Research on Simulation of Credit Risk of Credit Card Based on Multi-Agent

Transcription

1 Available online a ScienceDirec Procedia Compuer Science 31 ( 2014 ) Informaion Technology and Quaniaive Managemen (ITQM 2014) Research on Simulaion of Credi Risk of Credi Card Based on Muli-Agen Shuai Li a, Xu Chao a, Lai Hui a, Zhou Zongfang a, * a School of Economics and Managemen, UESTC, ChenDu,610054,China Absrac Banks are no able o idenify every cardholder s credi informaion in heir credi card business, so hey can conrol he credi risk effecively. To solve his problem, his paper esablishes a complex credi risk sysem of credi card business based on muli-agen. Simulaion of differen iniial sae of cardholders and ypes of cardholder in he credi card marke is carried ou. Through he analysis of he simulaion, we believe ha more aenion should be paid o he "overdrafs coningency" cardholders by banks, in order o conrol he credi risk beer and develop he credi card business Published by Elsevier B.V. Open access under CC BY-NC-ND license The Auhors. Published by Elsevier B.V. Selecion and peer-review under responsibiliy of he Organizing Commiee of ITQM Selecion and/or peer-review under responsibiliy of he organizers of ITQM 2014 Keywords: Credi card business, Credi risk, Muli-agen, Simulaion 1. Inroducion Wih he developmen of marke economy, credi cards convenien paymen and credi funcions have gradually been acceped. Therefore, more and more consumers regard credi cards as necessiies. No doub ha he credi card has become one of he fases growing financial businesses in oday's sociey. The rapid expansion of he credi card marke no only makes us see is prosperiy, however, he grea poenial credi risks i brings are quiely gahered as well. Therefore, wihou good risk conrol in he process of expansion, i is easy for banks o fall ino he siuaions of bad rae increasing and huge amoun of bad and doubful debs. Credi risk is he main risk ha faced by he conemporary economy and sociey, while he credi risk of credi card is he mos imporan source of credi card business loss 1. The credi risk of credi card can be defined as he risk ha due o he cardholders "can afford" objecively or "do no wan o" subjecively in paymens, he card issuer will suffer a grea loss. A presen, he sudies on he credi risk of credi card home * Corresponding auhor. Tel.: address: xcwdl@126.com Published by Elsevier B.V. Open access under CC BY-NC-ND license. Selecion and peer-review under responsibiliy of he Organizing Commiee of ITQM doi: /j.procs

2 518 Shuai Li e al. / Procedia Compuer Science 31 ( 2014 ) and abroad are mainly carried ou from he following aspecs: he analysis of macro risk facors and he microscopic personal credi assessmen mehods. Jusin W. Evans (2008) from he perspecive of foreign banks, qualiaively analyzed he causes of he credi risk of credi card and he keys o srenghening he conrol of he risk 2. Sun Lei and Sun Yingjun (2012) made recommendaions accordingly for he credi risk and he fraud risk in he credi card business of China 3. Chi Guoai e al (2006) esablished an evaluaion index sysem of he credi risk of personal credi card, according o he acual siuaion in China 4. Deanna L. Sharpe (2012) furher in-deph discussed he imporan facors affecing he credi risk of credi card of China's urban residens 5. In order o reduce he subjecive facors in he evaluaion process as far as possible and reflec he dynamic changes of he credi risk, regression analysis 6, neural nework 7, suppor vecor machines 8, he combinaion forecasing model 9 and non-parameric random fores classificaion mehod 10, and oher mahemaical mehods have been gradually applied ino he field of credi risk evaluaion. However, among hose mahemaical mehods, he forecas capaciy of he applicans fuure credi is limied, and mos of hem do no consider he associaion and ineracion beween he consumers, so i is difficul o dig ou he complex behavior paerns of and inernal evoluion mechanism ha he sysem generaes. In order o solve hose problems, he analysis and modeling of agen-based complex financial sysems mehods come ino being 11. The basic idea of modeling and simulaing based on muli-agen is ha he sysem is composed of he adapable agens, and he ineracions beween agens. While he agens and environmen will generae new srucures and more complex behaviors. A presen, he mehods based on agen have been widely used in he complex sysem of he invesmen behavior in he financial markes 12, he credi risk conagion of iner-banks 13 and he Enerprise Group credi risk managemen and conrol 14. Zhang Wei e al (2012) combined consumers behavior wih he changes in willingness o repay and repaymen behavior, and buil a credi card predicion model 15. We also did some basic research work on he managemen and conrol of he credi risk of credi card hrough simulaion mehods 16. China's credi card business is on he sage of rapid developmen, in order o sudy he credi risk of credi card in our counry, we ook he behavior of he micro-credi card holders and card-issuing bank as he enry poins. Through muli-agen simulaion mehod, we did simulaion abou he credi card risk managemen of personal credi card consumer behavior and he issuing bank. Then we analyzed he impac of consumer behavior paerns in differen ypes of credi card users on he credi risk of he bank credi card business. A he meanime, we also analyzed differen expansion sraegy ha banks adoped. Finally, he findings of his paper provided a more reasonable choice for he issuing banks in credi policies for differen ypes of applicans and he expansion sraegies. In ha case, i will effecively improve he level of risk managemen in he credi card marke of China, so he problems ha his paper sudied have imporan heoreical and pracical significance. 2. Simulaion model of he credi risk of credi card based on muli-agen 2.1. Basic assumpion of he sysem environmen of he credi card marke Nowadays, he vas majoriy of cardholders in he marke will ake acive seps o mainain is credi saus, and reurn a leas he minimum amoun of he bank repaymen in whaever circumsances, in order o mainain he use of credi cards. In ha case, we make he following basic assumpions: 1. Cardholders acively mainain heir credi saus, on he condiions ha" can afford", he curren reurn of bank amoun no less han he minimum amoun (Minimum-Repaymen, abbreviaed as min-r). 2. Cardholder wih a sable income (Income, abbreviaed as I), and may own some savings (Wealh, abbreviaed as W). 3. Cardholders credi card spending and cash wihdrawal wih daily ineres rae of five en housandhs (r = 0.05%) remain unchanged, and charge monhly wih compounding ineres. According o "he bank card business managemen approach" Aricle 23credi card overdrafs charge monhly wih compounding ineres,

