KR102633684B1 - Method for automatic credit scoring calibration considering macro-economic index and recofding medium storing program to execute the method - Google Patents

Abstract

A method of adjusting a credit rating model that automatically reflects economic conditions is provided, which includes: a first step in which a computer prepares a model that generates a monthly credit rating score based on TTC (Through-The-Cycle); A second step in which the computer calculates the structural risk and relative risk of the TTC-based monthly credit rating score; A third step in which the computer generates a correlation between the structural risk amount and the relative risk amount calculated in the second step and a macroeconomic indicator; And a fourth step in which the computer converts the TTC-based monthly credit rating score into a credit rating score calculated using the correlation equation generated in the third step.

Description

A method of adjusting a credit rating model that automatically reflects economic conditions and a recording medium storing a program for executing the method {METHOD FOR AUTOMATIC CREDIT SCORING CALIBRATION CONSIDERING MACRO-ECONOMIC INDEX AND RECOFDING MEDIUM STORING PROGRAM TO EXECUTE THE METHOD}

The present disclosure relates to a method of adjusting a credit rating model that automatically reflects the game situation and a recording medium storing a program for executing the method. More specifically, it relates to an approval strategy through credit rating score adjustment that automatically reflects the game situation. It relates to a credit rating model adjustment method that enables objective and stable risk management without a separate decision on change, and a recording medium that stores a program that executes the method.

The credit rating model also needs to be adjusted in relation to the rapid economic fluctuations coupled with the recent interest rate hike. The reason is that if the difference between the data used when developing a credit rating model and the data obtained during actual operation increases, the discrimination and stability of the model cannot be maintained at the level expected at the time of initial model development.

In the field of data science, such model fluctuations are called data drift. The data drift referred to here includes both feature drift, which refers to changes in input data, and label drift, which refers to changes in target values (defect rate/delinquency rate).

For example, when the loan interest rate increases, the principal and interest repayment amount increases, and accordingly, the DSR (principal and interest repayment amount to income ratio) increases, which may lead to a decrease in disposable income. In terms of feature drift, in order to increase insufficient cash liquidity, the holding ratio of cash advances or low-income credit loans (long-term card loans, savings banks/capital/loans, etc.) may increase, and in terms of label drift, the potential insolvency rate of customers may occur. This means that the ratio of (delinquent payment by 30 days or more) or defect rate (delinquency by more than 90 days, registration of Shin Jeong-won/CB company default) is increasing.

Depending on these changes in macro conditions, changes in the loan market, and changes in the meaning of items, the amount of risk may vary even if the credit score is the same. For example, although the average defect rate for a customer with a score of 800 is expected to be 2%, it may be 3% under certain economic conditions, or 1% under more favorable economic conditions.

Considering the fluctuations in real risk due to the characteristics of the economic cycle, financial companies do not formulate approval strategies based on cut-off standards of fixed credit evaluation scores, but rather develop loan approval strategies through complex decision-making. or adjust the loan interest rate and limit. However, the impact of these adjustments tends to depend on the qualitative experience of the risk management department or the internal decision-making system rather than being determined through the relationship between past economic fluctuations and credit rating models.

The present invention is intended to support objective and stable risk management without a separate decision to change the approval strategy through automated credit evaluation score adjustment that reflects the economic situation.

Existing credit rating models are divided into service methods according to the ideology of Point-In-Time (PIT) and Through-The-Cycle (TTC). PIT refers to a credit score model that evaluates a borrower's creditworthiness at a specific point in time. The model considers the borrower's credit history, income, debt-to-income ratio, and other factors to determine the likelihood that the borrower will default on the loan within a certain period of time. The PIT model can generally be used for short-term loans such as credit cards or personal loans. TTC, on the other hand, is a credit scoring model that assesses credit risk over long periods of time, typically business cycles or several years. The TTC model takes into account economic and market conditions that may affect a borrower's ability to repay debt over the long term. The TTC model can generally be used for long-term loans such as mortgages.

