Recently　years,　the　instability　of　soybean　futures　prices　has　become　more　and　more　serious.　How　to　dynamically　predict　the　price　trend　of　soybean　futures　according　to　the　dynamic　correlation　characteristics　of　fluctuations　between　domestic　and　foreign　soybean　futures　markets　has　become　a　research　hotspot.　Taking　the　soybean　futures　of　China　and　the　United　States　as　the　research　object,　this　paper　puts　forward　an　EMD-CTA　prediction　model　based　on　EMD　and　Transformer　method　to　realize　the　prediction　of　multivariate　time　series.　In　the　process　of　model　construction,　the　EMD　method　based　on　the　improved　continuous　mean　square　error　criterion　is　used　to　decompose　and　de　noise　the　time　series.　On　this　basis,　the　periodic　characteristics　are　used　to　convolute　the　sequence　using　the　filters　with　the　size　of　the　global　time　step　and　whose　size　is　related　to　period　length　respectively,　and　then　the　long-term　dependence　of　the　sequences　and　correlation　among　them　is　captured　by　the　Transformer　component.　Finally,　it　is　integrated　with　the　linear　components　output　by　AR　model　to　get　the　returns　forecast　result　of　the　closing　price　of　the　next　trading　day.　The　experimental　results　show　that　the　prediction　results　of　MAE,　RMSE　indexes　are　better　than　other　baseline　models　and　their　variants.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.