Credit Risk Evaluation Model
-
Upload
mihai-enescu -
Category
Data & Analytics
-
view
88 -
download
4
Transcript of Credit Risk Evaluation Model
1
The Credit Risk
Evaluation Model
By Maria A Canaveras Galdon, Jasdeep Kaur, Ashwani Dua, Ankita Jiwrajka, Shruti Choudhary, Mihai Enescu
2What is Credit Risk?
The risk of loss of principal or loss of a financial reward stemming from a borrower's failure to repay a loan or otherwise meet a contractual obligation.
Credit risk arises whenever a borrower is expecting to use future cash flows to pay a current debt. Investors are compensated for assuming credit risk by way of interest payments from the borrower or issuer of a debt obligation.
The higher the perceived credit risk, the higher the rate of interest that investors will demand for lending their capital. Credit risks are calculated based on the borrowers' overall ability to repay.
3Data UsedVariable Name Description TypeSeriousDlqin2yrs Person experienced 90 days past due delinquency or
worse Y/N
RevolvingUtilizationOfUnsecuredLines
Total balance on credit cards and personal lines of credit except real estate and no installment debt like car loans divided by the sum of credit limits
percentage
age Age of borrower in years integerNumberOfTime30-59DaysPastDueNotWorse
Number of times borrower has been 30-59 days past due but no worse in the last 2 years.
integer
DebtRatio Monthly debt payments, alimony,living costs divided by monthy gross income
percentage
MonthlyIncome Monthly income realNumberOfOpenCreditLinesAndLoans
Number of Open loans (installment like car loan or mortgage) and Lines of credit (e.g. credit cards)
integer
NumberOfTimes90DaysLate Number of times borrower has been 90 days or more past due.
integer
NumberRealEstateLoansOrLines Number of mortgage and real estate loans including home equity lines of credit
integer
NumberOfTime60-89DaysPastDueNotWorse
Number of times borrower has been 60-89 days past due but no worse in the last 2 years.
integer
NumberOfDependents Number of dependents in family excluding themselves (spouse, children etc.)
integer
4Trained Data
Test Data
Apply Discriminant Function Analysis
Apply Neural Networks
Apply Neural Networks
Apply Discriminant Function Analysis
Apply Logistic Regression Model
Risk rate of each Candidate
Maximize Return of Organization and find loan amounts
5
Whom to deny the loan after preliminary screening?
6
Artificial Neural network
7What is ANN? A neuron is a special
biological cell that processes information.
Receives signal from one neuron.
Transmits information to another neuron.
ANN is inspired from biological neurons.
Consists of extremely large number of processors with many interconnections.
Uses:- Function Approximation- Pattern Classification- Prediction/Forecasting
8Architecture Inputs represented by “i”.
Inputs connected to neuron using weights(w)
Output at neuron:function using SumProduct
(inputs,weights)
ACTIVATES the neurons.
Decided based on output.
In our model we need an output of 0 or 1.
Function =>Logistic Sigmoidal
Where, x = w1i1 + w2i2 + w3i3
Bias: y = mx + c
Activation Function
9Implementation in Excel - Solver Set of inputs interact to produce the
output.
The output can be 0 [ Not Default] or 1 [Default].
7 input variables and single output variable.
ANN used to predict the output [0,1].
Steps
Create an ANN- 7 inputs, 6 hidden neuron layer and 1 output.
Select the Activation function.
Optimize for the weights using Solver
Minimize the error.
Error Function
E= [Expected – Actual]2
Minimize the sum of these errors.
10
ANN - Solver Demo
ANN_test_modelANN_Training_model
11Discriminant AnalysisDiscriminant analysis is a
statistical tool used to clustered together members of a population because of similar
characteristics. the clusters (usually called groups) are
predefined
Steps
Create Score function = w1*p1+w2*p2+…+wn*pn
Get results of Score function for all the samples
Compare results of score function with a cutoff value. If sample score function result is above cutoff, samples will be classified in group one, otherwise group two
Decision variables will be weights and cutoff value.
Objective Function
Objective Function = Correct Classified/Total Samples
Maximize percentage of Correct Samples Classified. Evolutionary Solver used
Implementation in Solver
12Discriminant Analysis Sensitivity Analysis
Biased Data
Algorithm Improvement Possibility - Really good Performance, however possibility to improve detection of positive cases
Actual/Expected
1 0
1 3 0
0 185 2812
Steps
Modify model adding a Constraint for % of positive correctly classified.
Still objective function still is to maximize Total % of correct classifications
Run solver table to see maximum total classified when varying the % of positive correctly classified
0.69
9999
9880
7907
1
0.71
9999
9690
0558
5
0.74
0000
0095
3674
3
0.75
9999
9904
6325
7
0.77
9999
9713
8977
1
0.80
0000
0119
2092
9
0.81
9999
9928
4744
3
0.83
9999
9737
7395
6
0.86
0000
0143
0511
5
0.87
9999
9952
3162
8
0.89
9999
9761
5814
20.000.400.80
Sensitivity Positive vs. Total Correctly Classified
% Positive Correctly Classified
% Total Correctly
Classi-fied
Class Prob
1 93.3%
0 6.6%
Implementation in Solver Table
13
Discriminant Analysis - Solver Demo
Microsoft Excel Worksheet
1414
TRAINING TEST
% Correct % Correct
93.3% 93.83
TRAINING TEST
% Correct % Correct
94.3 90%
Actual/Expected 1 0
1 10 106
0 194 2690
Actual/Expected 1 0
1 3 0
0 185 2812
Discriminant Analysis Neural NetworksV/S
15
What credit interest to charge??
