Credit risk prediction method, device and equipment for listed company and medium

Abstract

The invention discloses a listed company credit risk prediction method and device, equipment and a medium, and relates to the technical field of security market credit risk metrology, and the method comprises the steps: determining a correlation coefficient between companies according to the stock quotation data of each company; determining a company association network diagram based on the correlation coefficient between the companies; for any company, based on the company association network diagram, determining associated companies associated with the company; according to a stock information encoder, a financial information encoder, and to-be-detected data corresponding to the company and associated companies related to the company, determining feature data of the company; the to-be-detected data comprises stock quotation data and financial data; and inputting the feature data of each company into a trained graph neural network model, and outputting a prediction result of the credit risk of each company. According to the invention, the accuracy of credit risk prediction of listed companies is improved.

Description

Technical Field

[0001] The present application relates to the technical field of credit risk measurement in the securities market, and particularly to a method, device, equipment and medium for predicting the credit risk of listed companies. Background Art

[0002] As a participant in the securities market and an important participant in the financial market, the credit risk of a listed company poses a great threat to the entire financial market environment. The occurrence of credit risk shows a certain degree of contagion and aggregation, which is very harmful. Accurately measuring the credit risk of a company and dealing with it in a timely manner can provide early warning and reduce the impact of credit risk. The problem of credit risk measurement is a classification problem, and supervised methods are often used for modeling, and the prediction results are used to represent the results of credit risk measurement.

[0003] Regarding the problem of credit risk measurement of listed companies, previous risk measurement methods relied on the experience of analysts, had strong subjectivity, did not comprehensively use the characteristics that could reflect credit risk, regarded each company as an independent individual, and ignored the contagion of credit risk between companies. Therefore, they could not accurately and comprehensively measure the credit risk of listed companies. Summary of the Invention

[0004] The purpose of the present application is to provide a method, device, equipment and medium for predicting the credit risk of listed companies, which can solve the problem of low prediction accuracy of the credit risk of listed companies in the prior art.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In a first aspect, the present application provides a method for predicting the credit risk of a listed company, including:

[0007] Determine the correlation coefficient between companies according to the stock market data of each company; the company is a listed company;

[0008] Based on the correlation coefficient between companies, determine a company association network diagram;

[0009] For any company, based on the company association network diagram, determine the associated companies related to the company;

[0010] According to the stock information encoder, the financial information encoder, and the data to be detected corresponding to the company and the associated companies related to the company respectively, determine the characteristic data of the company; the data to be detected includes stock market data and financial data; the financial data includes financial indicators reflecting the credit risk of the company;

[0011] Input the characteristic data of each of the said companies into the trained graph neural network model, and output the prediction results of the credit risk of each company; wherein, the trained graph neural network model is constructed based on the historical characteristic data and the true credit risk results of each company, and the historical characteristic data is determined based on the historical stock market data and the historical financial data; the true credit risk results of each company are obtained from the rating reports determined by rating agencies for rating each company.

[0012] In a second aspect, the present application provides a device for predicting the credit risk of listed companies, which is characterized by including:

[0013] A company correlation coefficient determination module, configured to determine the correlation coefficient between companies according to the stock market data of each company; the company is a listed company;

[0014] A company association network graph construction module, configured to determine the company association network graph based on the correlation coefficient between companies; and is also configured to, for any company, determine the associated companies related to the company based on the company association network graph;

[0015] A feature mining module, configured to determine the characteristic data of the company according to the stock information encoder, the financial information encoder, and the data to be detected respectively corresponding to the company and the associated companies related to the company; the data to be detected includes stock market data and financial data; the financial data includes financial indicators reflecting the credit risk of the company;

[0016] A classification module, configured to input the characteristic data of each of the said companies into the trained graph neural network model, and output the prediction results of the credit risk of each company; wherein, the trained graph neural network model is constructed based on the historical characteristic data and the true credit risk results of each company, and the historical characteristic data is determined based on the historical stock market data and the historical financial data; the true credit risk results of each company are obtained from the rating reports determined by rating agencies for rating each company.

