Researchbackground andmotivation
Non- performingloans(NPL)areasevereproblemforthewholeeconomyoftheworldsincet h e y l e a d t o t h e f i n a n c i a l crisisi n E a s t Asianc o u n t r i e s , AmericaandS u b -
S a h a r a n Africa(Farhan,Satta,C h a u d r h y & Khalil,2012).Therefore,f i n d i n g o u t t h e m a i n determinantso f N P L playsani m p o r t a n t r o l e i n p o l i c y m a k i n g i n ordert o preventt h e futurebaddebts(Adebola,WanYusoff,&Dahala(2011) inFarhanetal (2012)).Previousstudiesidentifythatmacro- economicconditions,bankandborrowersspecificcharacteristics,loanscharacteristics,relationshipb anking 1 ,andcollateralsarekeydrivers ofdefaultrisksandhenceNPL.
Therelationbetweencollateralcharacteristicsandloandefaultisinvestigatedinmanystudiesoverthe world.However,thefindingsareinconsistentamongdifferentpapers,someofwhichs h o w p o s i t i v e relationshipwhiletheothersprovideevidenceof anegativeeffect.
Research by Berger, Frame, and Ioannidou (2011) reveals a positive correlation between collateral pledged and ex-post non-performing loans (NPL) in Bolivia from 1998 to 2003, a finding echoed by Jiménez and Saurina (2004) in Spain Conversely, Berger and Udell (1990) highlight that borrowers who pledge collateral are often riskier, as explained by Leitner (2006), who attributes this to banks' collateral requirements for higher-risk borrowers In contrast, John, Lynch, and Puri (2003) examine the yield differences between secured and unsecured loans in the U.S., concluding that secured loans yield higher returns, indicating that borrowers who pledge collateral may be more efficient Additionally, Kugler and Oppes (2005) explore the role of collateral in group lending within developing countries, finding that individuals use collateral to mitigate the risk of loan defaults in joint borrowing situations.
(2011)arguethatthediversifiedfindingsaboutthisrelationship arisefromthevariationo f data sampleswhichi n c l u d e differenttypesandcharacteristicso f collaterals.Moreover,p re vi ou s pa persinvestigateonlytheimpacto f collateralso n N P L bycomparingt h e defaultr i s k ( p r o b a b i l i t y ofdefault)betweensecuredl o a n s andu n s e c u r e d l o a n s T o myknowledge,thereisveryli mitedworkontheimpactofdifferentcollateraltypesandcharacteristicso n d e f a u l t r i s k Bergere tal.
( 2 0 1 1 ) f i n d t h a t l i q u i d collateralsd e c r e a s e thep r o b a b i l i t y o f defaultw h e n comp aredt o n o n - l i q u i d collaterals.H o w e v e r , p r e v i o u s p a p e r s m a i n l y focuson loans forcompanies/enterprisesratherthanindividualandconsumerloans.
InVietnam,baddebthasincreasedsharplysince2011andstillbeenseriousuntilnow.Aswecanseein Figure1and2,NPLratiohasrisenfrom4.08%inDec-2012upto4.67%inApril-
2016.However,thisdoesnotrepresentanimprovementinloanqualityofbanksbutduetobank s’s w i t c h i n g too t h e r assettitlesi n t h e balanceshe et t o h i d e badd e b t s T h e VietN a m Ass etsManagementCompany(VAMC)wasestablishedonJuly-
% o f t h e s e baddebtswerecollected(VAMC,2016).Hence,thesepurchasedbaddebtshelptored uceNPLofbanksbutt h e y werenotcollectedinrealityandstillharmthewholeeconomy.Further more,manybaddebtshavebeenrestructuredbutstillclassifiedasnormaldebtsinsteadofbaddebtinal mostVietnamesebanks(thisproblemispermittedbytheStateBankofVietNam)andthereforet h e s e baddebtswerehidden.
