Abstract: Commercial banks, which play roles in financing and promoting economic development, directly affect the stability of the banking system and economy. Therefore, it is of great utility to do research on bank idiosyncratic risk prediction. Existing works predict bank risk mainly based on financial indicators. However, financial indicators are unable to provide timely and effective information for bank idiosyncratic risk prediction as they are low-frequency and published with a considerable lag. With the promotion of technology and the improvement of modern finance system, interconnectedness among banks becomes more diverse. The effect of closely connectedness among banks and network effect on bank risk cannot be ignored. Related studies argue that higher centrality helps with increasing return on asset ratios, and reducing financial distress probabilities and liquidity risk. That is because higher centrality provides banks with greater information and resource advantages, and makes these banks disperse their own risk exposure better. Thus, centrality is closely related to bank idiosyncratic risk. However, no study has predicted bank idiosyncratic risk from the perspective of bank network.
Existing works mostly construct bank network based on direct linkages (e.g., liability exposure, interbank asset, payment) or indirect linkages (co-movements in market, e.g., return connectedness, volatility connectedness, tail risk connectedness). However, direct linkage data are difficult to obtain, while bank networks based on indirect linkage data rely on market efficiency and the sample is just limited to listed banks. News reports contain wide soft information about economic linkages among banks, e.g., bank performance, common asset holding and business patterns. Besides, when multiple banks are co-mentioned in the same news report, investor recognition will be spilled over from one bank to other co-occurrence banks, thereby changing investors’ behaviors. Thus, the Internet news co-occurrence network can be constructed based on co-occurrence relationships to reflect more abundant business and non-business relations among banks. In contrast to bank networks based on direct linkage data or indirect linkage data, the Internet news co-occurrence network has high-frequency and high-availability data, and it is not constrained by market efficiency. However, the effect of the Internet news co-occurrence network on bank idiosyncratic risk prediction has not been examined yet.
Motivated by these, this study examines prediction capability of Internet news co-occurrence network on bank idiosyncratic risk. A web crawler program is developed based on Scrapy framework and XPath, based on which we collect more than 2 million news reports about 899 commercial banks. Notably, we propose a new co-occurrence measure for bank interrelations. To be precise, bank interrelations are measured by co-occurrence frequency adjusted by the number of banks appearing in a single piece of news. Dynamic quarterly co-occurrence networks from 2015Q1 to 2021Q4 are then constructed. Furthermore, we calculate information centrality that makes full use of all paths between pairs of nodes to measure the degree of banks’ core position in the co-occurrence network. Finally, we explore the predictive capability of information centrality on bank idiosyncratic risk. The results show that information centrality has incremental predictive capability for bank idiosyncratic risk and it can predict bank idiosyncratic risk for future three quarters. Higher information centrality helps with reducing bank idiosyncratic risk. It is also noteworthy that information centrality calculated under our proposed co-occurrence strategy has better performance in bank idiosyncratic risk prediction than that under existing co-occurrence strategy.
This study is the first attempt to predict bank idiosyncratic risk from the perspective of Internet news co-occurrence networks, which enriches studies on bank networks and bank idiosyncratic risk. Besides, we provide a more effective co-occurrence strategy than that in existing literature. According to our empirical results, it is necessary to take network information into account in bank idiosyncratic risk prediction. Banks with lower information centrality should strengthen their linkages with other banks to reduce their own bank idiosyncratic risk.

Key words: Internet news report, bank co-occurrence network, information centrality, bank idiosyncratic risk, forecast

摘要： 互联网新闻共现网络具有数据频率高、广泛可得,不受市场有效性约束等优点,能反映银行间更广泛的业务和非业务关联。基于Scrapy爬虫框架和XPath网页源码解析工具获取互联网新闻,以经单篇新闻中银行数量调整后的共现次数测度银行间关联强弱,构建银行共现网络。进一步,考虑信息在所有可能路径并行传播,构建信息中心性指标以衡量银行在网络中处于核心地位的程度,最后检验信息中心性对银行个体风险的预测能力。实证研究结果表明:信息中心性显著为未来三季度银行个体风险带来增量预测信息,并且信息中心性提高有利于降低银行个体风险。研究结论为银行个体风险管理带来重要启示:在银行个体风险预测中应纳入银行网络位置信息;信息中心性较小的银行可以通过加强与其他银行的业务和非业务关联以降低自身风险。

关键词: 互联网新闻, 银行共现网络, 信息中心性, 银行个体风险, 预测