JP4338958B2 - Personal credit rating system and personal credit rating program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a system for implementing a credit rating for an individual customer of a financial institution and a program thereof.
[0002]
[Prior art]
In banking business, the importance of risk management of loan receivables is recognized, and external corporate rating information is often used to make credit decisions for business entities. For example, for listed corporations, there are specialized corporate rating agencies such as Japan Credit Rating Agency, Ltd., and for unlisted corporations, there are credit rating agencies such as Teikoku Databank Co., Ltd. Credit rating information is provided based on the above.
For example, as shown in Non-Patent Document 1, a final rating can be obtained through a procedure of evaluation by a financial quantitative model, adjustment by a qualitative factor, and consideration of external information.
[0003]
[Non-patent literature]
“Special Issue Credit Risk (Part 1) Development of Credit Risk Management System Utilizing Credit Ratings” (Toshio Miyashita and Naoto Shimoda, “Securities Analyst Journal March 2002” pages 32 to 35, association (Japan Securities Analysts Association, issued on March 20, 2002)
[0004]
As described above, for corporations, there are institutions that give credit ratings, but for individuals, there are no specialized institutions that conduct credit ratings on a large number of individuals in the country and perform ratings.
However, for the purpose of optimizing individual credit, there are individual credit information institutions for each type of financial institution such as banks, credit card companies, consumer finance companies, etc. The company can know personal credit information regarding the borrowing status and settlement status of personal financial products of loan applicants and customers in business. The personal credit information can be used as an indicator of the relative creditworthiness of the individual who has applied for the borrowing status and settlement status of the individual who has applied from other companies for the individual credit information.
[0005]
[Problems to be solved by the invention]
However, from the viewpoint of personal credit rating, there is a problem that this personal credit information cannot be obtained for individuals who have not applied for use of personal financial products at member companies.
[0006]
In the first place, the personal credit rating represents the credit level related to repayment when assuming the use of personal financial products, and should be given as a certain value in advance for each customer regardless of whether or not the personal financial products are used. It is. From this point of view, any index that represents the degree of personal credit given for the first time when applying for the use of financial products is not within the category of personal credit rating.
[0007]
The present invention has been made in view of the above problems, and does not depend on information outside the financial institution, and does not depend on information from the outside of the financial institution. And providing a system and program that uses a bank account usage information to perform a personal credit rating for customers who have passed a certain period of time after opening a banking account, regardless of whether or not a personal financial product is used. Objective.
[0008]
[Means for Solving the Problems]
In general, the present invention comprises two means. The first means is to construct a poor transaction occurrence probability estimation model by using past financial product transaction case data for individuals stored in a financial institution as a training example, and output from the bad transaction occurrence probability estimation model. Define a rule to convert a bad transaction occurrence probability estimate into a personal credit rating. The second means applies the individual credit rating setting rule obtained from the first means to the individual training cases up to the present time and assigns the individual credit rating at that time to each customer.
[0009]
That is, an object of the present invention is a personal credit rating system that calculates a personal credit rating indicating the degree of expected repayment after a transaction for personal financial products as an order category and assigns it to an individual customer. Estimate of the probability of unsatisfactory transactions occurring as a result of continuous delinquencies over a specified month and / or the occurrence of bad debts that can be applied to individual customers regardless of the presence or absence of financial product transactions based on training cases extracted from customer cases where transactions were conducted A rule generation means for generating a rule for calculating a value, the first conversion rule applied to a credit customer who is an individual customer with a transaction of a financial product, and a bank transaction account but a financial product Rule generating means for generating a second conversion rule to be applied to a non-credit customer who is an individual customer without a transaction, and a first rule for storing the first conversion rule A storage means, a second rule storage means for storing the second conversion rule, and an individual customer who has passed a certain period of time after opening a trading account in a bank and for which some account transaction has been performed, Rule applying means for applying a rule stored in the first rule storage means or the second rule storage means, and when the individual customer is a credit customer, the first rule storage means And applying the first conversion rule stored in the table to assign a rating to the individual customer. On the other hand, if the individual customer is an uncredited customer, the rating is stored in the second rule storage unit. A rule applying means for assigning a rating to the individual customer by applying a second conversion rule, and a rating data storage means for storing the assigned rating in association with the individual customer. Pieces It is achieved by the credit rating system.
[0010]
In a preferred embodiment, the rule generation means is a bad transaction occurrence probability model for estimating a bad transaction occurrence probability from a training example, from a past time point “a” to a time point “b” (where a <b). A training example satisfying the condition that the consumer financial product is being used in the first period and satisfying the condition that no foul transactions have occurred within the period is used as a common training example, Account usage information including liquidity deposit balance, fixed deposit balance, ATM withdrawal count in the first period up to “b”, personal basic attributes including age, gender and job type at time “b”, and the time point The use / settlement information of the personal financial product in the first period from “a” to time “b” is used as an explanatory variable, and the information from the time “b” to the current “c” (where b <c) In period 2 A first unsatisfactory transaction occurrence probability estimation model with the presence or absence of unsatisfactory transaction settlement as an objective variable, the account usage information in the first period from the time “a” to the time “b”, and Second unsatisfactory transaction occurrence with the individual attribute at time “b” as an explanatory variable and the presence or absence of occurrence of unsatisfactory transaction in the second period from time “b” to time “c” as an objective variable A probability estimation model is generated.
[0011]
In the rule generation means, the model construction method used to estimate the probability of occurrence of unsatisfactory transactions is a discriminant tree model etc. with the presence or absence of unsatisfactory transactions occurring as an objective class and other items as explanatory variables. Even a (Classification) model can be applied to the present invention.
[0012]
Further, in a preferred embodiment, the rule generation means converts the unsatisfactory transaction occurrence rate on the training example estimated by the first unsatisfactory transaction occurrence probability model into an arbitrary number of order scale category values. A second conversion rule that generates one conversion rule and converts the unfavorable transaction occurrence rate on the training example estimated by the second unfavorable transaction occurrence probability model into an arbitrary number of order scale category values Is generated.
[0013]
One method of conversion to a general rating is to adjust the appearance distribution of credit ratings to satisfy a certain composition ratio in the customer population. Here, after obtaining the cumulative distribution function of the unsatisfactory transaction occurrence probability estimated value in the training example, the unsatisfactory transaction occurrence probability estimated value corresponding to the accumulated appearance probability may be used as the threshold increment. For example, the rating “A” is the top 5%, the rating “B” is the next 10%, the rating “C” is the next 20%, the rating “D” is the center 30%, and the rating “E” is the next 20%. If you want to distribute the rating “F” to the next 10% and rating “G” to have the remaining 5%, the cumulative distribution value in the cumulative distribution function of the unfavorable transaction occurrence probability value is A rule for conversion to personal credit rating may be defined using the estimated unfavorable transaction occurrence probability corresponding to 5%, 15%, 35%, 65%, 85%, 95% and 100% as a threshold value.
[0014]
Alternatively, as a conversion rule, a method may be employed in which the size of the unsatisfactory transaction occurrence probability estimation value itself is used as a threshold value. For example, if the estimated probability of occurrence of unfair transactions is 0%, the rating is “A”, if it is more than 0% and less than 1%, the rating is “B”. If it is more than 3%, the rating is “D”. If it is more than 3%, 5% or less, the rating is “E”. If it is more than 5%, the rating is “F”. Any other conversion rule including the above technique can be used in the present invention.
[0015]
In another preferred embodiment, the rule applying means is a third point in time for the credit customers out of the individual customers whose at least the first period has elapsed since the opening of the trading account. Account usage information of the existing credit customer in a second period corresponding to the first period from “d” to the current “c” (where d <c), personal basic attributes at the current “c”, And based on the use / settlement information of the financial product in the second period, the rating corresponding to the estimated unfavorable transaction occurrence probability is obtained by referring to the first conversion rule stored in the first rule storage means. And the account usage information of the non-credit customer in the second period corresponding to the first period from the third time point “d” to the current time “c”, “C Based on the personal basic attributes in, with reference to the second conversion rule stored in the second rule storage means, and is configured to provide a rating for that FuKaoru transaction occurrence probability value.
[0016]
Further, when the individual customer is a non-credit customer, the rule application means applies the second conversion rule to give the individual customer a preliminary rating, and the individual customer A management means for causing the rule applying means to apply the first conversion rule and to give a rating based on a loan performance when a period corresponding to the first period has elapsed after starting the transaction of It is desirable to have it. Thus, for example, when a non-credit customer who applied for a financial product transaction (for example, applied for a mortgage loan) is given a preliminary rating and a predetermined period of time has elapsed since the start of the financial product transaction It is possible to give a more accurate new rating.
[0017]
In addition, an object of the present invention is a computer-readable program that calculates a personal credit rating indicating the degree of expected repayment after a transaction for a personal financial product as an order category and operates the computer to give to an individual customer. And defined by the occurrence of continuous delinquencies over a predetermined month and / or the occurrence of bad debts that can be applied to individual customers regardless of the presence or absence of financial product transactions based on training cases extracted from customer cases in which transactions with financial products for individuals were conducted in the past A rule generation step of generating a rule for calculating an estimated value of the probability of occurrence of unsatisfactory transactions, wherein a first conversion rule to be applied to a credit customer who is an individual customer with a transaction of a financial product is generated The first conversion rule is stored in the first rule storage means, and the bank transaction account is owned but there is no financial product transaction. A second conversion rule to be applied to a non-credit customer who is a human customer, a rule generation step for storing the second conversion rule in the second rule storage means, and a certain period of time after opening a bank transaction account A rule applying step of applying the rule stored in the first rule storage means or the second rule storage means to an individual customer who has passed and some account transaction has been performed in the meantime. If the customer is a credit customer, the first conversion rule stored in the first rule storage means is applied to assign a rating to the individual customer, and the rating is associated with the individual customer. If the personal customer is an uncredited customer, the second conversion rule stored in the second rule storage is applied to the personal customer. With the grant, the rating, and rule application storing the rating data storage means in association with the individual customers, also achieved by a program for causing the computer to perform.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below with reference to the accompanying drawings. The personal credit rating system according to the present embodiment extracts a financial institution transaction customer and executes a predetermined process on the customer to assign a personal credit rating.
In this specification, first, the principle of personal credit rating by the personal credit rating system will be described, and then the processing by an actual program will be described.
[0019]
In the personal credit rating system, the current time “c” is set to a certain date, and the first past time “a”, the second past time located between the first past time and the current time Appropriate values are assigned to “b”. Here, by way of example, the current time “c” is October 1, 2001, the first time “a” is retroactive from the current time “c” 12 months ago (October 1, 2000), the second The time “b” is set to be 6 months ago (April 1, 2001) from the present time.
[0020]
In this embodiment,
(1) During the period from the first time point “a” to the second time point “b”, the consumer financial product is still being used, and there have been three or more delays of 4 days or more within that period. As a training example, we use customers that were extracted based on the above conditions, or the condition that no delinquent transactions have occurred more than once a month,
(2) Establish a bad transaction occurrence probability estimation model that employs a discriminant tree method based on entropy criteria,
(3) Adopting a conversion method to personal credit rating based on the probability of occurrence of unfair transactions, defining conversion rules,
(4) Assign individual credit ratings to uncredited customers and already credited customers at the second time point “b”.