Figure1:NPLrate ofVietNamforthe period fromDec-2012to Jun-2013
Source:StateBankof Viet Nam inhttp://tapchitaichinh.vn/
Figure2:NPLofViet Nam forthe periodfromJun-2014 toDec-2015
(2014)arguethatlendinginlowincomecountriesisnotoriouslyriskybecauseo f informatio nasymmetryproblemwhicha r e highi n d e v e l o p i n g countries Un i t e d NationsConference onTradeandDevelopment-
In Vietnam, the banking sector faces challenges such as a lack of credit management skills and underdeveloped financial intermediation, exacerbated by a complex regulatory environment and a large informal cash-based economy Most loans in Vietnam are collateralized due to information asymmetry, with banks prioritizing collateral over borrowers' payment ability in practice As a result, loans are 100% secured, and the difference in non-performing loan (NPL) rates among banks is influenced by the quality of their screening processes More efficient banks focus on screening quality and use collateral primarily to enhance borrowers' repayment incentives, while smaller banks may compromise screening quality, relying heavily on collateral as protection against loan losses Collateral plays a crucial role in NPL control, particularly for smaller banks However, the relationship between collateral requirements and the ability to distinguish between optimistic and realistic borrowers remains unclear, as both types may accept collateral conditions to secure lower-cost loans Additionally, the liquidity of collateral types, such as real estate and vehicles, varies and impacts the probability of default (PD) on loans Therefore, a thorough investigation into the impact of different collateral characteristics on PD in the Vietnamese context is necessary.
Researchobjectivesandresearchquestions
Theobjectiveofthispaperistoexaminetheimpactofcollaterals’liquiditycharacteristicsonl o a n s ’ probabilityofdefault (PD)atthecommercialbanks inVietnam.
ResearchMethodologiesand Data
Thispaperappliest h e logitm o d e l t o e x a m i n e t h e responseso f d i f f e r e n t l i q u i d i t y levelso f collaterals,loansamountandranksofprotectionratesofloansonPDofpersonalsloans.Allpredicto rsarecategoricalvariablesandtheresponsetakesonlyoneoftwocategoriesatthesametime:defa ultandnon -default.
Dataofthis researchis collectedfrominternalloans accountdatasourceofa me di um size Viet namesebank.Loansaccountsarefirstgeneratedin3yearsas2010,2011,2012andfrombusinessunits placedinHoChiMinhcityandHanoi.Inordertoinvestigatethedirectimpacto f collateralsonloansd efault,otherfactorsthatpotentiallyaffectloan’sPDareminimizedbycollectingloanaccountsthatare securedbyonlyonecollateralatonepointoftimeduringtheperiodfrom2010to2012.Hence,2,295o bservationsareincludedintheresearch’sempiricalanalysis.
ResearchContribution
Thisstudyprovidesempiricalevidenceabouttheimpactofcollaterals’liquiditycharacteristicso n t h e PDo f personall o a n s i n a V i e t n a m e s e bank.Differentfrompreviousresearches,thisthesist riestofindouthowPDofloansresponsetovariousliquiditylevelsofcollateralsincasesoffullypro tectedloanswhilepreviousstudiesonlyfocusondifferenceinP D betweensecuredandunsecuredl oans.Dataofthisthesisarecollectedfrominternaldatao f onebankinVietNamanddifferfromother researchesofwhichdataalmostwereprovidedbyNationalCrediti n f o r m a t i o n centerso f o t h e r countrieso r collectedf r o m questionnaire.Furthermore,tomyknowledge,therehavebeenlimi tedstudiesabouttherelationshipbetweencollateralcharacteristicsandl o a n d e f a u l t r i s k hasn o t b e e n w i d e l y studiedi n VietNam.Oneofthemostimportantreasonsisweakcreditinformationinf rastructureand
5 ineffectivepublicrecordsasmentionedaboveandthereforemakesthedatacollectioncostly.Howe ver,banksmayh a v e t h e i r o w n researchesaboutt h i s t o p i c b u t d o n o t p u b l i s h d u e t o inform ationprivacy.
NegativeresponsesofhighliquiditylevelonPDarefoundinthispaper,stronglysupportingt h e dominanceofborrowerselectionandriskshiftingeffectinthisbank.Thehigherliquiditylevelso f c ollaterals,t h e lowerp r o b a b i l i t y ofdefaulto n i n d i v i d u a l l o a n s Fromt h i s result,i m p r o v i n g t h e s c r e e n i n g q u a l i t y i n c a s e o f l o w e r l i q u i d collateralsi s suggestedf o r investi gatedbank Moreover,status ofhiddensubprimeloans existenceiswarnedtopolicy maker s.Therefore,somerecommendationsaresuggestedfortheStatebankofVietNaminordertopre ventsevereloan loss ifassetspriceshavedevaluated.
