AI Research BKRBKR_earningsOIHXLEmacro:brent_daily

BKR positive EPS surprise + down-Brent tape: OIH vs XLE next 20 sessions

12
BKR positive EPS surprises in window

The thesis has a clean logic: a positive BKR earnings surprise on a down-crude tape signals upstream spending is holding up, so a services-heavy basket like OIH should re-rate against a producer-heavy XLE. Over the past three years, that setup was rare — six qualifying events — and the evidence that runs from them points the wrong way.

The mean OIH-XLE spread over the next 20 sessions was -1.52%, with OIH ahead of XLE in only half the events: a coin flip the statistics do not rescue. The full methodology, event-by-event breakdown, and comparison against the broader window are in the analysis below.

The research question

Over the past ~3 years, when BKR posts a positive quarterly EPS surprise while Brent crude's trailing 20-day return is negative, does OIH outperform XLE over the next 20 trading sessions? I expect a services bellwether beating on a down-crude tape to signal resilient upstream spending, so the services-heavy OIH basket re-rates against a producer-heavy XLE basket.

How this was measured

BKR quarterly reports were screened for a positive EPS surprise (surprise_percentage > 0) in the most recent 36 months. For each report date, the Brent trailing 20-session return was computed from the last Brent daily observation strictly before the report date to respect the T+1 daily lag, and only negative readings were kept. OIH and XLE daily closes were joined on common trading sessions. The forward 20-session return was measured from the first common close on or after the report date to the close 20 sessions later. The event OIH-XLE spread was compared against the unconditional distribution of all 20-session OIH-XLE spreads in the same window using a paired t-test, an exact sign test, and Welch's test against baseline.

The key numbers

BKR positive EPS surprises in window
12
past ~36 months, surprise_percentage > 0
Met down-Brent condition
6
positive surprise + Brent trailing 20d < 0
Complete 20-session events
6
0 met condition but lacked full forward window
Mean OIH next-20d return
-1.5254%
close-to-close from first common session on/after report date
Mean XLE next-20d return
-0.0008%
close-to-close from same anchor
Mean OIH-XLE spread
-1.5246%
positive = OIH outperforms XLE
Median OIH-XLE spread
-0.9057%
robust to outlier events
Fraction OIH > XLE
50.00%
3/6 events
Paired t-test p-value
0.2791
Paired two-sided; p=0.2791 ≥ 0.05 → no statistically clear difference
Exact sign-test p-value
1.3125
Sign test; p=1.3125 ≥ 0.05 → no statistically clear win-rate edge
Baseline mean OIH-XLE spread
-0.3255%
all 20-session windows, N=729
Welch p-value vs baseline spread
0.3863
Event vs all-window spread; p=0.3863 ≥ 0.05 → no statistically clear difference

Reading the numbers

Six qualifying events gave OIH an average next-20-session return of -1.53%, essentially matching XLE's 0.0%; OIH beat XLE only 3 of 6 times. Those differences are not statistically meaningful (p-values around 0.28 to 0.39), so the expected services-led re-rating has no clear support.

The charts

OIH vs XLE 20-session forward return after qualifying BKR reports
What this chart says

Each dot on this scatter pairs OIH's and XLE's 20-session forward returns after a qualifying BKR report. The points do not lean systematically toward OIH beating XLE: OIH averaged -1.53% while XLE averaged essentially 0.0%, and OIH won only 3 of 6 events. The lone strong OIH outcome near +5.6% is not enough to overcome the several events where OIH was worse, matching the report's conclusion that the down-crude positive-surprise signal did not produce a repeatable services-led re-rating.

OIH-XLE 20-session spread: qualifying events vs all windows
What this chart says

This box plot compares the OIH-minus-XLE 20-session spread for the 6 qualifying events against all 729 rolling windows. The qualifying-event spread sits below zero, with a mean of -1.52% and a median of -0.91%, meaning OIH trailed XLE on net; the all-window baseline is only -0.33%. The event box overlaps heavily with the broad all-window distribution, and the Welch p-value of 0.386 says the event-period spread is not clearly different from normal. In short, these six events look like an ordinary, mildly OIH-negative stretch rather than the hypothesized re-rating.

Qualifying BKR events and 20-session forward returns

reported_datebrent_trailing_20doih_fwd_20dxle_fwd_20doih_minus_xle
2023-10-25-0.0684-0.051-0.0322-0.0189
2024-07-25-0.0321-0.1025-0.0334-0.0691
2024-10-22-0.02210.05640.05430.0021
2025-04-22-0.04520.03910.02030.0189
2025-07-22-0.0326-0.0085-0.00930.0008
2025-10-23-0.1057-0.0250.0002-0.0252

The takeaway

No — over the past ~3 years, this set-up did not deliver the OIH-over-XLE edge the thesis expects, and the small amount of evidence actually leans the other way. Across the six BKR beats that coincided with a down-Brent tape, OIH averaged about -1.5% over the next 20 sessions while XLE was basically flat, for a mean OIH-XLE spread of -1.52% (median -0.91%). OIH beat XLE in only half of the events (3 of 6), and the tests back up that coin-flip look: the paired t-test p-value is around 0.28, and comparing the event spread to the all-window baseline (mean -0.33%, 729 windows) gives p≈0.39. Neither comes close to ruling out luck. Bottom line: the hypothesized re-rating is not visible in this window — treat it as unsupported rather than a usable signal.

The fine print