OIH vs XLE after rising CPI + positive 20d Brent momentum (last ~3 years)
The average looks persuasive until you count the winners. When CPI printed hotter than the prior month and Brent's 20-day return was positive, OIH out-earned XLE by an average of 1.10% over the following 20 sessions — versus -0.62% on non-signal days. A 1.7-point swing between regimes is the kind of gap that gets called a pattern.
But OIH beat XLE on just 50.4% of the 135 signal days, and the median spread was +0.03%, essentially flat. Nearly all of the average is being carried by a few outsized episodes rather than a repeatable day-in, day-out advantage.
The full breakdown — signal construction, the distribution of spreads, and why the significance test flatters the result — is in the analysis below.
Over the past ~3 years, when the latest CPI inflation reading is higher than the prior month and Brent crude's 20-day return is positive, does OIH outperform XLE over the next 20 trading sessions? I expect an accelerating-inflation regime with oil still bid to favor high-beta oil-services leverage over diversified energy equities, so OIH leads until the macro impulse fades.
How this was measured
Built a daily panel from OIH and XLE ETF closes. Brent daily closes were aligned to that trading calendar and a 20-session return was computed. CPI inflation was measured as year-over-year change in the CPI index, and a rising-inflation state was defined as the latest available CPI YoY reading exceeding the prior month's reading, with a 15-calendar-day release lag. A signal day occurs when that state coincides with a positive 20-session Brent return. Forward 20-trading-session returns were then computed for both ETFs, and the outcome is the OIH-minus-XLE spread. Signal-day spread distribution is compared with non-signal days using a Welch t-test, though overlapping 20-day windows make the p-value descriptive rather than strictly inferential.
The key numbers
Reading the numbers
On the 135 days when inflation was accelerating and oil was bid, OIH beat XLE by about 1.1% on average over the next 20 sessions, versus a 0.6% shortfall on other days. But the win rate was 50.4% — essentially a coin flip — so the average edge came from a few big wins, not steady outperformance.
The charts
This box plot compares the 20-session OIH-minus-XLE spread on signal days (mean +1.1%) against non-signal days (mean -0.6%). The detail worth your eye is that the two boxes sit almost on top of each other: signal days run from -9.3% to +16.7% and non-signal days from -11.3% to +14.2%, and the signal-day median is essentially zero even though its mean is positive. That combination — a positive mean with a near-zero median and a 50.4% win rate — means the typical signal day does not deliver OIH leadership; a handful of large oil-services rallies pull the average up, which is why the edge is directionally supportive of your thesis but not dependable day to day.
These bars break the average 20-session return into each ETF, split by day type. On signal days the tall bar is OIH at 1.35% versus XLE at 0.26%, the pattern your thesis predicts — high-beta services leverage leads when CPI is rising and crude is rallying. Off-signal days flip the picture: OIH returns 0.96% while XLE returns 1.58%, so the diversified energy name is the better performer when the macro impulse is absent. The takeaway is that the condition itself, not simply owning energy, is what lines up with OIH leadership.
