一种新的地方政府债务风险预警系统设计与应用

A New Early-warning System of the Local Government Debt Risk and Empirical Research

本文围绕构建地方政府债务风险预警系统,首先综合运用TOPSIS法和德尔菲法确定了样本的债务风险综合评价值;然后利用支持向量机,提出了基于结构风险最小化的地方政府债务风险预警模型,并将该模型的求解转化为非线性规划仅有线性约束问题,解决了传统方法中忽略模型置信范围、需要样本数量大及过度学习等缺陷。在实证研究中,基于训练样本的模型平均绝对百分精度达99.69%,基于检验样本的模型平均绝对百分精度达96.99%,数值结果表明本文所设计的地方政府债务风险预警系统是有效的,可行的。

The goal of this paper is to construct the local government debt risk early- warning system and give the solution method. Firstly, using the Topsis method and Delphi method to determine the comprehensive evaluation value of the debt risk of the sample. Sec- ondly, the early-warning model of the local government debt risk based on the structural risk minimization is to be established in this paper by using the support vector machine （SVM）. And the model is solved by transforming the model into a nonlinear programming problem with only linear constraints. This model avoids the traditional model of ignoring the confi- dence range of the model, and requires large number of samples, excessive learning and oth- er defects. In empirical research, the average absolute percent accuracy of training sample sreach is 99.69%, the average absolute percent accuracy of the test sample is up to 96.99%, The numerical results show that the model we proposed is effective and feasible.