The CS CRT Compendium

Transcription

1 Americas Fixed Income Research Securitized Products FOR INSTITUTIONAL CLIENTS ONLY The CS CRT Compendium Research Analysts Marc Firestein Jonathan Corwin Freddie Mac s STACR 2013-DN1 was a landmark transaction for the nonagency mortgage market for a number of reasons. It was the first post-crisis transfer of mortgage credit risk on newly originated mortgage loans. It provided the GSEs an outlet for lowering their credit exposure in their mortgage portfolio. Pricing levels gave insight on the market level of guarantee fees. The STACR structure, in its most important components, calls on similarly built transactions from Freddie s past. The MODERNs came nearly 20 years prior where Freddie bought credit protection on newly originated loans and used a fixed severity schedule. The RESIF and ARMOR programs, used by other entities, had similar structures for portfolios of loans. The first STACR transaction brought a two-tranche sequential structure with no ratings. Loans are removed from the pool if they reach 180 days delinquent; a fixed severity schedule, varying with cumulative defaults, determines losses. The M1 and M2 are uncapped LIBOR floaters, which priced at +340 and +715, respectively. Freddie retained the bottom 30 basis points of risk as well as some of the M1s and M2s (through pari-passu, non-interest-bearing exposures). The next transactions, including the first deal from Fannie Mae s CAS program, brought rated tranches, three tranche structures, and MI-covered, higher-ltv pools. In 2015, Freddie Mac began selling first-loss pieces to investors and brought the first actual loss transaction. Fannie Mae followed suit later in The GSEs have issued over $25B in CRT transactions to date, of which over $10B carries an IG rating. These transactions have transferred credit risk on roughly $850B in UPB since the program s inception. In addition, Freddie and Fannie have received reinsurance on some of their retained M tranches, as well as MI pool policies on other loans. (Please see the appendix for issuance and pricing data.) Hand in hand with new transactions and new structures comes new valuation techniques and new approaches to comparing different tranches. The combination of credit and convexity in CRT structures presents an interesting challenge which we have delved into at length. This compendium includes every piece we have written on CRT, in addition to issuance data and pricing spreads. We intend to update this document as we publish more pieces on this expanding market. DISCLOSURE APPENDIX AT THE BACK OF THIS REPORT CONTAINS IMPORTANT DISCLOSURES AND ANALYST CERTIFICATIONS. CREDIT SUISSE SECURITIES RESEARCH & ANALYTICS BEYOND INFORMATION Client-Driven Solutions, Insights, and Access

2 Appendix: Issuance data Figure 1: Deal Spreads Data as of issuance Deal (Group) Risk Transferred ($B Original Deal Balance) Source: Credit Suisse, Fannie Mae, Freddie Mac Pricing Date M1 Spread M2 Spread M3 Spread B Spread STACR 2013-DN /23/ N/A N/A STACR 2013-DN /7/ N/A N/A CAS 2013-C /22/ N/A N/A CAS 2014-C /14/ N/A N/A CAS 2014-C02 (G1) /21/ N/A N/A CAS 2014-C02 (G2) /21/ N/A N/A CAS 2014-C03 (G1) /17/ N/A N/A CAS 2014-C03 (G2) /17/ N/A N/A CAS 2014-C04 (G1) /19/ N/A N/A CAS 2014-C04 (G2) /19/ N/A N/A CAS 2015-C01 (G1) /19/ N/A N/A CAS 2015-C01 (G2) /19/ N/A N/A CAS 2015-C02 (G1) /19/ N/A N/A CAS 2015-C02 (G2) /19/ N/A N/A CAS 2015-C03 (G1) /16/ N/A N/A CAS 2015-C03 (G2) /16/ N/A N/A CAS 2015-C04 (G1) /21/ N/A N/A CAS 2015-C04 (G2) /21/ N/A N/A STACR 2014-DN /6/ N/A STACR 2014-DN /1/ N/A STACR 2014-DN /6/ N/A STACR 2014-DN /28/ N/A STACR 2014-HQ /6/ N/A STACR 2014-HQ /11/ N/A STACR 2014-HQ /28/ N/A STACR 2015-DN /29/ ,150 STACR 2015-HQ /24/ ,075 STACR 2015-DNA /22/ STACR 2015-HQ /2/ STACR 2015-DNA /24/ STACR 2015-HQA /24/ STACR 2015-DNA /3/ STACR 2015-HQA /2/ ,050 The CS CRT Compendium 2

