Credit Risk Assessment of Bank Customers Using DEMATEL and Fuzzy Expert System - Case Study Example

Credit Risk Assessment of Bank s using DEMATEL and Fuzzy Expert System Credit Risk Assessment of Bank Customers using DEMATEL and Fuzzy Expert System Abstract Islamic banking is essential in the Islamic realm that deploys Islamic moral acts that steer Muslims in their daily lives. This banking system is very beneficial to its users but its contains risks like any other firm that requires profitability. One most evident type is credit risk that is greatly assessed in this paper with the use of Dematel method and fuzzy expert systems. This paper consists of credit risk management that is carried out with the implementation of computed technology to increase credibility of customer’s therefore increasing profitability of the banking institution. The research is carried out using data from the Middle East Region of the Asia. Introduction Banking has become a necessity to individuals who require safe storage of their financial assets(Bilal,7). Among commercial banking various benefits, some risks are associated with them especially credit risk which requires management for any bank institution to make profits for its stakeholder and investors. Assessment of credit risk is necessary for sustainability of banking institutions especially in the Islamic realm where moral laws such as shari’ah and social justice are adopted in order to preserve customer associations, reasonable dealing, protection and precautions of staff amongst others(Bilal,7). Management of credit involves mitigating the exploitation of the risk, which is applied by including credit scoring models that serve as structures for providing credit to customers. Research has been conducted on how to evaluate credit risk with success factors being evident in the use of GA-based SVM and Rough set theory that provided effectiveness in data mining therefore contributing a positive impact on risk restructuring(Jianguo and Bai, 3). Other studies included credit risk assessment with support vector machines and hybrid neutral systems that resulted in robustness in the use of fuzzy logic in real time applications to solve problems specifically in credit risk management(Shin, Lee and Kim, 130). This paper concentrates on credit risk assessment using Dematel and fuzzy expert systems applying credit scoring models. The objective of this paper is to study consumers’ credit risks that are obtained from previous research that includes financial ratios obtained from banking balance sheets. Rules are used to determine the correlation between consumers’ financial credit risk levels and resultant financial circumstances that is improved by expert decision making that is from filtered financial ratios(Amorim, Vasconcelos and Brasil, 455). Credit Scoring This is a technique applied to by banking institutions to assess feasible risk created by providing money to consumers and therefore mitigate any loss associated with dire debt(Jianguo and Bai, 3). This technique is divided into two types specifically behavioral scoring and application scoring. Behavioral scoring involves evaluation of existing consumer data that involves payment history and payment patterns on previous loans. Application scoring involves evaluation of applicant data that involves financial and demographic data that categorize the applicants to either ‘good’ or ‘bad’(Shin, Lee and Kim, 130). Basel III regulations of the Basel accords are applied to credit scoring with this regulation superseding Basel II regulation by including proper scenario evaluation in terms of supervisors, ranking bureaus and firms instead of calculation credit risk with reference to outsourcing to firms that are not supervised that is in Basel II(Min and Lee, 605). The Basel III regulation is a necessity that is applied in every banking institution where credit scoring models are used to improve the competence of principal distribution. Credit scoring models are therefore important for profitability of a banking institution in terms of applying credit scores that involve numerical calculations on statistics of a customer’s credit history therefore symbolizing the customer’s credibility in terms of credit and loan application(Min and Lee, 605). A suitable model must therefore be applied for effective credit risk assessment of a customer that specifically relates numerical calculations on existing customer data that is financial and demographic basing it on their credibility to pay back their debt. This relationship is eventually used to evaluate new applicants on subsequent credibility that is based on a specific criterion(Amorim, Vasconcelos and Brasil, 455). Dematel Method This technique involves numerical calculations that result to graphical theory that facilitate arranging and deciphering of problems with visual aid therefore segregating several principles into a cause and effect cluster that increases the competency of contributory associations used to draw network diagrams such as diagraphs used to exhibit direct associations of systems and their sub counterparts.