Many existing measures of vulnerability lack a theoretical basis. In this paper we propose to measure vulnerability rigorously as the welfare of a household which solves an intertemporal optimisation model under risk.In such models, in essence a stochastic version of the Ramsey model, an important part of chronic poverty may be caused by the ex ante response of households to risks. Our simulation results indicate that whether or not a household is to be classified as vulnerable depends strongly on the time horizon considered. We use the model to assess the accuracy of existing regression-based vulnerability measures. We find that these methods can be vastly improved by including asset measures in the regression.