Foto sezione
Logo Bocconi

Events of Department of Economics

October 8, 2019 at 12:45 - 14:00
Seminar Room 3.e4.sr03 - Via Roentgen 1

Seminar (Joint with the Department of Decision Sciences)

Economic Theory, Decision Theory and Experimental Economics

Persuasion with Correlation Neglect

Ronny Razin (London School of Economics and Political Science)


In this paper we investigate how coordination across information sources - when unknown to voters and consumers - can be used strategically to affects their beliefs. In particular we analyse a model of persuasion when the receiver has correlation neglect. We use an information design model in which a sender, who cares about the action of the receiver as well as about the state of the world, can design (and commit to) a joint information structure for m experiments or signals, as a function of the state of the world. A receiver who attempts to learn the state of the world will observe the realizations of these m signals, but she believes that the signals are all conditionally independent. Our first result is to provide a graphical characterisation for the set of distributions over posterior vectors that arise from information structures with m signals. The characterisation highlights a trade-off that we emphasise throughout the paper; while the sender may want to negatively correlate signals, it is always easier to positively correlate signals. In particular, with fully correlated signals any distribution over posteriors that satisfies Bayes Plausibility is possible to achieve, negative correlation always puts more constraints on the distributions that can be induced. Next, we use the characterization above to shed light on the optimal information structure for the sender. We illustrate how the optimal full correlation solution can be derived by a modification of the concavification method suggested in Kamenica and Gentzkow (2011). But full correlation is not always optimal: We characterise conditions under which the Sender would want to use negative correlation. Finally we provide limit results when the sender has many signals at her disposal. We show that in this case, the Sender can achieve her first best utility.