JP2020071835A - Credit determination support system - Google Patents

Abstract

To provide a system that supports credit determination of customers.SOLUTION: A system of the present invention includes: means for scoring a customer using at least one of customer information, financial account information, and credit information on the customer; means for determining the state of the customer on the basis of the score and storing it in a database; means for extracting a predetermined number of pieces of information on the customer corresponding to a state for each of states of the customer from the database, the states of the customer including at least a first state of the customer, a second state showing a state relatively worse than the first state, and a third state relatively worse than the second state; means for extracting the information on the customer at substantially the same percentage for the second state and the third state, and for extracting the information at a greater percentage for the first state than the second state or the third state; and means for generating a scoring model for default probability on the basis of the extracted information on the first state, the second state, and the third state.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a system for supporting credit judgment, and more particularly, to a system for supporting credit judgment in order to realize protection of a loanee through comprehensive financial services.

  In recent years, contact points with customers have decreased in financial institutions. In other words, the personnel of financial institutions are less likely to have direct sales to customers, asset / credit management, and financial education. This is because the number of customers visiting ATMs and the number of ATM users are decreasing due to the declining population, the progress of digital technologies such as smartphone applications, the rise of FinTech companies, and the diversification of payment methods. Furthermore, each financial institution is reducing stores and personnel, and it is expected that the number of customer contacts will be further reduced.

  Regarding customer protection, there is a demand for customer protection and user protection such as customer-oriented business operation (fiduciary duty).

  Conventionally, there are systems that support credit judgment, as shown in Patent Documents 1 and 2. However, when giving credit, the customer's condition must be considered from various points of view, such as normal (customers who do not have a problem with lending), mild delinquency, severe delinquency, and bad debt. The system did not exist. That is, it is required to facilitate credit examination and improve credit accuracy.

  Conventionally, respective scoring models such as an overdue model and a bad debt model have been used to judge whether the customer's condition is normal, light delinquency, severe delinquency, bad debt, or the like. However, a scoring model must be created for each state, and the final judgment of the customer's state is made by referring to each scoring model and making a subjective judgment by the person in charge of the financial institution. Was entrusted to.

  In addition to follow-up counseling such as suspension of lending and control of deterioration of the state, there is also an aim as consulting such as grasping condition changes and attribute changes, hearings, financial literacy education.

  Also, it becomes possible to judge the customer's status within the own system without using external credit information.

  Since it takes a huge cost to regularly acquire personal credit information of all customers (loans), it is possible to reduce the frequency of acquisition, but this may delay the detection of credit risk. In addition, as described in Patent Document 1, if a score is used in two stages and only high-risk customers are acquired in the first stage, information of customers whose risk is not manifested is not acquired. Inadequate from a protection point of view.

JP, 2006-344047, A JP, 2017-182284, A

  In terms of operations, particularly regarding personal loans such as card loans, most of the customer support in developing credit management is such that it is difficult to say that user protection such as credit suspension using a credit loss prediction model.

  In terms of the model, the conventional scoring model predicts targets such as “bad debts” and “delinquencies” independently of each other, and it is not possible to continuously represent the deterioration of creditworthiness of customers of financial institutions. In other words, there are many cases in which a person is in a mild or severe delinquency until it becomes a bad debt. In addition, the acquisition of "personal credit information" required for credit management on the way is extremely expensive.

  Therefore, in the conventional credit credit management, credit is suspended for customers with a certain score or less in the scoring model. This is insufficient from the viewpoint of user protection.

  The present invention provides a system for facilitating credit examination and improving credit accuracy.

In a system that supports customer credit decisions,
Means for scoring a customer using at least one of the customer information of the customer, the financial account information and the credit information of the customer;
Means for determining the customer's condition based on the score and storing it in a database;
From the database, for each state of the customer, means for extracting a predetermined number of the information of the customer corresponding to the state,
Here, the customer's condition is at least a first condition of the customer, a second condition indicating a condition relatively worse than the first condition, and a condition worse than the second condition. Including 3 states,
With respect to the second state and the third state, substantially the same percentage of the information of the customer is extracted, and the first state is more than the second state or the third state. Means to extract in large proportion
Means for generating a scoring model relating to a predicted default probability based on the extracted information of the first state, the second state, and the third state;
There is provided a system comprising:

  In the on-going credit management of the present invention, while the face-to-face relationship with the customer decreases, the model that can predict the route of the deterioration of creditworthiness enables the customer to be followed according to the condition.

  Other objects, features and advantages of the present invention will become apparent from the following description of embodiments of the present invention with reference to the accompanying drawings.

