JP2003316950A - Credit evaluation system and credit evaluation method - Google Patents

Abstract

(57) [Summary] [Problem] To provide a credit evaluation system and a credit evaluation method capable of appropriately performing a credit evaluation and making a comprehensive judgment. For an index for evaluating the creditworthiness of a debtor, an index information file for storing past index data including default destination data and normal destination data is provided. First, a division value for dividing default destination data and normal destination data is input. The ratio determined as the default destination among the data of the default destination is calculated as signal information. From the data of the normal destination, the ratio determined as the default destination is calculated as noise information. Calculate the ratio of signal information to noise information.
Further, it is determined whether the divided value belongs to the default side or the normal destination side, and the reciprocal of the ratio of the signal information to the noise information is given to the default side as a point.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to a credit evaluation system and a credit evaluation method for obligors.

[0002]

2. Description of the Related Art Currently, various financial variables (sales, tangible fixed capital, etc.) and their combined financial indicators (sales ordinary income ratio, equity capital) are used as a means of rating or scoring debtors at financial institutions. The mainstream method is to judge the possibility of default of the target company by using statistical methods such as discriminant analysis and logistic analysis using ratios. These statistical methods assume that the same financial variables are obtained from all companies analyzed.

However, in general, in medium-sized enterprises, small and medium-sized enterprises, small enterprises and individuals, the financial statements are often poorly prepared, and there is a high possibility that an abnormal value will be included in the statistical data due to statistically settled settlement. . Therefore, when analyzing the data of these companies and individuals, the financial indicators necessary for the analysis are often not available. In this case, it is necessary to either give up using the financial index or use a certain other numerical value as a substitute, but this may cause a problem in analysis from the viewpoint of data accuracy.

In addition, existing statistical methods are premised on the use of quantitative numerical values such as financial indicators. Incorporation of qualitative information into the model requires analytical measures such as setting dummy variables.

Furthermore, in the existing statistical method, when the correlation between financial statements is strong, unfavorable influences such as the sign of coefficients being reversed due to the occurrence of collinearity are likely to occur, and therefore extreme caution is required. There is.

[0006]

As described above, the conventional credit evaluation system has qualitative data, quantitative data,
Financial data with many missing values, financial data with few missing values,
We could not comprehensively judge data with many outliers. That is, in the conventional technology, the creditworthiness is measured by using only a part of the information of the debtor. Moreover, a credit evaluation system and a credit evaluation method capable of calculating a comprehensive evaluation within a certain economic concept have not been established.

The present invention has been made in order to solve such problems, and can appropriately perform credit evaluation, and particularly qualitative data, quantitative data, financial data with many missing values, and missing data. It is an object of the present invention to provide a credit evaluation system and a credit evaluation method capable of comprehensively judging financial data having a small value, data having a large outlier, and the like.

[0008]

A credit evaluation system according to the present invention comprises a first processing means (for example, a processing means by the primary discriminant model 100) and a second processing means (for example, a processing means).
A credit evaluation system for evaluating creditworthiness of a debtor, comprising: a processing means based on a secondary scoring model 200, wherein the first processing means processes the data of the debtor into a plurality of data groups according to the degree of creditworthiness. The second processing means relates to an index for evaluating the creditworthiness of the obligor, and the first data distribution leading to a low creditworthiness state (for example, default) and the creditworthiness. An index information file that stores past index data including the second data distribution that has reached a high state, and a division value that divides each of the first data distribution and the second data distribution acquired from the index information file. And a means for dividing the first data distribution based on the division value and calculating the ratio of the low credit side of the first data distribution as signal information. Based on the divided value, the second
Means for dividing the data distribution of the second data distribution and calculating the ratio of the side of low reliability in the entire second data distribution as noise information, and discrimination capability information (for example, S / N ratio) based on the ratio of the signal information and the noise information. A means for calculating, based on the divided value, a means for determining whether the evaluation target belongs to the side with high credit rating or the side with low credit rating, and a point is given based on the determination result of the determination means, and the score With respect to the above, a means for reflecting the coefficient based on the discrimination capability information is provided.

Further, it is preferable to provide a means for giving a score for a plurality of indexes, obtaining a specific total value, and dividing the score by the number of the indexes to calculate a score per unit.

Preferably, the index for evaluating the creditworthiness of the debtor includes a profit-related index.

On the other hand, another credit evaluation system according to the present invention relates to an index for evaluating the creditworthiness of an obligor, and the first data distribution that has reached a low credit status and the first data distribution that has reached a high credit status. An index information file which stores past index data including two data distributions, and means for determining a division value for dividing each of the first data distribution and the second data distribution acquired from the index information file, A means for dividing the first data distribution based on the division value and calculating a ratio of the low credibility side in the entire first data distribution as signal information; and based on the division value,
A means for dividing the second data distribution and calculating a ratio of low reliability in the entire second data distribution as noise information; and a means for calculating discrimination ability information based on the ratio of the signal information and the noise information. , A means for judging whether the evaluation object belongs to the side with high credibility or the side with low credibility based on the division value, and a score is given based on the judgment result of the judging means, and the judgment is made for the score. A credit evaluation system comprising means for reflecting a coefficient based on capability information, characterized by comprising selection means for selecting an index for evaluating the creditworthiness of the obligor,
The selecting means calculates the discriminating ability information for each of the indicators from the indicator information file, and selects the first indicator having the highest discriminating ability based on the discriminating ability information;
And a means for determining whether or not the explanatory power to the default sample has been improved based on the second index and the first index after selecting the second index having high discriminating ability. In this case, a means for further selecting an index is continued, and continues until the explanation power for the default sample is no longer improved, and if the explanation power is not improved, the index selection process is stopped.

