Small and micro enterprise risk control method based on tax data
By collecting and verifying the tax data of small and micro enterprises, build a logical regression model, obtain classification scores and coefficients, adjust risk control strategies, solve the problem of difficulty in using tax data for traditional risk control methods, realize real -time real -time risks of small and micro enterprises. Monitor and accurate early warning.
Patent Information
- Application Number
- CN202411384706.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-21
AI Technical Summary
Traditional risk control methods are difficult to effectively utilize the tax data of small and micro enterprises, resulting in the inability to detect potential risks in a timely manner.
By collecting the three -party data source of small and micro enterprises involved in taxation, the HC rules verification is performed. Based on the logical regression algorithm, the classification model is established, the classification score and classification coefficient are obtained, and the risk control strategy is adjusted.
Real -time monitoring and accurate early warning of small and micro enterprises risks have improved the scientificity and effectiveness of risk control strategies.
Smart Images
Figure CN119338575A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of risk control management of small and micro enterprises, and in particular to a risk control method for small and micro enterprises based on tax data. Background Art
[0002] With the growing financing needs of small and micro enterprises, small and micro enterprises occupy an important position in the national economy. However, due to their small scale, opaque financial information, and relatively fragile capital chain, small and micro enterprises face higher risks in credit. Traditional risk control methods often rely on static data such as financial statements and credit ratings, which makes it difficult to detect potential risks of small and micro enterprises in a timely manner. In recent years, with the development of big data technology, tax data has become one of the important data sources for risk control of small and micro enterprises due to its authenticity and timeliness, providing strong support for the adjustment of risk control strategies of small and micro enterprises. Therefore, how to use tax data for risk control of small and micro enterprises has become an urgent problem to be solved. Summary of the invention
[0003] The present invention aims to provide a risk control method for small and micro enterprises based on tax data, so as to deeply mine the tax data of small and micro enterprises and build a risk control model based on the tax data, so as to provide strong support for the risk control strategy adjustment of small and micro enterprises.
[0004] In order to achieve the above object, the present invention adopts the following technical scheme:
[0005] A risk control method for small and micro enterprises based on tax data, comprising:
[0006] S1, collects third-party data sources related to taxation of small and micro enterprises;
[0007] S2, verify the HC rules of the three-party data sources;
[0008] S3, building an enterprise classification model based on the logistic regression algorithm, and obtaining classification scores based on the enterprise classification model;
[0009] S4, divide the classification levels based on the classification scores, and issue real-time warnings to small and micro enterprises that meet the warning conditions;
[0010] S5, obtain the classification coefficient based on the classification level, and adjust the risk control strategy of small and micro enterprises according to the classification coefficient.
[0011] The principles and advantages of this solution are: in actual application, traditional risk control models mainly rely on basic financial data and enterprise qualifications, but the data of small and micro enterprises are generally incomplete and non-standard, which makes it impossible for traditional models to accurately assess their risks. By collecting third-party data sources related to taxation of small and micro enterprises, deeply mining and integrating the external and internal data of small and micro enterprises, the data integrity is improved; the third-party data sources are verified by HC rules to improve the availability of data; the enterprise classification model is constructed based on the logistic regression algorithm, the classification score is obtained based on the enterprise classification model, the classification coefficient is obtained based on the classification level, and the risk control strategy of small and micro enterprises is adjusted according to the classification coefficient. It can use the real-time nature of tax data to realize real-time monitoring of small and micro enterprise risks, improve the accuracy and timeliness of risk warnings through in-depth mining of tax data, quantify the results of enterprise classification, and facilitate financial institutions and regulatory authorities to carry out risk management and decision support.
[0012] Preferably, as an improvement, the HC rule verification includes: if the difference between the income in the corporate income tax annual report and the income in the tax data exceeds a threshold, the application is rejected; if the income in the profit statement and the income in the tax data exceed a threshold, the application is rejected; if the corporate address is inconsistent with the address in the tax data, the application is rejected; and cross-checking is performed based on the tax data and the application form data.
[0013] Technical effect: effectively avoid the problems of overfitting and underfitting and improve the generalization ability of the model.
