This paper develops statistics for testing the null specification of an economic agent’s intertemporal optimum problem, that incorporate information about a specific non-nested alternative specification. The disturbances may be serially correlated and conditionally heteroskedastic. An optimal test in developed for a local alternative that reflects the specification of the alternative optimum problem through a term representing the ‘best’ approximation of the null model by the non-nested alternative model. Then, the procedures are implemented to study monetary and non-monetary models of consumption and portfolio decisions.