The key factors of outstanding credit balances among revolvers: a case study of a bank in China

Transcription

1 Avalable onlne at ScenceDrect Proceda Computer Scence 91 (2016 ) Informaton Technology and Quanttatve Management (ITQM 2016) The key factors of outstandng credt balances among revolvers: a case study of a bank n Chna Changzheng He a, Bng Zhu a, Mngzhu Zhang a, Xaol He b a Busness School, Schuan Unversty, Chengdu , Schuan, Chna b West Chna Hosptal, Schuan Unversty, Chengdu , Schuan, Chna Abstract The purpose of ths study s to fnd the key factors of the amount of outstandng balances among revolvng credt card users n Chnese credt card market. A Heckman procedure s used to analyze a dataset of a bank n Chna. The small amount of revolver coursng mbalanced problem, and we try to use the rebalanced method n machne learnng doman to deal wth the problem. Results show there are dfferences n the determnants of beng a revolver and the amount of the outstandng balance. Age, housng condton, ndustry, and average cash advance amount per tme, etc. are sgnfcant related to the outstandng credt card balance Publshed The Authors. by Elsever Publshed B.V. Ths by Elsever s an open B.V. access artcle under the CC BY-NC-ND lcense ( Selecton and/or peer-revew under responsblty of the organzers of ITQM 2016 Peer-revew under responsblty of the Organzng Commttee of ITQM 2016 Keywords: Credt card; revolver; outstandng balance; Heckman procedure 1. Introducton A credt card s both a payment tool and a convenent source of credt (Garman & Forgue, 1997). All major credt card ssuers, to a greater or lesser extent, are holdng a portfolo consstng of three types of credt card holder: () non-actve card holders; () non-nterest payng actve card holders ; and () nterest payng actve card holders (Hamlton & Khan, 2001). When credt card holders have an outstandng credt card balance after the last monthly payment, they are known as revolvng credt card users, namely revolvers (Brd, Hagstrom & Wld, 1997). Accordng to the Survey of Consumer Fnances (SCF), n 1998 more than two-thrds of U.S. households had a bank-type credt card, compared to only 43 percent n the 1983 Survey (Bertaut & Halassos, 2002). Credt card holdng has ncreased steadly over the past 20 years. Over tme, the number of credt card revolvers and the amount of outstandng credt card debt has ncreased (Km & DeVaney, 2001). Chnese banks have recognzed that credt card s proftable and wth brllant prospect, and many resources had njected snce Up to the end of November 2011, the fnancal nsttutons had ssued 285 mllons credt cards (Jang, Correspondng author. Tel.: ; fax: E-mal address: hechangzheng@163.com; hechangzheng@scu.edu.cn Publshed by Elsever B.V. Ths s an open access artcle under the CC BY-NC-ND lcense ( Peer-revew under responsblty of the Organzng Commttee of ITQM 2016 do: /j.procs

2 342 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) ). However, Chnese credt card market s stll n ts nfancy and Chnese are restrcted by ther consumpton concepton. As a McKnsey s survey report, most Chnese people use credt cards as a convenent way to make purchases (or accumulate ponts toward a small gft), not as a credt faclty (Atsmon et al., 2011). Unlke US mature credt card market, ther revolvng nterests account for about 70% of card revenue n 2001 (Zhao et al, 2009), there are only about 14% revolvng credt card users n Chnese credt card market (Cu, 2009) and they contrbute about 47% profts (Wang, 2010). Fewer revolvers are one of the characterstcs of Chnese nfancy credt card market. Thus, to study the problem of revolvers n Chnese credt card market makes sense to card ssuers. Much of the prevous research addressng the determnants of credt card use examnes revolvng behavor by dvdng the sample between those holdng a balance and those wth no balance on ther credt cards (Be, 1993; Canner, 1988; Stedle, 1994). However, credt card balances vary from $1 to more than $100,000, so t s more approprate to estmate the amount of outstandng balances on credt cards (Km & DeVaney, 2001). However, lttle research has been conducted n Chna, where the tradtonal consumpton value s save frst and spend later. Based on Bryant's exposton of consumpton and borrowng (1990, pp ) and the lfe cycle theory (Ando & Modglan, 1963), then consder the characterstcs of Chnese nfancy credt card market. Usng the data set obtaned from a bank n Chna, we attempt to fnd the results of the two problems proposed by (Km & DeVaney, 2001): 1) to examne the patterns of Chnese credt card use (transactors versus revolver), and 2) to dentfy the factors affectng the amount of the outstandng credt card balance among Chnese revolvers, and do comparson between the two markets. We fnd there s mbalance problem n dataset whch courses bad results. Thus, we use undersamplng to solve the problem. The results should contrbute to our understandng of Chnese credt card use and provde educators wth nformaton to assst Chnese consumers. Further, polcy makers can use the nformaton to evaluate current regulatons n the Chnese credt card market. 2. Credt card n Chna Snce 1985 Chna ssued the frst credt card, the Chnese credt card ndustry has experenced arduous development. At the end of 2011, people only hold 0.21 cards on average n Chna, even Bejng, Shangha are less than 2 cards (Jang, 2012). However, Amercans hold 4.39 cards on average n 2009 (Ln, 2010). It appears that Chnese credt card market s n ts nfancy. Accordng to MasterCard, they expect that the total number of credt cards n crculaton n Chna n the next 15 years would ncrease at an average annual rate of 11%, wth the total number of cards reachng 1.1 bllon n 2025 (.e cards a person) and the total amount of credt card spendng to ncrease at an average annual rate of 14% reachng US$2.5 trllon by 2025 (MasterCard 2010 Forum). Thus Chna has great market potental. Central bank statstcs show there are more than 30% overall retal sales of socal consumpton products are pad by card, Bejng and Shangha, moreover, are 50% (Cu, 2009). Table 1.Comparson between Chnese credt card market and mature markets Indcators Chnese market Mature markets The rate of credt card consumpton 3% 30%~50% Merchant rate 4% 100%(US), 87%(Korea) The per capta rate (bank card) Charge tmes n very three months (US), 10.8(Korea) However, Chnese have an ngraned cultural reluctance to borrow, even for large expenses such as cars, property, and educaton (Atsmon et al., 2011). The card consumpton amount s far less than Unte Stated. The

3 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) merchants have not recognzed the advantage of the credt card. Chnese credt card market, therefore, s stll n ts nfancy and has many problems. Table 1 s the comparson between Chna and mature markets (Yang, 2010). The above factors wth the sprted competton made the man contrbutes of the credt card annual - annual fees, nterest, nterchange fees merchant dscounts unreasonable. More than 50% are revolvers n mature market, and the nterests obtaned from the revolvers are accounted for about 75% of revenue for the ssuer n US (Durkn, 2000). But there are only 14% revolvers n Chnese credt card market, the nterests are lmted. The market showed an upward turn n recent years. For ncome of Chnese credt card market, the nterests ncreased to 47% n 2009 from 22% n 2003, whle US s about 70% n 2005 (Zhao et al., 2009). In many respects, therefore, the nterests contrbuted by revolvers stll reman at rsng, but not enough. 3. Lterature Revew Why would a revolver arse? Ando and Modglan s lfe cycle model assumes that consumers try to maxmze utlty from lfetme consumpton. The model mples that rather than havng consumpton each year be related to that year s ncome, consumers even out ther household ncome stream over ther lfetme through borrowng or savng. Consumpton needs can be changed by household demographc characterstcs such as age, household sze, and martal status (Km & DeVaney, 2001). Bryant (1990, p. 80) offers a detaled explanaton to show that borrowng s the transfer of future resources to the present to ncrease current consumpton. Under a budget constrant, consumers make borrowng decsons to maxmze utlty. The research about revolver can be roughly dvded nto three groups: 1) studyng who are revolvers and those who are not (Johnson, 2004; Keen, 1998; Lee & Hogarth, 1998; Steddle, 1994;); 2) the behavor of revolver (Hamlton and Khan, 2001; Lee and Kwon, 2002; Hseh, 2004; Reynolds et al., 2006); 3) the amount of outstandng balance among revolvers (Chakravort and Emmons, 2003; Durkn, 