MODELING CREDIT CARD BORROWING BY STUDENTS

Transcription

1 Modelng Credt Card Borrowng By Students MODELING CREDIT CARD BORROWING BY STUDENTS Kathleen G. Arano, Fort Hays State Unversty Carl Parker, Fort Hays State Unversty ABSTRACT Credt card use has become accepted practce n the US even for students. Ths places students at rsk of creatng debt that they may be unable to repay. Ths paper estmates the borrowng and amount of borrowng decsons jontly usng the Tobt model, and the two-step estmaton approach that allows for separate estmatons. The two models yelded smlar results, although qualtatve dfferences were found n four varables: ncome, race, atttude towards debt and college major. After controllng for soco-economc and atttude effects, other credt card characterstcs appear to be more mportant than nterest rate n nfluencng student credt card borrowng. INTRODUCTION Easy access to credt cards, assocated wth aggressve marketng by the credt card ndustry, combned wth the ncreasng socal pressure to enjoy a more extravagant college lfe style are beleved to ncrease the lkelhood of students buldng up debt. Ths adds to the total debt burden faced by college students when combned wth student loans to fnance ther educaton. Average credt card usage among college students n terms of outstandng debt, the percentage of students wth cards, and the average number of cards per student has ncreased to sgnfcant levels n recent years. The Nelle Mae report [16] suggests that a market saturaton pont may have been reached and that the percentage of students who have cards may hover around 80 percent. Studes that are specfc to credt card use among college students nvestgate and relate spendng habts and money atttudes to credt card use and how ths could lead to a potental fnancal problem. Hayhoe, et. al. [11, p. 122] found that atttudes towards usng credt cards and gender nfluenced purchasng on credt by college students. Roberts and Jones [18, p. 225], usng a sample of Amercan college students, examned the role of atttude towards money and credt card use n compulsve buyng and found results that suggest credt cards facltate compulsve buyng of unnecessary goods. Related studes by Cunnngham [8] and Norvlts and Santa Mara [17] lkewse found that although a majorty of the students are responsble and can handle ther credt well, there s a sgnfcant mnorty who were havng problems. Thus, these studes emphaszed the need for on-campus fnancal educaton or counselng. The demand for credt s related to the desre to fnance hgher levels of current consumpton than would otherwse be avalable. The demand for credt card 27

2 Southwestern Economc Revew borrowng s a part of an ndvdual s demand for credt. Ths study wll focus on dentfyng and estmatng varous factors that nfluence the unversty student s demand for credt card borrowng. Much of the research on college credt card debt has been devoted to tabulatons of the number of credt card accounts and balances by demographc characterstcs. Very few studes consder the factors that nfluence ndvdual student decson-makng about credt card usage. In ths paper, econometrc models and emprcal estmaton procedures examne credt card borrowng among college students. These models ncorporate demographc, economc, and atttude consderatons nto student decson makng. MODEL SPECIFICATION Based on consumer choce theory, the ndvdual demand for credt card usage, CCU, among college students can be specfed as: CCU = f( P,D,E,A ) (1) where P s the prce of credt measured by the average annual percentage rate (APR) of credt card accounts; D s a vector of demographc varables; E s a vector of economc varables ncludng ncome; and A s a vector of dummy varables for student atttudes toward credt card use and debt n general. The hstorcally slow response of credt card nterest rates changes compared to changes n money market rates has been nvestgated by Ausubel [1, p. 53], Calem and Mester [4, p. 1327] and Cargll and Wendel [5, p. 375]. Ausubel attrbutes the stckness of credt card nterest rates to the rratonalty of credt card holders,.e. they do not look for lower nterest rates because they expect to use the cards as only a transactons medum, and even after they dscover that they use credt cards as a borrowng medum, they do not learn from ther experence. Calem and Mester add that search and swtch costs for the card holders and adverse selecton for frms contrbutes to the stckness of credt card rates, confrmng the clam that credt card rates are stcky because consumers are not responsve to rate cuts. Cargll and Wendel, on the other hand support the vew that household ndfference to nterest rates s a ratonal behavor f n fact other credt card characterstcs are more mportant than nterest rates. Mn and Km [15, p. 136] found that households are responsve to nterest rates when decdng to borrow, but not when decdng how much to borrow. Economc theory ndcates that credt card balances should fall as the average APR ncreases, thus, a negatve coeffcent estmate for P s expected. However, prevous studes suggest that a statstcally nsgnfcant coeffcent mght be expected. Key economc varables under E are ncome, employment status, and credt constrant. In theory, the relatonshp between credt card balances (CCB) and ncome s uncertan as hgher levels of ncome may mply no borrowng constrants or a lesser need to borrow. Both Cargll and Wendel [5, p. 384] and Mn and Km [15, p. 136] emprcally show that ncome negatvely affects the lkelhood of borrowng on credt cards, but s postvely related to the amount borrowed. Intutvely, students who are credt constraned wll have a hgher demand for credt card borrowng, because they have greater need and are lkely to be dened other forms of credt. 28