Here, in a preferred embodiment, the obligor is a small and medium-sized enterprise.

A credit evaluation method according to the present invention is a credit evaluation method comprising a first processing step and a second processing step for evaluating creditworthiness of a debtor, wherein the first processing step is a control step. The program classifies the debtor's data into a plurality of data groups according to the degree of credit,
In the processing step of, regarding the index for evaluating the creditworthiness of the obligor, the past index data including the first data distribution having a low credit quality and the second data distribution having a high credit quality are collected. Determining a division value for dividing each of the first data distribution and the second data distribution acquired from the stored index information file by a control program, and dividing the first data distribution based on the division value Then, a step of calculating the ratio of the low credibility side in the entire first data distribution as signal information by a control program, and dividing the second data distribution based on the division value, and dividing the second data distribution in the entire second data distribution. The step of calculating the ratio of the low credibility side as noise information by the control program, and the determination based on the ratio of the signal information and the noise information Calculating the force information by a control program, determining by the control program whether the evaluation target belongs to the side with high credibility or the side with low credibility based on the division value, and the determination of the determining means by the control program The method has a step of giving a score based on the result and reflecting a coefficient based on the discrimination ability information on the score.

Another credit evaluation method according to the present invention relates to an index for evaluating the creditworthiness of an obligor, and a first data distribution that has a low credit quality and a second data distribution that has a high credit quality. Determining a division value for dividing each of the first data distribution and the second data distribution acquired from the index information file storing the past index data including the data distribution by the control program, and based on the division value Dividing the first data distribution and calculating a ratio of the low credibility side of the entire first data distribution as signal information by a control program; dividing the second data distribution based on the division value. The second concerned
A step of calculating the ratio on the low credibility side of the entire data distribution of as a noise information by the control program,
A step of calculating discrimination ability information by a control program based on the ratio of the signal information and the noise information, and a control program determines whether the evaluation target belongs to the side with high credibility or the side with low credibility based on the division value. A credit evaluation method comprising: a step, and a step of giving a score based on the judgment result of the judgment means and reflecting a coefficient based on the discrimination ability information to the score by a control program, It is characterized by including a selection step of selecting an index for evaluating creditworthiness by a control program, and this selection step calculates discriminating ability information for each indicator from the index information file, and based on this discriminating ability information, A step of selecting a first index having high discriminating ability, and a step of selecting the first index having high discriminating ability following the first index. Selecting the index of, and determining whether the explanatory power for the default sample is improved based on the second index and the first index; and if it is improved, selecting the subsequent index. The step is continued until the explanatory power for the default sample is not improved, and if the explanatory power is not improved, the index selection process is stopped.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1 of the Invention FIG. 1 shows a conceptual diagram of an evaluation model in the credit evaluation system according to the present invention. As shown in FIG. 1, the model according to the present invention includes a primary discriminant model 100 and a secondary scoring model 200, through which a final scoring model 300 is constructed.

Here, the primary discriminant model 100 is a model for dividing the target enterprise group into excellent and non-excellent. Two
The next scoring model 200 is a model for calculating the default probability of each company for each group subdivided into the result of the primary discriminant model and the type of industry. In the final scoring model 300, the models 100 and 200 are integrated, and for example, scoring for each industry type, scale, and group is integrated.

Further, the primary discriminant model 100 will be described in detail. The primary discriminant model is a model that separates SMEs that are judged to be excellent from financial data at first glance and non-excellent SMEs that are expected to have severe management. Primary discrimination model 10
The reason for providing 0 is as follows.

(1) This is because there is a possibility that there is a difference in the processes leading to bankruptcy, such as the settlement of accounts, etc., between excellent small and medium-sized enterprises and non-excellent enterprises.

(2) It is generally known that the default rate increases exponentially as the credit quality deteriorates.
This exponential default rate is divided into good and bad by the quality of the debtor, and is approximated by two straight lines. As a result, it is possible to improve the fitting of the estimation result.

For the above reasons, the first-order discriminant model is used to divide into a group of relatively good companies and a group of not good companies. Here, for the division of the excellent company and the non-excellent company, for example, the rating of the rating company or the evaluation point may be used. For example, a model for obtaining the probability that the rating is worse than BB + is created and the result is used. In addition, in this example, although it is divided into two, it is not limited to this and may be divided into three or more.

Next, the secondary scoring model 200 will be described in detail. The secondary scoring model 200 is a model for obtaining the theoretical default probability of each obligor. Inevitable with traditional statistical models,
Build a model that minimizes the effects of outliers and missing values. In addition, as many index values as possible are targeted for evaluation. In order to realize this, in this model, one of the explanatory variables includes the total default index.