1 shows a credit judgment support system according to an embodiment of the present invention. 6 is a flowchart of information processing regarding scoring model creation according to an embodiment of the present invention. 3 shows a flowchart of information processing used for scoring model creation according to an embodiment of the present invention. (A) A comparative example of a data structure of a scoring model according to an embodiment of the present invention and (b) a conventional data structure is shown. An example in which a credit judgment support system according to an embodiment of the present invention is made into a consortium (consortium system) will be shown. An example in which a credit judgment support system according to an embodiment of the present invention is made into a consortium (consortium system) will be shown.

  Examples of the present invention will be described.

  FIG. 1 shows a credit judgment support system 1000 according to an embodiment of the present invention. The credit determination support system 1000 according to the present exemplary embodiment includes a screen display unit 1110, an input unit 1120, a personal information management unit 1130, an extraction unit 1140, a model calculation unit 1150, a follow determination unit 1160, and a follow processing unit 1170. An external information acquisition unit 1180, a financial account DB 1210, a personal information DB 1220, a model storage DB 1230, a payment information DB 1240, a follow-up result DB 1250, a control unit 1310, and an interface unit 1320. Here, DB is an abbreviation for DataBase (database), and is a means for storing respective information.

  The screen display unit 1110 is means for presenting necessary information for credit.

  The input unit 1120 is a means for providing necessary information to the credit judgment support system 1000 as data.

  The personal information management unit 1130 is means for managing personal information (annual income, work place, etc.) of a customer. In addition, in this embodiment, the user, the loanee, and the customer are used synonymously.

  The extraction unit 1140 is means for extracting necessary information.

  The model calculation unit 1150 is means for constructing a scoring model based on the extracted data and calculating a comparable numerical value such as a specialization coefficient.

  The follow determination unit 1160 is a unit that determines the content of follow-up to a customer.

  The follow processing unit 1170 is means for executing a follow for a customer.

  The external information acquisition unit 1180 is means for accessing an external database or the like to acquire information. For example, the customer's credit information is acquired.

  The financial account DB 1210 is a means for storing information on a customer's financial account.

  The personal information DB 1220 is means for managing personal information (annual income, work place, etc.) of customers.

  The model storage DB 1230 is a means for storing an overdue model, a bad debt model, and the like.

  The payment information DB 1240 is means for storing financial payment information.

  The follow result DB 1250 is a means for storing the follow executed to the customer and the result of the follow (correspondence of the customer).

  The control unit 1310 is a unit that controls each unit and each DB inside the credit judgment support system 1000. For example, it may be a processor or the like.

  The interface unit 1320 is means for transmitting / receiving data to / from a server, a terminal or the like outside the credit judgment support system 1000.

  In this embodiment, “delinquency” means that repayment is delayed beyond the repayment date determined by the financial institution. “Mild delinquency” means that the number of days of “delinquency” is short, and “severe delinquency” means that the number of days of “delinquency” is long, but in this specification, for the sake of convenience of explanation, it is decided by the financial institution. Within a prescribed period after the repayment date, it is considered as "mildly overdue", and after the prescribed period, it is considered as "severely overdue". "Credit" means that a loan cannot be collected. "Normal" indicates other states, that is, states other than "mildly overdue", "severely overdue", and "bad debt".

A logistic regression model is used to create the scoring model in this embodiment. The logistic regression model is a general method for calculating a default probability (predicted default probability: PD: Probability of Default).
The default probability is calculated by the following formula.

  Next, the personal credit information specification model will be described. The personal credit information specification model is an example of a predetermined scoring model of the present invention, and is a scoring model in which data based on information included in personal credit information is input data and default probability is output data. The personal credit information specification model can be obtained by subjecting a plurality of personal credit information accumulated in advance to statistical processing or the like. The personal credit information specification model of this embodiment is constructed using logistic regression analysis, and is represented by the following equations 1 and 2.

  Z = β0 + β1 × x1 + β2 × x2 (Equation 1)

  PD = 1 / (1 + exp (-Z)) (Equation 2)

  In Formula 1, Z is a score, x1, x2 ... Are explanatory variables, and β0, β1 ... Are parameters that are predetermined by logistic regression analysis. PD in Expression 2 is a probability of default, and exp represents an exponential function whose base is e (base of natural logarithm).

  The explanatory variables x1, x2, ... Are input data of the personal credit information specification model, and the default probability is output data of the personal credit information specification model. The default probability is the default probability of the customer represented by the personal credit information that is the basis of the explanatory variables x1, x2, .... The probability of default is the probability that repayment from the lender will be delayed under the conditions originally contracted, such as bad debts or delinquency, or changes in the obligor category or modification of conditions, that is, the lender will fall into a state of default in the future. It is a probability. In the present embodiment, it is the probability of falling into the state of “mildly overdue”, “severely overdue” or “bad debt”.