3 Modelng Credt Card Borrowng By Students Student preferences n terms of wllngness to borrow are also accounted for n the vector of atttudnal varables, A. In general, more acceptng and postve student atttudes towards credt card use and debt overall s expected to lead to hgher demand for credt card borrowng. To control for demographc dfferences, the vector of explanatory varables under D ncludes age, racal group, gender, martal status, student classfcaton, and major. Age and student classfcaton, lke the ncome varable, can have a postve or negatve mpact on credt card balances. Both of these varables affect the amount of desred credt and the ablty to obtan credt. Younger and lower level students may have a stronger desre for credt n the sense that they have less maturty or experence n dealng wth the mplcatons of credt card debt and are more lkely to want mmedate gratfcaton and consumpton fnanced through such debt. On the other hand, they may have less ablty to obtan the credt that they want. It s also hypotheszed that students majorng n busness are less lkely to ncur hgher credt card debt relatve to other majors gven that they have better exposure and nformaton on the mportance of responsble personal fnancal management from ther course work. Table 1 Varables and Summary Statstcs VARIABLE DEFINITION MEAN STANDARD DEVIATION CCB Credt card balance ($) , BD 1 f student currently carres credt card balance INT Average nterest on all credt cards (%) INC Student ncome from all sources n the last year ($) 15, , AGE Age of student SEX 1 f respondent s male SO 1 f respondent s a sophomore JR 1 f respondent s a junor SR 1 f respondent s a senor GR 1 f respondent s a graduate student RACE 1 f whte MS 1 f marred WORK 1 f currently employed MAJOR 1 f student s from College of Busness and Leadershp LOAN 1 f student do not have student loans BAL 1 f student pays more than the mnmum but not the full monthly balance NCC Number of credt cards CC 1 f appled for credt card last year and rejected and f requested a credt lne ncrease and rejected COST 1 f extensve knowledge of how much debt wll ultmately cost STYLE 1 f agree that debt s a normal part of today s lfestyle Note: bnary varables have zero values f otherwse from codng value of 1. 29

