There are many different credit organizations, offering a wide variety of credit products and managing credit portfolios in alternative strategic ways. Whilst it is fair to say that there are definite "rights" and "wrongs" in managing credit portfolios, it is also important to realize that there is no single, ideal approach to setting up your credit policies and strategies. Once an initial best practice strategy has been set up, constant changes in portfolio and economic dynamics, coupled with the need to maximize profitability, make continual re-evaluation of current procedures necessary.

As credit management professionals, the challenge is to identify ways to continually improve credit policies to become more profitable. This may be by increasing revenue, reducing costs, improving bad debt, making more effective use of resources or more likely a combination of all of these factors.

Often credit managers have ideas for changes to existing procedures that they believe will result in improved profitability. However, there is typically no empirical proof that the new strategies will work and it is necessary to take a "leap of faith" to implement them. This often constitutes a business or operational risk and as such managers are less likely to test the more radical ideas. These ideas often offer large potential benefits but also have larger potential risk.
It is essential that portfolio testing can occur, to continually improve and test new ideas, whilst not taking unnecessary risk and suffering negative impacts. It is for this reason that the principle of Champion/Challenger testing was developed.

Simply stated, Champion/Challenger testing allows the credit professional to test new business strategies on a small but representative sample of the account base, in order to prove the benefits whilst limiting the risks of new policies. It facilitates the on-going evolution of new credit strategies in the continual search to become ever more profitable.

Establishing Test Groups

In order to segment the portfolio to facilitate the implementation of the new strategies on smaller groups of accounts, test groups need to be created within each company or organization. This is achieved by allocating each account a random digit number. Typically this is a randomly assigned number from 00 to 99 inclusive. It is important that the number is assigned completely randomly to enable valid testing to be conducted. In the future this number may need to change as part of a re-randomization of the portfolios so any methodology associated with account number allocation should also be avoided.

The allocation of the random digit will normally require two separate developments; the first to allocate a random digit to all of the existing accounts and the second to ensure that all subsequent new accounts are automatically assigned a random digit at account creation. Once the random digit numbers have been assigned, it is possible to group accounts into test areas that are both representative of the total portfolio and statistically valid in terms of size. Research has shown that typically these test groups should contain at least 30,000 accounts although this number may vary based upon the business area being tested, delinquency levels, and the activity rates on the portfolio.

In the following example it can be seen that the revolving credit portfolio has been broken down into 5 separate test areas each representing 30,000 accounts, or 20% of the total portfolio.
Change in Behavior Over Time.

The core strength of any account management system is its ability to allow for strategy testing on small groups of accounts. Over time these small groups of accounts tend to "adapt" or alter behavior as a result of the test in place. An example of this may be a lower tolerance for oversells in one group at the same risk level as compared to a different group. Account holders in one group will become familiar with the strategies in place and as a result of the differing tests applied to different groups, the groups? behaviors are no longer similar i.e. the groups are no longer made up of random samples.

This becomes more of an issue in portfolios where aggressive testing is performed on groups of accounts. It is this factor that leads to the requirement to re-randomize all accounts in a portfolio from time to time. Where more aggressive testing is undertaken in a portfolio, this re-randomization may take place as frequently as once per annum just before the new strategy tests are put in place. In other less aggressive portfolios, this may be less often.

The re-randomization procedure re-numbers the test digit for each account. Typically, the re-randomization technique chosen will ensure that accounts are evenly re-distributed across the entire portfolio. It must be stressed that simple re-randomization techniques such as swapping the first and last digits of the number around etc do not provide a statistically valid randomization and should not be used.

The above provides an illustration of the type of process that should be followed in order to re-randomize the portfolio. It really serves as guide only and in fact the actual re-randomization procedure will entail far more complex check routines and reporting to ensure that the portfolio is in fact split up into scientifically valid random samples.

Testing the Outcome

For each measure, the client must list the value before and after re-randomization for each digit group. A digit group is a pre-defined % grouping e.g. a 5% group may include all accounts with test digits 00 - 04 or 40 - 44. The larger the total volume of accounts, the smaller the digit group will be, as they will provide a more representative sample. For example, a portfolio of 1,000,000 accounts can realistically be split into random digit groups of 5%, or 50,000 accounts. However, a portfolio of 100,000 accounts would require bigger random digit groups of say 20% for accurate testing.

In the example below, "% Difference" is the difference between the average for all Random Digit Groups and the actual Random Digit Group value. Examples of the types of tests that should be conducted are as follows:

- Total Accounts & % Difference
- Current Accounts & % Difference
- 1 Cycle Accounts & % Difference
- 2 Cycle Accounts & % Difference
- 3 Cycle Accounts & % Difference
- 4+ Cycle Accounts & % Difference
- Accounts with Debits & % Difference
- Merchandise Sales & % Difference
- Average Sale & % Difference
- Total Payments & % Difference
- Accounts with Lines & % Difference
- Average Credit Limit & % Difference
- Current Balances & % Difference
- 1 Cycle Balance & % Difference
- 2 Cycle Balance & % Difference
- 3 Cycle Balance & % Difference
- 4+ Cycle Balance & % Difference
- Total Balance & % Difference

A good final test is to show (in a matrix format with old Random Digit Group and new Random Digit Group on the 2 axes) the distribution of accounts before and after re-randomization by Random Digit Group so that the client can look at how the accounts are redistributed by the re-randomization.

Final Considerations

Once the re-randomization exercise has been completed, any reporting by random digit group that has been in place will be null and void and a new measurement period begins. It is because of this that the timing for any re-randomization exercise is critical. The portfolio should only be re-randomized at the point of implementing new account management strategies.