Research On Gold Price Trend Based On Improved PEEMD-BPNN Model

With the internationalization of the Renminbi,the gold reserve,almost equivalent to a country's international credit,is particularly important for the country's monetary policy;and then gold,as an important precious metal species in the global trading market,it is also an important way for companies to participate in global financial and personal investment transactions.Therefore,it is of great significance to analyze the international gold price trend.This paper utilizes a hybrid model to study the international gold price trend,so that the investors can make some reference in the investment decision-making process.In order to more rationally establish a hybrid model for gold price trend,this paper makes an exploratory analysis of the data characteristics of gold price at first,it tests its memory via moving statisticV-and analyzes its stationarity via the autocorrelation coefficient.It denotes that the gold price data is nonlinear and non-stationary.Based on these data characteristics,the partly ensemble empirical mode decomposition-BP neural network(PEEMD-BPNN)hybrid model is selected.Then PEMMD is used to decompose the glob gold price data to obtain the orthogonal the intrinsic mode functions(IMFs),then the T test and the bilateral permutation entropy(BPE)method are utilized to reconstruct the IMFs for the high-frequency data,low-frequency data and the trend data according to the frequency domain.The correlation analysis between different reconstruction data and its influencing factors,the reconstructed highfrequency data and global gold mine output,low-frequency data and world crude oil price,low-frequency data and world inflation,are conducted by calculating the correlation coefficient in comparison.Finally,the reconstructed data is treated as an input to the hybrid model for trend analysis,and the performance of the combined model is evaluated.As the perspective of memory analysis in gold price data,the reconstructed data analysis and the prediction via the hybrid model suggest that the accuracy of the PEMMD-BPNN model is higher overall,especially.The composed data reconstructed by bilateral permutation entropy shows better performance in terms of correlation analysis with relevant influencing factors and prediction effects of the hybrid predictive model as an input.