JP2004355236A - System and method for evaluating accuracy of credit risk evaluation model - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To allow a model evaluator to evaluate the accuracy of a credit risk evaluation model requested from a model owner without making it necessary for the model owner to disclose the credit risk evaluation model, or for the model evaluator to own the financial data of a large number of enterprises. <P>SOLUTION: This system is configured to perform the accuracy evaluation of a model requested from the owner of a model for credit risk evaluation for evaluating the bankruptcy probability of a company without grasping the overall contents of the model. This system is provided with a data fetching part 34 for acquiring credit risk evaluation calculation operated by substituting data for accuracy evaluation with response with each of a plurality of partial model formulas configuring a model formula to be used for the model, a ranking result integrating part 40 for integrating the results of the acquired credit risk evaluation calculation and a model accuracy evaluating part 46 for comparing the result integrated by the ranking result integrating part 40 with a preliminarily decided accuracy evaluation standard, and for evaluating the accuracy of the model based on the comparison result. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

【００９９】
上記式にしたがって得られる尤度Ｌは、ＣＲＡＦＴスコアの場合
Ｌ＝（１−０．１）×０．３×（１−０．２）×（１−０．１）×０．５＝０．０９７２ であり、
信用リスク評価モデルの場合
Ｌ＝（１−０．３）×０．２×（１−０．１）×（１−０．２）×０．３＝０．０３０２４ である。
【０１００】
尤度が大きい方が優れたモデルであるので、この場合ＣＲＡＦＴスコアの方が優れたモデルであるものと判定することができる。
【０１０１】
次に、以上のように構成した本実施の形態に係る信用リスク評価モデルの精度評価方法を適用した信用リスク評価モデル精度評価システムの動作について図８に示す処理フロー図を用いて説明する。
【０１０２】
まず、本実施の形態に係る信用リスク評価モデルの精度評価方法を適用した信用リスク評価モデル精度評価システムによって、信用リスク評価モデルの精度評価を行う場合には、モデル分割部１８において、信用リスク評価モデルに使用されている例えば式（ｉ）乃至（ｖｉｉ）からなるモデル式が、式（ｉ）と、式（ｉｉ）乃至（ｖｉｉ）とからなる２つのグループに分割され、分割された各グループがデータ送信部２０へと出力される（Ｓ１’）。
【０１０３】
モデル分割部１８から各グループの式情報がデータ送信部２０へと出力されると、式（ｉ）が例えば計算者サーバ１４（＃２）へ、式（ｉｉ）乃至（ｖｉｉ）が例えば計算者サーバ１４（＃１）へとそれぞれ送信される（Ｓ２’）。
【０１０４】
一方、財務データ格納部２２に格納されている財務データの中から精度評価に用いる企業の財務データが、またこの企業の倒産情報データが倒産情報格納部２４から、それぞれデータ送信部２０によって取得される（Ｓ３）。これら取得されたデータは、通信ネットワーク１０を介して評価者サーバ１６側へと送信される（Ｓ４）。
【０１０５】
データ送信部２０から送信された財務データおよび倒産情報データは、データ取込部３４によって取り込まれる（Ｓ５）。そして、財務データは、データ取込部３４から財務データ格納部３６へ（Ｓ６）、倒産情報データは、データ取込部３４から倒産情報格納部４４へ（Ｓ７）とそれぞれ格納される。
【０１０６】
財務データ格納部３６に財務データが格納されると、格納された財務データがデータ送信部３８によって取得され（Ｓ８）、取得された財務データが、通信ネットワーク１０を介して計算者サーバ１４（＃１）へと送信される（Ｓ９’）。
【０１０７】
次に、ステップＳ２’において、データ送信部２０から式（ｉｉ）乃至（ｖｉｉ）が送信された計算者サーバ１４（＃１）のデータ取込部２８によって、ステップＳ２’においてデータ送信部２０から送信された式（ｉｉ）乃至（ｖｉｉ）と、ステップＳ９’においてデータ送信部３８から送信された財務データとが受信される（Ｓ１０’）。そして、受信した式（ｉｉ）乃至（ｖｉｉ）と、財務データとが格付等算出部３０へと出力される（Ｓ１１’）。
【０１０８】
計算者サーバ１４（＃１）の格付等算出部３０では、データ取込部２８から出力された式（ｉｉ）乃至（ｖｉｉ）に、データ取込部２８から出力された財務データを代入した演算が行われ、演算結果であるＫＡ１〜ＫＡ６がデータ送信部３２へと出力される（Ｓ１２’）。
【０１０９】
上述したように、計算者サーバ１４（＃１）には式（ｉｉ）乃至（ｖｉｉ）が送信されるものの式（ｉ）は送信されない。これによって計算者サーバ１４（＃１）には、信用リスク評価モデルのモデル式の全容が開示されることはない。また、自分が演算した結果であるＫＡ１〜ＫＡ６がどのような計算に用いられるのかも分からない。
【０１１０】
計算者サーバ１４（＃１）のデータ送信部３２では、格付等算出部３０から出力された演算結果であるＫＡ１〜ＫＡ６が取得される。そして、この演算結果ＫＡ１〜ＫＡ６が通信ネットワーク１０を介して計算者サーバ１４（＃２）に向けて出力される（Ｓ１３’）。
【０１１１】
ステップＳ２’において、データ送信部２０から式（ｉ）が送信された計算者サーバ１４（＃２）のデータ取込部２８では、ステップＳ２’においてデータ送信部２０から送信された式（ｉ）と、ステップＳ１３’において計算者サーバ１４（＃１）のデータ送信部３８から送信された演算結果ＫＡ１〜ＫＡ６とが受信される（Ｓ１０’’）。そして、受信した式（ｉ）と、演算結果ＫＡ１〜ＫＡ６とが格付等算出部３０へと出力される（Ｓ１１’’）。
【０１１２】
計算者サーバ１４（＃２）の格付等算出部３０では、データ取込部２８から出力された式（ｉ）に、演算結果ＫＡ１〜ＫＡ６を代入した演算が行われ、その演算結果Ｚがデータ送信部３２へと出力される（Ｓ１２’’）。
【０１１３】
上述したように、計算者サーバ１４（＃２）には式（ｉ）が送信されるものの財務データは送信されない。これによって計算者サーバ１４（＃２）側では、演算結果ＫＡ１〜ＫＡ６がどのような式によって、どのような財務データが用いられて得られたものであるか一切知ることができない。したがって、やはり計算者サーバ１４（＃２）にもまた信用リスク評価モデルのモデル式の全容が開示されることはない。
【０１１４】
計算者サーバ１４（＃２）のデータ送信部３２では、格付等算出部３０から出力された演算結果Ｚが取得され、更に演算結果Ｚが通信ネットワーク１０を介して評価者サーバ１６に向けて出力される（Ｓ１３’’）。
【０１１５】
データ取込部３４は、このようにして計算者サーバ１４（＃２）のデータ送信部３２から送信された演算結果Ｚが通信ネットワーク１０を介して受信され（Ｓ１４’）、格付結果統合部４０を介して格付結果格納部４２へと書き込まれる（Ｓ１６’）。
【０１１６】
モデル精度評価部４６では、格付結果格納部４２に格納された演算結果と、倒産情報格納部４４に格納されたこの企業の倒産情報データとがそれぞれ取得される（Ｓ１７，Ｓ１８）。更に、評価ルール格納部４８に予め格納されている評価ルールの情報が取得される（Ｓ１９）。
【０１１７】
そして、ステップＳ２０’では、ステップＳ１７’で取得した演算結果Ｚと、ステップＳ１８で取得した倒産情報データと、ステップＳ１９で取得した評価ルールとに基づいて、信用リスク評価モデルの精度が評価される（Ｓ２０’）。評価方法としては例えばＡＲ値に基づく評価や、ＣＲＡＦＴスコアとの比較による評価がある。上述したような評価結果に基づいて信用リスク評価モデルの精度評価が行われる。
【０１１８】
評価結果送信部５０では、モデル精度評価部４６から出力された評価結果データが、通信ネットワーク１０を介して所有者サーバ１２に向けて送信される（Ｓ２１）。
【０１１９】
ステップＳ２１において評価結果送信部５０から送信された評価結果データは、所有者サーバ１２に備えられた評価結果取込部２６によって受信される（Ｓ２２）。そして、この評価結果データが評価者に参照されることにより、信用リスク評価モデルの精度が評価者によって把握される。
【０１２０】
上述したように、本実施の形態に係る信用リスク評価モデルの精度評価方法を適用した信用リスク評価モデル精度評価システムにおいては、上記のような作用により、計算者サーバ１４（＃１）には式（ｉｉ）乃至（ｖｉｉ）が送信されるものの式（ｉ）は送信されない。これによって計算者サーバ１４（＃１）には、信用リスク評価モデルのモデル式の全容が開示されることはない。また、自分が演算した結果であるＫＡ１〜ＫＡ６がどのような計算に用いられるのかも分からない。
【０１２１】
一方、計算者サーバ１４（＃２）には式（ｉ）が送信されるものの財務データは送信されない。これによって計算者サーバ１４（＃２）側では、演算結果ＫＡ１〜ＫＡ６がどのような式によって、どのような財務データが用いられて得られたものであるか一切知ることができない。したがって、やはり計算者サーバ１４（＃２）にもまた信用リスク評価モデルのモデル式の全容が開示されることはない。
【０１２２】