In summary, the PIT model focuses on assessing credit risk at a specific point in time, while the TTC model looks more broadly at credit risk over a longer period of time. Both approaches are useful for credit evaluation and can be used for different types of loans depending on the loan term and characteristics.

Speaking of risky characteristics, in the case of the credit evaluation score developed by TTC, the credit evaluation score does not change during a down-turn in the game, and the average defect rate of the credit evaluation score increases. On the other hand, in the case of credit evaluation scores developed with PIT, the credit evaluation score tends to change to a credit evaluation score according to the standard defect rate or default rate (PD) during a down-turn in the economy. In other words, in the PIT standard, the default rate of the rating or credit score section is not sensitive to the economy and the component ratio is sensitive to the economy, but in TTC, the default rate of the rating or credit score section is sensitive to the economy and the component ratio is not sensitive to the economy.

Figure 1 is a schematic diagram of the composition ratio and PD (default rate) of PIT and TTC.

Referring to Figure 1, in general, corporate loans are developed according to the TTC ideology, and retail loans are basically developed according to the PIT ideology. However, even in most PITs, the TTC ideology is generally included. For example, from the perspective of data drift, if feature drift occurs due to economic fluctuations, changes in credit rating scores may occur similar to PIT, and if changes in target value occur, TTC Depending on the idea, the effect of the average defect rate of the same credit evaluation score can be felt.

At this time, in practice, the credit score is used in a short-term loan approval strategy to approve or reject a customer's loan application or credit card application, and basically has a frame of rejecting customers with a score below a certain level. And, when the score for rejecting such customers is called a cut-off, the cut-off is set at a level that can manage the failure rate of approved customers below a certain level. However, in the credit evaluation score of the PIT system, there is no major problem without adjusting the cut-off according to the game situation because the risk amount for each score is maintained, but in the TTC system, the defect rate based on the same cut-off standard changes, so it is dependent on the game situation. If the cut-off is not adjusted accordingly, risk management indicators such as delinquency rate are bound to change.

In summary, if you follow the loan decision based on the defect rate in the TTC ideology, you may underestimate the possibility of default or be unable to determine an appropriate loan interest rate, so the process of converting the credit rating score in the TCC ideology to a form similar to the PIT ideology is necessary. It may be necessary.

The present disclosure aims to solve the needs and/or problems described above.

This disclosure converts the credit rating score developed based on the existing TTC ideology into a form similar to the PIT ideology by utilizing a technique for dynamically adjusting the credit score by reflecting the economic situation, thereby changing the approval strategy in the event of economic fluctuations. We provide a way to maintain the customer's risk level without risk.

In one form of the present disclosure, a method for adjusting a credit rating model that automatically reflects economic conditions is provided, wherein a computer prepares a model that generates a monthly credit rating score based on Through-The-Cycle (TTC). first step; A second step in which the computer calculates the structural risk and relative risk of the TTC-based monthly credit rating score; A third step in which the computer generates a correlation between the structural risk amount and the relative risk amount calculated in the second step and a macroeconomic indicator; And a fourth step in which the computer converts the TTC-based monthly credit rating score into a credit rating score calculated using the correlation equation generated in the third step.

In one embodiment, the model for generating the TTC-based monthly credit rating score prepared in the first step may be developed based on a logistic regression analysis model.

In one embodiment, the structural risk calculated in the second step is defined as the y-intercept value of a linear relationship having the TTC-based monthly credit rating score as the x-axis value and the logit value as the y-axis value, The relative risk amount is defined as the slope of a linear relationship that has the TTC-based monthly credit rating score as the x-axis value and the logit value as the y-axis value, and the logit value may be the logarithm of odds.

In one embodiment, the correlation equation generated in the third step is generated by a model that has the structural risk amount and the relative risk amount calculated in the second step as dependent variables and the macroeconomic indicator as an independent variable. It could be.

