Các thuộc tính tài chính gây ra nợ xấu tín dụng của khách hàng cá nhân tại ngân hàng thương mại Việt Nam

Nội dung text: Các thuộc tính tài chính gây ra nợ xấu tín dụng của khách hàng cá nhân tại ngân hàng thương mại Việt Nam

1. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 DETERMINANTS OF BAD DEBT IN PERSONAL CREDIT SEGMENT: EVIDENCE FROM VIETNAMESE COMMERCIAL BANKS CÁC THUỘC TÍNH TÀI CHÍNH GÂY RA NỢ XẤU TÍN DỤNG CỦA KHÁCH HÀNG CÁ NHÂN TẠI NGÂN HÀNG THƯƠNG MẠI VIỆT NAM Nguyen Tien Nhat, Le Ngoc Quynh Anh University of Economics, Hue University ntnhat@hce.edu.vn ABSTRACT This study applies the Decision Tree model to analyze the financial data of 500 individual customers who are granted by commercial banks in Vietnam. Based on the study of credit-scoring models for individual customers and previous relevant-research, the authors select the financial attributes of individual customers that are likely to cause the bad debt to put into the initial model of Decision Tree, including bad-debt group, loan frequency, loan purpose, loan term (month), overdue-loan frequency, collateral, Loan/Collateral ratio, Debt/Total Asset ratio, total income, Reserve Expense/Debt, credit policy. The result of decision tree illustrates eight attributes leading to the bad debt, specifically loan frequency, loan purpose, loan term, overdue-loan frequency, Loan/Collateral ratio, Debt/ Total Asset ratio, Reserve Expense/Debt ratio, and credit policy. According to the research result, this paper proposes a number of solutions to prevent the bad debt of individual customers during the process of credit appraisal at commercial banks in Vietnam. Keywords: Financial attributes, personal credit, bad debt, non-performing loans, commercial banks, Vietnam, Decision Tree model. TÓM TẮT Nghiên cứu này sử dụng mô hình Cây quyết định để xử lý dữ liệu tài chính của 500 khách hàng cá nhân đang vay nợ tại một số ngân hàng thương mại Việt Nam. Dựa trên việc nghiên cứu các mô hình chấm điểm tín dụng khách hàng cá nhân và một số tiền nghiên cứu liên quan, tác giả lựa chọn các thuộc tính tài chính của khách hàng cá nhân có khả năng gây ra nợ xấu tín dụng để đưa vào mô hình, bao gồm: nhóm nợ xấu, tần suất vay, mục đích vay, kỳ hạn vay (tháng), lịch sử tín dụng của khách hàng, số lượng và giá trị tài sản đảm bảo, tỷ lệ khoản vay/tài sản đảm bảo, tỷ lệ nợ phải trả/tổng tài sản cá nhân, tổng thu nhập, tỷ lệ chi phí dự phòng/nợ phải trả, chính sách tín dụng. Kết quả nghiên cứu chỉ ra 8 thuộc tính dẫn đến nợ xấu, bao gồm: tần suất vay, mục đích vay, kỳ hạn (tháng), lịch sử tín dụng của khách hàng, tỷ lệ khoản vay/tài sản đảm bảo, tỷ lệ nợ phải trả/tổng tài sản cá nhân, tỷ lệ chi phí dự phòng/nợ phải trả, chính sách tín dụng. Dựa trên kết quả nghiên cứu, tác giả đề xuất một số giải pháp nhằm ngăn ngừa nợ xấu tín dụng của khách hàng cá nhân trong quá trình thẩm định hồ sơ tín dụng tại các ngân hàng thương mại Việt Nam. Từ khóa: Thuộc tính tài chính, tín dụng cá nhân, nợ xấu, ngân hàng thương mại, Việt Nam, mô hình Cây quyết định. 1. Introduction Since 1986 Vietnam's economy has developed impressively accompanied by enormous capital demands mainly focusing on fields of infrastructural construction, manufacture and service business. In order to meet such growing demands, Vietnamese banking system with two main activities, specifically mobilization and lending, has efficiently perform to provide sufficient capitals for the economy and create a considerable momentum for the economic development in the long term. Business activities have become stagnant, thus causing varied financial risks to the banking system. One of major risks that commercial banks pay much attention to is the credit risk, in which the most serious situation is as the rate of non-performing loans (NPLs) increases higher and higher to a point of impossible solution. The 98
2. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 bad debt not only block the capital flows into the economy, but also damage the banking efficiency of commercial banks. Therefore, in 2013 the Vietnam Asset Management Company (VAMC) was established to buy and sell bad debt of all commercial banks in Vietnamese banking system. However, up to now VAMC has not been able to fulfill its mission because of an insufficient authorization of legal framework, meaning that it has not allowed commercial banks as well as debt-trading companies to handle assets of bank-borrowing enterprises. Hue, N. T. M. (2015) claims that VAMC could not find a real solution to the problem of high NPLs ratio in Vietnam, actually it just reformed the NPL, leading to an increase in the bad-debt rate. The nature of a bank's bad debt is defined as a wrong or ineffective spend of the borrowers who are granted by banks. When a large number of borrowers faces the financial dilemma, the amount of bad debt of the whole banking system would increase dramatically, causing serious consequences for the banks themselves as well as the domestic economy. In Vietnam, the ratio of bad debt has tended to increase since the end of 2007 and became more serious since the end of 2011. According to the annual reports of financial institutions, in 2012 