Research On Gold Futures Price Forecasting Based On Time Series

This article choose gold futures AU99.95 closing price as the research object,this paper integrated the research content of financial time series at home and abroad,first of all,through normal test was carried out on the yield,stationarity test,heteroscedasticity testing after that yield sequence is stationary time series,the typical "peak fat-tail" features,and exist since the correlation.Compared with GARCH family models with different residual difference cloth,the AR(1)-SGARCH model with the smallest BIC value was selected from AIC criterion and BIC criterion because the model had fewer coefficients.The coefficient estimation of the model was carried out and the significance of the coefficient was observed.The rate of return was re-converted into price data.5000 simulations were carried out and the price path of the simulation was analyzed.AU99.95 has a 75% chance of peaking again after 12 years.However,the GARCH model is not ideal for predicting specific price values,and the model residuals still have autocorrelation and volatility.With artificial neural network in the financial data of research more and more deeply,the artificial neural network is applied to the research in gold futures price prediction is becoming more and more widely,in this article,using the moving average data decomposition,the decomposed low volatility sequences using traditional linear linear time series modeling prediction,The Long short-term Memory(LSTM)network prediction based on high volatility sequence was constructed.By adding the predicted values of the two models,the evaluation indexes of the mixed model and the single model were compared and analyzed,and the prediction effect of the mixed model was able to be better than that of the single model.