Executive Summary

Problem identification:

American Express (Amex) has disclosed an increase in credit card loans delinquency
rate for its U.S. consumer and U.S. small business card member loans. This trend
might cause potential risks and revenue losses to its company. Therefore, it is
important to implement data-driven methods to predict the credit card default rate of
the new customers and optimize the credit limits under the control of default risks.

Research Methodology:

To address this problem, this essay will discuss two key Management Science
Techniques: Logistic Regression and Linear Programming. Logistic Regression is
utilized to predict the probability of Credit Card Default Rate, based on the following
factors: Age, Owns Car, Owns House, Number of Children, Net Yearly Income, Total
Family Members, Yearly Debt Payment, Credit Limit, Credit Limit Used (%), Credit
Score, Previous Defaults and Default in Last 6 Months. Linear Programming is
utilized to optimize the credit limit allocation which could increase the interest
revenue while reducing the default exposure.

Key Finding:

Through the Logistic Regression, it highlights that Credit Limit used, Credit Score,
and Number of Children are the most influential factors in determining default risk. In
addition, the accuracy rate in predicting the default probability is 98%. By leveraging
these insights, Amex could predict the new customers’ default and make a decision
whether to approve the application of the new customers’ credit card. Through Linear
Programming, credit reallocation strategy could reduce default risk by 56.5% and
increase the interest revenue by 31.8%.

Problem Description & Identification

Recently, high inflation and rising interest rates have increased the consumers’
financial pressure. The credit card market suffers the brunt of the economic downturn.
Amex reported that in September 2024, the 30-day delinquency rate for U.S.
consumer credit cards rose to 1.4%, while the rate for U.S. small business credit cards
increased to 1.5% (Investing.com, 2024). This upward trend indicates a potential
decline in cardholders’ repayment capacity, posing financial risks to credit card
issuers. Therefore, it is an important issue for Amex to address this trend, otherwise, it
may face higher bad default losses and increase the operation costs, which would lead
to a decline in investor confidence and market competitiveness.

To effectively identify customers with high default risks and adjust credit limits
appropriately, sometimes over-approval of credit limits may lead to increased default
rates, while insufficient credit limits may cause Amex to miss opportunities to acquire
creditworthy customers, ultimately affecting profitability. Amex has to implement a
Logistic Regression Analysis to predict credit card default rates among new
customers and Linear Programming to optimize credit limit decisions while
maintaining risk control.
Management Science Techniques

Method 1: Logistic Regression for Default Probability Prediction

Through the Python to train a Regression model based on the American Express
customer data provide by Kaggle (Basak,2021). The dependent variable of this model
is the credit card default rate. The independent variables are Age, Owns Car, Owns
House, Number of Children, Net Yearly Income, Total Family Members, Yearly Debt
Payment, Credit Limit, Credit Limit Used (%), Credit Score, Previous Defaults and
Default in Last 6 Months.

After analyzing the data, it generates an equation to predict the credit card default
probability (Table 1):

Logit(P) = -8.0658+ (0.0399 * Age) + (-0.0414 * Owns Car) + (-0.0279 * Owns House) +
(0.2153 * Number of Children) + (0.2223 * Net Yearly Income) + (-0.1468 * Total Family
Members) + (-0.0001 * Yearly Debt Payments) + (0.3004 * Credit Limit) + (2.1947 * Credit
Limit Used (%)) + (-4.5172 * Credit Score) + (1.3780 * Previous Defaults) + (1.3046 *
Default in Last 6 Months)

Table 1.
According to Table 1., the overall accuracy of the model is 98.03%, indicating that the
model has good predictive ability. The statistical significance test of the variables
shows that Credit Limit Used Rate, Credit Score and Number of Children all reach a
significant level (p < 0.05). The Credit Limit Used Rate is significantly positively
correlated with the Default Probability, indicating that the higher the Credit Limit
Used Rate, the higher the default risk will be; while the Credit Score is strongly
negatively correlated with the Default Probability, indicating that the higher the Credit
Score, the lower the default risk. The Number of Children shows a significant positive
relationship with the Default Probability, suggesting that an increase in the Number of
Children will lead to an increase in the family's financial pressure, thereby increasing
the risk of default.

Based on the above result, it could provide American Express as a reference to decide
whether to approve new credit cards to new customers. Amex could use credit limit
used rate and credit score as core risk prediction index to improve the accuracy in
measuring the probability of default. For instance, if a new customer’s default rate is
above 30% calculated under this model, Amex should reject new customer
applications to achieve effectiveness of risk management.

According to American Express's 2023 financial report, the credit card default rate is
approximately 2.5% (American Express, 2023). However, the predicted default rate
does not represent the actual default rate. The reason for setting the approval standard
for new customers at 30% is based on the Defensive risk management. It involves a
comprehensive, proactive approach to safeguarding stakeholder interests through
integrated activities like governance, compliance, and security. It anticipates,
prevents, detects, and reacts to threats, ensuring organizational resilience and effective
corporate defense management (Lyons, 2009). Therefore, in the current financial
market, the strategy of moderately relaxing certain standards to balance risks and
returns while strictly controlling risks can not only meet the needs of business growth,
but also quickly adjust risk tolerance when the economic environment fluctuates,
thereby achieving the optimal configuration between risk and return.
Method 2: Linear Programming to optimize the credit limit allocation.

After calculating the default rate of new customers through the default rate prediction
model of Method 1, Amex could select out customers with a default rate below 30%
and approve their credit cards. Then, optimizing the credit limit through linear
programming to minimize risk and maximize interest revenues. Assume that the new
customer’s existing Credit Limit is based on the information of other card issuers and
then adjust it to minimize risk & maximize profit.

