Research on Network Lending Risk Analysis Based on Platform Efficiency

Optimizing the financial risk control mechanism is very important for small and medium-sized enterprises. It is an important foundation for lending companies to continue to operate steadily and ensure the safety of credit assets. Only by optimizing the financial risk control mechanism can small and medium-sized enterprises develop better and healthily. This study designs the data envelopment analysis (DEA) model (CCR and BCC) to evaluate the relative credit risk of the enterprise. According to the selected core indicators, comprehensive technical efficiency, pure technical efficiency and scale efficiency of each lending platform are calculated. In order to further explore the efficiency status of each lending platform and the reasons for its inefficiency, the efficiency pedigree was drawn according to the pure technical efficiency value and scale efficiency value of the platform. The main reason for the low efficiency of online loan platform was the low management level. Therefore, we need to improve the comprehensive management level of the platform, and strengthen the risk management control to improve the operation efficiency of the online loan platform. Finally the related policy proposal was put forward.

With the help of P2P technology and ecommerce technology, it lends idle funds owned by fund suppliers to fund demanders on the Internet platform, which breaks away from the traditional financial media. The Internet technology is used to match the fund supplier with the fund demander, which is different from the traditional loan mode of financial institutions.
The rapidly developing P2P network loan industry has solved the problem of insufficient financing needs and personal financial needs of small and medium-sized enterprises in China. From 2017 to 2019, the number of P2P online lending platforms has increased from 6584 to 6656, as shown in Figure 1. We can see that the number of platforms in P2P online lending industry has almost stopped growing. In 2017, there were 422 new platforms while only 71 new platforms were added in 2018 and only one new platform was added in 2019. It can be found that the number of new entrants in the P2P online lending industry has decreased significantly and even there has been no new platform for several consecutive months, which is in sharp contrast to the popularity in 2017.
In 2018, "Filing" has become the main keynote of the development of P2P online lending industry. In December 2017, the P2P platform was required to register and rectify. The qualified P2P platform is allowed to be filed and registered to ensure its normal operation. In the case of regulatory requirements for P2P online lending platform to reduce the business scale, the number of lenders and borrowers, the scale of P2P online lending industry is shrinking, and the number of exiting platform is increasing. The number of functioning platforms dropped sharply from 2448 in 2017 to 344 in 2019, in which not only a large number of platforms went out of business for transformation, but also a large number of problem platforms appeared, as shown in Figure 2. Among these problem platforms, according to the "zero one data" (https://www.01caijing.com), the number of websites closing platforms is the most, followed by the announcement of liquidation. In addition, the number of problem platforms with difficulty in cash withdrawal, closure of business, filing of cases, loss of contact and running away is also relatively high.
There are three main reasons for the significant decrease in the number of platforms. Firstly, the delay of filing has caused great uncertainty to the whole industry and many platforms choose to take the initiative to liquidate, transform or be dissuaded because they know that there is no hope of filing. Secondly, the  supervision has become stricter, such as the notice on carrying out compliance inspection of P2P online lending institutions issued by the online lending rectification office in August 2018. Finally, with the continuous emergence of problem platforms such as poor management, self-financing, and capital pool, investors lose confidence in the industry and the industry runs and the capital flow becomes tight, which makes a large number of platforms unable to adhere to operation due to lack of comprehensive strengthen (Gao et al., 2021;Gao et al., 2018).
Through the above analysis, we can see that under the new regulatory policies and domestic situation in China, many P2P companies have withdrawn from the market, and some companies are also facing transformation. Importantly, most platforms lack experience in risk management. They have no risk dynamic assessment system and risk quantitative indicators. This leads to the risk management method being too fragmented, simplified and unsystematic. Therefore, it is very important for the healthy development of lending platforms, especially small and medium-sized enterprises, to optimize the financial risk control mechanism.
Herein, in this paper, we first analyze the current situation and development trend of P2P platforms in China. Then, the methods of risk control analysis reported in the literature are reviewed. Besides, the models and methods used in this study are proposed. Subsequently, we analyzed the selected quantitative indicators and evaluated the risks in the lending business. Finally, we discussed the results and put forward suggestions for improving risk control management.