An example of the explanatory variables is shown below.
Explanatory variable x1: Customer attribute information Explanatory variable x2: Customer financial account information Explanatory variable x3: Customer credit internal information (borrowed amount, credit application content, contract content, payment status, balance etc. Credit information generated by
Explanatory variable x4: Customer credit external information (credit information acquired from outside the financial institution, etc.)
Explanatory variable x5: Customer financial behavior information (convenience store ATM usage frequency, frequent overtime usage, etc.)
Explanatory variable x6: Customer follow information (customer follow result)

  The information (parameters) used for these explanatory variables is correlated with delinquencies and bad debts. For example, it is general that a customer with a low financial balance entrusted to a financial institution has a tendency to be overdue or bad (ie, the default probability tends to be high). In this example, a model for predicting overdue and bad debts is created by utilizing the information having such a tendency.

  The parameters β0 to β6 are determined, for example, by the credit judgment support 1000 performing logistic regression analysis using basic information that is personal credit information accumulated in advance by the credit judgment support system 1000. The parameters β0 to β6 may be determined by an information processing device different from the system of the present invention.

  The explanatory variable x1 is customer attribute information and is information held by the financial institution. The explanatory variable x1 is calculated based on age. Further, it may be calculated based on the work style, address change history, surname change history, and gender. Further, for example, information such as name, date of birth, sex, address, telephone number, work place, etc., customer age, working form, and the like held by the financial institution. Customer age is calculated based on the date of birth of the customer. The employment form is determined based on whether or not the employee is working for a corporation or a group such as a company, or whether the person is a manager who runs the company, or whether or not he / she is unemployed. The model to be compared with the personal credit information specification model will be described below. Here, an attribute model and an attribute personal credit information detail model are given as scoring models to be compared with the personal credit information detail model.

  The attribute model differs from the personal credit information detailed model in that only attribute data is used as an explanatory variable. Attribute data is data that a customer has filled in on an application form for a contract such as a loan. In the attribute model, occupation data, number of employees, years of service, type of industry, type of housing, type of work, unmarried, age at work, number of children, number of family members living together, age, gender, type of industry and scale, family status, annual income as attribute data. , And borrowing information is used.

  The attribute personal credit information detailed model is different from the personal credit information detailed model in that attribute data is used as an explanatory variable in addition to the explanatory variable of the personal credit information detailed model. In the attribute personal credit information specification model, information on a residence form, years of service, and annual income is used as attribute data.

  Hereinafter, an explanation variable parameter, a description of the parameter, and an example of the relationship between the parameter and the default probability will be described. When using the following explanatory variables, one parameter may be used or a combination of a plurality of parameters may be used.

The explanatory variable x2 is customer financial account information, such as the deposit balance and transaction history of the customer. The customer financial account information (explanatory variable x2) may be subdivided into customer financial account information (financing) and customer financial course information (deposit). Further, it may be subdivided into customer financial information (others). In such a case, the customer financial account information (financing) is calculated based on the contract status and the loan balance (a mortgage, a car loan, an educational loan, etc. may be separately divided). Further, it may be calculated based on other parameters.
The customer financial account information (deposit) is calculated based on the balance at the end of the deposit (the ordinary deposit, the time deposit, etc. may be separately divided) and the average deposit during the month.

  Customer financial account information (others) may be calculated based on the contract status of investment trusts, insurance, etc., and the contract status of credit cards issued by the company (own bank).

  The explanatory variable x3 is customer credit internal information (credit information generated by being credited inside the financial institution credited from the loan amount, credit application content, contract content, payment status, balance, etc.). For example, it may be calculated based on the borrowed amount, the payment balance, the digestion rate (the ratio of how much of the borrowed limit is used), and the repayment history (whether there is overdue). The personal credit information is information indicating the contract content such as a loan and the payment status.

  The explanatory variable x4 is customer credit external information, which is credit information acquired from outside the financial institution. The explanatory variable x4 may be calculated based on the number and frequency of inquiries of personal credit information, the borrowing balance, the repayment record (whether there is arrears other than the company account), and the presence or absence of negative information.

  The explanatory variable x5 is customer financial behavior information, and may be subdivided into customer financial behavior information (passive) and customer financial behavior information (active).