In one embodiment, the model that generates the correlation equation refines the macroeconomic indicator, ensures normality, confirms causality between the independent variable and the dependent variable, and determines whether the independent variable is the dependent variable. It may be developed by selecting a significant time point that affects the variable and creating a regression model using the dependent variable and the independent variable of the refined optimal time lag.

In one embodiment, refining the macroeconomic indicator may include checking seasonality included in the macroeconomic indicator and removing the seasonality if present.

In one embodiment, confirming causality between the independent variable and the dependent variable may be performed through causality analysis through Granger causality.

In one embodiment, selecting a significant time point at which the independent variable affects the dependent variable may be performed by a method of selecting a significant time point through cross-correlation analysis.

In one embodiment, the macroeconomic indicator may include items such as economic growth rate, unemployment rate, inflation rate, domestic stock price, housing price increase/decrease rate in the metropolitan area, and credit loan interest rate.

In another form of the present disclosure, a computer-readable recording medium is provided that records a computer program for executing each step of any one of the above-described methods to adjust a credit rating model to automatically reflect the game situation. .

According to this disclosure, the credit rating score developed with the existing TTC is converted into a form similar to the PIT by utilizing a technique for dynamically adjusting the credit score by reflecting the economic situation, and through this, the credit rating score is converted into a form similar to the PIT in the event of economic fluctuations without changing the approval strategy. , can provide a way to maintain the risk level of approved customers.

Figure 1 is a schematic diagram of the composition ratio and PD (default rate) of PIT and TTC.
Figure 2 is a flowchart illustrating a method for adjusting a credit rating model that automatically reflects economic conditions according to the present disclosure.
Figure 3 is a diagram showing structural risk changes and relative risk changes to explain the credit rating model adjustment method according to the present disclosure.
Figure 4 is a diagram for explaining the score conversion concept applied to the credit rating model adjustment method according to the present disclosure.
Figure 5 is a diagram illustrating a data purification process applying the credit rating model adjustment method according to the present disclosure.
Figure 6 is a diagram for explaining the process of measuring the structural risk amount and relative risk amount for each score by applying the credit rating model adjustment method according to the present disclosure.
7A and 7B are diagrams showing structural risk prediction results and relative risk prediction results, respectively, by applying the credit rating model adjustment method according to the present disclosure.
FIG. 8 is a diagram illustrating a dynamic score adjustment process applying the credit rating model adjustment method according to the present disclosure.

The advantages and features of the present disclosure and methods for achieving them will become clear with reference to the embodiments described in detail below in conjunction with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below and will be implemented in various different forms, and the embodiments are merely intended to ensure that the disclosure of the present disclosure is complete and that those skilled in the art It is provided to fully inform the person of the scope of the invention, and the present disclosure is defined only by the scope of the claims.

The terminology used herein is only used to describe specific embodiments and is not intended to limit the disclosure. For example, a component expressed in the singular should be understood as a concept that includes plural components unless the context clearly indicates only the singular. In addition, in the specification of the present disclosure, terms such as 'include' or 'have' are only intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, and such The use of the term does not exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

Additionally, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which this disclosure pertains.

Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless clearly defined in the specification of the present disclosure, they shall be interpreted in an ideal or excessively formal sense. It doesn't work.

Hereinafter, embodiments of the present disclosure will be described in more detail with reference to the attached drawings. However, in the following description, detailed descriptions of well-known functions or configurations will be omitted if there is a risk of unnecessarily obscuring the gist of the present disclosure.

It will be understood that the prior literature described in this disclosure is incorporated herein by reference in its entirety, and that a person of general knowledge in the art can apply the content described in the prior literature to the portions briefly described in this disclosure. .

Hereinafter, a credit rating model adjustment method that automatically reflects economic conditions according to the present disclosure will be described with reference to the drawings.

Figure 2 is a flowchart of a credit rating model adjustment method that automatically reflects economic conditions according to an embodiment of the present disclosure.