[0017] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the method for predicting the credit risk of listed companies described above.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the method for predicting the credit risk of listed companies described above.

[0019] According to the specific embodiments provided by the present application, the present application discloses the following technical effects:

[0020] The present application provides a method, apparatus, device, and medium for predicting the credit risk of listed companies. First, based on the stock market data of each company, the correlation coefficient between companies is determined. Then, based on the correlation coefficient between companies, a company association network graph is determined. Through the company association network graph, any company and all associated companies related to the company can be determined. That is, the present application takes into account the impact of the contagion of credit risk between companies on the credit risk of a company, no longer regarding the company as an independent individual, but synchronously considering the associated companies with contagious risk of the company, and based on the stock information encoder and the financial information encoder, feature extraction is performed on the stock market data of the company and the financial data that can reflect the credit risk of the company to obtain accurate company feature data. Finally, the feature data of each company is input into the trained graph neural network model, and the prediction result of the accurate credit risk of each company is output. The accuracy of predicting the credit risk of listed companies is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a schematic flowchart of a method for predicting the credit risk of listed companies provided in an embodiment of the present application;

[0023] Figure 2 It is a schematic structural diagram of a stock information encoder provided in an embodiment of the present application;

[0024] Figure 3 It is a schematic diagram of a device for predicting the credit risk of listed companies provided in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0026] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0027] As Figure 1 shown, the present application provides a method for predicting the credit risk of listed companies, including:

[0028] Step 101: Determine the correlation coefficient between companies based on the stock market data of each company; the companies are listed companies.

[0029] Step 102: Determine the company association network diagram based on the correlation coefficient between companies.

[0030] Step 103: For any company, based on the company association network diagram, determine the associated companies related to the company.

[0031] Step 104: According to the stock information encoder, the financial information encoder, and the data to be detected corresponding to the company and the associated companies related to the company respectively, determine the characteristic data of the company; the data to be detected includes stock market data and financial data as the data to be detected; the financial data includes financial indicators reflecting the credit risk of the company.

[0032] Among them, select solvency indicators, profitability indicators, operating ability indicators, etc. from the financial data that can reflect credit risk, and standardize the risk indicators.

[0033] Step 105: Input the characteristic data of each company into the trained graph neural network model, and output the prediction result of the credit risk of each company; among them, the trained graph neural network model is constructed based on the historical characteristic data and the true credit risk results of each company, and the historical characteristic data is determined based on the historical stock market data and historical financial data; among them, the true credit risk results of each company are obtained from the rating reports determined by rating agencies for each company.

[0034] The stock market data can be obtained from the stock exchange, the company's financial data is regularly disclosed by the company quarterly, and the credit rating data is obtained from the rating reports of rating agencies on the company. The above three types of data are publicly available and easy to obtain. According to the credit rating of the company, the credit risk status of the company can be divided into: risky and risk-free. For companies with credit risk, the financial indicators that can reflect credit risk will deviate from the normal values and the stock market will also show abnormal fluctuations. According to the classification results of the credit rating, the characteristic data can be labeled as risky and risk-free, and then used as the training samples of the graph neural network model. This enables the trained graph neural network model to directly predict or classify the credit risk status of the company to be detected. Compared with the traditional classification model, the classification by the graph neural network model takes into account the contagion of credit risk and can predict the credit risk more comprehensively and accurately.

[0035] In some embodiments, step 101 specifically includes steps 201 - 203.

[0036] Step 201: For any two companies, calculate the logarithmic returns of each trading day of the two companies within a preset number of days respectively.

[0037] Step 202: Determine the Spearman correlation coefficient between the companies according to the logarithmic returns of each trading day of the two companies.

[0038] Step 203: Use the Spearman correlation coefficient between the companies as the correlation coefficient between the companies.