  The customer financial behavior information (passive) is calculated based on the number and amount of salary transfers and the number of failed account debits. Furthermore, in addition to the number of salary transfers and the amount of money, it may be calculated based on the number of deposits and the amount of money such as the number and amount of pension transfers, the number of dividends received, and the amount of money. Furthermore, in addition to the number of deposits and the amount of money, the number of public utility bills deducted, the amount, type (electricity, gas, water, etc.), the number of tuition deductions, the amount, type (tuition fees, childcare fees, etc.), other companies (Other bank) It may be calculated based on the number of withdrawals and the amount of money, such as the number of issued credit card accounts, amount of money, and type. Furthermore, in addition to the number of failed bank account debits, the number of failed bank bill debits, amount, type (electricity, gas, water, etc.), number of failed school debits, amount, type (tuition fees, childcare fees, etc.) It may be calculated based on the number of unsuccessful credit card debits issued by other companies (other banks), the amount, type, and the like.

  The customer financial behavior information (active) is calculated based on the number of times ATM is used, the number of cases, the place, the day of the week, and the time zone. Furthermore, it may be additionally calculated based on the number of times of use, the number of cases, the place, the day of the week, the time zone, and the number of times of the Internet banking, the number of cases, the place, the day of the week, and the time zone. The number of times of using ATMs at convenience stores and frequent overtime usage are recorded by recording the number of times of using ATMs (Automatic Teller Machine). For example, the number of times ATMs are used by financial institutions and convenience stores at a certain period. The number of uses and the number of uses of ATMs installed in the etc. are recorded.

  The explanatory variable x6 is customer follow-up information, and is calculated based on the previous follow-up content and follow-up result.

  Predict changes in personal credit information and determine the need to obtain personal credit information. Specifically, in the scoring model of the present embodiment, “withdrawal failure” is assumed in the first step leading to bad debts, and concretely, it is predicted because convenience store ATMs and overtime use frequently occur. As a result, the same effect as obtaining information of substantially all customers can be obtained.

  In the present embodiment, the relationship (with the probability of default) in each parameter of the above-mentioned explanatory variables needs to have the same tendency whether it is overdue or in bad debt. Don't have Here, the similar tendency is, for example, a relationship in which the default probability increases (or decreases) both in delinquency and in bad debt when the value of a certain parameter increases (however, The degree of change may differ between delinquencies and bad debts). On the other hand, different tendencies mean that, for example, when the value of a certain parameter changes so as to increase, the default probability changes in the case of delinquency, but the default probability decreases or does not change in the case of bad debt. Is.

  In the present embodiment, an example using the logistic regression model has been described, but as another embodiment, a method (scoring model) such as support vector machine, decision tree, random forest, neural network, deep learning is used. However, the default probability can be calculated using the same input variables (parameters) as in the present embodiment.

FIG. 3 shows a processing table referred to by the follow processing unit 1170.
First, the previous determination result is referred to. In the present embodiment, the previous determination result is “none”, “normal”, “slightly overdue”, “severely overdue”, and “bad debt”. Then, with respect to the result of the previous determination, it is confirmed whether or not the customer has responded to the follow-up executed by the system of this embodiment for the customer. In the present embodiment, if the customer responds, it is “ready”, and if the customer does not, it is “not ready”. Then, the system of the present embodiment executes a new follow for the corresponding item according to the matrix of the determination result of the previous determination, the customer response, and the current determination. Further, the customer's response to the new follow may be stored in any storage means (not shown) of this embodiment. For example, if the previous determination result is “none” (that is, the current determination is the first determination), the previous customer response is also “none”, so the current determination is referred to. If the judgment is normal this time, "wait and see" (that is, the system of the present embodiment does not follow anything to the customer), and if "mildly overdue", the smartphone used by the customer You may want to display a notification to the application that "Mild Overdue" has occurred, or you may automatically send an email that says "Mild Overdue" to the customer's email address. .. In the case of “severe delinquency”, it may be configured such that a customer representative of a financial institution makes a follow-up call (direct phone call to make a proposal for independence support or repayment support) to the customer. Good. In addition, in the case of "credit", "credit suspension" may be performed immediately. In addition, when judging and recording customer correspondence, it may be different for each follow, and in the case of app / email notification, when the notification of the application is judged as read, when replying to the automatically sent mail, It may be determined that the system is “ready”, and otherwise it may be “not ready”. Also, in the case of a follow-up call, if the call for the follow-up call arrives at the call of the other party registered in this system, it may be automatically judged as “ready”. Is configured so that the person in charge of the customer decides “ready” or “not ready” or changes (content of automatically registered customer correspondence) according to the customer's reaction in the dialogue in the follow-up call with the customer. May be.

  Data construction will be described.