FIG. 2 shows the structure of the secondary scoring model 200. First, a comprehensive default index is calculated from various financial index values. Then, using this comprehensive default index as one of the explanatory variables, along with other financial index values,
Calculate the default probability using a given logistic model.

Here, the signal model for deriving the total default index will be described. First, for example, it is assumed that the distribution of a specific index for evaluating credit risk such as an equity ratio is given in FIG. This distribution is, for example,
Given based on sample data of a large number of obligors, distribution data for default destinations and distribution data for normal destinations are shown. Here, the “default destination” refers to a sample that has actually reached the default at the reference time, and the “normal destination” refers to a sample that has not actually reached the default at the reference time. In FIG. 3, the horizontal axis is the equity ratio, and the vertical axis is the number of samples. The capital adequacy ratio decreases toward the left, that is, there is a general tendency to default. First, the sample is divided into two at appropriate points. For example, division is performed at the portion indicated by the dotted line in FIG.

Here, the left side of the dotted line is the "default area". The area on the right side of the dotted line is the “normal area”. Then, the number of samples in the default area in the actual default sample is A. In addition, the number of samples in the default area of the actual normal samples is set to B. Let C be the number of samples in the normal region with the actual default samples. Further, the number of samples in the normal region which is the actual normal destination sample is D. These relationships are summarized in Table 1 below.

[0025]

[Table 1] When the sample is divided into two parts by this dividing line, the ratio of the number of samples in the default region to the entire default destination sample, A / (A + C), is called the signal (S). This signal (S) is the ratio of correctly discriminating the actually defaulted sample as the default destination.

Further, B / (B + D), which is the ratio of the number of samples in the default area to the normal destination sample, is called noise (N). The noise (N) is a ratio of the normal destination which is not actually defaulted but is mistakenly determined to be the default destination.

The transition of signal and noise is shown in FIG. The vertical axis shows the ratio, and the horizontal axis shows the equity ratio. The N / S ratio at the division points in FIG. 4 is obtained. The N / S ratio is calculated as noise (N) / signal (S). The higher the discriminative ability between the normal sample and the default sample at the division point, the smaller the noise and the larger the signal, so N / S
The ratio is a small value. Conversely, the lower the discrimination ability is, the more N / S
The ratio increases. When the N / S ratio exceeds 1, it indicates that there is no discriminating ability at the division point. A dividing point where the calculated N / S ratio has a minimum value, that is, a minimum misjudgment point is obtained, and the dividing point is determined. Regarding the setting of the division points, a method other than this method will be described in other embodiments of the invention.

When the dividing points are determined in this way, the credit risk evaluation for a specific obligor is then performed by scoring each index. In the preferred embodiment, all the samples on the left side of the dividing point have 1 point, and the samples on the right side of the dividing point have 0 point. Specifically, a debtor having a capital adequacy ratio lower than the capital adequacy ratio at the split point is given one point. In addition, 0 points are given to obligors who have a higher capital adequacy ratio than the split point. In these, N / at the determined dividing point
The number obtained by multiplying the reciprocal of the S ratio is used as a score. The higher the discriminating ability, that is, the lower the N / S ratio, the higher the score of the sample on the left side of the division point. This is to weight the score so that the credit risk becomes higher when it is determined that the possibility of default is high by a method having high determination ability.

Similarly, scores are calculated for other indicators other than the capital adequacy ratio and they are all summed. For example, 5
Scores are calculated for indices of 0 or more and summed. This total value is called the total default index. The maximum value of the total default index is equal to the sum of the reciprocals of the N / S ratio when all the indicators emit a signal of 1.

However, in some samples, some indexes cannot be calculated due to insufficient data. In this case, the overall default index cannot be directly compared between samples. Therefore, in the credit evaluation system according to the present invention, a unit score (unit total default index) is calculated by dividing the total default index by the number of indexes used when calculating the total default index (that is, the number given 0 or 1). is doing. This would allow a direct comparison of credit risk among obligors. The larger this unit comprehensive default index, the higher the probability of default.

The credit evaluation system according to the present invention using such a unit integrated default index is extremely useful for evaluating credit risk of small and medium-sized enterprises. In the usual credit risk analysis, when the reliability of data is poor and there are many missing values for which financial indicators are not sufficiently aligned, such as financial statements of small and medium-sized enterprises, credit risk analysis becomes difficult. In this case, in the credit evaluation system according to the present invention, each financial index value is divided into a group of companies belonging to a relatively good side and a group of companies belonging to a bad side, and 1 is assigned to the companies belonging to the bad side.
Since the creditworthiness of companies is evaluated by assigning 0 to the companies that belong to the good side and uniting the various financial index values, the creditworthiness of SMEs can be evaluated comprehensively.

With respect to the credit evaluation system according to the present invention, the parameters of the unit comprehensive default index were actually calculated using a large-scale financial database that is essential for highly accurate calculation of the unit comprehensive default index. This financial database has about 750,000 obligors (the default number is about 50,000) and the number of financial statements is about 2250,000 (the default number is about 130,000) for corporations. Has about 240,000 debtors (of which about 2 are defaults)
The number of financial statements is about 500,000 (of which the default number is about 3
10,000). As a result, conventional credit rating systems
It is rarely used because it has low correlation with default, operating profit margin on sales, business profit after interest payment per person,
It was confirmed that profit-related indicators such as the total profit ratio for the current period and depreciation rates can also be evaluated by the credit evaluation system according to the present invention.