2000; Km & Devaney, 2001; Yoo,1998; Zhu and Meeks 1994). Zhu and Meeks (1994) focused on low-ncome households usng the data from the 1983 and 1986 Survey of Consumer Fnances. They found sgnfcant determnants of the amount of credt outstandng n 1986 were household head s employment status and age, credt balance n 1983, and two nteracton varables: specfc atttude toward credt wth head s educatonal level and wth debt balance n Yoo (1998) used the data form Survey of Consumer Fnances to report descrptve statstcs of credt card debt between 1992 and Durkn (2000) analyzed the credt card outstandng ssue (use and consumer atttudes) from Km (2001) used the data from the 1998 Survey of Consumer Fnances to analyze the determnants of outstandng balances among credt card revolvers. Carow and Staten (1999) found consumers are more lkely to use cash when they have less educaton, lower ncomes, are mddle-aged, and own fewer credt cards. Stavns (2000) also used the ndcators, such as: age, ncome, marred status, resdental stuaton, and educaton, to estmate f a credt card holder wll be delnquent. Hamlton and Khan (2002) found that the card holder s behavor,.e. cash advances, mnmum payment due, nterest pad n prevous perods, wll dscrmnate a card holder to be revolver or not. Adaptng the lfe cycle model and Bryant s exposton to credt card use produces the followng proposton. Takng all the factors consdered n the studes of the amount of outstandng balance among revolvers and wth the consumer behavor theory and Chnese characterstcs are also consdered. Credt card payment behavor and the outstandng credt card balance are affected by: consumpton needs (age, martal status, housng condton, support populaton), current resources and future resources (ncome, educaton, ndustry, credt lmt) and behavor (average repayment amount, average consumpton amount, average cash advance amount, delnquent tmes).

4 344 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) Hypotheses Based on the theoretcal framework and prevous research and the Chnese characterstcs, several hypotheses are formulated. Consumpton Needs Cardholders who are younger and marred are hypotheszed to be more lkely to be revolvers and to have a larger outstandng credt card balance than cardholders by those who are older and those who are not marred. Cardholders who have more pressure on house and more populaton to support are hypotheszed to be more lkely to be revolvers and to have a larger outstandng credt card balance than cardholders by those who have less pressure on house and less people to support. Resources Educaton s expected to have a postve effect on the lkelhood of beng a revolver and the amount of the outstandng credt card balance based on theoretcal framework. Cardholder wth less ncome s expected to be more lkely to be a revolver and to have a larger outstandng credt card balance. Cardholders who work n government agency and state run company are less lkely to be revolvers and have smaller outstandng credt card balance. The credt lmt s hypotheszed to have a postve effect on the lkelhood of beng a revolver and the amount of the outstandng credt card balance. Behavor It s hypotheszed that there wll be a negatve relatonshp between the average repayment amount and beng a revolvng credt card user and the amount of the outstandng credt card balance. The average consumpton amount and average cash advance amount are hypotheszed to have a postve effect on the lkelhood of beng a revolver and the amount of the outstandng credt card balance. It s hypotheszed that there wll be a postve relatonshp between the delnquent tmes and beng a revolvng credt card user and the amount of the outstandng credt card balance. 