The method of adjusting a credit rating model that automatically reflects the economic situation according to an embodiment of the present disclosure is a first step in which a computer prepares a model for generating a TTC (Through-The-Cycle)-based monthly credit rating score ( S201); A second step in which a computer calculates the structural risk and relative risk of the TTC-based monthly credit rating score (S203); A third-stage process (S205) in which a computer generates correlations between the structural risk amount and relative risk amount calculated in the second stage and macroeconomic indicators; It includes a fourth step (S207) in which the computer converts the TTC-based monthly credit rating score into a credit rating score calculated using the correlation equation generated in the third step.

1. Method of adjusting the credit rating model to automatically reflect economic conditions

Hereinafter, with reference to FIGS. 2 to 8, each step of the credit rating model adjustment method that automatically reflects the economic situation according to an embodiment of the present disclosure will be described.

(1) Step of preparing a model to generate TTC-based monthly credit evaluation score (S201)

The method for adjusting a credit rating model that automatically reflects economic conditions according to the present disclosure includes, as a first step, preparing a model for generating a TTC-based monthly credit rating score (S201).

In one embodiment, the model that generates the TTC-based monthly credit rating score in this step may be developed based on a logistic regression model. The logistic regression model is a model that linearly regresses the logit value (log of odds), and in this model, the credit evaluation score and the logit value appear in a linear relationship.

In one embodiment, the model that generates the TTC-based monthly credit score in this step may be prepared using an artificial neural network model taking retraining into account. In one embodiment, the artificial neural network models used in the model that generates the TTC-based monthly credit score are Convolutional Neural Network (CNN), Deep Neural Network (DNN), and Recurrent Neural Network. , RNN), Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), Bidirectional Recurrent Deep Neural Network (BRDNN), Variational Auto Encoder (VAE) ) or a Deep Q-Network, etc., or a combination thereof, but is not limited thereto.

However, as explained previously, when the economic situation changes rapidly, the difference between the data used to develop the model to generate the TTC-based monthly credit rating score and the data obtained from actual operation increases, resulting in a greater difference than expected at the time of initial model development. Since the level of model discrimination and stability cannot be maintained, it is necessary to additionally perform the second to fourth steps of the present disclosure as described later.

(2) Calculating the structural risk amount and relative risk amount of the TTC-based monthly credit evaluation score (S203)

The credit rating model adjustment method that automatically reflects the economic situation according to the present disclosure includes, as a second step, a step (S203) in which a computer calculates the structural risk amount and relative risk amount of the TTC-based monthly credit rating score. .

As described above, the model that generates the credit rating score applied to the credit rating model adjustment method according to the present disclosure may be developed based on a logistic regression analysis model, and in this model, the score and the logit value appear in a linear relationship. do.

Figure 3 is a diagram showing structural risk changes and relative risk changes to explain the credit rating model adjustment method according to the present disclosure.

Referring to Figure 3, in this step, the relative risk amount is defined as the slope of a linear relationship with the monthly credit evaluation score as the x-axis value and the logit value as the y-axis value, and the structural risk amount is the monthly credit evaluation It is defined as the y-intercept value of a linear relationship with the score as the x-axis value and the logit value as the y-axis value.

Odds are defined as good/bad, so if the odds are high, credit can be considered to be relatively good. Therefore, it can be defined that the greater the amount of structural risk, the better the economic situation. In addition, the relative risk amount refers to the difference between the risk level of customers with a high credit score and the risk level of customers with a low credit score. Depending on changes in economic conditions, the relative risk of the risk-vulnerable class compared to the risk-stable class is determined by how much change in credit risk there is. It can be interpreted as a quantitatively measured value.

In the process of calculating logit values for each score range and finding the relationship between scores and logit, the structural risk amount and relative risk amount may appear differently depending on how many points are set as the unit for grouping scores. Therefore, in the credit rating model adjustment method according to the present disclosure, as an example, score ranges can be grouped into sections with a minimum composition ratio or number of cases.

In the case of customers who do not have a credit score due to no credit activity, the credit score may be set to 0, and in the case of customers who are in default (customers who are already listed as delinquent), a specific score (less than 350 points based on NICE rating information) may be assigned. there is. Since customers receiving such policy scores may act as noise, in the credit evaluation model adjustment method according to the present disclosure, as an example, such customers' scores may be excluded when measuring the amount of risk.

