EMD-ARIMA Model And Its Application In The Prediction Of Commodity Price Index

This article uses Empirical Model Decomposition and time series analysis model to study the prediction of Yiwu commodity price index.First,we describe the development history and research status of time series analysis,Empirical Mode Decomposition and Yiwu commodity price index.Then,we introduce the basic theory of time series analysis,including the smoothness test of time series and pure random process,stationary time series model and nonstationary time series model.Next,we introduce the theoretical knowledge of EMD in detail.We explain three basic concepts which are instantaneous frequency,intrinsic modal function and characteristic time scale.We explain the basic idea of EMD,the algorithm flow of EMD,the stopping criteria for IMF screening process,and end of EMD.We expound four main characteristics of EMD,which are adaptive property,filtering property,completeness and orthogonally of IMF.According to these four characteristics,we design two kinds of modeling methods based on EMD-ARIMA model.Both of two methods are use EMD to decompose the original sequence and analyze the effective components.One method establishes the time series prediction model one by one for each effective component,and then obtains the final forecast result by reconstructing the effective component forecast results.It is referred to as "EMD-ARIMA-Reconstruction" modeling method.The other method will reconstruct the effective component to get a new sequence first,and build the prediction model for the new sequence to obtain the final prediction result.It is referred to as "EMD-Reconstruction-ARIMA" modeling method.In the empirical study,we first use the "EMD-ARIMA-Reconstruction" modeling method to predict the Yiwu Commodity Price Index series,and obtain the first group of forecast results.Then we use the "EMD-Reconstruction-ARIMA" modeling method to predict the same original sequence,and obtain the second group of forecast results.Finally,we use the GARCH model to predict the original sequence,and obtain the third group of forecast results.The prediction errors of the three modeling methods were evaluated by MAPE and RMSE.The result show that the prediction error of the GARCH model is almost twice than the prediction error of two methods of the EMD-ARIMA model.The prediction error of the “EMD-ARIMA-Reconstruction” method is the smallest.Finally,we summarize the research results,and conclude that EMD can greatly improve the prediction accuracy of time series analysis model,and the subdivision prediction modeling method has the highest accuracy.At the end of this article,we propose that the application of EMD in medium and long term time series analysis can be further studied.