Credit risk measurement is the basis and key link of modern credit risk management. The measurement of credit risk has gone through three main stages: expert judgment, credit scoring model and default probability model analysis. In particular, the New Basel Capital Accord encourages qualified commercial banks to use the method based on internal rating system to measure default probability and default loss and calculate the capital requirements corresponding to credit risk, which has effectively promoted the in-depth development of internal rating system and credit risk measurement technology of commercial banks.
The measurement of credit risk by commercial banks depends on the evaluation of borrowers and transaction risks. The new Basel Capital Accord clearly requires that the internal rating of commercial banks should be based on a two-dimensional rating system: one is customer rating and the other is debt rating.
3.2. 1 Customer Credit Rating
1. Basic concept of customer credit rating
Customer credit rating is a measure and evaluation of the solvency and willingness of customers by commercial banks, which reflects the risk of default of customers. The evaluation subject of customer rating is commercial banks, the rating target is customer default risk, and the evaluation result is credit rating and default probability (PD).
Regarding the customer's credit rating, the following statement is wrong ().
A. The appraisal subject is a commercial bank.
B. the evaluation target is the default risk of customers.
C. the evaluation results are credit rating and default probability.
D. the evaluation content is the loss of specific debts after the customer defaults.
Answer: d
(1) Definition of breach of contract
According to the definition of the New Basel Capital Accord, when one or more of the following events occur, the debtor is deemed to be in breach of contract:
(1) The commercial bank determines that unless recourse measures are taken, such as realizing the collateral (if any), the borrower may not be able to repay the debt to the commercial bank in full.
② The debtor's substantial credit debt to the commercial bank is overdue for more than 90 days (inclusive). If the debtor exceeds the stipulated overdraft limit or the newly approved limit is less than the current balance, all overdrafts will be considered overdue.
(3) The following situations will be regarded as possible failure to repay debts in full:
The bank stops paying the loan interest;
After the credit relationship, due to the sharp decline in credit quality, banks write off loans or withdraw special reserves;
Banks sell loans and bear relatively large economic losses;
Banks agree to negative debt restructuring, which may lead to a large-scale reduction or delay in repayment of principal, interest or expenses, resulting in a reduction in debt scale;
As far as the borrower's debt to the bank is concerned, the bank lists the debtor as a bankrupt enterprise or something similar;
The debtor applies for bankruptcy, or has gone bankrupt, or is in a similar state, so it will not perform or delay the repayment of bank debts.
(2) Default probability
The probability of default refers to the possibility that the borrower will default in a certain period in the future. In the new Basel Capital Accord, the default probability is defined as the higher of the borrower's internal rating 1 year and 0.03%. The Basel Committee set the lower limit of 0.03% in order to reset the lower limit of risk right, and also considered the difficulties faced by commercial banks in testing small probability events.
In the new Basel Capital Accord, the default probability is specifically defined as the higher of the borrower's internal rating 1 year default probability and ().
A.0. 1%
B.0.0 1%
About 0.3%
D.0.03%
Answer: d
The estimation of default probability includes two levels: one is the default probability of a single borrower; The second is the default probability of all borrowers under a certain credit rating. The new Basel Capital Accord requires commercial banks that implement the internal rating method to estimate the default probability of borrowers with various credit ratings. Commonly used methods include historical default experience, statistical model and external rating mapping.
A concept that is easily confused with default probability is default frequency, which is usually called default rate. The frequency of default is the result of backtesting, while the probability of default is the prior prediction made by the analysis model, which is essentially different.
Another concept that is easily confused with the probability of default is non-performing rate, which makes the ratio of non-performing debt balance to total debt balance not comparable.
2. The development of customer credit rating
(1) expert judgment method
Expert system is a traditional credit analysis method developed and perfected by commercial banks in the process of long-term credit business and taking credit risks.
① Factors related to the borrower:
Reputation (reputation)
Lever (lever)
Income volatility (income volatility)
② Market related factors
Economic cycle (economic cycle)
macroeconomic policy
Interest rate level (interest rate leverage)
At present, among the expert systems used, the 5Cs system for enterprise credit analysis is the most widely used. 5Cs system refers to:
Character (character)
Capital (capital)
debt paying ability
Collateral (mortgage)
Operating environment (conditions)
In addition to 5Cs system, widely used expert systems include 5Ps system for enterprise credit analysis and CAMEL analysis system for commercial banks and other financial institutions.
The 5Ps include: people factor, purpose factor, repayment source factor, protection factor and enterprise prospect factor.
In the customer's credit rating, the single choice consists of personal factors, fund use factors, repayment source factors, guarantee factors and enterprise prospect factors. Enterprise credit analysis expert system is ().
A.5Cs system
B.5Ps system
C.CAMEL analysis system
D.4Cs system
Answer: b
CAMEL analysis system includes capital adequacy ratio, asset quality, management level, profitability and liquidity.
