Back-propagation neural network and ARIMA algorithm for price trend analysis

Abstract

Forecasting models have a high value in the field of finance, we can better gain in investment and it provide a better basis for the national macroeconomic control strategy. In this paper we build three forecasting models based on a combination of linear ARIMA time series and nonlinear Back-Propagation neural networks to improve the accuracy of forecasting. The first model uses direct summation; the second uses ARIMA and Back-Propagation neural network forecasting results as independent variables and actual prices as dependent variables to build a multiple linear regression model; the third uses bp neural network to indirectly compensate for the residuals of ARIMA results. The more representative gold in the international market was selected as the forecasting object, and the final errors were respectively, proving that the new combination method can improve the accuracy.