Literature Review
As technologies of big data and blockchain advancing, the financial credit risk  (Zhao et al., 2021;Song et al., 2018;Liu, 2020;Ahelegbey et al., 2019 (Zhang et al., 2020). It is worth noting that platform background is related to operational risk (Wang et al., 2020b;Ma et al., 2020).

Common Models of Credit Risk Measurement
There are many ways to measure credit risk, including expert analysis, Fisher linear discriminant analysis, Z-score model and so on.
For protecting the obtaining its interests and obtaining the confidence, banks must carefully evaluate the nature, capabilities, capital, collateral, and business prospects of their debts or their customers, called "5C" (Wahyuni, 2017). The "5C" analysis method is commonly used in expert analysis, which quantifies the relevant factors that affect the business performance, such as character, capital, capacity, condition and collateral, and then obtains the credit rating of the enterprise by weighted average. This method needs to rely on the professional opinions of experts. When experts evaluate enterprises, they mainly consider the above five elements. The most common problem in financial service industry is risk. One of the most critical risks is the default risk. One evaluation model used by Franata, Faturohman and Rahadi is 5C credit analysis (Franata et al., 2018).
According to the analysis of data by 5C analysis tools, researchers found that most borrowers could obtain a credit from PD. In general, the credit approval process includes credit application, research documents, bank information search, site visits, credit research analysis (Mardhotillah, 2019). The expert method is relatively comprehensive, but it is not difficult to see that this method will bring some subjectivity, too many qualitative components, and the opinions to provide a basis for whether a credit provider grants credit to a customer.
Moreover, a linear mixed model (LMM) was proposed as a new method to evaluate the credit risk of financial companies (Casin, 2018;Pérez-Martín et al., 2018). The premise of Fisher linear discriminant method is that the independent variables need to obey normal distribution, and the total covariance matrix of each group should be equal, which is not true in practical problems.
Altman (1968) put forward the Z-score Model. The model screened out five representative financial indicators of the same number of normal operating companies and defaulted companies through selection analysis. It established a multivariate linear model to get a total discriminant value, which is defined as Z-value. The conclusion is that when Z value is less than 1.81, the borrower is expected to default, indicating the great risk. When Z value is greater than 2.99, the risk of default is small. When Z value is between 1.81 and 2.99, the financial situation of the enterprise is unstable, so we need to continue to pay attention to the operation of the enterprise (Panigrahi, 2019). Although the Altman Z-score is "outdate", it is still the standard to measure most other bankruptcy or default prediction models. Furthermore, in numerous empirical studies, it is also regarded as the benchmark of credit risk measurement by researchers (Altman, 1968;Altman, 2018a;Altman, 2018b Model Using Altman has only 87.4% effectiveness. Therefore, the advantage of Z-score Model is to assess the credit status of companies according to the financial data of enterprises. The data is easy to obtain and has strong operability. However, in this multivariate linear model, the weight of each index and the benchmark setting of Z value need to be adjusted according to the actual situation, which leads to the enhancement of subjectivity and the lack of a unified standard.

Credit Risk Model Based on Specific Theory
As time goes on, more and more scholars used different tools and measurement methods to study credit risk and many advanced credit risk measurement models emerged (Yeh et al., 2012;Hamerle and Rösch, 2006;Tian, 2018  of PV model is to use a mark to market method, which not only applies to a single sample, but also can describe the risk of a group of samples and give the specific loss distribution. However, CPV model requires to obtain the default data of each country or industry in advance, so it is not easy to gain the data. CreditMetrics model was proposed by JP Morgan in 1997 (Morgan, 1997).
The model is based on the theory of value at risk (VaR). According to the historical default rate of the borrower and the credit rating transfer matrix given by the rating company, the distribution curve is obtained and the VaR value is cal-  (Tian, 2018). The following key indicators need to be calculated to assess the CR of PPP projects, including the default probability (DP), the recovery rate (RR) and the exposure at default (EAD). The DP was analyzed by using improved Credit Metrics model and Monte Carlo simulation. Moreover, the evaluation method was successfully applied in the case of PPP financial project (Wang et al., 2020c).
CreditRisk+ model is based on actuarial theory to study the default risk of loan. Debali explains the different characteristics of CreditRisk+ model, and thinks that it can calculate the default probability of credit portfolio (Derbali, 2018). In the credit risk model, the numerical calculation of the risk measure is equivalent to evaluating the tail expectation of the investment loss distribution.
Although the moment generating function of the loss distribution was obtained with analytical closed form in CreditRisk+ model, it is still a challenge to calculate risk measurement and risk contribution efficiently, accurately and reliably (Kwok, 2020

