What Credit Scores Carry Forward
The question at the center of this series is what it means for a credit score to carry the past into the future. The answer is more precise than “your history matters,” and less absolute than “your score determines your life.” Selected observations become an uncertain belief about future repayment; that belief changes the price and availability of credit; those terms change behavior and outcomes; and the next observable outcome can become part of the history. This final synthesis connects that chain to the measurement limits, distributional trade-offs, and policy choices that determine which past events are allowed to matter.
What exactly travels forward?
Two different processes are easy to confuse. The first is the expansion of credit visibility across calendar time. The CFPB update compares the estimated U.S. adult population with and without records in 2010 and 2020, using a corrected data source and a consistent treatment of inquiry-only files. On that comparable basis, the estimated share with a scored record rose from 81.6 percent to 87.5 percent, while the estimated share without a record fell from 5.8 percent to 2.7 percent. The unscored share also fell, from 12.7 percent to 9.8 percent.
The second process is learning within a person’s life. The Philadelphia Fed paper documents that mean credit rankings rise across age brackets and that the dispersion of rankings within those brackets also grows, especially between ages 20 and 40. The paper interprets this pattern partly through a model in which lenders gradually learn from observable choices and outcomes. The two processes are related because more visible records create more opportunities for scoring, but they are not the same evidence. A calendar-time change in the share of adults with scores does not show that each individual’s lender learned faster. An age pattern does not show that the population became more visible between two dates.
The first process is about whether usable information exists. The CFPB distinguishes no record, scored record, stale unscored record, and insufficient unscored record. These categories are produced by the coverage of a reporting panel, the treatment of files such as inquiry-only records, the definition of recent or sufficient activity, and the validity rules of a commercial scoring model. A person can move from invisible to recorded because the data system captures a closed account or collection, while remaining unscored because that record lacks enough payment history. Visibility and prediction are nested but different conditions.
The second process is about what a lender infers from the information that is available. In the Philadelphia Fed model, a hidden type represents a difference in patience. Observable saving, borrowing, income, repayment, and default provide noisy evidence about that type. The resulting type score is a posterior belief: the lender’s current assessment of the probability that the borrower belongs to one type rather than another. It is not the same object as a credit score.
The model’s credit score is a prediction of avoiding a default in the next period. It depends on the hidden type but also on income, wealth, and the probability that the person will borrow. A person can therefore have a high predicted repayment rate partly because they are unlikely to borrow, not only because they are unlikely to default after taking a standard loan. A credit ranking then places that predicted repayment probability within the distribution of scores across the model population. A type score, a credit score, and a ranking may be related, but they answer different questions.