[0014] Preferably, as an improvement, S3 includes:
[0015] S31, preprocessing the third-party data sources and extracting characteristic variables related to enterprise classification;
[0016] S32, confirming the target feature variable;
[0017] S33, based on the logistic regression algorithm, construct and train the enterprise classification model according to the characteristic variables and the target characteristic variables, and the logistic regression algorithm is:
[0018] Y = sigma (beta _0 +beta _1 X _1 +beta _2 X _2 +...+beta _n X _n )
[0019] Among them, Y is the target feature variable, sigma is the sigmoid function, beta _0 , beta _1 , ..., beta _n is the model parameter, X_1 , X _2 , ..., X _n is the characteristic variable.
[0020] Technical effect: Easy to obtain classification scores.
[0021] Preferably, as an improvement, the characteristic variables include the borrower's tax credit rating, recent tax return amount, stability of tax record, company establishment date, tax category, taxpayer status and taxpayer identification number.
[0022] Technical effect: Can provide richer and more accurate input for subsequent risk control models.
[0023] Preferably, as an improvement, in S31, transfer learning technology is also used to mine new feature variables or combined feature variables, and the enterprise classification model is updated in real time.
[0024] Technical effect: Through in-depth analysis of tax data, some new features that are highly relevant to corporate risks are discovered, such as operating income growth rate, tax payment volatility, etc.; multiple features are combined to form composite features to better reflect the risk status of the enterprise.
[0025] Preferably, as an improvement, in S33, reinforcement learning technology is also used to dynamically acquire model parameters.
[0026] Technical effect: It facilitates continuous optimization through interaction with the environment, enabling the model to automatically adjust parameters and weights based on new data, maintaining sensitivity and adaptability to the market.
[0027] Preferably, as an improvement, it also includes S6, which combines the classification level, risk control strategy and portrait parameters of small and micro enterprises for visual display, wherein the portrait parameters include business owner dimension parameters and enterprise dimension parameters, the business owner dimension parameters include multiple conduct, consumption behavior, asset portrait, performance ability, identity characteristics, and relationship network, and the enterprise dimension parameters include enterprise development potential, enterprise risk index, enterprise regional index, enterprise operating ability, and enterprise debt repayment ability.
[0028] Technical effect: Visual display can enable financial institutions to fully understand the real risk situation of small and micro enterprises, so as to make more scientific risk decisions. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 The figure is a flowchart of a risk control method for small and micro enterprises based on tax data. DETAILED DESCRIPTION
[0030] The following is further described in detail through specific implementation methods:
[0031] The embodiment is basically as shown in the attached Figure 1 As shown:
[0032] A risk control method for small and micro enterprises based on tax data, comprising:
[0033] S1, collect third-party data sources related to taxation of small and micro enterprises; obtain tax declaration data, tax penalty records and other relevant information of small and micro enterprises from various channels, such as tax return forms reflecting the basic financial status of enterprises, including operating income, cost and expenses, total profit, etc.; invoice issuance records recording the sales activity information of enterprises, including sales, customer information, transaction time, etc.; tax exemption and reduction amounts and tax exemption and reduction items reflecting the tax preferential policies enjoyed by enterprises.
[0034] S2, perform HC rule verification on the three-party data source; HC rule verification refers to running rule verification, as shown in Table 1. In this embodiment, HC rule verification is performed based on the obtained three-party data related to taxation and the financial data filled in the application form. The HC rule verification includes: if the difference between the income in the corporate income tax annual report and the income in the tax data exceeds a threshold, it is rejected; if the income in the profit statement and the income in the tax data exceed a threshold, it is rejected; if the corporate address is inconsistent with the address in the tax data, it is rejected; cross-check based on the tax data and the application form data, and cross-validate in the data preparation stage to improve the reliability of the evaluation and maximize the data utilization. For small and micro enterprises with limited data, cross-validation can make full use of all data for training and testing.