3 Figure 2: Issuance data Data as of issuance, $MM Source: Credit Suisse, Fannie Mae, Freddie Mac Deal Note issuance IG UPB STACR 2013-DN ,584 STACR 2013-DN ,327 CAS 2013-C , Total 1, ,667 CAS 2014-C ,309 STACR 2014-DN1 1, ,441 STACR 2014-DN ,147 CAS 2014-C02 1, ,818 CAS 2014-C03 2, ,224 STACR 2014-DN ,746 STACR 2014-HQ ,974 STACR 2014-HQ ,434 STACR 2014-DN ,741 STACR 2014-HQ ,001 CAS 2014-C04 1, , Total 10,764 5, ,709 STACR 2015-DN ,644 CAS 2015-C01 1, ,193 STACR 2015-HQ ,552 STACR 2015-DNA1 1, ,876 CAS 2015-C02 1, ,009 STACR 2015-HQ ,325 STACR 2015-DNA ,986 CAS 2015-C03 1, ,326 STACR 2015-HQA ,377 CAS 2015-C04 1, ,045 STACR 2015-DNA3 1, ,706 STACR 2015-HQA , Total 12,579 5, ,142 The CS CRT Compendium 3

4 Appendix: Performance Snapshot Figure 3: CRT performance summary As of 11/2015 remittance Deal (Group) WALA (mos) WAC (%) CPR (1mo) CPR (3mo) CPR (6mo) Source: Credit Suisse, Remittance Reports, the BLOOMBERG PROFESSIONAL service DQ (bps) 60+ DQ (bps) Cum Credit Events (bps) CAS 2013-C CAS 2014-C CAS 2014-C02 (G1) CAS 2014-C03 (G1) CAS 2014-C04 (G1) CAS 2015-C01 (G1) CAS 2015-C02 (G1) CAS 2015-C03 (G1) CAS 2015-C04 (G1) CAS 2014-C02 (G2) CAS 2014-C03 (G2) CAS 2014-C04 (G2) CAS 2015-C01 (G2) CAS 2015-C02 (G2) CAS 2015-C03 (G2) CAS 2015-C04 (G2) STACR 2013-DN STACR 2013-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2015-DN STACR 2015-DNA STACR 2015-DNA STACR 2015-DNA STACR 2014-HQ STACR 2014-HQ STACR 2014-HQ STACR 2015-HQ STACR 2015-HQ STACR 2015-HQA The CS CRT Compendium 4

5 Jan-14 Mar-14 May-14 Jul-14 Sep-14 Nov-14 Jan-15 Mar-15 May-15 Jul-15 Sep-15 Nov-15 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 Jun-15 Jul-15 Aug-15 Sep-15 Oct-15 Nov-15 CPR CPR Nov-13 Jan-14 Mar-14 May-14 Jul-14 Sep-14 Nov-14 Jan-15 Mar-15 May-15 Jul-15 Sep-15 Nov-15 Jun-14 Aug-14 Oct-14 Dec-14 Feb-15 Apr-15 Jun-15 Aug-15 Oct-15 CPR CPR 2 December 2015 Appendix: Prepayment Speeds Figure 4: CAS <80 OLTV transactions (groups) C C C02 (G1) 2014-C03 (G1) 2014-C04 (G1) 2015-C01 (G1) 2015-C02 (G1) 2015-C03 (G1) 2015-C04 (G1) Figure 5: CAS >80 OLTV transactions (groups) C02 (G2) 2014-C03 (G2) 2014-C04 (G2) 2015-C01 (G2) 2015-C02 (G2) 2015-C03 (G2) 2015-C04 (G2) Source: Credit Suisse, remittance reports Source: Credit Suisse, remittance reports Figure 6: STACR <80 OLTV transactions Figure 7: STACR >80 OLTV transactions DN DN DN DN DN DN DN DNA DNA DNA HQ HQ HQ HQ HQ HQA1 0 0 Source: Credit Suisse, remittance reports Source: Credit Suisse, remittance reports The CS CRT Compendium 5