(Min, Lee and Han, 654). This technique involves four stages of quantitative evaluation based on a specific criterion. The steps begin by determining a standard matrix, which is used to determine a standardized preliminary direct connective matrix that is eventually used to determine a total connective matrix that determines a verging numerical finally used in drawing the network diagram specifically showing connectivity in a network relationship map (NRM). Determining a standard matrix is the first stage that involves numerical calculation of a number of experts who have an opinion on a specific criterion(Lee, 69). The number of experts (N) give an opinion on how a specific criterion affects another one specifically criterion a and b respectively.(Min, Lee and Han, 654). This evaluation is represented by Hab with each opinion given a rank from 0-4 whose numbers symbolize a null opinion, low opinion, medium opinion, high opinion and excellent opinion respectively. The standard matrix symbolized by yxy= S is then computed with the following formula [Hab]yxy=1/N ∑Nk=1[xkab] yxy, which is also referred to as the preliminary direct connective matrix that shows an impact of one criterion on the other(Lee, 69). The second stage involves determining the standardized preliminary direct connective matrix C using the following formula E= max (max1≤a≤y∑b=1Hab, max1≤b≤y∑a=1Hab) that is used in the next formula, C=S/E. E determines the computation of rows a and b of matrix S therefore symbolizing the total impact of criterion a with the other criterion and vice versa with criterion b on a(Lee, 69). The third stage involves computing the a total connective matrix M of the initial yxy by the following formula max1≤a≤y ∑yb-1Hab, with M=[mab] and a,b=1,2……., which finally develops a new formula M=C+C2+Ct=C(1+ C+C2+….+Ct-1)=C|(1+ C+C2+….+Ct-1)(1-C)|(1-C)-1=C(1-C)-1, as t ∞ and |(1+ C+C2+….+Ct-1)(1-C)|= (1-C)m (Burges, 956). Finally, the rows and columns are defined in terms of o and l vectors that are used to compute the sum of the total connective matrix developing the following formulas o=[o a] yx1=(∑yb=1 mab) yx1 and l=[l b]’ yx1=(∑ya=1 mab)’yx1. This rows and columns lead to the equalization between a=b therefore determining the sum (oa+lb) giving an indicator that symbolizes the impact specified and expected by criterion a. The difference that is (oa-lb) provides a net impact of a on the application tested. The following correlations are therefore formed from (oa-lb) where if positive and negative cause them to be a net specified and expected respectively(Burges, 956). The fourth stage that involves determining a verging numerical, z elaborates the organizational association among the criteria above therefore managing the convolution of the application (Credit risk assessment) by sorting negligible impact in matrix M. The criteria whose impact on matrix M is greater than the verging numerical should be used on the network relationship map according to expert opinions. The final verging numerical is finally chosen and used on the map(Xiao and Wu, 166). Fuzzy expert system This system uses fuzzy logic in terms of rules and functions to reason customer data concerning credit risk assessment(Li and Love, 170). The following fuzzy expert system will be used in the credit risk analysis, an input will be provided and an output a as consequence which are customers’ financial ratios and predicted credit risk respectively. Fuzzy inference systems are part of the fuzzy expert systems used where fuzzy if-the-rules, set theory and reasoning is applied. Its involves a structure of a rule base, database and reasoning method that serves as the model used for credit risk assessment. There are three different types of fuzzy inference models that mainly include the Sugeno fuzzy model, Mamdani fuzzy models and Tsukamoto fuzzy model. The Mamdani fuzzy model is used in conducting the credit risk evaluation where its rules symbolizing the expert knowledge is