This dissertation studies public policy in coordination environments, where there is complementarity in the agents' actions. The first chapter studies a model of currency attacks in which the government can choose a credible signal about the fundamentals of the economy. Public signals create partial common knowledge that can lead to multiple equilibria. The optimal policy with commitment is characterized when, if there is multiplicity, the government only cares about its lowest equilibrium payoff. In this case, the public signal is informative and leads to a unique equilibrium, which is preferred to a full disclosure policy. Our results indicate that the government has incentives for being vague in its communication. The highest equilibrium payoff for the government can be achieved with a two-signal policy. In equilibrium, agents follow the public signal and take the same action: either there is a coordinated attack, or all speculators refrain from attacking. The second paper develops a model where short-term reputation concerns guide the public disclosure of information. There are and high and low states that determine the productivity of investment, and the high state is more likely if the government is efficient rather than inefficient. Governments know the state and make public reports with the objective to be perceived as efficient. I find that the inefficient government is never completely truthful in equilibrium. When the efficient government is truthful, the inefficient government sends false reports of a high state with positive probability. This creates uncertainty following the report of a high state: if the true state is high, productivity is underestimated; if the true state is low, productivity is overestimated. This bias reduces welfare in the high state, but there is a tradeoff in the low state: marginal entrepreneurs lose from overestimating productivity; all entrepreneurs gain from a higher aggregate investment. I show that when the trust in the government's report is low, the inefficient government can improve welfare in the low state by sending false reports that increase investment. However, as the trust in the false reports rises, the bias in entrepreneurs' beliefs becomes large and welfare decreases.