5. Methodology Heckman procedure appeared n the lterature as a smpler alternatve to maxmum lkelhood methods to estmate the parameters of a model of selecton. Ths model ncludes a second equaton, called the selecton equaton, besdes the one to be estmated, whch determnes whether an observaton makes nto the sample, causng the sample to be non-random. To avod bas, the estmaton must take nto account the phenomenon of sample selecton. (Hoffmann & Kassouf, 2005) Consder the followng equaton, whch causes sample selecton. L z u (1) where L s the dfference between the credt card ssuer offered to an cardholder, and hs outstandng balance, and z s a vector of varables that affects L. The varable L s not observed, but we observe f the cardholder s a revolver or not, n the way that: L 1 f L 0 L 0 f L 0 Let W represent the natural logarthm of the outstandng balances of each cardholder, assumng that: W x (2) where x s the vector of varables determnng outstandng balances. Assumng that u and have a bvarate normal dstrbuton wth zero means, standard devaton u and and correlaton, and that L and z are observed for a random sample of cardholders, but W s observed only when L 1,.e, when the cardholder s revolver, then EW ( L 1) EW ( L 0) EW ( u z ) Where x E u z x u ( ) ( ) (3)

5 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) ( u) ( u) ( z / u) ( u) 1 ( ) ( ) ( z / ) u u u (4) and and are respectvely, the normal densty functon and the normal dstrbuton functon. The functon ( u) s called the nverse of Mll s rato. Due to the correlaton between x and ( u), a least squares regresson of W on x, omttng the term n ( u), would produce an nconsstent estmator of. If the expected value of the error was known, t could be ncluded n the regresson as an extra explanatory varable, removng that part of the error correlated wth the explanatory varables and avodng nconsstency. The Heckman s procedure, n ts frst stage, conssts of estmatng the expected value of the error and, n ts second stage, usng t as an extra varable n the regresson equaton. In other words, usng a probt model, parameters of the L equaton are estmated by maxmum lkelhood. Havng the estmates of, ˆ s obtaned for each observaton and used as an exogenous varable n the W equaton, allowng the parameters to be consstently estmated by least squares n the regresson of W on x and ˆ. Consderng that we obtan, from equaton (3), that the condtonal expected value of W s, EW ( L 0) x ( u) (5) Above all, there are two steps n the Heckman method. The frst step s the development of a selecton equaton (.e. a model of factors assocated wth survey non-response). Ths step ncludes dervaton of a varable from the selecton equaton called the Inverse Mlls Rato (IMR). The second step of the Heckman method s the nserton of the IMR varable nto the ntal regresson models (e.g. those not accountng for potental bas due to mssng surveys) from a gven study n order to assess for, and attempt to control for, selecton bas. (Sales et al. 2004) In our study, almost half of the sample used credt cards solely as a convenent means of payng for goods and servces, and therefore had zero credt card balances carred over. An Ordnary Least Squares (OLS) regresson analyss of the amount of outstandng credt card balance that ncluded all of the households wth credt cards would be based because of sample selecton bas (Maddala, 1983). If sample selecton bas s not consdered n the equaton for outstandng balance, there wll be an omtted varable specfcaton. To correct the sample selecton bas of the OLS regresson on the outstandng balance on the explanatory varables, the Heckman procedure was employed n ths study (Maddala, 1983). 6. Data and varables A bank n the cty of western Chna provde the data set used n ths study. Ths bank has ssued about 33,000 credt cards, and about 3000 users are revolvers. About 20% credt card data (6,083 samples) are used n ths study. Our sample ncludes nformaton from credt applcatons and monthly statements from November 2010 through January Among the samples, 586 card holders wth an outstandng credt card balance. Two dependent varables were used for the Heckman two-step procedure. In the frst step, the dependent varable s equal to 1 f the cardholder has an outstandng credt card balance (revolver) and s equal to 0 f otherwse. In the second step, the dependent varable s the average amount of credt card debt remanng after the last monthly payment n the 15 months. The three groups of ndependent varables are consumpton needs, resources and behavor. Consumpton needs ncluded age, martal status and support populaton. As Km & DeVaney (2001), age and age squared are ncluded to measure f there s a possble curvlnear effect on outstandng credt card balances n Chnese credt card market. Martal status s coded as marred and non-marred. Resource varables ncluded ncome, educaton, housng condton and credt lmt. All the varables are contnuous varables. Income was placed