concerned with fuzzy sets in their respective effects(Li and Love, 170). The Mamdani fuzzy model’s main goal was to run a steam engine by a combination of linguistic control configurations and rules provided by experts. The fuzzy model includes fuzzy sets for example X,Y and Z and rules such as a, b and c and a correlation between them like if a is A AND/OR b is Y then c is Z. The final fuzzy set is therefore defuzzified, which involves extracting a crisp input numeral from a fuzzy set that eventually becomes a representative numeral with the application of various methods like centroid, bisector, mean of maximum, weighted average, smallest of maximum, largest of maximum. The centroid is the common method used with the defuzzified numeral of fuzzy set X, j(X) is computed using the following formula therefore determining its membership functions. The formula includes j(X)= ᶴXa.μX(a)ja/ᶴXμX(a)ja. Empirical Study The research was conducted using Saman bank data located in Iran, which was recently founded in the year 2002. Data collected from the financial ratios was computed by other researchers by using Kolmogrove-Smirnov tests to verify financial ratios allocation therefore choosing significant values based on a specific allotment category. Effectiveness of the financial ratios is determined using the Dematel method and the credit assessment with the fuzzy expert system. The fuzzy system is first developed by choosing appropriate input and output values, determining the values’ membership functions that identify the rules of the correlation between the input and output. This is made possible with the implementation of the MATLAB software used in developing fuzzy inference models(Xiao and Wu, 166). The analysis begins by extracting financial ratios by experts from the Saman bank balance sheets for credit assessment of potential applicants. They include Current Ratio(D1), Quick Ratio (D2), Asset Turnover(D3), Cash Ratio(D4), Working Capital Turnover(D5), Average Collection Period(D6), Inventory Period(D7), Debt Coverage Ratio(D8), Debt Ratio(D9), Current Debt to Net worth(D10), Gross Profit Ratio(D11), Return on Equity(D12), Return on Assets(D13), Payout Ratio(D14), Return On Sales(D15) and Debt to equity ratio(D16). The Dematel method was used to determine meaningful variables in the financial ratios that can be used for credit risk assessment(Xiao and Wu, 166). Table 1: The sum and differences of opinions specified and expected criteria Criteria Oa Lb oa+lb oa-lb D1 3.1372 1.9415 5.0787 1.1957 D2 3.3106 2.1526 5.4632 1.158 D3 1.9003 3.064 4.9643 -1.1637 D5 2.2439 2.8586 5.1025 -0.6147 D6 2.7273 2.1389 4.8662 0.5884 D8 2.4636 2.9979 5.4615 -0.5343 D9 2.7695 1.7077 4.4772 1.0618 D10 2.1161 2.689 4.8051 -0.5729 D11 2.2176 2.4201 4.6377 -0.2025 D13 1.6539 2.9217 4.5756 -1.2678 D14 2.3669 2.495 4.8619 -0.1281 D15 2.4525 1.6464 4.0989 0.8061 The table displays the meaningful financial ratios that will be applied in the fuzzy expert system with the elimination of Cash Ratio, Inventory Period, Return on Equity and Debt to equity ratio. The next step involves constructing a causal diagram as a result of the Dematel method determining the most efficient financial ratios that is D1, D2, D6, D9 and D15. Figure 1: Causal diagram From the diagram above, filtering of the financial ratios has been conducted therefore choosing efficient criteria that will be used for decision making by the experts using the fuzzy expert system is enabled. The financial ratios chosen are Current Ratio, Quick Ratio, Average Collection Period, Debt Ratio and Return On Sales. These financial ratios are used as input and credit risk degree as the output used for construction of the fuzzy expert system. Table 2: The Inputs and Output of Fuzzy Expert System Sign Inputs Interval Type of Membership function Linguistic Terms CR Current Ratio [0 2] Gbell Low(L)Medium(M)High(H) DR Debt Ratio [0 1] Gbell Low(L)Medium(M)High(H) ROS Return on Sales [0 3] Gbell Low(L)Medium(M)High(H) ACP Average Collection Period [0 300] Gbell Low(L)Medium(M)High(H) QR Quick Ratio [0 1] Gbell Low(L)Medium(M)High(H) CRDC Credit Risk of Customer [0 1] Gaussian 1 Low(L)Medium(M)High(H) Table 3: The Rules of Designing Fuzzy Expert System CR DR ROS ACP QR CRDC 1 H M M H L H 2 M H H L M L 3 H L M H H M 4 M H H M M H 5 L H M L M H 6 L L H H L L 7 H M L M M M 8 M L H L H L 9 H H L M M M 10 M H M H L H The fuzzy expert system is finally used with the input and output and membership functions to determine credit risk by customers therefore using prediction. Figure 