Consider a principal (say a central government) which allocates a fixed budget among multiple agents (say local governments). Each agent chooses a policy or technology. After observing the success or failure of each agent, the principal allocates a larger share of the budget to agents who succeeded than to those who failed. Under these conditions, a Nash equilibrium may have all agents herd, all choosing the same technology, even though the principal would prefer that they experiment, with different agents choosing different technologies. Relatedly, under some conditions all risk-averse agents may choose a technology whose outcome is risky, over a technology whose outcome is certain.