AI Research OIHXOPmacro:brent_daily

OIH vs XOP after Brent -5% / OIH positive 5-day divergence (next 20 sessions)

16
Qualifying trigger days

When Brent dumps more than 5% in a week while oil-services stocks are still climbing, the reflex is to expect a catch-down. Over the past three years, the opposite tended to happen. On the 16 days that fit that divergence, OIH beat XOP by an average of 2.11 percentage points over the next 20 sessions, winning 62.5% of the matchups.

The question is whether an oil-services basket advancing while crude sells off is pricing backlog and pricing power rather than oil beta. The conditional spread leans that way, but with a thin sample and p-values near 0.20, it reads as a suggestive pattern rather than a confirmed edge.

The full analysis below lays out the triggers, the return distributions, and the tests behind that lean.

The research question

Over the past ~3 years, when Brent crude's 5-day total return is below -5% while OIH's 5-day total return is positive, does OIH outperform XOP over the next 20 trading sessions? I expect an oil-services basket that advances while crude sells off to be pricing backlog and pricing power rather than oil beta, so OIH should keep leading the E&P-heavy producers instead of mean-reverting.

How this was measured

Daily closes for OIH and XOP were assembled from minute bars and merged with Brent daily prices. A trigger day was defined as an ETF trading day where Brent's 5-session return, shifted one Brent session for T+1 settlement, was below -5% while OIH's own 5-session return was positive. Forward performance was measured as close-to-close total return over the next 20 ETF trading sessions for both OIH and XOP, with OIH−XOP defined as the difference in those simultaneous 20-session returns. Conditional trigger-day performance was compared with the full valid sample using means, medians, win rates, a one-sample t-test against zero, and a two-sample Welch t-test against the baseline.

The key numbers

Qualifying trigger days
16
2023-09-08 to 2026-08-03
Valid base days
685
Days with 5d and 20d-forward data available
Conditional OIH 20d mean
4.2236%
N=16 trigger days
Conditional XOP 20d mean
2.1149%
N=16 trigger days
OIH−XOP 20d mean
2.1087%
Positive means OIH leads XOP
OIH−XOP 20d median
2.3166%
N=16 trigger days
OIH−XOP win rate
62.50%
Share of 16 trigger days where OIH beat XOP over next 20 sessions
Baseline OIH−XOP 20d mean
-0.1164%
N=685 all days with sufficient forward data
Edge vs baseline
2.2251%
Conditional OIH−XOP minus unconditional OIH−XOP
One-sample t-stat (vs 0)
1.272
Positive favors OIH
One-sample p-value
0.2226
p=0.2226 ≥ 0.05 → no statistically-clear conditional OIH−XOP is nonzero
Two-sample Welch t-stat (vs baseline)
1.332
Positive favors conditional trigger days
Two-sample p-value
0.2020
p=0.2020 ≥ 0.05 → no statistically-clear conditional OIH−XOP differs from baseline

Reading the numbers

On the 16 days in the past three years when crude fell more than 5% over a week while oil-services stocks rose, OIH beat XOP by 2.1 percentage points on average over the following month and won 62% of the time, versus a slightly negative edge on all other days. But with a p-value of 0.22, a gap that size is still easy to write off as chance in a 16-day sample.

The charts

Average cumulative return after trigger (next 20 sessions)
What this chart says

This traces the average path of both baskets over the 20 sessions after a trigger day, and the key feature is that the two lines never cross back: OIH starts at roughly 0.5% after one session and grinds to 4.2% by session 20, while XOP ends at 2.1%. The gap between them widens rather than closes, with the spread running between 0.2% and 2.4% along the way — the opposite of the mean reversion you would expect if this were just a delayed catch-down to crude. That steady divergence is the chart evidence for the backlog-and-pricing-power story: the services basket keeps leading the E&P-heavy producers instead of giving the gains back.

Trigger-day OIH vs XOP next-20-session returns
What this chart says

Each of the 16 trigger days is one dot, plotted by what OIH did over the next 20 sessions on the horizontal axis and what XOP did on the vertical. The averages sit 2.1 points apart in OIH's favor, but the ranges are almost identical — OIH spans about -9.9% to +14.2% and XOP about -9.7% to +13.2% — so the two baskets frequently move together and both swing far wider than the average edge. That overlap, not the headline means, is why the 62% win rate and the 2.1-point edge come with a p-value of 0.22 rather than a clean statistical result.

Conditional vs baseline next-20-session metrics

MetricTrigger daysAll valid days
OIH mean 20d forward (%)4.220.85
XOP mean 20d forward (%)2.110.96
OIH-XOP mean (%)2.11-0.12
OIH wins vs XOP (%)62.547.3
Observations (N)16685

Trigger events

datebrent_5d_ret_pctoih_5d_ret_pctoih_fwd20_pctxop_fwd20_pctoih_minus_xop_pct
2023-10-09-8.351.63-3.49-1.44-2.05
2024-05-08-6.283.41-5.87-3.36-2.51
2024-08-23-5.160.58-2.7-2.67-0.04
2025-04-11-9.881.758.813.15-4.35
2025-05-02-7.61.94-0.085.52-5.6
2025-06-30-11.890.1610.994.66.39
2025-07-01-8.332.814.651.573.08
2025-08-08-8.771.067.484.632.85
2025-10-06-7.061.348.9-4.913.8
2025-10-17-9.154.8214.218.455.76
2025-10-21-5.364.369.087.371.71
2026-03-26-7.585.85.6-9.6515.24
2026-04-10-6.722.461.26-2.283.54
2026-04-21-16.131.4310.068.271.79
2026-04-22-10.573.58.553.774.78
2026-06-12-6.23.42-9.870.79-10.65

The takeaway

Directionally, the answer is yes — when Brent dumps 5% while OIH is still up, OIH does tend to keep beating XOP over the following month, but the sample is too thin to call it a real edge rather than luck. On the 16 qualifying days since September 2023, OIH averaged a 4.22% gain over the next 20 sessions against 2.11% for XOP, a 2.11% spread in OIH's favor. That spread looks reasonably broad-based rather than one luckbox day: the median was 2.32%, slightly better than the average, and OIH won 10 of the 16 matchups (62.5%). For context, across all 685 comparable days in the window the pair was essentially a wash — OIH trailed XOP by 0.12% — so the conditional days did run about 2.2 points hotter than normal. But that gap is not statistically distinguishable from noise: the p-values come in around 0.20–0.22, meaning roughly a 1-in-5 chance you'd see a spread this size purely by chance, and with only 16 triggers a couple of big moves can swing everything. Treat this as a suggestive lean that fits your backlog/pricing-power story, not a confirmed signal — it's a pattern worth watching and paper-trading, not a basis for sizing up.

The fine print