Prediction of Macroeconomic indicators in China's Market based on Traditional Time Series model and LSTM Model

Forecasting macroeconomic indicators is a core reference for financial decision-making and policy formulation, but traditional time series analysis methods have modelling bottlenecks in dealing with the complex dynamic characteristics of indicators such as GDP and CPI. Aiming at the shortcomings of traditional models such as ARIMA, which can hardly deal with the nonlinear correlation and cross-cycle dependence of economic data, this study constructs a multivariate forecasting framework based on long- and short-term memory networks. The ARIMA and LSTM models are constructed and compared by integrating the quarterly GDP and CPI core data of the past two decades in China, and incorporating 27 important indicators such as enterprise sentiment index, PMI, and total issued equity. Firstly, the Wind database was used to carry out the domestic 20-year GDP, CPI and other 27 indicators data for statistics. In order to ensure the accuracy of the model data, the missing sequences are supplemented by Lagrange interpolation method, and the data are processed by noise reduction, the scientificity and relevance of the data are ensured by the correlation analysis chart of economic data characteristics and the importance analysis chart of GDP characteristics, and finally, the traditional time series model and LSTM model are compared by modelling to provide objective evaluation and outlook of the model.