Research On Credit Risk Quantification And Credit Decision Of Small And Medium Sized Micro-enterprises In Multi Scenarios

Along with the increasing development of China’s financial market,Micro,Small and Medium Enterprises(MSMEs)have become a major force in China’s economic and social development,playing a crucial role among market entities.MSMEs boast advantages such as flexible operations,low organizational costs,and convenient transfers,and they are capable of adapting to various market and consumer needs.They play a crucial role in expanding society’s employment,improving people’s livelihood,promoting stability and national tax revenues.However,the current challenges faced by most MSMEs,including smaller scales,weaker strengths,lower creditworthiness,and unstable business conditions,result in a lack of effective collateral.The mismatch between their economic contributions and the scale of bank financing perpetuates difficulties and expensive financing problems.Therefore,exploring the present development scenario of MSMEs and establishing a scientific and effective credit and risk evaluation methodology system are of paramount importance.This endeavor holds significance for innovating financing modes,responding to risk management,and overseeing financial services for MSMEs.Financial institutions currently rely heavily on financial data to assess credit and manage risks,with particular attention to credit risk as one of the most significant concerns in their lending practices.The main problems of the existing credit scoring model for MSMEs include the following aspects: the scoring model has a poor credit rating effect when there is a scarcity of information,and it requires a large amount of historical data of MSMEs,while MSMEs generally have inadequate or missing financial data;the existing indicator weighting method cannot reflect the indicator’s ability to discriminate between defaulting and non-defaulting enterprises;Furthermore,the existing models do not adequately consider the changes in the quantitative strategy of MSMEs’ credit risk response during unexpected events.Most notably,research on credit assessment tends to overlook the interaction influences between affiliated enterprises.Lastly,in the era of big data,a variety of non-financial data sources should be introduced into the credit assessment as well,with the goal of improving the reasonableness and effectiveness of the credit assessment strategy.In conjunction with the questions above,this thesis’ s research primarily unfolds in five perspectives.(1)Research on Credit Strategies for MSMEs Based on Improved Gradient Descent and Integrated LearningWhen banks consider whether to lend money to MSMEs,they focus on the selfstrength and credit rating of the MSME.In light of this,the thesis combines highfrequency indicators recognized by authoritative financial institutions both domestically and internationally,screening core indicators in quantifying credit records.With the goals of maximizing bank loan proceeds,minimizing customer turnover default risk,and considering constraints such as bank loan amount,lending interest rate,and enterprise default probability,the thesis establishes a nonlinear multi-objective planning model for the bank’s credit strategy.It constructs a credit rating system for MSMEs with the aim of achieving a higher credit rating and a lower loss rating.Utilizing the concept of integrated learning and employing the gradient descent algorithm for iterative optimization,the system solves the model to determine the optimal credit strategy for the bank.Finally,a comparison is made with the decision tree algorithm and BP(Back Propagation)neural network model.The results demonstrate that the predictive accuracy of the model designed in this chapter reaches 96.76%,signifying a significant improvement and validating the model’s effectiveness.(2)Research on Credit Strategies in Information Incomplete Scenarios Based on Double-Layer XGBoostA double-layer XGBoost(e Xtreme Gradient Boosting)prediction method is proposed using a small amount of collected data of MSMEs with credit records.The dataset of enterprises with credit records is trained by XGBoost,and thus designing a double-layer XGBoost algorithm for enterprises without credit records,which realizes credit ratings of enterprises without credit records and the prediction of defaults.With the objectives of maximizing bank loan proceeds and minimizing customer default risk,a credit strategy model for MSMEs is established.The model employs the particle swarm algorithm for solving.At the same time,the k-fold cross-validation method is used to solve the inaccurate prediction problem that may be brought by the small-scale data set,with the target of improving the prediction accuracy of the model.Finally,a comparative analysis is conducted using performance indicators such as AUC(Area Under Curve),ACC(Accuracy),and MSE(Mean Squared Error).The results demonstrate that,compared to other prediction models,the dual-layer XGBoost model exhibits superior predictive performance in different prediction processes,highlighting the excellence of the dual-layer XGBoost model constructed in this study.(3)Research on Risk Quantification Strategies under the Influence of Unexpected Factors Based on the Mutation Level MethodUnforeseen factors have varying impacts on MSMEs in different industries.In order to examine how the credit strategy of financial institutions can be optimized and adjusted,this study,grounded in the real context of the COVID-19 pandemic,utilizes collected data related to MSMEs and categorizes enterprises accordingly.By comprehensively considering macro and micro impacts,the mutation level method is applied to the adjustment of the bank’s credit strategy for enterprises in unforeseen circumstances.Using principal component analysis and combination weight method,the credit risk of enterprises is quantitatively analyzed,categorizing the impact of the epidemic on different enterprises into five classes to determine the bank’s credit adjustment strategy.Genetic algorithms are then employed to obtain the final optimal credit strategy.Finally,through simulation experiments,the effectiveness and feasibility of the model are verified.(4)Research on Credit Strategies Based on Markov Random Fields and Knowledge Graphs of Firm LinkagesUsing a small amount of collected data related to MSMEs with credit records,the degree of association between enterprises is fully excavated by collating the sales data of MSMEs,and the credit rating is jumped out from the limitation of only considering the enterprise itself,the influence of the association between enterprises on the credit rating of enterprises is increasingly considered.By establishing an improved Markov Random Field model and utilizing the Bean Search algorithm,the study updates known enterprise credit ratings.Then,using genetic algorithms,it solves the credit strategy model,offering guidance to the bank for lending strategies regarding MSMEs.The experiments indicate that after combining the consideration of inter-enterprise correlation,the constructed model can produce more comprehensive and accurate credit decision-making results.(5)Research on Credit Strategy in Big Data Volume Scenarios based on Multimodel FusionBased on previous studies,the thesis constructs a dual-model fusion algorithm based on feature selection and channel attention mechanisms,which uses the correlation coefficient method and GBDT(Gradient Boosting Decision Tree)to screen the original features.Compared to the general base model,stacking model,and fusion model,the prediction results of the constructed NABS(NN-ATT-Bayesian-Stacking)model have a very ideal distribution,with an AUC value of 0.9675,which is 4%～5% higher than that of the convolutional neural network combined with a short and long memory neural network model based on the attention mechanism designed in the stacking model,and it is demonstrated that the constructed model exhibits good robustness,enabling accurate and reliable assessments in scenarios related to financing and lending for MSMEs.