更に、評価者サーバ１６は、計算者サーバ１４（＃２）から演算結果のみを知らさせ、この演算結果に基づいて信用リスク評価モデルの精度の評価を行う。したがって、やはり信用リスク評価モデル式の内容については全く知り得ることはない。
【０１２３】
その結果、信用リスク評価モデルの所有者は、各計算者および評価者のいずれに対しても信用リスク評価モデルの全体内容を開示することなく精度評価を委託することが可能となる。
【０１２４】
以上、本発明の好適な実施の形態について、添付図面を参照しながら説明したが、本発明はかかる構成に限定されない。特許請求の範囲の発明された技術的思想の範疇において、当業者であれば、各種の変更例及び修正例に想到し得るものであり、それら変更例及び修正例についても本発明の技術的範囲に属するものと了解される。
【０１２５】
【発明の効果】
以上説明したように、本発明によれば、モデル保有者が信用リスク評価モデルを公開することなく、かつモデル評価者が多くの企業の財務データを保有することなく、モデル評価者がモデル保有者から委託された信用リスク評価モデルの精度評価を行うことが可能な信用リスク評価モデルの精度評価システムおよび精度評価方法を実現することができる。
【図面の簡単な説明】
【図１】第１の実施の形態に係る信用リスク評価モデルの精度評価方法を適用した信用リスク評価モデル精度評価システムの構成例を示す機能ブロック図。
【図２】収益性ポイントと安全性ポイントとによる指標を説明するための図。
【図３】財務データの一例を示す図。
【図４】第１の実施の形態に係る信用リスク評価モデルの精度評価方法を適用した信用リスク評価モデル精度評価システムの動作を示す処理フロー図。
【図５】財務指標とその係数の一例を示す図。
【図６】ＡＲ値に基づく評価に用いられるスコアと累積倒産率との関係を示す図。
【図７】ＣＲＡＦＴスコアとの比較評価を説明するための図。
【図８】第２の実施の形態に係る信用リスク評価モデルの精度評価方法を適用した信用リスク評価モデル精度評価システムの動作を示す処理フロー図。
【符号の説明】
１０…通信ネットワーク、１２…所有者サーバ、１４…計算者サーバ、１６…評価者サーバ、１８…モデル分割部、２０…データ送信部、２２…財務データ格納部、２４…倒産情報格納部、２６…評価結果取込部、２８…データ取込部、３０…付等算出部、３２…データ送信部、３４…データ取込部、３６…財務データ格納部、３８…データ送信部、４０…付結果統合部、４２…付結果格納部、４４…倒産情報格納部、４６…モデル精度評価部、４８…評価ルール格納部、５０…評価結果送信部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an accuracy evaluation system and an accuracy evaluation method of a credit risk evaluation model for evaluating a bankruptcy probability of a company, and more specifically, an owner who owns a credit risk evaluation model reads the entire contents of the credit risk evaluation model. A system and method for allowing a third party to evaluate the accuracy of a credit risk assessment model without disclosing.
[0002]
[Prior art]
Needless to say, it is important in business activities and the like to ascertain whether or not the financial status of the business partner or the borrower is sound. Therefore, some companies (hereinafter, referred to as “model holders”) have a credit risk evaluation model for evaluating the bankruptcy probability of the company based on the financial situation of the business partner or the borrower. As a credit risk evaluation model of this type, there is, for example, a CRAFT scoring method shown in Non-Patent Document 1 below.
[0003]
Generally, the accuracy of this type of credit risk evaluation model deteriorates as time elapses after the credit risk evaluation model is constructed. Therefore, when the accuracy falls below a certain value, it is necessary to replace this model and maintain it so that a highly reliable evaluation result is always obtained.
[0004]
However, it goes without saying that replacing the model involves costs. Normally, the model is replaced once a year, that is, when the financial statements are published. However, if it is shown from an objective point of view that the deterioration of the model accuracy is small, that is, the model has sufficient accuracy, there is no need to replace the model, and the cost associated with model replacement will be reduced. Can be avoided.
[0005]
Therefore, a third party (hereinafter, referred to as a “model evaluator”) is required to evaluate the accuracy of the credit risk evaluation model from an objective viewpoint.
[0006]
[Non-patent document 1]
Toshiba Review Vol. 57 No. 4 (2002) p. 43-46
[0007]
[Problems to be solved by the invention]
However, when the accuracy of the credit risk evaluation model is evaluated by such a third party, that is, the model evaluator, there are the following problems.
[0008]
That is, when a model holder entrusts a model evaluator to evaluate the accuracy of a credit risk evaluation model owned by the model evaluator, there are generally two methods.
[0009]
In the first method, a model holder evaluates bankruptcy probabilities for many companies using a credit risk evaluation model held by the model holder, and passes the result to a model evaluator. Then, the model evaluator is a method of evaluating the accuracy of the credit risk evaluation model based on the evaluation result.
[0010]
In the second method, the model holder discloses the logic of the credit risk evaluation model to the model evaluator. Then, the model evaluator evaluates the probability of bankruptcy of a company by applying actual financial data of various companies to this credit risk evaluation model, and based on the evaluation result, evaluates the accuracy of the credit risk evaluation model It is.
[0011]
In the first method, the model holder needs to evaluate the bankruptcy probability of many companies, and thus needs financial data of these companies. However, in general, model holders often do not have financial data for many companies. For this reason, the bankruptcy probability evaluation results of many companies cannot be presented to the model evaluator, and the model evaluator cannot perform highly reliable accuracy evaluation.