Next, the configuration and processing example of the credit evaluation system according to the first embodiment of the present invention will be described. Figure 5
Shows the configuration example of the credit evaluation system.

This credit evaluation system can be realized by a computer such as a dedicated computer or a personal computer (PC). However, the computer does not need to be physically single, and when performing distributed processing,
There may be a plurality. This credit evaluation system includes, for example, a CPU 101, an input / output device 102, a control program 103, an index information file 104, a model information file 105, a debtor information file 106, an evaluation information file 107, and a bus 108 connecting them. .

The CPU 101 is a central control unit that executes various processes in the credit evaluation system based on the control program 103. The input / output device 102 is, for example,
It is composed of an input means such as a keyboard and a mouse and an output means such as a CRT, a liquid crystal display and a printer.

The control program 103, the index information file 104, the model information file 105, the debtor information file 106, and the evaluation information file 107 are ROM and RA.
It is stored in an internal or external storage means such as M or a hard disk. These programs and databases may be acquired from a remote database or the like via a communication network such as the Internet.

The control program 103 is a program for causing the CPU 101 to execute processing relating to this system.

The index information file 104 is a file for storing index information relating to indices related to credit risk such as equity ratio, creditworthiness of parent company, presence / absence of owner assets. For example, the index information includes information in which index data for a large number of past samples for obtaining the distribution information of FIG. 3 and data as to whether each sample has defaulted at the reference time point are associated with each other.

The model information file 105 is a file for storing information necessary for calculating the N / S ratio by the above-mentioned signal model. For example, this information
Information about the determined dividing points, a data replenishment rate, a reference N / S ratio, and the like are included.

The debtor information file 106 is a file for storing information about the debtor who is going to evaluate credit risk. For example, this information includes data corresponding to each index of the obligor and information specifying an index to be included in calculating the total default index.

The evaluation information file 107 is a file for storing information indicating the evaluation result of credit risk. For example, this information includes the comprehensive default index and the unit comprehensive default index calculated by the method described above.

Next, the flowchart shown in FIG. 6 and FIG.
An example of the processing of the credit evaluation system according to the present invention will be described based on the screen examples shown in FIGS.

First, the processing relating to the primary discriminant model 100 is performed. For example, the control program 103 displays the number of debtors to be evaluated from the debtor information file 106 on the screen based on the operator's instruction. The operator inputs rating information and evaluation points that serve as a reference for dividing the group. The control program 103 classifies the debtor information in the debtor information file 106 according to the input of the reference, and displays the number of debtors belonging to each group.

Subsequently, the processing relating to the secondary scoring model 200 is performed. This processing is performed by the primary discrimination model 100.
It is executed according to a different model for each group divided by. First, for example, a specific index is selected from the indices used to evaluate credit risk (S10).
1). For example, the equity ratio is selected. In particular,
Selection of a specific index is executed based on the screen display example shown in FIG. 7. In the index item column of FIG. 7, indexes used in advance in evaluating credit risk are listed. The operator assigns a group number to each index item to perform grouping and select a specific index. For example, assuming that the item assigned with 1 as the group number is the selected index, the comprehensive default index and the unit comprehensive default index corresponding to the group number 1 are calculated. If the input completion button is clicked when the group number has been input for each index item, the information is stored in the model information file 105 by the control program 103.

You may make it perform the process which further narrows down the selected index. For example, as shown in the screen display example of FIG. 8, the target criteria for the limiting elements of the index such as the data replenishment rate and the N / S ratio are input. Control program 103
On the basis of the target criteria input in this way, the selected indexes are narrowed down, and information indicating that is not added to the indexes that are not targeted, and is not targeted for the processing in step S102 and thereafter.

Further, at this point of time, an instruction for counting processing may be given. For example, as shown in the screen display example of FIG. 9, information such as an index type, the number of divisions, a division standard, etc. is input as the processing item and the instruction content therefor. In the index type, select either the unit comprehensive index or the comprehensive index.

Next, the N / S ratio is calculated (S1
02). First, an appropriate division point for dividing the sample into two is input. For example, the distribution points as shown in FIG. 3 may be displayed and the division points may be input by designating an arbitrary position. Further, the distribution may be represented by a numerical value, and the division points may be entered by a numerical value. Then, the signal (S) is calculated by calculating A / (A + C). Further, the noise (N) is calculated by calculating B / (B + D). Then, the N / S ratio at the division point is obtained. The N / S ratio is calculated as noise (N) / signal (S). The calculation of these signal (S), noise (N) and N / S ratio is realized by appropriately storing in a storage means such as a buffer and reading out. That is, the signal (S) is calculated, and the calculation result is stored in the storage means such as a buffer. Calculate the noise (N),
The calculation result is saved in a storage means such as a buffer. The calculation result stored in this way is read, the N / S ratio is calculated, and the result is stored again in the buffer or the model information file 105. A division point at which the calculated N / S ratio has the lowest value is obtained, and the division point is determined. Information about the division points is stored in the model information file 105 together with the N / S ratio information.