[0039] In some embodiments, step 102 specifically includes: for any two companies, determine the Euclidean distance according to the correlation coefficient between the companies; compare whether the Euclidean distance is less than or equal to a preset threshold to obtain a first result; if the first result is yes, determine that there is an association relationship between the companies; if the first result is no, determine that there is no association relationship between the companies; determine the company association network diagram according to whether there is an association relationship between any two companies.

[0040] Specifically, calculate the correlation coefficient between company i and company j to construct a correlation coefficient matrix. Since the correlation coefficient cannot directly measure the topological relationship of the network diagram, it needs to be appropriately transformed to calculate its Euclidean distance. The calculation formula is: Then, select an appropriate threshold θ. The value range of the Euclidean distance is [0, 2]. When the Euclidean distance is less than θ, it is considered that there is an association relationship between the two companies. When the Euclidean distance is greater than θ, it is considered that there is no association relationship between the two companies. In the company association network diagram, e ij = 1 indicates that there is an edge connection between company i and company j, and e ij is obtained from the following formula:

[0041]

[0042] For different thresholds θ, the connection situation of the edges is different, and different network structures can be obtained. When the threshold θ continuously decreases, the number of edges in the graph also decreases accordingly until the number of nodes in the largest connected subgraph is less than the number of nodes in graph G, breaking the connectivity of the graph and splitting into multiple unconnected graphs. Combine the number of nodes in the largest connected subgraph and the number of unconnected graphs to determine the final value of θ. Through the above operations, an undirected graph can be obtained to determine the structure of the company association network diagram.

[0043] In some embodiments, step 104 specifically includes: inputting the stock market data in the data to be detected into a stock information encoder to extract the first feature vector corresponding to the company and the second feature vectors corresponding to each company associated with the company; inputting the financial data in the data to be detected into a financial information encoder to extract the third feature vector corresponding to the company and the fourth feature vectors corresponding to each company associated with the company; concatenating the first feature vector and the third feature vector to determine the feature vector of the company; concatenating the second feature vectors corresponding to each company associated with the company and the fourth feature vectors corresponding to each company associated with the company to determine the feature vectors corresponding to each company associated with the company; fusing the feature vector of the company and the feature vectors corresponding to each company associated with the company to generate the target feature vector of the company; and using the target feature vector as the feature data.

[0044] Among them, the stock market is a semi-strong efficient market, and stock prices have fully reflected all publicly available relevant information. If there is unpublished insider information, and if investors can quickly obtain this information, stock prices can react quickly. The time-series data related to stocks contains features related to the credit risk of listed companies. Using a stock information encoder to mine the hidden features in stock market data, for high-dimensional long-sequence data, in order to extract effective features through the encoder, a stock information encoder with Transformer as the core is constructed. In analyzing the credit risk of listed companies, in addition to considering forward-looking stock time-series information, basic financial information needs to be combined. For example, the current ratio index. The higher the current ratio, the stronger the company's solvency; the lower the current ratio, the weaker the company's solvency. The financial information encoder can extract effective features from the indicators that can reflect credit risk. By fusing the feature vectors extracted by the two encoders, the features of the company's credit risk can be comprehensively represented.

[0045] In some embodiments, the construction process of the trained graph neural network model specifically includes: determining the historical feature data of the company according to the stock information encoder, the financial information encoder, and the historical data to be detected corresponding to the company and the associated companies related to the company; inputting the historical feature data of the company into the graph neural network model to output the credit risk prediction result of the company; determining the loss between the credit risk prediction result of the company and the true result of the company's credit risk, and taking minimizing the loss as the goal to determine the trained graph neural network model.

[0046] Among them, this application constructs a classification model, which consists of three parts: a stock information encoder, a financial information encoder, and a graph neural network model. The input of the stock information encoder is stock market data, and the input of the financial information encoder is financial data. The outputs of the two encoders are jointly used as the input of the graph neural network model, and the output of the graph neural network model is the final output of the classification model, that is, the prediction result of the credit risk of each company. Classify the credit risk status of listed companies according to the prediction results. The specific process is as follows.

