Several investors face an irreversible investment opportunity whose value V is governed by Brownian motion with upward drift and random expiration. The �rst investor i to seize the opportunity before expiration receives the current V less a privately known cost Ci; the other investors receive nothing. We characterize Bayesian Nash Equilibrium (BNE) for this game, extending previously known results.

We also report a laboratory experiment with 72 subjects randomly matched into 600 tri- opolies. As predicted in BNE, subjects in triopolies invested at lower values than in monopolies, changes in Brownian parameters signi�cantly altered investment values in monopoly but not in triopoly; and the lowest cost investor in a triopoly usually preempted the others. Evidence was mixed on other BNE predictions, e.g., whether higher cost brings smaller markups. Overall, subjects' earnings came rather close to the BNE prediction.

An irreversible investment opportunity has value V governed by Brownian motion with upward drift and random expiration. Human subjects choose in continuous time when to invest. If she invests before expiration, the subject receives V−C : the final value V less a given avoidable cost C . The optimal policy is to invest when V first crosses a threshold V ∗ = (1 + w∗ )C , where the option premium w∗ is a specific function of the Brownian parameters representing drift, volatility and discount (or expiration hazard) rate.

We ran 80 periods each for 69 subjects. Subjects in the Low w∗ treatment on average invested at values quite close to optimum. Sub jects in the two Medium treatments and the High w∗ invested at values below optimum, but with the predicted ordering, and values approached the optimum by the last block of 20 periods. Behavior was most heterogeneous in the High treatment. Subjects underrespond to differences in both the volatility and expiration hazard parameters. A directional learning model suggests that sub jects react reliably to ex-post losses due to early investment, and react much more heterogeneously (and on average more strongly) to missed investment opportunities. Simulations show that this learning process converges on a nearly optimal steady state.