Determinants of default possibility of individual customers at Co - Operative Bank of Vietnam

tudies such as Bennell et al. (2006) using comprehensive data sets of rating agencies and countries between 1989 and 1999; Finch & Schneider (2007) undertook the study of accuracy classification of artificial neural network (ANN) compared with numerical analysis, logistic regression and classification plants; Pacelli & Azzollini (2011) research on the use of artificial neural networks for credit risk management in Italy; Zang (2011) implemented using ANN neural neural network model to predict the default of customers in commercial banks in China 1.2. Credit issues of the bank 1.2.1. Banking credit concept Bank credit is the relationship of asset transfer between the bank and other economic entities. According to Article 20 of the Law on Credit Institutions, 2010, the regulation: “Credit extension is an agreement by a credit institution to allow customers to use an amount of money with the principle of repayment by lending, discounting operations discount, financial leasing, bank guarantee and other operations ”. 1.2.2. Characteristics of bank credit Firstly, the subject in the credit relationship; Secondly, the subjects of commercial credit transactions include lending with money and leasing real estate or real estate; Thirdly, the capital transfer is based on "belief" and on the principle of unconditional repayment in a certain period of time; Fourth, the repayment value must be greater than the value at the time of lending; Fifthly, banking credit activities pose many risks 6 1.2.3. The role of bank credit Firstly, bank credit ensures the production process takes place on a regular and continuous basis; secondly, this is a strong lever to promote the process of capital accumulation and concentration; thirdly, bank credit helps promote the equilibrium rate of profit among industries in the economy and is an important tool in organizing people's life. 1.2.4. Bank credit classification Classification by time of credit extension; Classification by capital use purpose; Classification by the mode of repayment; Classification by the level of assurance; Classification by subjects of credit extension; Classification by credit origin; Classification by economic sector 1.3 Personal customer credit 1.3.1. Individual customer credit Based on the definition of "bank credit", science and technology credit can be understood as a form of credit in which a commercial bank acts as a transferor of its right to use capital to science and technology for the most used time repayment of both principal and interest (Nguyen Dang Don 2013). 1.3.2. Personal customer credit policy The science and technology credit policy at banks will depend on the objectives and operational policies and there will always be changes to suit the socio-economic conditions as well as ensure the development, sustainable and lucrative for the bank. 1.3.3. Individual customer credit process Contact and guide customers to prepare loan documents; Appraisal of loan conditions; Determine the method of lending, consider the ability of the capital, interest rates; Loan appraisal; Approve the loan; Disbursement; Checking and supervising the loan; Collecting debts, principals and handling arising; Liquidation of credit contracts, loan security contracts, collateral security; Keep credit and loan guarantee records 1.4. Credit risk Credit risk is the inherent potential loss generated when granting credit to customers. Any credit granted must comply with the following three basic principles: (1) The credit must be used for the right purpose and effectively; (2) The credit must be secured by assets; (3) The credit must be repaid both principal and interest within the committed term. 1.5. Effect of credit default In general, the main impact of non-performance loan (NPLs) on banks is that the increase in NPLs limits the financial growth of banks (Karim et al., 2010; Kuo et al., 2010). 1.6. Credit rating activities in banks 1.6.1. Concept The definition of a credit rating is to make a statement about the level of creditworthiness of a financial liability or to assess the level of credit risk depending 7 on factors including the ability to meet financial commitments, the ability to default when economic conditions change, the consciousness and willingness to repay the borrower. The credit rating system is used to assess the financial responsibility of both corporate and S&T customers. Within the scope of this dissertation, the author focuses on analysis and research on credit ratings for science and technology group. 1.6.2. The role of credit rating Banks will control customers' creditworthiness, evaluate the effectiveness of their loan portfolios by monitoring changes in outstanding loans and loan classification of customers thanks to the credit rating system. 