the banking system's bad debt is 117,723 billion VND, accounting for 4.47%. However, the State Bank of Vietnam announced that the ratio of bad debt was 8.82% in 2012, far exceeding the figures reported by commercial banks. In addition, according to Fitch Ratings, Vietnam's bad-debt ratio in 2012 was 13% of total outstanding loans. As comprehensively reassessing the figure of bad debt of banking system in 2015, the SBV issued that the ratio of bad debt was twice as high as that in 2012, at 17.21%. In an attempt to quickly and completely eliminate bad debts of financial institutions within five years (from 2016 to 2020), the SBV determined that the completion of the legal framework on restructuring weak banks would play a central role of this period. Accordingly, the SBV formulated and submitted the Resolution No.42/ 2017/ QH14 to the National Assembly for approval in 2017, with aiming to launch the pilot project of dealing with bad debts of credit institutions. The Resolution No.42 took effect in the period of 2012-2017. As a result, over the 10-month period from mid-August 2017 to June 2018, the whole system of credit institution has handled nearly 140 trillion VND of bad debt. Handling bad debt and minimizing the credit risk of are top priorities of Vietnamese banking system. However, in order to avoid bad debt, especially in terms of bank credit for the individual customers, commercial banks must identify the financial attributes of customers that are likely to cause bad debt, and build appropriate principles of the credit policy based on their analyses. Over the recent years, Vietnamese commercial banks have applied various business strategies and policies to expand the credit operation and improve the credit quality, with aiming to increase the capital supply for the capital market in Vietnam. However, in the process of credit growth, especially for individual customers, commercial banks have encountered several challenges and risks that lead to a sharp increase in bad debt related to the credit segment of individual customer which in turn affects the banking efficiency of these banks. Therefore, it is imperative for Vietnamese commercial banks to pay much attention to the financial attributes of individual customers that are considered to be major factors causing bad debt in order to formulate an appropriate policy of preventing the credit risk. The author finds this issue to be a research gap because there are not much thorough research covering it. Determining the financial attributes of individual customers causing bad debt not only to help commercial banks find the appropriate solution, but also to fill the research gap in terms of the credit risk and bad debt issue. This paper aims to study the theoretical basis regarding the credit segment of individual customer, bad debt, as well as the financial attributes of individual customer that can be used to predict the probability of causing bad debt. The authors then apply the Decision Tree model to identify these individual attributes and clarify how they are likely to trigger bad debts. Finally, based on the research results, this study proposes some possible solutions to help commercial banks minimize the credit risk of individual customer. 99
3. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 2. Literature Review 2.1. Definition of personal credit The credit is defined as an economic category that reflects the transaction relationship between two (or more) entities. In which one side transfers an amount of money to the other for a specified period of time, following a principle that the receivers must pledge to return it on (or ahead of) time shown in a credit agreement. Accordingly, the banking credit is defined as a credit relationship between a certain bank and another party in the economy, in which the bank acts as both a borrower and lender. In other words, the bank can be seen as a financial intermediary that connect capital flows from the side having temporary surplus capital to the another being in capital shortage. The price (interest rate) of a loan set by the bank is the amount of interest that the customer obliges to pay throughout the loan cycle. Personal credit is individual‟s debt taken on to purchase goods and services (www.investopedia.com). 2.2. Definition of non-performing loan The bad debt (or doubtful debt) are commonly considered to be overdue loans, meaning that the borrowers cannot afford to repay or recover their loans. This phenomenon normally happens when the debtors declare themselves to be bankrupt or to have dispersed properties. Furthermore, they are accompanied by overdue interest and/ or principal repayments categorized into different groups based on the length of time out of date. However, the categories of bad debt are fairly broad and varied in opinion on it, thus each country or each economy has a different perspective about the bad debt. A nonperforming loan (NPL) is a certain loan upon which the borrower has not made the scheduled payments for a specified period (www.investopedia.com). According to European Central Bank (ECB), the bad debt is defined as an irrecoverable loan, such as expired