Forward 20-session summary
| Group | N | Mean OIH fwd20 | Mean XLE fwd20 | Mean spread | Win rate OIH>XLE |
|---|---|---|---|---|---|
| Signal days | 135 | 0.0135 | 0.0026 | 0.011 | 0.5037 |
| Non-signal days | 576 | 0.0096 | 0.0158 | -0.0062 | 0.4705 |
| All valid days | 711 | 0.0103 | 0.0133 | -0.0029 | 0.4768 |
Signal-day event log (most recent 25)
| signal_date | cpi_yoy_latest | cpi_yoy_prev | brent_20d_ret | oih_fwd20 | xle_fwd20 | spread_fwd20 |
|---|---|---|---|---|---|---|
| 2026-03-19 | 3.26 | 2.41 | 0.5177 | 0.0217 | -0.0692 | 0.0909 |
| 2026-03-20 | 3.26 | 2.41 | 0.6278 | 0.0424 | -0.0635 | 0.1058 |
| 2026-03-23 | 3.26 | 2.41 | 0.4435 | 0.0296 | -0.0601 | 0.0897 |
| 2026-03-24 | 3.26 | 2.41 | 0.5225 | 0.021 | -0.0662 | 0.0872 |
| 2026-03-25 | 3.26 | 2.41 | 0.5439 | 0.0398 | -0.0602 | 0.1 |
| 2026-03-26 | 3.26 | 2.41 | 0.5823 | 0.056 | -0.0766 | 0.1325 |
| 2026-03-27 | 3.26 | 2.41 | 0.7032 | 0.0606 | -0.0939 | 0.1545 |
| 2026-03-30 | 3.26 | 2.41 | 0.5779 | 0.0981 | -0.0692 | 0.1673 |
| 2026-03-31 | 3.26 | 2.41 | 0.5213 | 0.1062 | -0.0303 | 0.1365 |
| 2026-04-01 | 3.26 | 2.41 | 0.4659 | 0.1378 | 0.0182 | 0.1195 |
| 2026-04-02 | 3.26 | 2.41 | 0.4405 | 0.1093 | -0.0072 | 0.1165 |
| 2026-04-07 | 3.26 | 2.41 | 0.4649 | 0.1077 | 0.0349 | 0.0729 |
| 2026-04-08 | 3.26 | 2.41 | 0.3592 | 0.0508 | -0.0201 | 0.0708 |
| 2026-04-09 | 3.26 | 2.41 | 0.3083 | 0.0195 | -0.0121 | 0.0316 |
| 2026-04-10 | 3.26 | 2.41 | 0.163 | 0.0126 | -0.0228 | 0.0354 |
| 2026-04-13 | 3.26 | 2.41 | 0.1942 | 0.0445 | 0.0019 | 0.0426 |
| 2026-04-14 | 3.26 | 2.41 | 0.1747 | 0.0817 | 0.031 | 0.0506 |
| 2026-04-15 | 3.26 | 2.41 | 0.0603 | 0.0766 | 0.0325 | 0.0441 |
| 2026-04-21 | 3.81 | 3.26 | 0.0226 | 0.1006 | 0.0927 | 0.0078 |
| 2026-04-22 | 3.81 | 3.26 | 0.0463 | 0.0855 | 0.0555 | 0.03 |
| 2026-04-23 | 3.81 | 3.26 | 0.0377 | 0.0458 | 0.0367 | 0.0091 |
| 2026-04-30 | 3.81 | 3.26 | 0.0391 | -0.072 | -0.0557 | -0.0163 |
| 2026-05-15 | 3.81 | 3.26 | 0.1554 | -0.0543 | -0.0707 | 0.0164 |
| 2026-05-18 | 4.25 | 3.81 | 0.1289 | -0.1087 | -0.0923 | -0.0164 |
| 2026-05-19 | 4.25 | 3.81 | 0.0801 | -0.1202 | -0.1126 | -0.0076 |
The takeaway
Directionally, yes — when inflation is accelerating and Brent has been bid, OIH has out-earned XLE over the following month — but the edge is thinner than the headline average suggests, and it rests on a handful of outsized episodes rather than broad consistency. On the 135 signal days in the sample, OIH averaged a 1.35% forward 20-session gain against XLE's 0.26%, an average spread of +1.10%, while non-signal days actually ran the other way at -0.62% — a 1.71-point swing that's hard to dismiss as noise on its face. The catch is that OIH beat XLE on only 50.4% of those signal days, essentially a coin flip, and the median spread was +0.03% — basically flat. In other words, the positive average is doing all its work through a few big winners (the recent March 2026 cluster shows spreads of +9% to +10% on days when XLE fell hard while OIH held up), not through a repeatable day-in, day-out advantage. The Welch test comes back at p = 0.0011, which sounds decisive, but overlapping 20-day windows mean that number is descriptive rather than a true significance test — with that many overlapping observations, the honest effective sample is closer to a few distinct CPI-and-oil episodes. Treat this as a suggestive lean, not a proven signal: the regime tilt is real in the averages, but the consistency isn't there to trade it mechanically.
The fine print
- OIH beat XLE on only ~50% of signal days and the median spread is roughly zero — the positive average comes from a few large winners, not a steady edge.
- Overlapping 20-day forward windows mean the 135 signal days are far from independent; the effective sample is closer to the number of distinct rising-CPI episodes.
- The p-value of 0.0011 is descriptive only — it isn't a valid significance test given the window overlap.
- Three years, one pair, one macro regime, plus a 15-day CPI release-lag assumption and a year-over-year CPI definition that a different choice would change.