When the division points are determined in this manner, the credit risk evaluation relating to a specific obligor is then performed by scoring each index (S103). Specifically, we compare the equity ratio at the split point with the equity ratio of the obligor to be evaluated. The equity ratio concerning the division point is read from the model information file 105.
The equity ratio to be evaluated is the debtor information file 106.
Read more. When the control program 103 determines that the equity ratio of the debtor to be evaluated is lower than the equity ratio of the dividing point, the reciprocal of the N / S ratio is calculated for the debtor. Then, the calculation result is stored in the evaluation information file 107. Further, 0 points are stored in the evaluation information file 107 for obligors having a higher equity ratio than the division point.

When the score for the selected index is calculated,
The control program 103 determines whether or not the score calculation has been completed for all the indexes (S104).
When the control program 103 determines that the score calculation has not been completed for all the indexes, another index other than the equity ratio is selected (S101), and step S10 is performed.
2. The processes of step S103 and step S104 are repeated.

When the scores have been calculated for all the indexes, the total value of the calculated scores is calculated and stored in the evaluation information file 107 as the total default index (S).
105). Further, a calculation is performed in which the total default index stored in the evaluation information file 107 is divided by the number of indices calculated in calculating the total default index,
A unit integrated default index is calculated (S106). This unit comprehensive default index is also stored in the evaluation information file 107.

As described above, according to the credit evaluation system and the credit evaluation method according to the embodiment of the present invention,
Information such as large companies, small and medium-sized enterprises, micro enterprises, individual enterprises, etc., where there is a difference in the amount of qualitative / quantitative data available, or even in the same business category, data with a large number of missing values or data with few missing values In cases where there is a difference in the quality of the debt, it becomes possible to measure the credit risk of the obligor by comprehensively evaluating various information.

Embodiment 2 of the Invention The second embodiment of the present invention relates to index value selection processing used in the credit evaluation system according to the present invention.

FIG. 10 shows the flow of the index value selection processing. First, a target index is primarily selected from a plurality of indexes prepared in advance (S201). Specifically, in the credit evaluation system, the index information is read from the index information file 104 according to the control program 103, and the input / output device 1
The index information is displayed on the display device 02. The operator selects specific index information from the displayed index information. For example, the index information is selected by checking the check box of the index information displayed on the screen using the keyboard. The selected index information is stored in the model information file 105.

Next, the credit evaluation system calculates the minimum misjudgment point for each of the primarily selected indexes (S202). First, an appropriate division point for dividing the sample into two is input. And the signal (S) is A /
It is calculated by calculating (A + C). Further, the noise (N) is calculated by calculating B / (B + D). Then, the N / S ratio at the division point is obtained.
The N / S ratio is calculated as noise (N) / signal (S). These signals (S), noise (N) and N / S
When calculating the ratio, it is realized by appropriately storing it in a storage means such as a buffer and reading it. That is, the signal (S) is calculated, and the calculation result is stored in the storage means such as a buffer. The noise (N) is calculated, and the calculation result is stored in a storage unit such as a buffer. The calculation result stored in this way is read, the N / S ratio is calculated, and the result is stored again in the buffer or the model information file 105. A division point where the calculated N / S ratio has the lowest value is obtained, and the division point is determined (S203). The method of determining the division points is not limited to this, and other methods may be used. The information regarding the division points is stored in the model information file 105 in association with each index together with the N / S ratio information. Here, the division points may be generated by the control program 103 automatically even if they are not input, and the N / S ratio may be calculated for each division point at any time.

Next, the control program 103, based on the N / S ratio information stored in the model information file 105,
Select the index with the lowest N / S ratio (S20
4). Then, the control program 103 calculates a unit comprehensive default index for the selected index (S20).
5). Specifically, S10 of the flowchart shown in FIG.
The unit integrated default index is calculated by executing the processing shown in 2 to S106 and its description.

Thereafter, the control program 103 confirms the explanatory power for the default sample based on the unit comprehensive default index calculated (S20).
6). This explanatory power is confirmed by, for example, extracting 20% from the bad unit comprehensive default index for a large number of samples, and calculating to what extent the actually defaulted sample is included.

Next, the control program 103 selects the index having the second lowest N / S ratio (S207). Then, the control program 103 calculates the unit comprehensive default index based on this index and the index selected in S204 (S208). Then, based on this calculation result, the control program 103 confirms whether or not the explanatory power for the default sample is improved as compared with the previous calculation result (S208). Control program 103
If it is determined that is not improved (S210), the addition of the index is stopped (S211). In the case of stopping, the index selected so far is output by the input / output device 102 (S212).

When it is determined that the control program 103 has been improved (S210), the third index having the lowest N / S ratio is selected (S207), and the unit total default index is calculated by adding it (S208). The explanatory power of the default sample is compared (S209). This series of operations is continued until the explanatory value for the default sample is not improved even if the index value is added.