debts or unwarranted debts of the debtors who might be missing (or flee). They might also terminate or liquidate their businesses that certainly suffer losses; thus, the remaining assets are not enough to pay the debts. In these cases, the banks normally cannot contact or find the debtors. The bad debt might also be unrecovered loans, with having no collaterals or the assets offered as collaterals are not sufficient value to cover the debts, meaning that the banks cannot fully recover the debt. There are various circumstances that might lead to the bad debt eventually as follows: - The debtors agreed to pay the debts in the past but the rest cannot be compensated. Or the collaterals are transferred for regularly payments of interest and principal but the remaining value are not sufficient to cover the whole debt. - The debtors are troubled to repay the debts and require the banks to reschedule the payment plan. However, eventually they cannot afford to compensate for their debts within the agreed time. - The debtors‟ mortgaged properties are insufficient in value to pay off the debts or the mortgaged assets, thus the debtors are unable to fully repay the debts. - The debtors declare themselves to be bankrupt, with having the payoff be less than the outstanding balance. According to the International Monetary Fund (IMF), the bad debt is defined as an unprofitable loan (bad debt) accompanied by the interest and/or principal payments that are overdue for more than 90 days, or accepted to be restructured/rescheduled after 90 days. It might also be the case that the interest and/or principal payments within are not overdue but there are clear reasons to doubt the debt payment will be fully implemented. According to the State Bank of Vietnam (SBV), based on the Decision No. 493/2005 of the Governor of the SBV about how to classify banking debt. The NPL includes three groups, specifically Substandard debt (group 3), Doubtful debt (group 4), and Irrecoverable debt (group 5). They are also determined based on two factors: (i): overdue for more than 90 days or (ii): unsettling repayment capacity. 100
4. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 Based on these viewpoints mentioned above, this paper considers the bad debt to be a loan that might cause the debtors to be unable to stay solvent. In other words, even though a loan is still on due or just lent, but there are clear signals regarding the financial ability of individual customer that might help the banks affirm the repayment ability of customer to be suspicious, so it can also be considered as a possible bad debt. All in all, this research follows the definition and categorization of the State Bank of Vietnam about the NPLs that are applied to Vietnamese banking system. 2.3. Attributes of causing the bad debt in the segment of personal credit Antwi et al. (2012) studied the debt repayment capacity of customers using the data of 800 observations from 2006 to 2010. The study concluded that there were two variables, specifically forms of borrowing and assets, that actually affect the borrower's ability to repay. Therefore, the banks should focus on the borrower's ability to secure the loan with assets and improve the risk of default of the borrower. In addition, Pasha, S. A. M., & Negese, T. (2014) researched the microfinance related to the provision of small loans, savings, and other services to the poor, excluding commercial bank collateral and other reasons. The authors collected data from credit institutions and analyzed using the Logistic model. The study shows that there are 14 determinants of loan repayment performance, nine of them are statistically significant. The results suggest that the level of education of customers would help them to use capital more effectively. Moreover, their labor-age and good business experience will enable them to better repay the debts. Nawai, N., & Shariff, M. N. M. (2012) investigated the factors that affect the ability of borrowers using the research sample of 309 customers participating in micro-credit programs in Malaysia from 2010 to 2011. In which, the status of loan repayment is categorized into three specific types: punctual repayment, overdue repayment and non-performing loan. The study used Logit regression with 12 independent variables, for example gender, age, education level, education/religion, total income, loan term, monthly sales, and legal business registration affecting customers' ability to pay debts. The results show that age, education/ religion, monthly sales, and legal business registration reduce the ability to repay loans, while gender, total debt, and the repayment of a loan according to customer needs have a positive effect on the customer's ability to repay loan. This study selects the attributes regarding the financial information of individual customers who are on loan from commercial banks based on various systems of credit criteria which Vietnamese banks apply to assess customers‟ financial capacity and previous studies in terms of the personal credit and bad debt. Although the factors causing the bad debt of personal credit are quite different for each commercial