Credit assessment method

Abstract

The aim is to provide a credit assessment method that improves the accuracy of credit assessment for companies with potential repayment ability by taking into account not only basic features of company external information but also auxiliary features including customer personal information. [Solution] The present invention provides a credit assessment method that is performed by a computer to support credit assessment of small and medium-sized enterprises, startups, or venture companies as customers, and is characterized by comprising the steps of: the computer obtaining, based on information input by an operator, at least a plurality of basic features as company appearance information and a plurality of auxiliary features as company initiative information, as input elements for calculating the default probability; and the computer calculating the default probability by inverse logit transformation by multiplying each basic feature and each auxiliary feature by a regression coefficient.

Description

Technical Field

[0001] The present invention relates to a credit judgment method for assisting in the credit judgment of customers.

Background Art

[0002] Patent Document 1 discloses an invention related to a credit system aimed at providing a highly accurate credit system.

[0003] In the credit system of Patent Document 1, it determines whether to start a transaction with an enterprise analyzed by a financial information-based credit analysis device.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Thus, according to Patent Document 1, since it is a credit model using a financial information-based credit analysis device, small-scale and financially weak enterprises cannot receive sufficient credit transactions.

[0006] However, even among such enterprises, there are enterprises with potentially high repayment ability. In a credit model that makes credit judgments only based on basic external company information as in the past, it was not possible to conduct credit transactions with such enterprises.

[0007] In view of this point, the present invention is made. By taking into account not only the basic feature quantities of external company information but also auxiliary feature quantities that are company activity information for measuring activities and efforts corresponding to recent management issues as a company, such as diversity and SNS dissemination, it aims to provide a credit judgment method that improves the credit accuracy for enterprises with potentially repayment ability. [Means for solving the problem]

[0008] One embodiment of the credit assessment method of the present invention is a credit assessment method performed by a computer to support credit assessment of small and medium-sized enterprises, startups, or venture companies as customers, wherein the computer obtains, based on information input by an operator, at least a plurality of basic features as company appearance information and a plurality of auxiliary features as company initiative information, which serve as input elements for calculating the default probability; and the computer calculates the default probability by multiplying each basic feature and each auxiliary feature by a regression coefficient and performing an inverse logit transformation. Furthermore, between the step of acquiring the basic and auxiliary features and the step of calculating the default probability, it is determined whether the necessary basic and auxiliary features have been acquired. If all the necessary basic and auxiliary features have been acquired, the process proceeds to the step of calculating the default probability. Alternatively, for basic and auxiliary features that cannot be acquired, specific features are input, and the process proceeds to the step of calculating the default probability. It is characterized by the following: [Effects of the Invention]

[0009] According to the credit assessment method of the present invention, it is possible to improve the accuracy of credit assessment for companies that have the potential ability to repay, and to make decisions quickly. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a block diagram of the credit decision system (information processing device) in this embodiment. [Figure 2] Figure 2 is a flowchart of the credit assessment method in this embodiment. [Figure 3] Figure 3 is a flowchart of the credit assessment method in this embodiment. [Figure 4] Figure 4 is a table showing an example of features and regression coefficients used to calculate the default probability. [Figure 5] Figure 5 is a step-by-step diagram showing the process of acquiring features. [Modes for carrying out the invention]

[0011] The embodiments of the present invention will be described in detail below, but the following description is merely an example (representative example) of the embodiments described herein, and the present invention is not limited to these contents unless it exceeds the gist of the invention.

[0012] <Background leading to the credit assessment method of this embodiment> Current credit scoring models determine creditworthiness based on a company's current financial situation. In other words, they calculate default rates based on past performance. As a result, small and medium-sized enterprises (SMEs) with limited financial resources often receive low credit ratings and are unable to obtain sufficient credit.

