CN117853225A - Credit evaluation method for debt subject - Google Patents

Abstract

The invention discloses a credit evaluation method for a debt subject, which comprises the following steps: s1, selecting a sample; s2, index analysis and screening; s3, identifying a scoring interval; s4, calculating weights; s5, verifying a result; s6, dividing the level; the technical problems that research normalization is poor, evaluation result acceptance is low, risk analysis accuracy is low and external evaluation does not meet self-organization preference in the prior art are solved; the method makes up the basic evaluation difference of different researchers on the main body, and all parties are involved in the formulation and review processes, so that the compliance of the model result is ensured, the basic judgment accuracy of the researchers on the main body can be solved through a back-testing scheme, meanwhile, the manual identification of the researchers is allowed to have a certain influence on the model, and meanwhile, the risk preference of the model applicable mechanism is fully considered.

Description

Credit evaluation method for debt subject

Technical Field

The invention belongs to the technical field of credit evaluation of debt subjects, and particularly relates to a credit evaluation method of a debt subject.

Background

The investor's judgment on the marketing company is not limited to the traditional financial indexes such as income, profit capability and debt paying capability, but is also an important consideration factor whether to have the external effects such as negative environment and society, etc., and the credit evaluation method of the debt body is a technical means for evaluating the debt paying capability and willingness of the debt body, and can help the investor, the supervision organization and the market participants to know the credit risk condition of the bond market and promote the healthy development of the bond market.

The conventional solid collection and research internal evaluation system for the financial institutions has the following defects:

1. study normalization was poor: part of institutions do not develop an internal evaluation model of a ripening system, each evaluation mainly depends on external rating or collective decision, and different personnel cannot obtain a uniform evaluation scheme due to analysis experience differences of own industries.

2. The acceptance of the evaluation result is low: investment, risk and credit research departments play different roles in the investment process, the angle and the method for risk management are different, and satisfactory results of all parties are difficult to form.

3. The risk analysis accuracy is low: different researcher evaluation systems are different, and the risk evaluation mode of the current domestic debt and market cannot be met based on the subjective evaluation or objective evaluation.

4. The critique does not satisfy its own institutional preferences: at present, whether the market is a third-party rating institution or the investment market is evaluated rapidly in recent development, model schemes and results are provided in the rating institution, and the risk preference of different institutions cannot be met.

Disclosure of Invention

The present invention aims to solve at least one of the technical problems existing in the prior art; therefore, the invention provides a credit evaluation method for a debt main body, which is used for solving the technical problems of poor research standardization, low evaluation result acceptance, low risk analysis accuracy and insufficient external evaluation for self-organization preference.

To achieve the above object, a first aspect of the present invention provides a credit evaluation method for a liability agent, comprising the steps of:

s1, sample selection: acquiring industry data and annual report financial data of debt and marketing enterprises within a plurality of years, marking the industry data and the annual report financial data as sample data, and grouping the sample data according to industry types to obtain k groups of basic models; extracting an external evaluation result of the debt subject;

wherein the industry data comprises a plurality of major indexes, and the major indexes comprise a plurality of sub indexes; the major category of indicators includes capital structure indicators, operational capability indicators, profitability indicators, payability indicators, growth capability indicators, and regional indicators, the annual report financial data includes a profit-and-loss table, a liability table, and a cash flow table, and k is a positive integer;

s2, index analysis and screening: screening a plurality of large indexes and a plurality of corresponding sub indexes to obtain a modeling index;

s3, identifying a scoring interval: selecting a plurality of preset scoring points corresponding to the sample data as scoring standards based on the modeling indexes and the basic model to obtain scoring intervals;

s4, calculating weights: carrying out weight identification on a plurality of large-class indexes in the industry data based on an analytic hierarchy process to obtain index weights, and identifying a plurality of sub-index weights based on the index weights and the number of sub-indexes in the large-class indexes to obtain final weights;

s5, verifying results: generating h evaluation models by using the scoring interval and the final weight corresponding to the modeling index, and checking the evaluation models based on consistency and AUC values to obtain a checking result; calibrating the evaluation model according to the test result; wherein h is a positive integer;

s6, level division: and inputting the industry data and annual report financial data of a plurality of subjects to be tested into an evaluation model to generate an evaluation result, classifying the evaluation result, and distinguishing the level subjects of different credits by using the rating symbol.

