On The Gold Price Modeling By Deep Learning And Its Applications

With so many black swan events occurring in 2016,the gold has became global investors' preferred haven.So it's feasible and valuable to make a quantitative analysis and prediction of gold price movements with deep learning models.This paper has built a deep learning model with a research object of U.S.Comex gold futures from 2010 to 2016.After theoretical analysis and empirical test,we have found that gold price is a typical nonstationary time series which is extremely sensitive to external financial market.So we first select Brent crude futures,Goldman sachs commodity index,U.S.dollar index and actual interest rate of 10-year Treasury note as 4 input characteristics of the model.After that,we have chosen a double-layers Elman neural network with delay operators to build our gold deep learning models considering that gold price has a nonlinear dynamic characteristic.Then we used R program to make the empirical analysis of the gold future data and found that Elman neural network models has made a good job with strong generalization capability on different data sets.Further,we added a 5-order-lag gold price data set to build a "memorial" gold deep learning model.The empirical results showed that the prediction accuracy of new model improved.Finally,we tried making a quantitative trading strategy based on the forecasting tendency of gold price,giving the model more practical values.