6 346 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) wth ther natural logarthms to reduce heteroskedastcty (unequal varance of the resduals). Housng condton s the money one should pay on ther house. Behavor factors ncluded average cash advance amount, average consumpton amount and average repayment amount. The measurement of varables s presented n Table 2. Table 2.Measurement of Varables Varable Dependent varables Usng of credt card The amount of outstandng credt card balance Independent varables Sexual dstncton Age Age squared Martal status Housng condton Support populaton Ln (Income) Educaton Industry Credt lmt Ln (average repayment amount) Ln (average consumpton amount) Ln (average cash advance amount) Delnquent tmes Measurement 1 f an outstandng credt card balance,0 0therwse Contnuous 1 f male, 0 otherwse Contnuous Contnuous 1 f marred, 0 otherwse 1 f one has no debt on house, 2 f do not need to pay on the house, but the house s not belong to the cardholder, 3 f one has to pay on the house Contnuous Contnuous, log(1+total ncome) The levels of educaton attaned by cardholder 1 f government agency, 2 f state run company, 3 f others contnuous Contnuous, log(1+ average repayment amount) Contnuous, log(1+ average consumpton amount) Contnuous, log(1+ average cash advance amount) Contnuous We present summary statstcs of these varables n Table 3. Among credt card users, 56.53% are male and 70.68% are marred. Meanwhle, about 51.88% credt card users have bachelor degree or above. The average repayment amount per month s CNY 1, The average consumpton amount per tme s CNY 1, The average cash advance amount per tme s CNY Results We use the 6,083 samples to fnd the relatonshps between the dependent varables and ndependent varables. The results are not half satsfactory, because most of the varables are not statstcally sgnfcant. However, the research about the determnants of usng credt card usually ncludng demographcs such as: age, educaton etc. (Zhu and Meeks, 1994; Km and DeVaney, 2001; Wang and Ren, 2004). Ths result, therefore, may have lttle sgnfcance for us. We try to use the undersamplng method n machne learnng to rebalance the data. At the data level, the objectve s to re-balance the class dstrbuton by resamplng the data space, ncludng oversamplng nstances of the small class and undersamplng nstances of the prevalent class. We use the undersamplng to rebalance the data. Then we obtan 1095 credt card users, and 589 are revolvers.

7 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) Table 3.Descrptve Statstcs of credt cardholders Varable Mean (SD) or % Credt card use Convenence user 90.37% Revolver 9.63% Amount of outstandng credt card balance ( ) Sexual dstncton Male 56.53% Female 43.47% Age (8.97) Martal status Marred 70.68% Not marred 29.32% Housng condton 1 (Has no debt on house) 59.91% 2 (Do not need to pay on the house, but the house s not 12.69% belong to the cardholder) 3 (Has to pay on the house) 27.4% Support populaton 0.57 (0.78) Income 6,510 (8,693) Educaton 1 (Junor mddle school and below) 1% 2 (Senor mddle school) 10.96% 3 (College) 36.16% 4 (Bachelor) 41.10% 5 (Master) 8.31% 6 (Doctor) 2.47% Industry 1 (Government agency) 39.36% 2 (State run company) 14.52% 3 (Others) 46.12% Credt lmt 23, ( ) Average repayment amount per month 1, ( ) Average consumpton amount per tme 1, ( ) Average cash advance amount per tme ( ) Delnquent tmes 0.04 (0.23) 7.1. Probt analyss for revolver The results of the probt analyss were obtaned to provde the nverse Mll's rato for the second stage analyss (Table 4). As hypotheszed, age s also curvlnearly related to the lkelhood of beng a revolvng credt card user. However, ths s just contrary wth the results n US. In Chna, younger credt card user s more lkely to be a revolver. Ths mght be that credt card market s stll n ts nfancy; meanwhle younger people are more lkely to study and accept new conceptons and thngs. Moreover, younger people may have worse fnancal problems than older people. In common wth US, cardholder wth lower ncome s more lkely to be a revolver. Martal status, educaton, ndustry and housng condton have no statstcally sgnfcant nfluence on weather a card user s a revolver.

