Thread by @shoshievass

@RevEconStudies @SNageebAli @PennStateEcon @econ_greg @MSFTResearch @StanfordGSB @nberpubs A few years ago, I wrote a thread introducing a paper on "Voluntary Disclosure & Personalized Pricing" with @SNageebAli and @econ_greg (threadreaderapp.com/thread/1186778…).
Now that the paper has been accepted for publication, an update on what the paper is about 👇 1/n

The gist: In the debate about privacy, info disclosure & price discrimination, it's important to think about the structure and verifiability of the disclosure technologies. 2/n

Classical intuitions from Grossman (81) and Milgrom (81) suggest that voluntary disclosure is ultimately self-defeating and unhelpful to consumers. These intuitions are often invoked in arguments for strong privacy protections meant to keep consumer data out of firms' reach 3/n

More recent work has focused on disclosure as a vehicle for customization (e.g. Hidir and Vellodi ('21); Ichihashi ('21)) or disclosure through an all-powerful intermediary who can act on behalf of the consumer (e.g. Bergemann, Brooks & Morris ('15)) 4/n

We focus on a simple model of disclosure in which i) consumers can send convex messages about their types to receive an offer that they can accept or reject; ii) Messages are verifiable by a 3rd party (eg type 1/3 can claim to be in the interval [0,.5] but not [.5, 1]) 5/n

We show that the structure of the message technology makes a big difference in what can result. 6/n

In a world w/ only "simple" messages (e.g. "tell me everything or tell me nothing") voluntary disclosure cannot help consumers in monopolistic markets; the classical intuition holds. 7/n

By contrast, partial disclosure through "rich" messages (e.g. "nothing but the truth, but not necessarily the whole truth") allows for equilibria in which nearly all consumer types strictly benefit from disclosing information. 8/n

This idea generalizes: it suffices for there to be a way to generate "group pricing" (e.g. info that separates low value types from high types in equilibrium) in order for voluntary disclosure to produce Pareto improvements. 9/n

These results extend to natural relaxations of our basic model, such as when firms have external info about their consumers prior to interacting. If instead of a monopolist, there are multiple differentiated firms, even simple messages are beneficial to consumers. 10/n

A natural question is where this all fits in discussions of real life policy. For this, it's useful to revisit the assumptions in our model. 11/n

There are two key features in our messaging technologies: (i) messages are verifiable: there is a 3rd party who can ensure that consumers don't lie; (ii) consumers are never committed to sending a particular message or executing a given personalized offer. 12/n

First, note that there is verifiable disclosure all around us. Folks can submit gov't docs (EBT cards; W-2 receipts) to apply for discounts big and small. People regularly receive promotions based on their internet browsing record, or even their recent purchasing history. 13/n

In some cases, voluntary disclosure is offered as a direct promotional program by the seller, as in the auto-insurance monitoring program that Yizhou and I study. 14/n

Second, our characterization has guidance for regulatory design.

Consumer control over info, rather than rigid privacy, is what drives Pareto gains. But to implement gains, a "verifier" the gov't, a platform or a firm) can choose the coarseness of the message space. We argue that each potential verifier has interests in consumer gains 15/n

There's a lot more in the paper of course. I hope you read it and let us know what you think. This project started when I interned at @MSFTResearch as a PhD student. It's been a pleasure to work on. Thank you to all the amazing folks who've discussed it with us over the years n/n

@MSFTResearch @threadreaderapp unroll