[0013] Incidentally, among small and medium-sized enterprises, startups, and venture companies, there are many excellent companies that, despite having limited financial resources at present, are expected to grow in the future. If such companies can receive sufficient credit, their businesses will expand, which in turn will contribute to the development of Japan's industries. However, under the current credit scoring model, potentially high-quality companies were unable to receive sufficient credit, hindering their business expansion.

[0014] Therefore, as a result of diligent research, the inventors have developed a credit assessment method that allows credit transactions to be made with companies that are expected to have a high repayment capacity, either potentially or in the future, by focusing not on current or short-term evaluations, but on whether there are elements that can contribute to the growth and improvement of the competitiveness of the organization.

[0015] In other words, the credit assessment method in this embodiment is characterized by improving the accuracy of credit assessment for companies with potential repayment ability by taking into account not only basic (qualitative) features of company appearance information, but also auxiliary features of company initiatives, which are information that measures activities and efforts by companies to address recent management challenges, such as diversity and social media communication, when determining the default probability as a predicted value indicating the possibility of default in the future.

[0016] <Regarding the credit decision system (credit decision device) in this embodiment> FIG. 1 is a block diagram of a credit judgment system (information processing apparatus) according to the present embodiment. The credit judgment system 1 shown in FIG. 1 includes an input unit 3, an arithmetic unit 4, a storage unit 5, and a display unit 6. Although the credit judgment system 1 is divided into a plurality of functional units, this is for convenience of explanation and is not strictly divided, and the plurality of functional units may operate as an integrated functional unit.

[0017] The input unit 3 inputs feature amounts necessary for calculating a default probability. The input value of the feature amount is, for example, "0: no, 1: yes", but is not limited thereto.

[0018] The arithmetic unit 4 calculates a default probability based on the feature amounts input from the input unit 3. The default probability represents the probability of falling into a default (default in debt repayment) state as a percentage (%).

[0019] The storage unit 5 stores the feature amounts input from the input unit 3 and the regression coefficient a. The regression coefficient can also be referred to as an expected ratio or the like. Further, the regression coefficient can be updated periodically or randomly by re-learning considering economic conditions and the like. The larger the regression coefficient is as a positive value, the higher the probability of default, and the larger the regression coefficient is as a negative value, the lower the probability of default.

[0020] The display unit 6 displays on the screen the default probability calculated by the arithmetic unit 4, the credit judgment result of the company information, and the like. In the input unit 3, a plurality of basic feature amounts as company appearance information and a plurality of auxiliary feature amounts as company effort information are input.

[0021] The company appearance information includes at least any one of at least governance information readable from a register transcript, company information readable from an HP, and growth information readable from a company newsletter of a register transcript or an HP. Note that the company appearance information is not limited to these.

[0022] Governance information includes at least one of the following: the number of directors, whether the company is owner-managed, the length of time directors have served, the number of directors who have left office, the company's location, whether it is family-owned, whether it has an auditor or external auditor, and whether it has a board of directors. However, governance information is not limited to these. Governance information can also be obtained from, for example, the Employees' Pension Insurance and Health Insurance Applicable Business Establishment Search System. Company information includes at least one of the following: the number of full-time employees, the number of stores, and the capital. However, company size information is not limited to these.

[0023] Growth information includes at least one of the following: the status of issuance of preferred stock / stock options, the growth rate of the number of employees and / or the number of stores, and the frequency of press releases. However, growth information is not limited to these.

[0024] In addition, information such as whether or not an official gazette announcement has been made, target values ​​that can be gleaned from IR reports, business segments, new business development status, ROE, compliance information (such as whether or not the company is associated with anti-social forces, a history of legal violations, and negative reputation), social insurance office information (such as whether or not health checkups have been conducted), and the details of receivables (such as infrastructure / rent, entertainment, procurement, loan repayment, equipment / development investment, and outsourcing fees) can be obtained. Company profile information can also be obtained online, and can include, for example, whether or not the company's website has been updated within the last three months, and whether or not a photo of the company representative is posted.