Credit Risk Model Based on Machine Learning
As the development of science and technology and the more and more extensive application of computer, the method of learning from data has also been introduced to the field of credit risk measurement, among which back propagation (BP) neural network and support vector machine (SVM) are more active.
As many machine learning and data mining technologies are applied to financial decision-making, the research on credit risk assessment has attracted a lot of attention. BP neural network has become a common choice for credit risk assessment, but many researchers point out that classifier integration is better than a single classifier (Shen et al., 2019). BP neural network is mainly composed of two processes of information forward propagation and back propagation. The neural network is composed of input layer, middle layer and output layer. Guo assessed the traditional financial risk and information technology risk of P2P lending companies by analyzing the large-scale lending data from 2015 to 2019.
They found that the algorithm based on BP neural network can effectively decrease the risk of investors (Guo, 2020). In Zhou's study, they have proposed a big data mining method based on particle swarm optimization (PSO) back propagation (BP) neural network. This method is used for financial risk management of commercial banks deploying the Internet of things. On the data sets  (Du et al., 2021). The advantage of this method is that the basic idea is simple and easy to understand and it can deal with the problem of data classification and complex relationship. However, BP neural network in the setting of learning mechanism is complex and time-consuming, and requires sufficient training samples and professional computer and mathematical basis, which increases the difficulty of dealing with the problem.
Corinna and Vapnik proposed support vector machine (SVM) method (Cortes and Vapnik, 1995). With the development of time, this method was widely used in nonlinear separable classification problems. The basic idea of SVM is to minimize the structural risk and separate the two kinds of sample data accurately.
The biggest advantage of support vector machine is that it can solve the nonlinear and small sample problem. Considering that the parameters we choose have a great impact on the performance of SVM, so it is our previous work to optimize it (Wang and Li, 2019).

Research on the Theory and Method of Efficiency Evaluation
Financial inclusion is based on the requirements of equal opportunities and the principle of business sustainability to provide appropriate and effective financial services for all social strata and groups with financial service needs at an affordable cost. It is crucial to a country's competitiveness . After analyzing the advantages and disadvantages of different models, we can find that there is no unified scheme for credit risk measurement. The purpose is to propose the research scheme of credit risk measurement of P2P network lending platform and introduce the models and main methods involved in the empirical study.
A. Charnes and W. W. Cooper have put forward the data envelopment analysis, which is an input-output ratio to study the effectiveness of decision-making units. The advantage of data envelopment analysis is that subjective factors can be eliminated because there is no need to know the form of production function.
Moreover, it has less research constraints and is useful for the efficiency assessment of multiple inputs and outputs (Charnes et al., 1997). DEA model began to be applied in the field of business efficiency evaluation and financial health evaluation (Mendelová and Tatiana, 2017 risk, but also evaluate a single production unit according to the whole sample (Horváthová and Mokrišová, 2018).
Davis et al. believe that P2P online lending has injected fresh vitality into the traditional lending market to meet the needs of investors and consumers. The platform's operation and management capacity, organizational production capacity and environment are very important. The imbalance of these factors will lead to the platform's resources and opportunities ineffective, resulting in the collapse of the platform (Razi et al., 2004). The survival of the platform depends on the age, scale and life cycle of the enterprise (Bruton and Rubanik, 2002). The management ability of platform operators plays a key role in the success or failure of small and micro platforms (Honjo, 2000). There is a relationship between the scale of the platform and its failure, which the smaller the scale of the platform, the greater the possibility of its failure (Kale and Arditi, 1998;Stokes and Blackburn, 2002). Therefore, data envelopment analysis is widely used. Based on the above research literature, this study considered the availability of platform operational data and operational risk factors. In this study, DEA-CCR and DEA-BCC model were used to analyse the data of the sample platform.
China's internet finance is developing rapidly and on a large scale, which has great influence on China's economy. The central bank has successively formulated the regulatory system on the Internet lending platform, marking that the Internet financial lending platform is basically under systematic supervision. Therefore, this study chose the Internet financial lending platform as the research object to conduct risk control evaluation research. According to the selected core indicators, the comprehensive efficiency, pure technical efficiency and scale efficiency of each lending platform are calculated by using the classic DEA model (CCR and BCC). According to the comprehensive efficiency performance, it can include three grades: excellent, good and medium. Moreover, different from the traditional machine learning methods, this study analyzes and optimizes the DEA model from the perspective of multi input and multi output.
Online lending banks are developing rapidly, and many platforms have defaulted when new platforms are established. On the one hand, the rapidly increasing online lending platform provides investors with more convenient and diversified services. On the other hand, it makes choice more difficult. Compliant online loan platforms generally operate in a standardized manner. The possibility of default is relatively small, and its risk control audit is also relatively strict. Compliant online loan platforms generally operate in a standardized manner. The possibility of default is relatively small, and its risk control audit is also relatively strict. The risk assessment of online loan platform needs to comprehensively consider many factors. This paper proposes to conduct risk analysis from the perspective of platform risk efficiency.