4 Southwestern Economc Revew DATA The data used n the emprcal analyss s from a self-admnstered mal survey of Fort Hays State Unversty (FHSU) students that collected nformaton about student fnancal resources, spendng patterns, employment actvtes, and credt card debt. The survey was admnstered n the sprng semester of 2005 from a random sample of 1500 undergraduate and graduate students at FHSU. There were 411 responses. Snce ths study examned credt card borrowng, students must have had at least one bank credt card (Vsa, MasterCard, etc.) and as a consequence, the sample sze was reduced to 278 students. 1 After consderng complete tem responses to the varables specfed n the emprcal models, 243 students remaned n the fnal sample. Detaled varable descrptons and summary statstcs are provded n Table1. 2 METHODOLOGY Credt card debt functons have been estmated usng the tradtonal Tobt model (Knsey [13, p. 173], Calem and Mester [4, p. 1331], Chen and DeVaney [6, p. 169]). Because the Tobt model restrcts the sgn and magntude of the coeffcents of partcpaton and utlzaton to be the same, although there s no a pror reason to assume ths, the two-step estmaton approach has been used to overcome the possblty of model msspecfcaton. Blaylock and Blsard [2, p. 700] have used the two-step estmaton approach to model cgarette consumpton and Byrne, Capps, and Saha [3, p. 616] to model food-away-from-home expendtures. Wth regard to studes about debt, the two-step estmaton procedure has been utlzed by Duca and Whtesell [9, p. 606] to model credt cards and money demand; Crook [7, p. 84] to model demand for household debt; and Cargll and Wendel [5, p. 381] and Mn and Km [15, p. 132] to model credt card borrowng. Ths paper wll present both the Tobt and the two-step estmaton approaches n the statstcal analyss. The Tobt model estmates the borrowng decson and the amount borrowed jontly whch requres the sgns of the parameter estmates for these two separate decsons to be the same. Ths may not necessarly hold. The ncome varable, for example, may have a negatve sgn f hgher ncome reflects the lkelhood of borrowng (whch reflects need ); however, the sgn may be postve f hgher ncome ncreases the lkelhood of borrowng (whch reflects ablty to pay ). By utlzng the two-step approach, dfferent margnal effects for the borrowng decson and for the amount borrowed can be estmated. The two-step approach wll address the possblty of sample selecton bas snce the dependent varable (credt card balance) s observed only for a restrcted, nonrandom sample (those who report postve balances). Tobt Model Specfcaton The general formulaton of the Tobt model s gven by: (Greene [10, p. 908]) ' Y = β X + ε, Y = 0 f Y 0 and Y = Y f Y f 0 (2) where 30 Y s a latent varable and Y s the observed varable, n ths case the amount X nclude nterest rate, ncome, of credt card balances. Explanatory varables ( )

5 Modelng Credt Card Borrowng By Students demographcs, atttudes, and other factors descrbed n the prevous secton. Consderng the model gven n (2) and the observed varabley, we have: (Maddala [14, p. 158]) ( Y 0) E Y φ f = β ' X + σ (3) Φ where φ and Φ are the standard normal densty functon and dstrbuton functon ' evaluated at β X / σ. On the bass of the theoretcal consderatons, credt card borrowng s specfed as: Y = β + β INT+ β INC+ β AGE+ β SEX + β MS + β RACE 0 + β SO + β JR + β SR + β GR+ β WORK+ β COST 7 + β STYLE+ β MAJOR+ β LOAN+ β CC + β NCC+ β BAL+ ε (4) where the explanatory varables are defned (wth summary statstcs) n Table 1. A postve credt card balance ( Y = CCB ) s observed f the latent varable Y > 0 and a zero balance f Y 0. Two-step Model Specfcaton Buldng on the specfcaton of Heckman [12, p. 154], modfed specfcally for credt card borrowng gves: Y ' 1 = β X 1 + ν selecton equaton (5) 1 Y ' 2 = α X 2 + μ man equaton (6) where Y s a latent varable, where we observe Y1 = 1 when students are borrowng on credt cards and Y1 = 0 when students have zero credt card balance. Y 2 s the credt card balance held whch s observable only when Y 1. The frst step nvolves the probt regresson of (5) to determne the probablty of borrowng. OLS estmaton of the sub-sample n (6) wll result n sample selecton bas (from an omtted varable) and Heckman provdes a specfcaton for the omtted varable called the nverse mlls rato (MILLS). For nonzero observatons, Y 2 ' φ ( βx ), the nverse mlls rato s 1 ', a rato of the value Φ βx 1 = ( ) 1 31