[0025] "Company initiative information" refers to information that measures a company's activities and initiatives in response to recent management challenges, such as diversity and social media communication. For example, whether or not a company engages in ESG activities or CSR activities can be considered company initiative information because they are activities and initiatives that address recent management challenges. It is preferable that company initiative information is publicly available, and "public" means that even if procedures such as account registration are required, it can still be considered "public."

[0026] Company initiative information should preferably include at least one of the following: company diversity information and social media information that disseminates information about the company's business activities.

[0027] Diversity information includes at least one of the following: the percentage of women on the board of directors, the gender ratio of employees, the percentage of non-Japanese nationals, the age ratio, the employment rate of people with disabilities, the percentage of women in management positions, the use of teleworking, and the existence of parental leave systems. However, diversity information is not limited to these.

[0028] Diversity information can be gleaned from sources such as a company's website, and this includes not only the company's own website but also those of other companies. For example, job search websites can be cited as examples of other companies' websites.

[0029] For example, a social networking service (SNS) can be selected from one or more of the following: Facebook (registered trademark), Instagram (registered trademark), and X (formerly Twitter) (registered trademark). However, this is not limited to SNS.

[0030] SNS accounts may be company accounts or personal accounts of employees belonging to the company. While not limited to employees, accounts of decision-makers such as CEOs and executive officers are more likely to communicate the company's business activities and are therefore more valuable and preferable.

[0031] For example, input information can be obtained as auxiliary features from social media profile sections, account posts, and the number of followers and following accounts.

[0032] As an example, as shown in Figure 4, the following information is obtained from social media: "percentage of emojis," "text length of profile information," "whether the company name is included in the profile," "average number of characters in a tweet," "number of tweets in the past 3 months," "average number of likes in the past 3 months," "whether the tweet content tends to be positive," "number of people being followed," "number of followers," and "number of followers of followers."

[0033] In this input information, if the value is above a certain level, below a certain level, or within a certain range, "1: Yes" is entered as an auxiliary feature; otherwise, "0: No" is entered. If it is "1: Yes," it can be assumed that it is highly likely to contribute to the communication of the company's business activities, and if it is "0: No," it can be assumed that it is unlikely to contribute to the communication of the company's business activities.

[0034] As shown in Figure 1, the operator can input information that forms the basis of multiple basic features as company appearance information and multiple auxiliary features as company initiative information from the input unit 7 of the credit assessment system 1. The computer automatically selects and inputs "0: No, 1: Yes" based on this input information. Alternatively, the operator can directly input the basic and auxiliary features. That is, a terminal device capable of communicating with the credit assessment system 1 can access the credit assessment system 1 and input multiple basic features as company appearance information and multiple auxiliary features as company initiative information from the terminal device.

[0035] Furthermore, the credit assessment results from the credit assessment system 1 can be confirmed on the contract terminal device 2. The contract terminal device 2 is envisioned to be a terminal device of a credit institution such as a financial institution or credit company.

[0036] <Regarding the credit assessment method in this embodiment> Next, the credit assessment method in this embodiment will be explained using the flowchart in Figure 2. Figure 3 is a table showing an example of features and regression coefficients for calculating the default probability.

[0037] In step ST1 of Figure 2, several basic features are acquired as company appearance information. For this acquisition, numerical values ​​of 0 or 1 can be entered, for example, as shown in Figure 4: "0: No, 1: Yes".

[0038] The operator can input basic features, or the computer can acquire features based on the information entered by the operator. For example, the operator can select the appropriate answer from multiple options displayed on the screen, and the credit scoring system 1 can automatically convert this into features.

[0039] While not limited to information about the company's outward appearance, it is preferable to include at least one of the following: governance information that can be read from the company registration certificate, company size information that can be read from the company's website, or growth potential information that can be read from the company registration certificate or the company's public relations materials on the website.