6 Cum. Credit Events (bps) Cum. Credit Events (bps) Cum. Credit Events (bps) Cum. Credit Events (bps) 2 December 2015 Appendix: Cumulative Credit Events Figure 8: CAS <80 OLTV transactions (groups) C C C02 (G1) 2014-C03 (G1) 2014-C04 (G1) 2015-C01 (G1) 2015-C02 (G1) 2015-C03 (G1) 2015-C04 (G1) Mos. Since Origination Source: Credit Suisse, remittance reports Figure 9: CAS >80 OLTV transactions (groups) C02 (G2) C03 (G2) C04 (G2) C01 (G2) C02 (G2) C03 (G2) C04 (G2) Mos. Since Origination Source: Credit Suisse, remittance reports Figure 10: STACR <80 OLTV transactions DN DN DN DN DN DN DN DNA DNA DNA Mos. Since Origination Source: Credit Suisse, remittance reports Figure 11: STACR >80 OLTV transactions HQ HQ HQ HQ HQ HQA Mos. Since Origination Source: Credit Suisse, remittance reports The CS CRT Compendium 6

7 Table of contents Appendix: Issuance data 2 Appendix: Performance Snapshot 4 Appendix: Prepayment Speeds 5 Appendix: Cumulative Credit Events 6 Relative Value Frameworks 8 Breakup Value a component framework for CRT... 8 What do GSE credit risk bonds tell us about housing risk? An IOS framework for STACR/CAS Collateral Performance 18 CRT, revisited deal performance and high LTV tiering A look at CRT prepayment exposure Fat tails CRT prepays driven by higher loan balances and WAC dispersion CRT prepays flip the script Fast and furious drift the impact of HPA on high-ltv CRT transactions Historical pieces/introductions 33 STACR Implications for g-fees and the Future Mortgage Credit Market Exploring the fair basis between STACR and CAS GSE Risk Sharing capital market execution is more efficient than MI Pool Policy The CS CRT Compendium 7

8 This is an exact excerpt from the Global Securitized Products Weekly, published 09 April Relative Value Frameworks Breakup Value a component framework for CRT The CRT transactions present a challenge in evaluating relative value across the various deals. Given the pricing differences across each deal, as well as the prepayment and credit variance, the possibilities for relative value frameworks exist. However, they tend to focus on either the credit or prepayment aspect of CRT, but rarely both. Herein, we present a CMO-inspired framework for comparing tranches up and down the stack in STACR transactions. We had previously compared CRT LCFs on a coupon swap basis, using IOS as a benchmark. STACR 2013-DN2 saw the introduction of MACRs, whereby investors could exchange the floaters for lower-margin floaters and an IO. With these MACR IOs, we can apply IO valuations that incorporate the structure of the STACR transactions. We first estimate the margin on a comparable agency-guaranteed par-priced floater using agency CMO benchmarks. Given the margin on each STACR bond, this leaves an excess IO remaining, which we can value using an equal OAS approach to agency IOs. This creates a benchmark price for a theoretical agency-guaranteed high-coupon floater, which controls for prepayment exposure and provides a unique comparison for each STACR tranche. We compare this benchmark price to the market price for each STACR bond to calculate the market-implied discount from our theoretical agency-guaranteed high-coupon floater. Even though MACR IOs do not exist in CAS transactions, we acknowledge that a similar approach could be used on CAS deals if the IO components are modeled out separately. Although a number of factors could cause the market-implied discount calculated here, it allows for a powerful comparison of transactions on a relative basis. A relatively smaller discount would imply a more favorable view of collateral, while a relatively larger discount would imply a more negative view. We believe that this metric can expose the market s view on credit across different transactions and tranches as comparing market-implied discounts controls for CRT market technicals. Additionally, this approach can be used on new deals as they come to market to compare new tranches to seasoned ones. This analysis, at current market levels, indicates that the market prefers the credit exposures in STACR 2014-DN4 over the exposures in 2014-DN1 and DN2, for example. However, we can take this result in a more general light. The market ascribes a significant premium to faster-paying transactions, as their market-implied discount to the theoretical agency high-coupon floater is much smaller. Finding the benchmark creating the comparable agency floater The calculation of the theoretical high-coupon benchmark requires a few steps. Given that we are creating a comparable cash flow that is derived from agency benchmarks, we will assume no losses on the structure. The first step is to use our agency model to project out the weighted-average life for each tranche and to assign comparable IOS tranches based on WAC and WALA (Figure 12). The CS CRT Compendium 8

