DVN downside vs upside beta to Brent: asymmetry test over three years
The hypothesis was elegant: an E&P like DVN should bleed harder on crude drawdowns than it pumps on rallies. Over 741 trading days, the numbers line up — downside beta of 0.362 versus 0.275 upside — but the statistics refuse to call it a fact. The 0.087 gap carries a p-value of 0.46, meaning you'd see this split nearly half the time even if asymmetry were absent.
That distinction between a directional lean and an established skew is the real story. Rolling 126-day windows show downside beta beating upside in three of every four windows, so it isn't a few sessions doing the work. But the full-sample beta of 0.39 is the cleanly measured number; the asymmetry is not.
Below is the full study — the sign-interacted regression, the rolling analysis, and the charts separating what DVN's price action shows from what the hypothesis expects.
For DVN over the past ~3 years, is its downside beta to Brent crude on days Brent falls higher than its upside beta on days Brent rises? I expect the E&P to be sold more aggressively on crude drawdowns than it is bid on equal-size Brent rallies, creating persistent downside-beta skew.
How this was measured
Daily DVN close-to-close returns were joined with daily Brent returns from the global Brent macro series. Days with Brent return greater than zero were classified as up days, and days with Brent return less than zero as down days. A sign-interacted OLS regression was estimated: DVN return on Brent return, a Brent-down dummy, and a Brent-down interaction term. The upside beta is the Brent-return slope on up days; the downside beta is that slope plus the interaction coefficient. HC1 robust standard errors were used, and the interaction p-value tests whether the downside-up downside beta gap differs from zero. Rolling 126-trading-day regressions were also computed to assess persistence.
The key numbers
Reading the numbers
Over 741 trading days, DVN's downside beta (0.362) was larger than its upside beta (0.275), an asymmetry of +0.087. But the p-value of 0.46 means the gap is not statistically distinguishable from zero: the direction supports the thesis, the evidence is not conclusive.
The charts
The scatter plot shows daily Brent and DVN returns as a cloud centered near zero, with most days inside a roughly +/-10% band and a clear positive tilt because DVN generally moves with crude. The thing to look at is whether the cloud on the left, where Brent falls, drops more steeply than the cloud on the right, where Brent rises; the raw scatter is noisy enough that this asymmetry is hard to see by eye. This chart sets the stage: the relationship exists, but isolating downside versus upside behavior needs the sign-split betas shown next.
The taller bar is the downside beta, 0.362, versus 0.275 for the upside beta. Reading it plainly: on a day Brent falls 1%, DVN tends to fall about 0.36%, while on a day Brent rises 1%, DVN tends to rise only about 0.27%. That gap is in the direction the question predicted, but with a p-value of 0.4636, it is not statistically clear — a gap this size could easily arise from noise.
The two rolling lines move around a lot: across the 126-day windows, downside beta averaged 0.615 and upside beta averaged 0.493, with the downside line spending much of the period above the upside line. The lines cross frequently, so the dominance is not constant, but downside beta was still higher in 75% of rolling windows. That consistency supports the skew story in most sub-periods, while the wide swings and occasional inversions echo the statistical uncertainty shown by the full-sample p-value.
Conditional daily-return and beta summary
| Brent day type | N | Mean Brent ret | Mean DVN ret | Beta vs Brent |
|---|---|---|---|---|
| Brent up days | 372 | 0.0186 | 0.0088 | 0.2745 |
| Brent down days | 365 | -0.0179 | -0.0086 | 0.3619 |
| Full sample | 741 | 0.0005 | 0.0001 | 0.3918 |
Asymmetric-beta regression (HC1 robust errors)
| Term | Coefficient | Std err | t-stat | p-value |
|---|---|---|---|---|
| Upside beta (Brent>0) | 0.2745 | 0.0656 | 4.185 | 0 |
| Downside incremental beta | 0.0874 | 0.1192 | 0.733 | 0.4636 |
| Downside beta (Brent<0) | 0.3619 | 0.0995 | 3.638 | 0.0003 |
The takeaway
Over the past three years, DVN's downside beta to Brent came in a bit higher than its upside beta — 0.362 versus 0.275, a gap of about 0.087 — so the direction matches your hypothesis. But the evidence is not conclusive: the p-value is 0.46, meaning there's roughly a 46-in-100 chance you'd see this gap even if downside and upside betas were truly identical. That's basically a coin flip, not a persistent skew you can hang a trade on. The one point in the hypothesis's favor is that in rolling 126-day windows, downside beta exceeded upside beta in 75% of windows, so the pattern wasn't driven by just a few days. Still, the full-sample beta of about 0.39 is cleanly measured, while the asymmetry is not. Net takeaway: DVN clearly moves with Brent, but the downside-beta skew is a lean at best, not an established feature of the stock.
The fine print
- Up and down days have slightly different average Brent move sizes (0.0186 vs -0.0179), so this tests sign-conditioned slopes rather than exactly equal-size rallies and selloffs.
- Brent and DVN are aligned on the same calendar day; a one-day lag robustness check would be useful if Brent settles after the US close.
- Daily close-to-close returns can miss intraday dynamics; minute-level alignment could tell whether the skew also shows up inside the trading day.
- The ~36-month window covers one oil-price regime, so persistence outside this period is untested.