bank or research period, they converge into some fundamental attributes. According to the results of previous studies, this paper chooses four groups of attributes that might lead to the bad debt in the segment of personal credit as follows: - The financial capability of customer. - The overdue-loan frequency of customer. - The loan characteristics. - The credit policy of bank. The financial capability of customer The attributes of financial capability are used to determine whether the debtors can pay the interest and principal on time, including the total income, personalized collateral (in value and quantity). (Owen, 2006) indicates that the total income probably affects the credit risk of loan portfolio in a positive way. In great detail, the credit risk would be reduced if the total income of customer is high, meaning that the customer has more capability to pay the interest and principal on time regardless of the cost of living and other expense. To sum up, the higher the income of customer, the smaller the probability of causing the bad debt. 101
5. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 Similarly, the quantity and quality of collaterals owned by the customers have a positive impact on their credit conditions. (Owen, 2006) also claims that the larger the value of collateral, the less likely the banks face the credit risk. The credit history of customer The detailed information on the credit history reflects the number of loans, the frequency of borrowing, the description of overdue loans. (Li, 2012) concludes that they are considered to be significant to assess the creditworthiness of a certain customer in the credit relationship with (various) commercial banks. In other words, as the customers enhance their own reliability and credit worthiness in the credit relationship with credit situations, they will be lent by the banks many times, showing in an increase of loan frequency. However, in reality, the burden of repayment and other expense might also rise when the customers are accepted to get as many as they need, leading to the surge of the credit risk that directly causes the bad debt (Xiao, 2006). The loan characteristics This group of attributes presents the characteristics of the loan, including the loan limit, the purpose of loan, the loan term (month), the rate of loan/collaterals rate, the rate of loan/collateral. There are different opinions about the effect of the loan limit (for a certain customer) to the bad debt in the field of personal credit. (Steenackers, 1989) indicates that the larger the allowable limit of loan, the higher the profit that customers can generate. The smaller limit of loan is likely to cause the bad debt because the customers do not have enough capital to yield expected profits on their businesses so as to cover the financial expense. In contrast, the large loans are considered to be less risky because they are commonly strictly monitored by the banks, thus reducing the probability of causing the bad debt. Regarding the loan term, (Steenackers, 1989) demonstrates that the banks find more difficult to manage the credit risk of long-term loans. As the customers consider their credit risks to be low, they will prefer short-term loans to long-term loans so as to reduce the financial expense. As determined the loan purpose, the loans can be served the business or living purposes. If the customers spend the loans on their own consumptions that often do not generate the reciprocal income, the credit risk of these loans might be higher than that of business-purpose loans. It is clear that this category of loan enables the customers to gain expected profits so that they can pay the interest and principal on time, thus eliminating the possibility of causing the bad debt. According to the traditional view, there is a positive relationship between the credit risk and the collateral, implying that if the repayment capacity of customer is judged to be weak, the bank would require larger collateral in value (or more collaterals in quantity). (Vo, 2015) indicates that the ratio of loan to collateral has a positive effect to the rate of bad debt in the segment of personal credit. The credit policy of bank This group of attributes introduces the credit policies that the banks apply to the customers of personal credit, including the rate of reserve expense/repayment, the credit rank of customer. They can provide the detailed information on the budget of risk reserve and favorable policies applied to the individual customer. These rates also indicate the risk appetite or the amount of bad debt in total that banks can accept. (Capon, 1982) concludes that the credit policy helps banks not only monitor the effectiveness of their business activities, but also identify the goodwill of customer in repayment and other factors that might affect customer's ability to repay debts, thus reducing the probability of causing the bad debt. 3. Methodology and data 3.1. Decision Tree model This research uses the Decision Tree (DT) model because it helps to identify key attributes that affect the probability of repaying individual customers‟ debts on time. The DT model makes explicitly all 102
6. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 possible alternatives and traces each alternative to its conclusion in a single view, allowing for easy comparison among the various alternatives. It also allows for a comprehensive analysis of the consequences of each possible decision, such as what the decision leads to, whether it ends in uncertainty or a definite conclusion, or whether it leads to new issues for which the process needs repetition. Moreover, the DT approach is not tied by any theoretical assumption. It is a series of probabilistic algorithms in which the probability of an event occurs in different "if-then" scenarios is calculated and showed in the form of information trees. Finally, none of the previous studies in this field used the DT model in the case of Vietnam. Technically, a tree is either a leaf node labelled with a class, or a structure containing a test, linked to two or more nodes (or subtrees). Each subtree from start to finish reflects a rule of an event and goes with a separate event, mutually exclusive, meaning that if this event occurs when the other event will not. In fact, a basic decision tree might have several internal (non-leaf) nodes denoting tests on the attributes or input features, branches representing the outcome of tests, and leaves (or terminal nodes) holding class labels. The probability algorithm can help to determine the major nodes and leaves by eliminating the disturbance properties or elements. A tree which can be considered a good outcome must include the attributes with maximum information gain ratio - the probabilities of each case with a particular value for the attribute being of a particular class. The DT model is supported by a software named Weka- an open source software issued by the University of Waikato, thus this study uses Weka J48 which is an open Java implementation of the C4.5 algorithm, based on Ross Quinlan‟s ID3 algorithm. It is ranked among the top 10 algorithms in Data Mining (Wu, et al., 2008). The experts in the field of machine learning consider C4.5 algorithm a powerful learning algorithm because it is able to form a mapping from indicator values to classes, dedicated to classifying unobserved variables. Korting (2006) showed a brief description about the C4.5 algorithm, used to create Univariate Decision Trees and Multivariate Decision Trees, as following: The premises on which this algorithm is based are: - If all cases are of the same class, the tree is a leaf and so the leaf is returned labelled with this class. - For each attribute, calculate the potential information provided by a test on the attribute (based on the probabilities of each case having a particular value for the attribute). Also calculate the gain in information that would result from a test on the attribute (based on the probabilities of each case with a particular value for the attribute being of a particular class). - Depending on the current selection criterion, find the best attribute to branch on. Counting gain: This process uses the “Entropy”, i.e. a measure of the disorder of the data. The Entropy of is calculated by Iterating over all possible values of The conditional Entropy is: and finally, we define Gain by: 103
7. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 The aim is to maximize the Gain, dividing by over-all entropy due to split argument by value . Pruning: This is an important step to the result because of the outliers. All data sets contain a little subset of instances that are not well-defined, and differs from the other ones on its neighborhood. After the complete creation of the tree, that must classify all the instances in the training set, it is pruned. This is to reduce classification errors, caused by specialization in the training set; this is done to make the tree more general. For a Multivariate Decision tree, we can represent the multivariate tests with a Linear Machine (LM). Considering a multiclass instance set, we can represent the multivariate tests with a LM. LM: Let be an instance description consisting of 1 and the n features that describe the instance, . Then each discriminant function has the form: where is a vector of n+1 coefficients. The LM infers instance to belong to class if: The weights vector is assigned a default value for all One approach for updating the weight of the discriminant functions is the absolute error correction rule, which adjusts where is the class to which the instance belongs, and where is the class to which the LM incorrectly assigns the instance. The correction is accomplished by and where is the smallest integer such that the updated LM will classify the instance in a correct way. Thermal Perceptron: For not linearly separable instances, one method is the “thermal perceptron”, that also adjusts and , and deals with some constants The basic idea of this algorithm is to correct the weights-vector until all instances become correct, or in the worst case, a certain number of iterations is reached (represented by the actualization of value, decreasing according to the equation , as and is also a linear small decreasing of the value For this research, the key instrument for collecting data is the questionnaire. A valid questionnaire must be designed in accordance with the contextual adaptation and relevant information. For this 104
8. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 issue, the decision tree approach offers an advantage over other models such as exploratory factor analysis or structural equation modelling. Designing a questionnaire for an EFA or SEM is bounded by multiple technical procedures and criteria. For example, their measurements are commonly tightened with a Likert scale which is uni-dimensional and only gives 5-7 options of choice, and the space between each choice cannot possibly be equidistant. Therefore, it might fail to measure the true attitudes of respondents in some cases. In contrast, the tree approach is free with the aforementioned elements. The tree approach could deal with a set of attributes with different measurement scales. Indeed, the tree approach gives greater flexibility to the researcher in establishing a questionnaire which collects as much as possible of the needed information and in establishing the whole set of relations between a different kind of information. Despite these advantages, there are still some recommended practices. - The attributes in a decision tree approach may be nominal categorical, or continuous. If it is a continuous-valued attribute, for example “bus waiting time”, it must be discretized prior to attribute selection (Fayyad & Irani, 1992). It is possible to use simple clustering algorithms for one-way data to divide the values into small clusters, then a continuous-valued attribute is converted into categorical properties. The discretization algorithm is automatically available with ID3, C.4.5 and Fuzzy ID3 algorithm. - If the attribute is assigned to many values, it might result in a tree with several possible outcomes, meaning that it becomes less certain and less representative. Therefore, when an attribute has a finite list of possible values, it is necessary to screen carefully, select only the typical values for each attribute. These technical notes mentioned above are useful for designing a valid questionnaire for our decision tree analysist. 3.2. Data This study was conducted in Hue City nested in the middle part of Vietnam. Given an access to the database of personal credit by various branches of commercial banks in Hue City, the research amassed a comprehensive data that meet the requirements of the Decision Tree approach, containing the information of loan commitment of borrowers who got loans in the period of 2017-18. The research data includes the financial information of 500 individual customers being on loan from five joint-stock commercial banks in Vietnam, specifically Vietin, BIDV, ACB, Sacom and VIB. This sample is composed of customers who repay debts on time and those whose debts are reported being overdue. The authors apply the Decision Tree model including 11 attributes- that is to say, bad debt group, loan frequency, loan purpose, loan term, overdue-loan frequency, customer‟s collateral (in quantity and value), the rate of loan/collateral, the rate of liability/total asset, customer‟s income, the rate of reserve expense/repayment, to assess the probability of causing the bad debt in the segment of personal credit. These attributes belong to the string type which represents a dynamically expanding set of nominal values, usually used in text classification as shown in Table 1. Table 1: Description of attributes used in Decision Tree model Attribute Meaning Characteristics Type 1 The groups of bad debt according to the standard of Bad-debt Group Type 2-3 State Bank of Vietnam Type 4-5 Quantity Collateral The collateral of individual customer Value Normal The credit policies are being applied to the segment of Under control Credit policy personal credit during research period Under control & Awareness 105
9. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 Debt/Total Asset ratio The ratio of debt to customer‟s total assets One-time How regular the customers receive personal loans Loan Frequency Average from the banks Frequent Business Loan Purpose The purposes of personal loans Consumption Household Spend Short Loan Term The different periods of personal loans (month) Medium Long Loan/Collateral ratio The ratio of loan to collateral Rare Overdue-loan Whether the customer have overdue loan during the Frequent Frequency credit history in banks‟ database Punctual Reserve The ratio of reserve expense to debt Expense/Debt ratio Low Total income The total income of individual customer per month Average High Source: Author’s description 4. Results and discussion 4.1. Results The decision tree is a tool of decision making that uses a tree-like graph or model of possible consequences and displays an algorithm containing conditional control statements. The common illustration of a certain decision tree is a flowchart-like structure in which each internal node represents a “test” on an attribute, each branch represents the outcome of the test, and each leaf node represents a class label (outcome given after computing all attributes). The paths