9 Figure 12: STACR tranche WALs and IOS counterparts Projected WAL based on agency prepayment models Tranche Source: Credit Suisse, Freddie Mac Projected WAL WAC WALA IOS Counterpart STACR 2014-DN1 M STACR 2014-DN2 M STACR 2014-DN3 M STACR 2014-DN4 M STACR 2014-HQ1 M STACR 2014-HQ2 M STACR 2014-HQ3 M STACR 2015-DN1 M STACR 2015-HQ1 M STACR 2013-DN2 M STACR 2014-DN1 M STACR 2014-DN2 M STACR 2014-DN3 M STACR 2014-DN4 M STACR 2014-HQ1 M STACR 2014-HQ2 M STACR 2014-HQ3 M STACR 2015-DN1 M STACR 2015-HQ1 M STACR 2013-DN2 M STACR 2014-DN1 M STACR 2014-DN2 M STACR 2014-DN3 M STACR 2014-DN4 M STACR 2014-HQ1 M STACR 2014-HQ2 M STACR 2014-HQ3 M STACR 2015-DN1 M STACR 2015-HQ1 M The next step is to build a similar-coupon bond using agency CMO levels. We first create a par-priced floater using an estimated DM, given the WAL of the tranche. The difference between the STACR bond s coupon and the estimated DM of the par-priced floater leaves behind an excess IO. To value the excess IO, we apply a multiple found by using the IOS benchmark to solve for an equal OAS dollar price on the IO MACR. We aggregate these two components to estimate the dollar price of the comparable agency-guaranteed highcoupon floater. The CS CRT Compendium 9

10 Figure 13: Walkthrough of calculating equivalent agency-guaranteed highcoupon floater price and corresponding market implied discount Indicative prices, market levels for OAS and CRT as of 4/7/15 Source: Credit Suisse, Freddie Mac STACR 2014-DN2 M2 Breakdown Margin (bp) 165 A WAL (years) 4.3 B Equivalent Par Floater Margin (bp) 30 C Excess IO (bp) 135 D = A-C OAS on IOS Counterpart (bp) 162 E MACR IO Tranche Coupon (bp) 75 F MACR IO Price at equal OAS 2.50 G MACR IO multiple 3.33 H = G/F Equal OAS Excess IO Price 4.49 I = H*D Implied Agency Floater Price J = I+100 Current Market Value K Market-Implied Discount 5.23 L = J-K As an example, we have provided a full walkthrough of the calculation for STACR DN2 M2 (Figure 13). Our model estimates the M2 cash flow to have a 4.3-year WAL at current rate levels; comparable WAL par-priced agency floaters price indicatively around 30 DM. The coupon on STACR 2014-DN2 M2 is L+165, so the par-priced agency floater leaves us with 135 bp of excess IO. The CRT structure requires cash flow projections on the excess IO to properly ascribe a multiple to it, particularly in the IG tranches. We can use the STACR 2014-DN2 M2I, a 75 bp IO strip off of the M2, to model out the excess IO. Using the OAS on IOS 3.5s of 2013, we can solve for the price on the M2I and, therefore, the IOS-implied multiple. We can then apply this multiple to the 135 bp excess IO and add the value to that of the parpriced floater to find the theoretical agency high-coupon floater price that we use to benchmark STACR 2014-DN2 M2. Outcomes higher-speed tranches trade at a premium We can apply this methodology to every STACR deal since 2013-DN2, as all of those transactions have these IO MACRs available. This allows us to compare tranches across STACR transactions on a more even footing because it controls for both tranche margin and the prepayment exposure underlying them. Figure 14: Average concession by transaction type and bond rating Indicative prices, market levels for OAS and CRT as of 4/7/15 Source: Credit Suisse, Freddie Mac Rating DN HQ Aggregate A Rated BBB Rated BB & Unrated Generically speaking, the market tends to favor higher prepayment speed deals across the stack, as well as the DN deals versus the HQ deals. In particular, STACR 2014-DN4 currently trades at the tightest discount, while STACR 2014-HQ2 trades at the widest discount. Of note, the market applies a concession to STACR 2015-DN1 M1 and M2 similar to that of other comparable cash flows but ascribes a much smaller discount to the more enhanced, BB-rated M3. The CS CRT Compendium 10