[0012]
In the second method, the model holder must disclose the credit risk evaluation model, which is his own know-how, to the model evaluator, and there is a problem that his own know-how leaks. Further, there is a problem that the model evaluator must have financial data of many companies in order to evaluate the bankruptcy probability of many companies.
[0013]
The present invention has been made in view of such circumstances, and has been developed without the model holder disclose the credit risk evaluation model and without the model evaluator holding financial data of many companies. Of the present invention is to provide a credit risk evaluation model accuracy evaluation system and an accuracy evaluation method capable of evaluating the accuracy of a credit risk evaluation model commissioned by a model holder.
[0014]
[Means for Solving the Problems]
In order to achieve the above object, the present invention takes the following measures.
[0015]
That is, according to the invention of claim 1, the owner who owns the credit risk evaluation model for evaluating the bankruptcy probability of the company can evaluate the accuracy of the credit risk evaluation model by two or more calculators and one evaluator. A method whereby the evaluator evaluates the accuracy of the credit risk evaluation model without disclosing the entire contents of the credit risk evaluation model to each of the evaluators and the evaluator. First, the model formula used in the credit risk evaluation model is divided into the same number of partial model formulas as the calculator, and each of the divided partial model formulas and the data for the credit risk evaluation calculation substituted into each partial model formula Is given to each computer. Next, each computer is caused to execute a credit risk evaluation calculation by substituting data into the partial model formula. Then, the evaluator is caused to compare the result of the credit risk evaluation calculation by each computer with a predetermined accuracy evaluation criterion, and to evaluate the accuracy of the credit risk evaluation model based on the comparison result.
[0016]
Therefore, in the method for evaluating the accuracy of the credit risk evaluation model according to the first aspect of the present invention, by taking the above measures, each computer is given only a partial model expression, and the evaluator is provided with the model expression. Not disclosed at all. Therefore, the owner can outsource the accuracy evaluation to each of the calculators and the evaluator without disclosing the entire contents of the credit risk evaluation model.
[0017]
According to a second aspect of the present invention, the owner who owns the credit risk evaluation model for evaluating the bankruptcy probability of the company can evaluate the accuracy of the credit risk evaluation model by one or more partial calculators and one total calculator. By outsourcing to one evaluator, the evaluator can improve the accuracy of the credit risk evaluation model without disclosing the entire contents of the credit risk evaluation model to each evaluator and evaluator. A method of evaluating a credit risk by first providing one or more partial calculators with each term constituting a model formula used in a credit risk assessment model and a credit risk assessment calculation corresponding to each term. And data are given. Next, while allowing each partial calculator to calculate the value of each term by substituting data for each term, the total calculator is applied to the model formula composed of each term by each partial calculator. The credit risk evaluation calculation is executed by substituting the calculated values of the terms. Further, the evaluator causes the result of the credit risk evaluation calculation to be compared with a predetermined accuracy evaluation criterion, and the accuracy of the credit risk evaluation model is evaluated based on the comparison result.
[0018]
Therefore, in the method for evaluating the accuracy of the credit risk evaluation model according to the second aspect of the present invention, by taking the above measures, the entire model formula is not disclosed to any of the computer users. Discloses only the final result obtained by substituting a value into the model formula. Therefore, the owner can outsource the accuracy evaluation to each of the calculators and the evaluator without disclosing the entire contents of the credit risk evaluation model.
[0019]
According to the invention of claim 3, the accuracy evaluation of the credit risk evaluation model commissioned by the owner who owns the credit risk evaluation model for evaluating the bankruptcy probability of the company can be performed without grasping the entire contents of the credit risk evaluation model. The system includes an acquisition unit, an integration unit, a comparison unit, and an evaluation unit. The obtaining means obtains a credit risk evaluation calculation performed by substituting data for accuracy evaluation for each of a plurality of partial model expressions constituting the model expression used in the credit risk evaluation model. The integrating means integrates the results of each credit risk evaluation calculation acquired by the acquiring means. The comparing means compares the result integrated by the integrating means with a predetermined accuracy evaluation criterion. The evaluation unit evaluates the accuracy of the credit risk evaluation model based on the comparison result by the comparison unit.
[0020]
Therefore, in the credit risk evaluation model accuracy evaluation system according to the third aspect of the present invention, by taking the above-described measures, each computer is given only a partial model expression, and the evaluator is provided with the model expression. Not disclosed at all. Therefore, the owner can outsource the accuracy evaluation to each of the calculators and the evaluator without disclosing the entire contents of the credit risk evaluation model.
[0021]
The invention according to claim 4 performs the accuracy evaluation of the credit risk evaluation model commissioned by the owner who owns the credit risk evaluation model for evaluating the bankruptcy probability of the company without grasping the entire contents of the credit risk evaluation model. An acquisition unit for acquiring a credit risk evaluation calculation performed by substituting data for accuracy evaluation into a model formula used in a credit risk evaluation model, and a result integrated by the acquisition unit and A comparison means for comparing the accuracy with the determined accuracy evaluation standard, and an evaluation means for evaluating the accuracy of the credit risk evaluation model based on the comparison result by the comparison means.
[0022]
Therefore, in the accuracy evaluation system for a credit risk evaluation model according to the fourth aspect of the present invention, by taking the above-described measures, the entire model formula is not disclosed to any of the calculators. Discloses only the final result obtained by substituting a value into the model formula. Therefore, the owner can outsource the accuracy evaluation to each of the calculators and the evaluator without disclosing the entire contents of the credit risk evaluation model.
[0023]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0024]
(First Embodiment)
A first embodiment of the present invention will be described with reference to FIGS.
[0025]
FIG. 1 is a functional block diagram showing a configuration example of a credit risk evaluation model accuracy evaluation system to which a credit risk evaluation model accuracy evaluation method according to the first embodiment is applied.
[0026]
That is, the credit risk evaluation model accuracy evaluation system to which the accuracy evaluation method of the credit risk evaluation model according to the present embodiment is applied is a method in which the owner who owns the credit risk evaluation model for evaluating the bankruptcy probability of a company has a credit risk evaluation model. Disclose the entire contents of the credit risk evaluation model to both each evaluator and each evaluator by delegating the accuracy evaluation of the evaluation model to two or more evaluators and one evaluator. This is a system that allows the evaluator to evaluate the accuracy of the credit risk evaluation model.
[0027]
This system includes an owner server 12, a plurality of computer servers 14 (# 1, # 2,..., #N) connected to each other by a communication network 10 such as the Internet or a LAN, and an evaluator server 16 It consists of:
[0028]
The owner server 12 is provided on the owner side, and includes a model division unit 18, a data transmission unit 20, a financial data storage unit 22, a bankruptcy information storage unit 24, and an evaluation result acquisition unit 26.
[0029]
The computer server 14 is provided on each computer side, and includes a data acquisition unit 28, a rating calculation unit 30, and a data transmission unit 32.
[0030]
The evaluator server 16 is provided on the evaluator side, and includes a data acquisition unit 34, a financial data storage unit 36, a data transmission unit 38, a rating result integration unit 40, a rating result storage unit 42, and a bankruptcy information storage. It includes a unit 44, a model accuracy evaluation unit 46, an evaluation rule storage unit 48, and an evaluation result transmission unit 50.
