A novel approach for oil price forecasting based on data fluctuation network

Characterizing nonlinear time series using complex network science is a new multidisciplinary methodology. This paper puts forward a new time series prediction method based on data fluctuation network, named data fluctuation networks predictive model (DFNPM). The basic idea of the method is: first map time series into data fluctuation network and extract the fluctuation features of time series according to the topological structure of the networks, and then construct models with useful information extracted to predict time series. With Cushing, OK Crude Oil Future Contract 1 (Dollars per Barrel) and New York Harbor Regular Gasoline Future Contract 1 (Dollars per Gallon) as its sample data as well as DFNPM as its prediction model, the research makes a prediction on crude oil and gasoline futures prices from December 30, 2014 to February 26, 2015. A comparison is conducted between the result of the prediction and such traditional prediction models as grey prediction (GM) model, exponential smoothing model (ESM), autoregressive integrated moving average (ARIMA) model and radial basis function neural network (RBF) model, which shows that DFNPM performs significantly better than the above four traditional prediction models in both the direction and level of prediction.