Structureofthesis
Chapter1 introducesaboutt h e backgroundandm o t i v a t i o n o f t h i s research.Inconsistente mpiricalevidencea b o u t t h e impacto f collateralso n P D i s s h o w n a n d t h e c a u s e o f t h i s i n c o n s i s t e n c y isdiscussedshortly.BaddebtsituationinVietNamfrom2011untilnowisalsopr esentedto show theneedandmotivation of aresearchaboutdeterminants ofbaddebts.
Chapter2willreviewthetheoreticalandempiricalliteraturesabouttherelationshipbetweencollate ralsandnon- performingloans(NPL).Briefexplanationoftheinteractionsof4channelsthroughwhichcollat eralsaffectloandefaultisdiscussedin thispart.
Researchmethodologyanddatawillbepresentedclearlyinchapter3.Briefexplanationaboutt h e m eaningandsuitabilityoflogitmodelforthisresearchanddatacollectionisdiscussed.T h i s cha pteralsoshows thelimitation ofdatasource.
Chapter4discussesempiricalresultsandchapter5summarizestheresearch’smainfindingsfromwhi ch policyimplicationissuggested.
Loan risk (probability of default)
Theoreticalreviewofrelationshipbetweencollateralsandloanrisks
Therea r e t w o s t r a n d s oftheoriest h a t e x p l a i n differenteffectsofcollateralrequiremento n l o a n risk,exantandexposttheory.Theex- anttheoryinterpretstheborrowerselectioneffectandtheex- postoneexplainsthelenderselectioneffect,riskshiftingeffectandlossmitigationeffect(Bergeretal.2 011).
TheexanttheoryexplainsthenegativerelationshipbetweencollateralrequirementandNPLd u e totheborrowerselectioneffect.Inthischannel,higherqualityborrowerstendtopledgem o r e l i q u i d collateraltotakelowerinterestratesonloansthankingtolowerscreeningcost.In
7 thiscase,decisionofbanksinapprovingloansisbasedonsignalingwhichmeansthatbanksobserve behaviorofborrowersbetweensecuredandunsecuredloansinordertoclassifytheq u a l i t y o f borrowers(JaphetandMemba,2 0 1 5 ) A n d accordingt o Bergeretal(2009)i n J a p h e t and Memba(2015),thisexanttheoryisonlyapplicableincasesofshortrelationshipbetweenborrower sandlenderswhichimpliesahighlevelofasymmetryinformationbetweent w o parties.
Thechoiceofpledgingcollateralofborrowersisbasedontheexpectationofavoidingscreeningcostof banksandthereforelowerinterestrate.Thescreeningprocedureofbanksisc o s t l y ando b v i o u s l y t h i s costw i l l b e accountedi n t h e interestrate.Manoveetal.
In 1998, it was argued that banks screen all projects but only fund those deemed good, charging an interest rate that reflects the cost of funds, the screening costs of approved projects, and a prorated share of the screening costs for unapproved projects Banks often lose their screening effectiveness when collateral is highly protected Consequently, high-quality applicants tend to offer collateral to bypass screening, resulting in lower interest rates This borrower selection effect suggests a negative relationship between collateral pledging and loan risk, indicating that a higher quantity and liquidity of collateral are linked to a lower probability of default However, the strength of this effect may diminish due to the optimism of certain borrowers.
Wishful thinking can lead to biased perceptions among borrowers, as noted by Manove and Padilla (1999) DeBondt and Thaler (1995) highlight that a significant finding in the psychology of judgment is the tendency for individuals to be overconfident This overconfidence is particularly evident in optimistic borrowers, who often exaggerate the efficiency of their projects while underestimating the likelihood of defaulting on loans Consequently, when collateral requirements ease project approvals and lower interest rates, these optimistic borrowers tend to pledge more collateral However, this practice can diminish economic and social welfare by diverting resources into low-quality projects, as indicated by Manove and Padilla (1999) Therefore, while secured loans might suggest a lower probability of default due to high-quality projects and responsible borrowers, this does not apply to optimistic borrowers.