[0021]
First, training examples used in (1) will be described. As the definition of the occurrence of unsatisfactory transactions, which is the target class of the model, 4 days in the period from the second time point “b” (April 1, 2001) to the day before the current time “c” (September 30, 2001) It is assumed that the above-mentioned delinquency is 3 times or more or the monthly overdue is 1 or more times.
[0022]
The explanatory variable of the first unsatisfactory transaction occurrence probability estimation model applied to the existing credit customer is from the first time point “a” (October 1, 2000) to the day before the second time point “b” (March 2001). Monthly total of liquid deposits, fixed deposits, ATM withdrawals, total credit self-payments, salary transfers, automatic transfers, remittances, utility charges deductions, and pension transfer amounts Bank account usage information to be stored, personal attribute information to store the date of birth, gender, and initial account opening date at the second time point “b” (April 1, 2001), and in addition to the first A loan that stores the total monthly loan balance from the time “a” (October 1, 2000) to the day before the second time “b” (March 31, 2001), and the average value of loan income Use / settlement information.
[0023]
On the other hand, the explanatory variable of the second unsatisfactory transaction occurrence probability estimation model applied to the non-credit customers is also the same as the previous day from the first time point “a” (October 1, 2000) to the second time point “b” ( Monthly total of liquid deposits, fixed deposits, ATM withdrawals, credit self-transfer totals, payroll transfers, automatic transfers, remittances, utility charges deductions, pension transfer amounts up to March 31, 2001) The bank account usage information that stores the average value, and the personal attribute information that stores the date of birth, gender, and date of initial account opening at the second time point “b” (April 1, 2001).
[0024]
Next, the discriminant tree method based on the entropy criterion employed in the present embodiment will be described in more detail. This is described in detail in, for example, “C4.5: Programs for Machine Learning (1993, published by Morgan Kaufmann Publishers, Inc. by J. Ross Quinlan)”.
[0025]
A set S of individual customers is represented by a purpose attribute (generally N class C ₁ , ..., C _N However, in this case, it is assumed that there are two classes for the occurrence of unfair transactions. )
So that k can be properly explained, k disjoint subsets T ₁ , ..., T _k Think about dividing it. In general, an entropy function representing the degree of mixing of classes is defined as follows. freq (C _j , S) gives class C in set S _j Represents the number of cases. At this time, one case is selected at random from the set S of cases, and this is the class C. _j Prob (C _j ) Is represented by Equation 1.
[0026]
[Expression 1]
Here, | S | represents the number of cases in the set S.
At this time, class C _j Amount of information Info (C _j ) Is represented by Equation 2.
[0027]
[Expression 2]
When the amount of information is defined in this way, the probable event (Prob (C _j ) Is large), the amount of information is small and rarely occurs (Prob (C _j ) Is small), the amount of information is large. What is represented by Equation 3 that averages the amount of information by the ease of selecting a class is referred to as entropy (S) of the set S.
[0028]
[Equation 3]
X, set S to T ₁ , ..., T _k Is one of the methods of dividing into two. Here, k may be from 2 to the number of categories of X. At this time, the divided subset T _i Entropy expected value of entropy _x (S) is represented by Entropy (T _i ) Is given as a weighted average.
[0029]
[Expression 4]
Here, the gain Gain (X) defined by the difference in entropy caused by the division X is given as shown in Equation 5, which increases the amount of information for predicting the class on average by the division. Represents.
[0030]
[Equation 5]
When the gain itself is used as a reference, there is a drawback that X having a large division number k is easily adopted when dividing the set S. In the above-mentioned literature, in order to prevent this, division is performed using the following criteria. First, the division information amount Split (X) is defined as shown in Equation 6.
[0031]
[Formula 6]
Next, a gain ratio GainRatio (X) whose gain is adjusted by Split (X) is defined as shown in Equation 7.
[Expression 7]
[0032]
Further, X that maximizes the gain ratio GainRatio (X) is selected as the division attribute of the customer set S under the constraint that the gain Gain (X) is equal to or greater than the average. Specifically, it follows the following algorithm.
[0033]
Step 1. First case set S ₀ Are assigned to the set S to be divided.
Step 2. If the number of cases in the set S to be divided is greater than a predetermined minimum case number criterion, all attributes X for the set S _i The divided gain and gain ratio are calculated for (i = 1, 2,..., M,..., M).
Step 3. For an attribute whose gain is greater than a predetermined minimum gain criterion, the attribute X having a gain ratio greater than average and a maximum gain ratio _m Is selected, and the set S to be divided is divided by using it.
Step 4. Subset T after splitting _j (J = 1, 2,..., K) are assigned to the set S to be divided, and the processes in steps 2 and 3 are repeated recursively.
[0034]
Algorithm stop condition: When the number of cases in all the division target sets S is less than the minimum number of cases criterion in Step 2, or in any division target set S in Step 3, no matter what division attribute is selected If no gain criterion is obtained, the algorithm is terminated.
[0035]
FIG. 8 is a diagram illustrating an example in which a subset is obtained from the set S including 100 cases by executing the processes of Step 1 to Step 5 described above. Here, there are two contradictory subsets, that is, the entire set S (see reference numeral 801), the customer subset by application in the campaign (see reference numeral 811), and the customer subset by application in the store (reference numeral 812). Has been generated.
Further, for the former, a subset of customers having the attribute of male to be rejected (see reference numeral 821) and a subset of customers having the attribute of female (see reference numeral 822) are generated, while the latter is rejected. A subset of customers having an attribute of less than 30 years old (see reference numeral 831) and a subset having an attribute of 30 years or older (see reference numeral 832) are generated.
