The empirical method is one of the best-performing demand forecasting methods in inventory decision making for products with intermittent demands. However, this method has di_culties obtaining high _ll rates since the used corresponding empirical distribution cannot handle extreme values that has not been observed. To improve the empirical method on this issue, this thesis _rst briey intro- duces the extreme value theory (EVT) which provides a solid theoretical basis and framework for tail estimation and extrapolation, and then proposes three extrapolation methods. Veri_ed by simulation studies, the extrapolation method applying EVT works the best among the proposed methods and successfully ex- trapolates the empirical distribution in the tail area