6 Southwestern Economc Revew of the standard normal densty functon to the value of the standard normal cumulatve dstrbuton functon, and s ncluded as a regressor n the estmaton of (6). Ths two-step estmaton technque allows separate parameter estmates for the frst stage (borrowng decson) and the second stage (amount borrowed) but lnks them through the nverse mlls rato (Byrne, Capps, and Saha [3, p. 617]). For the two-step estmaton model, two regresson equatons are run, the borrowng decson (probt selecton equaton) n (5) and the amount borrowed (man equaton) n (6) gven below: Y 1 = β + β INT + β INC + β AGE + β SEX + β MS + β RACE 0 + β SO + β JR + β SR + β GR + β WORK + β COST 7 + β STYLE + β MAJOR + β LOAN + ν (7) where the borrowng decson ( ) and BD s observed BD 1f Y f 0 and BD = 0 f Y 0, = 1 1 Y 2 = α + α INT + α INC + α AGE + α SEX + α MS + α RACE 0 + α SO + α JR + α SR + α GR + α STYLE 7 + α MAJOR + α CC + α NCC + α BAL + α MILLS + μ (8) Y 2 s the credt card balance ( ) BD = 1. The two equatons are lnked together by the nverse mlls rato ( ) where the dependent varable CCB gven that MILLS 3. If ths varable s statstcally sgnfcant, then sample selecton bas s present. EMPIRICAL RESULTS 4 The emprcal results derved through the Tobt model and the two-step estmaton process descrbed above s presented n Table 2. Unlke the parameter estmates n OLS regressons, whch represent margnal effects of the explanatory varables, the coeffcents estmated from the Tobt and probt models do not make mmedate ntutve sense. Probt coeffcents measure how much a unt change n the ndependent varable changes the cumulatve normal probablty of the dependent varable (Z scores). As such, the magntude of the parameter estmate vares wth the magntude of the ndependent varables. Tobt coeffcents, on the other hand, measure the margnal effect of the ndependent varable on the latent varable, not the observed varable, Y. It s shown n Greene [10, p. 909]; however, that the margnal effect on the observed varable s smply the parameter estmate multpled by the probablty of the noncensored observaton. Snce the estmated parameters cannot be drectly nterpreted, only the margnal effects of the explanatory varables on credt card borrowng are reported n Table 2, computed at the means of the X s. Actual parameter estmates are reported n Appendx A. 32

7 Modelng Credt Card Borrowng By Students TABLE 2 MARGINAL EFFECTS FOR TOBIT AND TWO-STEP ESTIMATION MODELS VARIABLES TOBIT MODEL TWO-STEP MODEL BORROWING DECISION (SELECTION EQN) HOW MUCH TO BORROW (MAIN EQN) Constant (589.90) 0.22 (0.27) (870.22) INT (11.76) (0.01) (17.00) INC 0.01 (0.01) ( ) 0.01 (0.01) AGE (13.22) (0.006) (18.69) SEX (152.62) (0.06) (225.83) MS (251.73) (0.13) (380.49) RACE (318.30) (0.04) (424.93) SO (315.72) 0.04 (0.11) (486.66) JR (283.85) 0.08 (0.09) (440.28) SR (270.71) 0.18 (0.09) (430.77) GR (326.48) 0.07 (0.11) (493.54) WORK (199.08) 0.29 (0.09) COST (150.60) (0.06) STYLE (160.35) 0.13 (0.07) (282.38) MAJOR (156.13) (0.07) (236.92) LOAN (166.75) (0.07) CC (204.87) (285.57) NCC (69.98) (98.78) BAL (246.92) (338.59) MILLS (554.70) LR stat = p-value = 0 Pred Accuracy χ 17 = χ 16 = p value = 0 p value = 0 and ndcate 10% and 5% sgnfcance levels, respectvely. Standard errors are n parentheses. The regresson results presented below support the general concluson that student credt card balances are systematcally related to the explanatory varables n the two models as demonstrated by the overall goodness of ft measures for the models. In the Tobt model, the lkelhood rato statstc (LR) of (p-value =0) confrms that ths model provdes a better ft versus the alternatve model where all the parameter estmates are jontly zero. The predcton accuracy of 76% also mples a good ft for the selecton probt equaton model and the overall model sgnfcance test for the man equaton n the two-step model also returned a sgnfcant model. From the last column n Table 2, t can be concluded that sample selecton bas s not a problem (model msspecfcaton), because the parameter estmate for the nverse mlls rato (MILLS) s not statstcally sgnfcant (p-value = 0.97). Thus, for our data set, the Tobt model, whch restrcts the borrowng and amount of borrowng decsons nto a sngle parameter estmate, s an acceptable specfcaton of the credt card borrowng equaton. It s nterestng to note that one of the key varables, nterest rate (INT), does not sgnfcantly affect the students decson to borrow or how much to borrow n ether the Tobt or the two-step models. Ths s consstent wth Ausubel s [1, p. 71] argument of credt card consumer rratonalty and Calem and Mester s [4, p. 1327] 33