2: Three Gbell Membership functions for current Ratio Figure 3: Three Gbell Membership Function for Debt Ratio Figure 4: Three Gbell Membership functions for Return on Sales Figure 5: Three Gbell Membership functions for Average Collection Period Figure 6: Three Gbell Membership functions for Quick Ratio Figure 7: Three Gaussian Membership function for Credit Risk Degree An example of the data from the system is taken as CR=1.31, ROS=0.14, QR=0.23, DR=0.42, ACP= 161 and CRDC=0.52. This example can be used to determine credit risk of any potential loan applicant. Conclusion The research conducted in this paper concluded to need to increase credit risk assessment in banking institutions to decrease the chances of granting loans to customers of low credibility. The use of Dematel method and a fuzzy expert system has provided an insight in decision making that is required by banking institutions in order for them to evaluate customers as they apply for loans. The methods have deemed fruitful therefore can be used for further evaluation in the banking institution. Works Cited 1. Ahmed. Bilal. Managing Credit Risk in Islamic Banking. Directory of Social Sciences Articles e.Publications. 1.1 (2011): 1 - 8. Print. 2. Ahn. B, Cho. C, and Kim. C. The integrated methodology rough set theory and artificial neural network for business failure prediction. Expert Systems with Application. 18.2 (2000): 65-74. Print. 3. Blum, M. Failing company discriminant analysis. Journal of Accounting Research.12. 1 (1974): 1-25. Print. 4. Burges, C.J.C. A tutorial on support vector machines for pattern recognition. Data Mining and Knowledge Discovery. 2 (1998): 955-974. Print. 5. Canbas, S., A. Cabuk, and S.B. Kilic, “Prediction of commercial bank failure via multivariate statistical analysis of financial structure: The Turkish case,” European Journal of Operational Research vol. 16.6 (2005): 528–546. Print. 6. Chiang, L.H., M.E. Kotanchek and A.K. Kordon. Fault diagnosis based on fisher discriminant analysis and support vector machine. Computers and Chemical Engineering. 28 (2004): 1074-1389. Print. 7. Cristiamini, N. and Shawe-Taylor J. An Introduction to Support Vector Machines. Cambridge university press, Cambridge: England, 2000. Print. 8. De Amorim, B. P., G. C. Vasconcelos and Brasil L. M. Hybrid Neural Systems For Large Scale Credit Risk Assessment Applications. Journal of Intelligent & Fuzzy Systems. 18 (2007) 455–464. Print. 9. Gold, Carl, and Peter Sollish. Model Selection for Support Vector Machine Classification. Neurcomputing. 55 (2005). 221-249. Print. 10. Lak, Paw. Rough Sets and Intelligent Data Analysis. Information Science. 147.11 (2002): 1-12. Print. 11. Lee, Y. C. Application of support vector machines to corporate credit rating prediction. Expert Systems with Applications. 33 (2007): 67-74. Print. 12. Li, Heng, and Peter E. D.Love. Combining Rule-Based Expert Systems and Artificial Nueral Networks for Mark-up Estimation. Construction Management and Economics. 17 (1999): 169 – 176. Print. 13. Min, J. H. and Y.C. Lee. Bankruptcy prediction using support vector machine with optimal choice of kernel function. Expert System with Applications. 28 (2005): 603-614. Print. 14. Min, S. H., J.M. Lee, and I. Han. Hybrid genetic algorithms and support vector machines for bankruptcy prediction. Expert Systems with Applications. 31 (2006): 652-660. Print. 15. Shin, K. S., and Y. J. Lee. A genetic algorithm application in bankruptcy prediction modeling. Expert Systems with Applications. 23.3 (2002): 321–328. Print. 16. Shin, K. S., T. S. Lee, and H.J. Kim. An application of support vector machines in bankruptcy prediction model. Expert System with Applications. 28 (2005): 127-135. Print. 17. Shin, Kyuan-Shik, Taik Soo Lee, and Hyun-jung Kim. An Application of Support Vector Machines in Bankruptcy Prediction Model. Expert System with Application, 28 (2005) 127-135. Print. 18. Tipping, M. E., and Bishop C. M. Mixtures of Probabilistic Principal Component Analyzers, Neural Commutation. 11. (1999): 443-482. Print. 19. Xiao, J. H. and J. P. Wu. SVM model with unequal sample number between classes. Computer Science. vol. 2 (2003): 165-167. Print. 20. Xu, Jianan, and Bao Xi. AHP-ANN Based Credit Risk Assessment for Commercial Banks. Journal Harbin University Science & Technology. 6 (2002) 94-98. Print. 21. Zhou, Jianguo, and Tao Bai. Credit Risk Assessment Using Rough Set Theory and GA-Based SVM. The 3rd International Conference on Grid and Pervasive Computing-Workshops. IEE Computer Society. 2008. Print. Read More