[0031]
The model dividing unit 18 divides the model formula used for the credit risk evaluation model into the same number of partial model formulas as two or more calculators outsourcing the calculation, and transmits information on each divided partial model formula to data. Output to the unit 20 (S1).
[0032]
For example, as shown in FIG. 2, the model formula of the credit risk evaluation model is a partial model formula (1) for calculating the A index and the B index which are profitability points, and the C index and D which are security points. When it is composed of the partial model formula (2) for calculating the index, the model formula used for the credit risk evaluation model is represented by the partial model formula (1) and the partial model formula (2). And outputs the information of the divided partial model formulas (1) and (2) to the data transmission unit 20.
[0033]
Further, if the partial model equation (1) can be divided into a partial model equation (1A) for calculating the A index and a partial model equation (1B) for calculating the B index, the divided partial model equation (1A) ) And the information of the partial model formula (1B) may be output to the data transmitting unit 20. Similarly, if the partial model formula (2) can be divided into a partial model formula (2C) for calculating the C index and a partial model formula (2D) for calculating the D index, the divided partial model formula (2D) 2C) and the information of the partial model formula (2D) may be output to the data transmitting unit 20.
[0034]
Each computer is provided with a computer server 14 (# 1, # 2,..., #N), and the owner is any one of n (n ≧ 2) computers. The calculation can be entrusted to two or more calculators. Accordingly, not all of the computer servers 14 (# 1, # 2,..., #N) shown in FIG. 1 are necessarily required, and if at least two exist, the present embodiment is applied. The credit risk evaluation model accuracy evaluation system is operable. For example, when the model dividing unit 18 divides the model formula of the credit risk evaluation model into two partial model formulas (1, 2), the calculation is entrusted to any two calculators, and the model dividing unit 18 Divides the model formula of the credit risk evaluation model into four partial model formulas (1A, 1B, 2C, 2D), the calculation is entrusted to any four calculators.
[0035]
The financial data storage unit 22 stores the financial data of the company in advance, and the bankruptcy information storage unit 24 stores the bankruptcy information data of the corresponding company in advance. For example, as shown in FIG. 3, the financial data corresponds to general financial indicators such as a return on total assets business, a return on shareholders' equity, a gross profit margin, a recurring return on sales, a total assets turnover, and an immediate ratio. Data. Bankruptcy information data is information on whether or not the company has gone bankrupt.
[0036]
When the information of each partial model formula is output from the model dividing unit 18, the data transmitting unit 20 transmits the information of each partial model formula to each of the outsourced computer servers 14 (S2). Further, from among the financial data stored in the financial data storage unit 22, the financial data of the company used for the accuracy evaluation is obtained, and the bankruptcy information data of this company is obtained from the bankruptcy information storage unit 24 (S3). . Then, the financial data and the bankruptcy information data are transmitted to the evaluator server 16 via the communication network 10 (S4).
[0037]
For example, the model formulating unit 18 divides the model formula into two partial model formulas, a partial model formula (1) and a partial model (2), and calculates the partial model formula (1) using the computer server 14 (# 1). When the information of the partial model equation (1) is transmitted to the computer server 14 (# 1), and the calculation of the partial model equation (2) is entrusted to the computer server 14 (# 2). Transmits the information of the partial model formula (2) to the computer server 14 (# 2). Similarly, the model formulating unit 18 divides the model formula into four partial model formulas (1A, 1B, 2C, and 2D), and calculates the partial model formula (1A) by using the computer server 14 (# 1). When the information of the partial model expression (1A) is transmitted to the computer server 14 (# 1) and the calculation of the partial model expression (1B) is entrusted to the computer server 14 (# 2). When the information of the partial model formula (1B) is transmitted to the computer server 14 (# 2) and the calculation of the partial model formula (2C) is entrusted to the computer server 14 (# 3), When the information of the equation (2C) is transmitted to the computer server 14 (# 3) and the calculation of the partial model equation (2D) is entrusted to the computer server 14 (# 4), the partial model equation (2D) Is transmitted to the computer server 14 (# 4).
[0038]
The data capturing unit 34 captures the financial data and the bankruptcy information data transmitted from the data transmitting unit 20 (S5), transfers the financial data to the financial data storage unit 36 (S6), and transfers the bankruptcy information data to the bankruptcy information storage unit 44. (S7).
[0039]
When the financial data is stored in the financial data storage unit 36, the data transmitting unit 38 acquires the stored financial data (S8), and transmits the acquired financial data to the computer server 14 via the communication network 10. It is transmitted (S9). In this case, the data may be transmitted to all the computer servers 14 (# 1, # 2,..., #N). The information may be transmitted only to the computer server 14 to which the information has been transmitted.
[0040]
In step S2, the data capturing unit 28 of each of the computer servers 14 to which the information of the partial model formula has been transmitted from the data transmitting unit 20 includes the information of the partial model formula transmitted from the data transmitting unit 20 in step S2, In step S9, the financial data transmitted from the data transmitting unit 38 is received (S10). Then, the received information of the partial model formula and the financial data are output to the rating etc. calculating unit 30 (S11).
[0041]
The rating calculation unit 30 performs an operation by substituting the financial data output from the data acquisition unit 28 into the partial model formula output from the data acquisition unit 28, and outputs the operation result to the data transmission unit 32. (S12).
[0042]
For example, the rating calculation section 30 of the computer server 14 to which the partial model formula (1) has been transmitted calculates the A index and the B index by substituting the financial data into the partial model formula (1), and calculates each calculation result. The information on the profitability points (A index XX points, B index △△ points) is transmitted to the data transmission unit 32. Further, the rating etc. calculating unit 30 of the computer server 14 to which the partial model formula (2) is transmitted calculates the C index and the D index by substituting the financial data into the partial model formula (2), and calculates each calculation result. The information of the safety points (C index xx points, D index □□ points) is transmitted to the data transmission unit 32.
[0043]
Thus, only the partial model formula is transmitted to each computer server 14. For this reason, no computer server 14 discloses the entire model formula of the credit risk evaluation model.
[0044]
The data transmission unit 32 acquires the calculation result output from the rating etc. calculation unit 30, and outputs the obtained calculation result to the evaluator server 16 via the communication network 10 (S13).
[0045]
When the data acquisition unit 34 receives each operation result transmitted from the data transmission unit 32 of each computer server 14 in this way via the communication network 10 (S14), it ranks each received operation result. The result is output to the result integration unit 40 (S15).
[0046]
The rating result integration unit 40 integrates the respective calculation results output from the data acquisition unit 34 to substitute the financial data into the model formula to obtain the calculation results, and stores the calculation results in the rating result storage unit 42. Write (S16). It should be noted that the rating result integration unit 40 only performs the integration processing of the calculation results transmitted from the data transmission unit 32 of each of the computer servers 14 entrusted with the calculation. I don't know the content.
[0047]
For example, when the data capturing unit 34 captures the information on the profitability points (A index ○ point, B index △△ point) from the data transmission unit 32 of the computer server 14 (# 1), the data capture unit 34 The result is output to the rating result integration unit 40. Similarly, when the data capturing unit 34 captures the information of the security points (C index XX points, D index □□ points) from the data transmitting unit 32 of the computer server 14 (# 2), the data capturing unit 34 Is output to the rating result integration unit 40. Thereby, the rating result integration unit 40 grasps the profitability points (A index XX points, B index △△ points) and safety points (C index XX points, D index □ points) of the company. However, the rating result integration unit 40 can know the profitability points (A index XX points, B index △△ points) and security points (C index XX points, D index □ points) of the company. However, they do not know what these points are as a result of what kind of model formula and what kind of financial data is input.
