基于RBF网络的中国信贷规模稳健预测

将M-估计稳健损失函数与基于统计贡献度的动态确定核函数方法相结合, 提出一种有效的非参数RBF预测模型, 该方法克服了稳健性缺失问题, 在估计参数的同时动态确定最佳网络结构, 并且在学习中自动消除噪声和异常点的影响, 加快了网络的学习和收敛速度. 利用中国月度信贷数据进行实证分析表明, 本文模型与基准模型相比具有最好的预测稳健性和准确性, 对于提高货币政策有效性和前瞻性具有很好的应用价值.

Chinese credit scale prediction using M-estimator based robust radial basis function neural networks

An M-estimator based robust radial basis function (RBF) learning algorithm with growing and pruning techniques using the concept of statistical contribution was proposed. This method not only eliminates the influence of noise and outliers, but also dynamically adjusts the number of neurons to approach an appropriate size of the network in estimating the parameters at the same time, and so improves the speeds of learning and convergence. Compared with SARIMA and SVR approaches based on empirical RMB monthly loan data series in China, the proposed method has the strongest forecast ability among all methods, and has an important value in applications to improve the effectiveness and perspectiveness of the monetary policy.