Preferably, the screening the plurality of major indexes and the corresponding plurality of sub indexes to obtain the modeling indexes includes:

a1, screening deletion rate: extracting sample data, and removing large indexes with the loss rate of annual report financial data exceeding a% in industry data to obtain screening indexes, wherein a is a positive integer, and a is less than 100;

a2, screening consistency of external evaluation: judging whether the numerical value of the sub-index in the screening index is consistent with the external evaluation result or not according to the external evaluation Somer' sD of the corresponding sample, and obtaining a judgment result;

a3, explanatory screening: judging whether the change area of the neutron index along with the external evaluation result is consistent with the corresponding economic meaning or not, and obtaining a judgment result;

a4, screening relevance: judging whether the relevance of any two sub-indexes in the same large-class index is greater than a preset relevance threshold value or not to obtain a judging result; the covariance method is adopted in the calculation mode of the relevance of any two sub-indexes.

And A5, recognizing the judging results of the external evaluation consistency screening, the interpretation screening and the relevance screening as reference indexes to obtain target indexes.

Preferably, the selecting a plurality of preset scoring points corresponding to the sample data based on the modeling index and the basic model as the scoring standard to obtain the scoring interval includes:

b31: extracting the numerical values of corresponding sub-indexes of all enterprises in a z-th basic model in a plurality of years to obtain a plurality of comparison data; wherein the corresponding sub-index is a sub-index of interest in the industry; wherein z is a positive integer, and z is less than or equal to k;

b32: sequencing a plurality of comparison data from large to small to obtain a sequencing result, and calculating a point value corresponding to a preset dividing point based on the number of submitted data;

b33: and marking the numerical value corresponding to the point value as a demarcation value according to the sorting result, and setting a scoring interval based on the demarcation value and the corresponding sub-index.

Preferably, the identifying the plurality of sub-indicator weights based on the indicator weights and the number of sub-indicators in the major indicator includes:

c41: determining the reserved number u of the sub-indexes according to the large-class index weight and the number of the sub-indexes, wherein u is more than or equal to 1 and less than or equal to 5;

c42: determining the weight value of each sub-index according to the consistency of the external evaluation in the mathematical screening process;

c43: and marking the product of the weight index of the sub-index corresponding to the large-class index and the weight value of the corresponding sub-index as the final weight of the sub-index.

Preferably, the checking the evaluation model based on the consistency and the AUC value includes:

calculating a model result of the evaluation model, comparing the model result with the outer evaluation symbols 'D and ROC curve AUC values corresponding to the outer evaluation grades, and comparing the model result, the outer evaluation symbols' D and the respective outer evaluation grades with corresponding AUC values and preset thresholds; if the model result, the outer rating detectors' D, the corresponding AUC value of each outer rating and the preset threshold are all larger than the preset threshold, the evaluation model passes statistical verification; otherwise, the evaluation model is adjusted.

Preferably, the classifying the evaluation result includes:

summarizing the evaluation models with the required classification level based on a preset model sample classification basis, listing the first m classes as high-quality evaluation models, and classifying the evaluation models based on a preset normal distribution theoretical value to obtain a classification result; sorting the evaluation models from high to low according to the division result, grouping the sorted evaluation models according to the corresponding theoretical occupation ratio, extracting the upper and lower scores of the evaluation models in different grades, and marking the upper and lower scores as a grade main body; wherein m is a positive integer, and m is less than or equal to h.

