Analysis of Local Macroeconomic Early-Warning Model Based on Competitive Neural Network

Abstract

At present, the commonly used index selection methods for macroeconomic early-warning research include K-L information volume, time difference correlation analysis, and horse farm methods. These traditional statistical methods cannot cope with the continuous changes of economic indicators, and due to the existence of statistical errors, these methods are difficult to perform. Therefore, this paper proposes to use a self-organizing competitive neural network to select early warning indicators. Its self-learning and adaptive characteristics and fault tolerance overcome the limitations of the above statistical methods. This article proposes a method of selecting macroeconomic early-warning indicators using self-organizing competitive neural networks and designs a macroeconomic nonlinear early warning model of self-organizing competitive neural networks; using fuzzy logic reasoning to introduce economic experts’ experience into macroeconomic early warning analysis, the system has the ability to deal with nonlinear and uncertain problems and realizes the intelligence of the early-warning process, uses the national macroeconomic indicator data from January 1997 to March 2008 for empirical analysis, and compares the self-organizing competitive neural network method with the traditional KL information method. From the experimental results, compared with the KL information method, the self-organizing competitive neural network method selects more comprehensive indicators and has greater advantages in seismic resistance and stability.