Balanced Social Learning

When is social learning efficient if agents have heterogeneous preferences over a set of actions? When myopic agents generate information only about the action they choose, efficiency requires balance such that all high-value actions are discovered. We characterize the value of policies that control the flow of past information to future agents. Under full transparency of past information, the probability of efficient outcomes is approximately one if and only if the prior and preferences are balanced and the log-likelihood ratio distribution satisfies a set of intuitive moment conditions. Noisy bad news is a canonical example. By contrast, approximate efficiency is obtainable on the set of arbitrary adaptive information policies if and only if the prior alone is balanced. Moreover, it suffices to use adaptive evidence policies which permanently and fully disclose subsets of past information. When the prior is imbalanced and approximate efficiency is unattainable, we characterize the value of adaptive information and show that it is uniformly approximated by the value of adaptive evidence as the impact of each individual signal on beliefs vanishes.