Chart of the Week: Credit Score Transition Table

Source: MBA Analysis of Intercontinental Exchange (ICE) McDash Data

This week’s MBA Chart of the Week presents an analysis of the credit scores recorded in ICE loan application data from the first half of 2025. Each record includes up to three anonymized credit scores for both the borrower and co-borrower.

For this analysis, we keep only the most recent record for each application and limit the sample to observations with at least one reported borrower credit score. We further restrict the data to loans with a single borrower (i.e., no co-borrower) and exclude loans with any reported credit score below 500. These filters result in a final sample of nearly 105,000 records.

For each record, we calculate a decisioning credit score following Fannie Mae’s Selling Guide methodology. Specifically, when three credit scores are available, we use the middle score; when two scores are available, we use the lower score; and when only one score is available, we use that score as the decisioning credit score.

To create a random credit score for each observation, we use a uniform random number generator to select among the available credit scores with equal probability. This approach is intended to mimic a process in which a lender randomly selects the score from one of the three credit bureaus for each application.

Because the focus of this analysis is on pricing and not on automated underwriting and credit-risk decisioning (i.e., determining whether to approve or deny a loan application), we group both the decisioning credit scores and the randomly selected credit scores into the credit-score buckets defined in Fannie Mae’s 2026 Loan-Level Price Adjustment (LLPA) matrix and construct the swap-set table shown above.

For example, among borrowers whose decisioning credit score falls within the 700–719 LLPA bucket:

• Nearly 68% of randomly selected scores fall into the same LLPA bucket as the decisioning score.
• Approximately 91% fall either in the same bucket, one bucket higher, or one bucket lower in the pricing matrix.
• The shares of loans moving up one bucket and down one bucket are roughly comparable, suggesting that, in aggregate, the GSEs would realize little net gain or loss in LLPA revenue from these bucket shifts.

Interestingly, this math (2/3 in the same bucket as the actual decision score and 90% +/- one bucket) largely holds throughout the entire LLPA grid and implies that the move to a single file would have little impact on credit risk and little impact on LLPA revenue.