[0048]
The model accuracy evaluation unit 46 acquires the calculation result stored in the rating result storage unit 42 (S17), and the bankruptcy information data of this company stored in the bankruptcy information storage unit 44 and (S18). Further, information on the evaluation rules stored in advance in the evaluation rule storage unit 48 is obtained (S19). Then, the accuracy of the credit risk evaluation model is evaluated based on the calculation result obtained in step S17, the bankruptcy information data obtained in step S18, and the evaluation rule obtained in step S19.
[0049]
The evaluation rule is an evaluation criterion for evaluating the accuracy of the credit risk evaluation model based on the calculation result obtained by the rating result integration unit 40.
[0050]
As this evaluation criterion, for example, a total point obtained by adding a profitability point (A index XX point, B index △△ point) and a safety point (C index XX point, D index □□ point) is used. There is a bankruptcy probability based on: In this case, the rating result integration unit 40 calculates the sum of the profitability points (A index XX points, B index △△ points) and the safety points (C index XX points, D index □□ points). A pair of the point and the result of bankruptcy determined from the bankruptcy information data acquired in step S18 is compared with an evaluation criterion to evaluate the accuracy of the credit risk evaluation model. The model accuracy evaluation unit 46 outputs the evaluation result data to the evaluation result transmission unit 50 (S20).
[0051]
When the evaluation result data is output from the model accuracy evaluation unit 46, the evaluation result transmission unit 50 transmits the evaluation result data to the owner server 12 via the communication network 10 (S21).
[0052]
The evaluation result data transmitted from the evaluation result transmitting unit 50 of the evaluator server 16 in this manner is received by the evaluation result capturing unit 26 provided in the owner server 12 (S22). The evaluator grasps the accuracy of the credit risk evaluation model owned by the evaluator by referring to the evaluation result data received by the evaluation result capturing unit 26.
[0053]
Next, the operation of the credit risk evaluation model accuracy evaluation system to which the credit risk evaluation model accuracy evaluation method according to the present embodiment configured as described above is applied will be described with reference to the processing flowchart shown in FIG.
[0054]
First, when the accuracy evaluation of the credit risk evaluation model is performed by the credit risk evaluation model accuracy evaluation system to which the accuracy evaluation method of the credit risk evaluation model according to the present embodiment is applied, the credit risk evaluation model The model formula used in the model is divided into the same number of partial model formulas as two or more calculators entrusted with the calculation, and information on the divided partial model formulas is output to the data transmission unit 20 ( S1).
[0055]
Here, the credit risk evaluation model is composed of a partial model for evaluating profitability points and a partial model for evaluating security points as shown in FIG. The case of dividing into the model formula (1) and the partial model formula (2) will be described as an example.
[0056]
When the information of each partial model formula is output from the model dividing unit 18 to the data transmitting unit 20, the information of each partial model formula is transmitted to each of the outsourced computer servers 14 (S2). Here, it is assumed that the information of the partial model equation (1) is output to the computer server 14 (# 1), and the information of the partial model equation (2) is output to the computer server 14 (# 2).
[0057]
On the other hand, from the financial data stored in the financial data storage unit 22, the financial data of the company used for the accuracy evaluation and the bankruptcy information data of this company are obtained from the bankruptcy information storage unit 24 by the data transmission unit 20, respectively. (S3). These acquired data are transmitted to the evaluator server 16 via the communication network 10 (S4).
[0058]
The financial data and the bankruptcy information data transmitted from the data transmitting unit 20 are captured by the data capturing unit 34 (S5). Then, the financial data is stored from the data acquisition unit 34 to the financial data storage unit 36 (S6), and the bankruptcy information data is stored from the data acquisition unit 34 to the bankruptcy information storage unit 44 (S7).
[0059]
When the financial data is stored in the financial data storage unit 36, the stored financial data is acquired by the data transmission unit 38 (S8), and further transmitted to the computer server 14 via the communication network 10 (S9). .
[0060]
In step S2, the data capturing unit 28 of each of the computer servers 14 to which the information of the partial model formula has been transmitted from the data transmitting unit 20 includes the information of the partial model formula transmitted from the data transmitting unit 20 in step S2, In step S9, the financial data transmitted from the data transmission unit 38 is received (S10). Then, the received information of the partial model formula and the financial data are output to the rating etc. calculating unit 30 (S11).
[0061]
In the rating etc. calculating section 30, the financial data output from the data capturing section 28 is substituted into the partial model formula output from the data capturing section 28. The calculation result is output from the rating etc. calculating unit 30 to the data transmitting unit 32 (S12).
[0062]
Thus, only the partial model formula is transmitted to each computer server 14. Therefore, none of the computer servers 14 discloses the entire model of the credit risk evaluation model.
[0063]
The data transmission unit 32 acquires the calculation result output from the rating etc. calculation unit 30 in step S12. Then, the obtained calculation result is transmitted to the evaluator server 16 via the communication network 10 (S13).
[0064]
Each calculation result transmitted from the data transmission unit 32 of each computer server 14 in this way is received by the data acquisition unit 34 via the communication network 10 (S14). Further, the respective calculation results are output from the data capturing unit 34 to the rating result integrating unit 40 (S15).
[0065]
The rating result integration unit 40 integrates the respective calculation results output from the data acquisition unit 34, substitutes the financial data into the model formula applied to the credit risk evaluation model, and obtains the final calculation result. Further, the calculation result is stored in the rating result storage unit 42 (S16).
[0066]
As described above, the rating result integration unit 40 only performs the integration process of the calculation results transmitted from the data transmission unit 32 of each of the computer servers 14 entrusted with the calculation, and the content of the model formula is not limited to the partial model. I do not know the contents of the formula.
[0067]
The calculation result stored in the rating result storage unit 42 and the bankruptcy information data of this company stored in the bankruptcy information storage unit 44 are acquired by the model accuracy evaluation unit 46 (S17, S18). Further, information on the evaluation rules stored in advance in the evaluation rule storage unit 48 is obtained (S19). Then, the accuracy of the credit risk evaluation model is evaluated by comparing a pair of the calculation result obtained in step S17 and the result of bankruptcy determined from the bankruptcy information data obtained in step S18 with an evaluation criterion. Then, the evaluation result data is output from the model accuracy evaluation section 46 to the evaluation result transmission section 50 (S20).
[0068]
The evaluation by the model accuracy evaluation unit 46 as described above is performed for many companies, and the accuracy of the credit risk evaluation model is evaluated by comparison with the result of the bankruptcy information data.
[0069]
The evaluation result data output from the model accuracy evaluation unit 46 is transmitted by the evaluation result transmission unit 50 to the owner server 12 via the communication network 10 (S21).
[0070]
The evaluation result data transmitted from the evaluation result transmitting unit 50 in step S21 is received by the evaluation result capturing unit 26 provided in the owner server 12 (S22). Then, by referring to the evaluation result data by the evaluator, the accuracy of the credit risk evaluation model is grasped by the evaluator.
[0071]
As described above, in the credit risk evaluation model accuracy evaluation system to which the credit risk evaluation model accuracy evaluation method according to the present embodiment is applied, each computer server 14 has a partial model expression Only sent. Therefore, none of the computer servers 14 discloses the entire model of the credit risk evaluation model.