The outstanding feature of expert system is that the experience and judgment of credit experts are the main basis for credit analysis and decision-making. A prominent problem brought by this subjective method/system is the lack of consistency in the evaluation of credit risk. In addition, although the expert system has formed a mature analytical framework in the long-term development and practice of the banking industry, it lacks systematic theoretical support, especially in the selection of key elements, the determination of weights and comprehensive evaluation. Therefore, expert system is more suitable for two-dimensional decision-making of borrowers, and it is difficult to accurately measure credit risk.
(2) Credit scoring method
Credit scoring model is a traditional quantitative model of credit risk, which uses observable borrower characteristic variables to calculate a numerical value (score) to represent the debtor's credit risk, and divides borrowers into different risk levels.
Background knowledge: credit scoring model
In 1960s, the introduction of credit card promoted the great development of credit scoring technology, and quickly extended to other business fields. Altman (1968) proposed a Z-score model based on multivariate discriminant analysis. Martin (1977), Olson (1980) and Wigginton (1980) used Logit model to analyze enterprise bankruptcy for the first time.
The key of credit scoring model lies in the selection of characteristic variables and the determination of their respective weights. The basic process is:
Firstly, according to experience or correlation analysis, determine which economic or financial factors are mainly related to the credit risk of a certain type of borrower, and simulate a specific form of functional relationship;
Secondly, according to the historical data, regression analysis is carried out to get the weight of each related factor;
(3) Finally, the numerical value of the relevant factors of the potential borrowers belonging to this category is substituted into the functional relationship, and a numerical value is calculated. According to this numerical value, the credit risk level of the potential borrowers is measured, and accordingly, the borrowers are rated to decide whether to lend.
There are some outstanding problems:
① The credit scoring model is based on the simulation of historical data (rather than current market data), so it is a backward-looking model.
(2) The credit scoring model requires quite high historical data of borrowers.
③ Although the credit scoring model can give the score of customer's credit risk level, it can't provide the accurate value of customer's default probability, which is often the most concerned by credit risk management.
(3) Default probability model
Default probability model analysis belongs to modern credit risk measurement method. Among them, the representative models are Moody's RiskCalc and Credit Monitor, KPMG's risk-neutral pricing model and mortality model, which have aroused great repercussions in the banking industry.
The new Basel Capital Accord also clearly stipulates that commercial banks that implement the internal rating method can use models to estimate the default probability.
Compared with the traditional expert judgment and credit scoring methods, the default probability model can directly estimate the default probability of customers, so it requires higher historical data, requiring commercial banks to establish a consistent and clear definition of default and accumulate data for at least five years.
3. Enterprise customer rating model
(1) Altman's Z-score model and ZETA model
Altman (1968) thinks that there are five main factors that affect the borrower's default probability: liquidity, profitability, leverage, solvency and activity. Altman chose the five financial indicators listed below to comprehensively reflect the above five factors, and the final Z score function is:
X 1= (current assets-current liabilities)/total assets
X2= retained earnings/total assets
X3= earnings before interest and tax/total assets
X4= market value of stock/book value of debt
X5= sales/total assets
As an indicator of default risk, the higher the z value, the lower the default probability. In addition, Altman also put forward the critical value for judging the bankruptcy of an enterprise: if z is lower than 1.8 1, the enterprise has great bankruptcy risk and should be classified as a high default risk level.
In 1977, Altman, Hardeman and Narayanan put forward the second generation Z-scoring model-Zeta credit risk analysis model, which is mainly used in public or private non-financial companies, with wider application scope and more accurate calculation of default probability.
ZETA model increases the number of model indexes from five to seven, namely:
X 1: return on assets, equal to earnings before interest and tax/total assets.
X2: income stability index, which refers to the trend standard deviation of enterprise's return on assets within 5 ~ 10 years.
X3: solvency index, equal to earnings before interest and tax/total interest expense.
X4: profitability index, equal to retained earnings/total assets.
X5: Liquidity index, namely current ratio, equals current assets/current liabilities.
X6: Capitalization index, equal to common stock/total capital. The greater the ratio, the stronger the capital strength of the enterprise and the smaller the probability of default.
X7: Scale indicator, expressed by the logarithm of the total assets of the enterprise.
(2) Risk calculation model
RiskCalc model is a default probability model developed on the basis of traditional credit scoring technology, which is suitable for non-listed companies. Its core is to select a group of variables that can predict the default best from customer information through strict steps, and then use Logit/Probit regression technology to predict the default probability of customers through appropriate transformation.
① Collect a lot of company data;
(2) sample selection and outlier processing are carried out on the data;
③ Analyze and transform the monotonicity, default prediction ability and correlation of each risk factor one by one, and initially screen out 20 ~ 30 risk factors with strong default prediction ability and low correlation;
④ Using Logit/Probit regression technology, 9 ~ 1 1 risk factors are selected from the preliminary factors, and the regression coefficient is guaranteed to have clear economic significance, and there is no multicollinearity among variables;
⑤ Verify the ability to distinguish model default based on unmodeled samples and modeled samples in unmodeled samples, and ensure the horizontal applicability and vertical predictability of the model;
⑥ Correct the output of the model and get the default probability of each customer.
