Essay in Media Economics

Abstract

This thesis consists of three essays on how measurable signals of demand shape offline decisions across politics, news production, and financial markets. The unifying theme is gatekeeping in contemporary information systems, decision-makers increasingly rely on observable cues, some originating in digital discourse, others in audience composition and ratings, to decide who to amplify, what to cover, and when to act. Empirically, the chapters combine large-scale text and transcript data with quasi-experimental designs and simple theoretical frameworks, methodologically, they integrate modern NLP, classification and embeddings, with high-frequency outcomes to trace selection mechanisms across domains.

The first essay, Chapter 2, studies whether social-media virality serves as a booking signal for television producers. Using approximately 700,000 tweets from Italian MPs, a content-agnostic top-1 percent virality threshold, and a dynamic difference-in-differences design, it shows that viral politicians receive more next-day invitations and speaking time on TV. A simple model explains why, producers treat virality as a noisy proxy for audience interest, which in turn incentivises attention-maximising rhetoric online. The results reveal a cross-platform feedback loop whereby platform metrics reshape the supply of political speech by altering visibility incentives.

The second essay, Chapter 3, moves from ``who to invite'' to ``what to cover,'' examining editorial choices in U.S. local television news using the weather as an empirical laboratory where underlying facts are observable. By linking near-universe newscast transcripts to market-day temperature deviations, the chapter documents a clear ``man bites dog'' pattern, rarer events receive more coverage, and shows systematic heterogeneity with audience partisanship, markets emphasise deviations consistent with local ideological priors. Ownership switches to a large conservative broadcaster leave these gradients largely unchanged, pointing to audience-driven rather than owner-driven selection. A simple demand model with taste for surprise and belief-consistent content rationalises the findings.

The third essay, Chapter 4, turns to financial markets and timing, testing whether pre-announcement social-media discourse around central-bank meetings predicts announcement-day asset-price moves. Using supervised classification to isolate policy-relevant tweets and embedding-based measures of disagreement, the chapter shows, focusing on the ECB for institutional consistency, that higher pre-meeting dispersion of narratives foreshadows larger price reactions when policy news arrives. The evidence highlights an anticipation channel, markets partly react to the resolution of publicly visible disagreement, not only to the information in the announcement itself.

Together, the essays map complementary facets of gatekeeping, who is amplified, what is covered, and when reactions occur, and show that observable signals such as engagement metrics, audience composition, and narrative dispersion do not merely mirror demand, they feed back into editorial and policy choices.