[0047] In the first step, select the indicators that can reflect credit risk from the financial data and standardize the indicators. Among them, the financial indicators given in the financial data include current ratio, quick ratio, super quick ratio, cash ratio, asset-liability ratio, equity ratio, interest coverage ratio, etc.

[0048] In the second step, classify the credit risk status of listed companies according to the credit rating results. For example: a rating of AA or above is regarded as risk-free, and below is regarded as risky. Then, construct a classifier with the stock information encoder, financial information encoder, and graph neural network model as the core.

[0049] As Figure 2 shown, for the stock information encoder, in actual situations, the two main credit ratings of the same company will be separated by a long time, so it is necessary to consider data with a cycle of the past quarter or even half a year. According to the characteristics of long sequence data, use the sparse attention mechanism, followed by residual connection and normalization to improve the feature extraction ability. Then, use a one-dimensional convolutional neural network and pooling operation to reduce the dimension of the extracted features. Through multiple repetitions of the above operations, the main features in the sequence are highlighted. The input of the stock information encoder, that is, the stock market data, consists of input encoding, position encoding, and time encoding with the same dimension. The input encoding u is calculated by the stock line data through a one-dimensional convolutional network and is optimized with the training of the graph neural network model. Among them, for a certain company, u = {u1, u2,....u t} where u1, u2,....u t all represent a vector, t represents time, that is, the number of days of the data, and u1, u2,....u t vertically concatenated to form a matrix as the input encoding. In addition to the daily opening price, highest price, lowest price, closing price, and trading volume data of the stock, the stock market data also includes moving averages that reduce the impact of random short-term fluctuations on the stock price, including 5-day, 10-day, and 30-day price moving averages. For each position data (i.e., the data of a certain day t) and feature dimension e in the stock market data, PE (t,e) represents the position encoding of the data in the stock market data, and the calculation formula is as follows:

[0050]

[0051] Among them, the value range of e is [1, d], where d represents the dimension of the feature. PE1 represents fixing t as 1 and calculating the vector composed of each value of e within the range [1, d]. PE1, PE2,......, PE t The matrix formed by vertical concatenation is used as the positional encoding.

[0052] Inspired by the fact that the time encoding is sometimes directly related to the natural time due to the fluctuations of stocks. The time encoding maps the input discrete index to a continuous vector through the embedding layer from the natural time (month, day) corresponding to the data, obtaining the time encoding E, where E = {E1, E2,....E t}, E t represents the vector corresponding to time t, and E1, E2,......, E t The matrix formed by vertical concatenation is used as the event encoding. The embedding layer is optimized along with the training of the graph neural network model. The input of the stock information encoder is obtained by adding the input encoding, positional encoding, and time encoding.

[0053] For the financial information encoder, the financial indicators containing implicit credit risk information can be calculated using the basic information in the financial report. Table 1 shows some of the used indicators and their meanings.

[0054] Table 1 Financial Indicators and Their Meanings

[0055]

[0056] The financial information encoder uses a multi-layer perceptron structure for feature extraction.

[0057] For the graph neural network model, the constructed company association network graph is used as the graph structure of the graph neural network. Considering the contagion of credit risk, the graph neural network is used to learn the interactions between companies, obtaining the companies associated with the predicted company, and then classification is performed through a fully connected layer. Graph convolutional networks, graph attention networks, etc. can be used. The calculation method of the graph attention network is introduced below.

