This post provides an update on delinquency rate trends observed in the Non-Agency mortgage market with a deep dive on different segments of the fast growing Non-QM mortgage market. All of the figures in this post are based on queries of historical CoreLogic Non-Agency data via our proprietary RiskSpan Edge Historical module.

After reaching post-Covid highs in May 2025, delinquency rates have stabilized at slightly lower levels in August 2025, the most recent factor date available from CoreLogic: 

Figure 1. 

Figure 2.

Figure 3 shows the relative delinquency performance of mortgages across 4 segments of the Non-QM population, which represents the largest portion of the PLS 2.0 market. While loans with full documentation represent the largest segment of this market from a total outstanding balance perspective, originations have been shifting towards DSCR/Investor and Bank statement loans since 2022 (see Figure 4). In 2025, the combined volume of originations in the DSCR/Investor and Bank statement segments was about four times the volume of loans originated with full documentation. 

Figure 3. 

Figure 4. 

Figures 5 and 6 show the relative delinquency performance of Non-QM mortgages by year of origination. For these charts, we exclude vintages prior to 2021 to avoid the distorting impact of the COVID delinquency shock. 

Figure 5 shows the 60+ delinquency rate for each vintage by factor date. 

Figure 6 shows the 60+ delinquency rate for each vintage by loan age.

Figure 5. 

Figure 6.

Given the elevated delinquency rates of Non-QM mortgages relative to Agency and Prime Jumbo mortgages and the backdrop of housing and macroeconomic uncertainty, it is important for investors to monitor their portfolios that have Non-QM exposure. Our credit models at RiskSpan model these delinquency roll rates directly, and our modeling team calibrates our suite of models to capture both the overall trends and the differentiated performance across loan and product types. These models are just one component of our scaled analytics solutions to help our clients evaluate risk and make investment decisions.

Register here for next month’s call: Thursday, August 21st, 2025, 1 p.m.

Each month, we host a Models & Markets call to offer our insights into recent model performance, emerging credit risks, and broader economic indicators. This month, as interest rates remain elevated and economic uncertainty persists, we addressed how both conventional and AI-based modeling techniques are shaping decision-making processes across agency, non-QM, and ARM products.

Here’s a quick recap in case you missed it.

(Click here to listen to the entire 30-minute recording, or continue reading for a summary.)

Model Performance: Prepayment Dynamics in Focus

RiskSpan’s prepayment model continues to perform well based on benchmarking against actuals across coupon stacks. The team noted:

New ARM Model Launch

An enhanced ARM Prepayment Model (v3.8) is now live in production. It exhibits refined sensitivity to rate shocks and aims to provide improved accuracy for adjustable-rate portfolios in today’s volatile environment.

Claude the Research Assistant: AI in Action

One of the highlights of the call was a deep dive into how we are testing Claude (Anthropic’s well-known LLM) as a mortgage research assistant.

Using a dataset from RiskSpan’s Snowflake instance, Claude orchestrated an end-to-end analytical workflow, including:

Key Insights from Claude’s Analysis

Claude surfaced several noteworthy trends:

Claude correctly reasoned with the provided data but could not identify some features (like “Spread at Origination”). This raises interesting questions about LLMs’ capacity to reason beyond their training corpus.

Market Outlook: Economic Signals Turning Cautionary

The macro backdrop continues to weigh on securitization and borrower behavior. Highlights from July’s indicators:

On the Horizon

Contact us to learn more.

As generative AI becomes a powerful tool in Private asset-backed finance (ABF), the need for precision and transparency is more critical than ever. At RiskSpan, we’re applying Large Language Models (LLMs) to automate and accelerate private ABF deal modeling and surveillance. But speed is only half the battle—accuracy is non-negotiable.

That’s where Human-in-the-Loop (HITL) validation plays a vital role. While the RiskSpan platform incorporates sophisticated AI guardrails, we believe the right blend of automation and expert oversight ensures results that are not just fast—but reliable, auditable, and production-ready.

The AI-Powered Workflow: What’s Automated

Our private ABF modeling and surveillance system uses LLMs to handle several critical tasks:

But What About Hallucinations?