Good borrowers prevent this cost by pledging collaterals
Bank do not screen if fully protected do not pay this S
Secured loans can indicate a lower quality of borrowers and projects, primarily due to the lender selection effect Banks possess an advantage in assessing projects and distinguishing between reliable and optimistic borrowers To mitigate loan risk, they often require low-quality borrowers to pledge additional collateral However, the screening process employed by banks remains unobservable, and their decision to screen projects depends on the profitability of the information gained versus the costs incurred According to Manove et al (1998), if the benefits of screening do not outweigh the costs, banks are unlikely to undertake the screening process The expected loss equation, L = (1 - PH) * (R - K), illustrates this decision-making framework, where S represents screening costs, PH is the probability of success for a good borrower and project, R is the original loan amount, and K is the loan amount secured by collateral.
PH).Therefore,bankswillscreenifS0whichmeansborrowergethigherutilityin caseofdefault.y* it takestheform:y* it =α + x’ it β +z’ t γ + w’ i Ω+hcm+ ε it x’it:Liquiditylevelsof collaterals Thisis themainexplanationvariable. z’t:setofotherexplanatoryvariablessuchas:interestrates,loanstimes,protectedlevelsofl o a n s , l o a n s sizes,ownership. dummy variablefor theyear in thateachloan iscreated. w’ i : controlvariablesfortime,professionandregionsfactors.
Callyitisdecisioni n defaulto f borroweri w h o geta l o a n i n yeart yitisanendogenousvariableandd i c h o t o m o u s , whereyit=1 i f t h e l o a n i s delinquent(y*it>0)and0 otherwise(y*it0/ (xit,zt))=F(α+x’itβ +z’tγ +w’iΩ +hcm+ εit)
WhereProb(yit=1/(xi,zt))istheprobabilityofdefault(PD)oftheloaniwhichisgeneratedi n yeart.
Because dependentvariabletakesthe valueonlyfrom 0 to 1 whiletherighthand sidecantakea n y v a l u e , o d d s ratiosarecalculatedandtransformt o logiti n ordert o removet h e valuerestrictionof thelefthandside.Themodel now is:
OutcomeofthemodelnowisthepredictionofthechangesinPDcomparedtoProbabilityof notdefaultofaloanwhenitspredictorsvary,especiallyl i q u i d i t y l e v e l s ofcollaterals.Margi naleffectis calculatedasMG =Pr(y=1|x,xk=1)–Pr(y=1|x,xk=0)
Data
DataforthisthesisiscollectedfromtheinternalloanaccountsofamediumsizebankinVietNam.Diff erentiatingfrommanypreviouspaperswhichfocusonenterprisesloans,thispaperaimsati n d i v i d u a l borrowers.D a t a i s collectedino n l y fromo n e bankd u e t o t h e lackandd i f f i c u l t y in collectingdataaboutloansaccountsin all banks in VietNam.
(i) loans werefirstgeneratedinthe years2010, 2011,2012 andmaturedbeforetheday03/31/2 016;(ii)loanshaveshortandmediumterms(morethan12monthsto
60m o n t h s ) b u t durationo fl oa ns t i m e i s atleasto n e year,
( i v ) eachloanaccountissecuredbyonlyoneassetandthisassetisusedasacollateralforthatloanacc ountateachpointoftime,(v)locationswhereloansareappliedandapprovedareHo Chi Minh cityandHaNoi.
Thedatasampleincludes2,295observationswhichsatisfytheabovefilteredcriteria.Inthissampl e,m a j o r i t y o f professionso f borrowersareworkers,employeeso f enterprises,businessmani n s ervicesectorandtraders,thereforeb o r r o w e r s ’ professioni s groupedi n t o nonbusiness,servicea ndtrade.Loansofborrowerswhoseprofessionsarebelongtogroup2(service)and3 (trade) willcompareto thosebelongtogroup1 in theimpact on PD.
