Gold Futures Price Forecasting Based On Deep Learning Network

Gold has the function of avoiding risks and hedging.When the financial market or the international situation is turbulent,countries and investors often choose to buy gold.Therefore,it is of great practical significance to predict the price of gold.This thesis predicts the price of U.S.Comex gold futures from 2010 to 2019.By analyzing the historical price of gold,it is found that the price of gold is affected by external factors,and the price sequence has nonlinear characteristics.Due to the fast development of deep learning,and it performs well in processing nonlinear time series,so this thesis selects the deep learning model for prediction.Firstly,Using linear interpolation to fill the missing values of gold futures and 10 influencing factors,and selects silver futures,Goldman Sachs Commodity Index,U.S.dollar index,and the actual interest rate of 10-year Treasury note as the external factors of gold futures.After that,this thesis constructs a multi-factor BP neural network model and LSTM model that consider four influencing factors and the price of gold futures,and compares with the single factor BP neural network model and the LSTM model that only consider the gold futures price.Secondly,due to the prediction effect of single factor model is poor,this thesis attempts to improve it and proposes an ARIMA-LSTM hybrid model,which uses a MA filter to obtain the linear and non-linear components of gold futures,and gets the prediction values of the two components through ARIMA and LSTM,and combines the two predictions to get the final results.Finally,based on the results of one-day prediciton,this thesis further predicts the price of multi-day gold futures,and the models of multi-day prediction are obtained by improving the multi-factor LSTM model and the ARIMA-LSTM hybrid model in the one-day prediction.It can be seen from the empirical results that the model considering multiple influencing factors are better than the model that only considers the price of gold futures.The performance of LSTM model is also significantly better than that of BP neural network model.Compared with the best multi-factor LSTM model among the four single models,ARIMA-LSTM hybrid model has better results,and it has a certain improvement in prediction accuracy and direction prediction.