Here, regarding the index selection process, the index value is classified into several subcategories from the viewpoints of growth and profitability, and N is selected from each subcategory.
You may make it select some indices with a low / S ratio. For example, when there are 100 indices, they are divided into 20 sub-categories and one index is selected from each sub-category.

Further, the correlation coefficient of the index value to be used may be investigated in advance, and one of the indexes having a high correlation may be excluded.

Other Embodiments of the Invention. In the above example, the case where the dividing point is one, that is, the whole is divided into two, has been described, but it is also possible to divide into two or more places. In that case, the N / S ratio at each division point is calculated. In this case, the given score is not limited to 1 or 0, and can be set in any of three levels such as 0, 1, 2 in the case of three divisions. A debtor who wants to assess credit risk
It is determined which one of the three divided regions it corresponds to. Then, 0, 1, and
A value obtained by multiplying any of 2 by the reciprocal of the N / S ratio, that is,
A score (score) of 0, S / N or 2S / N is used. In the case of four divisions, it is preferable to use four-step numerical values, in the case of five divisions, five-step numerical values, and in the case of n-division, n-step numerical values. These calculation processes are executed based on the control program 103, and the final evaluation result is stored in the evaluation information file 107.

In the above example, the unit comprehensive default index is calculated in order to avoid the influence of the index having no numerical value, but other numerical value is used as a substitute for the numerical value which does not exist. Instead, it is possible to use the total default index as it is, for example, 0.5 point between 0 and 1 can be used as a substitute value.

Further, the division points may be such that the number of samples on the side of high credibility and the number of samples on the side of low credibility are equal. That is, the sample is divided into two at the point of halving it. In this case, 1 point is given to the sample on the side of low credibility.

It is also possible to divide the whole into 5 by 20% tiles, and give 1 point, 2 points, 3 points, 4 points and 5 points in order from the region of low credibility.

It is also possible to divide the whole into n and give the same score, for example, 1 point to each.

Regarding the division points, the optimum division points may be searched for by using Tree analysis such as CART method and QUEST method which are data mining methods. Where C
The ART method is a classification and regression binary tree, in which the data is 2
Divide into two subsets so that the cases in each subset are more homogeneous than the previous subset. Also, Q
The UEST method is a binary tree growth algorithm that handles variable selection and branch point selection separately.

The setting of N / S, which is indispensable for creating the unit comprehensive default index, may be as follows. That is, there are a plurality of division points, and 0 is set for each area.
If points or points including one point and other numerical values are given, the N / S ratio is made constant at all division points. In other words, when divided by a plurality of division points, different points are given to the respective areas. In that case, it is not necessary to multiply each score by the N / S ratio, and the score may be used as it is.

When there are a plurality of division points and the same score is given to each area, the N / S ratio may be given individually to the division points given the same score. . By doing so, it becomes possible to cause a difference in each region depending on the N / S ratio.

Further, division points may be given at all default sample points, and the N / S ratio may be calculated at each division point. That is, when there are 1000 samples, the N / S ratio for 1000 samples is calculated. A line approximated by a curve that smoothly connects these N / S ratios is used to calculate the N / S ratio at any point of the index value.

In the above example, continuous data is assumed, but it is possible to incorporate qualitative and discrete data. For example, when there is data indicating “owned house” or “rented house”, this index can be incorporated into the model by calculating the N / S ratio by the following formula. If the rented house is the default area, the default sample for the rented house is A, the normal sample for the rented house is B, the sample for the rented house is C, and the normal sample for the rented house is D. To calculate. That is, A / (A
The signal (S) is obtained by calculating + C), and B
The noise (N) is obtained by calculating / (B + D). Then, from these values, N / at a specific dividing point
Calculate the S ratio.

Further, using the unit comprehensive default index,
It is also possible to evaluate quantitative items such as “operating profit ratio” and qualitative items such as “presence or absence of successor” in the same model as the evaluation items of the business individual.

Further, in the case where the distribution of the default destination is polarized as shown in FIG. 11, the original data as shown on the horizontal axis of FIG. 4 is standardized by the distance from the intermediate value of the default destination to obtain N. The accuracy can be improved by calculating the / S ratio. Alternatively, if the variable itself is normalized, the N / S ratio can be calculated and the accuracy can be improved by using the absolute value of the variable.

It is also possible to use the calculated unit comprehensive default index as one of the explanatory variables of a statistical model such as discriminant analysis and logistic analysis. Alternatively, there is a reverse procedure or a repeating method in which the N / S ratio is calculated using the evaluation result of a statistical model such as discriminant analysis or logistic analysis, and the unit comprehensive default index is calculated.

Here, the statistical model to be combined with the unit comprehensive default index is, for example, evaluation result = x, or evaluation result = base e of natural logarithm to the power of x / (1 + base e of natural logarithm to the power of x), , X are explanatory variables composed of a constant and n explanatory variables such as financial indexes and attributes.

