| As an important power source and component of social and economic development,small businesses play an important role in promoting economic development,increasing employment opportunities and stimulating innovation.For a long time,the CPC Central Committee,the State Council,provincial and municipal governments and regulatory authorities have attached great importance to the financing of small businesses,and guided commercial banks to continuously strengthen credit supply through a series of policies and assessments.Commercial banks have also begun to actively explore the technical model of credit for small businesses and achieved certain results.However,on the premise of the mass development of small businesses credit,the smaller the average household credit amount of commercial banks to small businesses,the greater the inclusive benefit,but the difficulty increases.Compared with the huge credit financing demand of small businesses,it still presents the problem of insufficient credit supply.Therefore,it is difficult for commercial banks to change their lending preference and stimulate their lending willingness.This is also the key to the current financial supply-side reform.In this regard,from the perspective of the development of small businesses credit technology,this paper studies how the credit technology effectively stimulates the lending willingness to small businesses loans of commercial banks based on alleviating information asymmetry and reducing transaction costs.Moreover,the credit needs of long tail customers of small businesses can be solved and the long-term endogenous development of small businesses credit market can be realized.This is also the original intention of this paper to study the small business credit scoring.As a type of credit technology,small business credit scoring(hereinafter referred to as SBCS)has been gradually promoted and popularized in American commercial banks since the 1990s,and has been widely used in Japan,Canada,Australia and other countries.Mass experience has proved that SBCS technology plays an important role in improving approval efficiency and realizing mass development of credit business.Therefore,based on the current credit development model and technical path of small businesses in China,studying and learning from the experience of foreign SBCS practice is conducive to stimulating the lending willingness of commercial banks in China and breaking through the development bottleneck.It is not only the external requirement for the stable development of social economy in China,the deepening of reform in financial supply side,and the management responsibilities of Chinese government and bank supervision departments,but also the internal requirement for the diversified development of commercial banks,the adjustment of credit structure,and the balance of income and risk,which embodied both theoretical value and practical significance.For these reasons,combined with the credit model and technical characteristics of small businesses of commercial banks,this paper firstly uses the panel data model,analyzes and studies the support of credit technology development of different commercial banks for small businesses,and summarizes that as a quantitative evaluation technology,SBCS aligns with the credit technology path and can meet the internal needs of commercial banks’ development.At the same time,by introducing the principle and characteristics of SBCS,this paper uses "long tail effect" and evolutionary game theory to explain the mechanism and conditions that the lending willingness of commercial banks can be effectively stimulated by SBCS.Then this paper starts from the two features of "digitization" and "predictability" of SBCS,optimizing and integrating the binning results of K-means,Chi-merge and CART algorithms based on Ⅳ measurement,and then constructing the SBCS by using Elastic Net logical regression and onedimensional convolution neural network respectively.Finally,the paper puts forward some practical suggestions and summarizes main research results according to the actual situation in China.This paper is divided into six chapters.The first three chapters except for the introduction,mainly study how SBCS stimulates the lending willingness of commercial banks,the fourth and fifth chapters mainly study the relevant methods of SBCS implementation,and the last chapter is about suggestions and conclusion.The contents of each chapter are as follows:The first chapter is the introduction.This chapter mainly expounds the background and significance of the topic,systematically combs the relevant research literature of SBCS at home and abroad,and introduces the research framework,content,methods,main innovations and deficiencies of the paper.The second chapter mainly analyzes which type of small businesses credit technology can effectively stimulate the lending willingness of commercial banks.Firstly,through the analysis of the definition of small businesses and the characteristics of financing demand,it is concluded that there is a problem of insufficient credit supply in small micro credit market,especially the supply of credit to long tail customers in small businesses.Therefore,it is necessary to study the issue of commercial banks’ lending willingness from the perspective of supply.Based on this,taking the credit technology models formed by different types of commercial banks as the starting point,this paper summarizes and refines the characteristics of the three stages of credit technology development of small businesses,and uses the panel data model to analyze and demonstrate that the quantitative evaluation technology based on transaction loan technology is the development trend of credit technology of small businesses and meets the internal development needs of commercial banks,and SBCS belongs to this type of technology.The third chapter studies the mechanism that how SBCS effectively stimulating the lending willingness of commercial banks.By introducing the development history and principle of SBCS and combining with its basic characteristics and its impact on commercial banks,this paper demonstrates that SBCS can effectively stimulate the lending willingness of commercial banks from the aspects of both theory and practice.In theory,from the