Identifying discreditable firms in a large-scale ownership network

TL;DR

This paper constructs a large-scale ownership network of Chinese firms to analyze the network effects of discreditable activities, revealing that such activities tend to propagate through the network and can be more accurately identified using these effects.

Contribution

It introduces a novel large-scale ownership network analysis to identify discreditable firms by leveraging network effects, improving detection accuracy over traditional methods.

Findings

Discreditable activities are strongly correlated within ownership networks.

The probability of a firm being discreditable increases if connected to discreditable firms.

Network effects decay with topological distance, following a 'three degrees of separation' pattern.

Abstract

Violations of laws and regulations about food safety, production safety, quality standard and environmental protection, or negative consequences from loan, guarantee and pledge contracts, may result in operating and credit risks of firms. The above illegal or trust-breaking activities are collectively called discreditable activities, and firms with discreditable activities are named as discreditable firms. Identification of discreditable firms is of great significance for investment attraction, bank lending, equity investment, supplier selection, job seeking, and so on. In this paper, we collect registration records of about 113 million Chinese firms and construct an ownership network with about 6 million nodes, where each node is a firm who has invested at least one firm or has been invested by at least one firm. Analysis of publicly available records of discreditable activities show strong network effect, namely the probability of a firm to be discreditable is remarkably higher than the average probability given the fact that one of its investors or investees is discreditable. In comparison, for the risk of being a discreditable firm, an investee has higher impact than an investor in average. The impact of a firm on surrounding firms decays along with the increasing topological distance, analogous to the well-known "three degrees of separation" phenomenon. The uncovered correlation of discreditable activities can be considered as a representative example of network effect, in addition to the propagation of diseases, opinions and human behaviors. Lastly, we show that the utilization of the network effect largely improves the accuracy of the algorithm to identify discreditable firms.