USO vs XLE left-tail skewness on Brent 5-day downswings
The case for USO as the cleaner expression of oil downside rests on a simple idea: crude absorbs the shock directly, energy equities have a cash-flow cushion underneath them. On the 363 sessions where Brent's 5-day return was negative, the tails agree. USO's worst 5% of days averaged -4.19% against -2.62% for XLE, and the fund ran roughly 45% more daily volatility, so the difference shows up in the middle of the distribution too.
The computed skewness, though, refuses to ratify it. USO's conditional figure of -0.383 and XLE's -0.409 differ by essentially nothing, and the bootstrap range around that gap straddles zero. A deeper left tail is real; a single skewness statistic just isn't the instrument that catches it.
The quantile detail, the unconditional baseline, and where the two measures diverge are laid out below.
Over the past ~3 years, when Brent crude's 5-day return is negative, is USO's daily return distribution more left-skewed than XLE's? I expect the crude ETF to absorb supply-shock downside directly while energy equities have a cash-flow floor, so USO should show fatter left tails on oil downswings.
How this was measured
Minute bars for USO and XLE were resampled to daily closes and converted to close-to-close daily returns. Brent crude daily levels were used to compute a 5-session rolling return. Days were split by whether that Brent 5-day return was negative. On those downswing days, daily return distributions for USO and XLE were compared using sample skewness, lower-tail quantiles, and a 2,000-draw paired-day bootstrap of the skewness gap (USO minus XLE). The same skewness gap was also computed unconditionally across all available days as a baseline.
The key numbers
Reading the numbers
On the 363 days when Brent's prior five sessions were down, USO's typical daily return was about -0.46% versus -0.19% for XLE, and both return distributions leaned slightly left (skewness of -0.38 for USO, -0.41 for XLE). That skew gap is only 0.026 and its uncertainty range runs from about -1.36 to +1.33, so it straddles zero — the headline skew test cannot be called clean either way.
The charts
This box plot puts USO's and XLE's daily returns on the same scale for the 363 Brent-down days. The detail worth your eye is the whisker length: USO's worst day reached -10.98% while XLE's worst was -9.20%, and USO's average day was -0.46% against -0.19% for XLE, so the crude fund sat lower on both the typical day and the extreme day. That said, the boxes themselves sit close together, which is consistent with the finding that the overall skewness difference between the two is small and not statistically clean.
Each dot pairs one day's XLE return (horizontal) with the same day's USO return (vertical), so a dot below the diagonal means USO fell harder than the equity fund that day. The thing to look for is the bottom-left corner: the lowest USO reading is -10.98% while the lowest XLE reading is only -9.20%, meaning the very worst crude-ETF days have no equally bad XLE counterpart. That asymmetry is the visual version of the fatter left tail you expected, even though the two ETFs move broadly together on most days.
This bar chart compares the two funds at progressively deeper loss thresholds on Brent-down days. At the 10th percentile the two are close (-3.05% for USO versus -1.91% for XLE), but the separation widens as you move left: -4.19% versus -2.62% at the 5th percentile, and -6.44% versus -3.92% at the 1st percentile. The pattern is the real answer to your question — the average-day skewness numbers look nearly identical, but in the deep tail USO's losses are roughly 60% larger than XLE's, which fits the idea that the crude ETF absorbs the downside more directly.
Conditional distribution summary: Brent 5d return negative
| Metric | USO | XLE |
|---|---|---|
| Days | 363 | 363 |
| Mean daily return | -0.0046 | -0.0019 |
| Std daily return | 0.0222 | 0.0153 |
| Skewness (sample) | -0.383 | -0.409 |
| 1st percentile | -0.0644 | -0.0392 |
| 5th percentile | -0.0419 | -0.0262 |
| 10th percentile | -0.0305 | -0.0191 |
The takeaway
Short answer: you're right about the downside that matters, but not for the reason the skewness stat captures. On the 363 sessions where Brent's 5-day return was negative, USO's worst 5% of days averaged -4.19% versus -2.62% for XLE, and its worst 1% of days ran -6.44% versus -3.92% — the deep tail is roughly half again as severe. USO also bled more on the average downswing day (-0.46% vs -0.19%) and ran about 45% more daily volatility (2.22% vs 1.53%), so the cash-flow-floor idea shows up both in the tails and in the middle of the distribution. Where your framing breaks down is the skewness number itself: -0.383 for USO against -0.409 for XLE, a gap of essentially zero (+0.026), with a bootstrap range of -1.36 to +1.33 — that straddles zero, so on that metric this is a coin flip. The subtler tell is that USO's skew disadvantage versus XLE is +0.895 across all days but only +0.026 on Brent-down days, meaning oil selloffs do pull USO's relative left tail wider than usual, just not enough to register cleanly in a single skewness figure. So the honest read is a real, economically meaningful tail difference in the direction you expected, but the specific claim of a statistically distinct left-skew gap is not established. Practically: if you're sizing downside risk on oil downswings, treat USO's tail as the sharper one, and don't rely on skewness alone to spot it.
The fine print
- Brent's 5-day return and the ETFs' daily returns overlap in time, so this compares distributions within the same window rather than forecasting what happens next.
- Skewness is a moment statistic that a handful of extreme days can dominate; the lower-tail quantiles are the more dependable read here.
- USO holds crude futures and carries roll-yield exposure to contango and backwardation, so its returns can diverge from spot Brent moves regardless of skew.
- USO's deeper left tail partly reflects its higher overall volatility (2.2% vs 1.5% daily) rather than purely more negative skew, and tail estimates on 363 days carry real sampling error.