Research on the Credit Risk Assessment of Small and Medium

Enterprises Based on BP Neural Network

Xiaoying Zhang

and Xing Tan

School of Management, Wuhan University of Technology, Wuhan, China

Keywords: Small and Medium-sized Enterprises, BP Neural Network, Credit Risk, Evaluation Model.

Abstract: Small and medium-sized enterprises play an important role in China's economic development. However, they

face many problems in their development, among which credit financing is particularly important. This paper

mainly focuses on the correlation of small and medium-sized enterprise credit problems. Combined with the

characteristics of small and medium-sized enterprises in China, this paper builds a more scientific and

reasonable evaluation model based on the evaluation index system and uses BP neural network to grade the

credit capacity and risk of small and medium-sized enterprises. It also puts forward relevant suggestions from

the perspective of enterprises and banks to help banks make reasonable decision-making plans and find the

optimal credit strategy.

1 INTRODUCTION

With the rapid development of China's economy and

the adjustment of economic structure, China's small

and medium-sized enterprises have developed rapidly

and played an important role. In the development

process of small and medium-sized enterprises, there

are often various problems due to their scale and

organizational structure. High credit risk, difficult

financing, and expensive financing are still the

primary problems faced by China's small and

medium-sized enterprises. Through investigation and

research, the existing internal management

regulations of commercial banks in China are not

very perfect and lack scientific and reasonable credit

risk assessment methods and systems for the

evaluation and prevention of small and medium-sized

enterprise creditability. Therefore, it is particularly

important to establish a perfect credit risk evaluation

model for small and medium-sized enterprises in

commercial banks.

Based on the theoretical research at home and

abroad, Wu Jingmei puts forward an evaluation

system suitable for the company's financial quality

and solvency. She is the first economist in China to

deeply study the credit evaluation index system.

https://orcid.org/0000-0001-5251-9718

https://orcid.org/0000-0002-4631-9987

Based on a commercial bank in Zhejiang Province,

Shi Zhen and Xu Feng used BP neural network

training sample model to predict and evaluate their

risk evaluation. Through empirical research, the

model can be used as the basis for credit risk decision-

making of urban commercial banks. Based on the

credibility research theory, Liu Cheng and Liu

Xiangdong combined analytic the hierarchy process

with credibility measurement to establish a

comprehensive risk index system from multiple

levels and form a framework and model that can

reflect the characteristics of small and medium-sized

enterprises. Wu Jingru selects and modifies the

indicators that can reflect its characteristics,

establishes and tests the credit evaluation index

system of small and medium-sized enterprises by

using fuzzy analytic hierarchy process.

Based on previous studies, combined with the

characteristics of small and medium-sized enterprises

in China, this study intends to build a more scientific

and reasonable evaluation model based on the

establishment of the evaluation index system, and use

BP artificial network to evaluate the credit capacity

and risk of small and medium-sized enterprises in

China.

Zhang, X. and Tan, X.

Research on the Credit Risk Assessment of Small and Medium Enterprises based on BP Neural Network.

DOI: 10.5220/0011159900003440

In Proceedings of the International Conference on Big Data Economy and Digital Management (BDEDM 2022), pages 82-86

ISBN: 978-989-758-593-7

2 DATA SOURCES AND INDEX

SELECTION

2.1 Index System Construction

This paper will build a credit strategy system from

two aspects of enterprise-strength and credit level

(see Table 1). The hierarchy under "credit level" is

divided into the following two categories: credit level

and default. The hierarchy of "enterprise-strength" is

divided into the following four categories: capital

flow, transaction vitality, profitability, and supply,

and demand. To analyze and solve the problems more

specifically, these four aspects are further subdivided,

and it is decided to select 10 quantities such as total

annual profit, profit margin, price and tax stability of

purchase (Sales), the efficiency of purchase (Sales)

invoice and invoicing stability of purchase (Sales)

invoice as the evaluation index.

Table 1: Index System.

Primary index Secondary index

Enterprise strength

Input price tax stability

Output price tax stability

Input invoicing stability

Output invoicing stability

Total profit for the year

Profit margin

Credit level

Credit rating

Default or not

2.2 Data Sources and Assumptions

The data in this paper are all from the relevant invoice

data of 123 enterprises with credit records and 302

enterprises without credit records provided in the

appendix of question C of the "2020 Higher

Education Association Cup National Mathematical

Contest in Modeling for College Students" released

by the organizing Committee.

In order to ensure the scientificity and rationality

of the system construction, the following assumptions

are established:

(1) It is assumed that the screening data is true and

effective without deviation.

(2) It is assumed that the enterprise has a certain

risk control ability.

(3) It is assumed that in addition to the selected

index system, other factors have little impact on risk

assessment and decision-making.

(4) It is assumed that the optimal loan amount

should maintain the same change rate as the

enterprise as a whole.