[0072]
Further, the evaluator server 16 only performs the integration processing of the operation results transmitted from the data transmission unit 32 of each of the evaluator servers 14, and cannot know the contents of the model formulas or the partial model formulas.
[0073]
As a result, the owner of the credit risk evaluation model can outsource the accuracy evaluation to each of the calculators and evaluators without disclosing the entire contents of the credit risk evaluation model.
[0074]
(Second embodiment)
A second embodiment of the present invention will be described with reference to FIGS.
[0075]
In the first embodiment, the model formula of the credit risk evaluation model is a partial model formula (1) for calculating the profitability points A and B as shown in FIG. Although a description has been given of the case where the model is composed of the partial model formula (2) for calculating the C index and the D index, in the present embodiment, for example, the model formula of the credit risk evaluation model is shown below. The case represented by the equations shown from (i) to (vii) will be described.
Z = KA1 + KA2 + KA3 + KA4 + KA5 + KA6 + C (i)
KA1 = a (1) * b (1) (ii)
KA2 = a (2) * b (2) (iii)
KA3 = a (3) * b (3) (iv)
KA4 = a (4) * b (4) (v)
KA5 = a (5) * b (5) (vi)
KA6 = a (6) * b (6) (vii)
Here, a (x): (x = 1 to 6) is a coefficient for each financial index name x as shown in FIG. B (x): (x = 1 to 6) is an index value for each financial index name x as shown in FIG. C is a constant term as shown in FIG.
[0076]
As described above, the credit risk evaluation model accuracy evaluation system to which the credit risk evaluation model accuracy evaluation method according to the present embodiment is applied is a credit risk evaluation model to which the credit risk evaluation model accuracy evaluation method according to the first embodiment is applied. This is a modified example of the risk evaluation model accuracy evaluation system, the configuration of which is the same as that of FIG. In the first embodiment, the operation steps are shown as steps S1 to S22. Hereinafter, only steps having different operations will be described, and description of steps having the same operation will be omitted.
[0077]
That is, in the present embodiment, the model dividing unit 18 divides each of the above equations into two groups consisting of equation (i) and equations (ii) to (vii), and transmits each of the divided groups for data transmission. Output to the unit 20 (S1 '). Note that step S1 'indicates that the content of step S1 in the first embodiment is different in the present embodiment. In the present embodiment, the steps that are the same as those in the first embodiment and the steps that are different are clarified by indicating the step numbers according to such rules.
[0078]
When the formula information of each group is output from the model dividing unit 18, the data transmitting unit 20 converts the formula (i) to, for example, the computer server 14 (# 2), and converts the formulas (ii) to (vii) to, for example, the computer. Each is transmitted to the server 14 (# 1) (S2 ').
[0079]
When acquiring the financial data from the financial data storage unit 36 in step S8, the data transmitting unit 38 transmits the acquired financial data to the computer server 14 (# 1) via the communication network 10 (S9 ').
[0080]
In step S2 ', the data capturing unit 28 of the computer server 14 (# 1) to which the expressions (ii) to (vii) have been transmitted from the data transmitting unit 20 has been transmitted from the data transmitting unit 20 in step S2'. The formulas (ii) to (vii) and the financial data transmitted from the data transmitting unit 38 in step S9 'are received (S10'). Then, the received formulas (ii) to (vii) and the financial data are output to the rating calculation unit 30 (S11 ′).
[0081]
The rating etc. calculating unit 30 of the computer server 14 (# 1) substitutes the financial data output from the data capturing unit 28 for the formulas (ii) to (vii) output from the data capturing unit 28. And outputs the calculation results KA1 to KA6 to the data transmission unit 32 (S12 ').
[0082]
As described above, the formulas (ii) to (vii) are transmitted to the computer server 14 (# 1), but the formula (i) is not transmitted. As a result, the entire model formula of the credit risk evaluation model is not disclosed to the computer server 14 (# 1). Also, it is not clear what kind of calculation KA1 to KA6, which are the results of the calculation, are used.
[0083]
The data transmission unit 32 of the computer server 14 (# 1) acquires the calculation results KA1 to KA6 output from the rating calculation unit 30, and transmits the calculation results KA1 to KA6 to the computer via the communication network 10. Output to the server 14 (# 2) (S13 ').
[0084]
In step S2 ', the data capturing unit 28 of the computer server 14 (# 2) to which the formula (i) has been transmitted from the data transmitting unit 20 transmits the formula (i) transmitted from the data transmitting unit 20 in step S2'. And the calculation results KA1 to KA6 transmitted from the data transmission unit 38 of the computer server 14 (# 1) in step S13 ′ (S10 ″). Then, the received equation (i) and the calculation results KA1 to KA6 are output to the rating etc. calculating unit 30 (S11 ″).
[0085]
The rating etc. calculating unit 30 of the computer server 14 (# 2) performs an operation by substituting the operation results KA1 to KA6 into the expression (i) output from the data capturing unit 28, and transmits the operation result Z to the data transmitting unit. 32 (S12 '').
[0086]
As described above, the computer server 14 (# 2) cannot know at all what kind of formulas and what kind of financial data the calculation results KA1 to KA6 are obtained. Therefore, the entire model formula of the credit risk evaluation model is not disclosed to the computer server 14 (# 2).
[0087]
The data transmission unit 32 of the calculator server 14 (# 2) acquires the calculation result Z output from the rating calculation unit 30, and outputs the calculation result Z to the evaluator server 16 via the communication network 10. (S13 '').
[0088]
The data acquisition unit 34 receives the operation result Z transmitted from the data transmission unit 32 of the computer server 14 (# 2) in this way via the communication network 10 (S14 '), and receives the rating result integration unit 40. (S16 '). That is, in the present embodiment, there is no operation corresponding to step S15 and step S16 in the first embodiment.
[0089]
In step S20 ', the accuracy of the credit risk evaluation model is evaluated based on the calculation result Z obtained in step S17', the bankruptcy information data obtained in step S18, and the evaluation rules obtained in step S19. Examples of the evaluation method include an evaluation based on an AR value and an evaluation based on a comparison with a CRAFT score.
[0090]
The evaluation based on the AR (Accuracy Rate) value is an evaluation method using a statistical index proposed by Moody's. An evaluation method based on the AR value will be described with reference to FIG.
[0091]
FIG. 6 shows the calculation result Z on the horizontal axis and the cumulative bankruptcy rate of the company on the vertical axis. Here, the calculation result Z is a parameter indicating that a larger value indicates a lower bankruptcy probability, and is hereinafter referred to as a score. That is, the probability of bankruptcy is highest on the origin side on the horizontal axis. While the score Z is obtained for many companies, the cumulative ratio of the number of companies that have actually gone bankrupt ideally takes a tendency as indicated by R1 in the figure. That is, a certain low score Z ₀ All companies up to should be bankrupt, so score Z ₀ , The cumulative bankruptcy rate becomes 100%. Therefore, the credit risk evaluation model closer to the tendency as indicated by R1 in the figure has higher accuracy. Conversely, the credit risk evaluation model having the tendency as indicated by R3 in the figure indicating that the score Z is a low value but has not actually gone bankrupt has no discriminating ability, and This indicates that the accuracy is low.