3 Shuai Li e al. / Procedia Compuer Science 31 ( 2014 ) quasi-credi card overdrafs charge monhly wih simple ineres, overdraf ineres rae wih daily ineres rae of five en housandhs, and in accordance wih he ineres rae adjusmen provision of China People's Bank if need adjusmen. 4. Cardholders cash wihdrawal is relaed o he credi line, i is assumed ha he cash wihdrawal (Wihdrawal Cash, abbreviaed as WC) is 50% of is credi line (Credi Line, abbreviaed as CL). The cash wihdrawal proporions in differen banks are differen, ranging from 30% o 100%, in order o faciliae he simulaion, we ake 50%. 5. Credi card annual fee is negligible. In China, many credi cards carry no annual fee or no annual fee wih fixed number of consuming, generally, i is he easy for he cardholders o achieve, so we ake he credi card annual fee wih Properies and behavior of inelligen economic agens in he credi card marke The credi card marke conains wo ypes of inelligen economic agens: cardholder (individual consumers) and he issuing bank Cardholder seings (1) Cardholders propery seing The cardholder's properies include: consumer ypes, income levels, he iniial savings, and credi line. Consumer ype (B): we divide credi card consumers ino hree ypes according o he lieraure [17]: "overdraf Preferences "," overdrafs coningency "and "overdrafs averse ". (2) Cardholders behavior seing Cardholders behavior includes: credi o income, credi card spending, cash wihdrawals, and curren repaymen four Issuing bank seing (1) Issuing banks properies seing The issuing bank's properies include: he cumulaive amoun of bad debs, he number of disabled credi card. The cumulaive amoun of bad debs (Debs): he accumulaed bad debs of bank because of he credi card owners losing he abiliy o repay, his indicaor can reflec he size of he credi risk of credi card faced by he bank. The number of disabled credi card (Number): he number of he cardholders ha can repay he curren specified minimum repaymen amoun and are sopped using he card by he card issuing bank, his is anoher indicaor o reflec he credi risk of bank credi card. (2) Issuing banks behavior seings Issuing banks behavior includes: issuing credi cards, deermining credi line, disabling credi card Issuing credi cards: banks marke credi cards o consumers in order o expand is business. Deermining credi line: banks deermine he credi line wih a cerain mehod, according o he consumers applicaion. Disabling credi card: by seing he minimum repaymen amoun, monioring cardholder repaymen, banks disable he cardholder's credi card when he cardholder can repay he curren minimum repaymen amoun Simulaion model of he credi risk of credi card The cardholder s curren spending should be refleced he wo aspecs of he curren consumpion frequency and each consuming amoun, because each consumpion frequency and consuming amoun will affec he

