I uploaded the prospectus for a Deephaven Residential Mortgage Trust deal (DRMT 20XX-X) and, within minutes, Graam's agent produced a full investment committee report.
The agent parsed the cashflow waterfall, extracted the capital structure, and converted the deal into an executable model directly from the document. It then ran the full underwriting workflow — collateral stratification, empirical benchmarking with NQM adjustments, loss and prepayment modeling, and cashflow projections across multiple scenarios.
Once the model was built, I could immediately interact with the deal: adjust pricing assumptions, run stress scenarios, inspect tranche-level cashflows, and evaluate breakeven loss across the capital structure.
Result: all senior tranches (A1–A3) are well-protected under both base and stress assumptions, while the B2 tranche emerges as the marginal credit risk position under elevated loss scenarios.
Here's the full IC-style report generated from that workflow.
Prepared for: Investment Committee
Asset Class: Non-Qualified Mortgage (NQM) RMBS
Deal Reference: @deephaven_nqm
1. Executive Summary
Deephaven Residential Mortgage Trust 20XX-X is a $480.9 million non-qualified mortgage (NQM) securitization. The pool comprises 1,116 residential mortgage loans with a weighted average coupon range of 4.375%–10.375%, a mix of 43.9% fixed-rate and 56.1% ARM collateral, and a notably high self-employed borrower concentration (60.5%). The deal is structured as a sequential-pay transaction with 7 offered tranches (A1 through B3) plus ancillary classes.
Key Pool Metrics
| Metric | Value |
|---|---|
| Issuance Balance | $480,856,235 |
| Loan Count | 1,116 |
| WAC Range | 4.375%–10.375% |
| Fixed/ARM Split | 43.9% / 56.1% |
| Self-Employed Borrowers | 60.5% (primary NQM risk driver) |
| Investment Property Exposure | 19.9% |
Under the base scenario (CDR 1.00% agency proxy, NQM-adjusted), all offered tranches (A1–B3) fully amortize with no principal shortfall. Senior tranches (A1–A3) exhibit short WALs of ~4.2 years at par, while mezzanine and subordinate tranches extend to 13–21 years. The AIOS and XS reference/excess classes show structural shortfalls consistent with their design as excess spread vehicles.
2. Deal Structure
2a. Capital Stack
The deal is structured as a sequential-pay transaction. Principal is distributed A1 → A2 → A3 → M1 → B1 → B2 → B3. Losses are allocated in reverse sequential order (B3 → B2 → B1 → M1 → A3 → A2 → A1).
2b. Subordination Levels
| Tranche | Balance | Coupon | Sub Below ($) | Sub Below (%) | Type |
|---|---|---|---|---|---|
| A1 | $296,688,000 | 2.79% | $184,168,235 | 38.3% | Senior |
| A2 | $32,458,000 | 2.94% | $151,710,235 | 31.5% | Senior |
| A3 | $61,790,000 | 3.05% | $89,920,235 | 18.7% | Senior |
| M1 | $32,457,000 | 3.48% | $57,463,235 | 11.9% | Mezzanine |
| B1 | $25,486,000 | 3.99% | $31,977,235 | 6.6% | Subordinate |
| B2 | $22,600,000 | 4.92% | $9,377,235 | 1.9% | Subordinate |
| B3 | $9,377,235 | SOFR+0bps | $0 | 0.0% | First Loss |
Subordination % = balance of tranches junior to each class / total collateral balance.
