Traditional poverty measures fail to indicate the degree of risk of becoming or remaining poor that households are confronted to. They can therefore be misleading in the context of implementing poverty reduction policies. In this paper I propose a method to estimate an index of ex ante vulnerability to poverty, defined as the probability of being poor in the (near) future given current observable characteristics, using panel data. This method relies on the estimation of the expected mean and variance of future consumption conditional on current consumption and observable characteristics. It generates a vulnerability index, or predicted probability of future poverty, which performs well in predicting future poverty, including out of sample. About 80% of households with a 2000 vulnerability index of 100% are actually poor in 2007. This approach provides information on the population groups that have a high probability of becoming or remaining poor in the future, whether currently poor or not. It is therefore useful to complement traditional poverty measures such as the poverty headcount, in particular for the design and planning of poverty reduction policies.