8 Southwestern Economc Revew emprcal fndng that credt card rates are stcky because consumers are not responsve to rate cuts. On the one hand, ths may be more a reflecton of the populaton consdered n ths study students. It s perhaps more acceptable to beleve the argument of consumer rratonalty f the consumer group s students. Students are more lkely to be mpulsve and fall prey to nstant gratfcaton by usng credt cards to fnance consumpton, and may place less mportance on the future mplcatons and consequences of gong nto debt. Mn and Km [15, p. 136] found that for households, nterest rates negatvely affect the amount of borrowng n the Tobt model, whereas n the two-step model, nterest rate had no sgnfcant effect on the amount of borrowng, only on the borrowng decson. Ths fndng was not confrmed n ths study. Another mportant economc varable s ncome (INC). In the Tobt model, ncome s a sgnfcant determnant of credt card balances for students, and the magntude seems reasonable: for each $100 ncrease n a student s ncome, credt card balances ncrease by $1. For the two-step model, ncome postvely affects the decson to borrow, but has no sgnfcant effect on the amount borrowed. These results would support the hypothess that hgher ncome may mply no borrowng constrant. Consstent wth the coeffcent on INC, a student who s currently employed s 29% more lkely to borrow accordng to the result from the two-step model selecton equaton. In the Tobt model, an employed student has on average, $390 more n credt card debt, than a student not currently workng. Students who have been rejected by another credt card applcaton and/or a credt lne ncrease wthn the last year (CC) have, n the Tobt model, $554 more debt and $835 more debt n the two-step model. Calem and Mester [4, p. 1333] nvestgated the smultanety between credt card debt (CCB) and credt constrant (CC), whch would mply that the parameter estmates here wll be nconsstent f CC s n fact not exogenous. A test for the exogenety of CC was performed where the null hypothess s CC s exogenous followng the procedure from LIMDEP whch s a smple t-test for the hypothess that ρ [ ε1, ε 2 ] = 0, where ε 1 and ε 2 are the error terms from the CCB Tobt model and the model where CC s the dependent varable, respectvely. The t statstc was calculated to be wth a p-value = , therefore, CC s exogenous. As hypotheszed and emprcally supported by Mn and Km [15, p. 137], credt constraned households who are lkely to be dened other forms of credt have hgher demand for credt card borrowng. The results of ths study also show a smlar stuaton for students. Multple credt cards (NCC) ncrease the avalable funds for borrowng and the lkelhood of usng the funds whch may be manfested n hgher balances held. The Tobt model shows that an addtonal credt card held by a student wll ncrease credt card balances by $330, whle the two-step model estmates a $588 ncrease. On the other hand, the premse that people obtan more cards to allow for larger balances (Cargll and Wendel [5, p. 384]) may ndcate NCC s endogenous n ths model. Followng the procedure for testng for exogenety performed above for the varable CC, a smlar test s performed here where results show NCC s n fact exogenous (tstat = 0.002, p-value = ). Smlarly, students who regularly pay more than the mnmum but not the full balance monthly (BAL) 5 have $521 and $838 less balance than students who fal to pay the mnmum balance as per the Tobt model and the two-step model, respectvely. Credt card balances are postvely related to the student s age, consstent wth the hypothess that older students have more credt card debt snce they may 34