  As shown in the table below (examples of the number of extracted cases / ratio), the number (ratio) of each state (normal, slight delinquency, severe delinquency, and bad debt) to be analyzed is as follows. For example, the number of normal cases is arbitrarily set (300,000 cases in this embodiment). Further, the total number (ratio) of abnormal states is smaller than the number (ratio) of normal states (in this embodiment, 10,000 cases, which is 1/10 of the normal number). Set to. Here, the number (percentage) of each of the abnormal states (mild arrears, severe arrears, and bad debts) is set to be equal to each other.

  The adjustment of the weighting of the predicted state of the data used when constructing the scoring model will be described.

  In general, the number of cases in each predicted state decreases in the order of “mild delinquency”, “severe delinquency”, and “bad debt”, so the scoring model that was simply constructed follows the order of the number of cases, so “mild delinquency” The characteristics of are strongly reflected. Therefore, as shown in the table below, the number (percentage) of each condition (normal, mild delinquency, severe delinquency, and bad debt) to be analyzed is set as follows. For example, the number of normal cases is arbitrarily set (300,000 cases in this embodiment). And abnormal state (mildly overdue, severely overdue, bad debt). It is set to be smaller than the normal state (10,000 in this embodiment, which is 1/10 of the normal number). Here, the number (percentage) of each of the abnormal states (slightly overdue, severely overdue, and bad debt) is set to be equal to each other.

  In order to construct such a scoring model, in this embodiment, first, the number of (or desired) loan losses that can be collected from an arbitrary database is determined. Then, the same number of light arrears and severe arrears as the number of bad debts are extracted. The extraction method may be randomly determined. Then, the number of normal cases (and the number of cases of abnormal delinquency, severe delinquency, total bad debts) or the number of cases larger than the number of single bad debts (for example, a predetermined multiple such as 10 times) (and It is also possible to determine the number of cases that can be collected from the database) and randomly extract normal individual data. As another example, the number of overdue or bad debts may be determined based on the number of normal cases. If the number of cases of normal, overdue, and bad debts recorded in the database does not reach the number of cases determined in the above procedure, it is not possible to extract, so the value (number of cases or multiple) can be changed manually or automatically. The system may be configured to display that it is possible.

  By constructing the scoring model as in the present embodiment, it is possible to substantially reduce the weighting of “mildly overdue”, which originally has a large number of cases, and substantially increase the weighting of “bad debt”, which has a small number of cases originally.

  Generally, the number of normal cases is the highest, followed by the cases of mild delinquency, severe delinquency, and bad debt. Therefore, in this embodiment, first, the number of collectable (or desired) bad debts is determined. Then, the same number of light arrears and severe arrears as the number of bad debts are extracted. The extraction method may be randomly determined. Then, determine the number of cases that are not normal (total number of cases of light delinquency, severe delinquency, and bad debts) or that are larger than the number of cases of bad debts alone (for example, at least 10 times in statistical processing). However, normal individual data may be randomly extracted. As another example, the number of overdue or bad debts may be determined based on the number of normal cases. As yet another example, after determining the total number of cases to be acquired and first randomly acquiring the number, the ratio of the number of cases of normal, delinquent and bad debts is calculated, and the ratio is within a predetermined range (for example, normal: delinquent: The necessary number may be added so that the number of cases is within ± 10% of each case based on bad debt = 100: 10: 1.

  Here, the relationship of the ratio between the number of normal cases and the number of abnormal (non-normal) cases can be adjusted in relation to the ranking described later. For example, if the rank is divided into 10 stages, it can be determined that the 1st stage to the 6th stage (upper rank) is normal, and the 7th stage to the 9th stage (medium rank) is overdue (severe delinquency is severe). When the customer specifies to the system that the last 10 stages (lower rank) can be determined to be bad debts, the system specifies Whether or not the model can be constructed is determined, and when such a ranking cannot be performed, the ratio between the number of normal cases and the number of abnormal cases is adjusted. For example, when the range of ranks that can be judged to be normal is narrow, adjustment may be made to increase the number of normal cases (in the present embodiment, increase it to more than 10 times) or decrease the number of abnormal cases. If the range of ranks that can be determined to be normal is wide, adjustment may be made to reduce the number of normal cases (decrease to less than 10 times in this embodiment) or increase the number of abnormal cases. The numbers shown in the table below are approximate.