Compared with the prior art, the invention has the beneficial effects that:

according to the invention, a large amount of sample data and external evaluation results of a debt main body are obtained, a large amount of sample data are screened, each sub-index in the sample data is weighted and identified, an evaluation model can be obtained, the evaluation model is inspected, and the main scale of the evaluation results is divided, so that the basic evaluation differences of different researchers on the main body are made up, the participation of each party is brought in the process of making and evaluating, the fairness of the model results is ensured, the basic judgment accuracy of the researchers on the main body is solved through a back-testing scheme, meanwhile, the manual identification of the researchers is allowed to have a certain influence on the model, and meanwhile, the risk preference of the model applicable mechanism is fully considered.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow chart of the present invention.

Detailed Description

The technical solutions of the present invention will be clearly and completely described in connection with the embodiments, and it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1, an embodiment of the first aspect of the present invention provides a credit evaluation method for a liability agent, including:

s1: acquiring industry data and annual report financial data of debt and marketing enterprises within a plurality of years, marking the industry data and the annual report financial data as sample data, and grouping the sample data according to industry types to obtain k groups of basic models; extracting an external evaluation result of the debt subject;

wherein the industry data comprises a plurality of major indexes, and the major indexes comprise a plurality of sub indexes; the major category of indicators includes capital structure indicators, operational capability indicators, profitability indicators, payability indicators, growth capability indicators, and regional indicators, the annual report financial data includes a profit-and-loss table, a liability table, and a cash flow table, and k is a positive integer;

s2: screening a plurality of large indexes and a plurality of corresponding sub indexes to obtain a modeling index;

s3: selecting a plurality of preset scoring points corresponding to the sample data as scoring standards based on the modeling indexes and the basic model to obtain scoring intervals;

s4: carrying out weight identification on a plurality of large-class indexes in the industry data based on an analytic hierarchy process to obtain index weights, and identifying a plurality of sub-index weights based on the index weights and the number of sub-indexes in the large-class indexes to obtain final weights;

s5: generating h evaluation models by using the scoring interval and the final weight corresponding to the modeling index, and checking the evaluation models based on consistency and AUC values to obtain a checking result; wherein h is a positive integer;

s6: and inputting the industry data and annual report financial data of a plurality of subjects to be tested into an evaluation model to generate an evaluation result, classifying the evaluation result, and distinguishing the level subjects of different credits by using the rating symbol.

It should be noted that: industry information of debt and marketing enterprises is obtained. Because the financial characteristics of enterprises in different industries are different, the main bodies are required to be grouped according to the actual financial characteristics of the industries, and the models are sequentially built. For example, real estate enterprises and banking enterprises have large differences in income, liability performance and profitability, so that real estate and banking enterprises establish two sets of models for analysis respectively. Enterprises such as agriculture, forestry, animal husbandry, fishery and food manufacturing industries are consistent in characteristics, and the main bodies of the industries are uniformly classified into groups of 'agriculture, animal husbandry, fishery and food'.

In this embodiment, the screening of the plurality of major indexes and the corresponding plurality of sub indexes in S2 to obtain the modeling index includes:

a1, screening deletion rate: extracting sample data, and removing large indexes with the loss rate of annual report financial data exceeding a% in industry data to obtain screening indexes, wherein a is a positive integer, and a is less than 100;

a2, screening consistency of external evaluation: judging whether the numerical value of the sub-index in the screening index is consistent with the external evaluation result or not according to the external evaluation Somer' sD of the corresponding sample, and obtaining a judgment result;

a3, explanatory screening: judging whether the change area of the neutron index along with the external evaluation result is consistent with the corresponding economic meaning or not, and obtaining a judgment result;

a4, screening relevance: judging whether the relevance of any two sub-indexes in the same large-class index is greater than a preset relevance threshold value or not to obtain a judging result; the covariance method is adopted in the calculation mode of the relevance of any two sub-indexes.

And A5, recognizing the judging results of the external evaluation consistency screening, the interpretation screening and the relevance screening as reference indexes to obtain target indexes.

