In many predictive analytics situations, deriving the cost matrix for a particular problem is an important pre-cursor to developing predictive models. For the Credit Screening problem, giving a positive rating to a negative case could result in bad debts (i.e., non-credit-worthy people defaulting on their loans). Conversely, failing to lend to a credit-worthy customer has an associated opportunity cost (interest revenue from the loan).

Construct a Cost Matrix for two types of classification errors

Assume that the credit screening model will be used to determine which customers will receive home mortgage loans. Compute the cost matrix based on the following information:

• Assuming an average loan amount of $300,000 and a fixed interest rate of 5% for a 30-year loan with monthly payments. Assume $0 down payment.
• For the profits for lending to a credit-worthy customer, you can use an online mortgage calculator to get payments on a 30-year amortization schedule. No need to consider PMI or secondary mortgages. NPV calculations will also be needed to get the present value of those payments, using a discount rate of 2%, which we will assume to be the risk free rate. DO NOT simply assume that if the bank doesnt lend to this worthy customer, that they can simply lend to anotherthe available lending pool always exceeds the number of worthy prospective customers!
• For the cost of lending to an unworthy customer, assume the average time to default is 2 years. If a loan defaults, the property goes into foreclosure/short sale. Assume 80% of the original value of the home ($300,000) can be recovered. Also include $50,000 of foreclosure costs for real-estate agents, marketing, maintenance, etc.

Fill in the following cost matrix. You can use whatever baseline makes most sense to you (e.g. actual costs/benefits, relative to approving all, approving none, etc.):

Q1: Cost Matrix for Classification Errors

Cost Matrix	Actual Unworthy (+)	Actual Worthy (-)
Predicted Unworthy
Predicted Worthy