We experimentally test an endogenous-timing investment model in which subjects privately observe their cost of investing and a signal correlated with the common investment return. Subjects overinvest, relative to Nash. We separately consider whether subjects draw inferences, in hindsight, and use foresight to delay profitable investment and learn from market activity. In contrast to Nash, cursed equilibrium, and level-k predictions, behavior hardly changes across our experimental treatments. Maximum likelihood estimates are inconsistent with belief-based theories. We offer an explanation in terms of boundedly rational rules of thumb, based on insights about the game, which provides a better fit than quantal response equilibrium. (JEL C72, D82, D83, G11)