2c. Key Structural Features
- Payment Priority: Sequential interest (A1→B3), then scheduled principal, unscheduled principal, and recovery principal
- Excess Turbo: Excess spread directed sequentially to pay down principal (A1→B3) before release
- Excess Release: Residual excess spread released to XS, then R
- Clean-Up Call: Optional redemption when pool balance falls below 10% of original balance
- Triggers: 1 trigger (clean-up call only); no performance-based step-down triggers identified in the model
3. Collateral Analysis
3a. Pool Composition
| Characteristic | Detail |
|---|---|
| Loan Count | 1,116 |
| Issuance Balance | $480,856,235 |
| WAC Range | 4.375% – 10.375% |
| Fixed Rate | 43.9% |
| ARM | 56.1% |
| Lien Position | 100% First Lien |
| Delinquency at Cutoff | 100% Current |
3b. Property Type
| Type | % |
|---|---|
| Single Family Detached | 52.1% |
| Planned Unit Development | 34.7% |
| 2–4 Family | 6.6% |
| Condominium | 5.5% |
| Single Family Attached | 0.9% |
| Townhouse | 0.1% |
3c. Occupancy & Purpose
| Occupancy | % |
|---|---|
| Owner-Occupied | 77.5% |
| Investment Property | 19.9% |
| Second Home | 2.6% |
| Loan Purpose | % |
|---|---|
| Purchase | 58.9% |
| Cash-Out Refi | 31.7% |
| Rate/Term Refi | 9.4% |
3d. Employment Type
| Employment | % |
|---|---|
| Self-Employed | 60.5% |
| Salaried/Wage Earner | 25.9% |
| Unknown | 13.6% |
3e. Key Risk Factors
- Self-Employed Concentration (60.5%): The dominant risk driver. Bank statement and alternative documentation borrowers exhibit materially higher default rates than full-doc borrowers, particularly in economic downturns. Primary basis for NQM multiplier adjustments.
- Investment Property Exposure (19.9%): Non-owner-occupied properties have higher default propensity and lower recovery rates in distress scenarios.
- ARM Exposure (56.1%): Rate reset risk on ARM collateral could elevate defaults if rates rise materially at reset dates.
- Cash-Out Refinance (31.7%): Higher LTV at origination relative to rate/term refis; reduced equity cushion.
- No Initial OC: The deal relies entirely on subordination for credit enhancement with no excess collateral buffer at issuance.
4. Historical Performance (Empirical Benchmarks)
4a. Matched Deal Universe
The empirical engine matched 10 comparable deals (avg FICO 706, avg rate 6.0%) drawn from Freddie Mac SFLLD data across vintages 2010–2025. Seasoning ranges from 8 months (2025 vintage) to 188 months (2010 vintage).
4b. Empirical CNL Milestones (Agency Proxy)
| Month | CNL | CDR (Ann.) | Severity | VPR (CPR) |
|---|---|---|---|---|
| 12M | 0.00% | 1.00% | 25.00% | 3.42% |
| 24M | 0.01% | 1.00% | 25.00% | 10.50% |
| 36M | 0.03% | 1.00% | 25.00% | 17.70% |
| 48M | 0.10% | 1.00% | 38.28% | 14.00% |
| 60M | 0.15% | 1.00% | 31.06% | 10.09% |
Source: 10 matched deals, 119 months of CDR data, 113 months of severity data. These are historical observations, not projections.
Key Observations
- Agency CDR is flat at 1.00% annualized throughout — consistent with high-quality conforming collateral
- Severity rises from 25% in early periods to ~38% at 48 months, reflecting liquidation timing
- VPR peaks at ~17.7% at 36 months, consistent with refinancing activity in the 2018–2022 rate environment
- CNL through 60 months is only 0.15% for agency collateral — NQM adjustments are critical
5. Loss & Prepay Forecast
5a. NQM Adjustment Framework
The agency SFLLD data serves as a base proxy. Per the empirical engine's NQM adjustment guidance, the following multipliers are applied:
| Metric | SFLLD Agency Base | NQM Multiplier | NQM Adjusted | Rationale |
|---|---|---|---|---|
| CDR | 1.00% | ×5.0 | 5.00% | Alt-doc/bank stmt; 60.5% self-employed; investment property |
| CPR | 12.03% avg | ×0.6 | ~7.2% avg | Limited refi access for NQM borrowers |
| Severity | 28.30% avg | ×1.5 | ~42–57% | Higher LTV, less MI, investment property discounts |
5b. Base Scenario Curve Assumptions
| Curve | Average | Range | Data Basis |
|---|---|---|---|
| CDR | 1.00% | [1.00%, 1.00%] | 119 months empirical, 10 deals |
| CPR | 12.03% | [0.66%, 21.31%] | 10 similar deals |
| Severity | 28.30% | [25.00%, 38.28%] | 113 months empirical, 10 deals |
5c. Projected vs. Empirical CNL Comparison
| Milestone | Agency Empirical CNL | Base Projected CNL | NQM Stress Estimate |
|---|---|---|---|
| 12M | 0.00% | ~0.00% | ~0.05% |
| 24M | 0.01% | ~0.01% | ~0.20% |
| 36M | 0.03% | ~0.03% | ~0.50% |
| 48M | 0.10% | ~0.10% | ~1.20% |
| 60M | 0.15% | ~0.15% | ~2.00% |
NQM stress estimates are indicative based on ×5 CDR and ×1.5 severity multipliers; not model-computed.