4 520 Shuai Li e al. / Procedia Compuer Science 31 ( 2014 ) calculaion of ineres, he curren consuming amoun is he oal each consuming amoun. If cardholder s consuming number is a random number, and each consuming amoun is random number as well, hen he oal monhly consuming amoun we obain will also be a random number, which is equivalen wih ha we direcly se he oal monhly consuming amoun a random number. So we se ha consuming many imes per monh for consuming once per monh, and consuming amoun is he adding liabiliies of his monh. According o he basic assumpions as well as he descripions of he properies and behavior of inelligen economic agens in he credi card marke, we have developed he following ineracion rules: C b 1 CL (1) b 0 100%, is he proporion of credi card random consuming wihin he credi line. 1 b2 r1 CL b1 1 r1 T (2) b2 CL C b1 1 r1 Because he acual cash wihdrawal is affeced by boh bank regulaions and credi card consumpion. b %, represens he proporion of random cash wihdraw wihin he, cash wihdrawal; r %, means he proporion he bank ses abou ha he cash wihdrawal accouns for he overall credi line. C d1 T d2 Y 1 3 P 0.05% d (3) Y -1 represens ha he repay amoun in he previous period; d 1, d 2 and d 3 respecively represens consumpion, cash, and he random days wih ineres ha is generaed from he cardholder overdue in he previous period. min R P T Y 10% (4) According o he "bank card business managemen approach ", he overdrafs period of credi card is up o 60 days, and in he firs monh, he minimum paymen of he credi card shall no be less han 10% of is overdraf balance for he monh. R min R b3 S I S min R I min R min R I I S S (5) S Y 1 C T P S is he amoun ough o repay currenlyy W 1 unpaid off currenlyw I T R W is he savings of cardholder currenly. S R is he amoun According o he above ineracion rules, we buil he basic simulaion model based on he NeLogo simulaion plaform, hrough adjusing he corresponding parameers. Due o he fac ha he model conains a lo of random variables, we need o repea he experimen many imes. Through removing he minimum and maximum values and averaging, we go he final resuls of he saisics. Y

5 Shuai Li e al. / Procedia Compuer Science 31 ( 2014 ) Simulaion experimen and resuls analysis According o lieraure [17], he cardholders are divided ino hree ypes, heir specific feaures are as follows: 1. overdraf preferences, i owns he highes consuming willingness, he highes overdraf emoion, he highes repaymen abiliy, he lowes financial pressure, he mos perfec social securiy, he lowes selfconrolling, he lowes ineres rae sensiiviy, he lowes herd environmen; 2. overdraf coningency, i owns he average consuming willingness,, he highes overdraf emoion, he average repaymen abiliy, he average financial pressure, he average perfec social securiy, he average self-conrolling, he average ineres rae sensiiviy, he average herd environmen; 3 overdrafs averse, i owns he lowes consuming willingness, he lowes overdraf emoion, he lowes repaymen abiliy, he highes financial pressure, he wors social securiy, he highes self-conrolling, he highes ineres rae sensiiviy, he average herd environmen;. According o he survey of lieraure [17], hese hree ypes of cardholders respecively accoun for 21.9% 38.5% and 39.6%, and he saisics provide us he corresponding inerval of he income. Accordingly, our simulaion experimens deermined differen iniial parameers for he differen hree ypes of consumers: overdraf preferences, income [5000,8000], he credi consumpion inerval [0.5,0.8]; 2 overdraf coningency, income [2000, 5000], credi consumpion inerval [0.25,0.55]; 3 overdrafs averse, income [1000,2000 ],credi consuming inerval [0.1,0.4]. When he credi line reaches 8000, ha is he condiions mee he applicaion requiremens of all consumers. When we faced he independen and he adjused mixed consumers afer he acual invesigaion, we did specific simulaion abou he credi risk of banks in he 36 monhs, and he resuls are as follows: Number of he disabled credi cards Credi line wih 10000, figure of changes in he number of he disabled credi cards Time (36 monhs overdrafs averse overdraf coningency overdraf preferences acual siuaion Fig. 1. Changes in he number of differen ypes of defauled cardholders in he credi card marke From he Figure 1, we can see he iniial parameers seing for he hree ypes of cardholders are realisic, he" overdraf preferences "has he sronges solvency; he "overdrafs averse" has he weakes solvency; while he "overdrafs coningency" in he middle. From Figure 1, he acual defaul of various ypes of cardholders exised wih cerain proporions is closes o he "overdrafs coningency" consumers. The issuing bank should focus on he "overdrafs coningency" consumers informaion and daa. Therefore, he defaul of he credi card marke can be prediced hrough analyzing is rules and rends. 4. Conclusion This paper sudied i based on muli-agen simulaion mehod: firsly, we did simulaion of he changes of he bank's credi risk in he case of ha he homogeneous cardholders are differen in credi line and iniial savings. The findings show ha he banks should credi o he applicans in he appropriae range; he banks also should consider wheher he applicans have applied for credi and ake he applicans savings condiion ino accoun when doing credi assessmen. Secondly, we furher divided he cardholders ino hree ypes, and consruced he acual siuaion of he marke according o he proporion of he hree ypes; and hen he

