Research On The Credit Risk Assessment Of Farmer's Microfinance Based On BP Neural Nerwork

At present, more than 90% of the Rural Credit Cooperatives offer microfinance business to the farmer all over China. The microfinance for farmers has played an important role in solving farmer's loan difficult problem, increasing farmer's income and promoting rural economic development. But its unique characteristics make Rural Credit Cooperative face great credit risk during the implementation process of microfinance business. The default phenomenon often occurs and the rate of non-performing loan remains high. Now, it exists a lot of subjectivity when assess the farmer's credit status. This easily leads to credit rating inaccurate and the loans can't recall in time. So this paper takes credit risk in the farmer's microfinance as the research object and uses BP neural network to build a credit risk assessment model, and so as to improve the credit risk management level of Rural Credit Cooperatives, reduce the default risk likelihood of farmer and promote the sustainable development of microfinance.This paper firstly introduces the concept of microfinance and its development in China, analyzes the causes of credit risk on the basis of the characteristics description of Chinese farmer's microfinance. Secondly, introduces the basic theories of artificial neural network, such as neuron model and its transformation function and learning style, at the same time emphatically introduces the BP neural network learning algorithm and its improved algorithm, and summarizes the characteristics of artificial neural network. This makes a theory basis for the following empirical research part. Then makes an empirical research of farmer's microfinance credit risk assessment through on-the-spot investigation data: divides the samples into the training sample group and the test sample group; to screen the strongest discriminative indicator variables between the default farmers and no default farmers, eliminate the multiple linear problem between indicator variables and improve the training speed of BP neural network model, the paper respectively makes a normality test, otherness test and multicollinearity test of indicator variables; then uses BP neural network to build a farmer's credit risk assessment model with MATLAB7.0 software, and compares with the credit risk assessment model established with Logit method. Finally, to make a good popularization and application of the farmer's credit risk assessment model based on BP neural network, and effectively reduce the credit risk of farmer's microfinance, the paper gives some corresponding policy recommendations.This study finally establishes a farmer's credit risk assessment model based on BP neural network of 8-14-1 structure. The model's overall discrimination accuracy of the training samples is 100 percent, the discrimination accuracy of default farmers of the test samples is 90 percent, and its overall discrimination accuracy of test samples is 84.09 percent; the Logit model's overall discrimination accuracy of the training samples is 79.33 percent, the discrimination accuracy of default farmers of the test samples is 75 percent, and its overall discrimination accuracy of test samples is only 63.64 percent. Over all, whether for training samples or test samples, the discrimination accuracy of BP neural network model is higher than Logit model. This confirms that the BP neural network is effectiveness and applicability in the farmer's microfinance credit risk assessment field. It can provide the better reference to the Rural Credit Cooperatives when they identify the credit risk of the farmer.