1 0pt0.4pt Persuasion through Selective Disclosure: Implications for Marketing, Campaigning, and Privacy Regulation Florian Hoffmann, Frankfurt Roman Inderst, Frankfurt & Imperial College London Marco Ottaviani, Bocconi
2 Motivation Technological progress in collecting and processing personally identifiable data > Firms, political campaigners, etc. (=senders) can better tailor their communication to the preferences and orientations of individual consumers, voters, etc. (=receivers). What are the positive and normative implications? Identified determinants: Implications depend on whether there is (intense) competition among senders; receivers make individual or collective decisions (voting); firms are able to personalize prices; receivers are wary of the senders incentives to become better informed.
3 Sketch of the Model 1. Senders decide whether to acquire better information about receiver preferences. (Absent regulation this is not observed by receivers.) 2. Senders and receivers play a disclosure game. Notably, we consider two "classic" specifications of selective disclosure: Selective Non-Disclosure: Sender can increase probability of becoming informed about receiver preferences and then, if informed, decide whether or not to disclose (Dye 1985, Jun/Kwon 1998, Shavell 1994). Selective Targeted Disclosure: Sender can decide whether to observe preferences and then which of two attributes to disclose (continuous version of Fishman/Hagerty 1990).
4 Sketch of the Model 1. Senders decide whether to acquire better information about receiver preferences. (Absent regulation this is not observed by receivers.) 2. Senders and receivers play a disclosure game. Notably, we consider two "classic" specifications of selective disclosure: Selective Non-Disclosure: Sender can increase probability of becoming informed about receiver preferences and then, if informed, decide whether or not disclose (Dye 1985, Jun/Kwon 1998, Shavell 1994). Selective Targeted Disclosure: Sender can decide whether to observe preferences and then which of two attributes to disclose (continuous version of Fishman/Hagerty 1990). Note: No full unravelling (as in Grossman 1981, Milgrom 1981), as seller may be uninformed; or there is limited scope for full disclosure (due to costly/limited airtime, screen space, or limited attention).
5 Non-Disclosure: Setting Single receiver must make decision between M 2 alternatives: Respective utility u m is independently distributed according to F m (u m ) with expectation E[u m ]. Sender m learns about u m with probability θ m and then decides whether or not to disclose the respective information to the receiver. Sender gains when own alternative is chosen.
6 Non-Disclosure: Wary Receiver Definition for now: Knows likelihood θ = θ m with which sender learns preferences (u). Equilibrium "threshold" disclosure strategy: No information when sender has not learnt anything, but also no information when sender has learnt but u < u d.
7 Non-Disclosure: Wary Receiver Definition for now: Knows likelihood θ = θ m with which sender learns preferences (u). Equilibrium "threshold" disclosure strategy: No information when sender has not learnt anything, but also no information when sender has learnt but u < u d. Wary receiver s (updated) expected utility U: If information is disclosed: U = u If not disclosed: U = (1 θ)e[u] + θe[u u u d ]. Ex-ante distribution of U: G(U) > How does this change in likelihood that sender is informed?
8 Non-Disclosure: Naive Receiver Misperceives likelihood θ with which sender learns his preferences: > Misperceives likelihood with which non-disclosure is strategic! Perceived utility Û: Ex-ante distribution G(Û). > How does this change in likelihood that sender is informed?
9 Targeted Disclosure (1) Product with two attributes, i = 1, 2, which can be disclosed: > Unravelling avoided as disclosure of both attributes not feasible/too costly. Selective = Choice of disclosed attribute, based on knowledge of receiver preferences.
10 Targeted Disclosure (2) Short-cut formalization of non-targeted disclosure Disclose same attribute (e.g. i = 1) independently of receiver. Perceived utility when respective u 1 disclosed: U = u 1 + E[u 2 ] (additive). Ex-ante distribution of U: G(U) = F 1 ( U E[u 2 ] ). Possible alternative foundation: "2 Salop circles" [Location of product attributes & receiver preference > Distances]
11 Targeted Disclosure (3) With targeted disclosure, disclose "best fit": max i=1,2 u i. Wary receiver: Perceived utility: U = u i + E[u j u j u i ]. Unwary/Naive receiver: Û = ui + E[u j ]. Again ask: How does ex-ante distribution of U better informed? (Û) change as sender becomes
12 General Properties of (More) Selective Disclosure Receiver is aware: G(U) changes in the sense of a Single-Crossing, Mean- Preserving Spread. G(Û) changes in the sense of First-Order Sto- Receiver remains unaware: chastic Dominance.
13 Targeted Disclosure: Uniform Distribution
14 Selective Non-Disclosure: Uniform Distribution
15 Analysis: Two Clear-Cut Cases (1) 1. Naive/unwary consumers, personalized pricing and monopoly: Firm always prefers to become better informed [and consumer naively does not anticipate this] and exploits inflated valuation of consumer. (Consumer) Welfare higher when practice prohibited.
16 Analysis: Two Clear-Cut Cases (2) 2. Wary consumers/receivers and no personalized pricing: Receivers always benefit when senders better informed. [Intuition: SCMPS and binary yes/no decision over each alternative.] Thus, prohibition reduces (consumer) welfare.
17 Analysis: Two Clear-Cut Cases (2) 2. Wary consumers/receivers and no personalized pricing: Receivers always benefit when senders better informed. [Intuition: SCMPS and binary yes/no decision over each alternative.] Thus, prohibition reduces (consumer) welfare. Interesting twist: Even light-touch regulation (asking for receiver consent) can reduce (consumer) welfare. If sender cannot commit not to become better informed, he will always become better informed and disclose more selectively ["off equilibrium" FOSD shift]. But a sender who could commit would become better informed only if competition is weak ["mass vs. niche market" intuition].
18 Role of Personalized Pricing (selected results) Firms always choose to become better informed and disclose more selectively. [Additional benefit: Increased (horizontal) differentiation.] Even wary consumers can be hurt. Clear-cut comparison: With M = 2, consumer realizes > no personalized pricing: max[u 1, U 2 ] > with personalized pricing: min [U 1, U 2 ]
19 Role of Competition (selected results) Wary consumers/receivers: Always better off, even with personalized pricing, when competition is suffi ciently intense (high M). Naive consumers/receivers: Symmetric competition "debiases" decision making. Then same decision made as wary receivers, and with personalized pricing even better off [less perceived differentiation of firms].
20 Short Summary Framework for studying the implications when firms/campaigners collect more personal information for personalized communication (here: disclosure). [Applications: Targeted disclosure and selective non-disclosure.]
21 Short Summary of Presented Results Framework for studying the implications when firms/campaigners collect more personal information for personalized communication (here: disclosure). [Applications: Targeted disclosure and selective non-disclosure.] Some insights for policy and regulation: When receivers are aware of this, a seemingly more distorted communication is actually more informative. > Even light-tough regulation may backfire. When receivers are/remain unwary of this, there is scope for exploitation, notably with personalized pricing. But competition acts as a remedy - through various channels! Ask: 1) Awareness? 2) Competition? 3) Scope for personalized pricing?
22 Collective Decision Making: Changes? Considered application: Voting (with iid preferences). Key change: Expected utility depends also on likelihood of being pivotal. > How does this change when campaigners use personal information for communication to other voters? Side aspect: Possible negative externality from presence of unwary voters.
23 Persuasion through Selective Disclosure: Implications for Marketing, Campaigning, and Privacy Regulation Florian Hoffmann, Frankfurt Roman Inderst, Frankfurt & Imperial College London Marco Ottaviani, Bocconi