XLE forward 5-day returns after high-volume Brent-led selling: intraday close-location effect
The thesis was that XLE, closing in the top half of its range on high-volume, Brent-led selling, marks institutional accumulation and outperforms over the following five sessions. The measured gap between top-half and bottom-half closes: 0.13 percentage points. Welch's p-value: 0.87. That is a coin flip.
This study examined 754 daily XLE sessions, isolating 67 days where volume hit the top quintile and Brent fell more than 1%. Each event was split by where XLE closed within its high-low range — 20 top-half closes, 47 bottom-half — and compared on close-to-close 5-day forward returns.
What the data do show: both groups bounced, with roughly 70-72% of events followed by positive forward returns, versus a 0.31% mean baseline for all days. The edge, such as it exists, comes from the high-volume selloff itself. The full statistics, charts, and robustness checks follow below.
Over the past ~3 years, when XLE's daily volume is in the top quintile and Brent crude falls more than 1%, does XLE's forward 5-day return depend on whether it closes in the top or bottom half of its intraday range? I expect high-volume crude-led selling that closes in the top half to mark institutional accumulation and outperform, while close-at-lows selling signals continued de-risking.
How this was measured
Daily XLE bars were resampled from minute data. A day entered the event sample when XLE daily volume was in the top quintile of the available daily-volume distribution and Brent crude's daily return was below -1%. Each event day's close location within its own high-low range was computed as (close - low)/(high - low), and split into top half (>=0.5) versus bottom half (<0.5). The outcome is XLE's close-to-close 5-trading-day forward return. Group means, medians, positive-hit rates, Welch's t-test, and a Mann-Whitney U rank test compare top-half versus bottom-half closes. The same-calendar-day Brent return is used for the economic trigger, with the timing caveat noted below.
The key numbers
Reading the numbers
Top-half closes averaged 1.04% over the next five days versus 0.91% for bottom-half closes, but the 0.13 percentage-point gap is weak evidence (p=0.87) — far too noisy to call a real edge. The medians (2.13% vs 1.24%) point the same direction but the sample is only 20 vs 47 events.
The charts
The box plot shows the two distributions of 5-day forward returns side by side. The top-half group has a higher mean (1.04%) but a much tighter range, from about -3.6% to +4.3%, while the bottom-half group stretches from -11.7% to +8.1%. Despite the higher average, the top-half results sit well inside the bottom-half's spread, so the difference is not clearly separable from random noise. That directly undercuts the idea that closing in the top half of the range marks a reliably better forward return.
The bars compare mean forward returns: top-half closes at 1.04%, bottom-half closes at 0.91%, and the all-day baseline at 0.31%. Both event groups beat an ordinary day, but the top-half edge over bottom-half is only 0.13 percentage points. With 20 top-half versus 47 bottom-half events and a p-value of 0.87, the bar difference is well within random sampling variation. This chart supports the 'both recover after crude-led selling' part of the story, but not the 'top-half recovers more' part.
Event-day and baseline summary
| group | n | mean_fwd5 | median_fwd5 | std_fwd5 | fraction_positive |
|---|---|---|---|---|---|
| Top half close | 20 | 0.0104 | 0.0213 | 0.0287 | 0.7 |
| Bottom half close | 47 | 0.0091 | 0.0124 | 0.0366 | 0.7234 |
| All event days | 67 | 0.0095 | 0.0134 | 0.0342 | 0.7164 |
| All-day baseline | 749 | 0.0031 | 0.0051 | 0.0313 | 0.5661 |
The takeaway
The short answer is no: across 67 high-volume Brent-led selling days, XLE's 5-day forward return did not depend on whether it closed in the top or bottom half of its intraday range. Top-half closes averaged 1.04% versus 0.91% for bottom-half closes—a gap of about 0.13 percentage points that is indistinguishable from noise (Welch p=0.87). Medians told the same story (2.13% vs 1.24%), and the non-parametric rank test also came up empty (p=0.59). With only 20 top-half and 47 bottom-half events, plus overlapping forward windows, this sample cannot resolve a close-location effect; it is essentially a coin flip. What the data do show: both event groups bounced more than a typical day, with ~70-72% of events followed by positive 5-day returns and mean forward returns near 0.95% versus 0.31% for the all-day baseline. The edge, if any, comes from the high-volume Brent selloff itself, not from where XLE closed within the day's range.
The fine print
- Small sample: 67 events (20 top-half, 47 bottom-half); overlapping 5-day forward windows make the effective evidence thinner than the event count implies.
- Brent trigger uses calendar-day data with T+1 convention; if the signal must be known at XLE's close, the trigger should be lagged one business day.
- Volume cutoff is the in-sample 80th percentile; out-of-sample volume trends and share counts could shift the event set materially.
- Close location uses daily high/low only, ignoring path and auction prints; single-ticker, no controls for equity beta, OPEC/Fed events, or WTI-Brent basis.