11 Figure 15: Market-implied discounts for STACR transactions with MACR IOs Indicative prices, market levels for OAS and CRT as of 4/7/15. "Difference From Mean" calculated by rating band Rating A Rated BBB Rated BB & Unrated Tranche Source: Credit Suisse, Freddie Mac Equivalent Agency Floater Price Market Level Market Implied Discount Difference From Mean (Aggregate) Difference From Mean (DN/HQ) 1 month vcpr STACR 2014-DN1 M STACR 2014-DN2 M (0.03) STACR 2014-DN3 M (0.50) (0.31) 37.6 STACR 2014-DN4 M (0.57) (0.38) 21.7 STACR 2014-HQ1 M (0.16) (0.40) 30.7 STACR 2014-HQ2 M STACR 2014-HQ3 M (0.38) (0.61) 16.3 STACR 2015-DN1 M (0.22) (0.03) 39.0 STACR 2015-HQ1 M N/A STACR 2013-DN2 M (3.01) (1.63) 5.4 STACR 2014-DN1 M STACR 2014-DN2 M STACR 2014-DN3 M (2.98) (1.61) 37.6 STACR 2014-DN4 M (3.08) (1.70) 21.7 STACR 2014-HQ1 M (1.06) 30.7 STACR 2014-HQ2 M STACR 2014-HQ3 M (0.01) (2.07) 16.3 STACR 2015-DN1 M (2.01) (0.63) 39.0 STACR 2015-HQ1 M (0.37) N/A STACR 2013-DN2 M STACR 2014-DN1 M STACR 2014-DN2 M STACR 2014-DN3 M (2.99) (2.07) 37.6 STACR 2014-DN4 M (4.09) (3.16) 21.7 STACR 2014-HQ1 M (0.50) 30.7 STACR 2014-HQ2 M STACR 2014-HQ3 M (0.82) (2.21) 16.3 STACR 2015-DN1 M (6.30) (5.38) 39.0 STACR 2015-HQ1 M (1.50) (2.88) N/A We note that the market-implied discount metric only truly holds meaning in the relative sense. The discount from the equivalent agency floater could include a number of technical factors that distort the value in the absolute sense. However, comparing the discounts for similarly rated tranches against each other should, in our view, control for most, if not all, of these factors. 1 We acknowledge that there are a handful of issues to address with the methodology. We assume an equal OAS for all tranches, up and down the stack, to their comparable IOS tranches. While this does control mostly for structure and WAL as a spread-based metric, one could argue that M1 and M2 excess IOs could trade at different OAS due to the structured nature of the cash flows. One could also argue that the HQ and DN excess IOs would trade at different OAS as well, given the somewhat different prepayment exposure in higher LTV loans. However, we believe that our approach controls for most issues involved in comparing these transactions, while introducing as little additional complexity as possible. 1 One additional note is the value of the uncapped floater in STACR versus agency-guaranteed par-priced floaters we have used DMs on 6.5% cap floaters. The value of the cap increases with WAL as we go down the capital structure. The impact is de minimus on the IG tranches but could be as high as 1.25 points on the BB and unrated tranches. In turn, this would understate the market-implied discount by the value of the cap. Although this would have limited impact on comparing the BB and unrated tranches to each other, it would have a modest impact on comparing BB and unrated tranches to IG tranches. The CS CRT Compendium 11

12 Conclusions As more and more CRT transactions have come to market, more differences in prepayment exposures and margins exist up and down the stack. These new deals create relative value discussions between tranches, and we believe that this framework will allow investors to compare the market-implied credit risk in a transaction. Although it does not currently apply to CAS and has some limitations, it does facilitate a relatively quick comparison of tranches that controls for a number of factors. Additionally, this approach can be used on new deals as they come to market to compare new tranches to seasoned ones. The CS CRT Compendium 12