Collateralsarec l a s s i f i e d basedo n t h e internalpolicy’sr a n k i n g o f b a n k aboutcollateral s(Regulationn u m b e r 1782/2011/QĐ-TGĐand591/2012/QĐ-
TGĐ).Accordingly,t h e collateralsarerankedinto11levelsbasedontheliquidity ofcollateralsa nddiscountedrisklevelo f collateralvalue.Highestleveli s 1 1 t h a t is t h e m o s t l iq ui d anddesirab lecollateral.Datacollecteddonotincludeloansthataresecuredbydepositsissuedbyinvestigated bankandthereforerank11collateralsdonotexistinsample.Collateralsofwhichlevelsarefrom2t o 1
0 a r e collected.However,level2 , 3 , 4 w i l l b e combinedi n 1 groupbecauseo f f e w observatio nsi n l e v e l 2 and4 Level9 , 1 0 arealsocombinesast h e samereason.Consequently,t h e r e a r e 6 levelso f l i q u i d i t y ofcollateralsi n sample.M o r e detailaboutcollateralsclassificationis below:
Therea r e 4 r a n k s c o r r e s p o n d i n g t o 1 1 levelso f collateralliquiditiesaccordingt o abov eregulationsof investigated bank:
RankA(A1,A2,A3): most easilytransferinto cashandlowestdiscountedrisk.RankB(B1,B2):easilytransfer into cashandlowdiscountedrisk.
RankD(D1,D2)andrankE(E1,E2):HigherriskinliquidityanddiscountedthanC.Theset w o ty pesofcollateralsappearrarelyinindividualloans ofinvestigatedbank.
Equal1if aloanis downgradedtothe rate3- 5(overdue morethan90days)orloanisoverduemoreth an90daysanytimeafterorigination.
Thisi s ordinalvariablet h a t t a k e s valuef r o m 1 t o 6,co rrespondingt o i n c r e a s i n g l y l i q u i d i t y l e v e l s o f c ollaterals
Ownerofcollateralistheborrowerorfamilymemberso f t h e borrower.T h i s i s a d u m m y variablew h i c h i s e qual1 if theowner is notborrowerand0otherwise.
6 Loantime Numbero f m o n t h s betweenl o a n o r i g i n a t i o n d a y a ndma tu ri ty day.
8 Middle-term Dummyvariablewhichtakesvalueof1ifloantimeism o r e tha n 12 months, 0 if not.
Adummyvariabletakingthevalueof1ifratioofloansamou ntoncollateralsvalueisequaltoorlessthan50%and0otherw ise.
10 Protect75 Ad u m m y variabletakingt h e valueof 1i f ratioo f
21 loansamountoncollateralsvaluevariesfromequalandhigherth an 50%to 75%and0otherwise.
Ad u m m y variablet a k i n g t h e valueof1 i f ratioo f l o a n s amountoncollateralsvaluevariesfromequalandhigher than 70%and0otherwise.
Threedummyvariablesareconstructedtorepresentthreedi fferentgroupsoflenders’professions:T r a d e - profession,Service-professionandNonbusiness 4
3 year Equal1if loans arefirstgeneratedin2010;equal2 i f t ha t in 20 11 ; equal3 ifthatin 2012.
The model predicts Probability of Default (PD) based on various explanatory variables, revealing that interest rates and loan duration have an insignificant impact on PD Due to strong correlations among these variables, indicated by a high Variance Inflation Factor (VIF), they are excluded from the model Interaction variables related to Housing Construction Market (HCM) and two professions, Trade and Service, are included to enhance model fit and relevance A link test confirms the model's suitability, evidenced by a low p-value of 12.8% for the prediction square Additionally, a Wald test verifies differences among collateral rankings, loan amount groups, and varying protected rates To strengthen robustness, the liquid rank variable is replaced with five dummy variables representing different levels of collateral liquidity, allowing for a refined prediction of PD.
4 Nonbusiness: borrowerswhoclassifiedinthisgroupareworkers,employeesingovernmentagenciesandcompanies.Thisvaria bleisconsideredasbasevariableandthereforebedroppedoutfromregressionmodelinordert o preventmulticollinearityproble m.
