融合局部聚类特征的银行间网络重构研究

准确地重构银行间网络是开展银行系统性风险研究的首要工作，实证研究发现最小密度方法低估了真实银行间网络的连边密度且无法刻画其“核心?外围”结构。该文提出一种融合局部聚类特征的银行间网络重构方法。通过引入局部增强矩阵对最小密度方法的概率矩阵进行修正，增强核心银行间的连接倾向。进而，引入自适应因子法对银行间关系重构过程进行权重分配，从而建立具有局部聚类特征的低密度银行间网络。在2018年中国272家银行年报数据集上的实验结果表明，在核心银行连边密度和平均聚类系数上，融合局部聚类网络的银行间网络相比于未融合的银行间网络提升了83.9%和60.1%。而且重构的银行间网络具有稀疏性、异配性和无标度等实证网络结构特征。

Interbank Network Reconstruction Based on Local Clustering Features

Accurately reconstructing the interbank network is the primary task of carrying out research on systemic risks in banks. The empirical study found that the minimum density method underestimated the density of the real interbank network and could not reproduce its "core-periphery" structure. This paper proposes an interbank network reconstruction method that integrates local clustering features. By introducing a local enhancement matrix, the probability matrix of the minimum density method is modified to enhance the connection tendency among core banks. Furthermore, the adaptive factor method is introduced to distribute the weights of the interbank relationship reconstruction process, so as to establish a low-density interbank network with local clustering characteristics. The experimental results on the 2018 annual report data set of 272 Chinese banks show that, compared with the unintegrated interbank network, the interbank network integrated with the local clustering network has increased by 83.9% and 60.1% in terms of density of core bank’s and average clustering coefficient. Moreover, the reconstructed interbank network has empirical network structure characteristics such as sparsity, disassortativity and scale-free properties.