P2.T6.604. Retail credit scoring models (Crouhy)

Nicole Seaman

Director of CFA & FRM Operations
Staff member
Subscriber
Learning objectives: Define and describe credit risk scoring model types, key variables, and applications. Discuss the key variables in a mortgage credit assessment and describe the use of cutoff scores, default rates, and loss rates in a credit scoring model. Discuss the measurement and monitoring of a scorecard performance including the use of cumulative accuracy profile (CAP) and the accuracy ratio (AR) techniques.

Questions:

604.1. In applying for a mortgage loan, a borrower who is seeking approval prefers a lower number for each of the following variables in her application, ceteris paribus, EXCEPT for which does she not necessarily prefer a lower number?

a. DTI ratio
b. FICO score
c. LTV ratio
d. Number of recent inquiries


604.2. Suppose the population of credit applicants, which is ordinarily unknowable, divides into either "good" or "bad" accounts, to use Crouhy's terminology. Good accounts have a mean credit score of 730.0 with a standard deviation of 25.0. Bad accounts have a mean credit score of 600.0 with a standard deviation of 40.0. Both distributions are approximated by a normal distribution. A bank is evaluating whether to set its cutoff score at 680.0; i.e., applicants with a score of 680.0 or greater will be approved. There are two types of underwriting error: a "false bad" is the rejection of good account; a "false good" is the acceptance of a bad account.

a. This cutoff of 680.0 ensures an error rate of less than 1.0%
b. The bank can reduce the fraction of false bads by increasing the cutoff score
c. The bank can reduce the fraction of false goods by increasing the cutoff score
d. The bank can simultaneously reduce the fraction of errors (ie, both false bads and false goods) by increasing the cutoff score


604.3. About the cumulative accuracy profile (CAP), each of the following statements is true EXCEPT which is not?

a. A perfect credit scoring model generates an accuracy ratio (AR) of 1.0, which is the upper bound on the AR
b. A purely random model that cannot differentiate between good and bad customers is likely to generate an accuracy ratio (AR) of 0.40 to 0.60; i.e., 50% +/- 10%
c. The CAP curve, which plots the actual rating model as a cumulative percentage of defaults, is monotonically increasing (aka, nondecreasing or weakly increasing)
d. The CAP curve plots the fraction of defaulted customers (y axis) against the fraction of entire customer population sorted by score from highest risk (left) to lowest risk (right)

Answers here:
 
Top