Scribd
Upload
88 views24 pages

Model Building in Creditcard and Loan Approval V

This document discusses credit risk model building. It outlines the typical 4 steps: 1) studying historical customer data to identify causes of default or charge-offs, 2) identifying the most impacting customer attributes, 3) using logistic regression to quantify the impact of each attribute, and 4) using the model to predict probabilities of default for new customers based on their attribute values. An example model is described that uses attributes like income, number of loans, payment history, credit utilization, age, and debt ratio to predict default probability. Other applications of model building discussed include fraud detection, employee attrition prediction, and marketing response modeling.

Uploaded by

mike110*
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
88 views24 pages

Model Building in Creditcard and Loan Approval V

This document discusses credit risk model building. It outlines the typical 4 steps: 1) studying historical customer data to identify causes of default or charge-offs, 2) identifying the most impacting customer attributes, 3) using logistic regression to quantify the impact of each attribute, and 4) using the model to predict probabilities of default for new customers based on their attribute values. An example model is described that uses attributes like income, number of loans, payment history, credit utilization, age, and debt ratio to predict default probability. Other applications of model building discussed include fraud detection, employee attrition prediction, and marketing response modeling.

Uploaded by

mike110*
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Credit Risk Model Building

Venkat Reddy
Note
• This presentation is just the lecture notes for the talk on Credit Risk Model building
conducted for MBA students
• The best way to treat this is as a high-level summary; the actual session went more
in depth and contained other information.
• Most of this material was written as informal notes, not intended for publication
• Please send your questions/comments/corrections to
venkat@trenwiseanalytics.com or 21.venkat@gmail.com
• Please check my website for latest version of this document
-Venkat Reddy
Contents

• Applications of Statistics in Business


• The Model Building Problem
• Credit Risk Model Building
• Other Applications of Model Building
Applications of Statistics in Business
Increasingly, business rely on intelligent tools and techniques to analyze data systematically to
improve decision-making.

 Retail sales analytics


 Financial services analytics
 Telecommunications
 Supply Chain analytics
 Transportation analytics
 Risk & Credit analytics
 Talent analytics
 Marketing analytics
 Behavioral analytics
 Collections analytics
 Fraud analytics
 Pricing analytics
The Problem
Who will run away with my money?
• Citi Bank : Present in more than 90 countries.
• More than 100,000 customers apply for credit
cards/loans every month
• All of them have different characteristics
• Out of 10o,000 customers, who all have the higher
probability of default/ Charge off?
• Basically, who will run away with my money?
• We need to predict the probability of “Running away”
• Who all have ‘Gupta Bank’ credit card applications in
this room?
Bank builds a model that gives a score to each
customer
Applicants

“Developing set of equations or mathematical formulation to forecast future


behaviors based on current or historical data.”
Predictive Modeling
Lets try to understand predictive modeling

Predictive Modeling – Fitting a model to the data to predict the future.

• Predicting the future –and it is so easy some times


• Who is going to score more runs in IPL-2013?
• That’s it you predicted the future..
• BTW how did you predict?
• Predicting the future based on historical data is
nothing but Predictive modeling
Predictive Modeling
Lets try to understand predictive modeling

Predictive Modeling – Fitting a model to the data to predict the future.

• Who is going to score more runs in IPL 2013?


• Predicting the future …well it is not that easy …
Predictive Modeling
Horse Race Analogy

How to bet on best horse in a horse race


The Historical Data
Win vs. Loss record in past 2 years

• Long legs: 75% (Horses with long legs won 75% of the times)
• Breed A: 55%, Breed B: 15 % Others : 30%
• T/L (Tummy to length) ratio <1/2 :75 %
• Gender: Male -68%
• Head size: Small 10%, Medium 15% Large 75%
• Country: Africa -65%
Given the historical data
Which one of these two horses would you bet on?
Kalyan Chethak
Length of legs 150 cm 110 cm
Breed A F
T/L ratio 0.3 0.6
Gender Male Female
Head size Large Small
Country India India
Given the historical data
Which one would you bet on….now?
Kalyan Chethak
Length of legs 110 cm 150 cm
Breed C A
T/L ratio 0.45 0.60
Gender Male Female
Head size Small Large
Country Africa India
Given the historical data
What about best one in this lot?

Horse-1 Horse-2 Horse-3 Horse-4 Horse-5 Horse-6 Horse-7 Horse-8 Horse-9 Horse-10
Length of
109 114 134 130 149 120 104 117 115 135
legs
Breed C A B A F K L B C A
T/L ratio 0.1 0.8 0.5 1.0 0.3 0.3 0.3 0.6 0.7 0.9
Gender Male Female Male Female Male Female Male Female Male Female
Head size L S M M L L S M L M
Country Africa India Aus NZ Africa Africa India India Aus Africa
Citi has a similar problem?
Who is going to run away with my money?
• Given Historical of the customers we want to predict the probability of bad
• We have the data of each customer on
• Customer previous loans, customer previous payments, length of account credit
history, other credit cards and loans, job type, income band etc.,
• We want to predict the probability of default
Credit Risk Model Building
Four main steps

1. Study historical
data
3. Find the exact
• What are the 2. Identify the most 4. Use these
causes(Customer impact of each
impacting factors coefficients for future
Characteristics) factor(Quantify it)
• What are the
effects(Charge off)
The Historical Data of Customers
Attribute Value

• Contains all the information about customers


SSN 111259005
Age 27
• Contains information across more than 500
variables Number of dependents 2

• Portion of data is present in the application form Number of current loans 1

• Portion of the data is available with bank


Number of credit cards 1
Number Installments 30days late in 4
• Lot of data is maintained in bureau last 2 years
• Social Security number –in US Average utilization % in last 2 years 30%
• PAN Number in India Time since accounts opened 60 months
Number of previous applications for 2
credit card
Bankrupt NO
Identifying most impacting factors
Out of 500 what are those 20 important attributes
Model Building
logistic regression model to predict the probability of default

• Probability of bad = w1(Var1)+w2(Var2) +w3(Var1)……+w20(var20)


• Logistic regression gives us those weights
• Predicting the probability
• Probability of bad=0.13(number of cards)+ 0.21 (utilization)+…….+0.06(number of loan
applications)
• That’s it ….we are done
Credit Risk Model Building
Real-time Example
Attributes used on the model
1. MonthlyIncome
2. Number of loans
3. Number of times 30days late in last 2 years
4. Utilization in last 2 years
5. Age
6. DebtRatio – Monthly Debt / Monthly Income
‘Gupta Bank’ Credit Cards Approval

• Lets use above model for gupta bank


• Who are the applicants here?
• Lets get the bureau information for the applicants
Actual model Building Steps
Objective & Customer score Observation
Portfolio vs. account Exclusions point and/or
identification score window

Performance Validation plan


Bad definition Segmentation
window and samples

Variable Scoring and Score


Model Building
Selection Validation Implementation
Marketing Example
Predicting the response probability to a marketing Campaign

• Selling Mobile phones – Marketing campaign


• Who should we target?
• Consider historical data of mobile phone buyers
• See their characteristics
• Find top impacting characteristics
• Find weight of each characteristic
• Score new population
• Decide on the cut off
• Try to sell people who score more than cut off
Other Applications of Model Building

• Fraud transactions scorecard – Fraud identification based on attributes like


transaction amount, place, time, frequency of transactions etc.,

• Attrition modeling – Predicting employee attrition based on their characteristics


Thank You
Venkat Reddy
21.venkat@gmail.com

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy