Forecasting　the　stock　market　price　movements　is　now　popular　in　the　field　of　financial　research.　A　large　number　of　scholars　has　carried　on　the　positive　exploration.　Only　these　people　are　more　focused　on　selection　of　prediction　methods　and　algorithm　optimization.　In　view　of　the　stock　market　time　series　has　the　nature　of　the　multi-scale　features,　nonstationary　and　nonlinear　properties　and　low　signal-to-noise　ratio　of　some　different　from　other　general　characteristics　of　time　series,　this　paper　puts　forward　building　a　multi-scale　technique　index　method　for　preprocessing　of　the　input　data　and　then　used　very　popular　in　recent　years　the　output　of　the　neural　network　technology　to　the　pre-processed　data　to　make　predictions.