Informational externalities with lump-sum sampling

Abstract .  We consider a game of strategic experimentation where agents are restricted to an all-or-nothing sampling strategy. The strategic interaction between agents due to informational externalities is affected by the sizes of the experimentation samples and the sensitivity of information to changes in sample sizes. There is experimentation only if the overall sample is large enough. Equilibrium may involve optimal, insufficient or excessive experimentation relative to a second-best welfare benchmark. This unusual over-experimentation result is associated not necessarily with large samples but with a low elasticity of the value of information with respect to the sample size.