The　volatility　of　exchange　rate　is　very　important　to　a　country＇s　trading.　Accurately　forecasting　exchange　rate　time　series　appears　to　be　a　challenging　task　for　the　scientific　community　on　account　of　its　nonstationary　and　nonlinear　structural　nature.　In　order　to　improve　the　performance　of　exchange　rate　forecasting,　this　study　develops　two　evolutionary　support　vector　regression　models　to　forecast　four　typical　RMB　exchange　rates　(CNY　against　USD,　EUR,　JPY　and　GBP),　and　employs　four　evaluation　criteria　to　assess　the　performance　of　outof-sample　exchange　rate　forecasting.　In　this　study,　the　evolutionary　algorithm　optimizes　the　SVR　parameters　by　balancing　search　between　the　global　and　local　optima.　However,　the　inputs　of　models　are　selected　though　phase　space　reconstruction　method　from　historical　data　of　exchange　rate　series.　The　empirical　results　demonstrate　that　our　proposed　evolutionary　support　vector　regression　outperforms　all　other　benchmark　models　in　terms　of　level　forecasting　accuracy,　directional　forecasting　accuracy　and　statistical　accuracy.　As　it　turns　out,　our　proposed　evolutionary　support　vector　regression　is　a　promising　approach　for　RMB　exchange　rate　forecasting.　(C)　2019　Elsevier　B.V.　All　rights　reserved.