BKR vs OIH after OPEC/quota/output-cut headline-intensity spikes
The hypothesis was reasonable: OPEC supply headlines should force service-company backlogs to re-price faster than the broad oil-services basket, letting BKR lead OIH. Across roughly three years of overlapping BKR and OIH sessions, twelve days cleared the top-quintile bar for OPEC, quota, or output-cut headline intensity.
Those twelve days did not cooperate. BKR averaged +0.97% over the following 10 sessions while OIH averaged +1.40%, a spread of about -0.43%, and BKR came out ahead only a third of the time. On ordinary days the same spread averaged +0.53% — the opposite sign from what the idea predicts.
The catch is that none of it clears noise: p-values of 0.70 and 0.41, on twelve observations with overlapping windows. The full event construction, baselines, and spread tests are laid out below.
Over the past ~3 years, when headline intensity for OPEC/quota/output-cut keywords spikes into the top quintile of its prior 20-session distribution, does BKR outperform OIH over the next 10 trading sessions? I expect OPEC supply-action news to force faster re-pricing of service-company backlog than broad oil-services beta, so BKR leads the services basket.
How this was measured
Combined BKR_news and OIH_news, flagged headlines whose title or summary matched OPEC, quota, or output cut(s), and assigned each headline to the next common BKR/OIH trading session. Daily keyword-headline counts were then compared with the 80th percentile of their own prior 20 sessions, excluding the current session. On the resulting spike days, BKR and OIH close-to-close forward 10-session returns were measured, and the BKR-minus-OIH spread was tested against zero and against the non-event baseline spread.
The key numbers
Reading the numbers
On the 12 sessions where OPEC/quota/output-cut headlines spiked, BKR rose 0.97% over the next 10 sessions while the services basket OIH rose 1.40% — so BKR lagged by 0.43% on average. That gap is statistically meaningless (a p-value of 0.70 means a result this size would show up by chance most of the time), and BKR only beat OIH in 4 of the 12 cases.
The charts
This tracks how many OPEC/quota/output-cut keyword headlines appear each day against the trailing 80th-percentile threshold that defines an 'event'. The detail worth your eye is the thin threshold line: it sits at zero for the entire period (its minimum, maximum and average are all 0), because most days have no qualifying headlines at all while the daily count peaks at just 3. In practice that means any day with even one or two hits clears the bar, so the 12 'spike' events are better described as 'headline days' than as a rare top-quintile surge — a caveat that matters when judging how selective the test really was.
This histogram spreads out the 12 event outcomes for BKR minus OIH over the following 10 sessions. The range is wide — from about -6.2% to +6.7% — with a mean of -0.43% and a median of -0.55%, so the negative tilt is not just one bad observation. Only 4 of the 12 events were positive for BKR, and the bars lean left of zero, which is the opposite of the backlog-repricing lead the question expected.
The three bars put the event-day averages side by side: BKR at +0.97%, OIH at +1.40%, and the difference at -0.43%. The OIH bar is visibly the taller of the two, meaning the broad services basket, not the single service name, captured more of the post-headline move. For your question, that bar gap is the headline: OPEC supply-action news moved the whole services complex, and BKR did not outrun it.
Event-level 10-session forward returns
| event_date | keyword_headlines | BKR_fwd10 | OIH_fwd10 | BKR_minus_OIH |
|---|---|---|---|---|
| 2024-07-22 | 1 | -0.015 | -0.082 | 0.067 |
| 2024-08-02 | 1 | -0.0191 | -0.01 | -0.0092 |
| 2024-08-30 | 1 | -0.037 | -0.0735 | 0.0365 |
| 2025-04-22 | 1 | -0.0315 | -0.0002 | -0.0313 |
| 2025-08-07 | 1 | 0.0168 | 0.0184 | -0.0016 |
| 2025-10-14 | 2 | 0.0684 | 0.114 | -0.0456 |
| 2026-01-05 | 1 | 0.0438 | 0.0457 | -0.0019 |
| 2026-01-23 | 1 | 0.1081 | 0.0855 | 0.0226 |
| 2026-02-13 | 1 | 0.0555 | 0.0252 | 0.0303 |
| 2026-04-29 | 3 | -0.0409 | -0.0268 | -0.0141 |
| 2026-05-04 | 1 | -0.0372 | 0.0245 | -0.0617 |
| 2026-05-06 | 1 | 0.0042 | 0.0465 | -0.0423 |
The takeaway
No — on this data, OPEC/quota/output-cut headline spikes did not give BKR an edge over OIH; if anything the relationship ran the other way. Across the 12 spike days that cleared the top-quintile bar, BKR averaged +0.97% over the following 10 sessions while OIH averaged +1.40%, leaving a BKR-minus-OIH spread of about -0.43% — and the median was worse, around -0.55%. BKR beat the basket in only a third of those events (33% win rate), while on ordinary non-event days the same spread averaged +0.53%, which is the opposite sign from what the hypothesis predicts. The catch is that none of this is statistically distinguishable from noise: the event spread's p-value of 0.70 says a gap this size would show up roughly 70% of the time by luck alone, and testing event days against normal days gives p = 0.41, also nowhere near a real difference. Twelve observations with overlapping 10-day windows is thin enough that even that weak negative tilt shouldn't be trusted as a genuine reversal — the honest read is no measurable effect, not a confirmed anti-signal. Practically: don't build a BKR-over-OIH trade around OPEC headline intensity, at least not with a simple count-based top-quintile trigger; if there's a backlog re-pricing story here, this test doesn't capture it.
The fine print
- Just 12 spike events, so treat this as descriptive; nothing here supports a firm conclusion in either direction.
- Forward 10-session windows overlap when events cluster, so the effective sample is smaller than 12 and the p-values flatter the evidence.
- The spike threshold is calibrated on the same history it's tested on — not an out-of-sample rule you could trade.
- Headline counts come only from the BKR and OIH feeds, so the same story can be counted twice; and BKR may itself sit inside OIH, making this partly a constituent-versus-basket comparison.