[0040] For example, as shown in Figure 4, there are input fields such as "Is there an official gazette announcement?", "Has it been established for more than 10 years?", "Does it have multiple businesses?", "Is its capital stock 100 million yen or more?", "Is there a shareholder register administrator?", "Is it a family business?", "Does it issue stock options?", "Does it have a board of directors?", "Has its website been updated within the last 3 months?", and "Is the representative's photo posted?", and you enter "0: No, 1: Yes" for each input field.

[0041] In step ST2 shown in Figure 2, the credit assessment system 1 determines whether all necessary basic features have been acquired. If all features have been acquired, the system proceeds to step ST3; otherwise, it returns to step ST1. However, if there are items that cannot be acquired immediately, the acquisition can be postponed, and the system proceeds to step ST3.

[0042] In addition, the basic feature quantities necessary for credit judgment can be arbitrarily determined and can be reduced or increased. The necessary basic feature quantities can determine specific input items or can be the number of input items. That is, for example, “Is there an official gazette announcement?”, “Is it more than 10 years since establishment?”, “Is the capital more than 100 million yen?”, “Is there a shareholder register manager?”, “Is stock option issued?”, “Is a board of directors installed?” described in FIG. 4 can be set as mandatory input items, and “Are there multiple businesses?”, “Is it family-owned?”, “Has the HP been updated within the last three months?”, “Is there a headshot of the representative posted?” can be set as optional input items, and control can be performed so that it is impossible to transition to step ST3 if the mandatory input items have not been input. Or, although there are 10 input items described in FIG. 4, for example, it is also possible to make the input of 5 or more items mandatory and perform control by a number.

[0043] In addition, for example, when all 10 items shown in FIG. 4 are set as essential basic feature quantities, if there is an item that cannot be input anyway, it may be possible to select “0: No” or, in addition to “0: No, 1: Yes”, select “Unknown”. At this time, “Unknown” is within the range of “0 < a < 1”, and for example, according to the regression coefficient, the input value of the feature quantity of the input item can be given a width.

[0044] In step ST3 of FIG. 2, a plurality of necessary auxiliary feature quantities are acquired. The auxiliary feature quantity is company activity information as information measuring activities and efforts corresponding to recent management issues as a company, such as diversity and SNS dissemination.

[0045] Although not limited to diversity information, it preferably includes at least any one of the female ratio of directors, the male-female ratio of employees, the nationality ratio other than Japanese, the age ratio, the employment ratio of persons with disabilities, the female ratio of management positions, the presence or absence of utilization of telecommuting, and the presence or absence of a parental leave system.

[0046] For example, as shown in Figure 4, there are input fields such as "Does the percentage of women exceed 25%?", "Is there a teleworking system in place and is it being utilized?", and "Is there a parental leave system in place and is it being utilized?", and you enter "0: No, 1: Yes" for each input field.

[0047] Furthermore, as shown in Figure 5, the input screen includes input fields such as "Number of Directors," "Number of Female Directors," "Years Since Establishment," "Family-Owned Business," "Number of Employees," and "Number of Female Employees," and the operator enters specific numerical values ​​into each input field.

[0048] On the computer side, as a backend process, the calculation unit 4 shown in Figure 1 calculates whether the percentage of female directors exceeds 25% based on the input "number of directors" and "number of female directors". If it is 25% or more, it obtains the feature "1: Yes", and if it is 25% or less, it obtains the feature "0: No".

[0049] Furthermore, on the computer side, as a backend process, the calculation unit 4 shown in Figure 1 acquires the "1: Yes" feature if it has been more than 10 years since its establishment, and the "0: No" feature if it has been less than 10 years since its establishment, based on the input "Years since establishment".

[0050] Furthermore, on the computer side, as a backend process, the calculation unit 4 shown in Figure 1 calculates whether "the percentage of female employees exceeds 25%" based on the input "number of employees" and "number of female employees". If it is 25% or more, it obtains the feature "1: Yes", and if it is 25% or less, it obtains the feature "0: No".