Furthermore, the determination results of the external evaluation consistency screening, the interpretation screening and the relevance screening can be manually determined by taking the determination results as reference indexes, and 3-5 experts corresponding to the industry are required to be invited during manual determination, so that the evaluation model indexes commonly used in daily analysis work of the industry are manually determined. Such as: the building industry enterprises need to pay high attention to the liability rate of the assets, the medicine industry needs to pay high attention to the interest guarantee multiple of the EBITDA, and the expert carries out manual identification according to the analysis working experience in the industry. The screening process is as follows:

a51: for each index major class in the industry, 5 indexes at most and 1 index at least are selected as alternatives.

A52: the index is scored 1 score for 1 time each time it is selected by the expert.

A53: summarizing all expert selections to obtain sub-indexes of top 5 of expert ranks in each large-class index. For example: take the capital construction as an example. The following sample indices a to f are sub-indices of any large class of indices.

Index a

Index b

Index c

Index d

Index e

Index f

Expert A

√

√

√

√

Expert B

√

√

√

√

√

Expert C

√

√

√

√

The sub-index scores and the ranks are in turn

	Score of	Ordering of	Alternative to
				Index a	3	1	√
Index d	3	1	√
				Index b	2	3	√
Index c	2	3	√
				Index e	2	3	√
Index f	1	6

Sub-indices are classified as follows:

index classification	Manually-identified ranking	Consistency and interpretation of external evaluation
			Class 1	1～3	Both satisfy
Class 2	1～3	At least one does not satisfy
			Class 3	4～5	Both satisfy
Class 4	4～5	At least one does not satisfy

If a plurality of indexes exist under the same sub-index classification, indexes with lower relevance in relevance screening are preferentially used in the subsequent modeling.

In this embodiment S3, a plurality of preset scoring points corresponding to sample data are selected as scoring criteria based on the modeling index and the base model, so as to obtain a scoring interval, including:

b31: extracting the numerical values of corresponding sub-indexes of all enterprises in a z-th basic model in a plurality of years to obtain a plurality of comparison data; wherein the corresponding sub-index is a sub-index of interest in the industry; wherein z is a positive integer, and z is less than or equal to k;

b32: sequencing a plurality of comparison data from large to small to obtain a sequencing result, and calculating a point value corresponding to a preset dividing point based on the number of submitted data;

b33: and marking the numerical value corresponding to the point value as a demarcation value according to the sorting result, and setting a scoring interval based on the demarcation value and the corresponding sub-index.

For example, a value of "liability rate" in the construction industry capital structure index is selected, and preset quantiles are 20%, 40%, 60%, 80%, then the thresholds are specified as follows:

and taking out an index sample of the 'liability rate' of all subjects under the building industry classification for nearly 5 years. There is no need to pay attention to which enterprises these liability rates belong to, and only the numerical values are analyzed. Assume a total of 10 samples:

sample of	A	B	C	D	E	F	G	H	I	J
											Liability rate @％)	55	75	45	85	85	45	55	65	55	95

And a second step of: sorting according to the values from small to large;

ordering of	1	2	3	4	5	6	7	8	9	10
											Liability rate (%)	45	45	55	55	55	65	75	85	85	95

Calculating a point value of the quantile, for example: 20% quantiles, 10×20% = 2, and the rest are the same (if the calculation is not an integer, the average of the two ordered samples is taken, assuming the number of samples is 11, 20% ×11=2.2, i.e. the average of samples 2 and 3). Sample values of the sequences 2, 4, 6 and 8 are obtained and used as demarcation values.

Ordering of	2	4	6	8
					Liability rate (%)	45	55	65	85

Setting a scoring interval. Since the asset liability rate is a negative index, the scoring rule is as follows:

gear position	Lower limit (without)	Upper limit (with)	Index score
				First gear	-999,999	45	100
Second gear	45	55	80
				Third gear	55	65	60
Fourth gear	65	85	40
				Fifth gear	85	999,999	20

That is, if the liability rate of one construction industry subject is 75, the index score is 40 points since its value falls in the "fourth gear".