DescriptiveStatisticsandPre-estimationtests
Variable Obs Mean Std.Dev Min Max liquidrank 2,295 4.31024 1.162309 1 6 amount(m il.dong) 2,295 678.7174 841.6046 100 30,000 loantime
Table3willsupplyabriefdescriptivestatisticaboutliquiditylevelsofcollaterals,loansizes,l o a n duration,andinterestratei n sample.Meano f l i q u i d i t y levelsi n samplei s 4 3 whichs h o w apr eferableofliquidity4,5,6ininvestigatedbank.Investigatedbankprefershomeandresidentiallandsin urban,urbanizedadjacentareasandsuburbsareaofHoChiMinhcityandHanoi,andalsobrandnewca rs.Meanofinterestrateis21.9%andinterestvariesfrom16%t o 26%.Thisnumberissuitablewit hthemediuminterestrateincreditmarketintheperiod2010-
2012.Loans i z e s whicharel e s s t h a n 1 , 0 0 0 m i l l i o n s d o n g dominatei n samplewhenmeanv alueo f l o a n s a m o u n t i s 6 7 8 7 m i l l i o n s dong.M a j o r i t y ofl o a n s i n sampleares h o r t term loans(Meanofloansdurationis25.39).
Mostcollateralsinsamplearebelongtorank5(orliquiditylevel5)with1,166observations,equiva lent5 0 8 1 % o f t h e totaln u m b e r o f l o a n s accountsi n sample.T y p e s o f collaterali n sample arehomeandresidentiallands,agriculturallands,apartments,cars.Theseareregularassetsthatu suallypledgedbyindividuals.AscanbeseenfromTable4,defaultrateishigh
Total 1,856 439 2,295 0.19 forliquiditylevel1,2,3withthevalueof0.35,0.39,0.33respectively.Thisrateislowerfor level4, 5, 6(0.18,0.12,0.14respectively).
Defaultr a t e o f l o a n s w h i c h amounti s m o r e t h a n 1 b i l l i o n s d o n g t o 3 0 b i l l i o n s d o n g i s e x t r e m e l y high,reachestherateof42%.Alsoveryhighdefaultrateof20%happenst oloanso f whichamountv a r y from5 0 0 millionsd o n g to 1 b i l l i o n s dong.T h i s ratei n l o a n s w hichamountsareunder 500millions dongislower(12%).
Almostcollectedloans(2,274/2,295loans#99%)havetheratiobetweenloansamountandcol lateralv a l u e (calledprotectedr a t e i n t h i s t h e s i s ) equalo r l e s s t h a n 7 5 % Forl o a n s t h a t ha veprotectedrateequalorlessthan50%,defaultrateis11%.Thisrateincreasesto27%forl o a n s wit hprotectedratemorethan 50%andlessthanorequalto75%.
Intermsofloanstime,1,324loansaccounts,equivalentto57.69%samplesizeareshortterml o a n s (lesst h a n o r e q u a l 1 yeard u r a t i o n ) D e f a u l t ratei n t h i s l o a n t i m e g r o u p i s highest (24%).Thisratedecreasesin2yearsdurationloansgroup(10%)andincreasesagainin3-
Empiricalresults
Firstly,P D i s predictedbasedo n allfactorsbyregressingt w o m o d e l s u s i n g differentl o a n t i m e variables.Loanstimeinthefirstregressionismeasuredinmonthsbut inthesecondone,d u m m y variable(middle- term)whichr e p r e s e n t s m i d d l e terml o a n s i s usedt o controlf o r differenceinmaturitybet weenloans.Both2modelsshowinsignificantimpactofinterestrateo n P D whichi s d i f f e r e n t fro mmanypreviouspaperssuchasLouzis,VouldisandMetaxas,(2011)andFarhanetal.
(2012).Furthermore,interestfactorsuffersfromseriousmulticollinearityproblem andthereforebedropped out ofmodel.
