Scoring Your Customers: How Often Is Enough?

By Luke van Dijk and Erik Franco

Credit scores are dynamic. They can move up or down depending on changes in a consumer’s credit report. How quickly scores change and by how much depends on the consumer’s credit-related actions and behaviors as reported to the credit bureaus.

But how much do credit scores change in the short and long term? How well does a credit score pulled months prior to the lender’s decision date predict future consumer credit repayment risk? Recent FICO research looked at FICO Score migration on a nationally representative sample of active bank card accounts over the course of one month, three months, six months, and twelve months.

The key finding from this research is that a consumer’s most recent score is the most predictive. Lenders using credit scores for account management decisions should obtain refreshed scores frequently. And in the securitization market, institutions assessing the risk of a book of consumer credit debt are strongly encouraged to obtain fresh scores on the accounts in question, rather than relying on outdated originations scores.

For the majority of the population studied, the score did not change by more than 20 points upward or downward compared to the prior month or quarters. However, 12 percent of scores changed by more than 20 points over the preceding month, 23 percent changed more than 20 points over the prior quarter and 32 percent changed more than 20 points over the prior two quarters. The longer the time elapsed, the more likely a substantial migration (e.g., score changes more than 40 points up or down) had occurred.

A substantial drop in a score is typically the result of new delinquency being reported and/or the consumer’s accounts having higher balances or higher credit line utilization. The typical drivers of a substantial score increase include having lower balances or lower credit line utilization, as well as aging of the credit history (i.e., time since credit accounts were opened, and time since delinquencies were posted).

Higher scores are more likely to remain stable. For example, 86 percent of the records with a starting score of 750 and higher remained within 20 points of their starting scores when re-examined three months later. But records with lower starting scores are more likely to experience score movement over time. In the under-550 segment, a majority of records (57 percent) migrated by more than 20 points over the following quarter.

Of note is that many lower-scoring consumers who are experiencing notable score change over a given quarter are improving their scores over time. Of those with a starting score less than 650 that migrated more than 20 points over a quarter, a majority tended to migrate in a positive direction (29 percent upward vs. 13 percent downward).

Score migration in a particular direction did not typically signal an ongoing trend, i.e. the initial movement of a consumer’s score up or down did not necessarily yield continued movement in that same direction in the next quarter. In particular, scores that fall by more than 20 points one quarter are more likely the following quarter to rise upward, rather than falling further. On the other hand, scores that rise by more than 20 points one quarter are more likely the following quarter to fall downward, rather than rising further.

For example, a person’s score may drop in the December-January time frame due to higher credit card balances being reported to the credit bureaus resulting from holiday spending, but rebound higher in March-April as those holiday balances are paid down.

The movement in scores reflects the change in risk profile associated with the corresponding consumer. FICO conducted further research to determine whether a current score outperforms an older score as a predictor of risk. As evidenced by the results in Figure 2, the most current FICO Score for a given consumer is the most predictive.

The first column of Figure 2 shows sample score cutoffs. The next two columns show the percentage of accounts falling above and below each cutoff. For example, at a cutoff of 670, 77.2 percent of the accounts would pass and 22.8 percent would not.

The next two columns show the percentages of accounts that migrated above or below the cutoff three months after the initial observation date. At the same 670 cutoff, 2.5 percent of the accounts migrated from below to above the cutoff, and 1.7 percent migrated from above to below. These “swap sets” capture the accounts potentially receiving different treatment at an April 2015 decision date, depending upon whether the older or the current score was used to make the decision.

The last two columns illustrate the repayment odds on the “swap-in” (migrated above) and “swap-out” (migrated below) groups over the subsequent 24 months. At a cutoff of 670, the odds of those whose scores migrated above the cutoff was 15.5 to 1, versus 5.7 to 1 for those records that dropped below 670. (Repayment odds are defined as the number of good-paying accounts over the number of accounts that go more than 90 days past due on any credit obligation over the subsequent 24 months. The lower the odds, the worse the delinquency rate of the group being observed.)

The fact that the “migrated below” records have much worse odds than the “migrated above” records means that the change in score over the three-month interval reflected a real change in the risk level of the consumers involved. The group that migrated above cutoff did indeed pay better over the subsequent 24-month period than those whose scores fell below cutoff.

In today’s credit environment, there is less margin for error. Proactive lenders want to be sure they are not caught unaware of borrowing activity that may turn yesterday’s “good” customer into a future write-off. Thus, the industry’s largest lenders are making decisions based on current credit scores. Rigorous analysis of score migration, combined with experience from leading lenders, affirms that frequent score refreshes will help institutions cost-effectively make better decisions in managing portfolio risk.

Luke van Dijk is senior principal consultant in analytic science at FICO. Erik Franco is senior scientist in analytic science at FICO.