Methods
In 1978, the first classic DEA model, CCR model, was put forward by Charnes, where, X i is the i-th input indicator of DMU and X ij is the i-th input indicator of j-th DMU. Y r is the r-th output index of DMU, and Y rj is the r-th output index of j-th DMU.
The scale efficiency can be obtained by the following formula: Comprehensive technical efficiency Pure technical efficiency Scale efficiency = * Using the selected methods in this study, the results will not be affected by

Efficiency Analysis of P2P Platform
Firstly, this chapter introduced the screening principles, data sources and processing methods of sample indicators. Secondly, it introduced the DEA method and the reasons and empirical results of using this method to evaluate the relative credit score of enterprises. Finally, a summary of this chapter was given.

Calculation of the Efficiency Value
The data source is the first quarter of 2019 data of Internet financial companies provided by the website data platform of "Wangdaizhijia" (https://www.wdzj.com). There are 27 platforms in the sample whose pure technical efficiency is equal to 1. For some platforms where pure technical efficiency is not effective, attention should be paid to the rational allocation and effective use of input resources, and the maximum benefit of input factors should be brought into play by improving the structure of input factors.
The closer the scale efficiency is to 1, the closer the project is to the most ap- As the capital investment of the registered capital, the assets that can form the P2P platform are the guarantee and basis for it to bear the operating risks and risk responsibilities.

Expected Rate of Return (%)
It is used to characterize the variable cost of platform operation. Now that the online loan industry is increasingly standardized and operating costs are gradually rising, online loan platforms with high interest rates often have large investment traps, so investors get reasonable through P2P platforms.
To some extent, the expected return reflects the robustness of platform operations and the effectiveness of risk management and control.
3 Transaction Points An indicator used to characterize the trading volume of the platform. If the transaction score is higher, the transaction volume of the platform will be higher.

Popularity Points
An indicator used to characterize the number of investors and borrowers on the platform. The higher the popularity score, the more people will invest and borrow on this platform.

Liquidity Points
In addition to the liquidity of investors, the liquidity of the platform will also be considered. Net outflow indicators and future repayment pressure indicators will be established. For platforms with a shorter project period, the liquidity points will be lower.
6 Leverage Points An indicator used to characterize the risk tolerance of the platform. If the leverage point is higher, the potential capital leverage of the platform is smaller and the risk tolerance is higher.

Dispersion Points
An indicator used to characterize the diversification of platform borrowing and investment funds. If the diversification score is higher, it indicates that the investment and borrowers of the platform are more diversified, and the operating risk of the platform is lower.

Technology Points
An indicator used to characterize the technical strength of the platform. If the technical score is higher, the stronger the technical strength of the platform is.
9 Transparency Points An indicator used to characterize the transparency of platform information. The higher the transparency score is, the more open the platform information is and the more transparent the platform is.
10 Brand Points An indicator used to characterize the popularity of the platform. If the platform brand score is higher, it means that its popularity is higher and it can be recognized by the lender.