6. Breakeven Loss Analysis
6a. Subordination-Based Breakeven
| Tranche | Balance | Credit Support (Sub Below) | Breakeven CNL (Est.) | Base CNL (60M) | Cushion (vs. Base) |
|---|---|---|---|---|---|
| A1 | $296,688,000 | 38.3% | ~38.3% | 0.15% | >100× |
| A2 | $32,458,000 | 31.5% | ~31.5% | 0.15% | >100× |
| A3 | $61,790,000 | 18.7% | ~18.7% | 0.15% | >100× |
| M1 | $32,457,000 | 11.9% | ~11.9% | 0.15% | ~79× |
| B1 | $25,486,000 | 6.6% | ~6.6% | 0.15% | ~44× |
| B2 | $22,600,000 | 1.9% | ~1.9% | 0.15% | ~13× |
| B3 | $9,377,235 | 0.0% | 0.0% | First Loss | — |
Breakeven CNL ≈ credit support percentage. Cushion = Breakeven CNL / Projected Terminal CNL (base 60M = 0.15%).
Under the base (agency proxy) scenario: All tranches A1 through B2 have substantial cushion (13× to >100×) against the 0.15% projected 60-month CNL. B3 is the first-loss piece with no protection.
Under NQM stress (CDR ×5, severity ×1.5): Terminal CNL could reach 2–4% over the life of the deal. Even under this scenario, A1–B1 retain meaningful cushion. B2 (1.9% breakeven) would be at risk if NQM-adjusted losses approach or exceed 2%. B3 would absorb first losses immediately.
Key Conclusion: The senior tranches (A1–A3) are well-protected even under severe NQM stress scenarios. The B2 tranche represents the marginal credit risk position in a stress environment.
7. Waterfall & Tranche Analytics
7a. Waterfall Results (Base Scenario)
All offered tranches fully amortize under the base scenario with no principal shortfall:
| Tranche | Opening Balance | Closing Balance | Shortfall |
|---|---|---|---|
| A1 | $296,688,000 | $0 | None |
| A2 | $32,458,000 | $0 | None |
| A3 | $61,790,000 | $0 | None |
| M1 | $32,457,000 | $0 | None |
| B1 | $25,486,000 | $0 | None |
| B2 | $22,600,000 | $0 | None |
| B3 | $9,377,235 | $0 | None |
| XS | $480,856,235 | $281,234,334 | YES (by design) |
| AIOS | $480,856,235 | $480,856,235 | YES (by design) |
7b. Bond Analytics (Base Scenario, Price = 100.00)
| Tranche | Price | Yield | WAL | Notes |
|---|---|---|---|---|
| A1 | 100.00 | 15.88% | 4.21y | Short WAL; high yield reflects excess spread capture |
| A2 | 100.00 | 17.19% | 4.21y | Co-sequential with A1 |
| A3 | 100.00 | 18.09% | 4.21y | Co-sequential with A1/A2 |
| M1 | 100.00 | 9.61% | 12.96y | Mezzanine; extends significantly |
| B1 | 100.00 | 10.71% | 16.38y | Subordinate; long extension |
| B2 | 100.00 | 14.75% | 21.28y | Deep subordinate; very long WAL |
| B3 | 100.00 | -3.22% | 0.00y | First loss; negative yield reflects SOFR+0 coupon vs. loss absorption |
| XS | 100.00 | -14.74% | 6.01y | Excess spread vehicle; negative yield expected |
Yield Observations
- The unusually high yields on A1–A3 (15–18%) at par reflect the model's fast-prepay base case, which compresses WAL and front-loads cashflows to senior tranches
- M1 and B1 yields (9.6–10.7%) are more consistent with mezzanine NQM pricing
- B3's negative yield confirms its first-loss/residual character
8. Purchase Recommendation
Senior Tranches (A1–A3): At par (100.00), the model-implied yields of 15–18% are likely overstated due to the fast-prepay base case. Under NQM-adjusted assumptions (slower CPR, higher CDR), WALs would extend and yields would compress. These tranches carry strong credit protection (18–38% subordination) and are appropriate for investment-grade buyers. Recommend further analysis with NQM-calibrated curves before pricing.