8 348 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) There are statstcally sgnfcant explanatory varables related to Behavor. The average repayment amount s postvely related to the lkelhood of beng a revolvng credt card user, whle average consumpton amount per tme and average cash advance amount per tme are negatvely related to the lkelhood of beng a revolvng credt card user. Table 4.The Heckman Selecton Model: Probt Analyss for Revolver (N=1095) Varable Coeffcent P Sexual dstncton Age Age squared Martal status Housng condton Support populaton Ln(Income) Educaton Industry Credt lmt Ln(average repayment amount) Ln(average consumpton amount) Ln(average cash advance amount) Delnquent tmes Selectvty corrected OLS for the outstandng credt balance The second step of Heckman procedure s corrected for sample selecton bas. The nverse of the Mlls rato,, s estmated and used to reduce bas n the regresson (Table 5). However, the factors whch are statstcally sgnfcant n the second step are not the same as those n the frst stage. Ths may mean that there are dfferences n the determnants of beng a revolver and the amount of the outstandng balance. In the consumpton needs, only age and housng condton are statstcally sgnfcant. The relatonshp between age and the outstandng credt card balances s postve, whch s also contrary wth the results n US. Ths could be nterestng. Accordng to the dataset n ths study, the average ncome between age s 4,910; age s 7,560; age s 9,150; older than 50 s 11,410. The revolvers wth more ncome, hgher consumng ablty they could have. Among the resource factors, ncome, educaton, ndustry and credt lmt are statstcally sgnfcant. Educaton has postve effect on outstandng balance, whch s the same wth US. A one-level ncrease n educaton ncreases the average amount of an outstandng credt card balance by CNY606. Industry s a factor wth Chnese characterstcs. The Chnese have a common dea that one has a job n government agency means he has a lfelong job. And secondly, f one works n a state run company, then the job s more stable than workng n other knds of companes. We fnd the factor ndustry has postve relatonshp wth the outstandng credt card balances. Wang and Ren (2004) also fnd card users work n ndustry such as: bank, nsurance and foregn-funded company, are more lkely to overdraw. The two researches have smlar result, meanng that when ones job s more stable, then he has less outstandng credt card balances. In the behavor factors, the average repayment amount and average cash advance amount are statstcally sgnfcant. The hgher average repayment amount, the less outstandng balances are. The average cash advance amount per tme s postvely related to the outstandng balance. We know that usng credt card to get the cash should pay hgh servce charges compare wth other bank cards. Thus, the larger amount of cash advance from

9 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) revolver, the more eager hs/her need money. Ths may mean that the fnancal stuaton of the revolver s not so good at ths moment that he/she pay the debt less. Table 5.The Heckman Selecton Model: Selectvty Corrected for the Outstandng Credt Card Balances (N=586) Varable Coeffcent P Age Age squared Martal status Housng condton Support populaton Ln(Income) Educaton Industry Credt lmt Ln(average repayment amount) Ln(average consumpton amount) Ln(average cash advance amount) Delnquent tmes Conclusons Usng a dataset obtaned from a bank n Chna, ths study examnes the factors affectng outstandng credt card balance among revolvers usng Heckman two-step procedure. Three groups of factors are appled to the outstandng credt card balance: consumpton needs, resources and behavor The unbalanced dataset n our study mpels us to use undersamplng method to rebalance the data. The balanced data results show that factors nfluence weather beng a revolver and the outstandng credt card balances are dssmlar. Age has an opposte effect on Chnese credt card holders compare wth US. Industry s a Chnese characterstc factor and the results show that cardholders work n government agency and the state run company may have less outstandng balance. There s a negatve relatonshp between the average repayment amount and the amount of the outstandng credt card balance, but the average advance amount has a negatve effect. Acknowledgements Ths work s supported by Natonal Natural Scence Foundaton of Chna ( , ). References [1]Ando, A. & Modglan, F. (1963). The lfe cycle hypothess of savng: Aggregate mplcatons and tests.amercan Economc Revew, 53(1), [2]Atsmon, Y., L, L., Magn, M., St-Maurce, I., Wu, C., Annual Chnese Consumer Study - Capturng the Tradng-off Opportunty n Chna. McKnsey Consumer & Shopper Insghts, McKnsey & Company. [3]Be, L. (1993). Determnantsof consumer credt card repayment patterns. Consumer Interests Annual, 39, [4]Bertaut, Carol and Halassos, Mchael. "Debt Revolvers for Self Control." Board of Governers of the Federal Reserve System workng paper, 2002.