[0036]
The output of the unfavorable transaction occurrence probability estimation model that employs the discriminant tree method based on the entropy criterion as described above is obtained by dividing the entire case customer group into n1 segments with respect to the first model to be applied to the non-credit customers. It becomes. On the other hand, regarding the second model to be applied to the existing credit customer, the entire case customer group is divided into n2 segments. Without loss of generality, here, for example, n1 = 180 and n2 = 120.
[0037]
The conversion rule from the unsatisfactory transaction occurrence probability estimated value to the personal credit rating can be arbitrarily set, but here 7 to 7 from A to G depending on the size of the unsatisfactory transaction occurrence probability estimated value without losing generality. Set on the street. The threshold values of the corresponding unsatisfactory transaction occurrence probability estimates are 1%, 3%, 5.5%, 10%, and 20%, respectively.
Thereafter, the personal credit rating rule is applied to the uncredited customers and the credit customers at the second time point “b” (April 1, 2001) to give the personal credit ratings. As a result, for example, a rating distribution as shown in Table 1 below is obtained.
[0038]
[Table 1]
[0039]
A system and a program for actually realizing the rating described above will be described below. FIG. 1 is a block diagram showing a hardware configuration of a computer system for realizing a personal credit rating according to the present embodiment. As shown in FIG. 1, a computer system 10 includes a CPU 12, an input device 14 such as a keyboard and a mouse, a display device 16, a storage device 18 such as a hard disk, a RAM 20 used as a work area being processed, a CD-ROM, and the like. A storage medium driver 22 for accessing the portable storage medium 21 is provided. Of course, various interfaces (not shown) for this purpose may be provided in order to exchange data from the outside via a network (not shown).
[0040]
In the present embodiment, the personal credit rating program is stored in the CD-ROM, read by the storage medium driver 22, and stored in the storage device 18. Alternatively, it may be downloaded from outside via a network (not shown). The data stored in the storage device 18 is read as needed, and an image based on the data is displayed on the screen of the display device 16, and also via a storage medium driver 22, a CD-R or CD-R / W Or the like can be stored in a recordable portable storage medium. The storage device 18 has a capacity sufficient to accommodate these data. By installing a personal credit rating program, the computer system can function as a personal credit rating system.
[0041]
FIG. 2 is a block diagram describing the functions of the personal credit rating system according to the present embodiment. As shown in FIG. 2, the personal credit rating system according to the present embodiment includes a rule generation unit 24 that generates a personal credit rating rule for personal credit rating, and a predetermined number according to the generated personal credit rating rule. And a rule application unit 26 that executes customer ratings.
[0042]
In the present embodiment, as will be described in detail later, a loan has already been established (that is, a loan is used) and a repayment customer (existing credit customer) and a loan has not been established (that is, a loan). A rating can be given to each of the customers (non-credit customers) who are not using the service. In order to realize this, the rule generation unit 24 and the rule application unit 26 include components that execute processing related to credit customers and processing related to non-credit customers, respectively. The rule generation unit 24 includes an existing credit customer rating rule generation unit 28 that generates an individual credit rating rule for an existing credit customer, and an uncredit customer rating rule generation unit 30 that generates an individual credit rating rule for an uncredited customer. And have. Similarly, the rule application unit 26 includes an already-credit customer rating rule application unit 32 and an uncredit customer rating rule application unit 34. The functions of the rule generation unit 24 and the rule application unit 26 are realized mainly by the operation of the CPU 12 in FIG.
[0043]
As shown in FIG. 2, in the present embodiment, information indicating customer attributes, customer transaction status, and presence / absence of unsatisfactory transaction is stored as a DB. More specifically, the customer rating system includes a loan use / settlement information DB 36, a personal attribute information DB 38, a bank account use information DB 40, and a purpose class information DB 42. These DBs can be built in the storage device 18 of the system.
[0044]
In the loan use / settlement information DB 36, at least for each credit customer, the total monthly loan balance in the period from the first time point “a” to the second time point “b”, the average value of the loan income, etc. Information indicating so-called loan repayment status is stored. The attribute information DB 38 stores customer personal information such as the date of birth, sex, and date of initial account opening for each customer. Further, the bank account usage information DB 40 includes at least a monthly liquid deposit balance, a fixed deposit total, an ATM for each customer in a period from the first time point “a” to the second time point “b”. Bank account usage information including the average value of withdrawal, credit self-transfer total, salary transfer, automatic transfer, remittance, utility charge deduction, and pension transfer is stored.
[0045]
Note that information stored in the DB of the system in the bank can be extracted and used as the loan use / settlement information and bank account use information. Alternatively, the necessary information may be extracted by directly accessing the system DB in the bank and used for rule generation and rule application processing described later. In the former, for example, as pre-processing, from the information stored in the DB of the system in the bank, the loan use / settlement information from the first time point “a” to the second time point “b”, personal attribute information Bank account usage information can be calculated or obtained.
[0046]
In the purpose class information DB 42, for each credit customer, a flag indicating whether or not a foul transaction has occurred during the period from the second time “b” to the previous day of the current “c” is displayed in the purpose class information DB 42. It is stored as information. Also in this embodiment, as an example, the first time point “a” is October 1, 2000, the second time point “b” is April 1, 2001, and the current time point “c” is October 2001. Think of it as 1st of a month Moreover, unsatisfactory transaction means that overdue for 4 days or more has occurred 3 times or more or overdue for one month or more.
[0047]
In addition, the customer rating system is provided with an existing credit customer rating rule storage unit 44 and an uncredited customer rating rule storage unit 46 for storing the existing credit customer rating rules generated by the rule generation unit 24, and a rule application unit. A customer rating data storage unit 48 for storing customer rating data generated at 26 is provided. These storage units can also be realized by the storage device 18.