In this embodiment S4, the weighting determination is performed on a plurality of major indexes in the industry data based on the analytic hierarchy process, including:

defining an importance score: in general, the number of indexes to be compared is scored by adopting a symmetrical form value. 1 represents "importance up to all", 3, 5, 7 are used in turn to represent a gradual increase in importance, 1/3, 1/5, 1/7 represent a gradual decrease in importance;

filling in an importance evidence: an n×n rectangle is constructed, and the numbers in the ith row and the jth column represent the importance of the sub-index i to j. Scoring features are: 1) The diagonal lines each represent the importance of the ith index to the ith (self) index, and are therefore fixed at 1. 2) The value of the (i, j) th lattice is automatically the inverse of the value of the (j, i) th lattice.

For example, 5 values of 3, 2, 1/2, 1/3 are selected as the importance scores, which in turn represent very important, significantly important, equally important, significantly unimportant, very unimportant. That is, 1 is necessary for satisfying, and the "important number" and "unimportant number" are reciprocal.

Assume that in this embodiment, 3 indices are scored. A matrix of 3*3 is constructed and the numbers in the ith row and jth column represent the importance of index i to j. As can be seen by this definition, the diagonal of the matrix is all 1, and the j-th row and j-th column elements must be the reciprocal of the j-th row and i-th column elements. The hypothesis scores were as follows:

in this embodiment, a matrix of 3*3 is constructed, and the numbers in the ith row and jth column represent the importance of index i to j. As can be seen by this definition, the diagonal of the matrix is all 1, and the j-th row and j-th column elements must be the reciprocal of the j-th row and i-th column elements. The hypothesis scores were as follows:

	external risk	Basic face of enterprise	Financial risk
				External risk	1	2	2
Basic face of enterprise	1/2	1	1
				Financial risk	1/2	1	1

The scoring process ends. The calculation of weights is then started:

first, the geometric mean of the scores for each row is calculated.

	External risk	Basic face of enterprise	Financial risk	Geometric mean value
					External risk	1	2	2	1.587
Basic face of enterprise	1/2	1	1	0.794
					Financial risk	1/2	1	1	0.794

Next, the geometric mean of each index (per row) is normalized:

thus, the sum is 1.59+0.79+0.79=3.17, and the new geometric mean is 0.5, 0.25, respectively, which is the weight of the three indices.

The weights between each index are derived above. However, it should be noted that if the index is too many, the artificial scoring may be "inconsistent". For example, in this example, the external risk is 2 for both the business base and the financial risk is 2 for importance, which indicates that the relative importance of both the business base and the financial risk should be 1. However, if the index is too high, the association may be ignored, and "inconsistency" occurs in scoring. Therefore, the consistency test is performed after each AHP scoring (theoretically, this step should be performed after scoring, but it is more convenient to check mathematically after calculating the weights, and the consistency of the scoring is guaranteed in general, so it is also possible to do at last), and the method is as follows:

firstly, setting a scoring matrix as A, and calculating the characteristic value of A through w':

A*w’＝λ*w’

next, a consistency index CI is defined by λ:

wherein n is the number of scoring indexes (dimension of A) as the measurement CI, introducing a random consistency index RI, constructing 500 scoring matrixes of n order randomly, and calculating the average number of the 500 matrixes CI. Because each use is too cumbersome to construct, there are standard values for comparison.

Dimension n	1	2	3	4	5	6	7	8	9
										RI value	0	0	0.58	0.9	1.12	1.24	1.32	1.41	1.5

Finally, a consistency ratio CR is defined:

when CR <0.1, it is defined as passing the consistency check. If the consistency check is not passed, the scoring process needs to be rechecked and the A matrix corrected.

So far, the calculation process of the index scoring weight is completed.

Further, the identifying of the plurality of sub-indicator weights based on the indicator weights and the number of sub-indicators in the large category of indicators in S4 includes:

c41: determining the reserved number u of the sub-indexes according to the large-class index weight and the number of the sub-indexes, wherein u is more than or equal to 1 and less than or equal to 5;

c42: determining the weight value of each sub-index according to the consistency of the external evaluation in the mathematical screening process;

c43: and marking the product of the weight index of the sub-index corresponding to the large-class index and the weight value of the corresponding sub-index as the final weight of the sub-index.

