本文基于计量分析处理多指标数据的优势，结合神经网络最优化模型解的特点，构建了评估和预测国别风险的混合神经网络模型。其中相关性分析和Logit回归分析的结果显示，预测国别风险的两个重要指标是商业自由度（EBI）和外汇储备总额与外债总额之比（RED）；运用两个指标进行预测，多元感知器神经网络模型的预测准确度达到了100%，与其可相互替代的概率神经网络预测模型的准确度达到了90.91%。两个模型的预测结果相互支撑和验证，在一定程度上证明了混合神经网络模型在国别风险预测上具有较强的适用性和可信性。%Based on the advantages of econometric analysis methods in dealing with multiple indicators data and the advantages of neural network models in deriving optimal solutions, we build hybrid neural network models to predict country risks. We get two important predicting indicators—EBI (Ease of doing business index) and RED (Total reserves/total external debt) by the methods of correlation analysis and logistic regressions. The predicting accuracy of the binary-classification multi-layer perception network model is better than that of the probabilistic neural network model, the former reaches 100%and the latter is 90.91%. The predicting results of the two network models support and validate each other, which prove the reliability of hybrid neural network models in predicting country risk.
展开▼
