Record credit losses fuel the search for increased predictive power. Many risk managers have considered using score migration in new risk management strategies.
Score migration is a change in score value-negative or positive-that occurs in a short time, generally two to six months. This phenomenon has different implications for the pre-screen market, new account monitoring, and account management.
In marketing programs, the magnitude of migration is difficult to measure, since people who respond to credit card mail offers are not a random sample of the original prospect list. Many factors affect consumer responsiveness, including product offering, segmentation, and market trends. It is generally accepted that people with negative indicators in a credit report respond at a higher rate. Thus, individuals with unidentified or recently occurring negative information represent a disproportionate percentage of the responses, exaggerating the true level of migration.
In the pre-screen market, the change in score from pre-screen to post- screen can be linked to two factors: change in available data, and change in consumer behavior. Understanding the difference is important, since one situation might lead to a different decision than the other.
The principal cause of score migration from pre- to post-screen is that a credit bureau may have an incomplete file on a consumer and may gain information that makes the file more complete. Often the post-screen process allows for a more accurate report to be obtained-for instance, if the customer supplied a Social Security number or updated address. Therefore the data may be different from those in the file used in pre- screening. In some sense this is not score migration, since the consumer's behavior has not changed.
This is not an indictment of credit bureaus. They face staggering challenges in managing data from contributing institutions, yet they have lately improved file quality significantly. Data accuracy problems are likely to persist, but pre-screen marketers have tools and strategies to minimize the impact.
Sophisticated pre-screen marketers build models that predict the likelihood that a file in the data base is incomplete. The small percentage of the files that show a high probability of being file fragments are not solicited.
Another factor that causes change in the data is the use of a different credit bureau on the back end. This should be considered when drawing conclusions on score migration. Structured test and control programs can help determine the differences in data from various bureaus. For example, samples from all three credit bureaus can be mailed within specific ZIP codes to determine the incident of file fragments.
The average pre-screen marketing program takes nearly two months to reach the consumer after the original score is obtained. In this time a small percentage of these people will experience significant changes in credit data, resulting in a new score. This is "true" migration, reflecting changes in consumer behavior over time.
Each score range of a predictive system has some level of negative performance in the future, so movement should be expected. Again, the level of degradation is likely to be exaggerated: People with problems will respond at a higher rate than those who had no changes, or positive changes.
Most pre-screen marketers now use a post-screen process to reduce the impact of negative score migrations. Recently expanded interpretation of the Fair Credit Reporting Act and the increase in average lines and balances make this process economical.
Many lending institutions use a credit bureau score in application processing, then re-score an account in six months using an account monitoring program. The new calculation is often quite different from the score used in the underwriting process.
Depending on the credit bureau scoring system, the new account and inquiry may cause the consumer to be scored by a different algorithm. Some generic models have multiple scorecards that segment consumers by factors such as prior delinquency, time in file, and demand for credit.
Another complication in comparing the original score to the first account monitoring program is the score selection criteria for joint accounts. In general, a bureau score is obtained for only the primary applicant, but most account monitoring programs obtain scores for both primary and secondary account holders. The lender, or the monitoring program, then determines which score is entered in the data base or billing system. Many programs select the riskier score, exaggerating the decline in credit quality.
Careful analysis is required when determining score changes during the first few months of an account. It is usually too early to take decisive account actions or make portfolio evaluations based on score migration. Sophisticated data mining and decision engines are required to understand the dynamics of the score.
As accounts mature, scores tend to stabilize. The impact of change in available data and joint account processing diminish with time. Variations in score are more likely to represent changes in consumer behavior.
The most common miscalculation occurs when changes in delinquency and loss ratios are calculated during the same period as the change in score.
When a creditor reports an account delinquent to the bureau, the score will reveal an increase in risk. Conversely, if the consumer pays on time or reduces the balance, the risk indicated by the score will drop. In general, stable accounts show the best performance, those that have dropped are average, and those that appear to have "improved" are the highest risk. The pattern is somewhat similar to the stock market: consistency has its rewards; stocks at 52-week lows are likely to improve, and high stocks are likely to fall.
Considering the score dynamics and human behavior, this pattern makes sense. People with positive long-term credit may have temporary periods of decline, but recover quickly. Less stable people with temporary improvements are likely to return to old behaviors.
The desire to take action at the first sign of trouble is a noble pursuit that often drives risk managers to ignore conflicting data. To further complicate this quest, the true value of score migration should be measured by its marginal contribution, after account characteristics and behavioral scoring have been exhausted.
The marginal contribution analysis should be completed only on accounts that require risk management action. Accounts closed by collections, for example, are not likely to be eligible for a line increase and should not be included in such an analysis.
The marginal contribution can be calculated using score ranges of both behavioral score and current credit bureau scores. This results in mind- numbing three-way matrices. Only the largest institution can supply enough "bads" to make statistically valid calculations for this analysis.
The true nature of score migration is often misunderstood, and its value exaggerated. In pre-screened acquisition programs, it can be used to monitor credit bureau effectiveness. In new account monitoring, it provides little value. In account management, score migration should be used only if sufficient data are available to make marginal contribution calculations.