(3) Step of generating correlation between structural risk amount and relative risk amount and macroeconomic indicators (S205)

The credit rating model adjustment method that automatically reflects the economic situation according to the present disclosure is a third step, in which a computer generates correlations between the structural risk amount and relative risk amount calculated in the second step and macroeconomic indicators. Includes step S205.

Specifically, in this step, a model is created with structural risk and relative risk as dependent variables and macroeconomic indicators as independent variables.

In one embodiment, the method of creating such a model may use machine learning techniques, such as artificial neural networks, rather than standard time series analysis methods.

Alternatively, in another embodiment, in order to solve the problem of false fit due to the characteristics of macroeconomic indicators when using machine learning techniques, the model may be generated using the method below.

① Refining macroeconomic indicators: Due to the nature of macroeconomic indicators, they may have noise and seasonality. For example, in the case of the unemployment rate, frictional fluctuations occur at the end and beginning of each year, so this can be corrected by the average value of the previous month and the following month. Noise is removed using HF filters/wavelet filters, etc., and seasonality is checked through X12 ARIMA based on the moving average method developed by the US Census Bureau or TRAMO-SEATS based on the ARIMA model developed by the Bank of Spain. If present, it can be removed.

② Securing normality: If the items used in the dependent and independent variables have a unit root, there may be a pseudo-fit when fitting the model. Here, a pseudo-fit means that if an unstable time series with a unit root is used as is, the number of samples can be reduced. As it increases, the t-value of the regression coefficient also increases, indicating that there is a very strong correlation even between uncorrelated variables.

③ Causality confirmation: Causality analysis can be performed to identify variables that affect independent variables among a large number of dependent variables. This disclosure proposes causality analysis through Granger causality. Granger causality refers to the causal relationship between two groups of variables and can be used to identify exogenous or endogenous relationships between two groups of variables. In other words, when predicting x(t+k) at time t, the prediction error when using all past values of the two variables x(t) and y(t) is the same as when predicting x(t) using its own past data. If it is less than the prediction error, the random variable {y(t)} can be said to have Granger causality with respect to {x(t)}.

When defining RSS ₀ as the sum of squares when predicting x(t) alone, and RSS ₁ as the sum of squares when predicting with x(t) and y(t), the sum of squares in the entire period T, period p to check causality value created with follows the F(p, T-2p-1) distribution.

In this disclosure, variables with a P-Value that does not exceed the significance level based on the F distribution can be used when fitting a model.

④ Determination of optimal time lag: Since the time lag that affects the dependent variable is different for each independent variable, the most significant time point can be selected through cross-correlation analysis. However, this process can be omitted when fitting a model that includes multiple lags, such as VAR, or when learning by incorporating multiple lags into a machine learning model.

For example, Table 1 below is the result of an actual cross-correlation analysis. When considering the sign relationship (positive/negative) with the dependent variable, the actual past lag is increased and the lag with the highest correlation coefficient is selected. In the credit rating model adjustment method according to the present disclosure, as an example, items for which the sign of the correlation does not match may not be selected even if the value of the correlation coefficient is high.

[Table 1]

⑤ Model fitting: Develop a regression model using refined dependent and independent variables of optimal lag. The regression model can utilize simple linear multiple regression analysis, or it can also utilize machine learning models such as XGBoost or artificial neural networks. However, in the case of a macro model, it must be created while considering the relationship between the dependent variable and the independent variable, so in the case of a machine learning model where it is difficult to model the relationship with the variable in reality, the expected relationship after learning is comprehensively reviewed to create the final model. can be confirmed. For example, if the economic growth rate increases, the intuition that an overall increase in credit scores is expected can be expected, and the direction of the model actually created can be verified. Preferably, the directionality of a machine learning model can be verified using Integrated Gradients in the case of artificial neural networks, and explainable AI techniques such as SHAP or LIME (e.g. XAI) can be used for tree-based models such as XGBoost or LightGBM. You can also use etc.