How much loan to approve??
16Logistic Regression
Logistic Regression is a classification technique
The dependent variable which is in binary for (0 and 1)
Independent variables can be continuous or categorical
The logistic function provides logit Log (odds of event
happening)
The logit is used to calculate the probability of an event happening
17
Dependent Variable
Person denied the loan (Y/N?)
Independent Variable
Age
Debt Ratio
Revolving Utilization Of Unsecured Loans
Number Real Estate Loans Or Lines
Number Of Open Credit Lines And Loans
Number Of Time loan due more than 30_59 DaysNumber Of Time loan due more than 60_89 DaysNumber Of Times loan due for more than 90 Days
Logistic Regression- Our Model
18Methodology
Combined DATA from Discriminant Analysis
0: Loan Accepted1: Loan Denied
Test Trained
Run Logistic Regression
Using SAS
Logit value varied with income Divided applicants into income quartile and
ran separate logistic regression Record logit value for each variable and
each income band
Step I Step II
Probability of loan rejected Probability of Default
19Classification into Credit RatingProbability Non-
DefaultProbability
DefaultCredit
RatingsInterest Rates
84% 15.5% D 10.00%
91% 9.1% D+ 9.75%
94% 5.7% C 9.50%
96% 4.1% C+ 9.25%
97% 3.2% B 9.00%
97% 2.6% B+ 8.75%
98% 2.2% A- 8.50%
98% 1.8% A 8.25%
99% 1.5% A+ 8.00%
20
Credit Risk and Return Model Demo
Credit Risk and Return Model
21How much loan should be accepted?
Objective
Constraints
Decision Variable
Maximize total return= ∑(interest earned on loans approved)
Amount of loan accepted for each candidate
• Loan Approved < Loan Applied (For each candidate)• Loan Approved < Annual Income (For each
candidate)• Risk Taken*** < Maximum risk rate defined by
management• ∑ (Loan Approved ) < Funds Available to distribute
***Risk taken= ∑ (Probability of default X Loan Approved ) ∑ (Loan Approved )
22
Do I need any additional variables in calculating the credit rating ??
How can respond to instant line of credit increase??
Can I use the same model in new emerging countries ??
23Accommodate Additional Variable
Current Model is based on the numerical variables
Bank’s loan form collects Numerical variables Categorical variables Comments providing additional
information about the variables
Categorical Variable
Analyze Additional Variables
Assign Codes to Categorical Variables. Own Home – 1, Rented - 0.5
Are there any numerical variable that need code assignment - Number of Dependents ?
Include the Code assigned Categorical Variables in the current models
Comments / Explanation
How many potential customers have provided comments? Are codes generic or specific to a variable ?
Is any Public Information involved?
Run Text Analyzer on the comments
Assign Codes based on the historic data and include them in current model
Additional Steps
Prepare the Data
Review Codes regularly
Explore use of web services to access data from public info sites.
Benefits
Realistic, Dynamic automated Credit rating
Reduction in Cycle Time & Disputes !
Optimum Portfolio Selection !!
2424
Approve Disapprove V/S
Real time Credit Line increase Approval International Credit Line Request due to medical charges
Current Model is based on the data points for bank loan application
Accommodate credit card application data point
Run suggested models with the credit card information for Training and Test
Calculate Bias Decide on the additional Utility Factor for
credit card based on Bias Model is available to use.
Steps
Bank receives a request
Run past payment history through the model
Has customer updated bank on travel plans. No -Analyze Travel History from the model – location of past payments. Is he a frequent traveler .
Analyses type of request – Medical or Luxury ??
Run Logit Regression to Approve or Disapprove
Benefits
Dynamic, Instant credit line approval to increase the utility function for the credit card
25Use Model to Approve Loans in Emerging Market
Application of the Model
Is Data Available for Baseline?
Based on the financial history of the country , readjust the variables, weightage , codes .
Run Training Test and Training sets . Calculate Bias.
Confirm Clusters for Discriminant Analysis, Calculate Cut off and confirm on the eligibility?
Run Logit Risk and return Model to calculate the probability of Default ranges
Benefits
Comparative Analysis with home Country to design new products
Shorter Cycle Time
Reduction in Rate with Automation of the calculation and Batch processing.
Increase in Regulation
Intensifying competition
Increasing costs
Source: http://www.ey.com/Publication/vwLUAssets/EY_-_Banking_in_emerging_markets:_Investing_for_success/$FILE/EY-Banking-in-emerging-markets-Investing-for-success.pdf
26
Q & A
THANKS !!