HAL 20-day Brent beta around rising-10Y / positive-Brent regimes
When the 10-year yield jumps at least 20 basis points over 20 sessions while Brent is still climbing, does HAL's 20-day beta to crude actually fall? The intuition is reasonable: a rising discount rate should bite long-duration services backlog harder than spot oil exposure, so the stock ought to decouple from crude even with crude trending up.
The episode starts say yes. Across 16 such episodes, HAL's beta to Brent averaged 0.32 versus 0.45 in the prior 20 sessions, and 56% of those episodes showed beta falling. But widen to all 75 condition days and the average change shrinks to roughly -0.01, with only 44% falling, and the test statistic is nowhere near significant.
The full evidence, including the method and charts, is below.
Over the past ~3 years, when the 10-year Treasury yield rises by at least 20 basis points over the prior 20 trading days while Brent crude's 20-day return is positive, does HAL's 20-day realized beta to Brent crude fall relative to its prior 20-session beta? I expect a rising discount rate to hit long-duration services backlog harder than spot oil-beta, so HAL should decouple from crude even with oil trending higher.
How this was measured
Daily HAL close was computed from minute bars and joined to daily 10-year Treasury and Brent crude series. Both macro series were forward-filled onto HAL trading days. On each day, the trailing 20-session 10Y yield change and trailing 20-session Brent price return were computed. A condition day required the 10Y yield to have risen at least 20 bp over the prior 20 sessions while Brent's 20-session return was positive. HAL's 20-session rolling beta to Brent was estimated as rolling cov(HAL return, Brent return) / rolling var(Brent return) over the trailing 20 sessions, and its prior beta was the same estimate shifted 20 sessions earlier. Unique episodes were defined as the first day of each contiguous block of condition days. The primary comparison is beta at episode start versus its prior 20-session value.
The key numbers
Reading the numbers
On the 16 separate occasions when both conditions held, HAL's beta to Brent averaged 0.14 lower than in the prior 20 sessions and fell in 56% of them — but a p-value of 0.25 means a move this size could easily be noise, so it reads as a hint rather than proof.
The charts
This bar pair averages HAL's 20-day beta to Brent across the 16 episode starts: 0.4545 in the prior 20 sessions versus 0.3154 once the 10-year yield had risen at least 20bp while Brent's 20-day return was positive. The shorter bar on the left is the point — the current reading is roughly a third lower than the prior one, the same 0.14 gap that sits behind the headline decoupling figure. Against your question, that is the average shape you'd expect if the discount rate mattered more than spot crude, but it is a sample average over just 16 starts, not a rule that holds for each one.
Here all 75 condition days are counted by how much each day's beta moved versus the prior 20 sessions, and the pile sits essentially on top of zero — the average change is only -0.0084. The detail that matters is the width rather than the centre: readings run from -1.70 to +0.94, so individual days swing hard in both directions. That is why just 44% of event days show a fall; day by day, the decoupling signal is much weaker and noisier than the episode-level number suggests.
The 16 episode starts are plotted by the size of the 10-year yield rise (20 to 46 basis points, averaging about 30) against the beta change, which averages -0.139 and spans -1.48 to +0.39. The eye should go to the single point sitting far below the rest — it pulls the average down, while the remaining points cluster close to zero. So there's no visible tilt saying a larger yield jump produces a larger decoupling; the negative average leans on one episode rather than a broad pattern across the sample.
Unique condition episodes
| episode_start | 10y_20d_bps | brent_20d_ret | beta_current | beta_prior | beta_change |
|---|---|---|---|---|---|
| 2024-01-18 | 23 | 0.0547 | 0.248 | 0.332 | -0.084 |
| 2024-01-23 | 28 | 0.0116 | 0.257 | 0.314 | -0.057 |
| 2024-02-12 | 21 | 0.0499 | 0.57 | 0.355 | 0.215 |
| 2024-02-29 | 26 | 0.0192 | 0.356 | 0.349 | 0.007 |
| 2024-04-05 | 28 | 0.0718 | 0.191 | -0.199 | 0.39 |
| 2024-10-04 | 26 | 0.0893 | 0.447 | 0.565 | -0.118 |
| 2024-10-22 | 45 | 0.0085 | 0.593 | 0.451 | 0.142 |
| 2024-10-25 | 46 | 0.035 | 0.568 | 0.479 | 0.09 |
| 2024-11-18 | 34 | 0.022 | 0.305 | 0.609 | -0.305 |
| 2024-11-22 | 20 | 0.0246 | 0.365 | 0.595 | -0.23 |
| 2024-12-30 | 30 | 0.0045 | 0.39 | 0.199 | 0.191 |
| 2025-05-29 | 24 | 0.0087 | 0.296 | 1.78 | -1.484 |
| 2026-03-20 | 31 | 0.6278 | -0.105 | 0.578 | -0.683 |
| 2026-05-18 | 35 | 0.1835 | 0.39 | 0.006 | 0.384 |
| 2026-07-23 | 30 | 0.461 | 0.088 | 0.418 | -0.329 |
| 2026-07-31 | 26 | 0.4147 | 0.086 | 0.442 | -0.356 |
The takeaway
Directionally the answer is yes, but the numbers don't back it up strongly enough to call it a real effect. At the 16 episode starts where the 10-year yield rose at least 20bp over 20 sessions and Brent was up, HAL's 20-day beta to Brent averaged 0.32 versus 0.45 in the prior 20 sessions — a drop of about 0.14, and 56% of those episodes did show beta falling. Widen the lens to all 75 condition days, though, and the average change collapses to roughly -0.01 with only 44% falling, i.e. a slight majority rose. That gap tells you most of the episode-start decoupling comes from a handful of larger moves rather than a broad pattern. The formal test agrees: t of -1.21 and a p-value of 0.245 means roughly a 1-in-4 chance of seeing this by luck alone, so this is closer to a coin flip than a signal. With just 16 independent episodes and a noisy 20-day beta estimate, treat this as a suggestive anecdote, not a tradeable relationship — you'd need more episodes or a longer-window beta before leaning on the discount-rate-hits-backlog thesis.
The fine print
- Only 16 unique episodes and 75 condition days — well short of the ~30 independent observations you'd want for a firm statistical read.
- Episode starts still overlap when they fall within 20 sessions of each other, so the effective sample is even smaller than 16.
- Beta is single-factor to Brent only; it ignores market, rates, and oil-services factors that also drive HAL.
- The 20bp threshold and 20-session windows are arbitrary — results can flip at 15bp or 60-session settings.
- Macro series are date-aligned to HAL close without a release lag, which a strict point-in-time test would require.