(4) Converting the TTC-based monthly credit rating score into a credit rating score calculated using a correlation equation (S207)

The method for adjusting the credit rating model that automatically reflects the economic situation according to the present disclosure is the fourth step, which is a credit rating score calculated by a computer using the correlation equation generated in the third step, and is a TTC-based monthly credit rating score. It includes a step of converting (S207).

Figure 4 is a diagram to explain the concept of this step, where the TTC-based monthly credit rating score is indicated by a dotted line, and the credit rating score calculated using the correlation equation generated through the process of the third step described above. is indicated by a solid line. Hereinafter, the credit evaluation score calculated using the correlation equation generated through the above-described third step process will be simply referred to as “target credit evaluation score” or “target score.”

Referring to Figure 4, according to the TTC-based monthly credit rating score (dotted line), the score of S ₀ corresponds to the logit value (log of odds) of L ₀ , but the target score (or target credit rating score) According to , the score corresponding to the logit value of L ₀ corresponds to S ₁ .

Therefore, the credit rating model adjustment method according to the present disclosure performs processing to convert the score of S ₀ into the score of S ₁ through the process of this step, and the same method is processed for all TTC-based monthly credit rating scores. It is carried out.

Therefore, the TTC-based monthly credit evaluation score is dynamically adjusted in this form every month, and accordingly, the credit evaluation score that had a different defect rate for each monthly credit evaluation score (for example, changes in the corresponding defect rate due to economic fluctuations) is changed to external circumstances. It has the same defect rate regardless of (for example, economic fluctuations).

2. Embodiments of the Present Disclosure

(1) Data cleaning

Figure 5 is a diagram illustrating a data purification process applying the credit rating model adjustment method according to the present disclosure. 5, by applying the credit rating model adjustment method according to the present disclosure,

ㆍ We collected nationwide loan execution data from January 2013 to June 2021 and aggregated it by NICE CBS Score.

ㆍ After the loan execution date, Shin Jeong-won's default was registered, the credit information company's default was registered, and the credit information company re-registered for more than 90 days, and those that were not registered as defective were labeled as good.

ㆍ When checking the monthly defect rate and composition ratio of customers with a NICE CBS score of 850 points or higher, the composition ratio remains between 50 and 60% (average 56.4%), but the defect rate tends to gradually decrease, which means that the score has the characteristics of TTC. was able to confirm.

(2) Measurement of structural risk and relative risk by score

Figure 6 is a diagram for explaining the process of measuring the structural risk amount and relative risk amount for each score by applying the credit rating model adjustment method according to the present disclosure. 6, by applying the credit rating model adjustment method according to the present disclosure,

ㆍ The monthly structural risk amount and relative risk amount of the score were calculated.

ㆍ The Ln Odds value was calculated for each score section (10 points) every month, and the corresponding score section and Ln Odds value were analyzed through linear regression to calculate the y-intercept (relative risk) and slope (structural risk) values.

ㆍ Structural risk (orange line) is slightly increasing, but the relative risk, which refers to whether the risk-vulnerable class has a relatively high credit risk compared to the risk-stable class, is decreasing, confirming that the credit inequality index is continuously improving.

(3) Forecasting structural and relative risk amounts through macroeconomic indicators

7A and 7B are diagrams showing structural risk prediction results and relative risk prediction results, respectively, by applying the credit rating model adjustment method according to the present disclosure. In the figure, the dotted line represents the actual measured value and the solid line represents the predicted value. When explained with reference to FIGS. 7A and 7B, by applying the credit rating model adjustment method according to the present disclosure,

ㆍ The existence of causal relationships with each risk among various macroeconomic indicators was identified through Granger causality analysis.

ㆍ The time difference between each indicator and risk was performed through cross-correlation analysis.

ㆍ Through linear regression analysis, a regression model between the optimal lag variable and target was constructed, and the correspondence between the direction of general macro variables and the sign of the coefficient was verified.

