3. Model-Building and Impact on Business Costs.

1. Get your Instructions from Your Supervisor (video above)

2. Here is the background information on your task

When granting a payment holiday, Nedbank incurs several costs. The bank does not receive projected cash flows which impacts many projects and expected returns. There is higher credit risk and the bank runs the risk of higher impairments as the payment holiday period can serve as a default period. Furthermore, there are administration costs and operational costs associated with granting payment holidays.

On the other hand, if we do not grant payment holidays many more clients may default, and the bank may lose out on the initial amount loaned to the clients and the expected returns.

Costs

• It costs us R500 to grant a payment holiday.
• It costs us R2500 if a client defaults.
• Experience from all major banks has shown that if clients are granted a payment holiday, they have a 7% chance of defaulting within the year.
• Currently, business approves 50% of payment holiday applications. The remaining 50% have a 20% default rate.

Cost-to-business calculated on the currentapproval rate of 50%

Assume we have 100 people. If everyone gets a payment holiday, it will cost: 100 * 500 + 100 * 7% * 2 500 = 50000 + 17500 = R 67500

If no one gets a payment holiday, it will cost: 20% * 100 * 2 500 = 20 * 2 500 = R 50000

Under the current process it costs: (50% * 100 * (500 + 7% * 2500)) + (50% * 20% * 100 * 2 500) = (25000 + 8750) + 25000 = R 58750

3. Here is your task

As part of the analytics and modelling team at Nedbank, you are given this urgent task to provide a system or model to accept or decline payment holiday applications based on the manual data given. Your model must deliver a cost-effective way of granting payment holidays, and you must include a three-month payment holiday and a six-month payment holiday.

Your model must improve on the cost of approving 50% of the applications, which is what the case-by-case approach has achieved so far. You will be given 10 127 rows of data. You may ask for more if you need out-of-time data for testing and validations. The model can be made with any tool that yields effective results, like Excel or Python.

Remember that your model must save us money, while also helping clients find financial relief.

Hints:

• Build your model by considering the exploratory variables you deem relevant. Make sure that the data is separated for the three-month and six-month payment holiday. This includes looking at the relationship between the clients who defaulted and clients who did not default.
• Determine the weights and thresholds of each variable.
• Determine the score.
• Use score to accept or decline an application.
• Evaluate your model by considering the payment holiday approval rate and the cost to business.

Submit a two-page report, explaining the mechanics of your system or model, and explain why you think this system works and why it should be implemented.