[0051] Furthermore, as SNS information used to communicate the company's business activities, we obtain data such as, for example, as shown in Figure 4, "percentage of emoji usage," "text length of profile information," "whether the company name is included in the profile," "average number of characters in a tweet," "number of tweets in the past three months," "average number of likes in the past three months," "whether the tweet content tends to be positive," "number of people being followed," "number of followers," and "number of followers of our followers."

[0052] If the "Emoji Content Ratio" is 10% or more, the auxiliary feature is set to 0; otherwise, it is set to 1. If the "Profile Information Text Length" is less than 100 characters, the auxiliary feature is set to 0; otherwise, it is set to 1. If the company name is not included in the profile, the auxiliary feature is set to 0; if the company name / business name or service name is included, it is set to 1. If the "Average Tweet Character Count" is less than 100 characters, the auxiliary feature is set to 0; otherwise, it is set to 1. If the "Number of Tweets in the Past 3 Months" is less than 24, the auxiliary feature is set to 0; if it is 24 or more, it is set to 1. If the "Average Number of Likes in the Past 3 Months" is less than 20, the auxiliary feature is set to 0; if it is 20 or more, it is set to 1. If the tweet content has a negative tendency, the auxiliary feature is set to 0; if it has a positive tendency, it is set to 1. If the "Number of Following" is 2000 or more, the auxiliary feature is set to 0; if it is less than 2000, it is set to 1. If the "Number of Followers" is 300 or more, the auxiliary feature is set to 1; if it is less than 300, it is set to 0. If the average number of followers of a follower is less than 50, the auxiliary feature is set to 0; if the average is 50 or more, it is set to 1. Regarding the number of followings, if the number is too high, it will lack reliability and credibility, and the auxiliary feature will be evaluated as 0. Regarding the number of followers, if many of the following accounts are reliable and credible accounts, the account will be considered reliable and credible, and the auxiliary feature will be evaluated as 1. Note that the thresholds for each of the input information described above can be changed as needed.

[0053] Incidentally, in the case of startups and venture companies, there are many items that do not correspond to the basic feature input fields or are unknown, making it difficult to make accurate credit assessments. For example, features such as "Has it been established for more than 10 years?" and "Is its capital of 100 million yen or more?" as shown in Figure 4 have a low probability of being applicable, and the input value for these features would be "0: No".

[0054] Startups and venture companies often lack sufficient financial track records. Therefore, traditional credit assessment methods make it difficult to evaluate their future potential. It is impossible to determine business continuity or the feasibility of business plans based on financial statements whose reliability is uncertain.

[0055] Therefore, auxiliary features, which are company initiative information that measures activities and initiatives that address recent management challenges for companies, such as diversity and social media communication, are set as input items. For example, the focus is on whether there are elements that can improve organizational growth and competitiveness through diversity information and social media information, and the potential for long-term improvement of the overall value of the organization is evaluated. In addition, in this embodiment, the default probability is calculated by taking into account not only quantitative information but also qualitative information.

[0056] In step ST4 of Figure 2, the credit assessment system 1 determines whether all necessary company initiative information has been entered. If all necessary company initiative information has been obtained, it proceeds to step ST5; otherwise, it returns to step ST3. Here, whether or not "all information has been obtained" is determined by the essential features determined by the trained model at that time. For example, even if company diversity information can be obtained online but SNS information cannot be obtained, if the operator determines that there is no problem in calculating the default probability, the system may proceed to step ST5 to calculate the default probability without returning to step ST3 in Figure 2.

[0057] Alternatively, as shown in the flowchart in Figure 3, it is possible to first acquire basic features (step ST1) and then acquire auxiliary features (step ST3) in sequence, and then in step ST6, determine at once whether the necessary basic and auxiliary features have been acquired.

