I propose a new method of identifying firms' subjective beliefs in dynamic models. In settings where actions are continuous and substitutable in contributing to an endogenous state, I show how beliefs about exogenous state transitions can be separately identified from static payoffs. Applying this method to identifying beliefs about future SO2 allowance prices in the U.S. Acid Rain Program, I find that electric utilities underestimate the movements in the allowance price, leading to compliance strategies that respond too slowly to changing market conditions. Biased beliefs increase an average utility's compliance cost by equivalently 11.2% of its profits.