[0058] The edges of the company association network graph have no weights. The graph attention network can be dynamically adjusted according to the true importance degree, so as to better capture the relationships between companies. The calculation of attention in the graph attention network is only performed among adjacent nodes. For M companies, that is, M nodes, f i represents the feature vector of company i obtained by concatenating the outputs of the stock information encoder and the financial information encoder. In the graph attention network, the similarity coefficient S ij between company i and company j is: Among them, W is an initial value of learnable shared weights. Through this linear mapping, the feature vectors of nodes can be dimensionally increased. || represents the concatenation operation. Together with LeakyRelu, they represent a single-layer feedforward neural network. The calculated similarity coefficient S ij is normalized using softmax to obtain the attention weight α ij , and the calculation formula for the degree of association between companies is:

[0059]

[0060] Among them, ξ(i) represents the set of all associated companies related to company i; α ij represents the weight between company i and company j; e ik represents the weight between company i and associated company k.

[0061] Specifically, when updating the target feature vector (feature data) of company i, node i (company i) fuses the features of all adjacent nodes j (all associated companies). Since the graph attention network uses the multi-head attention mechanism, the features of all attention heads are concatenated to obtain the updated target feature vector f' of company i i , and the calculation formula is:

[0062] Among them, ξ represents the set of all associated companies related to company i; h j represents the feature vectors of company i and all associated companies related to company i. σ(∑ j∈ξ α ij Wh j ) represents the new feature obtained after company i fuses the features of all associated companies; K represents the number of heads of the multi-head attention mechanism; || represents the concatenation of all attention calculation results of company i.

[0063] Then, determine the network parameters of the stock information encoder, financial information encoder, and graph neural network model, mainly including the input sequence length, the number of hidden layers in the fully connected layer, the activation function, the optimization function, and the loss function. Generally speaking, the longer the input sequence length and the more hidden layers, the higher the accuracy of the result. Different activation functions will also have different effects on the model training rate and training accuracy. Commonly used activation functions include the tanh function, relu function, etc.

[0064] Finally, using the feature data as samples and the credit rating classification results as labels, determine the hyperparameters of the classifier, and construct a classifier with the stock information encoder, financial information encoder, and graph neural network model as the core.

[0065] This application constructs a corporate association network diagram based on the correlation coefficient between companies using stock market data. It comprehensively considers various risk factors through a graph neural network model, mines the feature vectors of each company through a stock information encoder and a financial information encoder, fuses the feature vectors to obtain feature data, uses the graph neural network model to mine the association risks between companies, and takes the feature data and the true results of credit risks as training samples for supervised learning and inputs them into a classification model to train a model with appropriate parameters and attribute information. It takes the stock market data and financial data of the listed company to be detected as the input of the trained graph neural network model, so as to obtain the result of whether the listed company to be detected has credit risks. This application can comprehensively and accurately predict the credit risks of listed companies.

[0066] In some embodiments, step 201 specifically includes: for any two companies, according to the daily closing price of company i on the trading day and the daily closing price of the previous trading day, determine the logarithmic return rate of company i on the trading day; wherein, the stock market data includes the daily closing price; the any two companies include company i and company j; according to the daily closing price of company j on the trading day and the daily closing price of the previous trading day, determine the logarithmic return rate of company j on the trading day; based on the preset number of days, obtain the logarithmic return rates of each trading day of company i and company j within the preset number of days.

[0067] In some embodiments, step 202 specifically includes:

[0068]

[0069] where x t represents the logarithmic return rate of company i on the t trading day; y t represents the logarithmic return rate of company j on the t trading day; represents the average value of the logarithmic return rates of each trading day of company i within the preset number of days N; represents the average value of the logarithmic return rates of each trading day of company j within the preset number of days N; ρ represents the Spearman correlation coefficient between company i and company j.

[0070] Specifically, by calculating the logarithmic return rate of the daily closing price of the stock market data, the influence of the absolute difference of the data is eliminated to make the data more stable. The calculation formula is as follows:

[0071] r(t) = ln(v(t) / v(t - 1)).

[0072] where r(t) is the logarithmic return rate of a listed company on the t trading day, v(t) represents the daily closing price of the listed company's stock on the t trading day, and v(t - 1)) represents the daily closing price of the listed company's stock on the t - 1 trading day.