In the above-described embodiment, an example in which the number of delinquencies and the number of bad debts are (substantially) equal has been described, but an example in which the number of delinquencies and the number of bad debts are different will be described.
In general, the number of delinquencies (severe delinquency and mild delinquency) is more than tens of times the number of bad debts (for example, delinquencies: bad debts = 40: 1), so simply Even if the number of cases is extracted, one scoring model cannot predict delinquencies and / or bad debts with high accuracy. On the other hand, since the number of bad debts is very small compared to the normal and overdue data, there are cases in which the number of delinquent and bad debts cannot be obtained in substantially equal numbers. In such a case, the number of overdue and bad debts may be acquired within the range permitted by the statistical processing. As shown in the table below, as an example, the embodiment 1 can be implemented if the difference between the number of overdue loans and the number of bad debts is up to about 10 times. The number of normal cases is about 10 times or more the number of cases of overdue.

  FIG. 2 shows a flowchart of information processing regarding scoring model creation according to an embodiment of the present invention. S2010-S2030 are executed by the procedure as described in the above table (examples of the number of extraction cases and ratios to be analyzed). As for S2010-S2030, the procedure may be interchanged as described above. In step S2040, the data extracted in steps S2010 to S2030 is used to construct a scoring model using the explanatory variables as described above.

  FIG. 3 shows a flowchart of information processing regarding periodic credit according to an embodiment of the present invention.

  S3010 acquires a scoring model created in advance.

  In step S3020, for any customer whose state is desired to be determined, the customer state such as normal, overdue, or bad debt is determined using the scoring model acquired in step S3010.

  In step S3030, follow-up is performed (follow call or the like) based on the previous recording and the current state (S3020). Here, “follow” is an action of a financial institution providing a loan to a customer, and is executed using a determination table as described later.

  S3040 records the follow result. Specifically, according to the determination table, the determined follow (action taken by the customer) and the customer's reaction to the follow (for example, whether or not the follow is handled) are stored in a predetermined database. Record.

  The relationship between the states of predicting the output result of the scoring model, “mildly overdue”, “severely overdue”, and “bad debt” will be described. The ratios of “mild delinquency”, “severe delinquency”, and “bad debt” and the ratio of “normal” that is the other state are adjusted so that they can be compared. In this embodiment, the specialization coefficient in which the average value of each state is set to 1 is used. It is not necessary to use the specialization coefficient such as the deviation value as long as it is a comparable adjustable value.

  The table below shows numerical examples of specialization factors. As shown in the table below, for example, when the score is divided into 10 stages, 1 to 6 stages (referred to as upper rank) can be determined to be normal, and 7 to 9 stages (medium level). The rank can be determined to be overdue (for example, 7 stages can be divided into mild delinquency, 8 stages and 9 stages can be divided into severe delinquency), and 10 stages (called lower ranks) can be determined to be bad debts. When a customer wants to build such a model to the system, the system determines whether such a model can be built. Adjust the ratio with the number of abnormal cases. Specifically, as mentioned above, adjustments will be made so that the number (percentage) of delinquencies and loan losses will be similar.

Conventionally, if the target of prediction is “credit” or “delinquency” (mildly overdue and / or severely overdue), simply trying to forcibly forecast using one model formula will not guarantee the order of scores and states ( See FIG. 4 (a). In other words, most of the lowest scores may be overdue rather than overdue, which makes it impossible to correctly predict the state of overdue and overdue. One of the reasons why this kind of problem arises is that although there are some cases of delinquencies, the number of cases leading up to bad debts is practically small, and if one model is used to predict bad debts or delinquencies, This is because they tend to be overdue. As an example, if there are over 10,000 out of 10,000 cases (30% of the total), it is known that there are 100 bad cases (1%). ..
On the other hand, in this example, as described in the example of the number of extracted cases / ratio of analysis targets, the number of cases of delinquencies and bad debts (ratio) is adjusted to be approximately the same, and in both cases of delinquencies and bad debts. By using parameters that have the same tendency of default probability, the above-mentioned problems do not occur, and the order of score and status (normal->delinquent-> bad debt) is The collateral can be secured (see FIG. 4 (b)).

  The follow determination table will be described. As for the first follow-up, there is no previous judgment, so follow the first judgment (see this judgment). For example, if the judgment is normal this time, keep a watchful eye (do nothing to the customer from the financial institution), and if it is a slight delinquency, inform the customer that a slight delinquency has occurred through the smartphone application or email. In the case of contact, and if the loan is severely overdue, the financial institution will call the customer (follow-up call), and if it is bad, the credit will be suspended (loan suspension). Here, in the case of mild overdue and severe overdue, the customer's response to the follow-up from the financial institution is recorded. For example, in response to a slightly overdue mail notification, the presence or absence of a mail reply from the customer is recorded in the database.