When the evaluation result shows a time-series tendency, by increasing or decreasing the constant term of x of the above statistical model according to the time-series fluctuation, the model is prevented from becoming obsolete and the future prediction is made. The evaluation result can be calculated. Also,
When the tendency of evaluation results can be predicted by macro variables or semi-macro variables, increasing or decreasing the constant term of x in the above statistical model according to the predicted values of macro variables or semi-macro variables will make the model obsolete. Prevention and future prediction evaluation results can be calculated. In addition to time-series factors, this constant term can reflect all individual factors such as debtor size, industry, regional characteristics, fiscal year, and customer base. For example, the statistical model has a large sample default rate (= evaluation result)
If you are estimating a small sample to be evaluated (for example,
If you adjust this constant term by calculating backward so as to calculate the expected value of the actual default rate and the expected default rate of the debtor's size, type of industry, regional characteristics, fiscal year, customer segment, etc.),
Modeling that reflects the characteristics of small samples is possible. In addition,
The evaluation results here refer to the estimated default rate and the score as a proxy variable.

Various selections can be made for the division points. For example, the minimum misjudgment point, the minimum N / S ratio point, the lower 20% tile point, and the like.

For example, in the search method for the minimum erroneous discrimination point, for example, the point where the value of the following erroneous discrimination rate evaluation formula is the smallest is determined as the division point. Default evaluation formula = (D2 / D1) + (s2 / s1) Non-investment grade evaluation formula = (n2 / n1) + (i2 / i
1) D1: Total default number s1: Total survival number D2: Number of total default number that could not be determined as default s2: Number of total survival number that could not be determined as survival D2 / D1: Default error rate s2 / s1 : Survival misjudgment rate

Further, in the above example, the division target is the default / non-default, but it is also possible to use an external rating. In this case, for example, the investment grade is equal to or higher than BBB- and the non-investment grade is equal to or lower than BB +. In addition, when external evaluation is used, it is divided into, for example, xx or more and xx or less. Furthermore, when the internal evaluation is used, the order is divided into two parts at an arbitrary position in the order of the normal destination, the caution-required destination 1, the caution-required destination 2, the bankruptcy concern, the real bankruptcy destination, the bankruptcy destination, and the like.

In the above example, the various programs installed in the storage means such as the hard disk and the memory of the system can be stored in various types of storage media, and transmitted via the communication medium. Can be done. Here, the storage medium includes, for example, a flexible disk, a hard disk, a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD, a ROM cartridge, a RAM memory cartridge with battery backup, a flash memory cartridge, a non-volatile RAM cartridge, and the like. The communication medium includes a wired communication medium such as a telephone line, a wireless communication medium such as a microwave line, and the Internet.

[0080]

According to the present invention, credit evaluation can be appropriately performed, and in particular, qualitative data, quantitative data, financial data with many missing values, financial data with few missing values, data with many outliers, etc. can be integrated. It is possible to provide a credit evaluation system and a credit evaluation method that can make a positive judgment. More specifically, the credit evaluation system according to the present invention is useful as a credit risk management means for portfolios. Further, if the future is predicted based on the default rate calculated by the present invention and reflected in the loan interest rate, it is effective as a criterion for credit examination and pricing. Moreover, since the credit evaluation system according to the present invention makes it possible to measure the theoretical default rates of medium-sized enterprises, small and medium-sized enterprises, and micro enterprises, if a large number of companies of the same class are collected, the actual default rate can be brought close to the theoretical value. By collecting and securitizing a large number of corporate bonds, the diversification of the default rate will be reduced, and it will be possible to formulate products in a world where it is easy to gain the confidence of investors. In particular, with the financial intermediary function declining due to various factors such as bad debts, the CLO (Collateralized Loan Ob) that securitizes lending bonds as a new financial product
It is very effective to be able to form a ligation) and to form not only a loan but also a loan secured by the accounts receivable of a small and medium-sized enterprise.

[Brief description of drawings]

FIG. 1 is a conceptual diagram of an evaluation model in a credit evaluation system according to the present invention.

FIG. 2 is a conceptual diagram of a secondary scoring model in the credit evaluation system according to the present invention.

FIG. 3 is a diagram for explaining a signal model in the credit evaluation system according to the present invention.

FIG. 4 is a diagram for explaining a signal model in the credit evaluation system according to the present invention.

FIG. 5 is a diagram showing a hardware configuration example of a credit evaluation system according to the present invention.

FIG. 6 is a flowchart showing processing of the credit evaluation system according to the present invention.

FIG. 7 is a diagram showing an example of an input screen in the credit evaluation system according to the present invention.

FIG. 8 is a diagram showing an example of an input screen in the credit evaluation system according to the present invention.

FIG. 9 is a diagram showing an example of an input screen in the credit evaluation system according to the present invention.

FIG. 10 is a flowchart showing a process of the credit evaluation system according to the present invention.

FIG. 11 is a diagram for explaining another embodiment of the credit evaluation system according to the present invention.