***p|z| [95%Conf.Interval] liquidrankm i d d l e t e r m amount2 a m o u n t 3 a m o u n t 1 p r o t e c t 7 5 p r o t e c t 1 0
Variable VIF 1/VIF liquidrank 7.62 0.131269 middleterm 2.86 0.349439 amount2 1.75 0.571355 amount3 1.72 0.580475 protect75 2.25 0.443539 protect100 1.05 0.951343 profession
Wecanseestrongmulticollinearityproblemhappensfor themainpredictorliquidrank.Becausemiddletermhasthesecondlevel ofmulticollinearityandin significantrelation,this variableisdroppedoutofmodel.
logitdefaultliquidrankamount2amount3amount1protect75protect100protect
>50i.professioni.yearownership note:amount1omittedbecauseofcollinearitynote:pr otect50omittedbecauseofcollinearityIteration0: loglikelihood=-1120.1469 Iteration1: loglikelihood=-945.28928
Loglikelihood=-929.20644 PseudoR2 = 0.1705 default Coef Std.Err z P>|z| [95%Conf.Interval] liquidrank amount2 a m o u n t 3 a m o u n t 1 p r o t e c t 7 5 p r o t e c t 1 0
Variable VIF 1/VIF liquidrank 4.13 0.241896 amount2 1.64 0.610917 amount3 1.59 0.626987 protect75 2.12 0.472023 protect100 1.05 0.953686 profession
Loglikelihood=-925.01698 PseudoR2 = 0.1742 default Coef Std.Err z P>|z| [95%Conf.Interval]
Theprediction_hati s significantat1 % whichs h o w s t h e s u i t a b i l i t y o f t h e u s i n g m o d e l However,thepredictionsquare_hatsqisalsosignificantat1%.Thisimpliesaspecification errorofthemodel.Ashcmand2professionsvariablesaresignificant,interactioneffectmayappearsi n m o d e l A d d i n g 2 interactionvariablest h a t arehcmtradea n d hcmservicei n t h e m o d e l , t h e n , regressingthemodelagain.
logitdefaultliquidrankamount2amount3amount1protect75protect100protect
>50i.professionhcmtradehcmserviceownershipi.year note:amount1omittedbecauseofcollinearitynote:pr otect50omittedbecauseofcollinearityIteration0: loglikelihood=-1120.1469 Iteration1: loglikelihood=-913.15293
Loglikelihood=-891.95505 PseudoR2 = 0.2037 default Coef Std.Err z P>|z| [95%Conf.Interval] liquidrank amount2 a m o u n t 3 a m o u n t 1 p r o t e c t 7 5 p r o t e c t 1 0
Loglikelihood=-890.80362 PseudoR2 = 0.2047 default Coef Std.Err z P>|z| [95%Conf.Interval]
Nowpredictionsquareisinsignificantat5%.Hence,themodelnow is suitableandmeaningful.
Variable VIF 1/VIF liquidrank 4.32 0.231384 amount2 1.64 0.610823 amount3 1.66 0.601216 protect75 2.14 0.467259 protect100 1.05 0.952621 profession
7 Subtituteliquidrankvariableby5dummyvariableswhichrepresentfor6ranks ofl i q u i d i t y ofcollaterals.Rank 5 will beomittedfrom themodelandtheresultwill showt h e differrencein PDbetweeneachof 5ranksandrank5.
logitdefaultrank1rank2rank3rank4rank6rank5amount2amount3amount1pro
>tect75protect100protect50i.professionhcmtradehcmserviceownershipi.year note:rank5omittedbecauseofcollinearitynote:amou nt1omittedbecauseofcollinearitynote:protect50om ittedbecauseofcollinearityIteration0: loglikelihood=-1120.1469 Iteration1: loglikelihood=-911.98164
Loglikelihood=-890.40995 PseudoR2 = 0.2051 default Coef Std.Err z P>|z| [95%Conf.Interval] rank1 r a n k 2 r a n k 3 r a n k 4 r a n k 6 r a n k
Loglikelihood=-889.23693 PseudoR2 = 0.2061 default Coef Std.Err z P>|z| [95%Conf.Interval]
Variable VIF 1/VIF rank1 1.05 0.949934 rank2 1.15 0.865966 rank3 1.48 0.675975 rank4 1.31 0.763534 rank6 1.19 0.842713 amount2 1.65 0.606421 amount3 1.72 0.579812 protect75 2.06 0.485412 protect100 1.05 0.951062 profession