9 Modelng Credt Card Borrowng By Students have more access to credt and more ablty to pay. However, age does not sgnfcantly affect the decson to borrow accordng to the result from the two-step model. On the other hand, Calem and Mester [4, p. 1332] and Mn and Km [15, p. 137] found that for households, balances are negatvely related to age, and older households are less lkely to use credt cards for borrowng than younger households. Ths s more consstent wth the premse that age reflects on the amount of desred credt, whch ntutvely should be lower for older households. In terms of race, whte students are 24% less lkely to borrow compared to nonwhtes 6 but do not sgnfcantly mpact the balance held as ndcated by the twostep model. The Tobt model, however, ponts out that whte students have approxmately $650 less credt card debt. These are all consstent wth household behavor and the hypothess that nonwhtes may have less access to alternatve types of credt than do whtes (Calem and Mester [4, p. 1332], Cargll and Wendel [5, p. 385], and Mn and Km [15, p. 136]). Martal status (MS) was not found to affect the amount of balances n both models, but the selecton probt equaton n the two-step model ndcates that a marred student s 22% less lkely to borrow than a sngle student. In the same manner, although a senor s 18% more lkely to borrow than a freshman, the student s academc classfcaton does not nfluence the amount of credt card borrowng. Results from the selecton equaton of the two-step model lkewse mply that a student majorng n busness has a lower lkelhood of borrowng (MAJOR) compared to students from other colleges, but has no sgnfcant effect on the amount of balances held (from the man equaton). The Tobt model, on the other hand, suggests that a student majorng n busness has $294 less credt card debt relatve to students from the three other colleges n the unversty. It mght be nferred that students who have more exposure to fnancal management as a result of beng a busness major, are more prudent n ther credt card debt compared to non-busness majors. A student wthout student loans s 22% less lkely to borrow (two-step model) and has $282 less n credt card debt (Tobt) than a student also burdened wth student loans. It would seem that credt card debt does add to the overall debt burden of students. Extensve knowledge on how much credt card debt wll ultmately cost does not sgnfcantly nfluence the borrowng decson (from selecton equaton of twostep model) and the amount of balance held (Tobt model). Ths s actually consstent wth the results on the INT varable, whch s lkewse not sgnfcant. However, a student who agrees that debt s a normal part of today s lfestyle s 13% more lkely to borrow and has almost $443 more debt, based on the two-step model and the Tobt model, respectvely. 7 CONCLUDING COMMENTS Ths research represents an ntal effort of modelng credt card borrowng of students usng econometrc models that are anchored on the economc theory of consumer behavor that ncorporates demographc, economc and preference consderatons nto the decson makng process. Prevous research usng ths approach has focused manly on household credt card borrowng, whle work on student credt card borrowng has been mostly restrcted to cross-tabulatons. Where econometrc models were ncorporated, the set of explanatory varables were lmted. 35