[0092]
Except for the extremely high precision shown in R1 in the figure and the extremely low precision shown in R3 in the figure, practically most credit risk evaluation models have the tendency shown in R2 in the figure, which is intermediate between these. take. In the evaluation based on the AR value, the area surrounded by R1 in the figure and R3 in the figure (S ₁ + S ₂ ), The area (S) surrounded by R2 in the figure and R3 in the figure. ₂ ) Is 0.7 or more (S ₂ / (S ₁ + S ₂ ) ≧ 0.7), it is determined that the credit risk evaluation model has sufficient accuracy.
[0093]
An evaluation method based on comparison with the CRAFT score will be described with reference to FIG. The details of the CRAFT score are described in Non-Patent Document 1 described above.
[0094]
According to the logit model, using the score Z, the bankruptcy probability p of the company is expressed by the following equation.
p = 1 / (1 + exp (-Z))
Therefore, the bankruptcy probability p evaluated by the credit risk evaluation model can be obtained by substituting the score Z into the logit model. FIG. 7 shows the bankruptcy probability p obtained for the companies M1 to M5 in this way. ₂ 3 shows an example of the result.
[0095]
Further, bankruptcy probabilities for the same companies M1 to M5 are also obtained from the CRAFT score described in Non-Patent Document 1 described above. FIG. 7 shows the bankruptcy probability p based on the CRAFT score. ₁ Examples of the results are also shown.
[0096]
Note that “bankruptcy or non-bankruptcy” shown in FIG. ₀ Indicates that "1" indicates bankruptcy and "0" indicates non-bankruptcy. The information indicating bankruptcy or non-bankruptcy is the bankruptcy information data itself acquired in step S18.
[0097]
Based on the results as shown in FIG. ₁ And bankruptcy probability p ₂ Is calculated according to the following equation.
[0098]
(Equation 1)

[0099]
The likelihood L obtained according to the above equation is the case of the CRAFT score
L = (1-0.1) × 0.3 × (1-0.2) × (1-0.1) × 0.5 = 0.0972,
Credit risk assessment model
L = (1-0.3) × 0.2 × (1-0.1) × (1-0.2) × 0.3 = 0.03024
[0100]
Since a model with a higher likelihood is a better model, it can be determined in this case that the CRAFT score is a better model.
[0101]
Next, the operation of the credit risk evaluation model accuracy evaluation system to which the credit risk evaluation model accuracy evaluation method according to the present embodiment configured as described above is applied will be described with reference to the processing flowchart shown in FIG.
[0102]
First, when the accuracy evaluation of the credit risk evaluation model is performed by the credit risk evaluation model accuracy evaluation system to which the accuracy evaluation method of the credit risk evaluation model according to the present embodiment is applied, the credit risk evaluation model For example, a model formula including formulas (i) to (vii) used in the model is divided into two groups including formula (i) and formulas (ii) to (vii). Is output to the data transmission unit 20 (S1 ′).
[0103]
When the formula information of each group is output from the model dividing unit 18 to the data transmitting unit 20, the formula (i) is converted to, for example, the computer server 14 (# 2), and the formulas (ii) to (vii) are converted to, for example, the computer. Each is transmitted to the server 14 (# 1) (S2 ').
[0104]
On the other hand, from the financial data stored in the financial data storage unit 22, the financial data of the company used for the accuracy evaluation and the bankruptcy information data of this company are obtained from the bankruptcy information storage unit 24 by the data transmission unit 20, respectively. (S3). These acquired data are transmitted to the evaluator server 16 via the communication network 10 (S4).
[0105]
The financial data and the bankruptcy information data transmitted from the data transmitting unit 20 are captured by the data capturing unit 34 (S5). Then, the financial data is stored from the data acquisition unit 34 to the financial data storage unit 36 (S6), and the bankruptcy information data is stored from the data acquisition unit 34 to the bankruptcy information storage unit 44 (S7).
[0106]
When the financial data is stored in the financial data storage unit 36, the stored financial data is acquired by the data transmission unit 38 (S8), and the acquired financial data is transmitted to the computer server 14 (#) via the communication network 10. 1) (S9 ').
[0107]
Next, in step S2 ', the data acquisition unit 28 of the computer server 14 (# 1) to which the formulas (ii) to (vii) have been transmitted from the data transmission unit 20 transmits data from the data transmission unit 20 in step S2'. The transmitted equations (ii) to (vii) and the financial data transmitted from the data transmitting unit 38 in step S9 'are received (S10'). Then, the received equations (ii) to (vii) and the financial data are output to the rating etc. calculating unit 30 (S11 ′).
[0108]
In the rating etc. calculating unit 30 of the computer server 14 (# 1), the calculation in which the financial data output from the data capturing unit 28 is substituted into the equations (ii) to (vii) output from the data capturing unit 28. Are performed, and the calculation results KA1 to KA6 are output to the data transmission unit 32 (S12 ').
[0109]
As described above, the formulas (ii) to (vii) are transmitted to the computer server 14 (# 1), but the formula (i) is not transmitted. As a result, the entire model formula of the credit risk evaluation model is not disclosed to the computer server 14 (# 1). Also, it is not clear what kind of calculation KA1 to KA6, which are the results of the calculation, are used.
[0110]
The data transmission unit 32 of the computer server 14 (# 1) acquires the calculation results KA1 to KA6 output from the rating etc. calculation unit 30. Then, the calculation results KA1 to KA6 are output to the computer server 14 (# 2) via the communication network 10 (S13 ').
[0111]
In step S2 ', the data capturing unit 28 of the computer server 14 (# 2) to which the formula (i) has been transmitted from the data transmitting unit 20 has the formula (i) transmitted from the data transmitting unit 20 in step S2'. And the calculation results KA1 to KA6 transmitted from the data transmission unit 38 of the computer server 14 (# 1) in step S13 ′ (S10 ″). Then, the received equation (i) and the calculation results KA1 to KA6 are output to the rating etc. calculating unit 30 (S11 ″).
[0112]
In the rating etc. calculating section 30 of the computer server 14 (# 2), the calculation is performed by substituting the calculation results KA1 to KA6 into the equation (i) output from the data capturing section 28, and the calculation result Z is The data is output to the transmission unit 32 (S12 ″).
[0113]
As described above, the formula (i) is transmitted to the computer server 14 (# 2), but the financial data is not transmitted. As a result, the computer server 14 (# 2) cannot know at all what kind of formulas and financial data are used for the calculation results KA1 to KA6. Therefore, the entire model formula of the credit risk evaluation model is not disclosed to the computer server 14 (# 2).
[0114]
The data transmission unit 32 of the computer server 14 (# 2) acquires the operation result Z output from the rating calculation unit 30, and outputs the operation result Z to the evaluator server 16 via the communication network 10. Is performed (S13 ″).
[0115]
The data capturing unit 34 receives the operation result Z transmitted from the data transmitting unit 32 of the computer server 14 (# 2) in this way via the communication network 10 (S14 '), and the rating result integrating unit 40 (S16 ').
[0116]
The model accuracy evaluation unit 46 obtains the calculation result stored in the rating result storage unit 42 and the bankruptcy information data of this company stored in the bankruptcy information storage unit 44 (S17, S18). Further, information on the evaluation rules stored in advance in the evaluation rule storage unit 48 is obtained (S19).
[0117]
In step S20 ', the accuracy of the credit risk evaluation model is evaluated based on the calculation result Z obtained in step S17', the bankruptcy information data obtained in step S18, and the evaluation rule obtained in step S19. (S20 '). Examples of the evaluation method include an evaluation based on an AR value and an evaluation based on a comparison with a CRAFT score. The accuracy evaluation of the credit risk evaluation model is performed based on the evaluation result as described above.
