Heterogeneous Interpretations and Collective Decisions under Model Persuasion (Job Market Paper) (Slides)
We study a model persuasion problem in a voting environment with heterogeneous default models and a common prior p ∈ (0, 1/2). A politician commits to a public menu of models before a common signal is realized. Voters select the best-fitting model from the menu and their own default, then vote according to its posterior. We characterize signal-specific support capacities and the weakly attainable joint support region. When p ≥ 1/4, one menu attains both capacities. Below 1/4, every nonempty menu leaves zero support after at least one signal, while silence preserves baseline support. We then characterize the politician’s optimal response and the planner’s optimal voting thresholds. When both capacities are below one, screening improves on rejection precisely when favorable evidence makes approval strictly preferable to rejection and its support capacity exceeds that of unfavorable evidence. Full capacity can prevent the planner from blocking passage. In a mixture-default application, mean-preserving dispersion can create approval opportunities or restore the planner’s ability to block passage, depending on the prior. These changes can improve welfare even under symmetric dispersion, which cannot reverse the capacity ranking.
Narrative Effect of Monetary Policy Announcements and Heterogeneous Investor Trading: Evidence from China's A-share Market (with Hengxu Song)
Monetary policy announcements shape investor expectations and financial markets, yet in the social media era, their transmission may have a “narrative effect,” particularly among retail investors. This paper develops a financial-market narrative model with heterogeneous investors. Theoretically, we show that retail and institutional investors may endogenously adopt different narratives to interpret the same public policy information, generating heterogeneous subjective beliefs and portfolio choices. In the model, following a supportive policy action, narrative-induced belief differences increase retail investors’ risky-asset holdings and reduce institutional investors’ holdings relative to a common-narrative benchmark through equilibrium price adjustment, thereby generating trading divergence. Using data from China’s A-share market, Loan Prime Rate (LPR) announcements, and the People’s Bank of China’s official WeChat account, our empirical analysis documents statistically significant trading divergence between retail and institutional investors. Across three complementary dimensions of narrative exposure, the timing and persistence of trading divergence vary with the surrounding information environment. This paper offers a first-moment, belief-based explanation for heterogeneous investor divergence that is distinct from the second-moment uncertainty mechanism emphasized in the existing literature and highlights the role of narrative interpretation in heterogeneous responses to monetary policy communication.
Interim Model Persuasion (new draft coming soon!) (Slides)
We study persuasion through the choice of interpretations rather than the design of information. Some evidence is publicly observed first, after which a persuader offers an explanation that the decision maker later uses to interpret additional, privately observed evidence. We analyze such interim persuasion problem in which, after a public signal is realized, a sender commits to a finite menu of probabilistic models (interpretations). After a subsequent private signal is realized, a boundedly rational receiver selects the model in the menu that best fits the realized signal pair, in the sense of maximizing joint likelihood, and then takes an action based on the resulting posterior. In a binary environment, we characterize the set of feasible posterior vectors across private signal realizations conditional on the public signal, and show that this feasible set expands relative to the ex-ante persuasion problem studied by Aina (2024). In particular, the sender can drive the receiver’s posterior on a target state to one, regardless of the private signal realization. We show that this extreme belief manipulation is robust in richer finite state and signal spaces. Finally, we discuss how a default interpretation and competition between opposing senders can discipline persuasion and prevent extreme manipulation.
Competition in Persuasion with Costly Verifiable Private Information (with Zhong Long) (new draft coming soon!)
We introduce a Bayesian persuasion model with two senders and one receiver into a costly verification setting. The receiver makes a decision on allocate one item to one sender, all players do not know about the true state but hold the same prior belief. Both senders can persuade The receiver by design an information structure; the receiver can pays a fixed cost to verifies one sender's true state. We apply concavification technology to characterize the optimal information structures when persuasion is valuable. We also provide a sufficient condition on the existence of symmetric equilibrium and show multiple equilibria exist for some values of the prior belief. Finally, we analysis the original information structure and intuitively explain why multiple equilibria exist in our model.
Home Bias and Narrative Effect (with Hengxu Song)