Upload
tess-brow
View
222
Download
1
Tags:
Embed Size (px)
Citation preview
Do Demographics Do Demographics Predict Predict
Creditworthiness?Creditworthiness?Presented by Kelli JonesPresented by Kelli Jones
ECON 616ECON 616
April 2, 2003April 2, 2003
IntroductionIntroduction
What is a What is a credit scorecredit score ? ? Measure of relative creditworthiness / Measure of relative creditworthiness /
credit performancecredit performance Based on items from credit history such Based on items from credit history such
as bankruptcies, delinquent payments, as bankruptcies, delinquent payments, revolving credit balancesrevolving credit balances
IntroductionIntroduction
How is a credit scoring system built?How is a credit scoring system built? It is determined how effective each risk It is determined how effective each risk
characteristic is in predicting credit characteristic is in predicting credit performanceperformance
Each element is given a weight depending on Each element is given a weight depending on that effectivenessthat effectiveness
The combination of each element and weight The combination of each element and weight results in the best predictor of credit results in the best predictor of credit performanceperformance
Generally, the higher the score, the better Generally, the higher the score, the better your credityour credit
IntroductionIntroduction
How are credit scores used?How are credit scores used? Credit applicationsCredit applications Mortgage loan applicationsMortgage loan applications Insurance underwriting and/or pricing Insurance underwriting and/or pricing
for personal auto and homeowners for personal auto and homeowners policiespolicies
Purpose of ResearchPurpose of Research
To test whether certain demographic To test whether certain demographic groups have a tendency to have groups have a tendency to have worse credit (i.e. lower credit worse credit (i.e. lower credit scores)scores)
Literature ReviewLiterature Review
Avery, Bostic, Calem, Avery, Bostic, Calem, CannerCanner
(1996, 2000)(1996, 2000) Data obtained from Equifax on 3.4 Data obtained from Equifax on 3.4
million individuals making up 2.5 million individuals making up 2.5 million householdsmillion households
incomeincome: : 33% of households in lowest income range 33% of households in lowest income range
have low credit scores, compared to 23% of have low credit scores, compared to 23% of households overall and 17% of households households overall and 17% of households in the highest income range in the highest income range
As median family income ↑, median credit As median family income ↑, median credit score ↑score ↑
RaceRace: : as the %age of minority households ↑, as the %age of minority households ↑,
median credit score ↓median credit score ↓ EducationEducation::
As the %age of high school graduates As the %age of high school graduates ↑, median credit score ↑↑, median credit score ↑
LocationLocation:: No statistically significant relationship No statistically significant relationship
shown between credit scores and shown between credit scores and urban/suburban/rural classificationurban/suburban/rural classification
AgeAge:: As the median age ↑, median credit As the median age ↑, median credit
score ↑score ↑
Kennickell, Starr-McCluer, Kennickell, Starr-McCluer, SuretteSurette(2000)(2000)
Comparison of family finances from data Comparison of family finances from data obtained from 1995 and 1998 Survey of obtained from 1995 and 1998 Survey of Consumer FinancesConsumer Finances
1998 survey samples 4,309 households1998 survey samples 4,309 households IncomeIncome::
As income ↑, the # of payments 60+ days As income ↑, the # of payments 60+ days past due ↓past due ↓
AgeAge:: As age ↑, the # of payments 60+ days past As age ↑, the # of payments 60+ days past
due ↓due ↓
Fair, IsaacFair, Isaac(1997)(1997)
Develops and markets credit scoring Develops and markets credit scoring systemssystems