Further, the customer rating system has a management unit 50 whose function is realized mainly by the CPU 12. In response to an operator instruction given from the input device 14, the management unit 50 gives instructions to start various components and execute predetermined processing according to a program. Further, the image data including the processing result can be given to the display device 16.
[0048]
The operation of the customer rating system configured as described above will be described. In the present embodiment, the processing of the customer rating system generally includes a rule creation process by the rule generation unit 24 and a customer rating process by the rule application unit 26. FIG. 3 is a diagram showing an outline of the rule creation process. The credit customer rating rule generation unit 28 and the non-credit customer rating rule generation unit 30 have different DBs that are referred to for rule generation. For this reason, the information that is used as an explanatory variable also differs during rule generation.
[0049]
Also, both of these rule generation units 28 and 30 are from the first time “a” (October 1, 2000) to the day before the second time “b” (March 31, 2001). (For convenience, also referred to as the “first period”), based on the case of a customer who is continuously using a personal financial product and has not had any foul transactions within that period Generate personal credit rating rules.
[0050]
Therefore, as shown in FIG. 5, the existing credit customer rating rule generation unit 28 first uses a loan for a record relating to a customer that satisfies the condition (extraction condition) that no illegal transaction has occurred in the first period described above. Extracting from the settlement information DB 36, the personal attribute information DB 38, and the bank account usage information BD 40, and the purpose class information related to the customer is also extracted from the purpose class information DB 42 (step 501). In other words, the existing credit customer rating rule generation unit 28 uses all the information related to the customer that satisfies the extraction condition in the various DBs 36 to 42 that are the rule creation source data (see reference numeral 301 in FIG. 3).
[0051]
Next, the credit customer rating rule generation unit 28 uses the information in the extracted customer record as a training example to construct a poor transaction occurrence probability model (step 502). Here, in the first period, the total monthly loan balance, the average value of loan income (loan use / settlement information), date of birth, gender, date of initial account opening at the second time point “b” ( Personal attribute information), monthly liquidity deposit balance, fixed deposit total, ATM withdrawal, credit self-transfer total, salary transfer, automatic transfer, remittance, utility charge deduction, pension transfer in the first period above The average value of the amount (bank account usage information) is used as an explanatory variable, and the purpose class information (whether or not a bad deal occurs) in the second period is the purpose attribute.
[0052]
More specifically, as described above, the following steps 1 to 5 are executed.
[0053]
Step 1. First case set S ₀ Are assigned to the set S to be divided.
Step 2. If the number of cases in the set S to be divided is greater than a predetermined minimum case number criterion, all attributes X for the set S _i The divided gain and gain ratio are calculated for (i = 1, 2,..., M,..., M).
Step 3. For an attribute whose gain is greater than a predetermined minimum gain criterion, the attribute X having a gain ratio greater than average and a maximum gain ratio _m Is selected, and the set S to be divided is divided by using it.
Step 4. Subset T after splitting _j (J = 1, 2,..., K) are assigned to the set S to be divided, and the processes in steps 2 and 3 are repeated recursively.
[0054]
Algorithm stop condition: When the number of cases in all the division target sets S is less than the minimum number of cases criterion in Step 2, or in any division target set S in Step 3, no matter what division attribute is selected If no gain criterion is obtained, the algorithm is terminated.
[0055]
Thereafter, the existing credit customer rating rule generation unit 28 defines a personal credit rating rule for converting the estimated unfavorable transaction occurrence probability in the second period into a personal credit rating for each generated subset (step 503). ). Thereby, each subset is associated with the personal credit rating. As a specific method, after obtaining the cumulative distribution function of the unsatisfactory transaction occurrence probability estimated value, the illegal transaction occurrence probability estimated value corresponding to the accumulated appearance probability may be used as the threshold increment. As described above, the rating is “A” for the subset located in the top 5%, the rating “B” for the subset located in the next 10%, and the subset located in the next 20%. Rating “C”, rating “D” for the subset located in the middle 30%, rating “E” for the next 20% subset, rating for the next 10% subset A rating “G” is given to the subset located at “F” and the remaining 5%.
[0056]
Alternatively, the size of the unsatisfactory transaction occurrence probability estimated value itself is used as a threshold value. For example, a rating “A” is assigned to a subset having an estimated value of 0%, and a rating “ B ”, a rating“ C ”for a subset with an estimated value greater than 1% and 2% or less, a rating“ D ”for a subset with an estimated value greater than 2% and less than 3%, and a subset with a rating greater than 3% and less than 5% A rating “E” may be assigned to, and a rating “F” may be assigned to a subset greater than 5%.
[0057]
In step 503, a table associating each subset with a rating may be created. Or you may create the table which linked | related the unfavorable transaction probability estimated value corresponding to a subset, and the table which linked | related the unfavorable transaction probability estimated value and the rating threshold value.
The rules (table) obtained in this way are stored in the existing credit customer rating rule storage unit 44 (step 504). Thereby, the process by the existing credit customer rating rule generation unit 28 ends.
[0058]
Next, processing by the non-credit customer rating rule generation unit 30 will be described. The non-credit customer rating rule generation unit 30 also basically executes the processing shown in FIG. However, in step 501, the non-credit rating rule generation unit 30 records a record relating to a customer that satisfies the condition (extraction condition) that no illegal transaction has occurred in the first period, the personal attribute information DB 38, the bank account use information BD40. And the purpose class information related to the customer is extracted from the purpose class information DB. In other words, the rule creation source data (see 301 in FIG. 3) is different from the process by the credit customer rating rule generation unit 28 in that only the personal attribute information and the bank account use information are used as explanatory variables. It is the same that the purpose class information in the second period becomes the purpose attribute.
