Center for Social Information Sciences (CSIS) Seminar
Abstract: We examine a stochastic growth process that can alternatively be interpreted as a model of economic growth, financial portfolio management, statistical inference, or biological population growth. For the economic interpretation, we find that the growth-maximizing policy satisfies a meritocracy principle: it minimizes the discrepancy between the resource shares allocated to the agents and the agents' "merits." For the statistical interpretation, the setting is equivalent to a model of predictive coding, in which a misspecified system maximizes the fit of data. A consistency principle analogous to the meritocracy principle requires the optimal fit to minimize a degree of Bayes inconsistency.
Joint work with Larry Samuelson.