from root to leaf represent classification rules. Unlike linear models, tree-based learning algorithms map non-linear relationships quite well. They are adaptable at solving any kind of problem at hand (classification or regression). Decision Tree algorithms are referred to as CART (Classification and Regression Trees). They have a natural “if then else ” construction that makes it fit easily into a programmatic structure, and would be well suited to categorization problems where attributes or features are systematically checked to determine a final category. The decision tree of this research is categorized into Categorical Variable Decision Tree which has categorical target variable. For example, in the scenario of the likelihood of bad debt, the target variable is “The categorization of bad debt “i.e. type 1 or 2-3 or 4-5. In order to show the paths from root to leaf which represent how different types of bad debt are likely to be incurred and identify key attributes of a certain individual customer that might cause the bad debt during the loan period, this study uses the Weka software to exploit the data including the financial information of 500 individual customers that are reported getting into the bad debt in their credit history in banks‟ database. The likelihood of causing the bad debt is expressed as the decision tree as follows: 106
10. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 Figure 1: The decision tree of probability causing the bad debt As mentions in Figure, the decision tree of bad-debt likelihood presents primarily financial attributes leading to different bad-debt groups that can be interpreted as follows: The “Rare” branch: If the Overdue variable falls into the „Rare‟ group, and combines with high Loan/Collateral rate, what group the bad debt would belong to depends on other factors shown in varied circumstances as bellows: - If the rate of Debt/Total Asset is high, the bad debt would belong to the group of debt type 4-5. - If the rate of Debt/Total Asset is medium with Loan Frequency being average and the rate of Reserve Expense/Debt being high, the bad debt would belong to the group of debt type 4-5. The rate of Reserve Expense/Debt is low or medium, the group of bad debt is type 2-3. - If the rate of Debt/Total Asset is medium with Loan Frequency getting the value of one, the bad debt would belong to the group of debt type 2-3. - If the rate of Debt/Total Asset is medium with Loan Frequency being high and the Income being high, the bad debt belongs to the group of debt type 1. The Income is average or low, then the group of bad debt is type 2-3. - If the rate of Debt/Total Asset is low with the Loan Term being short or long, the bad debt would belong to the group of debt type 4-5. The Loan Term is medium, then the group of bad debt is type 1. If the Overdue variable falls into the Rareness group, and combines with low Loan/Collateral rate, what group the bad debt would belong to depends on other factors shown in varied circumstances as bellows: - If the Credit Policy is labeled as Normality accompanied by high Loan Frequency, the bad debt belongs to the group of debt type 4-5. The Loan Frequency is average, then the group of bad debt is type 2-3. 107
11. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 - If the Credit Policy is labeled as Normality with the Loan Frequency getting the value of one and the rate of Reserve Expense/Debt being high or medium, the bad debt belongs to the group of debt type 4-5. The rate of Reserve Expense/Debt is low, then the group of bad debt is type 2-3. - If the Credit Policy is labeled as Under Control and Awareness, the bad debt belongs to the group of debt type 4-5and type 2-3respectively. If the Overdue variable falls into the „Rare‟ group, and combines with medium Loan/Collateral rate, what group the bad debt would belong to depends on other factors shown in varied circumstances as bellows: - If the Loan Purpose is categorized into the Business with the Loan Term being short and the Income being high, the bad debt belongs to the group of debt type 2-3. The Income is average or low, then the group of bad debt is type 1. - If the Loan Purpose is categorized into the Business with the Loan Term being medium and long, the bad debt belongs to the group of debt type 4-5 and type 2-3 respectively. - If the Loan Purpose is categorized into the Household Spend with the rate of Debt/Total Asset being high, the bad debt belongs to the group of debt type 2-3. The rate of Debt/Total Asset is medium or low, the group of bad debt is type 1. 