[0118]
In the evaluation result transmitting unit 50, the evaluation result data output from the model accuracy evaluating unit 46 is transmitted to the owner server 12 via the communication network 10 (S21).
[0119]
The evaluation result data transmitted from the evaluation result transmitting unit 50 in step S21 is received by the evaluation result capturing unit 26 provided in the owner server 12 (S22). Then, by referring to the evaluation result data by the evaluator, the accuracy of the credit risk evaluation model is grasped by the evaluator.
[0120]
As described above, in the credit risk evaluation model accuracy evaluation system to which the accuracy evaluation method of the credit risk evaluation model according to the present embodiment is applied, the above-described operation causes the computer server 14 (# 1) to use the expression Although (ii) to (vii) are transmitted, equation (i) is not transmitted. As a result, the entire model formula of the credit risk evaluation model is not disclosed to the computer server 14 (# 1). Also, it is not clear what kind of calculation KA1 to KA6, which are the results of the calculation, are used.
[0121]
On the other hand, although the formula (i) is transmitted to the computer server 14 (# 2), the financial data is not transmitted. As a result, the computer server 14 (# 2) cannot know at all what kind of formulas and financial data are used for the calculation results KA1 to KA6. Therefore, the entire model formula of the credit risk evaluation model is not disclosed to the computer server 14 (# 2).
[0122]
Further, the evaluator server 16 notifies only the calculation result from the calculator server 14 (# 2), and evaluates the accuracy of the credit risk evaluation model based on the calculation result. Therefore, there is no way to know the contents of the credit risk evaluation model formula at all.
[0123]
As a result, the owner of the credit risk evaluation model can outsource the accuracy evaluation to each of the calculators and evaluators without disclosing the entire contents of the credit risk evaluation model.
[0124]
Although the preferred embodiments of the present invention have been described with reference to the accompanying drawings, the present invention is not limited to such configurations. Within the scope of the invented technical concept of the claims, those skilled in the art will be able to conceive various changes and modifications, and those changes and modifications will be described in the technical scope of the present invention. It is understood that it belongs to.
[0125]
【The invention's effect】
As described above, according to the present invention, the model evaluator can use the model evaluator without releasing the credit risk evaluation model and without the model evaluator holding financial data of many companies. It is possible to realize a credit risk evaluation model accuracy evaluation system and an accuracy evaluation method capable of performing an accuracy evaluation of a credit risk evaluation model entrusted from a company.
[Brief description of the drawings]
FIG. 1 is a functional block diagram showing a configuration example of a credit risk evaluation model accuracy evaluation system to which a credit risk evaluation model accuracy evaluation method according to a first embodiment is applied.
FIG. 2 is a diagram for explaining indices based on profitability points and safety points.
FIG. 3 is a diagram showing an example of financial data.
FIG. 4 is a processing flowchart showing an operation of the credit risk evaluation model accuracy evaluation system to which the credit risk evaluation model accuracy evaluation method according to the first embodiment is applied.
FIG. 5 is a diagram showing an example of a financial index and its coefficient.
FIG. 6 is a diagram showing a relationship between a score used for evaluation based on an AR value and a cumulative bankruptcy rate.
FIG. 7 is a view for explaining comparative evaluation with a CRAFT score.
FIG. 8 is a processing flow chart showing an operation of a credit risk evaluation model accuracy evaluation system to which a credit risk evaluation model accuracy evaluation method according to a second embodiment is applied.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 10 ... Communication network, 12 ... Owner server, 14 ... Computer server, 16 ... Evaluator server, 18 ... Model division | segmentation part, 20 ... Data transmission part, 22 ... Financial data storage part, 24 ... Bankruptcy information storage part, 26 ... Evaluation result acquisition unit, 28 ... Data acquisition unit, 30 ... Attachment calculation unit, 32 ... Data transmission unit, 34 ... Data acquisition unit, 36 ... Financial data storage unit, 38 ... Data transmission unit, 40 ... Attachment Result integration unit, 42: attached result storage unit, 44: bankruptcy information storage unit, 46: model accuracy evaluation unit, 48: evaluation rule storage unit, 50: evaluation result transmission unit

Claims (4)

An owner who owns a credit risk evaluation model for evaluating the bankruptcy probability of a company delegates the accuracy evaluation of the credit risk evaluation model to two or more calculators and one evaluator. A method for causing the evaluator to evaluate the accuracy of the credit risk evaluation model without disclosing the entire contents of the credit risk evaluation model for any of the calculators and evaluators,
The model formula used for the credit risk evaluation model is divided into the same number of partial model formulas as the calculator, each of the divided partial model formulas, and data for credit risk evaluation calculation substituted into each of the partial model formulas. Is given to each of the calculators,
For each of the calculators, by performing the credit risk evaluation calculation by substituting the data in the partial model formula,
The evaluator is configured to compare the result of the credit risk evaluation calculation by each of the evaluators with a predetermined accuracy evaluation criterion, and to evaluate the accuracy of the credit risk evaluation model based on the comparison result. Evaluation method. The owner of a credit risk assessment model for assessing the probability of bankruptcy of a company may assign the accuracy assessment of said credit risk assessment model to one or more partial calculators, one total calculator, and one evaluator. By allowing the evaluator to evaluate the accuracy of the credit risk evaluation model without disclosing the entire contents of the credit risk evaluation model to each of the evaluator and the evaluator, The method
To each of the one or more partial calculators, each of the terms constituting a model formula used in the credit risk evaluation model and data for calculating a credit risk corresponding to each of the terms is given. ,
For each of the partial calculators, the value of each item is calculated by substituting the data for each item,
For the total calculator, a credit risk evaluation calculation is performed by substituting the value of each term calculated by each of the partial calculators into a model formula constituted by the respective terms,
An accuracy evaluation method, wherein the evaluator compares the result of the credit risk evaluation calculation with a predetermined accuracy evaluation criterion, and evaluates the accuracy of the credit risk evaluation model based on the comparison result. A system for evaluating the accuracy of the credit risk evaluation model commissioned by an owner who owns a credit risk evaluation model for evaluating the probability of bankruptcy of a company without grasping the entire contents of the credit risk evaluation model, ,
Acquisition means for acquiring a credit risk evaluation calculation made by substituting data for accuracy evaluation, for each of a plurality of partial model expressions constituting a model expression used in the credit risk evaluation model,
Integrating means for integrating the results of each credit risk evaluation calculation obtained by the obtaining means,
A comparison unit that compares the result integrated by the integration unit with a predetermined accuracy evaluation criterion,
An evaluation means for evaluating the accuracy of the credit risk evaluation model based on a comparison result by the comparing means. A system for evaluating the accuracy of the credit risk evaluation model commissioned by an owner who owns a credit risk evaluation model for evaluating the probability of bankruptcy of a company without grasping the entire contents of the credit risk evaluation model, ,
Acquisition means for acquiring a credit risk evaluation calculation made by substituting data for accuracy evaluation into a model formula used for the credit risk evaluation model,
A comparison unit that compares the result integrated by the acquisition unit with a predetermined accuracy evaluation criterion,
An evaluation means for evaluating the accuracy of the credit risk evaluation model based on a comparison result by the comparing means.

Images