10 Southwestern Economc Revew Credt card usage s a two part decson for any person: (1) shall the card be used (borrowng decson); (2) and f so, to what extent (amount of borrowng). The emprcal fndngs n ths paper suggest that the tradtonal Tobt model, whch combnes the borrowng and the amount of borrowng decsons nto a sngle estmaton, s an acceptable specfcaton of modelng student credt card borrowng. Results from the two-step estmaton approach dd not provde evdence of sample selecton bas (model msspecfcaton), but provded addtonal nformaton to the results of the Tobt model. Overall, the two approaches yeld smlar results, although there were some qualtatve dfferences on four varables INC, RACE, STYLE and MAJOR. From the Tobt model, credt card borrowng ncreases wth ncome, whle n the two-step model, hgher ncome ncreases the lkelhood of borrowng, but has no sgnfcant mpact on the credt card balance held. Whtes have lower balances accordng to the Tobt model; however, accordng to the two-step model, there s no statstcally sgnfcant dfference between whtes and nonwhtes n terms of the amount of borrowng, but whtes are less lkely to borrow. A more acceptng atttude towards debt as a lfestyle leads to hgher balances n the Tobt model, but n the twostep model, although t (STYLE) ncreases the probablty of borrowng, t has no mpact on the level of debt. A smlar relatonshp s demonstrated by the varable MAJOR a student from the busness major has less debt than a student from other majors as per the Tobt model; on the other hand, the two-step model predcts a student from the busness college s less lkely to borrow, but the balance held s no dfferent from a student from another college. Another key result s that student credt card borrowng does not seem to be nfluenced by nterest rate, consstent wth Ausubel s hypothess of consumer rratonalty, whch seems a more plausble explanaton for students than for households n general. After controllng for demographc, economc, and atttude (preferences) effects, other credt card characterstcs seem to be more mportant than nterest rate n nfluencng student credt card borrowng. These nclude the number of credt cards held, and credt card constrants faced by the student. Consequently, f polcy makers and unversty offcals were to pursue polces desgned to moderate credt card debt of students, t mght focus on credt card solctaton. Gven the fact that students do use credt cards for borrowng, t mght also help to mprove fnancal tranng and lteracy of students, partcularly on the ssue of how much ther debt ultmately costs them. 8 36

11 Modelng Credt Card Borrowng By Students APPENDIX A ACTUAL PARAMETER ESTIMATES FROM THE TOBIT AND TWO-STEP ESTIMATION MODELS. VARIABLES TOBIT MODEL TWO-STEP MODEL BORROWING DECISION (SELECTION EQN) HOW MUCH TO BORROW (MAIN EQN) Constant (895.21) 0.72 (0.92) (870.22) INT (17.75) 0.01 (0.02) (16.99) INC 0.02 (0.01) ( ) 0.01 (0.01) AGE (19.86) 0.01 (0.02) (18.69) SEX (230.28) (0.21) (225.83) MS (380.14) (0.36) (380.49) RACE (480.75) (0.69) (424.93) SO (476.71) 0.13 (0.39) (486.66) JR (428.65) 0.30 (0.35) (440.28) SR (408.47) 0.61 (0.33) (430.77) GR (493.01) 0.23 (0.43) (493.54) WORK (301.26) 0.84 (0.25) COST (227.27) (0.21) STYLE (241.39) 0.41 (0.21) (282.38) MAJOR (235.58) (0.21) (236.92) LOAN (252.36) (0.21) CC (307.72) (285.57) NCC (100.90) (98.78) BAL (371.03) (338.60) MILLS (554.70) and ndcate 10% and 5% sgnfcance levels, respectvely. Standard errors are n parentheses. 37

