AI Research HALmacro:brent_daily

HAL/Brent 20-day realized-vol ratio: top quintile and forward 10-day returns

192
Usable days after warmup

The comfortable story for oil-services names is that volatility shocks create washouts, and washouts mean-revert. So the test is straightforward: when HAL's 20-day realized volatility runs in its top quintile relative to Brent's, do the next ten days tend to beat the baseline?

Across the usable sample, the answer is no, and the signal points slightly the wrong way. The 31 top-quintile days averaged roughly 1.3% forward 10-day returns versus about 2.1% for the bottom 80% and 2.0% overall. But with a Welch p-value near 0.69 and a rank correlation of -0.04, that gap is noise, not edge. The full methodology, charts, and return breakdowns are below.

The research question

For HAL over the past ~3 years, does a top-quintile ratio of HAL's 20-day realized volatility to Brent crude's 20-day realized volatility predict above-baseline forward 10-day returns? I expect oil-services shares to overreact to crude volatility shocks, so extreme relative volatility marks a fear-driven washout that mean-reverts as oil stabilizes.

How this was measured

Daily HAL close-to-close returns were resampled from minute bars, and 20-day realized volatility was calculated as the rolling standard deviation of daily returns annualized by sqrt(252). The same 20-day realized volatility was calculated for Brent crude from the daily macro series. The signal is the ratio HAL_RV20 / Brent_RV20, evaluated each day against its trailing 252-day 80th percentile, so the top-quintile classification uses only data available at the close of day t. Forward 10-trading-day HAL returns were measured as close[t+10]/close[t]-1. Top-quintile and bottom-80% forward returns were compared with a Welch two-sample t-test and a one-sided p-value.

The key numbers

Usable days after warmup
192
2025-03-20 to 2026-07-17
Top-quintile signal days
31
Actual share 16.1%
Bottom-80% days
161
Top-quintile mean fwd 10d
1.2571%
n=31
Bottom-80% mean fwd 10d
2.1420%
n=161
Full-sample baseline mean
1.9991%
All days in usable post-warmup sample
Edge vs bottom 80%
-0.8849%
Positive = top-quintile outperforms bottom 80%
Top-quintile median fwd 10d
1.2870%
Top-quintile positive share
58.06%
Bottom-80% positive share
55.90%
Welch t-statistic
-0.490
Top vs bottom 80%; positive favors top-quintile
One-sided Welch p-value
0.6867
H1: top > bottom; p=0.6867 >= 0.05 or no positive edge -> no support
Spearman rho (ratio vs fwd 10d)
-0.043
Rank correlation across full usable sample

Reading the numbers

Days when HAL’s volatility was most extreme relative to Brent averaged +1.26% over the next 10 days, versus +2.14% for all other days — a -0.88% gap. With p=0.69, that gap is small enough to be pure chance, so the fear-driven rebound idea isn't supported.

The charts

HAL/Brent 20-day realized-vol ratio and trailing 80th percentile
What this chart says

This line chart tracks HAL's 20-day realized volatility relative to Brent's, with a trailing 80th-percentile line around 1.43 to 1.54 marking the top-quintile cutoff. Most of the time the ratio sits below that line — its average is about 1.11 — so extreme relative-volatility days are rare spikes rather than a persistent state. The occasional crossings above the line are exactly the signal days tested, but they don't cluster into obvious panic episodes.

Forward 10-day HAL return by RV-ratio bucket
What this chart says

The box plot compares the spread of forward 10-day returns for the bottom 80% of RV-ratio days versus the top quintile. The top-quintile group's average return is 1.26%, below the bottom group's 2.14%, and its range (-19.8% to +23.5%) is not shifted higher than the bottom group's range. So extreme relative volatility did not precede the kind of outsized rebound you'd expect from a fear-driven washout.

Mean forward 10-day return by RV-ratio bucket
What this chart says

The bar chart puts the headline comparison side by side: the top-quintile mean forward 10-day return (1.26%) is actually lower than the bottom-80% mean (2.14%). The gap is -0.88 percentage points, and the one-sided p-value of 0.6867 means there's no statistical evidence that extreme RV-ratio days beat ordinary days. In plain terms, high relative volatility at HAL didn't predict above-baseline forward returns in this sample.

Forward 10-day return summary by RV-ratio bucket

BucketNMeanMedianStdPositive share
Bottom 80%1610.02140.01390.10640.559
Top 20% RV ratio310.01260.01290.08920.5806
All usable days1920.020.01320.10360.5625

Recent top-quintile signal dates

DateRV ratioFwd 10d return
2025-10-271.5870.0358
2025-10-281.5380.0522
2025-10-291.59-0.0233
2025-10-301.6310.0052
2025-10-311.6350.0129
2025-11-031.654-0.0202
2025-11-041.619-0.0149
2025-11-051.581-0.0233
2025-11-061.54-0.0647
2025-11-071.541-0.0635
2025-11-121.549-0.038
2025-11-171.5450.0135

The takeaway

The short answer is no: over the roughly 15 months of usable data, days when HAL's 20-day realized volatility sat in its top quintile relative to Brent's did not predict above-baseline forward 10-day returns. In fact, the signal pointed the wrong way — top-quintile days averaged +1.26% over the next 10 days versus +2.14% for the bottom 80% and +2.00% for all days. With only 31 top-quintile signals and a Welch p-value of 0.687, this is basically a coin flip: there is no meaningful evidence that extreme relative volatility marks a washout that mean-reverts. The median forward return was nearly identical across buckets (about +1.3% for both), and the positive-hit rate was also similar (58% vs. 56%), so the slightly lower average is just noise. The rank correlation between the RV ratio and forward returns was essentially zero at -0.043. For a trader hoping for a fear-driven rebound edge, this sample offers no support; if anything, it suggests high relative volatility has been mildly associated with weaker, not stronger, forward returns.

The fine print