[0035] Table 1
[0036]
[0037]
[0038] S3, build an enterprise classification model based on the logistic regression algorithm, and obtain a classification score based on the enterprise classification model. The S3 includes: S31, preprocess the third-party data source and extract the characteristic variables related to enterprise classification; according to the business logic and data relevance, screen out the characteristics that have a significant impact on the credit characteristics of small and micro enterprises, such as tax credit level (X1), xx tax declaration amount in the past xx months (X2), the number of months in which the xx tax declaration amount in the past x months is greater than xx (X3), the number of months since the most recent / most distant record of xx tax declaration amount greater than xx yuan (X4), enterprise establishment date (X5), tax category (X6), taxpayer status (X7), taxpayer identification number (X8), etc.
[0039] Considering that the data support of small and micro enterprises is relatively weak, in S31, transfer learning technology is also used to mine new feature variables or combined feature variables, and the enterprise classification model is updated in real time. In the risk control scenario of small and micro enterprises, transfer learning can use the pre-trained model trained on large-scale enterprise data sets to improve the generalization ability of small and micro enterprise data. Specifically, financial data and other relevant information are collected from multiple large enterprises, and a label is assigned to each enterprise to indicate whether it has a bad financial record or a default event; similar data is collected from small and micro enterprises, including financial data and other relevant information, and a label is also assigned to each small and micro enterprise to indicate whether it has a bad financial record or a default event. Select feature variables from large enterprise data sets for feature engineering, use these feature variables to train a pre-trained model, evaluate the performance of the model on large enterprise data sets, and ensure that the model has good generalization ability; ensure the correspondence of large enterprise and small and micro enterprise data sets in feature variables, use the feature variable representation in the pre-trained model to mine new feature variables or combine feature variables, use the intermediate layer representation of the pre-trained model (such as hidden layer output) to generate new feature variable representation, and generate new combined feature variables by combining different features (such as product, summation and other operations).
[0040] S32, confirm the target characteristic variable (Y value); in this embodiment, the default value is mob4dpd30+, mob4 refers to the fourth month of the billing date, dpd30 refers to 30 days overdue, and mob4dpd30+ means that it has been 4 months and has been overdue for more than 30 days.
[0041] S33, based on the logistic regression algorithm, construct and train an enterprise classification model according to the characteristic variables and the target characteristic variables, take the small and micro enterprises that have successfully borrowed money as modeling samples, and extract the Y label; if the Y label is overdue, it is recorded as Y=1; if it is not overdue, it is recorded as Y=0. The logistic regression algorithm is:
[0042] Y = sigma (beta _0 +beta _1 X _1 +beta _2 X _2 +...+beta _n X _n )
[0043] Among them, Y is the target feature variable, sigma is the sigmoid function, beta _0 , beta _1 , ..., beta _n is the model parameter, X _1 , X _2 , ..., X _n is the characteristic variable.
[0044] In S33, reinforcement learning technology is also used to dynamically obtain model parameters. Traditional risk control models use fixed thresholds to classify customers, but this static method cannot adapt to the dynamic changes in the market. The concept of dynamic thresholds is introduced by dynamically obtaining model parameters, and the thresholds are adjusted dynamically according to market changes and corporate development, making corporate classification more accurate, and enabling the model to automatically adjust parameters and weights according to new data, maintaining sensitivity and adaptability to the market.
[0045] S4, divides classification levels based on classification scores, issues real-time warnings to small and micro enterprises that meet warning conditions, and provides corresponding risk management suggestions, so as to facilitate the use of the real-time nature of tax data to achieve real-time monitoring and accurate warning of risks of small and micro enterprises.
