We develop and explore a new model of the economics of privacy. Previous work has focused on "privacy of type," wherein an agent privately knows an immutable characteristic. We consider "privacy of action," wherein privacy means that an agent's choice of action is unobservable to others. To show how a policy of privacy can be socially optimal, we assume that an agent derives utility from an action he takes, from the aggregate of all agents' actions, and from other agents' perceptions of the agent's type (that are based on his action). If his action is observable, then he distorts it (relative to his full-information optimal action) so as to enhance the perceptions that others have of him. This contributes to aggregate welfare through increasing the public good, but the disutility associated with the distortion of agents' actions is also a social cost. If his action is unobservable, then he can take his full-information optimal action and still be "pooled" with other types. When the disutility of distortion is high relative to the marginal utility of the public good, a policy of privacy is optimal. We also consider a policy of waivable privacy, and find that equilibria exist in which some, but not all, types waive privacy. More significantly, if policies of privacy or publicity are costlessly enforceable, then a policy of waivable privacy is never socially preferred. Finally, we consider a number of examples (some of which involve a public bad and/or social disapproval): open-source software development; charitable giving; recycling; consumption of health services; DNA dragnets; student rankings; constraints on information disclosure at trial; electricity and water usage during periods of voluntary rationing; shaming of speeders; and the use of earmarks by Congress.