[0073] Specifically, calculate the logarithmic return of the daily closing price of the stock market data to eliminate the influence of the absolute difference in data and make the data more stable.

[0074] Since the absolute differences in the daily closing prices of stocks in different markets, different industries, and different companies are relatively large, this will inevitably cause problems in subsequent processing. Therefore, it is necessary to preprocess the data to eliminate the influence of absolute differences. The method that can be adopted is: obtain the daily line market data of all companies within the same date range, and calculate the logarithmic return using the daily closing prices of every two adjacent dates to make the data more stable and the fluctuation situation more intuitive.

[0075] As Figure 3 shown, the present application also provides a device for predicting the credit risk of listed companies, including: a company correlation coefficient determination module 1, configured to determine the correlation coefficient between companies according to the stock market data of each company; the company is a listed company; a company association network graph construction module 2, configured to determine a company association network graph based on the correlation coefficient between companies; and further configured to, for any company, determine the associated companies related to the company based on the company association network graph; a feature mining module 3, configured to determine the feature data of the company according to the stock information encoder, the financial information encoder, and the data to be detected corresponding to the company and the associated companies related to the company respectively; the data to be detected includes stock market data and financial data as the data to be detected; the financial data includes financial indicators reflecting the credit risk of the company; a classification module 4, configured to input the feature data of each company into a trained graph neural network model and output the prediction result of the credit risk of each company; wherein, the trained graph neural network model is constructed based on the historical feature data and the true credit risk results of each company, and the historical feature data is determined based on the historical stock market data and the historical financial data; wherein, the true credit risk results of each company are obtained from the rating reports determined by the rating agencies for each company.

[0076] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the above method is implemented.

[0077] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0078] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0079] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include Read-Only Memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRdM), magnetoresistive random access memory (MRdM), ferroelectric random access memory (FRdM), phase change memory (PCM), graphene memory, etc. Volatile memory can include Random Access Memory (RdM) or external cache memory, etc. By way of illustration and not limitation, RdM can be in various forms, such as Static Random Access Memory (SRdM) or Dynamic Random Access Memory (DRdM), etc.

[0080] The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., and are not limited thereto.

[0081] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0082] In this article, specific examples are used to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application. At the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for predicting credit risk of listed companies, characterized in that: include: Determine the correlation coefficient between companies based on the stock market data of each company; the companies are listed companies; Based on the correlation coefficients between companies, determine the company association network diagram; For any company, based on the company association network diagram, determine the associated companies related to the company; Determine the characteristic data of the company according to the stock information encoder, the financial information encoder, and the data to be detected corresponding to the company and the associated companies related to the company; the data to be detected includes stock market data and financial data; the financial data includes financial indicators reflecting the credit risk of the company; The characteristic data of each company is input into a trained graph neural network model, and the predicted results of each company's credit risk are output; wherein the trained graph neural network model is constructed based on the historical characteristic data and the actual results of credit risk of each company, and the historical characteristic data is determined based on historical stock market data and historical financial data; the actual results of credit risk of each company are obtained based on the rating report determined by the rating agency for each company.

2. The credit risk prediction method for listed companies according to claim 1, characterized in that: According to the stock market data of each company, the correlation coefficient between companies is determined, including: For any two companies, calculate the logarithmic returns of the two companies on each trading day within a preset number of days; Based on the logarithmic returns of the two companies on each trading day, determine the Spearman correlation coefficient between the companies; The Spearman correlation coefficient between the companies is used as the correlation coefficient between the companies.

3. The credit risk prediction method for listed companies according to claim 1, characterized in that: Based on the correlation coefficients between companies, determine the company association network diagram, including: Determine the Euclidean distance based on the correlation coefficient between the companies; Comparing whether the Euclidean distance is less than or equal to a preset threshold, obtaining a first result; If the first result is yes, it is determined that there is an association relationship between the companies; if the first result is no, it is determined that there is no association relationship between the companies; According to the first result, it is determined whether there is an association relationship between all the companies, and based on whether there is an association relationship between all the companies, the company association network diagram is determined.

