Abstract:The fluctuation of crude oil prices has an impact on the global economic and political security. Accurate predic ̄ tion of the future crude oil prices has been all parties′ concern. This paper proposes a multi-scale principal component a ̄ nalysis (MSPCA) -based ARIMA method for crude oil price forecastingꎬ which uses two-dimensional crude oil price se ̄ ries as data sourceꎬ uses the ARIMA for forecasting after MSPCA and fully considers the correlation between crude oil fu ̄ tures price and spot price. This method uses the multi-scale analysis ability of wavelet transformꎬ the dimension reduction statistical ability of PCAꎬ and the prediction ability of ARIMA model for non-stationary time series. The experiment proves that the ARIMA method based on MSPCA is better than ARIMA method and Holt′s exponential smoothing methodꎬ and thus can effectively improve the accuracy of forecasting crude oil prices.