RiskSpan has launched a credit-risk model built specifically for non-qualified mortgages, giving investors and issuers a tool that accounts for differences among bank-statement, debt service coverage ratio, and full-documentation loans.
Credit Model 7.1 is now generally available through the RiskSpan Platform. When paired with the company’s existing Non-QM prepayment model, the new model allows users to conduct credit and prepayment analyses and generate loan-level cash flow projections within a single environment, according to RiskSpan.
Morningstar DBRS reported that Non-QM residential mortgage-backed securities issuance reached a record $20.9 billion during the third quarter of 2025, up 97% from $10.6 billion one year earlier, according to figures cited by RiskSpan.
Fitch Ratings reported that issuance within its rated Non-QM and non-prime RMBS portfolio increased by more than 800% between 2020 and 2023. KBRA, meanwhile, projects that overall non-agency RMBS issuance — a broader category that includes Non-QM securities — will increase 15% in 2026 to $160 billion.
That growth has increased the volume of loan-level data investors and issuers must evaluate when pricing transactions and making allocation decisions. RiskSpan contends that models developed for agency mortgages or older non-agency collateral do not fully capture the borrower behavior found across different Non-QM documentation types.
Model Separates Collateral By Documentation Type
Credit Model 7.1 uses a transition-state framework, with loan-performance transitions estimated independently for bank-statement, DSCR, full-documentation, and other loan categories.
That segmentation is intended to capture the differences in borrower behavior and credit performance among products that may all fall under the broader Non-QM label.
“Non-QM borrower behavior varies meaningfully by documentation type, and generic credit and prepay frameworks simply don’t capture that,” said Divas Sanwal, head of modeling at RiskSpan. “Credit Model 7.1 was built from the ground up on Non-QM collateral, segmented by doc type, and validated with published backtesting — giving risk teams, auditors, and counterparties the transparency they need to stand behind the model.”
The model incorporates 10 loan- and borrower-level variables, including credit scores, mark-to-market loan-to-value ratios, debt-to-income ratios, and loan purpose. It also incorporates three macroeconomic drivers.
RiskSpan trained the model on approximately $87 billion in unpaid principal balance across roughly 226,000 Non-QM loans originated or outstanding between January 2018 and August 2025.
The distinction among loan types has become more consequential as Non-QM volume increases. NMP previously reported that delinquencies rose across the Non-QM and non-prime RMBS sector, with Fitch identifying deterioration among newer loan vintages.
Tape Analysis And API Access Included
The release also includes artificial intelligence-powered tools for processing loan tapes and analyzing collateral. Clients can access model results through an application programming interface and integrate the output into their existing systems.
RiskSpan said it plans to add a user-facing backtesting dashboard. Container deployment and additional integrations are also planned for later phases.
Credit Model 7.1 is currently available to RiskSpan Platform and Loans Module clients.
RiskSpan’s model reflects a more granular approach to Non-QM risk: bank-statement, DSCR, full-documentation, and other loans are modeled separately rather than treated as a single collateral category.
For mortgage bankers, that distinction could help identify which documentation types are driving expected losses or prepayments within a loan pool, informing aggregation, securitization, and capital allocation decisions.
*This article was primarily written by a human author. AI tools were used in a limited capacity for research assistance or light editing.