Generative models are powerful but imperfect. Without the right controls, they can fabricate securitization tranches or fees that are not present in the deal documents. They can also misinterpret waterfall rules or omit critical override conditions or generate semantically incorrect code for cashflow models. To address these challenges, RiskSpan employs a multi-layered safeguard framework, combining asset class based extraction; LLM-as-Judge; Rule-Based Guardrails and Inline Human Review

Humans in the Loop: Three Layers of Oversight

We’ve embedded human validation at three key points in the deal lifecycle:

  1. Pre-Modeling Validation: before LLM-generated outputs are finalized, RiskSpan analysts review extracted terms and model prompts—correcting anything misaligned with the source documents.
  2. Inline Oversight: during waterfall code generation, humans validate AI-generated logic in context, ensuring correct treatment of subordination, triggers, caps/floors and other.
  3. Post-Deployment Monitoring Surveillance: outputs are reviewed both by the RiskSpan team and client-side structuring or credit teams. Feedback is looped back into model tuning and prompt optimization.

Looking Ahead: RAG and Continuous Improvement

We’re actively exploring Retrieval-Augmented Generation (RAG) to reduce hallucinations even further. By grounding AI responses in deal-specific material such as offering documents, trustee reports, and internal risk memos—we aim to: 1) eliminate off-topic responses. 2) increase trust in model-derived outputs and 3) enable deeper customization per issuer or asset class.

The Takeaway

AI is transforming how private ABF deals are modeled and monitored—but it must be grounded and guided by human expertise and built for institutional rigor. At RiskSpan, we’re not just accelerating workflows—we’re raising the bar for accuracy and trust in AI-assisted private structures. Human-in-the-loop is not a fallback—it’s a strategic pillar. Want to see how our AI platform works in action? Reach out to schedule a demo or contact your RiskSpan representative to learn more.

Register here for next month’s call: Thursday, July 17th, 2025, 1 p.m.

Each month, we host a Models & Markets call to offer our insights into recent model performance, emerging credit risks, and broader economic indicators. This month, we showcased our responsiveness to shifting macroeconomic dynamics and introduced new transparency elements (i.e., back-testing tools) to our prepayment and credit modeling.

Click here to listen to the entire 23-minute recording, or continue reading for a summary.

Agency Prepay Model: Back-testing and Enhanced Control

We are launching a new loan-level prepayment back-testing tool using nearly all agency loans (FN/FH/GN) aged 10 years or less. The tool runs every month through our models with historical home prices and interest rates. Based on this data, we have an interactive dashboard that will allow users to drill down into model performance with far more granularity than currently possible.

Key Enhancements to Prepay Model v. 3.8

A soon to be released version of the prepay model will include:

A redesigned ARM prepayment framework, derived from the fixed-rate model. The new ARM component includes:

These updates give users the ability to more precisely tune model responses under a variety of macroeconomic and borrower scenarios.

Credit Model: V7 and Delinquency Transitions

The delinquency transition matrix incorporated into our new Credit Model V7 provides users a more nuanced credit risk assessment. This model works in conjunction with the enhanced prepayment model to better simulate the joint dynamics of default and prepay behavior across economic cycles.

Macroeconomic Context: Rates and Risk in a Holding Pattern

We remain cautious in our outlook for the remainder of 2025 and into 2026:

The Fed Funds Rate is expected to remain elevated—currently in the 4.25–4.50% range—with the first rate cut likely in September. By year-end 2025, the market expects it to settle around 3.75–4.00%.

Mortgage rates remain stubbornly high, hovering above 6.5%, putting pressure on origination volumes and reinforcing the value of accurate prepayment modeling.

Home prices and broader macro indicators like unemployment and PCE inflation remain stable, suggesting a “wait-and-see” mode for both consumers and investors.

What’s Next: More Models, More Tools, More Insights

We continue to expand our Platform with new analytics, model documentation, and client-facing tools. Users can soon access the new back-testing report directly within the Platform, alongside these updated prepayment and credit models. These developments reflect our commitment to model transparency, data-driven innovation, and practical tools for real-time market adaptation.

Contact us to learn more.

Limited Partners (LPs) continue to demand better data, faster, and with full transparency. At this week’s Private Credit Tech Summit in New York, I moderated a panel of industry leaders for a discussion on where LP expectations are heading and the challenges managers face trying to meet them.

