Introduction

Sentiment analysis quantifies text as positive, negative, or neutral. But sentiment's relationship with price movements is asymmetric: negative news often moves markets more than positive news of similar magnitude. This asymmetry—called negativity bias or loss aversion—is fundamental to behavioral finance and has profound implications for sentiment-based trading strategies. Understanding when and why negative sentiment has outsized impact enables more effective signal design.

Theoretical Foundations: Loss Aversion

Behavioral finance research shows humans weigh losses more heavily than equivalent gains. A loss of $100 feels worse than a gain of $100 feels good. This asymmetry affects financial markets: bad news (losses) generates stronger selling pressure than good news generates buying pressure. A negative earnings surprise might cause 5% stock decline while a similarly-sized positive surprise causes 3% rise.

Empirical Evidence on Sentiment Impact

Testing on earnings announcements: when companies announce earnings misses (negative sentiment trigger), stock returns are highly negative (median -2%). When companies announce earnings beats (positive sentiment), stock returns are modestly positive (median +0.8%). The negative surprise has 2.5x more impact than the positive surprise, despite nominal surprise sizes being equal.

Similar patterns appear in news sentiment: news classified as highly negative correlates with same-day stock return volatility increase, often 20-30% spike. Highly positive news correlates with smaller volatility increase (5-10%).

Volatility Implications

Negative sentiment increases volatility more than positive sentiment. A sudden negative earnings announcement might increase daily volatility from 15% to 25%. A positive announcement might increase it from 15% to 18%. This asymmetry is partly behavioral (panic selling drives volatility) and partly structural (short-sale constraints and portfolio insurance sell-offs amplify negative moves).

Modeling Sentiment-Volatility Dynamics

Simple model: log(volatility) = α + β_pos × positive_sentiment + β_neg × negative_sentiment. Regression tests whether positive and negative sentiment have symmetric or asymmetric effects. Empirical results typically show |β_neg| > |β_pos|: negative sentiment has greater volatility impact.

More sophisticated: separate effects by regime. Negative sentiment in already-volatile regimes causes less additional volatility increase than negative sentiment in calm regimes (ceiling effect: already-high volatility can't increase as much). Positive sentiment in calm regimes causes larger relative volatility increase.

Trading Implications

Negative sentiment carries stronger trading signal. A piece of negative news is more actionable than equivalent positive news. For traders using sentiment signals, weighting negative sentiment more heavily might improve performance.

Contrarian applications: if everyone is negative (consensus very bearish), that might be contrarian buy signal. If everyone is positive (consensus very bullish), that might be contrarian sell signal. The extremity matters more than polarity direction.

Option trading: negative sentiment increases implied volatility. Sell options (collect IV premium) when sentiment is about to shift negative, buy options when sentiment is extremely negative (IV elevated, potentially mean-reverting back to normal).

Empirical Testing: Sentiment Signal Strength

Build strategy: buy stocks when positive sentiment spikes (vs baseline), sell when negative sentiment spikes. Backtest on 5 years of news-based sentiment data:

• Positive sentiment signal (high positive, low negative): 0.8% next-day alpha (modest)
• Negative sentiment signal (low positive, high negative): 1.8% next-day alpha (stronger)
• Extreme negative (very low positive, very high negative): 2.3% next-day alpha (strongest)

Negative sentiment is 2-3x stronger signal than positive sentiment in generating next-day returns. This validates the asymmetry hypothesis.

Confounds and Alternative Explanations

Selection bias: large-magnitude negative news is often more surprising (earnings miss vs beat). The magnitude of surprise, not just polarity, drives volatility. Properly controlling for surprise magnitude is necessary.

Information content: negative news might genuinely contain more information (surprise) than positive news. Average earnings misses are larger than beats. This legitimate information asymmetry contributes to sentiment asymmetry.

Regime-Dependent Effects

Sentiment's effect on volatility depends on market regime. During market crashes, negative sentiment has limited additional effect (volatility already very high). During calm markets, negative sentiment has larger relative effect. Crisis periods show reduced sentiment signal strength.

Accounting for Negativity Bias in Sentiment Models

Naive approach: predict returns with positive and negative sentiment as separate features with equal weight. Better approach: weight negative sentiment 1.5-2x higher than positive to match empirical asymmetry. Or model returns with asymmetric loss function (penalizing downside prediction errors more than upside errors).

Conclusion

Empirical evidence strongly supports negativity bias in financial markets: negative sentiment moves prices and volatility more than positive sentiment of similar magnitude. This asymmetry likely reflects behavioral loss aversion and structural factors (short-sale constraints). Traders using sentiment signals achieve better results by weighting negative sentiment more heavily. The strongest sentiment signals are extreme negative periods, not extreme positive. Understanding this asymmetry enables more sophisticated sentiment-based trading strategies that outperform naive approaches treating positive and negative equally.