用人工神经网络方法预测中国外债适度规模

研究了中国外债适度规模及相关政策。在中国宏观外债同步时间序列模型基础上 ,建立了用 BP网络替代随机方程的模型 ,并结合求解一般抽象非线性方程组 F(x,y) =0的算法 ,进行仿真研究。样本期模拟检验结果指出 ,有 92 %的变量相对误差小于 5 %。应用该模型对中国外债管理方式(计划管理方式和余额管理方式 )和期限结构进行仿真研究 ,研究结果表明 ,这种研究方法是可行的、有效的。结果指出 :当中长期外债采取计划管理方式时 ,年借入外债量以不超过2 5 0亿美元为宜 ,建议不超过 2 0 0亿美元

Moderate size of the Chinese foreign debt predicted using artificial neural network approach

The moderate size of the Chinese foreign debt was analyzed using a new algorithm for general nonlinear equations incorporating the BP neural network method and common ways to solve nonlinear equations, like the Wegstein method, the Broyden method, etc. The algorithm was used in a macroeconomic model of the Chinese foreign debt to build an ANN model of the Chinese foreign debt. The model was used to analyze the simulation results of different management modes (planning and residual modes) and the term structure of the foreign debt. The test period results are better than those of the econometric method. The results suggest that the upper limit of foreign debt loans should be 25 billion dollars with the planning mode and payment of the foreign debt within 3 years should be avoided.