3 ESTABLISHMENT AND

SOLUTION OF THE MODEL

3.1 Model Overview

BP neural network was proposed by McClelland and

Rumelhart in 1986. It is a kind of multilayer feed-

forward neural network trained by an error

backpropagation algorithm. BP neural network does

not need to determine the mathematical equation of

the mapping relationship between input and output in

advance but can get the result closest to the expected

output value after giving the input value through its

training and learning of corresponding rules. Its core

idea is the gradient descent method, which uses

gradient search technology to minimize the error

mean square deviation between the actual output

value and the expected output value of the network.

The structure of the BP neural network is shown in

Figure 1:

(1)

Figure 1: Structure of multilayer neural network.

3.2 Model Establishment

For a single neuron, its structure includes input,

synaptic weight, sum plus bias, activation function,

and output.

Figure 2: Schematic diagram of the single-neuron structure.

Research on the Credit Risk Assessment of Small and Medium Enterprises based on BP Neural Network

1) Input vector of the input layer

𝑋=

(

𝑋



,𝑋



,…,𝑋



)

as 𝑋



=(𝑋



,𝑋



,…,𝑋



)

2) Hidden layer vector of layer 1:

𝐻



=ℎ





,ℎ





,…,ℎ





,…,ℎ









as 𝑙=2,3,…,𝐿 − 1,𝑗=1,2,…,𝑠



3) Output layer output vector:

𝑌=

(

𝑦



,𝑦



,…,𝑦



)

4) Set

𝑤





is the connection weight from the

𝑖th neuron in layer 𝑙−1 to the 𝑗th neuron

in the layer, is the

𝑏





offset of the 𝑗th

neuron in the layer, including:

ℎ





=𝑓(𝑛𝑒𝑡





)

𝑛𝑒𝑡





=𝑤













+𝑏





𝑛𝑒𝑡





is the input of the 𝑗th neuron in the lth

layer, and is the activation function. The activation

function can introduce nonlinear function. In this

paper, the Sigmod function is selected as the

activation function, that is:

𝑓

(

𝑥

)

1+𝑒



5) The error function is introduced to measure

the gap between the output result and the

expected output. Assuming that there are 𝑝

training samples of enterprise data, 𝑑(𝑖) is

the expected output corresponding to,

assuming that a single training sample has 𝑛

outputs, 𝐸(𝑖) is defined as the training

error of a single sample:

𝐸

(

𝑖

)

𝑑



(

𝑖

)

−𝑦



(

𝑖

)









therefore, the global error function:

𝐸=

2𝑝



𝑑



(

𝑖

)

−𝑦



(

𝑖

)













The weight and bias are updated. In BP neural

network, gradient descent method is generally used to

realize this step:

𝑤





=𝑤





−𝛼

𝜕𝐸

𝜕𝑤













−α

∂E

∂b





3.3 Solving the Model

In this paper, 123 known credit rating and default

record data are used as the training set and prediction

set. The proportion of training and prediction samples

is about 80% and 20% respectively. That is, the data

of the first 100 enterprises are used as the training set

and the data of the last 23 enterprises are used as the

prediction set. The training times are determined as

15000 times, the convergence error is 0.1 and the

learning rate is 0.0001. After the results of the

training set converge, the data records of the last 23

enterprises are put into the trained neural network for

the test, and the accuracy is 82.6%.

The prediction accuracy is high, so further

prediction can be carried out. The data of 302

enterprises with unknown credit ratings can be

brought into the trained BP neural network. The final

results are shown in Table 2:

Table 2: Evaluation Results.

Score Level Number

>0.9 Excellent 21

0.75-0.9 Good 114

0.6-0.75 Commonly 143

<0.6 Poor 24

Based on the analysis of the advantages of the

model algorithm, the value of AUC is 0.74, which

indicates that the model has a predictive value and the

establishment of the model is more reasonable

4 CONCLUSIONS

4.1 Conclusion

In real life, due to the scale and structure of small and

medium-sized enterprises, banks will evaluate the

loan risk of each enterprise when providing loans.

Banks prefer to provide loans to enterprises with

strong strength and stable supply-demand

relationships, and will give preferential interest rate

policies to enterprises with high reputations and low

credit risk. To do a good job in loan risk management,

banks need to establish corresponding models

according to the actual situation of enterprises,

comprehensively consider through the model

solution, and judge the risk of each enterprise, to

effectively avoid the emergence of non-performing

loans.

BDEDM 2022 - The International Conference on Big Data Economy and Digital Management

This study starts with the identification of credit

risk of small and medium-sized enterprises, and

mainly studies the application of the BP neural

network model in credit customer risk management

of commercial banks. This study takes the credit

rating of each enterprise as the evaluation standard of

loan risk, and the relationship between impact index

and credit rating can be trained through BP neural

network. It has the characteristics of simple structure,

fast convergence speed, easy implementation, and

high prediction accuracy. The establishment of this

model is convenient for banks to estimate the credit

rating of small and medium-sized enterprises more

scientifically and reasonably, and then formulate the

Credit Strategies of enterprises with different ratings

to achieve a win-win situation between banks and

enterprises.

4.2 Suggestion

(1) Enterprises should strengthen their

construction and improve their awareness and

ability of risk prevention and control.