In this embodiment S5, the evaluation model is checked based on the consistency and AUC values, including:

calculating a model result of the evaluation model, comparing the model result with the outer evaluation symbols 'D and ROC curve AUC values corresponding to the outer evaluation grades, and comparing the model result, the outer evaluation symbols' D and the respective outer evaluation grades with corresponding AUC values and preset thresholds; if the model result, the outer rating detectors' D, the corresponding AUC value of each outer rating and the preset threshold are all larger than the preset threshold, the evaluation model passes statistical verification; otherwise, the evaluation model is adjusted.

It should be noted that: somers 'D is an abbreviation for Somers' Delta. It is a measure of ordinal correlation between two possibly related random variables X and Y. When Somers' D equals-1, it means that all variable pairs are inconsistent. The case where the parameters' D is equal to 1 indicates that all variable pairs are identical. The calculation method comprises the following steps: assume that there are M samples, each of which has two parameters of X and Y values to be compared. The i and j elements (i. Noteq. J) in any sample are defined:

a consistent pair: xi > Xj and Yi > Yj; or Xi < Xj and Yi < Yj

Inconsistent pairs: xi < Xj and Yi > Yj; or Xi > Xj and Yi < Yj

Then the polymers' d= (Nc-Nd)/N

Wherein: nc: the total number of coincident pairs; nd: the total number of inconsistent pairs; n: the log is compared in total. If the total number of samples is M, n=m (M-1).

And AUC represents the area under the ROC curve and is mainly used for measuring the quality of the classification effect of the model. AUC is an evaluation index for measuring the quality of the two classification models, and represents the probability that the positive case is arranged in front of the negative case. In general, in a classification model, the prediction results are expressed in the form of probabilities, and if the accuracy is to be calculated, a threshold is manually set to convert the corresponding probabilities into the classes, and the threshold also affects the calculation of the accuracy of the model to a great extent.

The specific calculation method comprises the following steps: classification targets are divided into two categories: positive and negative examples. The following statistics are defined:

true Positves (TP): predicting the number of instances that are positive examples and are actually positive examples;

false Positives (FP): the number of instances predicted to be positive but actually negative;

false Negotives (FN): the number of instances predicted as negative but actually positive;

true Negotives (TN): the number of instances predicted to be negative and actually negative.

Based on the four statistics above, the following index is defined:

FPR: the false positive rate, predicted as positive but actually negative, accounts for the rate of negative in all real cases. The higher the FPR, the higher the likelihood of being predicted as a positive case in the negative case:

TPR: the true case rate, predicted as positive case and actually positive case, is the ratio of positive cases in all the true cases. The higher the TPR, the higher the likelihood of being predicted as positive in positive cases:

and a curve drawn by taking FPR as a horizontal axis and TPR as a vertical axis is an ROC curve. The area under this curve is the AUC (AreaUnderCurve) value.

Auc=1, a perfect classifier, when this prediction model is used, there is at least one threshold to yield perfect predictions. Most predictive cases do not exist as perfect classifiers.

Auc <1, 0.5< over random guesses. This classifier (model) can be predictably valuable if it is thresholded properly.

Auc=0.5, following machine guesses, the model has no predictive value.

AUC <0.5, reverse prediction, outperforms random guess.

Further, adjusting the evaluation model includes:

the unselected sub-indicators are replaced and the evaluation model is rechecked based on the consistency and AUC values.

In this embodiment S6, the classification of the evaluation result includes:

summarizing the evaluation models with the required classification level based on a preset model sample classification basis, listing the first m classes as high-quality evaluation models, and classifying the evaluation models based on a preset normal distribution theoretical value to obtain a classification result; sorting the evaluation models from high to low according to the division result, grouping the sorted evaluation models according to the corresponding theoretical occupation ratio, extracting the upper and lower scores of the evaluation models in different grades, and marking the upper and lower scores as a grade main body; wherein m is a positive integer, and m is less than or equal to h.

