Research

Live empirical notes built on the same dataset powering Sentiment Live.

Descriptive analytics only — not investment advice. Results may change as data updates.

Dataset snapshot

Research artifacts are generated from your latest scheduled pipeline output.

LIVE

Updated

2026-04-05

Date range

2024-10-31 .. 2026-04-03

Studies

10

Tickers

507

Obs (panel)

261179

Frequency

Daily

Scope

S&P 500 snapshot

Outputs

Reproducible JSON

Explore studies

Search titles / specs / tags. Filter categories like an actual research index.

Sort

10 / 10 studies

Category

Top tags

Contemporaneous relationships

1 studies

Empirical research notes built from the live Sentiment Live dataset.

Section takeaways

Same-day sentiment vs same-day returns: score_mean is positive in TS (β=0.0269142***, t=4.064152054961017, R²=0.07778476568949322).

Same-day sentiment vs same-day returns

TS (HAC) + ticker fixed-effects panel: y_ret(t) ~ score_mean(t) (+ n_total(t)).

Key finding: Same-day sentiment vs same-day returns: score_mean is positive in TS (β=0.0269142***, t=4.064152054961017, R²=0.07778476568949322).

time-seriespanelHACfixed effectsreturnssentiment

β(score_mean) TS

β(score_mean) FE

Cross-sectional pricing

1 studies

Empirical research notes built from the live Sentiment Live dataset.

Section takeaways

Fama–MacBeth estimates average cross-sectional pricing of sentiment signals using daily cross-sectional regressions.

Fama–MacBeth: next-day returns on sentiment (cross-sectional)

Per-day cross-sectional OLS y_ret(t+1) on score_mean(t) (+ n_total). Report mean slopes and NW t-stats.

Key finding: Fama–MacBeth estimates average cross-sectional pricing of sentiment signals using daily cross-sectional regressions.

Fama-MacBethcross-sectionalNewey-Westpredictive

Event study

1 studies

Empirical research notes built from the live Sentiment Live dataset.

Section takeaways

Event-study profiles help distinguish immediate reaction vs post-event drift.

Event study: return dynamics around extreme sentiment

Panel event study using within-ticker z-score of sentiment. Compare average CAR/AAR paths for extreme positive vs negative sentiment days.

Key finding: Event-study profiles help distinguish immediate reaction vs post-event drift.

event studyextremesCARAARpanel

2024-10-31..2026-04-03

Predictability

4 studies

Empirical research notes built from the live Sentiment Live dataset.

Section takeaways

Piecewise specification tests whether sentiment effects differ between positive and negative territory.
Distributed lag model estimates how sentiment impacts returns over multiple days (lags 0..5).
Horizon profiles show whether any predictability is short-lived or persists over longer windows.

Asymmetry: piecewise sentiment effects on next-day returns

Allow different slopes for positive vs negative sentiment: r(t+1) ~ score_pos(t) + score_neg(t) (+ controls).

Key finding: Piecewise specification tests whether sentiment effects differ between positive and negative territory.

asymmetrynonlinearpanelfixed effects

β(score_mean_pos) FE

β(score_mean_neg) FE

Distributed lags: next-day returns on lagged sentiment

Panel FE + TS HAC: y_ret(t+1) on score_mean lags 0..5 (+ n_total).

Key finding: Distributed lag model estimates how sentiment impacts returns over multiple days (lags 0..5).

distributed lagspanelfixed effectsNewey-West

Horizon sweep: predictability across 1–10 day horizons

Panel FE regressions of cumulative forward returns on score_mean across multiple horizons.

Key finding: Horizon profiles show whether any predictability is short-lived or persists over longer windows.

horizonpanelfixed effectscumulative returns

Sentiment vs next-day returns

Predictive check: y_ret(t+1) ~ score_mean(t) (+ n_total(t)) in TS + panel FE.

Key finding: Sentiment vs next-day returns: score_mean is positive in TS (β=0.00703729, t=1.277693110495327, R²=0.005332314025009222).

predictivepanelHACfixed effectsreturnssentiment

β(score_mean) TS

β(score_mean) FE

Robustness

2 studies

Empirical research notes built from the live Sentiment Live dataset.

Section takeaways

Placebo test: shuffling sentiment should remove any true predictive relationship.
Quantile slopes can reveal whether the signal is stronger in downside tails or upside tails.

Placebo test: shuffle sentiment within ticker

Shuffle score_mean within ticker (destroys time ordering) and re-run predictive panel FE.

Key finding: Placebo test: shuffling sentiment should remove any true predictive relationship.

placeborobustnesspanelfixed effects

Quantile regression (sample ticker): sentiment vs next-day returns

Quantile regression on the sample ticker to see whether sentiment loads differently in tails vs median.

Key finding: Quantile slopes can reveal whether the signal is stronger in downside tails or upside tails.

quantile regressiontailsrobustness

Volatility & attention

1 studies

Empirical research notes built from the live Sentiment Live dataset.

Section takeaways

Sentiment vs volatility proxy (abs returns): score_mean is negative in TS (β=-0.0170556***, t=-2.9597401096231226, R²=0.04357847269681925).

Sentiment vs volatility proxy (abs returns)

abs_ret(t) ~ score_mean(t).

Key finding: Sentiment vs volatility proxy (abs returns): score_mean is negative in TS (β=-0.0170556***, t=-2.9597401096231226, R²=0.04357847269681925).

volatilitypanelfixed effects

β(score_mean) TS

β(score_mean) FE