three averages are close to zero. The taller high-spread bar is partly consistent with the idea, but the positive low-spread bar undermines it.
Comparing weeks by whether XOP beat XLE or lagged it, the average next-week Brent return is 0.37% after a positive spread and 0.03% after a negative spread. That sounds like a slight edge for the thesis, but the two groups' ranges overlap heavily, and the Welch test gives p=0.73, meaning the difference is well within ordinary noise. With 78 positive-spread weeks and 77 negative-spread weeks, the averages would need to be much farther apart to count as evidence.
Next-week Brent return by spread sign
| spread_sign | mean_return | std | n |
|---|---|---|---|
| negative (XOP < XLE) | 0.0003 | 0.0693 | 77 |
| positive (XOP > XLE) | 0.0037 | 0.05 | 78 |
Regression summary
| alpha | beta | R² | p(beta) |
|---|---|---|---|
| 0.0022 | 0.28 | 0.0047 | 0.3951 |
The takeaway
Short answer: no — over 155 weeks, the XOP-vs-XLE spread has essentially no predictive power for next-week Brent moves. The correlation is r = 0.069, and the p-value of 0.40 means a value that small is comfortably consistent with pure randomness. Directionally, the spread called the next week's Brent sign correctly only 48.4% of the time — slightly worse than a coin flip — and the binomial test (p ≈ 0.75) says that miss rate is nothing special. Regression tells the same story: the slope is positive (0.28) but statistically indistinguishable from zero, and the model explains just 0.5% of the variance in Brent's next-week return. When XOP beat XLE, Brent averaged +0.37% the following week; when XLE beat XOP, Brent averaged +0.03% — a small gap in favor of the thesis, but with a Welch p-value of 0.73 there's no confidence it isn't noise. So: the front-running thesis is not supported. At best this is a faint, unreliable lean; as a trading signal it's basically useless.
The fine print
- Weekly Friday-close sampling; a different end-of-week convention could shift results.
- Only ~36 months of history — any lead-lag relationship could be regime-specific.
- Sign test treats weeks as independent; mild autocorrelation could nudge p-values down.
- Only linear association is tested; threshold or non-linear effects aren't captured.