Provided research paper in response Provided research paper in response to concerns that the use of credit to concerns that the use of credit scores results in unfair treatment to scores results in unfair treatment to low-to-moderate-income (LMI) and low-to-moderate-income (LMI) and high-minority area (HMA) high-minority area (HMA) populationspopulations
IncomeIncome:: At a given credit score, the level of risk is At a given credit score, the level of risk is
the same regardless of incomethe same regardless of income RaceRace::
Distribution of credit scores differs Distribution of credit scores differs between HMA and non-HMA populationsbetween HMA and non-HMA populations
For HMAs, 25.3% have scores < 620 For HMAs, 25.3% have scores < 620 compared to 13.8 % for non-HMA’scompared to 13.8 % for non-HMA’s
At any given score, the odds (ratio of At any given score, the odds (ratio of good to bad accounts) are lower for good to bad accounts) are lower for HMA’s; however, this difference seemed HMA’s; however, this difference seemed to be significant only at lower scoresto be significant only at lower scores
DatabaseDatabase
1998 Survey of Consumer Finances1998 Survey of Consumer Finances Complete sample is 21,525 Complete sample is 21,525
observationsobservations Reduced sample used for my Reduced sample used for my
analysis of those who have applied analysis of those who have applied for credit in the last 5 years consists for credit in the last 5 years consists of 13,664 observationsof 13,664 observations
Description of VariablesDescription of Variables
Creditworthiness / credit scoreCreditworthiness / credit score:: Y = 1 if credit denied or approved for Y = 1 if credit denied or approved for
lower amount based on credit history lower amount based on credit history Y = 0 if approved for full amount or Y = 0 if approved for full amount or
denied for reasons other than credit denied for reasons other than credit history history
LocationLocation: : No urban/suburban/rural classificationNo urban/suburban/rural classification 9 categories describing area of country 9 categories describing area of country
(e.g. New England, Midatlantic)(e.g. New England, Midatlantic) Not available in 2001 public datasetNot available in 2001 public dataset
EducationEducation:: 4 dummy variables to capture years of education4 dummy variables to capture years of education
High school diplomaHigh school diploma 1 – 3 years college1 – 3 years college 4 years college4 years college Graduate schoolGraduate school
Having less than high school diploma is base Having less than high school diploma is base casecase
RaceRace:: 3 dummy variables3 dummy variables
BlackBlack HispanicHispanic Asian / Native American / Hawaiian / otherAsian / Native American / Hawaiian / other
White is base caseWhite is base case
IncomeIncome:: Continuous variableContinuous variable
Age:Age: Continuous variableContinuous variable
Frequency TablesFrequency Tables
VariabVariablele
DescriptiDescriptionon
FrequeFrequencyncy
%%
YY CreditCredit
00 goodgood 1187611876 86.986.911
11 badbad 17881788 13.013.099
VariabVariablele
DescriptiDescriptionon
FrequeFrequencyncy
%%
EE Yrs. Of Yrs. Of EducatioEducationn
BaseBase < H.S. < H.S. diplomadiploma
1,3121,312 9.609.60
11 1212 3,0263,026 22.122.155
22 1 -3 yrs. 1 -3 yrs. CollegeCollege
3,1983,198 23.423.400
33 4 yrs. 4 yrs. CollegeCollege
3,1223,122 22.822.855
44 Grad. Grad. SchoolSchool
3,0063,006 22.022.000
VariablVariablee
DescriptiDescriptionon
FrequeFrequencyncy
%%
RR RaceRace
BaseBase WhiteWhite 11,44411,444 83.783.755
11 BlackBlack 1,0871,087 7.967.96
22 HispanicHispanic 691691 5.065.06
33 OtherOther 442442 3.233.23
Table of MeansTable of Means
VariaVariableble
Overall Overall MeanMean
Mean Y = Mean Y = 00
Mean Y = Mean Y = 11
E1E1 .221.221 .216.216 .257.257
E2E2 .234.234 .224.224 .304.304
E3E3 .228.228 .235.235 .183.183
E4E4 .22.22 .24.24 .087.087
R1R1 .080.080 .066.066 .169.169
R2R2 .051.051 .05.05 .054.054
R3R3 .032.032 .032.032 .036.036
II 402,414402,414 449,856449,856 87,30387,303
AA 4646 4747 4040
OLS Regression OLS Regression (Linear Probability Model)(Linear Probability Model)
ModelModel
YYi i = = αα + + ββXXii + + εεii
