AI Research HALXLE

Does HAL daily return lead XLE next-day return? (≈3‑year window)

752
Trading days analysed (lead panel)

The gap between same-day and next-day correlation tells the whole story at a glance. Over 752 overlapping trading days, HAL and XLE move in near lockstep on the same session—Pearson r of 0.78. But shift that relationship by a single day, and it collapses to essentially zero (r = -0.016). That is not a lagged signal that needs refining; it is the absence of one.

The thesis being tested is the familiar early-cycle story: that Halliburton's daily returns price in drilling trends before they ripple through the broader energy complex. The data here does not support it. XLE's next-day average after HAL up-days is barely higher than after down-days, and the directional gap is statistically insignificant.

This was a reasonably high-powered test, and the null came through cleanly. The full methodology, charts, and robustness checks are laid out in the analysis below.

The research question

Over the past ~3 years, does Halliburton's (HAL) daily return lead the Energy Select Sector SPDR Fund (XLE) by one trading day? Thesis: HAL acts as an early-cycle indicator for the energy sector, with its daily return predicting XLE's next-day move as oil-services activity prices in drilling trends before they ripple through the broader complex.

How this was measured

HAL and XLE daily close prices were resampled from minute bars to calendar‑day close. Daily simple returns were computed for both tickers. The lead hypothesis was tested by aligning HAL’s return on day t with XLE’s return on day t+1 (i.e. XLE return shifted forward by one trading day). Pearson and Spearman rank correlations quantify the linear and monotonic lead relationship, while same‑day (HAL_t vs XLE_t) correlation provides a no‑lead baseline. In addition, the sample was split into days when HAL closed up vs. down, and a Welch two‑sample t‑test compared XLE’s next‑day return distribution between the two groups — a directional test of the early‑cycle hypothesis. Only overlapping trading days were kept, and the analysis stays within the available 36‑month data window.

The key numbers

Trading days analysed (lead panel)
752
Days with both HAL return and valid XLE next‑day return
Pearson r (HAL_t → XLE_t+1)
-0.0160
Linear lead correlation
Pearson p‑value (lead)
0.6606
Two‑sided; p=0.6606 ≥ 0.05 → no statistically‑clear lead correlation
Spearman r (lead)
0.0344
Rank‑based, robust to outliers
Same‑day Pearson r (HAL_t ↔ XLE_t)
0.7807
n=753 days; baseline co‑movement without lead
XLE next‑day mean when HAL up
0.1055%
N=385 up‑days
XLE next‑day mean when HAL down
0.0183%
N=358 down‑days
Welch t‑stat (up vs down)
0.840
Positive → XLE next‑day higher after HAL up‑day
Welch p‑value (directional)
0.4014
Two‑sided; p=0.4014 ≥ 0.05 → no clear directional lead

Reading the numbers

Across 752 trading days, there is no meaningful one-day lead from HAL to XLE: the lead correlation is essentially zero (-0.02) and not statistically significant (p=0.66). HAL and XLE do move together the same day (r=0.78), but that is co-movement, not prediction.

The charts

HAL daily return vs XLE next‑day return
What this chart says

This scatter shows HAL's daily return on the x-axis and XLE's return the next trading day on the y-axis. If HAL led XLE, the points would tilt upward or downward; instead they form a round cloud with no visible slope, matching the -0.02 correlation. The cloud spans wide daily swings — HAL from about -11% to +18%, XLE next-day from about -9% to +9% — but those extremes are scattered randomly. In plain terms, today's HAL move tells you nothing about which way XLE goes tomorrow.

XLE next‑day return by HAL daily return direction
What this chart says

This box plot splits days into those when HAL ended up (385 days) versus down (358 days) and shows XLE's next-day return for each group. The average next-day return was only about 0.11% after HAL up days versus roughly 0.02% after HAL down days, and the ranges overlap heavily — XLE next-day returns went as high as 9.4% after a down HAL day. A t-test puts the p-value at 0.40, so this small gap is easily within normal random variation and does not support the lead thesis.

Lead‑effect summary statistics

MetricValuep‑value
Pearson r (lead)-0.0160.6606
Spearman r (lead)0.03440.3461
Same‑day Pearson r0.78070
Frac. XLE up after HAL up0.5584
Frac. XLE up after HAL down0.5279

The takeaway

Over 752 overlapping trading days, HAL's daily return does not lead XLE's next-day return — the thesis doesn't hold up. The lead correlation is essentially zero (Pearson -0.016, p≈0.66; Spearman 0.034, p≈0.35), while the same-day correlation is 0.78, meaning the two move together in the moment rather than one step ahead. Directional checks are just as weak: XLE averages +0.11% next day after HAL up-days versus +0.02% after HAL down-days, a gap nowhere near significant (p≈0.40), and XLE closes up 55.8% of the time after HAL up-days versus 52.8% after down-days — basically a coin flip. With 752 days, this is a reasonably confident null: if HAL led XLE by a tradable daily margin, a sample this size would likely have picked it up. The practical answer: don't use yesterday's HAL move as a next-day XLE signal. Same-day co-movement is real, but the early-cycle lead story isn't supported at daily frequency — any lead, if it exists, would need longer horizons or a different regime to show up.

The fine print