The personal credit rating rule (table) obtained by the non-credit customer rating rule generation unit 30 is stored in the non-credit customer rating rule storage unit 46. Thereby, the process by the non-credit customer rating rule generation unit 30 ends.
[0059]
As will be described later, the credit rating rule generation unit 28 and the non-credit rating rule generation unit 30 have a deviation within a predetermined range based on a comparison between the predicted value of the unsatisfactory transaction occurrence rate after the rule application and the actual value. When the management unit 50 determines that the credit limit has been exceeded, a personal credit rating rule (table) for an existing credit customer and a personal credit rating rule (table) for an uncredited customer are generated, respectively. It is stored in the credit customer rating rule storage unit 44 and the non-credit customer rating rule storage unit 46.
[0060]
Next, explanation will be given on the rating of customers by the rule application unit 26. FIG. 4 is a diagram showing an outline of the rule application process. The credit customer rating rule application unit 32 and the non-credit customer rating rule application unit 34 have different DBs to be referred to when applying the rules.
[0061]
Both of these rule application units 32 and 34 have a predetermined period (a period from the first time point “a” to the second time point “b” of the customer to be rated, that is, “ba”). Bank account usage information (see reference numerals 413 and 422 in FIG. 4) and customer attributes (reference numeral 412 in FIG. 4) from the time point back (for convenience, also referred to as “third time point”) to the previous day of the present time point. 421), and in the processing of the existing credit customer rating rule application unit 32, in addition to these, the loan use / settlement information (see reference numeral 411) is used as rule application data (see reference numerals 401 and 402). . The period from the third time point to the current day is also referred to as a “second period” for convenience. In the present specification, since the current time “c” is considered as October 1, 2001, the third time point coincides with the second time point “b” (April 1, 2001). Of course, if the current time is different, it goes without saying that the third time is also changed accordingly.
[0062]
As shown in FIG. 6, the credit customer rating rule application unit 32 searches the loan use / settlement information DB 36, the personal attribute information DB 38, and the bank account use information DB 40 for a customer record to be rated (step 601). And the total monthly loan balance in the second period, the average value of loan income (loan use / settlement information), date of birth at the current “c”, gender, date of first account opening (personal attribute information), Average monthly total of liquid deposits, total fixed deposits, ATM withdrawals, total credit self-payments, salary transfers, automatic transfers, remittances, utility charges deductions, and pension transfer amounts in the second period (banks) Account usage information) is extracted from the record (step 602).
[0063]
The loan use / settlement information, the bank account use information, and the current personal attribute information from the third time point to the previous day may be calculated or acquired from the bank system as preprocessing.
Next, the credit customer rating rule application unit 32 refers to the personal credit rating rule (table) stored in the credit customer rating rule storage unit 44 and sets the subset specified by the extracted various pieces of customer information. The associated rating is determined (step 603). The rating is associated with the customer and stored in the customer rating data storage unit 48 (step 604).
[0064]
Next, processing by the non-credit customer rating rule application unit 34 will be described. The non-credit customer rating rule application unit 34 also basically executes the processing shown in FIG. However, in step 601, the customer record to be rated is retrieved from the personal attribute information DB 38 and the bank account usage information DB 40. Also, in step 602, the date of birth, gender, date of initial account opening (personal attribute information) at the current time “c”, and the total amount of liquid deposits, total fixed deposits, and ATM withdrawals in the second period. The average value (bank account usage information) of money, credit self-transfer total, salary transfer, automatic transfer, remittance, utility charge withdrawal amount, and pension transfer amount is extracted from the record. The rating of the customer (uncredited customer) specified by the non-credit customer rating rule application unit 34 is also associated with the customer and stored in the customer rating data storage unit 48.
[0065]
The management unit 50 refers to the DBs 36 to 40 in response to the information specifying the customer and the rating instruction given by the operator operating the input device 14, and the customer is in a credit state. It is judged whether it is in an uncredited state or not. If the customer is a credit, the credit customer rating rule application unit 32 may be activated. On the other hand, if the customer is not a credit, the uncredit customer rating rule application unit 34 may be activated.
[0066]
Next, another process performed by the management unit 50 will be described. In the present embodiment, an example of a customer who is using a loan (existing credit customer) extracted under a condition in which no unfair transactions have occurred within a set period (first period) is used for rule generation. This is the range of data to be used. Therefore, the management unit 50 refers to the information in the loan use / settlement information DB 36 at a predetermined cycle (for example, from 3 months to a half year cycle), and calculates an actual value of the actual unsatisfactory transaction occurrence rate for each subset. Then, it is compared with the estimated illegal transaction occurrence probability value of each subset, and when the difference between these values exceeds a predetermined range, it is determined that a new personal credit rating rule should be generated. When such a determination is made, the management unit 50 activates the already-credit customer rating rule generation unit 28 and the non-credit customer rating rule generation unit 30 to execute processing.
[0067]
In addition, a customer who is not in a loan (uncredited customer) may become a credited customer by obtaining a loan when purchasing a house or a car. In consideration of this, the management unit 50 executes processing as shown in FIG. This process may be executed at a predetermined cycle (for example, once a month). The record for each customer in the personal attribute information DB 38 includes a flag indicating whether or not the loan is used and a loan use start date. Further, when the customer uses the loan, the management unit 50 sets the loan flag to “1” in the record associated with the customer in the personal attribute information DB 38 and records the loan use start date.