My fellow panelists included:

·      Charlie Tafoya, Co-Founder and CEO of Chronograph

·      Marios Tsiptis, Senior Portfolio Manager

·      Arnab Mazumdar, Partner at Pantheon

We explored the evolving expectations of LPs, the operational hurdles General Partners (GPs) face, and the technology shaping the next frontier of data transparency in private credit.

What follows is where we landed.

LP Expectations Are Outpacing the Status Quo

Quarterly performance reports and aggregate numbers used to be enough. But Marios Tsiptis explained that today’s LPs (particularly insurance companies) want detailed, timely insight into exactly what they own and what risks they’re carrying.

From Pantheon’s fund-of-funds vantage point, Arnab Mazumdar laid out three foundational data pillars:

1.     Performance metrics

2.     Operating and credit-level data

3.     Consistency across managers

Lagging information and inconsistencies across investment structures, especially between SMAs and feeder funds, create significant friction. Structure matters, and data must be complete, accurate, and delivered on LP terms.

The Operational Reality: Bridging Ambition and Execution

Charlie Tafoya, who works closely with GPs via Chronograph, provided a sobering view of the day-to-day realities. Many managers are eager to modernize, but they’re grappling with:

·      Delays in data delivery

·      Data quality and validation issues

·      Fragmented internal processes

There’s just no getting around the fact that investment teams simply must be embedded in the data flow. Too often, front-office insights are siloed from the operations teams responsible for reporting. “The investment team is the source of truth,” Charlie noted, making their engagement essential to any successful data transformation.

Meanwhile, Arnab called for a convergence of internal monitoring and external reporting. Aligning what GPs see internally with what they share externally could yield benefits across the board—but cultural and technological hurdles remain.

Technology as Enabler, Not Panacea

Tools that support data interoperability, real-time reporting, and workflow automation are rapidly maturing. But challenges persist around integration with legacy systems and data standardization.

From the LP side, Marios painted a picture of an ideal future. Intuitive dashboards, seamless access across portfolios, and fully integrated delivery pipelines all featured prominently in this future. But the road leading there is still under construction.

According to Arnab, the next few years will hinge on industry standardization and early adopters gaining a competitive edge, while laggards risk being left behind.

Transparency builds trust—but it requires real operational change.

More than just a tech problem, this requires a cultural shift. GPs and LPs need to work as true partners in designing data ecosystems that are not only robust and scalable, but also reflect the growing sophistication of the private credit space.

As the private markets evolve, so too must the infrastructure that supports them. Those who succeed in this transition will not only meet the expectations of today’s LPs, but also shape tomorrow’s.

AI is revolutionizing everything, and the UX design process is no exception. From the earliest conceptual ideas all the way through to final execution, the transformation is not just about speeding up workflows but also about enhancing creativity and collaboration.  

Here’s how. 

Initial Ideation

Every UX journey begins with the ideation process. AI tools like Claude have become a go-to starting point for brainstorming and generating initial design prompts. By feeding basic requirements and user journeys into the AI, I can quickly generate a list of potential features and pain points. For example, when working on a new ETL tool, Claude helped identify potential difficulties in data mapping, handling large datasets, and ensuring data accuracy during the transformation process. These pain points helped Claude generate a list of requirements and user journeys, which were then used to create a first-pass prototype 

This initial step is crucial as it sets the foundation for the entire design process. 

Rapid Prototyping

Once the ideation phase is complete, the next step is creating a first-pass prototype. Claude has helped me here by generating quick, functional prototypes that provide a visual representation of the overall application. Although not fully functional, these prototypes nevertheless offer a solid starting point for further refinement. This rapid prototyping capability allows me to iterate quickly and incorporate feedback more efficiently. 

After the initial prototype is created, I import it into Figma for refinement. This is where the design gets polished with logos, color schemes, and other branding elements. This is a highly collaborative phase of the process, where designers work closely with developers and test users to finalize the look and feel of the application. This step ensures that the design is not only functional but also visually appealing. 

Code Development

The final stage involves turning the refined design into a working application. Here, remarkably, AI tools like Claude and Cursor (an AI-enhanced version of VS Code) can actually generate and refine the code itself. By providing the AI with an image of the final design, it can produce a close approximation of the user interface, which can then be fine-tuned by developers. For example, I might ask Claude to generate a sample layout based on the refined design and then use Cursor to make specific changes, such as adjusting font sizes and colors. This significantly reduces the time and effort required to build the front end of the application. 