Mezzanine (M1): 9.61% yield at 100.00 with 12.96y WAL and 11.9% credit support. Reasonable risk/return for NQM mezzanine exposure. B2 breakeven (1.9%) provides a meaningful buffer above M1.
B2 Tranche: 14.75% yield at 100.00 with 21.28y WAL and only 1.9% credit support. Under NQM stress (terminal CNL 2–4%), this tranche is at risk of impairment. Not recommended at par without significant price discount.
B3 (First Loss): Negative yield at par; appropriate only for residual/equity investors with full understanding of first-loss exposure.
9. Key Risks
9a. Collateral/Credit Risks
| Risk | Severity | Detail |
|---|---|---|
| Self-Employed Concentration | HIGH | 60.5% of pool; bank statement/alt-doc borrowers have 3–5× higher default rates in stress |
| Investment Property Exposure | MEDIUM-HIGH | 19.9% of pool; higher default propensity and lower recovery rates |
| NQM Proxy Limitation | HIGH | Empirical benchmarks are agency data; actual NQM performance could be 3–5× worse on CDR |
| ARM Reset Risk | MEDIUM | 56.1% ARM collateral; rate resets could elevate defaults |
| Cash-Out Refi Concentration | MEDIUM | 31.7% cash-out refis; higher LTV, reduced equity cushion |
9b. Structural Risks
| Risk | Detail |
|---|---|
| No Initial OC | Zero overcollateralization; all CE is subordination-based |
| No Performance Triggers | No step-down or performance-based triggers; subordination cannot build through OC accumulation |
| WAL Model Calibration | 42/48 WAL tests failed; base scenario uses agency proxy curves that understate NQM risk |
| B3 SOFR+0 Coupon | Floating rate first-loss; in rising rate environment, coupon increases without collateral benefit |
9c. Macro/Rate Sensitivity
| Risk | Detail |
|---|---|
| Home Price Decline | 10–15% HPA decline would push severity from ~28% toward 40–50%, impacting B2 and B3 |
| Unemployment Shock | Self-employed borrowers disproportionately affected; recession could push CDR to 5–8% (NQM-adjusted) |
| Interest Rate Extension | Under NQM-adjusted slow CPR (~7%), WALs for M1–B2 extend to 15–25 years |
| Prepayment Lock-In | NQM borrowers have limited refi access; CPR may not accelerate even if rates decline |
| Rate Shock (–200bps) | CPR response muted (~×0.6 multiplier) vs. agency; limited WAL compression |
9d. Data & Model Caveats
- No loan tape provided: Collateral stratification (FICO distribution, LTV, state concentration) could not be independently verified. Characteristics sourced from prospectus supplement.
- Agency proxy for NQM: The 10 matched deals are all Freddie Mac conforming pools. NQM-specific empirical data was not available in the matched set.
- WAL calibration failure: The base scenario WAL tests failed significantly, indicating the model requires NQM-adjusted curves for accurate analytics. All yield/WAL figures should be treated as indicative pending recalibration.
- AIOS/R analytics unavailable: AIOS pricing failed due to settle date mismatch; R tranche has no cashflows in the model.
This report was generated using empirical data from 10 matched agency deals (Freddie Mac SFLLD, vintages 2010–2025) as a proxy for NQM collateral performance. All projected figures are model outputs and should not be relied upon as investment advice. NQM multiplier adjustments (CDR ×5, severity ×1.5, CPR ×0.6) are recommended before finalizing tranche analytics.