8 Replacing2dummyvariableswhichrepresent2 group ofloansamountbycontinuousvariablenamed“sizes”.
logitdefaultrank1rank2rank3rank4rank6rank5sizesprotect75protect100
>protect50i.professionownershipi.year note:rank5omittedbecauseofcollinearitynote:prot ect50omittedbecauseofcollinearityIteration0: loglikelihood=-1120.1469 Iteration1: loglikelihood=-935.72176
Loglikelihood= -914.5056 PseudoR2 = 0.1836 default Coef Std.Err z P>|z| [95%Conf.Interval] rank1 r a n k 2 r a n k 3 r a n k 4 r a n k 6 r a n k
logit,or note:rank5omittedbecauseofcollinearity note:protect50omittedbecauseofcollinearity
Loglikelihood=-914.5056 PseudoR2 = 0.1836 default OddsRatio Std.Err z P>|z| [95%Conf.Interval] rank1 r a n k 2 r a n k 3 r a n k 4 r a n k 6 r a n k
Variable VIF 1/VIF rank1 1.06 0.944015 rank2 1.12 0.894349 rank3 1.32 0.756533 rank4 1.30 0.769681 rank6 1.17 0.854506 sizes 1.89 0.528209 protect75 1.95 0.513305 protect100 1.08 0.928767 profession
(1)[default]rank1-[default]rank2=0 chi2(1)= 2.00
(1)[default]rank1-[default]rank3=0 chi2(1)= 7.42
(1)[default]rank1-[default]rank4=0 chi2(1)= 10.72
(1)[default]rank1-[default]rank6=0 chi2(1)= 15.71
(1)[default]rank2-[default]rank3=0 chi2(1)= 1.75
(1)[default]rank2-[default]rank4=0 chi2(1)= 3.92
(1)[default]rank2-[default]rank6=0 chi2(1)= 8.09
(1)[default]rank3-[default]rank4=0 chi2(1)= 1.07
(1)[default]rank3-[default]rank6=0 chi2(1)= 4.97
testliquidrankamount2amount3protect75protect100hcmtradehcmservice
(1)[default]amount2-[default]amount3=0 chi2(1)= 21.11
(1)[default]protect75-[default]protect100=0 chi2(1)= 6.94
: liquidrank = 4.31024 (mean) amount2 = 2196078 (mean) amount3 = 1834423 (mean) amount1 = 5969499 (mean) protect75 = 5089325 (mean) protect100 = 0091503 (mean) protect50 = 4819172 (mean) 1.profession = 4849673 (mean) 2.profession = 1664488 (mean) 3.profession = 3485839 (mean) hcmtrade = 2623094 (mean) hcmservice = 0976035 (mean) ownership = 1381264 (mean)
Expression :Pr(default),predict() dy/ dxw.r.t.:liquidrankamount2amount3amount1protect75protect1 00protect502.profession3.professionhcmtradehcmserviceowne rship2.year3.year at dy/dx
Err. z P>|z| [95%Conf.Interval] liquidrank amount2 a m o u n t 3 a m o u n t 1 p r o t e c t 7 5 p r o t e c t 1 0
The mean values for ranks indicate that rank 5 is the highest at 0.508061, followed by rank 3 at 0.1869281 and rank 4 at 0.1481481, while ranks 1 and 2 are lower at 0.0248366 and 0.0501089, respectively In terms of amounts, amount 1 leads with a mean of 0.5969499, while amount 2 and amount 3 are significantly lower at 0.2196078 and 0.1834423 For protection levels, protect 75 has a mean of 0.5089325, indicating strong protection, whereas protect 100 is notably low at 0.0091503 Regarding professions, the first profession has a mean of 0.4849673, followed by the third at 0.3485839, and the second profession at 0.1664488 The mean for hcmtrade stands at 0.2623094, while hcmservice is lower at 0.0976035, and ownership has a mean of 0.1381264 Lastly, the mean values for years show the second year as the highest at 0.5198257, followed by the first year at 0.2305011 and the third year at 0.2496732.
Expression :Pr(default),predict() dy/ dxw.r.t.:rank1rank2rank3rank4rank6rank5amount2amount3amount1prote ct75protect100protect502.profession3.professionhcmtradehcmservice ownership2.year3.year at dy/dx
Err. z P>|z| [95%Conf.Interval] rank1 rank2rank3 rank4rank6 rank5amoun t2amount3a mount1prot ect75prote ct100prote ct50 profession