  Here, in the follow-up determination table, the notification of notification is, for example, a notification of urgent repayment in writing, or notification that legal action will be taken if repayment is not made by the deadline. Further, the visit hearing is, for example, that the person in charge of the financial institution actually visits the loan recipient (home of the customer, etc.). It should be noted that when a notification of a notification or a visit hearing is given, the person in charge of the financial institution inputs data from the input unit 1120 of the credit judgment support system regarding the result of the customer's response (whether the customer responded or not). In the case of follow-up results that can be input from the input unit 1120 (notification notification, follow-up call, visit hearing, etc.), the provisional follow-up result (for example, “unsupported”) is pre-input as data as a countermeasure against non-input. It may have been done.

  Then, after the lapse of an arbitrary period after the first determination, the next determination (hereinafter referred to as the second determination) is performed. In the second judgment as well as in the first judgment, it is judged whether any of normal, mild delinquency, severe delinquency and bad debt is applicable. Then, the previous judgment result (first judgment result), its follow-up, and its customer correspondence are acquired from the database. Then, referring to the table in FIG. 8, the next follow-up content is determined in consideration of the determination result of the previous determination, the follow-up, and the customer correspondence in addition to the second-time determination. For example, when the first judgment is normal and the follow-up from the financial institution is normal, and when the second judgment is normal, the financial institution continues to wait or promote usage (from the financial institution to the customer. Propose further financing such as working capital). By determining according to such a procedure, it becomes possible to appropriately deal with the customer.

  A plurality of types of follow-up determination tables may be stored in an arbitrary database so that they can be selected for each customer and / or each financial institution.

  FIG. 5 shows a server / client system which is a consortium of the credit judgment support system according to the embodiment of the present invention. Note that, of the reference numerals shown in FIG. 5, the same reference numerals as those in the other embodiments have substantially the same functions, and thus the description thereof will be omitted.

  The consortium system according to the present embodiment is a system in which a part of the functions of the credit judgment support system according to the first embodiment is provided in a shared server and information processing is performed on specific information (data) held by a plurality of financial institutions. Is. By using the consortium system as in the present embodiment, it is possible to collect a large amount of information (data), so that an effect of improving the quality (value) of processed data can be expected.

  In the server / client system of FIG. 5, a plurality of credit judgment support systems 1010 (client side) and a consortium system 2010 (server side) are connected via a dedicated line or network 500.

  The credit determination support system 1010 may have a configuration substantially equivalent to the credit determination support system 1000 of the first embodiment, or may have a configuration in which the model calculation unit 1150 and / or the model storage DB 1230 of the credit determination support system 1000 are omitted.

  Each of the plurality of credit judgment support systems 1010 is installed in each financial institution or the like.

  The consortium system 2010 includes a management server 410, a model calculation unit 1150, and a plurality of model storage DBs 1230.

  The management server 410 controls the model calculation unit 1150 and the model storage DB 1230. The management server 410 further transmits data from the model calculation unit 1150 and / or the model storage DB 1230 to the plurality of credit judgment support systems 1020 via the network 500 and data received from the credit judgment support system 1020. Is transferred to the model calculation unit 1150 and / or the model storage DB 1230.

  The consortium system 2010 executes the same processing as S2010-S2040 of FIG. 2 via the management server 410 in response to a request from any credit judgment support system 1010. Here, when extracting the bad debt data (S2010), the overdue data (S2020), and the normal data (S2030), the management server 410, as an example, the management server 410 stores the model recording DB 1230 for bank A and the model for bank B. It may be randomly or averagely extracted from the recording DB 1230, the bank C model recording DB 1230, ... Or, as another embodiment, bank A's bad debt data (may be overdue data or normal data). If you want to create a scoring model based on, the delinquency data and normal data that are insufficient will be extracted from other bank model recording DB 1230 so as to be complemented based on bank A's bad debt data. Good.

  When using data from another bank model recording DB 1230, the management server 410 masks the data designated as personal information, and uses the masked data as the model calculation unit. It may be passed to 1150.

  FIG. 6 shows a server / client system in which a credit judgment support system according to an embodiment of the present invention is formed into a consortium (consortium system). In addition, among the reference numerals shown in FIG. 6, the same reference numerals as those in the other embodiments have substantially the same functions, and thus the description thereof will be omitted.

  The consortium system according to the present embodiment is a system in which a part of the functions of the credit judgment support system according to the first embodiment is provided in a shared server and information processing is performed on specific information (data) held by a plurality of financial institutions. Is. By using the consortium system as in the present embodiment, it is possible to collect a large amount of information (data), so that an effect of improving the quality (value) of processed data can be expected.

  In the server / client system of FIG. 6, a plurality of credit judgment support systems 1020 (client side) and a consortium system 2020 (server side) are connected via a dedicated line or network 500.