(3) credit monitoring model
Credit monitoring model is a default probability model developed on the basis of Merton model, which is suitable for listed companies. Its core lies in treating the loan relationship between enterprises and banks as an option transaction relationship, so the credit risk information in the loan relationship is implied in this option transaction, so we can apply the option pricing theory to solve the credit risk premium and the corresponding default rate, that is, the expected default frequency (EDF).
In the corporate customer rating model, (1) applies the option pricing theory to solve the credit risk premium and the corresponding default rate.
A. Altman z scoring model
B. Risk calculation model
C. Credit monitoring model
D. Mortality model
Answer: c
(4) KPMG risk neutral pricing model.
The core idea of risk-neutral pricing theory is to assume that every participant in the financial market is risk-neutral. Whether it is high-risk assets, low-risk assets or risk-free assets, as long as the expected returns of assets are equal, the attitudes of market participants are the same. This market environment is called the risk-neutral paradigm. KPMG applies risk-neutral pricing theory to the calculation of default probability of loans or bonds. Since the bond market can provide risk premiums corresponding to different credit ratings, the default probability of each loan or bond can be calculated according to the risk-neutral pricing principle of equal expected returns.
The annual yield of zero-coupon bonds in a certain year is 16.7%. Assuming that the debtor defaults, the recovery rate is zero. If the risk-free annual rate of return is 5% within one year, according to KPMG's risk-neutral pricing model, the default probability of the above bonds within one year is ().
A: 0.05
B.0. 10
C.0. 15
D.0.20
Answer: b
(5) mortality model
Mortality model is to calculate the default probability of loans or bonds with different credit ratings in a certain holding period in the future, that is, mortality, which is usually divided into marginal mortality (MMR) and cumulative mortality (CMR).
According to the mortality model, assuming that the marginal mortality of a three-year syndicated loan from 1 to the third year is 0. 17%, 0.60% and 0.60% respectively, the cumulative mortality for three years is ().
A.0. 17%
B.0.77%
C. 1.36%
2.32%
Answer: c
4. Individual customer scoring method
According to international practice, the credit evaluation of enterprises adopts rating method, and the credit evaluation of individual customers adopts scoring method. Due to the large number of individual customers and strong regularity of historical information, the scoring model based on historical data statistics is mainly used to measure the credit risk of individual customers.
Referring to international practice, individual customer rating can be divided into regression analysis, K- nearest value method, neural network model method and so on. According to the scoring object, it can be divided into customer layer, product layer and account layer, and according to the scoring purpose, it can be divided into risk score, profit score and loyalty score. According to the equal stages, it can be divided into customer development period (credit bureau score), customer approval period (application score) and customer management period (behavior score).
(1) Credit Bureau score
The commonly used models at this stage are:
① Risk score for predicting consumer default/bad debt risk;
(2) Revenue score, which predicts the potential revenue that consumers will bring to commercial banks after opening an account;
③ Bankruptcy score to predict the bankruptcy risk of consumers;
(4) scores of other credit characteristics.
(2) Application score
The application scoring model compares the credit performance of similar applicants in commercial banks by comprehensively considering all kinds of information filled in by the applicants on the application form, predicts the default probability of the applicants within a certain period of time after opening an account, and makes a decision of rejection or acceptance by comparing the default probability of customers with the acceptable default bottom line of commercial banks.
The risk scoring model and income scoring model of credit bureau are valuable decision-making tools, which are supplements to the application scoring model and can form a two-dimensional or three-dimensional matrix for credit approval decision. The difference is that the application scoring model is tailored by commercial banks for applicants of specific financial products, which can reflect the particularity of commercial bank customers more accurately and comprehensively, and can use more information to predict the future credit performance of customers; The scoring model of credit reporting agencies usually predicts the default probability of applicants in various credit relationships in the future.
(3) Behavior score
Behavior score is used to observe the behavior of existing customers to grasp the credibility of customers' timely repayment.
5. Verification of customer rating/scoring.
(1) Verification of Customer Default Risk Differentiation Ability
The basic principle of the period is to use a variety of mathematical analysis methods to test the accuracy of the rating system in judging whether customers default.
(2) Verification (correction) of prediction accuracy of default probability
Its basic principle is to use hypothesis testing in statistics. When the actual default situation exceeds the given threshold, the original assumption is rejected and the PD prediction is inaccurate. Commonly used methods are: binomial distribution test to test the accuracy of PD prediction in a given year; Chi-square distribution test, which tests the accuracy of PD prediction in different grades in a given year; Normal distribution test, which tests the accuracy of PD prediction of the same grade in different years; Expand the traffic light test to test the accuracy of PD prediction in different years and grades.