4.2. Discussion The decision tree of bad debt likelihood indicates that all selected attributes of individual customer that are presented in the research hypothesis cause the bad debt labeled into debt type 1 to 5. However, there are eight attributes that lead to the bad debt in the sector of individual credit, specifically the Overdue, the Loan/Collateral rate, the Debt/Total Asset rate, the Loan Frequency, the Reserve Expense/Debt rate, the Loan Term, the Credit Policy, the Loan Purpose. Analyzing the result of decision tree helps us find that the impacts of these attributes are varied due to their characteristics. In great detail, the Overdue and Loan/Collateral rate play a decisive role in the probability of causing Bad debt in the segment of individual credit. For example, if the Overdue is labeled as Rareness, High or Low accompanied by the medium Loan/Collateral rate, it will lead to various scenarios depended on the other attributes. Furthermore, the others such as the Debt/Total Asset, the Loan Frequency, the Reserve Expense/Debt, the Credit Policy, etc. also leads to the bad debt in the sub-branches. Although each attribute creates a certain branch itself, but the connection between them enable us to understand the root causes of the bad debt in the field of individual credit. Based on the results of the Decision Tree (Figure 1) which presents how various types of bad debt would be run up caused by essentially financial attributes of individual customers, this study proposes a number of solutions to help commercial banks prevent the bad debt from arising in the personal credit segment. First of all, banks should supervise the loan purpose after giving loans, meaning that they need to strengthen the process of supervision for individual customers so as to stop them from using loans for poor investments that might lead to their defaults. Secondly, banks need to tighten up the set of requirements for loan collateral. The result shows that the higher the value of loan collateral, the lower the likelihood of bad debt. Therefore, in order to minimize the bad debt in the future, the banks should focus on the loans accompanied by collaterals and enhance the value of collaterals for loans. Finally, banks should adjust their decisions on loan amount and loan term in accordance with customer‟s credit history. The result insists that the greater the size of loan the borrower get, the higher the likelihood of bad debt would incur. The main reason is that the banks have focused much on some economic sectors that are in need of large capital but with high levels of risk such as the real estate, the ship-building industry, etc. Therefore, the banks should carefully consider borrowers in each business field and stop granting too much capital on high-risk areas. Alternatively, they need to expand lending fields as well as subjects of loans, also to apply suitable loan terms to each type of customer. 108
12. INTERNATIONAL CONFERENCE FOR YOUNG RESEARCHERS IN ECONOMICS & BUSINESS 2019 ICYREB 2019 5. Conclusion The research creates a model that allows forecasting the solvency of individual borrowers in accordance with the characteristics at some commercial banks in Vietnam. The results properly assist the banks in the process of making decision on given-loan acceptance. Accordingly, there are three scenarios that embraces various attributes of personal credit, with ending up being different types of bad debts and NPLs as shown in the illustration of DT (Figure 1). This paper applies the Decision Tree model to identify the financial attributes of individual customers that are likely to cause the bad debt in the segment of personal credit. The research result provides not only the detailed description about the fact of bad debt in Vietnamese banking system, but also the different scenarios of causing the bad debt in the segment of personal credit for the banks to prevent bad debt. Despite deliberate efforts in the research process, this study cannot avoid some limitations due to the lack of data and research experience. REFERENCE [1] Capon, N. (1982). Credit scoring systems: A critical analysis. Journal of Marketing, 46(2), 82-91. [2] Fayyad, U. M., & Irani, K. B. (1992). Technical Note On the Handling in Decision Tree. Machine Learning, 87-102. [3] Hue, N. T. M. (2015). Non-performing loans: Affecting factor for the sustainability of Vietnam commercial banks. Journal of Economics and Development, 17(1), 93-106. [4] Kalra, S. (2015). Vietnam: the global economy and macroeconomic outlook. Journal of Southeast Asian Economies, 11-25. [5] Li, X. L. (2012). An overview of personal credit scoring: techniques and future work. International Journal of Intelligence Science, 2(04), 181. [6] Nawai, N., & Shariff, M. N. M. (2012). Factors affecting repayment performance in microfinance programs in Malaysia. Procedia-Social and Behavioral Sciences, 62, 806-811. [7] Owen, J. D. (2006). U.S. Patent Application No. 11/314,200. Chicago. [8] Pasha, S. A. M., & Negese, T. (2014). Performance of loan repayment determinants in Ethiopian micro finance-An analysis. Eurasian Journal of Business and Economics, 7(13), 29-49. [9] Steenackers, A. &. (1989). A credit scoring model for personal loans. Insurance: Mathematics & Economics, 8(1), 31-34. [10] Vo, H. (2015). Analysis of attributes affecting the bad-debt ratio of joint stock commercial banks in Vietnam. [11] Wu, X., Kumar, V., Quinlan, J. R., Ghosh, J., Yang, Q., Motoda, H., Steinberg, D. (2008). Top 10 algorithms in data mining. Knowledge and Information Systems, 14(1), 1-37. [12] Xiao, W. B. (2006). A study of personal credit scoring models on support vector machine with optimal choice of kernel function parameters. Systems Engineering-Theory & Practice, 10. 109