Traditional structural models cannot distinguish whether changes in activity are a function of altered expectations today or lagged responses to past plans. Polynomial-adjustment-cost (PAC) models remove this ambiguity by explicitly separating observed dynamic behaviour into movements that have been induced by changes in expectations, and responses to expectations, that have been delayed because of adjustment costs. In these models, agents' decisions are a function of forecasts of a desired level for the decision variable and, owing to frictions, this level is reached only gradually. In this paper, the authors use PAC models to analyze and forecast U.S. household spending. They find that the estimated models are rather rich from a theoretical and dynamic view-point. For example, the authors find that household spending is a function of forward-looking expectations, short- and long-term interest rates, human and non-human wealth, liquidity constraints, and uncertainty with respect to future business cycles. Moreover, out-of-sample forecasts and stability tests show that this theoretical structure is not added at the expense of the model's empirical features.