When Oil-Stock Signals Fail, the Data Leans Toward Humility

Today's tape is quiet on the headline side, but the research desk has been busy chopping up oil stocks. The pattern across the recent batch of findings is consistent and worth staring at: the signals that look obvious on a chart or in a sentiment feed tend to fail, while the more boring volatility mechanics still have a pulse.

The Curve Regime That Never Shows Up

Take COP. The question was whether its daily sensitivity to Brent changes when the Treasury curve is in a particular regime. The data's answer is effectively no — not because the beta is stable, but because there is almost no variation in the regime. Of 729 classifiable days, only 3 traded above the trailing 12-month average spread, meaning 726 sat at or below it. On those ordinary days, COP's daily beta to Brent was 0.3516, essentially the same as the full-sample beta of 0.3520. That's not a null result; it's a warning about asking questions the data can't answer.

Sentiment and Volatility Signals Invert

The sentiment work is more interesting because it runs in reverse. EOG's top-quintile news-sentiment days were followed by an average next-day return of -0.13%, while bottom-quintile days averaged +0.47%. That gap of roughly -0.60% has a t-stat of -1.98 and a p-value of 0.049 — just barely over the line of statistical significance, but directionally consistent with a fading or contrarian effect. The average next-day return steps down almost monotonically across sentiment quintiles, so the pattern isn't the product of one outlier.

Similarly, HAL's 20-day realized volatility relative to Brent tells you little about forward 10-day returns. Days when that ratio sat in the top quintile averaged +1.26% over the next 10 days, versus +2.14% for the bottom 80% and +2.00% for all days. The signal, if anything, points the wrong way.

Backtests That Bleed

Two backtests from the platform reinforce the same theme. When Brent closes up more than 1% and PSX closes down on the same day, the strategy returned +54.83% on $100,000 across 87 closed trades with a 56% win rate. That sounds decent until you compare it to SPY buy-and-hold over the same window: +68.30%. The strategy trailed by 13.47 points. The XOM version is worse: the strategy defined by XOM closing down more than 2% on a day with positive news sentiment returned +6.39% across 32 trades, also with a 56% win rate, but SPY's +68.30% left it in the dust — a 61.91-point shortfall. Best single trade +5.09%, worst -10.79%. These aren't disasters; they're just worse than doing nothing.

But not everything is broken. CVX's next-day range widens after extreme Brent volatility days: above its median range 67.6% of the time after extreme Brent days (50 of 74), versus 48.0% after non-extreme days. The odds ratio is 2.26, and the one-sided Fisher p-value is 0.001. That's a clear, honest signal — but it's a volatility forecast, not a money printer. It tells you the market will move more, not which way.

The lesson across all of this: the oil patch has absorbed the easy patterns. Crude correlations are pinned, sentiment has been arbitraged into a reversal, and obvious two-condition entries lag the index. The surviving edge is in measuring how much the market will move, not pretending to know where it's going. The data leans that way, and the past few days of research make a strong case for humility.