Through random simulation of 100 sets of new customer data, 10 customers with a
default rate below 30% calculated under the Prediction Model were selected (Table
2.), and credit limits were reallocated using linear programming. To analyze the
changes in risk and interest income, it is necessary to assume that these 10 customers
all enter a revolving interest rate and default in the end. According to American
Express's regulations, if a credit card holder fails to pay the bill in full, the outstanding
balance will be converted to revolving credit and incur interest. If a cardholder fails to
pay for 90-180 consecutive days, the outstanding balance will be considered in
default. Defaulting customers usually use up all their credit limits and are unable to
repay their credit card balances on time (American Express, 2023).

Table 2.
This Method’s Objective is to reallocate the credit limits, and the constraints are the
following conditions:

1. Set credit limits based on the customer’s default rate and introduce risk
factors. Under the Basel Agreement III, Risk-Weight exposure to individual is
50% to 150% (Basel Committee on Banking Supervision, 2017), thus, set the
β= 0.5 is to assume that American Express moderately control the credit
limits, so that the lower the default rate, the higher the credit limit. Here is the
equation to set up the credit limits’ cap of each customer by the different level
of default rate:

Ci=α×1/Pi×(1−βPi)

Ci = Final credit limit of customer.

Pi = Default probability of customer.
α= Coefficient based on the total credit issued to adjust credit limit.
β= Risk factor, controlling the extent of credit limit reduction for high default customers.
(1− βPi) = Risk adjustment factor, which becomes smaller as Pi increases, allowing high-risk customers
to receive lower credit limits.

2. Total Credit Issued is under 600 thousand.

3. Expected Default Exposure (EDE) ($) is under 5% of the total credit issued.
This is based on the 2023 annual report of American Express (American
Express, 2023). Credit loss reserves ($5.118 Billion) to Total credit card loans
($126 Billion) are around 5%.

EDE = Credit Limit * Default Rate * Loss Given Default (LGD)

LGD is around 80% because the credit card loans are Unsecured debt. The
company would face high ratio of loss when the customers are default.
According to the Basel Committee on Banking Supervision (2017), the LGD
of credit cards is approximately 75% - 90%, with an average of approximately
80%.
Result of the Linear Programming (Table 3.):

Ci=α×1/Pi×(1−βPi)

Sensitivity Report of the Linear Programming (Table 4.):

Key Findings and Recommendation

According to the results of linear Programming (Table 3.), it could be found that after
credit adjustment, the overall EDE dropped from 53,693 to 23,333, a decrease of
56.5%, which successfully reduced the risk of default loss of American Express.
However, EDE is still below the limit of 30,000, which means that American Express
could allocate more credit to each customer. From the Sensitivity Report (Table 4.), it
could be seen that if the total credit limit is increased by $1, the profit will increase by
$0.15. The allowed room for increase is $92,597, which means Amex could allocate
more credit to the customers and increase total revenue while keeping risks under
control. Although this analysis assumes that every customer delay payment and enter
the revolving interest period, increasing credit limits may also increase risk, but it can
increase overall interest income within an acceptable risk tolerance, which is a
favorable decision for American Express. In addition, the credit limit for each
customer is also in line with the risk control strategy, that is, low-risk customers A
and B are given higher credit limits, and high-risk customers H, I and J are restricted.

Based on the research of linear programming, American Express could take a

dynamic adjusted the risk factor (β) strategy to achieve an effective risk management
and increase the revenue. During the normal and mild economic condition, β equal
to 0.5 as mentioned above. When the economy slumps, the Amex could increase the
β to 0.8 to control the risk strictly. In addition, β could be adjusted person by
person. American Express needs to take the customer’s payment histories into
account. If the customer repays the loan regularly, this indicates that the default rate
of them is low and β could be adjusted lower. Similarly, if a customer’s credit score
is higher, it means that their default rate might be lower, and could be given higher
credit limits with lower β.
Conclusion

This essay explores the Logistic regression and Linear Programming Methods to
enhance credit risk management at American Express (Amex). Through logistic
regression, key predictors of credit card default rates were identified, including credit
limit used (%), credit score, and number of children, with a prediction accuracy of
98%. The model enables American Express to make data-driven credit approval
decisions, thereby reducing the financial risks. Additionally, through Linear
Programming, American Express could allocate the credit limits effectively.
According to the research results, Amex could achieve a 56.5% reduction in expected
default exposure (EDE) while increasing revenue by 31.8%. In practical, Amex could
adjust credit limits based on economic conditions and individual customer profiles,
ensuring a flexible and resilient risk management framework. Through these methods,
Amex could ensure sustainable business growth and maintain market competitiveness.
Reference list

American Express. 2023. Annual report 2023. Available at:

https://s26.q4cdn.com/747928648/files/doc_financials/2023/ar/American-Express-
Annual-Report-2023.pdf (Accessed: 26 February 2025).

Basel Committee on Banking Supervision (BCBS). 2017. Basel III: Finalising post-crisis
reforms. Bank for International Settlements. Available at:
https://www.bis.org/bcbs/publ/d424.pdf (Accessed: 28 February 2025).

Investing.com. 2024. American Express reports uptick in loan delinquencies. [Online].

[Accessed: 17 February 2025]. Available from: https://www.investing.com/news/company-
news/american-express-reports-uptick-in-loan-delinquencies-93CH-3664170

Lyons, S. 2009. Risk Management’s Role in Corporate Defense. Social Science

Research Network.

Pradip Basak. 2021. AMEXpert Codelab 2021 dataset. [Online]. [Accessed 17 February
2025]. Available from: https://www.kaggle.com/datasets/pradip11/amexpert-codelab-
2021/data?select=train.csv.