It should be noted that: results of numerical type (e.g., 73.5 minutes, 88.1 minutes) are inconvenient in practical applications. For example, enterprises at the same level may be different from each other by 1 to 2 points, but this is not an indication that there is indeed a risk difference between the two. Therefore, it is necessary to rank the rating results according to the usage habits, and the subjects at the same level are identified by commonly used rating symbols. Such as the common three-grade system (AAA/AA+/AA/AA-/A+/A/A-/BBB+/BBB-/BB-/B-/CCC/CC/C), L/R system (L1/L2/L3/L4/L5/L6/L7/L8/L9/L10, R1/R2/R3/R4/R5/R6/R7/R8/R9/R10), etc. The specific division is based on the following:

first, the rating of the credit of the assessment model reflects the probability of an breach of the assessment model over a period of time in the future. Theoretical studies referring to bond markets at home and abroad show that it is generally assumed that the evaluation models of different breach probabilities should satisfy normal distribution. Thus, the modeling process also typically assumes that the final result satisfies a normal distribution.

Second, different levels are typically applied to the criteria that use the rating results for admission in investment activity by institutions. Common examples are: the assets of the internal evaluation AA and above can be directly invested, the investment A-and below are non-investable, and if the investment is required to be independently approved in the middle area.

Therefore, when the evaluation model defines the gear, it is generally as follows: the division satisfies normal distribution; samples in the direct investment range are about 15-20% of samples in the market. The specific rules are determined according to the internal management method of each user organization.

For example: step one: we summarize the rating models that require classification levels.

Step two: assuming that 11 levels are expected to be divided, the first three levels are classified as high-quality jettisonable, i.e., the following normal distribution theoretical values are selected for division.

Step three: the model scores of all the evaluation models are ranked from high to low. Grouping according to the corresponding theoretical duty ratio, and setting the upper and lower score limits of the evaluation models in different grades.

It should be noted that the tail evaluation model score may be the same as the theoretical occupation ratio. At this time, the same-score evaluation model is required to be drawn into a first grade, so that the usability of the model is ensured. The situation that the same-score evaluation model is divided into different grades cannot occur.

For example: assuming a total of 1000 samples, according to the AAA theory, of 4.8%, the first 480 evaluation models should be assigned to the AAA size. However, it is possible to score the 480 th evaluation model in the same manner as the 481, 482, 483 evaluation models, and the 480 th evaluation model should be assigned to the aa+ stage at this time, so as to ensure that the AAA and aa+ evaluation models have clear score lines.

The partial data in the formula is obtained by removing dimension and taking the numerical value for calculation, and the formula is obtained by simulating a large amount of acquired data through software and is closest to the real situation; the preset parameters and the preset threshold values in the formula are set by those skilled in the art according to actual conditions or are obtained through mass data simulation.

The above embodiments are only for illustrating the technical method of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present invention may be modified or equivalent thereto without departing from the spirit and scope of the technical method a of the present invention.

Claims (6)

1. A credit evaluation method for a liability principal, comprising the steps of:

s1: acquiring industry data and annual report financial data of debt and marketing enterprises within a plurality of years, marking the industry data and the annual report financial data as sample data, and grouping the sample data according to industry types to obtain k groups of basic models; extracting an external evaluation result of the debt subject;

wherein the industry data comprises a plurality of major indexes, and the major indexes comprise a plurality of sub indexes; the major category of indicators includes capital structure indicators, operational capability indicators, profitability indicators, payability indicators, growth capability indicators, and regional indicators, the annual report financial data includes a profit-and-loss table, a liability table, and a cash flow table, and k is a positive integer;

s2: screening a plurality of large indexes and a plurality of corresponding sub indexes to obtain a modeling index;

s3: selecting a plurality of preset scoring points corresponding to the sample data as scoring standards based on the modeling indexes and the basic model to obtain scoring intervals;

s4: carrying out weight identification on a plurality of large-class indexes in the industry data based on an analytic hierarchy process to obtain index weights, and identifying a plurality of sub-index weights based on the index weights and the number of sub-indexes in the large-class indexes to obtain final weights;

s5: generating h evaluation models by using the scoring interval and the final weight corresponding to the modeling index, and checking the evaluation models based on consistency and AUC values to obtain a checking result; calibrating the evaluation model according to the test result; wherein h is a positive integer;

s6: and inputting the industry data and annual report financial data of a plurality of subjects to be tested into an evaluation model to generate an evaluation result, classifying the evaluation result, and distinguishing the level subjects of different credits by using the rating symbol.

