关键词: 财务危机预测, 金融网络指标, 信息技术类上市公司, lightGBM

Abstract: With the advent of the era of big data in recent years, the concepts of 5G, Internet+, big data, cloud computing, and blockchain have been put forward, and especially the rapid rise of Internet companies, brings full economic vitality to the country and becomes an important driver for the rapid development of China’s economic quality. At the same time, the information technology industry has gradually become a pillar industry for national economic growth, and a pioneering and strategic industry-leading national production and life. While the IT industry is booming, its high-growth and high-risk characteristics are also becoming more and more prominent. As the information technology industry has a large capital investment at the beginning of the listing, uncertainty in the process and timeliness of research and development, high requirements for technology iteration, the short life cycle of related products, weak solvency, uncertain future earnings of enterprises, unstable cash flow, etc., make the industry more prone to potential financial risks and even the outbreak of financial crises. 16 information technology enterprises were specially treated (ST) in 2019. The information technology industry is therefore in urgent need of financial crisis prediction models, targeting to help companies predict in advance whether there are serious financial risks. The information of indicators in traditional FDP is more limited to financial indicators, or adding non-financial indicators, but market information can also reflect the operation of information technology enterprises, and based on this, this paper proposes to introduce the market information of enterprises into the FDP model.
　Considering that it is difficult to capture the trend of market changes of information technology listed companies, while stock return information can reflect market changes promptly, this paper uses stock information to construct a financial network and introduces stock information into the model in the form of network indicators. To give full play to the role of the integration algorithm in the financial crisis prediction model, improve the generalization ability of the model, and at the same time solve the problem that a single classifier cannot fully use the data, this paper adopts the lightGBM algorithm to construct the financial crisis prediction model of information technology listed companies, and proposes the integration strategy of tuning parameters based on the lightGBM algorithm. Through parameter tuning, the lightGBM algorithm model with the highest accuracy is selected as the base model, and then a new model is obtained by single parameter tuning of the base model. The results of the tuned model and the base model are selected by the classical voting method to obtain the final prediction results. A total of 102 listed companies in the information technology industry in the A-share markets of Shanghai and Shenzhen in China from 2010—2019 are used as the research objects, and the financial and non-financial data of T-3 years are selected, as well as the stock price information of 500 trading days before T-3 years are selected, and the network is constructed by using complex network theory to extract the corresponding network indicators such as centrality and Pagerank value, and combining the enterprise’s financial and non-financial indicators to construct a comprehensive model indicator system.
　After empirical research and analysis, the following results are obtained: (1)Compared with the basic lightGBM model, the prediction performance of the integrated lightGBM model is better, with an accuracy rate of over 90% and a recall rate of 91.18%; (2)Compared with the integrated lightGBM model, the accuracy and recall rate of the integrated lightGBM model with the addition of financial network indicators are increased by 3.17% and 2.57%, respectively. The prediction performance of the model with the addition of financial network indicators is higher than that of the model with only financial and non-financial indicators, as verified by other benchmark models (Logistic Regression, Support Vector Machine, Random Forest). The above results show that the lightGBM algorithm with the integration of tuning parameters has higher prediction performance. Meanwhile, the introduction of stock information increases the effective information in the FDP model, which is beneficial to the financial crisis prediction of enterprises. The research in this paper broadens the application of lightGBM algorithm in enterprise financial risk management and provides new ideas for the construction of financial crisis prediction models.
　The research in this paper focuses on the indicator set and integrated model of FDP. Regarding the indicator study, the inclusion of textual information about enterprises, such as their annual reports, can be considered in future research. Text-based indicators (e.g., “bad debts”, “dividends”, “investment in fixed value assets”) are extracted from companies’ annual reports, and financial indicators, non-financial indicators, market information, and textual information are combined to build the model indicator system. Besides, the data in this study is for Chinese listed companies, and since each country has its judicial system and accounting rules, the data of companies in other countries can be used for the study. The classifier model selection can choose other integrated models or try to improve other machine learning algorithms to predict the financial crisis problem of information technology enterprises.

Key words: financial distress prediction, financial network indicators, listed companies of information technology, lightGBM