ㆍ The indicators actually included in the model are as follows, and the correlation coefficient was appropriate at the level of 0.72 for structural risk and 0.81 for relative risk.

① Economic growth rate (change rate compared to the same period last year, %)

② Unemployment rate (seasonally adjusted, %)

③ Inflation rate (change rate compared to the same period last year, %)

④ Domestic stock price (KOSPI increase/decrease rate compared to the same period last year, %)

⑤ Housing price increase/decrease rate in the metropolitan area (change rate compared to the same period last year, %)

⑥ Credit loan interest rate (new loan amount, %)

(4) Score dynamic adjustment

FIG. 8 is a diagram illustrating a dynamic score adjustment process applying the credit rating model adjustment method according to the present disclosure. 8, by applying the credit rating model adjustment method according to the present disclosure,

ㆍ The monthly score was dynamically adjusted using the relative/structural risk values predicted through the relevant macroeconomic indicators and changed into a PIT score.

ㆍ When looking at the defect rate of customers with a score of 850 points or higher before and after adjustment, the defect rate before adjustment was downward sloping, but the defect rate after adjustment did not change significantly at the 2.5% level.

ㆍ The results of the coefficient of variation (CV=standard deviation/average) values for each score range are as shown in Table 2 below, and it was confirmed that the CV value for each score range decreased in all sections.

[Table 2]

ㆍ In order to check the defect rate of approved customers, the CV value was observed for each cumulative approval defect rate when approval exceeded a certain score (see Table 3), and the CV value for each score range decreased in all sections, stabilizing the risk of customers who were actually approved. It was confirmed that it could be managed.

[Table 3]

ㆍ In order to check the defect rate of rejected customers, the CV value was observed for each cumulative rejected defect rate when rejection was below a certain score (see Table 4), and the CV value for each score range decreased in all sections, stabilizing the risk of actually approved customers. It was confirmed that it could be managed.

[Table 4]

ㆍ Through dynamic adjustment of macroeconomic indicators, some improvement in the discrimination power of the credit rating model was confirmed, with up to 2% improvement in indicators such as K-S statistics, AUC, and GINI Coefficient, which are the most commonly used discrimination indicators in credit evaluation. There was an improvement (see Table 5).

[Table 5]

3. Effect of the invention

The method of the present disclosure uses a technique to dynamically adjust the credit score to reflect the game situation, changing the score developed with the existing TTC into a form similar to the PIT, and through this, when the game changes, the approved score can be approved without changing the approval strategy. We suggest ways to maintain the customer's risk level. Therefore, when a financial institution using the existing CB credit rating uses the method of this disclosure, it does not need to change the cut-off level of the CB rating depending on the economy, but rather utilizes a score that dynamically changes depending on macro risks. , it is possible to quantitatively take the same amount of approval risk while maintaining the standard cut-off level.

The method and time series model of the present disclosure may be applied to the score as a whole, but may also be constructed separately in consideration of specific segments. For example, it can be developed only for mortgage loan customers, those who are new to society, or gig workers who do not have a fixed income. In the case of gig worker customers or self-employed people, the stability of income is not great, so the influence on macroeconomic indicators may be relatively large, and the score adjustment may be different from that in the full model. Through the method of the present disclosure, when adjusting the score, it is possible to quantitatively set a strategy for these vulnerable groups, thereby enabling more sophisticated credit loan operation.

4. Other embodiments

In the above description and drawings, the process of the credit rating model adjustment method that automatically reflects the economic situation according to the present disclosure is described as being executed sequentially, but this is merely an illustrative explanation of the technical idea of an embodiment of the present disclosure. . In other words, those skilled in the art will be able to apply various modifications and modifications by changing the order described above or executing one or more of the processes in parallel without departing from the essential characteristics of the embodiments of the present disclosure. Initiation is not limited to the chronological order described above.