  The credit determination support system 1020 may have a configuration substantially equivalent to the credit determination support system 1000 of the first embodiment, or may have a configuration in which the follow determination unit 1160 and / or the follow result DB 1260 of the credit determination support system 1000 are omitted.

  Each of the plurality of credit judgment support systems 1020 is installed in each financial institution or the like.

  The consortium system 2020 includes a management server 420, a follow determination unit 1160, and a plurality of follow result DBs 1260.

  The management server 420 controls the follow determination unit 1160 and the follow result DB 1260. The management server 410 further transmits the data from the follow determination unit 1160 and / or the follow result DB 1260 to the plurality of credit judgment support systems 1020 via the network 500, and receives the data received from the credit judgment support system 1020. It is transferred to the follow determination unit 1160 and / or the follow result DB 1260.

  The consortium system 2020 executes the same processing as S3010 to S3040 of FIG. 3 via the management server 420 in response to a request from any credit judgment support system 1020.

  For example, when determining the follow-up result in S3030 for the customer of Bank A, not only the follow-up result DB 1260 for Bank A, but also the follow-up result DB 1260 for other banks, regarding the previous judgment result, follow, and customer response. Search for. Here, when the management server 410 obtains the search result, it performs mask processing so that it does not know where the bank follow-up result DB 1260 is obtained, and then the follow-up determination unit 1160 receives it. Good.

  For example, when storing the follow-up result in S3040 in the follow-up result DB 1260 for the customer of the bank A, whether to store only in the follow-up result DB for bank A or in all follow-up result DBs for bank 1260 It may be configured so that it can be designated via the credit judgment support system 1020 of the bank.

  The configuration of the third embodiment may be implemented in combination with the configuration of the second embodiment.

  Although the embodiments of the present invention have been described above, various alternatives, modifications, and variations can be made by those skilled in the art based on the above description, and the present invention includes various modifications described above without departing from the spirit of the invention. It is intended to cover alternatives, modifications or variations.

1000 credit judgment support system, 1110 screen display unit, 1120 input unit 1130 personal information management unit, 1140 extraction unit, 1150 model calculation unit 1160 follow determination unit, 1170 follow processing unit, 1180 external information acquisition unit 1210 financial account DB, 1220 individuals Information DB, 1230 Model storage DB
1240 Payment Information DB, 1250 Follow Result DB
1310 control unit, 1320 interface unit

Claims (8)

In a system that supports customer credit decisions,
Means for scoring a customer using at least one of the customer information of the customer, the financial account information and the credit information of the customer;
Means for determining the customer's condition based on the score and storing it in a database;
From the database, for each state of the customer, means for extracting a predetermined number of the information of the customer corresponding to the state,
Here, the customer's condition is at least a first condition of the customer, a second condition indicating a condition relatively worse than the first condition, and a condition worse than the second condition. Any one of the three states,
Further, in the extracting means, the number of cases of extracting the information of the customer having the second state is determined based on the number of cases of the information of the customer having the third state, and The number of cases of extracting the information of the customer having the first state is determined based on the number of cases of the information of the customer having at least one of the second state and the third state,
Further, means for generating a scoring model regarding a default probability based on the information of the extracted first state, the second state, and the third state,
A system characterized by having.   The system according to claim 1, further comprising means for executing a new follow process based on a past follow process and the determination. Further, based on the scoring model, a means for determining the state of any customer who wants to determine the state,
The system according to claim 1, wherein the first state, the second state, and the third state are one of normal, overdue, and bad debt.   The system according to claim 3, further comprising means for changing the status of the arbitrary customer whose status has been determined based on a change in the behavior of the customer.   A method for implementing the system of claim 1.   A program for use in the system of claim 1. In a server / client system that supports customer credit decisions,
The client is
Means for scoring a customer using at least one of the customer information of the customer, the financial account information and the credit information of the customer;
Means for determining the customer status based on the score and storing it in a database;
Equipped with
The server is
From the database, for each state of the customer, means for extracting a predetermined number of the information of the customer corresponding to the state,
Here, the customer's condition is at least a first condition of the customer, a second condition indicating a condition relatively worse than the first condition, and a condition worse than the second condition. Any one of the three states,
Further, in the extracting means, the number of cases of extracting the information of the customer having the second state is determined based on the number of cases of the information of the customer having the third state, and The number of cases of extracting the information of the customer having the first state is determined based on the number of cases of the information of the customer having at least one of the second state and the third state,
Further, means for generating a scoring model regarding a default probability based on the information of the extracted first state, the second state, and the third state,
A system characterized by having. In a server / client system that supports customer credit decisions,
The client comprises the system of claim 1.
The system, wherein the server includes means for executing new follow processing based on past follow processing and the determination.