2. The credit evaluation method of a debt subject according to claim 1 wherein the screening of the plurality of major categories of indicators and the corresponding plurality of sub-indicators to obtain the model entering indicator includes:

a1, screening deletion rate: extracting sample data, and removing large indexes with the loss rate of annual report financial data exceeding a% in industry data to obtain screening indexes, wherein a is a positive integer, and a is less than 100;

a2, screening consistency of external evaluation: judging whether the numerical value of the sub-index in the screening index is consistent with the external evaluation result or not according to the external evaluation Somer' sD of the corresponding sample, and obtaining a judgment result;

a3, explanatory screening: judging whether the change area of the neutron index along with the external evaluation result is consistent with the corresponding economic meaning or not, and obtaining a judgment result;

a4, screening relevance: judging whether the relevance of any two sub-indexes in the same large-class index is greater than a preset relevance threshold value or not to obtain a judging result; the covariance method is adopted in the calculation mode of the relevance of any two sub-indexes.

And A5, recognizing the judging results of the external evaluation consistency screening, the interpretation screening and the relevance screening as reference indexes to obtain target indexes.

3. The credit evaluation method of a debt subject according to claim 1, wherein selecting a plurality of preset scoring points corresponding to sample data based on a model entry index and a basic model as scoring criteria, obtaining a scoring interval, comprises:

b31: extracting the numerical values of corresponding sub-indexes of all enterprises in a z-th basic model in a plurality of years to obtain a plurality of comparison data; wherein the corresponding sub-index is a sub-index of interest in the industry; wherein z is a positive integer, and z is less than or equal to k;

b32: sequencing a plurality of comparison data from large to small to obtain a sequencing result, and calculating a point value corresponding to a preset dividing point based on the number of submitted data;

b33: and marking the numerical value corresponding to the point value as a demarcation value according to the sorting result, and setting a scoring interval based on the demarcation value and the corresponding sub-index.

4. The credit evaluation method of a liability principal according to claim 1, wherein the identifying a plurality of sub-index weights based on the index weights and the number of sub-indexes in the major index includes:

c41: determining the reserved number u of the sub-indexes according to the large-class index weight and the number of the sub-indexes, wherein u is more than or equal to 1 and less than or equal to 5;

c42: determining the weight value of each sub-index according to the consistency of the external evaluation in the mathematical screening process;

c43: and marking the product of the weight index of the sub-index corresponding to the large-class index and the weight value of the corresponding sub-index as the final weight of the sub-index.

5. The method of claim 1, wherein said examining the assessment model based on consistency and AUC values comprises:

calculating a model result of the evaluation model, comparing the model result with the outer evaluation symbols 'D and ROC curve AUC values corresponding to the outer evaluation grades, and comparing the model result, the outer evaluation symbols' D and the respective outer evaluation grades with corresponding AUC values and preset thresholds; if the model result, the outer rating detectors' D, the corresponding AUC value of each outer rating and the preset threshold are all larger than the preset threshold, the evaluation model passes statistical verification; otherwise, the evaluation model is adjusted.

6. The credit evaluation method of a liability entity according to claim 1, wherein said classifying the evaluation result comprises:

summarizing the evaluation models with the required classification level based on a preset model sample classification basis, listing the first m classes as high-quality evaluation models, and classifying the evaluation models based on a preset normal distribution theoretical value to obtain a classification result; sorting the evaluation models from high to low according to the division result, grouping the sorted evaluation models according to the corresponding theoretical occupation ratio, extracting the upper and lower scores of the evaluation models in different grades, and marking the upper and lower scores as a grade main body; wherein m is a positive integer, and m is less than or equal to h.