10 350 Changzheng He et al. / Proceda Computer Scence 91 ( 2016 ) [5]Brd, E., Hagstrom, P. A. & Wld, R. (1997). Credt cards and the poor. Insttute for Research on Poverty Dscusson Paper, [6]Bryant, W. K. (1990).The economc organzaton of the household. Cambrdge: Cambrdge Unversty Press. [7]Canner, G. B. (1988). Changes n consumer holdng and use of credt cards Journal of Retal Bankng, 10, [8]Carow, K.A., Staten, M.E., Debt, Credt, or Cash: Survey Evdence on Gasolne Purchases. Journal of Economcs and Busness 51, [9]Chakravort, S., Emmons, W.R., Who Pays for Credt Cards? Journal of Consumer Affars 37(2), [10]Cu, S.F., Decreasng of the overdraft nterest and marketzaton of overdraft nterest. Chnese credt card 12, [11]Durkn, Thomas, A., Credt Cards: Use and Consumer Atttudes, Federal Reserve Bulletn. 86(9), [12]Garman, E.T., Forgue, R.E., Personal fnance (5thed.). Boston NY: Houghton Mffln Company. [13]Hamlton, R., Khan, M., Revolvng Credt Card Holders: Who Are They and How Can They Be Identfed? The Servce Industres Journal 21(3), [14]Hoffmann, R., Kassouf, A.L., Dervng condtonal and uncondtonal margnal effects n log earnng equatons estmated bu Heckman s procedure. Appled Economcs 37(11), [15]Hseh, N.-C., An ntegrated data mnng and behavoral scorng model for analyzng bank customers. Expert Systems wth Applcatons 27, [16]Jang, W.S., How can foregn banks ssue credt card n Chna better. Chnese credt card 11, [17]Johnson, K.W., Convenence or Necessty? Understandng the Resent Rse n Credt Card Debt. Federal Reserve Board, Fnance and Economc Dscusson Seres [18]Km, H. & DeVaney, S.A. (2001). The Determnants Of Outstandng Balances Among Credt Card Revolvers. Fnancal Counselng and Plannng, 12 (1), [19]Keen, C.N., Cohort effects on Credt Card Possesson and Use. Proceedngs of the Assocaton for Fnancal Counselng and Plannng Educaton. 9(11), [20]Lee, J., Hogarth J.M., Shoppng for a Credt Card: Does Informaton Search Pay Off? Consumer Interest Annual. 44, [21]Lee, J., Kwon, K.-N., Consumers Use of Credt Cards: Store Credt Card Usage as an Alternatve Payment and Fnancng Medum. The Journal of Consumer Affars 36 (2), [22]Ln, J. X Foregn banks speed up the layout of Chnese credt card market. Chnese credt card 6, [23]Maddala, G. S. (1983). Lmted dependent and qualtatve varables n econometrcs. Cambrdge: Cambrdge Unversty Press. [24]Sales, A.E., et al., Assessng response bas from mssng qualty of lfe data: The Heckman method. Health and Qualty of Lfe Outcomes, [25]Stavns, J., Credt Card Borrowng, Delnquency, and Personal Bankruptcy. Federal Reserve Bank of Boston New England Economc Revew (July/August), [26]Stedle, R.P., Determnants of bank and retal credt card revolvers: An applcaton usng the lfe-cycle hypothess. Consumer Interest Annual. 40, [27]Reynolds, L.M., Hogarth, J.M., Taylor, A., Cohort Analyss of Consumer Credt Card Behavors: Wll Consumers Be Ready for Retrement? Consumer Interest Annual 52, [28]Wang, M.H., Ren, X.W., The relatonshp between the demographc characterstc of credt card holder and overdraft. Journal of Fnancal Research 286(4), [29]Yoo, P. S., Stll chargng: The growth of credt card debt between 1992 and Federal Reserve Bank of Sant Lous 80, [30]Yang, H.H., The status and development of Chnese credt card market. Chnese credt card, 08, [31]Zhao, Y., Zhao, Y., Song, I., Predctng New Customers Rsk Type n thecredt Card Market. Journal of Marketng Research Vol. XLVI, [32]Zhu, L. Y., Meeks, C.B., Effects of low-ncome famles' ablty and wllngness to use consumer credt on subsequent outstandng credt balances. The Journal of Consumer Affars, 28(2),