Research On Prediction Of Chaotic Time Series Gold Futures Price

Economic time series is influenced by various factors.Thus,make it chaotic,that is a nonlinear system.The China's gold futures(AU.SHF)price is one of them.Base on this property,we provide two different methods of prediction.The first one fits the reconstructed phase space data into non-linear models like neural network to make prediction.The other approach is based on error compensation and it uses a linear and nonlinear hybrid models.More specifically,it starts with building an ARIMA(autoregressive integrated moving average)model and fits the data into it.The chaos will be represented by the estimation error of the model.Then we'll apply the first method to analyze the estimation error in order to compensate it.The temporal neighboring data has both linear and nonlinear components.It also has correlation between the data.So,we introduce the ARIMA-LSTM hybrid model that combines ARIMA and LSTM(Long and Short Memory Neural Network).Then we design three sets of experiments for each method.One for ARIMA-LSTM.The object of the study is the AU.SHF daily closing price,after conducting the experiments,we came to the conclusion that:(1)The China's gold futures price is chaotic and non-linear.(2)Both these two methods have better performance compared to traditional ARIMA model;(3)The second approach which applies the hybrid model based on error compensation and phase space reconstruction achieves significantly better results,especially the ARIMALSTM.