The credit rating model adjustment method that automatically reflects the economic situation described herein includes digital electronic circuits, integrated circuits, FPGA (field programmable gate array), ASIC (application specific integrated circuit), computer hardware, firmware, software, and/ Or it can be realized as a combination of these. These various implementations may include being implemented as one or more computer programs executable on a programmable system. The programmable system includes at least one programmable processor (which may be a special purpose processor) coupled to receive data and instructions from and transmit data and instructions to a storage system, at least one input device, and at least one output device. or may be a general-purpose processor). Computer programs (also known as programs, software, software applications or code) contain instructions for a programmable processor and are stored on a "computer-readable storage medium."

Computer-readable recording media include all types of recording devices that store data that can be read by a computer system. Additionally, the computer-readable recording medium may be distributed in a computer system connected to a network, so that the computer-readable code may be stored and executed in a distributed manner.

Various implementations of the methods described herein may be implemented by a programmable computer. Here, the computer includes a programmable processor, a data storage system (including volatile memory, non-volatile memory, or another type of storage system, or a combination thereof), and at least one communication interface. For example, a programmable computer may be one of a server, network device, set-top box, embedded device, computer expansion module, personal computer, laptop, personal data assistant (PDA), cloud computing system, or mobile device.

The above description is merely an illustrative explanation of the technical idea of the present embodiment, and those skilled in the art will be able to make various modifications and variations without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are not intended to limit the technical idea of the present embodiment, but rather to explain it, and the scope of the technical idea of the present embodiment is not limited by these examples. The scope of protection of this embodiment should be interpreted in accordance with the claims below, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of rights of this embodiment.

Claims (10)

A credit rating model adjustment method that automatically reflects economic conditions by a computer,
A computer prepares a model to generate a monthly credit score based on TTC (Through-The-Cycle), in which the credit score does not change regardless of the up-turn or down-turn of the economy. first step;
A second step in which the computer calculates the structural risk and relative risk of the TTC-based monthly credit rating score;
A third step in which the computer generates a correlation between the structural risk amount and the relative risk amount calculated in the second step and a macroeconomic indicator; and
A fourth step in which the computer converts the TTC-based monthly credit rating score into a credit rating score calculated using the correlation equation generated in the third step,
The model for generating the TTC-based monthly credit rating score, in which the credit rating score does not change regardless of the upturn or downturn of the game prepared in the first step, was developed based on a logistic regression model,
The structural risk calculated in the second step is defined as the y-intercept value of a linear relationship having the TTC-based monthly credit rating score as the x-axis value and the logit value as the y-axis value, and the relative risk amount is It is defined as the slope of a linear relationship that has the TTC-based monthly credit score as the x-axis value and the logit value as the y-axis value, and the logit value is the logarithm of odds,
The method wherein the correlation equation generated in the third step is generated by a model that has the structural risk amount and the relative risk amount calculated in the second step as dependent variables and the macroeconomic indicator as an independent variable. delete delete delete According to paragraph 1,
The model that generates the correlation equation refines the macroeconomic indicators, secures normality, confirms causality between the independent variable and the dependent variable, and determines whether the independent variable affects the dependent variable. A method developed by selecting a significant time point and generating a regression model using the dependent variable and the independent variable at the refined optimal lag. According to clause 5,
The method of claim 1 , wherein refining the macroeconomic indicator includes checking seasonality included in the macroeconomic indicator and removing the seasonality if it exists. According to clause 5,
A method in which the causality of the independent variable and the dependent variable is confirmed by causality analysis through Granger causality. According to clause 5,
The method of selecting a significant time point at which the independent variable affects the dependent variable is performed by a significant time point selection method through cross-correlation analysis. According to clause 5,
The macroeconomic indicators include the following items: economic growth rate, unemployment rate, inflation rate, domestic stock price, housing price increase/decrease rate in the metropolitan area, and credit loan interest rate. A computer-readable device that records a computer program for executing each step of the method according to any one of paragraphs 1, 5 to 9 on a computer to adjust the credit rating model to automatically reflect the economic situation. Recording media.