1.6.3. Principles of credit rating activities Firstly, credit analysis is based on the awareness and willingness to repay the borrower for each loan. Secondly, assess long-term risk based on the impact of the business cycle as well as the trend and the possibility of future default Thirdly, a comprehensive and unified risk assessment is based on a credit scoring system and rating symbols. Fourthly, the data collection for use in credit rating model should be done objectively and flexibly 1.6.4. Credit rating process Firstly, collect relevant information Secondly, analyze the information collected using models to draw conclusions about the credit rating of science and technology. Third, monitor the credit status of the customer is rated appropriate adjustment 1.6.5. Several credit rating models FICO's individual credit score model CreditKarma credit score model Credit Sesame credit score model VantageScore credit score model Kleimeier's credit score model 1.4.6. Credit rating model at the bank The credit rating system at Vietnamese commercial banks currently uses the grading method. The number of points customers achieve is the total score of financial and non-financial indicators with a certain proportion 1.4.7. A limited number of individual customer credit ratings today The practical model still has limitations that need to be overcome: Firstly, the criteria presented in the current credit rating model are still qualitative, the quantitative factors are still low because based on the method of experience and experts, and there have been no new updates for with quantitative statistical methods. 8 Secondly, the credit scoring results are not a strong basis to help the bank make a decision to grant credit limits to customers. The exact prediction level of customers' defaults is not high, so it can not eliminate the confusion, there are cases when customers are rated at a high, reliable and high level of credit. Low risk but in fact not able to repay. Thirdly, the actual credit rating model has encountered problems that are difficult to detect fraudulent acts of customers. The assessment of these behaviors is only based on the experience of the credit officer in customer contact and information gathering. Fourth, the current scoring model only provides the value of creditworthiness of customers at the time of credit extension, but not predictive for the future 1.7. Factors affecting the default of individual customers 1.7.1 Personal information of the customer The personal information about a customer is the internal information of that customer. The study of these factors helps banks assess the overall overview of the customer, the basic ability of the customer to meet the conditions that the bank requires, the level of reliability in the customer. The fact that customers make commitments with the bank and is also a source of information has a great influence on the decision of the bank about whether or not to provide credit to customers. Factors in this information group include: age, gender, marital status, education level, occupation, current position in the job, judicial record. 1.7.2. Factors of living conditions of customer Information about the living conditions of science and technology reflects the interaction of that customer with the society, thereby helping the bank to assess the impact of the external environment on its financial capacity as well like the perceived behavior of that customer. This group of information includes factors such as: household size, number of dependents, classification of place of residence, characteristics of place of residence, stability of accommodation, home ownership, and ownership of other types of valuable properties. 1.7.3. Financial factors of the customer Analyzing customers’ financial information and financial relationships is an important task for banks, which is crucial to assessing the likelihood of customers' default, affecting credit rating of customer as well as whether the bank makes a loan decision or not. The financial indicators of the customers are the banks 1.7.4. Customer behavior factor Factors in the group are analyzed such as: relationship with the bank, the number and type of banking services that the customer is using, the number of loans, repayment time, time of the loan application procedure, calendar loan history and repayment. 