Small and medium-sized enterprises should first start

from themselves, establish a set of management

structures suitable for the long-term development of

enterprises, and improve the internal control and

management system of enterprises. At the same time,

by constantly improving the enterprise system and

standardizing the financial system, enterprises can

improve the transparency and authenticity of

financial information, smooth the enterprise capital

chain, fundamentally control the generation of non-

performing loans and reduce the credit risk of

commercial banks. In terms of credit, enterprises

should repay the principal and interest on time,

strengthen communication and cooperation with

banks, and provide detailed and accurate financial

information and enterprise information to banks in

time. Enterprises should form a good cooperative

relationship with banks and maintain their own

reputation, so as to achieve win-win cooperation.

At the same time, enterprises should establish a

corresponding incentive system and supervision

system. Small and medium-sized enterprises through

the implementation of reward and punishment

measures to improve the enthusiasm of employees

and create more benefits for enterprises. The

corresponding supervision mechanism is conducive

to eliminating the improper behavior that is not

conducive to the development of enterprises.

(2) Banks should improve the credit evaluation

system and optimize the way of credit

management.

First, strengthen the pre-loan and post loan

supervision of small and medium-sized enterprises,

reduce the problem of information asymmetry with

enterprises, and establish a perfect information

disclosure platform to ensure the openness and

transparency of information. Second, optimize the

lending process of small and medium-sized

enterprises to achieve efficiency and effectively

avoid risks. Third, establish and improve the credit

risk early warning mechanism of small and medium-

sized enterprises in commercial banks, improve the

level of risk management, and use scientific and

effective risk identification and early warning models

to improve the Scientific and applicability of credit

decision-making.

(3) Improve the quality and ability of evaluation

personnel and optimize post setting

Banks should improve the quality of employees and

cultivate their risk assessment ability in various ways.

On the one hand, introduce high-quality risk

assessment talents, especially those with rich

experience, who can use and continuously improve

the risk assessment model to adapt to the actual

situation of different enterprises. On the other hand,

improve the bank's employee training system and

enhance employees' risk management awareness.

Relevant staff must master relevant legal knowledge,

credit rules and regulations, and deeply understand

the actual situation of the enterprise to evaluate and

handle credit business, so as to improve their own risk

management level.

A bank shall improve the comprehensive quality

of all staff in the whole process of credit business

handling, and improve their own ability and

professionalism. Only by strengthening the

awareness of relevant personnel from the level of

consciousness can they significantly improve their

personal ability and strengthen the risk control of

commercial banks.

At the same time, the bank shall improve the risk

management structure and clarify the responsibilities

of each department and post. Not only in quantity but

also in quality to ensure the matching of posts, so as

to form the organizational guarantee of risk

prevention and control.

(4) Strengthen external supervision and

communication and obtain data from multiple

channels

Banks should speed up the establishment of a perfect

information disclosure platform to ensure the

openness and transparency of market information.

Research on the Credit Risk Assessment of Small and Medium Enterprises based on BP Neural Network

Banks should constantly strengthen the information

exchange between enterprises and financial

institutions such as banks, implement effective

communication, and reduce the problem of

information asymmetry with enterprises, so as to

reduce financial risks and provide high-quality credit

services for enterprises.

A bank shall establish a responsibility system for

the authenticity of financial information, appoint

special personnel to supervise and implement an

accountability system. Banks should also strengthen

the management of the authorization system, clarify

the authorized personnel and operators, and conduct

strict supervision and inspection. At the same time,

with the advent and development of the big data era,

many information industries dominated by big data

are developing rapidly and playing an important role

in the modern Internet. Banks can use big data to

make credit data more specific, so as to promote the

construction of credit system more and more

comprehensive.

REFERENCES

Liu Cheng, Liu Xiangdong, Chen Gang Credit risk

assessment of small and medium-sized enterprises

based on credibility theory [J] Research on technology,

economy, and management, 2014,10:3-7.

Shi Zhen, Xu Feng Application of BP neural network in

credit risk assessment of Chenghai credit cooperatives

[J] Consumer guide, 2008 (21): 104-104.

Sun Qingwen, Yang Xuebin Research on credit risk

management of small and medium-sized enterprises in

China's commercial banks [J] Shanxi agricultural

economy, 2021 (03): 113-114.

Wu Jingmei Credit evaluation of Chinese enterprises and

securities [M] Beijing: CITIC Press, 1994.

Wu Jingru Construction of credit risk evaluation index

system for small and medium-sized enterprises [J]

Accounting newsletter, 2016 (26): 102-104.

Yao Nan Analysis on credit risk and Preventive

Countermeasures of small and medium-sized

enterprises in commercial banks [J] Modern marketing

(Business Edition), 2019 (09): 189.

Zhao Yu Causes and Countermeasures of the credit risk of

small and medium-sized enterprises [J] Investment and

cooperation, 2021 (07): 43-44.

BDEDM 2022 - The International Conference on Big Data Economy and Digital Management