[Explanation of symbols]

101 CPU 102 Input / output device 103 Control program 104 Index information file 105 Model information file 106 Obligor information file 107 Evaluation information file 108 Bus

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Shigeru Hikima             Kashiwabara Building, 1-3-3 Kyobashi, Chuo-ku, Tokyo             4F

Claims (7)

[Claims] 1. A credit evaluation system comprising first processing means and second processing means for evaluating creditworthiness of a debtor, wherein the first processing means is a debtor depending on whether the creditworthiness is high or low. And a second data processing unit, wherein the second processing unit has a first data distribution that has reached a state of low creditworthiness with respect to an index for evaluating the creditworthiness of the debtor. An index information file that stores past index data including a second data distribution that has reached a high credibility state, and each of the first data distribution and the second data distribution acquired from the index information file are divided. Means for determining a division value, means for dividing the first data distribution based on the division value, and calculating as a signal information the ratio on the low credibility side of the entire first data distribution; Based on , Means for dividing the second data distribution and calculating a ratio of low reliability in the entire second data distribution as noise information, and means for calculating discriminating ability information based on the ratio of the signal information and the noise information And a means for determining whether the evaluation target belongs to the side with high credit rating or the side with low credit rating based on the division value, and assigning a score based on the determination result of the determining means, and A credit evaluation system having means for reflecting a coefficient based on discrimination ability information. 2. A means for calculating a score per unit by assigning a score to a plurality of indexes, obtaining a specific total value, and dividing by a number of the indexes. The credit evaluation system described in 1. 3. The profit-related index is included in the index for evaluating the creditworthiness of the debtor.
The credit evaluation system described. 4. An index for evaluating the creditworthiness of an obligor, including past index data including a first data distribution that has reached a low credit quality and a second data distribution that has reached a high credit quality. An index information file to be stored, a means for determining a division value for dividing each of the first data distribution and the second data distribution acquired from the index information file, and the first data based on the division value. A means for dividing the distribution and calculating the ratio of the low credibility side in the entire first data distribution as signal information, and dividing the second data distribution based on the division value to calculate the ratio in the entire second data distribution. Means for calculating the ratio on the low credibility side as noise information, means for calculating the discrimination capability information based on the ratio of the signal information and noise information, and evaluation based on the division value A means for determining whether the elephant belongs to the side with high credibility or a side with low credibility, a score is given based on the determination result of the determining means, and a coefficient based on the discrimination ability information is reflected on the score. A credit evaluation system including means, comprising selection means for selecting an index for evaluating the creditworthiness of the obligor, the selection means being capable of discriminating each index from the index information file. Means for calculating information and selecting a first index having the highest discriminating ability based on this discriminating ability information, and a second indicator having a high discriminating ability following the first index, and selecting the second indicator And a means for determining whether or not the explanatory power for the default sample is improved based on the first index, and if it is improved, further subsequent indexes are selected and sequentially set as the default sample. Credit rating system continues until explanatory power is no longer improved, if the explanatory power does not improve, having means to stop the process of selecting an index for. 5. The credit evaluation system according to claim 1, wherein the obligor is a small and medium-sized enterprise. 6. A credit evaluation method comprising a first processing step and a second processing step for evaluating creditworthiness of an obligor, wherein the first processing step uses a control program to determine whether the creditworthiness is high or low. The debtor's data into a plurality of data groups, and the second processing step relates to an index for evaluating the creditworthiness of the debtor, and the first data distribution leading to a state of low creditworthiness and the creditworthiness of the debtor. The control program determines a division value for dividing each of the first data distribution and the second data distribution acquired from the index information file that stores the past index data including the second data distribution that has reached a high state. And a step of dividing the first data distribution on the basis of the division value, and using a ratio of the low credibility side in the entire first data distribution as signal information according to the control program. The step of calculating the second data distribution based on the division value, and calculating the ratio of the low credibility side of the entire second data distribution as noise information by the control program, the signal information And a step of calculating the discrimination capability information by the control program based on the ratio of the noise information, and a step of determining by the control program whether the evaluation target belongs to the side with high credibility or the side with low credibility, based on the division value, A credit evaluation method comprising a step of giving a score based on a judgment result of the judgment means by a control program and reflecting a coefficient based on the discrimination ability information to the score. 7. Regarding the index for evaluating the creditworthiness of the obligor, past index data including a first data distribution that has reached a low credit status and a second data distribution that has reached a high credit status Determining a division value for dividing each of the first data distribution and the second data distribution acquired from the stored index information file by a control program, and dividing the first data distribution based on the division value Then, a step of calculating the ratio of the low credibility side in the entire first data distribution by the control program as signal information, and dividing the second data distribution based on the division value, and dividing the second data distribution in the entire second data distribution. Calculating the ratio of the low credibility side as noise information by a control program, and discriminating ability based on the ratio of the signal information and noise information Calculating a report by a control program, based on the divided value, determining by the control program whether the evaluation target belongs to the high-credit side or the low-credit side, and a score based on the determination result of the determination means And a credit evaluation method including a step of reflecting a coefficient based on the discrimination ability information to the score by a control program, and selecting an index for evaluating the creditworthiness of the obligor by the control program. A step of calculating discriminating ability information for each indicator from the indicator information file, and selecting the first indicator having the highest discriminating ability based on the discriminating ability information. Then, a second index having high discrimination ability is selected following the first index, and the second index and the first finger are selected. The step of determining whether the explanatory power for the default sample is improved based on the above, and if it is improved, select the subsequent index and continue until the explanatory power for the default sample is not improved. A credit scoring system comprising the step of aborting the index selection process if not improved.