Considering　the　fact　that　crude　oil　markets　have　various　noise,　fluctuations　and　actual　needs　of　investors　in　different　trading　cycles,　in　this　study,　we　propose　a　multiwavelet　denoising-integration　MF-DCCA　method　by　three-phase　modeling　to　construct　a　portfolio　among　crude　oil　markets.　On　the　basis　of　noise　filtering,　this　method　extracts　the　effective　prediction　information　from　different　fluctuations　and　integrates　them　into　the　same　time　scale.　The　noise　reduction　and　out-of-sample　portfolio　performance　are　evaluated,　respectively.　The　robustness　of　out-of-sample　portfolio　performance　is　further　tested　by　changing　partial　conditions　The　empirical　results　indicate　that　the　denoising　performance　of　multiwavelet　denoising　method　is　clearly　better　than　the　traditional　wavelet　denoising　methods.　Furthermore,　in　any　sample　period,　the　multiwavelet　denoising-integration　MF-DCCA　method　also　performs　better　than　the　existing　popular　methods　in　terms　of　profitability　and　Sharp　ratio.　Besides,　the　long-term　scales　(s=60,　80)　are　superior　to　the　short-term　scales　(s=20,　40)　in　most　situations.　Robustness　results　verify　the　above　conclusions.　©　2019　Elsevier　B.V.