E(YE(Yii) = P) = Pii = P(Y = 1) = P( bad credit) = P(Y = 1) = P( bad credit) = = ααhathat + + ββhat hat XXii
ResultsResults
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
t Valuet Value
InterceptIntercept < H.S. < H.S. diplomadiploma
.230.230 25.0425.04
E1E1 1212 - .078- .078 - 7.10- 7.10
E2E2 1 -3 yrs. 1 -3 yrs. CollegeCollege
- .060- .060 - 5.53- 5.53
E3E3 4 yrs. College4 yrs. College - .125- .125 - 11.45- 11.45
E4E4 Grad. SchoolGrad. School - .178- .178 - 16.18- 16.18
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
t Valuet Value
InterceptIntercept WhiteWhite .116.116 37.0137.01
R1R1 BlackBlack .163.163 15.3615.36
R2R2 HispanicHispanic .023.023 1.771.77
R3R3 OtherOther .031.031 1.931.93
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
t Valuet Value
InterceptIntercept .326.326 32.9432.94
AA AgeAge - .004- .004 - 20.58- 20.58
InterceptIntercept .132.132 45.5245.52
II IncomeIncome - 2.857 E-9- 2.857 E-9 - 3.76- 3.76
Probit ModelProbit Model
ModelModel
ZZi i = = αα + + ββXXii + + εεii
ZZiihathat = = ααhathat + + ββhathatXXii = F = F-1-1(P(Piihat hat ))
PPiihathat = F( = F(ZZiihathat ) where F is the normal ) where F is the normal distributiondistribution
Probability modeled is Y = 1Probability modeled is Y = 1
ResultsResults
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
Chi-Chi-SquareSquare
InterceptIntercept < H.S. < H.S. diplomadiploma
- .738- .738 372.39372.39
E1E1 1212 - .290- .290 37.5537.55
E2E2 1 -3 yrs. 1 -3 yrs. CollegeCollege
- .217- .217 21.8221.82
E3E3 4 yrs. College4 yrs. College - .517- .517 112.43112.43
E4E4 Grad. SchoolGrad. School - .889- .889 270.93270.93
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
Chi-Chi-SquareSquare
InterceptIntercept WhiteWhite - 1.197- 1.197 6088.06088.044
R1R1 BlackBlack .610.610 198.64198.64
R2R2 HispanicHispanic .112.112 3.313.31
R3R3 OtherOther .148.148 3.903.90
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
Chi-Chi-SquareSquare
InterceptIntercept - 1.045- 1.045 5066.95066.900
II IncomeIncome - 0.000- 0.000 120.68120.68
InterceptIntercept - .154- .154 10.1310.13
AA AgeAge - .022- .022 410.12410.12
Logit ModelLogit Model
ModelModel
ZZi i = = αα + + ββXXii + + εεii
ZZiihathat = = ααhathat + + ββhathatXXii = ln (P = ln (Piihat hat / (1 - / (1 - PPiihat hat ))))
PPiihathat = exp( = exp(ZZiihathat) / (1 + ) / (1 + exp(exp(ZZiihathat) )) ) Probability modeled is Y = 1Probability modeled is Y = 1
ResultsResults
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
Wald Wald Chi-Chi-SquareSquare
InterceptIntercept < H.S. < H.S. diplomadiploma
- 1.207- 1.207 338.85338.85
E1E1 1212 - .512- .512 38.1338.13
E2E2 1 -3 yrs. 1 -3 yrs. CollegeCollege
- .380- .380 22.1322.13
E3E3 4 yrs. College4 yrs. College - .938- .938 114.09114.09
E4E4 Grad. SchoolGrad. School - 1.698- 1.698 260.61260.61
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
Wald Wald Chi-Chi-SquareSquare
InterceptIntercept WhiteWhite - 2.034- 2.034 4843.14843.199
R1R1 BlackBlack 1.0831.083 216.12216.12
R2R2 HispanicHispanic .210.210 3.393.39
R3R3 OtherOther .276.276 4.034.03
VariableVariable DescriptionDescription Parameter Parameter EstimateEstimate
Wald Wald Chi-Chi-SquareSquare
InterceptIntercept - 1.678- 1.678 3302.73302.700
II IncomeIncome - 1.55 E-6- 1.55 E-6 99.7199.71
InterceptIntercept - .111- .111 1.591.59
AA AgeAge - .042- .042 395.87395.87
Comparison of ResultsComparison of Results
VariablVariablee
OLSOLS
Sign Sign Signif.Signif.
ProbitProbit
Sign Sign Signif.Signif.
LogitLogit
Sign Sign Signif.Signif.
E1E1 -- XX -- XX -- XX
E2E2 -- XX -- XX -- XX
E3E3 -- XX -- XX -- XX
E4E4 -- XX -- XX -- XX
R1R1 ++ XX ++ XX ++ XX
R2R2 ++ ++ ++
R3R3 ++ ++ ++
VariablVariablee
OLSOLS
Sign Sign Signif.Signif.
ProbitProbit
Sign Sign Signif.Signif.
LogitLogit
Sign Sign Signif.Signif.
II -- XX -- XX -- XX
AA -- XX -- XX -- XX
Comparison of PComparison of Phathat
ECON 616 Comparison.ECON 616 Comparison.xlsxls
EnhancementsEnhancements
Update data to 2001 SCFUpdate data to 2001 SCF Look at multivariate resultsLook at multivariate results Analyze goodness of fit of modelsAnalyze goodness of fit of models