4. The credit risk prediction method for listed companies according to claim 1, characterized in that: Determine the characteristic data of the company according to the stock information encoder, the financial information encoder, and the data to be detected corresponding to the company and the associated companies related to the company, specifically including: Input the stock market data in the data to be detected into a stock information encoder, and extract the first feature vector corresponding to the company and the second feature vectors corresponding to each company associated with the company; Inputting the financial data in the data to be detected into a financial information encoder, extracting a third feature vector corresponding to the company and fourth feature vectors corresponding to each company associated with the company; Concatenate the first feature vector and the third feature vector to determine a feature vector of the company; Concatenate the second feature vectors corresponding to the companies associated with the company and the fourth feature vectors corresponding to the companies associated with the company to determine the feature vectors corresponding to the companies associated with the company; Fusion of the feature vector of the company and feature vectors corresponding to each company associated with the company to generate a target feature vector of the company; The target feature vector is used as the feature data.

5. The credit risk prediction method for listed companies according to claim 1, characterized in that: The process of constructing the trained graph neural network model specifically includes: Determine the historical characteristic data of the company according to the stock information encoder, the financial information encoder, and the historical data to be detected corresponding to the company and the associated companies related to the company; Inputting the historical feature data of the company into the graph neural network model, and outputting the credit risk prediction result of the company; Determine the loss between the credit risk prediction result of the company and the actual result of the credit risk of the company, and determine the trained graph neural network model with the goal of minimizing the loss.

6. The credit risk prediction method for listed companies according to claim 2, characterized in that: Based on the logarithmic returns of the two companies on each trading day, the Spearman correlation coefficient between the companies is determined, including: use Determine the Spearman correlation coefficient between companies; Among them, x t represents the logarithmic return of company i on trading day t; y t represents the logarithmic return of company j on trading day t; It represents the average value of the logarithmic return of company i on each trading day in the preset number of days N; represents the average value of the logarithmic return of company j on each trading day in the preset number of days N; ρ represents the Spearman correlation coefficient between company i and company j.

7. The credit risk prediction method for listed companies according to claim 2, characterized in that: For any two companies, the logarithmic returns of the two companies on each trading day within the preset number of days are calculated, including: For any two companies, the logarithmic rate of return of company i on the trading day is determined based on the daily closing price of company i on the trading day and the daily closing price of the previous trading day; wherein the stock market data includes the daily closing price; the any two companies include company i and company j; Determine the logarithmic rate of return of company j on the current trading day based on the daily closing price of company j on the current trading day and the daily closing price of the previous trading day; Based on the preset number of days, the logarithmic rate of return of company i and company j on each trading day within the preset number of days is obtained.

8. A credit risk prediction device for listed companies, characterized in that: include: A company correlation coefficient determination module is used to determine the correlation coefficients between companies based on the stock market data of each company; the companies are listed companies; A company association network diagram construction module is used to determine a company association network diagram based on correlation coefficients between companies; and is also used to determine, for any company, associated companies related to the company based on the company association network diagram; A feature mining module is used to determine the feature data of the company according to the stock information encoder, the financial information encoder, and the data to be detected corresponding to the company and the associated companies related to the company; the data to be detected includes stock market data and financial data; the financial data includes financial indicators reflecting the credit risk of the company; The classification module is used to input the characteristic data of each company into a trained graph neural network model and output the predicted results of each company's credit risk; wherein the trained graph neural network model is constructed based on the historical characteristic data and the actual results of credit risk of each company, and the historical characteristic data is determined based on historical stock market data and historical financial data; the actual results of credit risk of each company are obtained based on the rating report determined by the rating agency for each company.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the credit risk prediction method for listed companies according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the credit risk of listed companies described in any one of claims 1 to 7 is implemented.

Images