[0068]
The management unit 50 selects a record in the personal attribute information DB 38 (step 701), and determines whether or not the loan flag in the record is “1”, that is, the customer uses the loan, and from the uncredited customer. It is determined whether the credit customer has been changed (step 702). In the case of Yes in step 702, the management unit 50 refers to the loan use start date in the record and determines whether or not the second period has elapsed (step 703). When it is determined as Yes in step 703, the management unit 50 resets the loan flag in the record to “0” (step 704), and activates the existing credit customer rating rule application unit 32. In response to this, the existing credit customer rating rule application unit 32 assigns a rating to the customer and stores the data in the customer rating data storage unit 48. Thereby, the rating of this customer is updated from that based on the non-credit rating rule to that based on the credit rating rule. As described above, according to the present embodiment, when an uncredited customer becomes an already-credited customer, it is possible to give the customer a rating based on the loan performance by the processing of the management unit. The management unit 50 executes the processing in steps 701 to 705 for all records related to the customer to be processed (see step 706).
[0069]
According to the present embodiment, a rule for rating an existing credit customer is generated and given to the existing credit customer by applying the above rule, while another rule for rating an uncredited customer. For non-credit customers, the other rules are applied and given a rating. Therefore, it is possible to give a rating to a customer who does not use a loan, for example, a customer who intends to use a loan.
[0070]
Further, according to the present embodiment, when an uncredited customer becomes an already-credited customer, the rating as an already-credited customer can be updated after the elapse of a predetermined period. Thereby, it becomes possible to give a rating to a customer based on a loan performance.
The present invention is not limited to the above embodiments, and various modifications can be made within the scope of the invention described in the claims, and these are also included in the scope of the present invention. Needless to say.
[0071]
【The invention's effect】
According to the present invention, it is possible to assign a personal credit rating to a customer who has passed a certain period of time after opening a bank transaction account regardless of whether or not a personal financial product is used. As a result, it is possible to greatly contribute to the reduction in the risk of personal loan loss.
[0072]
Individual customer ratings for non-creditors serve as a kind of pre-screening. In financial institutions, personalized messages are output through ATMs to customers who are considered to be low-risk according to the obtained individual customer ratings, and loans are made almost immediately by only appropriate checks such as external credit information. Can be realized, and users of new financial products (loans) can be cultivated.
[0073]
In addition, based on the above-mentioned personal customer rating, it is possible to constantly monitor the “quality of receivables” at each branch office and the regional unit in which it is bundled. In particular, by observing changes over time in the quality of receivables, a substantial “early warning” can be made. In addition, when the deterioration of claims is foreseen, it can be used for portfolio management, such as conducting a campaign for a customer class that is considered to be of higher quality.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a hardware configuration of a computer system for realizing an individual credit rating according to the present embodiment.
FIG. 2 is a block diagram describing functions of the personal credit rating system according to the present embodiment.
FIG. 3 is a diagram showing an outline of rule creation processing according to the present embodiment.
FIG. 4 is a diagram showing an outline of rule application processing according to the present embodiment.
FIG. 5 is a flowchart showing processing executed by a customer rating rule generation unit according to the present embodiment.
FIG. 6 is a flowchart showing processing executed by a customer rating rule application unit according to the present embodiment.
FIG. 7 is a flowchart illustrating an example of processing executed by a management unit according to the present embodiment.
FIG. 8 is a diagram showing an example of acquiring a subset from a certain set S in the present embodiment.
[Explanation of symbols]
24 Rule generator
26 Rule Application Department
28 Credit customer rating rule generator
30 Non-credit customer rating rule generator
32 Credit customer rating rule application department
34 Non-credit customer rating rule application department
36 Loan use / settlement information DB
38 Personal attribute information DB
40 Bank Account Usage Information DB
42 Purpose class information DB
44 Credit customer rating rule storage
46 Non-credit customer rating rule storage
48 Customer rating data storage
50 Management Department

Claims (4)

Calculate a personal credit rating that indicates the degree of expected repayment after an individual financial product transaction as an order category, and assign a rating to an uncredited customer who owns a banking account but does not have a financial product transaction A personal credit rating system,
Non-definitions defined by the occurrence of continuous delinquencies and / or bad debts for a specified month or more applicable to individual customers based on training examples extracted from customer cases in which financial transactions for individuals were conducted in the past. A rule generation means for generating a rule for calculating a statistical value of a good transaction occurrence probability, wherein a first conversion rule to be applied to a credit customer who is an individual customer with a transaction of a financial product, and a bank transaction account A rule generation means for generating a second conversion rule to be applied to an uncredited customer who is an individual customer who owns but does not have a transaction of a financial product;
First rule storage means for storing the first conversion rule;
Second rule storage means for storing the second conversion rule;
Rules stored in the first rule storage means or the second rule storage means are passed to a personal customer who has passed a certain period of time after opening a transaction account with a bank and for which some account transaction has been performed. A rule applying means to apply, and when the individual customer is a credit customer, the first conversion rule stored in the first rule storage means is applied to give a rating to the individual customer On the other hand, when the individual customer is a non-credit customer, the rule applying means for applying a second conversion rule stored in the second rule storage means and assigning a rating to the individual customer. When,
A rating data storage means for storing the assigned rating in association with an individual customer;
The rule generation means is a poor transaction occurrence probability model for estimating a bad transaction occurrence probability from a training example,
Satisfies the conditions in which financial products for individuals are in use in the first period from past time point “a” to time point “b” (where a <b) and no foul transactions have occurred within that period. Training examples to be used as common training examples,
Account usage information including liquid deposit balance, fixed deposit balance, number of ATM withdrawals in the first period from time “a” to time “b”, and individual basic attributes including age and gender at time “b” , And use / settlement information of the personal financial product in the first period from the time point “a” to the time point “b” as explanatory variables, and from the time point “b” to the current time point “c” (where b < c) a first unsatisfactory transaction occurrence probability estimation model using whether or not unsatisfactory transactions have occurred in the second period until c),