9 CHAPTER 2. RESEARCH METHODS 2.1. Research process With the content of the dissertation, the research process is presented as follows: 2.2. Research models and hypotheses From previous studies, the author offers the following research model: In which: DLi: the dependent variable on default Xi: Independent variables can affect the default DLi Define research objectives Theorical Research model Data analysis Complete the report The research gap Relevant theories, factors affecting the default of individual customers The research variables obtained from the previous model, factors from qualitative interviews were included in the research model - Logit model - ANN, - Random forest - Check results on new data samples - So sánh các mô hình 10 2.3. Design 2.3.1. Sample Research data was collected from the database of Co-operative Bank of Vietnam Vietnam. Historical data about whether or not individual defaults of individual customers at the bank will be used. The sample collected 5498 customers at Co-operative Bank of Vietnam. With a sample size of 5498, we can guarantee the reliability of the minimum number of samples when analyzing multivariate data 2.3.2. Data collection With the research variables, the author proceeds to send to the bank branch in charge to ask for information on the situation of debt repayment of individual customers. Each bank is collected by NCS from about 100-500 customers. In particular, the number of customers who are able to pay the debt and default (not to pay the debt) is also collected in a balanced manner by the PhD student to analyze without bias in the sample. 2.4. Data analysis method 2.4.1. Data description 2.4.2. Correlations 2.4.3. Models of analysis and prediction of default of science and technology 2.4.3.1. Logit model 2.4.3.2. Probit model 2.4.3.3. Discriminant analysis 2.4.3.4. Predictive model of artificial neuron network (ANN) 2.4.3.4. Forecasting model by Random Forest 11 CHAPTER 3. RESULTS 3.1. Co-operative Bank of Vietnam 3.1.1. Introduction to Co-operative Bank of Vietnam The Co-operative Bank of Vietnam, formerly known as the Central People's Credit Fund, was established on August 5, 1995 and converted into a Co-operative Bank of Vietnam in accordance with License No.166/GP-NHNN dated June 4, 1995/2013 of the Governor of the State Bank of Vietnam. The full name in Vietnamese: Co-operative Bank of Vietnam of Vietnam 3.1.2. Capital use activities of Co-operative Bank of Vietnams Capital of Co-operative Bank of Vietnams from 2016 to 2017 increased by 2828 billion dong, equivalent to 10,49%. In which, equity increased by 63 billion dong, equivalent to 1,76%; The equity is mainly raised from funds but not from charter capital 3.2. Situation on individuals who borrow money at Co-operative Bank of Vietnams according to the research sample Descriptive statistics of the study variables continuously indicate that the average education of subjects is 21 years of schooling. The largest one is 36 years and the smallest is 12 years. The standard deviation of 6,1 indicates a relatively large level of educational inequality. Next, the age of science and technology loans borrowed on average is 32 years old. The largest one is 53 years old and the youngest is 20 years old. In terms of household size, the average number of borrowers in the family is 5, the largest is 6 and the smallest is 4. The average number of dependents in the family is 3, the largest is 5 and the smallest is 1 person. The average loan amount is 551 million, the smallest is 100 million and the largest is 10 billion. Average working time and experience of science and technology are 14 years, the minimum is 1 year of experience and the maximum is 30 years of working experience. The ratio of principal and interest payment to the average income is 0,397, equivalent to 39%, of which the largest is 60% and the smallest is 20% 3.3. The results of analyzing the factors affecting the default of individual customers 3.3.1. Logistic regression results Table. Logistic regression results for customers Estimate Std. Error z value p-value (Intercept) 14,2000 0,8484 16,7370 < 2e-16 *** Education_ Intermediate 0,2253 0,1951 1,1550 0,248272 College/University -0,6597 0,1915 -0,8339 0,71345 12 Estimate Std. Error z value p-value Graduate university -0,0898 0,1369 -0,6550 0,512187 Gender 0,8570 0,1421 6,0300 1,64e-09 *** Marriage -0,6551 0,1146 -5,7160 1,09e-08 *** Judicial Records 0,6929 0,2229 3,1090 0,001879 ** Business property -4,9570 0,1703 -29,1140 < 2e-16 *** Age -0,0729 0,0089 -8,1810 2,82e-16 *** Household size -0,2300 0,0966 -2,3800 0,017301 ** Number of dependents 0,0528 0,0707 0,7470 0,455127 Loan 0,0000 0,0003 0,0280 0,977847 Job 1,7440 1,0010 1,7420 0,081507 Location _TruongBoPhan -0,0234 0,2071 -0,1130 0,910178 Nhanvien 0,5550 0,3011 1,8430 0,065263* Work time -0,0226 0,0105 -2,1470 0,031771 ** Type of work -0,5450 0,1414 -3,8540 0,000116 *** Income -0,1218 0,0107 -11,3770 < 2e-16 *** Term_Mid-term -1,3230 0,2322 -5,6960 1,23e-08 *** Long-term -0,8374 0,2398 -3,4920 0,000480 *** Status_ Slow 1 time 1 -0,0890 0,1550 -0,5740 0,565849 Slow 2 times or more -1,4100 0,2621 -5,3780 7,54e-08 *** Purpose_Use the right purpose 0,2248 0,1134 1,9830 0,047378 ** Diversify the profession 0,5236 0,1137 4,6050 4,12e-06 *** Special properties are real estate -1,7930 0,1247 -14,3750 < 2e-16 *** Monthly payment rate 0,4287 0,4787 0,8960 0,370498 Life insurance -1,0980 0,2286 -4,8050 1,55e-06 *** Observations 5,498 Note:*p<0,1;**p<0,05;***p<0,01. 