Application Research Of Deep Learning In Gold Futures Price Forecasting

As a trading product in the precious metals trading market,gold futures is an indispensable investment and financial management tool for countries,institutions and individual investors.It is of great practical significance to forecast its prices.This paper analyzes the reasons why gold futures are difficult to predict.There are too many factors affecting the change of gold futures.How many factors play an unknown role at each moment,the influence of people's emotions on the market is uncertain,and the synergy of all factors is changing.It is very difficult to accurately predict such a nonlinear unsteady time series affected by many factors.The price of gold futures is affected by many factors such as oil,dollars,stocks,economic policies and emergencies,and they are added to the forecasting model as much as possible for the accuracy of the forecast.For investors,many technical indicators are also the basis for common trading decisions.Therefore,this paper also selects representative indicators(MACD,KDJ,RSI,MFI)to be added to the forecasting model.The recurrent neural network with memory ability is in line with the actual situation that the gold futures price will be affected by the past environment,so it is very suitable for the forecasting of gold futures prices.Thanks to the rapid development of neural networks,the variants of Circulating Neural Networks LSTM and GRU solve the problem that cyclic neural networks cannot solve long-term dependencies.This paper attempts to use them to build models to predict the price of COMEX gold futures.This paper first uses shallow BP neural network,LSTM neural network,GRU neural network and ARIMA to predict the price of gold futures based on gold price related data(opening price,closing price,trading volume,price increase,maximum price,lowest price).After that,the GRU-based deep neural network is designed.Through sufficient training and tuning,the gold price related data of the first 10 days and the technical indicators calculated according to this,plus the important influencing factors of the gold price as the data characteristics during training,The 11-day gold futures closing price was forecasted,and the final model's forecast accuracy was 61%.Compared with the shallow neural network that does not incorporate many influencing factors into the data characteristics,the final model obtains better prediction results and has certain innovation.It also proves the validity of the deep circulation neural network model in financial forecasting and prediction.The specific fluctuations can provide reference for investors and have certain practical value.