The account usage information in the first period from the time point “a” to the time point “b” and the personal basic attribute at the time point “b” are explanatory variables, and from the time point “b” to the time point “c”. Generating a second unsatisfactory transaction occurrence probability estimation model using the presence or absence of the unsatisfactory transaction occurrence in the second period as an objective variable,
The rule generation means generates a first conversion rule from an unfavorable transaction occurrence rate on a training example estimated by the first unfavorable transaction occurrence probability model to an arbitrary number of order scale category values, And generating a second conversion rule from an unfavorable transaction occurrence rate on the training example estimated by the second unfavorable transaction occurrence probability model to an arbitrary number of order scale category values, and
Among the individual customers whose at least the first period has elapsed since the opening of the trading account, the rule applying means is a credit customer, and the current customer “c” from the third time “d” that dates back to the first time from the current time. ”(However, d <c), in the second period corresponding to the first period, the account usage information of the credit customer, the personal basic attribute at the current“ c ”, and the second period Based on the use / settlement information of the financial product, refer to the first conversion rule stored in the first rule storage means, and assign a rating corresponding to the estimated unfavorable transaction occurrence probability,
For the non-credit customer, the account usage information of the non-credit customer in the second period corresponding to the first period from the third time point “d” to the current time “c”, and the current time “c” Based on the personal basic attributes in the above, the second conversion rule stored in the second rule storage means is referred to, and a rating corresponding to the unfavorable transaction occurrence probability value is assigned. Personal credit rating system.   Further, when the individual customer is a non-credit customer, the rule application means applies the second conversion rule to give the individual customer a preliminary rating, and the individual customer Management means for causing the rule applying means to apply the first conversion rule and to give a rating based on a loan performance when a period corresponding to the first period has elapsed after starting the transaction of The personal credit rating system according to claim 1, further comprising: Calculate a personal credit rating that indicates the degree of expected repayment after an individual financial product transaction as an order category, and assign a rating to an uncredited customer who owns a banking account but does not have a financial product transaction A computer-readable program for operating a computer for
Non-definitions defined by the occurrence of continuous delinquencies and / or bad debts for a specified month or more applicable to individual customers based on training examples extracted from customer cases in which financial transactions for individuals were conducted in the past. A rule generation step for generating a rule for calculating a statistical value of the probability of occurrence of a good transaction, generating a first conversion rule to be applied to a credit customer who is an individual customer having a transaction of a financial product, Storing one conversion rule in the first rule storage means, and generating a second conversion rule to be applied to an uncredited customer who is an individual customer who owns a bank transaction account but does not have a transaction of a financial product, A rule generation step of storing the second conversion rule in the second rule storage means;
Rules stored in the first rule storage means or the second rule storage means are passed to a personal customer who has passed a certain period of time after opening a transaction account with a bank and for which some account transaction has been performed. A rule applying step to apply, and when the individual customer is a credit customer, the first conversion rule stored in the first rule storage means is applied to assign a rating to the individual customer The rating is stored in the rating data storage means in association with the individual customer. On the other hand, when the individual customer is an uncredited customer, the second conversion stored in the second rule storage means is stored. Applying a rule to assign a rating to the individual customer, and executing a rule applying step of storing the rating in a rating data storage unit in association with the individual customer in the computer. ,
In the rule generation step,
A poor transaction occurrence probability model that estimates the probability of bad transaction occurrence from training examples,
Satisfies the conditions in which financial products for individuals are in use in the first period from past time point “a” to time point “b” (where a <b) and no foul transactions have occurred within that period. Training examples to be used as common training examples,
Account usage information including liquid deposit balance, fixed deposit balance, number of ATM withdrawals in the first period from time “a” to time “b”, and individual basic attributes including age and gender at time “b” , And use / settlement information of the personal financial product in the first period from the time point “a” to the time point “b” as explanatory variables, and from the time point “b” to the current time point “c” (where b < c) a first unsatisfactory transaction occurrence probability estimation model using whether or not unsatisfactory transactions have occurred in the second period until c),
The account usage information in the first period from the time point “a” to the time point “b” and the personal basic attribute at the time point “b” are explanatory variables, and from the time point “b” to the time point “c”. Generating a second unsatisfactory transaction occurrence probability estimation model using the presence or absence of the unsatisfactory transaction occurrence in the second period as an objective variable;
Generating a first conversion rule from an unsatisfactory transaction occurrence rate on a training case estimated by the first unsatisfactory transaction occurrence probability model to an arbitrary number of order scale category values; and Generating a second conversion rule from an unfavorable transaction occurrence rate on the training case estimated by the unfavorable transaction occurrence probability model to an arbitrary number of order scale category values;
In the rule application step,
Among the individual customers who have passed at least the first period from the opening of the trading account, the credit customers are already credited customers from the third time point “d” retroactive to the first time point to the current time point “c” (where d < Up to c) in the second period corresponding to the first period, the account usage information of the existing credit customer, the personal basic attributes at the current “c”, and the use / settlement of financial products in the second period Based on the information, with reference to the first conversion rule stored in the first rule storage means, to give a rating corresponding to the unsatisfactory transaction occurrence probability estimated value,
For the non-credit customer, the account usage information of the non-credit customer in the second period corresponding to the first period from the third time point “d” to the current time “c”, and the current time “c” And a step of assigning a rating corresponding to the unfavorable transaction occurrence probability value with reference to the second conversion rule stored in the second rule storage means based on the individual basic attribute in Credit rating program. Further, when the individual customer is a non-credit customer, the rule application unit causes the second conversion rule to be applied, and the computer is operated so as to give a preliminary rating to the individual customer. And,
Based on the loan performance, the rule application means applies the first conversion rule when a period corresponding to the first period elapses after the individual customer starts trading the financial product. The personal credit rating program according to claim 3, wherein the computer is operated to give a rating.