Emotions, rhetoric, and policy preferences
Yann Algan, Eva Davoine, Thomas Renault, and Stefanie Stantcheva investigate, in this paper, whether and how emotions causally shape citizens’ policy views, filling a gap in economics where policy preferences are traditionally explained by material self-interest or ideological priors. The motivation for this research is the striking rise of anger in political discourse since 2016 across social media, campaign speeches, and Congressional rhetoric which standard rational-choice frameworks cannot account for. The authors proceed in two steps:
- First, they apply NLP and machine learning to six large datasets of political speech and social media (2013–2025), covering 3 million citizen tweets, 1.5 million Congressional tweets, 1.8 million floor speech turns, and presidential campaign interventions, to document trends in emotional content over time. Emotions are classified by fine-tuning a supervised classifier on GPT-4o-mini-labeled sentences, then applied to the full corpus via sentence embeddings .
- Second, two pre-registered online experiments are run on nationally representative U.S. samples. Survey A (n = 3,512, November 2024) randomly assigns respondents to positive, negative, or neutral video treatments and measures effects on views across trade, immigration, redistribution, and democracy. Survey B (n = 6,366, November–December 2024) focuses specifically on climate change, using carefully designed videos to separately induce anger (disaster footage with institutional blame) versus fear (identical disaster footage without blame), isolating the distinct causal effect of each emotion.
Findings
Observational: A structural rise in anger.
The share of policy-related tweets expressing anger among citizens doubled from around 25% to 50% between 2013 and 2025. This rise begins sharply after the 2016 election and does not revert. Crucially, panel regressions with user fixed effects confirm the increase is genuine. On the supply side, Congressional tweets expressing anger generate 50% more retweets than non-emotional tweets, providing a structural incentive for political actors to amplify anger.
Strategic asymmetry between supply and demand.
Policymaker anger follows political power cycles — higher when in opposition — and is more prevalent on social media than in floor speeches, suggesting strategic deployment . Citizen anger, by contrast, is persistent across all partisan affiliations and has not reverted, suggesting that once triggered, citizen anger is self-sustaining.
Experiment A: Negative emotions broadly shift policy views.
Exposure to negative emotions increases support for protectionism, restrictive immigration, and redistribution, but does not reinforce populist attitudes. The redistribution effect works through a specific mechanism: negatively treated respondents are more likely to attribute poverty to external circumstances rather than individual failure. Positive emotions reduce populist inclinations but have little effect on most specific policy preferences
Experiment B: Anger, not fear, drives climate policy support.
The anger treatment significantly increases belief in human-caused climate change, support for carbon taxes, subsidies for electric vehicles, redistributive climate measures, and willingness to take private pro-climate actions (e.g., reducing meat consumption). The fear treatment has no significant effect on any of these outcomes, a striking result given that fear is the emotion most commonly invoked in climate communication .
Emotions and Policy Views
Authors: Yann Algan, Eva Davoine, Thomas Renault, Stefanie Stantcheva
From: HEC, UC Berkeley, University Paris 1 Panthéon-Sorbonne, Harvard University
Draghi’s smile, Lagarde’s frown: How central bank emotions move markets
This paper by Mamadou-Lamine Barry, Brenton Joey Bruns, Sinem Kandemir, Jens Klose, Victor Smirnov, and Peter Tillmann investigates whether the full scope of ECB presidential communication not just the spoken word but also facial expressions and vocal tone carry independent information that moves financial markets, addressing a gap in the central bank communication literature which has so far focused almost exclusively on text.
The authors analyze 100 ECB Governing Council press conferences from May 2012 to July 2023, covering both Mario Draghi and Christine Lagarde, at minute-by-minute frequency. Three measurement tools are applied in parallel :
- Facial expressions: SHORE software (Fraunhofer Institute) detects discrete emotions (happy, sad, angry, surprised, frightened, disgusted) frame by frame, aggregated to the minute level. A facial recognition algorithm confirms which face belongs to the ECB president.
- Vocal emotions: A pre-trained transformer model classifies audio into the Valence-Arousal-Dominance (VAD) framework at the minute level, robust to background noise and gender-fair across Draghi and Lagarde.
- Text hawkishness: A machine learning model measures the degree of hawkishness sentence-by-sentence from transcripts.
These three data streams are aligned at the minute level and combined with high-frequency asset price data, namely 10-year government bond yields for Germany, France, Italy, and Spain, the Bund future, Eurostoxx 50, and the EUR/USD exchange rate. The empirical model regresses asset prices on hawkishness, facial expressions, vocal arousal, and crucially, their interactions, estimated separately for Draghi and Lagarde and for the introductory statement versus Q&A .
Findings
- Draghi “kills with kindness”: hawkish statements only raise bond yields when delivered with a happy facial expression. Without it, hawkishness is statistically insignificant. The amplification effect is twice as large for Italian yields compared to German .
- Lagarde’s anger amplifies hawkishness: for Lagarde, an angry expression during the introductory statement strengthens the impact of hawkish language on bond yields across all countries .
- Forward guidance and happiness: Draghi’s happy facial expressions amplify the yield-reducing effect of forward guidance statements; a sad face undermines it. No equivalent effect is found for Lagarde .
- Emotions track inflation: across the full sample, angry facial expressions increase with inflation levels, even after accounting for the structural break between the two presidencies .
- Introductory statement > Q&A: facial expressions matter more during the scripted introductory statement than the unscripted Q&A, suggesting that in a stricter format, any emotional signal is more informative precisely because it is harder to fake .
The Emotions of Monetary Policy
Authors: Mamadou-Lamine Barry, Brenton Joey Bruns, Sinem Kandemir, Jens Klose, Victor Smirnov, Peter Tillmann
From: University of Giessen, THM Technische Hochschule Mittelhessen