13 Table. Forecast results for test sample Forecast results for 500 sample test Predict No-default Default No-default 181 27 Default 50 242 Accurate forecasts 83,46% 95% CI (81,13%; 87,65%) 3.3.2. Probit model results Probit model results show that the results are quite similar to the results of Logistic model. Table. Forecast results of Probit model Estimate Std. Error z value p-value (Intercept) 7,748 0,453 17,108 < 2e-16 *** Education_ Intermediate 0,131 0,107 1,223 0,221288 College/University -0,369 0,104 -1,213 0,20230 Graduate university -0,037 0,076 -0,481 0,63044 Gender 0,475 0,079 6,026 1,68e-09 *** Marriage -0,354 0,062 -5,666 1,46e-08 *** Judicial Records 0,374 0,116 3,212 0,001316 ** Business property -2,771 0,088 -31,491 < 2e-16 *** Age -0,037 0,005 -7,565 3,88e-14 *** Household size -0,115 0,053 -2,157 0,031033 * Number of dependents 0,022 0,039 0,582 0,560707 Loan 0,000 0,000 -0,294 0,768435 Job 0,998 0,516 1,932 0,053307 , Location _TruongBoPhan 0,007 0,113 0,064 0,949206 Nhanvien 0,309 0,164 1,884 0,059518* Work time -0,017 0,006 -2,895 0,003794 ** Type of work -0,278 0,078 -3,571 0,000355 *** Income -0,068 0,006 -11,781 < 2e-16 *** 14 Estimate Std. Error z value p-value Term_Mid-term -0,728 0,127 -5,740 9,46e-09 *** Long-term -0,479 0,131 -3,655 0,000257 *** Status_ Slow 1 time 1 -0,005 0,083 -0,061 0,950985 Slow 2 times or more -0,750 0,144 -5,208 1,91e-07 *** Purpose_Use the right purpose 0,103 0,062 1,653 0,098401 , Diversify the profession 0,307 0,063 4,917 8,80e-07 *** Special properties are real estate -0,954 0,068 -14,083 < 2e-16 *** Monthly payment rate 0,228 0,262 0,868 0,385316 Life insurance -0,583 0,123 -4,722 2,34e-06 *** Log Likelihood -131,042 Akaike Inf. Crit. 457,41 Note *p<0,1; **p<0,05; ***p<0,01 At the same time, the Probit model is capable of accurately forecasting for 500 tested customer samples which is 84,6% Bảng . Kết quả dự báo của mô hình Probit Predict No-default Default No-default 183 29 Default 48 240 Accurate forecasts 84,6% 95% CI (81,13%;78,5%) 3.3.3. Forecasting results based on artificial neural network At the same time, the forecast results on the accuracy of the model indicate that the ANN model is capable of accurately forecasting on the sample of 500 customers from the weight of ANN model is 83,86%. Table. ANN forecast results Predict No-default Default No-default 184 47 Default 34 235 Accurate forecasts 83,86% 15 3.3.4. Results of the Random Forest classification model From the forecast results on sample data 5.498, the PhD student conducted a random forest forecasting model on 500 test samples to obtain the following forecast results: Table . The forecast of the Random Forest model Predict No-default Default No-default 217 0 Default 14 269 Accurate forecasts 97,2% 95% CI (95,35%;98,46%) 3.3.5. Synthesize the results 3.3.6 Discussion With the estimated results from 4 models: Logisitic regression, Probit regression, ANN artificial intelligence network model and random forest model, indicating the ability of accurate prediction of the models are on 82%. Besides, through regression models Logistic and Probit showed the influence of each factor on the default of customers. To analyze the factors affecting customers' ability to pay debts, the PhD student conducted analysis based on the Logistic model (the model with good forecasting ability is equal to Probit but has more meaningful research variables). With the results of this analysis, the PhD offers a number of analyzes and discusses the research results as follows: The gender of the borrower also affects the likelihood of default. In particular, science and technology are male and more likely to default. It can be seen that male individuals often have more time for work than women (Because women have more daily tasks for families). Factors in marital status also have the opposite effect on the ability to default. With encryption as 1- married and 0- unmarried, this result indicates that married customers are more likely to pay off debt than unmarried people. The problem is that when getting married, financial resources can be mobilized from many sources: Spouse, parent or family member of two spouses (Ojiako & Ogbukwa, 2012) The better the criminal record, the lower the likelihood of defaulting, compared to customers with a bad criminal record. The factor of business ownership has an opposite effect on the ability of customers to default. This result indicates that customers who own a business (not renting) are less likely to default than customers who have to rent a business. 