AI Research PSXPSX_newsmacro:brent_daily

PSX top-quintile news sentiment on Brent down >1% vs rising-Brent days: forward 5-day return

63
Event-eligible top-quintile sentiment days

A refiner’s good news on a day crude gets slammed is the kind of setup that looks like a clean decoupling trade: if the market believes crack spreads are strong, PSX can rally while the raw input falls. That story drove the question here — does top-quintile PSX news sentiment on down-Brent days actually lead to bigger forward gains than the same sentiment on rising-crude days?

Over three years, the direction holds. Top-sentiment days when Brent fell more than 1% averaged a +2.21% forward 5-day return versus +1.63% when Brent rose, a +0.58 point gap. But the evidence is fragile: with only 18 down-crude days, the t-test p-value sits near 0.66, so the gap could easily be noise. The full stats and charts are below.

The research question

For PSX over the past 3 years, do top-quintile daily PSX news-sentiment scores on days when Brent crude falls more than 1% produce larger forward 5-day returns than top-quintile sentiment days when Brent rises? I expect bullish refiner coverage on a down-crude tape to signal crack-spread strength, drawing buyers who treat PSX as decoupling from crude rather than selling the oil beta.

How this was measured

PSX minute bars were resampled to daily closes, and forward 5-day return was computed as close[t+5]/close[t]-1 using observable trading-day closes. Daily PSX news sentiment was the mean of ticker_sentiment_score, falling back to overall_sentiment_score when missing, averaged across same-day headlines. Event-eligible days needed overlapping PSX sentiment, Brent return, and a 5-day forward PSX return. Top-quintile days are the upper 20% of the daily sentiment distribution. Brent daily return was computed from brent_daily_df after forward-filling the daily settlement to PSX trading dates. Down-crude group = top-quintile sentiment days with Brent return < -1%; rising-crude group = top-quintile sentiment days with Brent return > 0. The group means were compared with Welch's two-sample t-test and a Mann-Whitney U test.

The key numbers

Event-eligible top-quintile sentiment days
63
Days with PSX news, Brent return, and 5-day forward return available
Top-quintile sentiment threshold
0.3204
80th percentile of daily mean PSX news sentiment
Down-crude group N
18
Top-quintile sentiment days with Brent return < -1%
Rising-crude group N
36
Top-quintile sentiment days with Brent return > 0
Down-crude mean forward 5d return
2.2097%
Brent < -1%, N=18 top-quintile sentiment days
Rising-crude mean forward 5d return
1.6338%
Brent > 0, N=36 top-quintile sentiment days
Difference: down minus up
0.5759%
Positive means down-crude top-quintile days did better
Welch t-statistic
0.441
Unequal-variance two-sample test
Welch p-value
0.6616
p=0.6616 >= 0.05 -> no statistically clear difference
Mann-Whitney p-value
0.3834
p=0.3834 >= 0.05 -> no statistically clear rank difference
Down-crude positive share
72.22%
Fraction of down-crude top-quintile days with positive forward 5d return
Rising-crude positive share
55.56%
Fraction of rising-crude top-quintile days with positive forward 5d return

Reading the numbers

Top-quintile sentiment days with Brent down more than 1% averaged +2.21% forward 5-day returns versus +1.63% on rising-Brent days, a +0.58% gap. But the p-values (0.66 and 0.38) say that gap is not statistically clear.

The charts

PSX forward 5-day return: top-quintile sentiment by same-day Brent move
What this chart says

This box plot compares the spread of PSX forward 5-day returns on top-quintile sentiment days under the two Brent conditions. The down-crude group averaged +2.21%, a touch higher than the +1.63% average for rising-crude days, and had a positive-return share of 72.2% versus 55.6%. But the two distributions overlap heavily: the rising-crude group actually has the larger maximum return at +14.67%, while the down-crude group's best day is only +8.91%. That overlap is why the Welch and Mann-Whitney tests return p-values of 0.66 and 0.38, meaning the apparent edge in average return is not statistically distinguishable from luck.

Mean forward 5-day return by condition
What this chart says

This bar chart makes the headline comparison easy to see: the left bar for down-crude top-quintile days is 2.21% and the right bar for rising-crude days is 1.63%, a gap of 0.58 percentage points. The direction matches your hypothesis, but the bar heights are based on only 18 and 36 trading days respectively, and a Welch t-statistic of 0.44 with p=0.66 indicates the gap is small relative to the variation within each group. So while the pattern leans in your favor, the chart is not strong enough evidence to call it a reliable edge.

Top-quintile PSX sentiment days by Brent condition

ConditionNMean fwd5Median fwd5Std fwd5Positive share
TopQ & Brent < -1%180.02210.02150.0430.7222
TopQ & Brent > 0360.01630.01060.04940.5556
TopQ & Brent > +1% (robustness)240.02290.01270.05460.5833
All top-quintile days630.0154

The takeaway

Not reliably — the direction runs with your hypothesis, but the stats say the edge could easily be luck. Top-quintile sentiment days on a down-crude tape averaged +2.21% over the next 5 days versus +1.63% after rising-crude days, a +0.58 percentage-point gap, and the median tells the same story (+2.15% vs +1.06%). But that gap is far from conclusive: with only 18 down-crude days and 36 rising-crude days, the t-test p-value of 0.66 means a difference this size would occur by chance about two-thirds of the time, and the rank-based test (p=0.38) is still weak. The positive-day shares (72% vs 56%) tilt the same way but don't rescue it. Notably, top-quintile days with Brent up more than 1% also averaged +2.29%, so the footprint isn't unique to down tapes. Takeaway: the crack-spread/decoupling story looks plausible but is unproven in this sample — too thin, too overlapping, and too easily explained by chance to treat as a real edge.

The fine print