MSFT negativity-bias test: bottom-decile vs top-decile news-sentiment days (same-session absolute move)
Negative sentiment days do not appear to hit Microsoft harder than positive ones. Over roughly three years of overlapping price and news data (669 days; 67 days in each decile), the mean absolute same-session move on bottom-decile sentiment days was 0.93% versus 0.99% on top-decile days — a mean gap of -0.064 percentage points with a Welch p-value of 0.6878, so the difference is statistically indistinguishable from noise.
This study paired resampled daily MSFT closes with an aggregate daily sentiment score and compared the absolute same-day returns in the bottom and top deciles. Read on for the full sample description, distributions, robustness checks, and why this window and tagging approach may mask event-level asymmetries.
For MSFT over the past ~3 years, does bad news hit harder than good news helps — is the average absolute same-session price move on its most negative-sentiment days materially bigger than on its most positive-sentiment days? Thesis: the bottom-decile-sentiment sessions show larger swings than the top-decile ones, so Microsoft prices fear more violently than optimism in a textbook negativity bias.
How this was measured
Resampled MSFT minute bars to daily closes over the trailing ~3 years and computed same-session close-to-close returns. Aggregated MSFT_news to a daily mean sentiment (ticker_sentiment_score preferred; overall_sentiment_score as fallback). Joined days with both price and sentiment, then classified sentiment tails via deciles (robust rank-based qcut; fallback to 10th/90th percentile thresholds when sample/uniqueness is limited). Compared the absolute same-day return distribution on bottom-decile (most negative) vs top-decile (most positive) sentiment days. Reported mean gap and a Welch two-sample t-test for the difference in means.
The key numbers
Reading the numbers
Across 669 news+price days, the average absolute same-day return is about 0.010977. The bottom-decile mean |return| is 0.009282 and the top-decile mean is 0.009922; the gap (-0.0006396) is small and not statistically significant (p=0.6878).
The charts
This box plot compares the distribution of absolute same-day returns for the 67 most negative-sentiment days and the 67 most positive-sentiment days. The means are almost identical (bottom mean 0.0093, top mean 0.0099) and the ranges overlap — bottom min 0.0002 to max 0.0406, top min 0.0 to max 0.046. Look at the long tails: the top-decile group actually reaches a slightly higher maximum (0.046 vs 0.0406), so the negative tail does not show clearly larger swings here.
The side-by-side bars show mean absolute moves: bottom 0.0093 versus top 0.0099. The reported mean gap is -0.000639556 (bottom minus top), i.e., the top-decile mean is slightly larger, not smaller. Given the statistical result (p=0.6878), this small difference is not a reliable sign that bad news produces bigger same-day swings.
The scatter plots 669 days of daily mean sentiment (range -0.4486 to 0.7359, mean 0.2113) against absolute same-day return (range 0.0 to 0.116, mean 0.011). Absolute moves sit around the overall mean (~0.011, matching the baseline 0.010977) across both negative and positive sentiment values, with no visible upward trend at the negative extreme. In other words, more negative daily sentiment does not correspond to consistently larger absolute same-day returns in this sample.
Sentiment-tail absolute-return summary
| group | N | mean_abs_ret | median_abs_ret | std_abs_ret |
|---|---|---|---|---|
| Bottom 10% | 67 | 0.0093 | 0.0066 | 0.0087 |
| Top 10% | 67 | 0.0099 | 0.0082 | 0.0096 |
The takeaway
No — we don't see bad news hitting harder. The mean absolute same-session move on bottom-decile days was 0.93% vs 0.99% on top-decile days, so negative days were actually slightly smaller on average. The raw mean gap is -0.0006396 (about -0.064 percentage points); Welch t = -0.4028 with p = 0.6878, which translates to roughly a 69-in-100 chance the observed gap is just noise. Each tail contains 67 days out of 669 overlapping news+price days, and medians (0.66% vs 0.82%) plus similar stds show no hidden large skew favoring the negative tail. Practical takeaway: there is no clear negativity bias in same-session absolute MSFT moves over this window — any asymmetry is tiny and statistically indistinguishable, so test different timing windows or event-level tagging if you want a cleaner read.
The fine print
- Tails are only 67 days each — limited power in the extremes.
- Close-to-close grouping can misassign post-market headlines to the same session, blurring intra-day timing.
- Daily sentiment averages headlines, so opposing articles can cancel and understate true news dispersion.
- Equity returns are heavy-tailed; mean-based t-tests are sensitive to outliers despite reporting medians.