Organic portfolio growth starts with evaluating loan loss performance

July 8, 2013 by Andrew Beddoes

There are two core fundamentals of evaluating loan loss performance to consider when generating organic portfolio growth through the setting of customer lending limits. Neither of which can be discussed without first considering what defines a “customer.”

Customer lending limits starts with defining the customerDefinition of a customer

The approach used to define a customer is critical for successful customer management and is directly correlated to how joint accounts are managed. Definitions may vary by how joint accounts are allocated and used in risk evaluation. It is important to acknowledge:

  • Legal restrictions for data usage related to joint account holders throughout the relationship
  • Impact on predictive model performance and reporting where there are two financially linked individuals with differently assigned exposures
  • Complexities of multiple relationships with customers within the same household – consumer and small business

Typical customer definitions used by financial services organizations:

Checking account holders:

This definition groups together accounts that are “fed” by the same checking account. If an individual holds two checking accounts, then she will be treated as two different and unique customers.

Physical persons:

Joint accounts allocated to each individual. If Mr. Jones has sole accounts and holds joint accounts with Ms. Smith who also has sole accounts, the joint accounts would be allocated to both Mr. Jones and Ms. Smith.

Consistent entities:

If Mr Jones has sole accounts and holds joint accounts with Ms. Smith who also has sole accounts, then 3 “customers” are defined: Jones, Jones & Smith, Smith.

Financially-linked individuals:

Whereas consistent entities are considered three separate customers, financially-linked individuals would be considered one customer: “Mr. Jones & Ms. Smith”.

When multiple and complex relationships exist, taking a pragmatic approach to define your customers as financially-linked will lead to a better evaluation of predicted loan performance.

Evaluation of credit and default risk

Most financial institutions calculate a loan default probability on a periodic basis (monthly) for existing loans, in the format of either a custom behavior score or a generic risk score, supplied by a credit bureau. For new loan requests, financial institutions often calculate an application risk score, sometimes used in conjunction with a credit bureau score, often in a matrix-based decision.

This approach is challenging for new credit requests where the presence and nature of the existing relationship is not factored into the decision. In most cases, customers with existing relationships are treated in an identical manner to those new applicants with no relationship – the power and value of the organization’s internal data goes overlooked whereby customer satisfaction and profits suffer as a result.

One way to overcome this challenge is to use a Strength of Relationship (SOR) indicator.

Strength of Relationship (SOR) indicator

The Strength of Relationship (SOR) indicator is a single-digit value used to define the nature of the relationship of the customer with financial institution. Traditional approaches for the assignment of a SOR are based upon the following factors

  • Existence of a primary banking relationship (salary deposits)
  • Number of transactional products held (DDA, credit cards)
  • Volume of transactions
  • Number of loan products held
  • Length of time with bank

The SOR has a critical role in the calculation of customer level risk grades and strategies and is used to point us to the data that will be the most predictive for each customer.

Typically the stronger the relationship, the more we know about our customer, and the more robust will be predictive models of consumer behavior. The more information we have on our customer, the more our models will lean towards internal data as the primary source.

For weaker relationships, internal data may not be robust enough alone to be used to calculate customer level limits and there will be a greater dependency to augment internal data with external third party data (credit bureau attributes.) As such, the SOR can be used as a tool to select the type and frequency of external data purchase.

Customer Risk Grade (CRG)

A customer-level risk grade or behavior score is a periodic (monthly) statistical assessment of the default risk of an existing customer. This probability uses the assumption that past performance is the best possible indicator of future performance.

The predictive model is calibrated to provide the probability (or odds) that an individual will incur a “default” on one or more of their accounts.

The customerrisk grade requires a common definition of a customer across the enterprise. This is required to establish a methodology for treating joint accounts. A unique customer reference number is assigned to those customers defined as “financially-linked individuals”. Account behavior is aggregated on a monthly basis and this information is subsequently combined with information from savings accounts and third party sources to formulate our customer view.

Using historical customer information, the behavior score can accurately differentiate between good and bad credit risk individuals. The behavior score is often translated into a Customer Risk Grade (CRG). The purpose of the CRG is to simplify the behavior score for operational purposes making it easier for noncredit/ risk individuals to interpret a grade more easily than a mathematical probability.

Different methods for evaluating credit risk will yield different results and an important aspect in the setting of customer exposure thresholds is the ability to perform analytical tests of different strategies in a controlled environment. In my next post, I’ll dive deeper into adaptive control, champion challenger techniques and strategy design fundamentals.

Related content:

White paper: Improving decisions across the Customer Life Cycle