Extreme value theory has been applied to many areas of economics where the data is heavy tailed, e.g. in the analysis of market structure and risk forecasting. Accurate inference has, however, been hindered by the lack of consistent procedures for determining the start of the tail. A double subsample bootstrap procedure is proposed to solve this problem. The accuracy of the procedure is accessed with Monte Carlo experiments. Subsequently it is applied to Gibrat's and Zipf's laws, as well as the estimation of financial risk.