12 Southwestern Economc Revew ENDNOTES 1 68% of the students surveyed had at least one credt card n ther name; ths ncludes charge accounts from retal store cards and gas cards. 2 For a summary report on the credt card debt porton of the survey, see Arano (2006), whch shows behavoral patterns. Conclusons derved from these smple cross-tabulatons should however be treated wth cauton, and to better test causal relatonshp of credt card debt to demographc, economc, and atttudnal varables, consder results from ths paper. 3 For dentfcaton, t s requred that there should be some X n the selecton equaton not ncluded n man equaton. The varables excluded from the man equaton could stll affect the CCB equaton through the nverse mlls rato (Maddala [14, p. 233]. 4 The reported results, whle they perhaps yeld some nsght nto the natonal pcture, are based on a sample of students from a smaller unversty n western Kansas and may not be representatve of a natonal sample of students. 5 It s lkewse possble for BAL to be endogenous -- hgher credt card balances may ncrease the lkelhood that the student s not able to pay off the mnmum monthly balance. The exogenety of BAL was therefore tested usng smlar procedures dscussed earler and the test shows BAL s exogenous n ths model (t-stat = -0.91, p- value = ). 6 The RACE varable was represented wth ether Whte or Non-Whte (ncludes Black, Amercan Indan, and Asan/Pacfc Islander). Although the Non-Whte group mght show a sgnfcant varaton wthn tself, t was not further dsaggregated as the number of observatons wthn each of these sub groups was too small. The Non- Whte maybe consdered as the mnorty racal group. 7 Agan, t s possble that because a student has substantal credt card debt, the student s more lkely to agree that debt s a normal part of today s lfestyle. Ths mght be a ratonalzaton of the student s behavor. If ths s the case, STYLE maybe endogenous n ths model but the test for exogenety ndcate t s exogenous (t-stat = 0.031, p-value = ). 8 These results are based on a survey whch opens the possblty of bas n the form of self-selecton ssues for those who choose to answer the survey. Although the survey also ncluded questons other than those about credt card debt, ultmately, as wth other surveys, the researchers have no control on whether a student chooses to partcpate n the survey or not, whch may result n a nonrandom sample. The results presented should therefore be taken wth ths n mnd. 38

13 Modelng Credt Card Borrowng By Students REFERENCES Ausubel, L.M The Falure of Competton n the Credt Card Market. Amercan Economc Revew. (81): Blaylock, J.R., and W.N. Blsard U.S. Cgarette Consumpton: The Case of Low-Income Women. Amercan Journal of Agrcultural Economcs. (74): Byrne P.J., O. Capps Jr., and A. Saha Analyss of Food-Away-from-Home Expendture Patterns for U.S. Households, Amercan Journal of Agrcultural Economcs. (78): Calem, P.S., and L.J. Mester Consumer Behavor and the Stckness of Credt- Card Interest Rates. Amercan Economc Revew. (85): Cargll, T.F., and J. Wendel Bank Credt Cards: Consumer Irratonalty versus Market Forces. The Journal of Consumer Affars. (30): Chen, Y-W. and S. DeVaney The Effects of Credt Atttude and Socoeconomc Factors on Credt Card and Installment Debt. The Journal of Consumer Affars. (35): Crook, J The Demand for Household Debt n the USA: Evdence from the 1995 Survey of Consumer Fnance. Appled Fnancal Economcs. (11): Cunnngham, J College Student Credt Card Usage and the Need for On- Campus Fnancal Counselng and Plannng Servces. Undergraduate Research Journal for the Human Scences. (1). Duca, J.V., and W.C. Whtesell Credt Cards and Money Demand: A Crosssectonal Study. Journal of Money, Credt, and Bankng. (27): Greene, W Econometrc Analyss. Prentce-Hall, Upper Saddle Rver, New Jersey. Hayhoe, C.R., L. Leach, P. Turner, M. Brun, and F. Lawrence Dfferences n Spendng Habts and Credt Use of College Students. The Journal of Consumer Affars (34): Heckman, J.J Sample Selecton Bas as a Specfcaton Error. Econometrca (47): Knsey, J Determnants of Credt Card Accounts: An Applcaton of Tobt Analyss. Journal of Consumer Research. (8): Maddala, G.S Lmted-Dependent and Qualtatve Varables n Econometrcs. Cambrdge Unversty Press, New York. Mn, I., and J-H. Km Modelng Credt Card Borrowng: A Comparson of Type I and Type II Tobt Approaches. Southern Economc Journal. (70): Nelle Mae Undergraduate Students and Credt Cards n 2004, An Analyss of Usage and Trends. Norvlts, J.M., and P.S. Mara Credt Card Debt on College Campuses: Causes, Consequences, and Solutons. College Student Journal. (36): Roberts, J., and E. Jones Money Atttudes, Credt Card Use, and Compulsve Buyng Among Amercan College Students. The Journal of Consumer Affars. (35):

14 Southwestern Economc Revew 40