Real-World Application and Testing — Collaboration and Continuous Improvement

The iterative nature of AI tools allows for rapid prototyping and testing, leading to a more efficient development cycle. While AI-generated code might not be perfect, the ability to quickly identify and fix bugs makes the process much faster than traditional methods. For instance, I used Cursor to highlight and fix errors in the code by simply providing and asking it to correct the issues. 

But collaboration remains supremely important. AI tools facilitate cross-functional teamwork by making it easier to share prototypes and gather feedback. This collaborative approach ensures that the final product meets the needs of all stakeholders. Additionally, the iterative nature of AI tools means that the design can continuously evolve based on user feedback and testing. 

—————- 

AI is not just a tool for speeding up the UX design process; it’s a catalyst for innovation and collaboration. By leveraging AI for ideation, prototyping, and code development, designers can create smarter, more efficient workflows that lead to better user experiences. The future of UX design is not just about working faster but also about working smarter. 

As securitization models become increasingly complex and differentiated, validation becomes a critical challenge. We’ve experimented with an innovative approach that leverages large language models (LLMs) as impartial judges to validate models implemented across different platforms.

The Dual-Implementation Challenge

In cash flow modeling, we often maintain parallel implementations—typically in Python for flexibility and Excel for transparency. How do we ensure both versions produce consistent results?

Enter the “LLM as Judge” approach!

A Real-World Case Study: Residential Transition Loan Funding

Consider a portfolio of residential transition loans with a funding structure including:

We implemented this structure in both Python and Excel, then submitted both models to an LLM for validation.

The LLM Validation Process

The LLM first analyzed the conceptual alignment between models, confirming both followed the same fundamental approach to cash flow projection, default assumptions, reserve mechanics, and triggers.

Next came a rigorous numerical comparison. The LLM detected a $100,000 investor distribution discrepancy in Month 2:

Through logical analysis, the LLM determined this likely stemmed from differently evaluated trigger conditions. This kind of subtle implementation difference could easily go unnoticed in manual validation, potentially leading to significant valuation discrepancies over time.

Beyond Discrepancy Detection

The true power of this approach extends beyond finding differences. The LLM also provided:

  1. Stress testing recommendations tailored to our specific product, including scenarios for rapid defaults, extension waves, and interest rate shocks
  2. Model risk management insights highlighting documentation needs and suggesting a formal reconciliation process
  3. Code quality assessment noting strengths and weaknesses in both implementations

Why This Matters

For securitization professionals, this approach offers several advantages:

Contact us to discuss.

This is a monthly update on non-QM delinquency rate and roll rate trends based on the March 2025 remittance data. Similar to last month’s post, I use the CoreLogic Non-Agency loan data to split out the Non-QM population by loan type. I compare the relative delinquency performance of mortgages backed by Investor properties vs. loans with full documentation vs. other Non-QM loan types (this last bucket comprises mainly Bank Statement loans). I use a slightly revised and more inclusive definition of Non-QM this month so the overall balance figures are higher and delinquency rates and roll rates are slightly lower than those reported in last month’s post.

The first chart shows that the non-performing delinquency rate (60+ dpd loans as a percentage of the overall population) has risen from a post-COVID low of 0.85% in July 2022 to 3.09% as for the most recent remit month. This increase has been driven by deterioration in the credit performance across all Non-QM loan types. Notably, the delinquency rate for Investor loans increased to 3.56% as of March, up more than three-fold from post-COVID lows of 1.1% in October 2022 and up 91bp year over year. Full Doc Non-QM loans continue to outperform other segments significantly, but their delinquency rates still rose to 0.85%, a new post COVID recovery high.

The other driver of the increase in delinquency rates for the Aggregate Non-QM loan population is a gradual shift in their mix away from the Full Doc loans, which have a better credit profile. As shown in the graph below, Full Doc loans as a percentage of the overall NQM mix have fallen from over 50% of NQM population as of the end of 2018 to only just under 31% in March. Meanwhile, Investor loans have increased from only 3% of the Non-QM population as of the end of 2018 to 10% just before COVID to over 24% as of March.

The last graph considers the gateway transition of mortgages to non-performing status: the current to 30 roll rates, or the percentage of current loans that roll to 30 days delinquent in any given month. These trends are broadly in line with what we see for the overall delinquency rates: roll rates have increased significantly since their post COVID lows.