16 The working position of the customer who is an employee tends to default more with the customer as a manager. It can be seen that employees working and taking business loans bring high risks to banks The age factor has an opposite effect on the ability of individual customers to default when borrowing money from banks. Household size has a negative effect on customers' ability to default, indicating that the more members a household has, the more likely it is that it can repay debt. The income factor has the opposite effect on the ability to default, showing that the higher the income customers have, the higher the ability to repay debt to people with lower incomes. The type of business being employed (as an employee of the company or as a business owner), the results show that customers in the state sector are more likely to default on debt than those in the sector outside the state. Factors of work experience also negatively impact on the ability of individual customers to default on the banks. Factors about the loan term also affect the ability of customers to default. The logistic regression results show that the loans in the medium and long term tend to face a lower risk of default than short-term loans. Results of the effect of repayment of principal and interest on time also indicate that customers with a history of deferred payment of 2 or more times are more likely to default than customers who have not made any late payments or have only deferred payment 1 time Career diversification has a positive impact on the default of individual customers at Co-operative Bank of Vietnam Collateral has an opposite effect on the ability of individual customers to default. Results showed that customers with real estate collaterals were less likely to meet with their wives than customers with real estate collaterals. Factors involved in life insurance have a negative effect on customers' ability to default. This result shows that customers who participate in life insurance tend to have lower defaults than those who have not participated in insurance 3.3.7 Compare the level of accurate prediction of estimated models With the estimated results from 4 models: Logistic regression model, Probit model; ANN model and Random Forest model. NCS compares the forecast ability as well as ranking the forecast ability of the models. The results show that Logit, Probit and ANN models have the lowest predictability of the 4 estimation models with 83% accuracy forecast. The Random forest model is almost accurate at 97,2% prediction. It can be seen that with predictive analytical techniques of classification nature such as Random forest have better predictability than traditional estimation models such as Logistic or Probit. 17 Figure. Compare the forecast level of models From the results of the estimation models, the author also suggested that banks should refer to all 4 forecasting models capable of forecasting over 83%: Logit model of Probit model, artificial intelligence network and Random Forest. . In particular, ANN and Random forest model will help predict debt repayment quickly. Logit and Probit regression models will help banks build indicators to credit ratings and collect important customer-related data. 3.4 Interview with experts about the causes of credit risk In order to improve individual credit activities in banks, student conducted interviews with experts on issues related to limiting credit risks in banks to reduce customers' default. The interview results gave some main reasons for the customers' credit risks or default: Causes from customers: (1) Using capital for improper purposes; (2) Investment activities with low efficiency; (3) Fraudulent customers in applying for a loan; (4) Interaction with limited banks; (5) Diversify the portfolio from inefficient loans. Reasons from the bank: (1) Credit risk due to lack of customer information; (2) Due to the subjective will of the approver / lender; (3) Due to pressure to fulfill assigned annual targets, not really paying attention to credit quality; (4) Lack of supervision and management after lending, early warning system of loans with problematic issues, thus unable to intervene in time; (5) The credit scoring system is not appropriate; (6) Lending control system is not tight. 18 CHAPTER 4. CONCLUSIONS AND RECOMMENDATIONS 4.1. Solutions to improve the efficiency of science and technology credit activities at Co-operative Bank of Vietnams 4.1.1 Solutions to classify individual customers in the process of file preparation In order to manage credit risk of individual customers, based on the results of this study, the author offers a number of solutions to limit the cases of non- repayment: Initially customer data information needs to be captured in detail and accuracy. In other words, the b