[0058] Then, in step ST5 shown in Figures 2 and 3, the default probability is calculated. The regression coefficient 'a' necessary for calculating the default probability is pre-stored in the memory unit shown in Figure 1. The regression coefficient is a predetermined value that indicates the accuracy of the default prediction. It is calculated using machine learning based on a large amount of company data. In the calculation unit 4, the default probability is calculated based on the following (Equation 1) using the input features and the regression coefficients associated with each feature. inv_logit(a1x1+a2x2+···a n xn +b) (Equation 1) (Here, a1, a2…a n (n is an integer greater than or equal to 1) represents the regression coefficient, and x1, x2…x n (n is an integer greater than or equal to 1) represents the feature quantity, and b represents the intercept).

[0059] (Equation 1) is expressed as the probability obtained by taking the inverse logit of the feature quantity (input value) × regression coefficient + intercept. Note that the intercept b is used auxiliary to improve the accuracy of the default prediction value and is set individually according to each credit diagnosis model.

[0060] When conducting the previous credit review for the main finance, the market average default probability is 0.9 - 1.1%. In contrast, in the case of the credit review using the method of this embodiment that uses company activity information as information for measuring activities and efforts in recent years in response to management issues as a company, such as diversity and SNS dissemination, the probability is 0.09%, which is about one-tenth of the previous credit review.

[0061] As a result, even for companies that did not pass the previous credit review, by utilizing the credit review of this embodiment, they can pass the review and receive a speedy credit transaction.

[0062] The auxiliary feature quantity required for credit judgment can be arbitrarily determined, and can be reduced or increased. The required auxiliary feature quantity can determine specific input items, or can also be the number of input items.

[0063] Also, when there are items that cannot be input into the mandatory diversity information and SNS information, it may be possible to select "0: No" or, in addition to "0: No, 1: Yes", select "Unknown". At this time, "Unknown" is set within the range of "0 < a < 1", and for example, according to the regression coefficient, the input value of the feature quantity of the input item can be given a width.

[0064] Since the importance of diversity information and social media information changes depending on the era and industry, it is desirable to periodically review (retrain) the input items and regression coefficients to improve the accuracy of credit decisions made by the credit decision system 1. [Explanation of symbols]

[0065] 1: Credit assessment system 2: Contract terminal device 3: Input section 4: Arithmetic section 5: Storage section 6:Display section 7: Input section

Claims

1. A computer-based credit assessment method to support credit decisions for small and medium-sized enterprises, startups, or venture companies as customers, The computer, based on the information entered by the operator, takes the step of obtaining at least several basic features as company appearance information and several auxiliary features as company initiative information, which serve as input elements for calculating the default probability. The computer calculates the default probability by multiplying each basic feature and each auxiliary feature by a regression coefficient and performing an inverse logit transform. It has, Between the step of obtaining the basic features and the auxiliary features and the step of calculating the default probability, The system determines whether the necessary basic and auxiliary features have been obtained. If all the necessary basic and auxiliary features have been obtained, it proceeds to the step of calculating the default probability. Alternatively, for the basic and auxiliary features that cannot be obtained, specific features are input, and the system proceeds to the step of calculating the default probability. A credit assessment method characterized by the following features.

2. The credit determination method according to claim 1, characterized in that the basic feature quantity and the auxiliary feature quantity are 0 or 1, and the specific feature quantity is 0 or 1 or a value greater than 0 and less than 1.

3. The credit assessment method according to claim 1, characterized in that the aforementioned company initiative information includes at least one of the following: company diversity information and SNS information that disseminates information about the company's business activities.

4. The default probability is given by the credit determination method according to claim 1, as shown in formula 1 below. (Math 1) inv_logit(a 1 x 1 +a 2 x 2 + ··· a n x n +b) (Equation 1) (Here, a 1 , a 2 … a n (n is an integer greater than or equal to 1) represents a regression coefficient, x 1 , x 2 … x n (n is an integer greater than or equal to 1) represents a feature amount, and b represents an intercept).

Images