Compliance Points
An indicator used to characterize the degree of compliance of the platform and the difficulty of compliance transformation at this stage. According to regulatory opinions, each business requirement and prohibited items are quantified, and the degree of compliance of the platform is graded.

Rating Classification
According to the comprehensive efficiency value in Table 3

Efficiency Pedigree Analysis
In the DEA model, the comprehensive efficiency is equal to the multiplication of pure technical efficiency and scale efficiency. In order to further explore the efficiency status of each lending platform and the reasons for its inefficiency, an efficiency pedigree diagram is drawn based on the calculation results in Table 3.

Conclusion and Policy Recommendations
This study provides a theoretical explanation of the models involved in the empirical study in accordance with the logical sequence of writing. First of all, in the selection of indicators, not only the meaning of credit risk represented by the indicators is considered, but also the credit risk analysis indicators are finally selected based on the availability of data, and the indicators are handled differently from the previous literature in order to reduce the indicator distortion caused by time period or accidental reasons. In order to deeply analyze the credit risk of P2P network lending financial enterprises, this study designs the DEA model to evaluate the relative credit risk of the enterprise. In the combination of methods, a different research method is adopted from the predecessors, which explains the feasibility of the two used to assess corporate credit risk, and proposes a new idea for subsequent empirical research.
The difference in efficiency between online lending platforms is mainly due to the difference in risk control and management level of online lending platforms.
If the size of the enterprise remains unchanged, it is very important to strengthen the level and quantification of risk management to improve its comprehensive efficiency. The risk management level of the platform can be improved in terms of improving liquidity, anti-risk ability and platform openness. For platforms which are not of the "double high" type, it is necessary to improve the technical strength of the platform and the business level. Firstly, the platform should strengthen the pre-borrowing review and refine the review of borrowing information. Secondly, the platform should conduct professional training for employees to improve their awareness of risk control. Finally, the platform should pay more attention to post-loan supervision. It is necessary to understand the use of borrowers' funds to minimize the risk of default. In addition, efficiency indicators do not adequately represent platform risk. It cannot be used as a single evaluation of platform risk. Investors need to fully consider risks and make prudent decisions. P2P network lending platform not only provides a reliable investment channel for social idle funds, but also provides a stable source of funds for borrowers. This alleviates the imbalance between supply and demand of social funds, and meets the needs of the groups that traditional financial institutions can not cover. In China, regulatory authorities and the platform itself should take some measures to control the credit risk of the P2P Internet lending industry. The specific recommendations were as follows: 1) With more specific and pragmatic laws and regulations system, we will promote the standardized construction of online lending platform and realize the sound development of the industry. In the actual implementation process, the regulatory authorities must adhere to the principle of strict law enforcement and investigation when there are laws to abide by. We will severely punish and D. M. Li et al.
clear illegal platforms, and actively promote the implementation of policies. We should improve the market exit mechanism, which can effectively protect the interests of investors.
2) Government should accelerate the docking of the central bank's credit information system and P2P network lending platform and coordinate the cooperation of multiple departments to establish a full coverage, high standard and deep-seated social credit system, comprehensively recording and rating citizens' credit information and realizing information sharing.
3) For P2P platform, it should actively respond to national policy planning to bear social responsibility and actively innovate and strive for development. P2P platform should also fully play to its role as an information intermediary to improve public awareness of online lending industry and actively carry out risk education for investors.

Limitations and Future Research
Limitations: In the selection of indicators, in order to make the data objective, fair, accessible and operable, this study only selects some risk indicator data of P2P online loan platform for empirical analysis. Other environmental factors or other financial data that have an impact on the platform's credit status are not considered. If we consider all kinds of data, the empirical model may have better results.
Future research: In the credit evaluation of P2P network lending industry, we can use a reasonable method to bring some environmental factors, macroeconomic factors and other data into the model, making the index data more comprehensive. A new model to study the credit risk management of P2P network lending industry and platform, which makes a more detailed analysis of the credit risk factors of P2P network lending industry and platform.