perspective of static theory,this paper uses "long tail effect" to analyze how SBCS can improve credit rationing in the credit market of small businesses,and from the perspective of dynamic theory,evolutionary game theory is used to analyze the long-term state and effective conditions of SBCS.In practice,through the practice of SBCS in commercial banks in developed and developing countries,this paper analyzes the relevant experience and enlightenment of this technology in stimulating the lending willingness.The fourth chapter mainly analyzes the effective implementation methods of SBCS.Firstly,starting with the two characteristics of SBCS,this paper first studies the data sharing mode and how to use alternative data to solve the basic problem of building SBCS from "digitization",that is,the effective source of data.At the same time,this paper studies how to effectively construct the SBCS model from the "predictability" characteristics,including the use of K-means,Chi-merge and CART algorithms in feature analysis,and the basic method and optimization method of Elastic Net logistic regression and onedimensional convolution neural network in the construction of the scoring model.Finally,the paper introduces the related methods,experience and practical effect in developing SBCS from two typical development cases of micro practice,that is,the general SBCS"Small Business Scoring Service" developed by FICO and the customized SBCS"Business Direct" developed by Wells Fargo.The fifth chapter is the empirical research according to the effective implementation method of SBCS.The SBCS model is constructed by using the relevant data of small businesses of a regional commercial bank.Most of the features of the data have the attribute of "hard information",and include the application of relevant alternative data,such as tax data,which is close to the reality of the credit business of small businesses.In the feature analysis,K-means,Chi-merge and CART algorithms are used to implement feature binning respectively,and the results are integrated and optimized according to the IV measurement.Then Elastic Net logistic regression and one-dimensional convolution neural network are used to complete the model training respectively,and the effect of prediction is tested by the test set.Finally,the conversion of scores is realized and the construction of SBCS model is completed.The sixth chapter is the suggestions and main conclusion.By analyzing the current situation,existing problems and advantages of SBCS in China,this paper puts forward suggestions from the two levels of national guidance and commercial bank implementation and proposes to build a multi-dimensional risk assessment and value discovery system of small businesses based on the general SBCS of credit bureaus and the customized SBCS of commercial banks.Finally,the paper draws the main conclusion and prospect the future research direction.Combined with previous studies,the possible innovations of this paper are as follows:First,it analyzes the relationship between the development of credit technology of different types of commercial banks and the lending willingness of small businesses.After constructing the lending willingness index of small businesses by using the entropy weight method,the panel data model is used to analyze the incentive effect of the development of credit technology of state-owned commercial banks,national joint-stock commercial banks and urban commercial banks on the lending willingness of small businesses.Among them,the incentive to the willingness of national joint-stock commercial banks is the strongest,followed by state-owned commercial banks and finally urban commercial banks.By combining the credit technology characteristics and technological development path of three types of commercial banks,it is demonstrated that the quantitative evaluation technology as transaction loan is the development trend and direction of credit technology for small businesses,which can effectively stimulate the willingness of commercial banks to lend to small businesses.Secondly,it analyzes how the SBCS stimulates the lending willingness of small businesses of commercial banks from two aspects of static theory and dynamic theory.In terms of static theory,the "long tail effect" is used to analyze the economic effect and scale effect produced by SBCS under the condition of effectively identifying risks,which will help stimulate the lending willingness of commercial banks and improve the credit rationing in the credit market of small businesses.In terms of dynamic theory,when considering the time factor,the evolutionary game theory is used to analyze the conditions for realizing the effectiveness of SBCS technology in the long-term state.The conclusion is that in addition to the effective improvement of SBCS must be greater than a certain critical value,it is more important to ensure the convenience and effectiveness of data collection of small businesses.Thirdly,it tries to use different methods to build SBCS and achieve good results.In addition to using K-means,Chi-merge and CART algorithms in feature analysis,and optimizing the results of feature binning according to the Ⅳ measurement,this paper uses Elastic Net optimization method based on logistic regression,which can make the model as sparse as L1 regularization,while maintaining the regularization properties such as L2,and improve the generalization of the model.At the same time,in view of the fact that neural network as a deep learning method has good fitting ability in feature extraction and prediction and has begun to be used in credit fraud identification,few literatures are using one-dimensional convolutional neural network to solve prediction problems,in particular to build SBCS model.By sorting the importance of the binned features and after binning and converting them into sequence data,this paper tries to build SBCS model with onedimensional convolutional neural network and optimize the model by using the normalization method and the Dropout regularization method.The overall prediction effect of the model,it shows that the F1-score and the Accuracy are better than the logistic regression model and the neural network model under non sequential characteristic data. |
PDF Full Text Request