In the March remittance data, overall Non-QM C->30 roll rates increased to 1.18%, their highest level since December 2020. All 3 non-QM segments broken out in this graph also hit new post COVID highs, with Investor-backed loans experiencing a 1.43% C->30 roll rate, 2.9x the 0.50% roll rate experienced by Full Doc Non-QM loans.

As non-QM mortgages show signs of growing distress amid broader economic uncertainty, we recommend heightened vigilance for investors and risk managers with Non-QM exposure in their portfolios. RiskSpan’s credit models forecast delinquency roll rates directly, and our modeling team calibrates our suite of models to capture both the overall trends and the differentiated performance across loan and product types. These models are just one component of our scaled analytics solutions to help our clients evaluate risk and make investment decisions.

Contact me to discuss.

When agentic AI first hit the scene, we were intrigued—but skeptical. Was this just another over-hyped trend or something that could drive real value?

Fast forward a few months, and we’ve got our answer.

At RiskSpan, we’ve quietly integrated AI agents into our internal workflows through a dedicated Agent Desktop. These agents are now core to how we manage our business—monitoring client health, tracking system usage and perhaps most impressively, performing deep research across the massive datasets we store. What began as an experiment has become indispensable.

The real breakthrough is manifest, however, when Agents proactively uncover insights, flag anomalies, and automate routine analyses. Our Dev and Client teams are saving hours and making faster, more informed decisions because relevant information finds them.

Coming soon, our clients will be able to use Agents in the RiskSpan Platform to query their own data, analyze GSE performance data and run on-demand analysis instantly—all without waiting on a queue or building custom reports.

Designed for portfolio risk, surveillance, analyzing loan-level data, or exploring market trends, the AI agents will help you go from question to answer in seconds. Check out a sample below and reach out to learn more!

This is a follow-up to Bernadette Kogler’s short piece last month on stress in the Non-QM mortgage market. In this post, I use the CoreLogic Non-Agency loan data to split out the Non-QM population by loan type and look at the relative delinquency performance of mortgages backed by Investor properties vs. loans with full documentation vs. other Non-QM loan types (this last bucket comprises mainly Bank Statement loans).

As the following chart illustrates, the non-performing delinquency rate (60+ dpd loans as a percentage of the overall population) has risen from a post-COVID low of 1.01% to 3.59% as of March 2025. This increase has been driven by deterioration in the credit performance across all Non-QM loan types. Notably, the delinquency rate for Investor loans increased to 3.82% as of March, up more than three-fold from post-COVID lows of 1.1% in October 2022. While they remain the best performing loan type, even the Full Doc loans have seen a doubling of delinquency rate, to 1.11%.

The other driver of the sharp uptick in delinquency rates for the Aggregate Non-QM loan population is a shift in their mix away from the strongly performing Full Doc loans. As shown in the graph below, Full Doc loans as a percentage of the overall NQM mix have fallen from over 50% of NQM population as of the end of 2018 to only 22% in March. Meanwhile, Investor loans have increased from only 3% of the Non-QM population as of the end of 2018 to 10% just before COVID to 28% as of March.

Finally, we look at the gateway transition of mortgages to non-performing status: the current to 30 roll rates, or the percentage of current loans that roll to 30 days delinquent in any given month due to a missed payment. Not surprisingly, these trends are broadly in line with what we see for the overall delinquency rates: roll rates have increased significantly since their late 2022 lows.

But these roll rates give us a more real-time perspective on how different loan types are performing relative to each other than the delinquency rate levels, which represent the cumulative effect of historical performance. In the most recent remittance data, Investor-backed loans experienced a 1.42% C->30 roll rate, which was 2.5x the 0.58% roll rate experienced by Full Doc Non-QM loans. By contrast, that multiple was only 1.8x in October 2022 when NQM loans were experiencing their lowest post-COVID roll rate performance.

Given the deteriorating performance of Non-QM mortgages and backdrop of macroeconomic uncertainty, it is important for investors to monitor their portfolios that have Non-QM exposure. Our credit models at RiskSpan model these delinquency roll rates directly, and our modeling team calibrates our suite of models to capture both the overall trends and the differentiated performance across loan and product types. These models are just one component of our scaled analytics solutions to help our clients evaluate risk and make investment decisions.

Contact me to discuss.