6 522 Shuai Li e al. / Procedia Compuer Science 31 ( 2014 ) simulaion was carried ou, he resuls suggesed ha all banks should concern abou he "overdrafs coningency" consumers, in order o srenghen he monioring of credi risk. The model in his paper does no ake he banks' profiabiliy ino accoun, if we combine he profiabiliy and risk sudies, he sudy will be more meaningful. As for he cardholders heerogeneous discussion, we only consider he facors of consuming willingness and income difference, here sill exiss a huge gap wih he acual siuaion. Addiional, his paper did no consider he ineracion beween he consumpion conceps of he consumers, hese deficiencies will become he problems ha need o be focused on in furher research. Acknowledgemens This research has been suppored by Naional Naural Science Foundaion of China ( , ), he Special Research Foundaion of PhD Program of China References [1] Yan W. The macro and micro analysis of China's credi card business risk. Accouning Research; 2007,4, p [2] Jusin W. Evans. Challenging Confucius: Wesern banks in he Chinese credi card marke. Business Horizons; 2008, 51(6), p [3] Sun L, Sun Y. Analysis on China s Credi Card Risk Managemen and Prevenive Measures Based on Game Theory. Advances in Applied Economics and Finance; 2012, 2(4), p [4] Chi G T, Xu W,Sun X F. Research on Evaluaion Sysem and Model for Credi Risk of Individual Credi Card. Journal of Tong Ji Universiy(Naural science); 2006, 34(4), p [5] Deanna L. Sharpe, Rui Yao, Li Liao. Correlaes of credi card adopion in Urban China. Journal of Family and Economic Issues; 2012, 33(2), p [6] GeJun. Research on credi risk of credi card based on Logisic model.china Credi Card; 2010, 12, p [7] Wes D. Neural nework credi scoring models. Compuers and Operaions Research; 2000, 27 (11-12), p [8] Belloi, T., Crook J. Suppor Vecor Machines for Credi Scoring and Discovery of Significan Feaures. Exper Sysems wih Applicaions; 2009, 36(2), p [9] Bhekisipho. Muliple Classifier Applicaion o Credi Risk Assessmen. Exper Sysems wih Applicaions; 2010, 37, p [10] FangK N, Wu J B, Zhu J P,Xie B C. Forecasing of Credi Card Credi Risk Under Asymmeric Informaion Based on Nonparameric Random Foress. Economic Research; 2010, (S1), p [11] WangJun, YaoZheng,CuiCan. Analysis and Modelling of Financial Complex Sysems Based on Agen Approach. Sysems Engineering; 2010, 11, p [12] Li H, Cao H D,Xing H K. Modeling and simulaion of complex finance neworks based on minoriy game.sysems Engineering- Theory & Pracice; 2012, 32(9), p [13] Chen J X, Luo W Q,Pang S L. The risk infecion for he bank of me populaion model Based on he dynamic simulaion of he cellular auomaa. Sysems Engineering-Theory & Pracice; 2012, 32(3), p [14] Xu C, Yang Y, Zhou Z F. Simulaion of Risk Transfer in Enerprise Group Based on Cellular Auomaa. Chinese Journal of Managemen Science; 2012, 20(4), p [15] Zhang W, Zeng W, Xiong X. A Simulaion Sudy on he Risk of Credi Card wih Consumers Behavior. Eas China Economic Managemen; 2012, 26(4), p [16] Xu C, Zhou Z F. Simulaion of credi risk of credi card based on muli-agen, The sixh annual meeing of China managemen, Chengdu,China, paper #04. [17] Lu J Q. A Sudy on he Overdraf Behavior of he Psychological Qualiy and Group Classificaion of Credi Card Consumers[J]. Psychological Science; 2008, 31(2), p