[0046] S5, obtain the classification coefficient based on the classification level, and adjust the risk control strategy of small and micro enterprises according to the classification coefficient. In this embodiment, the classification levels include low risk, medium risk and high risk. The low-risk customer group has good post-loan performance, and the overdue rate / default rate is lower than the preset threshold (such as mob4dpd30+<0.5%); the medium-risk customer group has average post-loan performance, and the overdue rate / default rate is within the preset threshold range (such as 0.5%≤mob4dpd30+<1%); the high-risk customer group has poor post-loan performance, and the overdue rate / default rate is higher than the preset threshold (such as mob4dpd30+≥1%). For small and micro-enterprise customers who pay taxes, the basic credit limit is calculated based on the target multiple of their annual tax payment amount. In this embodiment, the credit is preferably magnified by 7-9 times the annual tax payment amount; the classification coefficients of low-risk, medium-risk and high-risk customer groups are: the classification coefficient of low-risk customer group is 2; the classification coefficient of medium-risk customer group is 1; the classification coefficient of high-risk customer group is 0.5; adjusting the risk control strategy for small and micro-enterprises includes adjusting the credit limit of small and micro-enterprise customers, and the credit limit = basic credit limit * classification coefficient.
[0047] It also includes S6, which combines the classification level, risk control strategy and the portrait parameters of small and micro enterprises for visual display. The portrait parameters include business owner dimension parameters and enterprise dimension parameters. The business owner dimension parameters include multiple conduct, consumption behavior, asset portrait, performance ability, identity characteristics, and relationship network. The enterprise dimension parameters include enterprise development potential, enterprise risk index, enterprise regional index, enterprise operating ability, and enterprise debt repayment ability. It is used to comprehensively and scientifically characterize the risks of small and micro enterprises, so that financial institutions can fully understand the real risk situation of small and micro customers, and thus make more scientific risk decisions.
[0048] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A risk control method for small and micro enterprises based on tax data, characterized in that: include: S1, collects third-party data sources related to taxation of small and micro enterprises; S2, verify the HC rules of the three-party data sources; S3, building an enterprise classification model based on the logistic regression algorithm, and obtaining classification scores based on the enterprise classification model; S4, divide the classification levels based on the classification scores, and issue real-time warnings to small and micro enterprises that meet the warning conditions; S5, obtain the classification coefficient based on the classification level, and adjust the risk control strategy of small and micro enterprises according to the classification coefficient.
2. According to claim 1, a risk control method for small and micro enterprises based on tax data is characterized by: The HC rule verification includes: rejection if the difference between the income in the corporate income tax annual report and the income in the tax data exceeds a threshold; rejection if the income in the profit statement and the income in the tax data exceed a threshold; rejection if the corporate address is inconsistent with the address in the tax data; and cross-verification based on the tax data and the application form data.
3. According to claim 1, a risk control method for small and micro enterprises based on tax data is characterized by: The S3 includes: S31, preprocessing the third-party data sources and extracting characteristic variables related to enterprise classification; S32, confirming the target feature variable; S33, based on the logistic regression algorithm, construct and train the enterprise classification model according to the characteristic variables and the target characteristic variables, and the logistic regression algorithm is: Y=sigma(beta _0 +beta _1 X _1 +beta _2 X _2 +...+beta _n X _n ) Among them, Y is the target feature variable, sigma is the sigmoid function, beta _0 , beta _1 , ..., beta _n is the model parameter, X _1 , X _2 , ..., X _n is the characteristic variable.
4. According to claim 3, a risk control method for small and micro enterprises based on tax data is characterized by: The characteristic variables include the borrower's tax credit rating, recent tax return amount, stability of tax record, business establishment date, tax category, taxpayer status and taxpayer identification number.
5. According to claim 3, a risk control method for small and micro enterprises based on tax data is characterized by: In S31, transfer learning technology is also used to mine new feature variables or combined feature variables, and the enterprise classification model is updated in real time.
6. According to claim 3, a risk control method for small and micro enterprises based on tax data is characterized by: In S33, reinforcement learning technology is also used to dynamically acquire model parameters.
7. A risk control method for small and micro enterprises based on tax data according to claim 6, characterized in that: It also includes S6, which combines the classification level, risk control strategy and portrait parameters of small and micro enterprises for visual display. The portrait parameters include business owner dimension parameters and enterprise dimension parameters. The business owner dimension parameters include multiple character, consumption behavior, asset portrait, performance ability, identity characteristics, and relationship network. The enterprise dimension parameters include enterprise development potential, enterprise risk index, enterprise regional index, enterprise operating ability, and enterprise debt repayment ability.