13 Cumulative Home Price Change 2 December 2015 This is an exact excerpt from the Global Securitized Products Weekly, published 13 February What do GSE credit risk bonds tell us about housing risk? The programmatic issuance of GSE credit-risk-sharing deals has expanded the menu for mortgage credit investors beyond legacy RMBS and loans. However, in addition to a different borrower base and structural quirks, the risk/return profile of GSE credit-risksharing deals differs from that on legacy RMBS. We use pricing differences between STACR/CAS and legacy RMBS to calculate a housing risk metric that reflects the marketimplied probability of a 2008-style bust. STACR/CAS M2s are unrated cash flows, price at par, and trade at higher yields than generic Subprime or other legacy RMBS bonds. The incremental yield on the STACR/CAS M2s can be potentially viewed as the put option premium on HPA and credit performance sold by the investor, in our view. In contrast to a somewhat symmetric/linear return profile of legacy RMBS versus HPA, upside is capped on the M2s and downside is much higher in highly adverse scenarios. We compare the return profiles of STACR/CAS M2s and legacy RMBS across a range of HPA scenarios to infer a market-implied probability of a housing downturn. At a 25% probability of an optimistic HPA environment, we estimate that current pricing on the M2s implies a 3.4% probability of a 2008-style housing bust. We plan to periodically update this metric to track trends in market pricing of housing risk. The Methodology The STACR/CAS M2 yield profile, due to its structure, is similar to selling out-of-the-money put options on home prices and credit performance. With most borrowers in these pools out of the money in terms of refinancing incentive, returns are likely to be driven less by refinancing speed and much more by credit performance. Under most home price paths, the M2s offer a fairly stable yield profile with zero loss. However, an immediately starting severe downturn scenario could cause enough defaults to result in losses on the M2, in our view. We project STACR 2013-DN2 performance under a number of home price scenarios. For the home price depreciation scenarios, we apply the price decline paths starting today. For the crisis scenario, we use the home price path. The yield remains stable up to a 10% HPD scenario, with a 20% HPD resulting in losses. The crisis scenario results in a full writedown on the M2s. In turn, the yield profile as a function of home price appreciation is flat in most home price scenarios, even optimistic ones, but highly negative in major downturns. Figure 16: We stress-test STACR with a number of housing downturn scenarios Home Price Down and Crisis scenarios follow experience, percentage is peak to trough 60% 40% 20% Optimistic Base Stress Down 10% Down 15% Down 20% Crisis 0% -20% -40% Year Source: Credit Suisse The CS CRT Compendium 13

14 Yield 2 December 2015 Figure 17: STACR has a capped upside to HPA, holds up well under modest downturns, a full blown housing crisis-scenario results in a full writedown Home Price Down and Crisis scenarios follow experience, percentage is peak to trough Source: Credit Suisse As a point of comparison, we ran the same set of scenarios on a Subprime last cash flow, because of both their sensitivity to home prices and similar WAL. The Subprime last cash flow is more responsive to smaller swings in home price responses than STACR, but is much less responsive in a crisis scenario. In addition, the Subprime last cash flow shows upside in a more bullish housing scenario, whereas STACR M2s have a capped upside. Figure 18: A Subprime LCF is more sensitive to smaller HPA movements, with a much smaller downside Home Price Down and Crisis scenarios follow experience, percentage is peak to trough WAL is based on principal payments Source: Credit Suisse Optimistic Base Stress Down 10% Down 15% Down 20% Crisis Yield DM ,244 WAL Duration Writedown Liquidation Optimistic Base Stress Down 10% Down 15% Down 20% Crisis Yield DM WAL Duration Writedown Liquidation The Option Model The yield differential between STACR M2s and Subprime last cash flows, in the base case, is an expression of the larger downside and capped upside of the GSE credit-risksharing bond, in our view. In particular, the massive underperformance in a major downturn allows us to estimate how much the market is pricing in such an event. To do so, we use a basic option pricing approach. Figure 19: While the Subprime LCF yield is much more linear with HPA, STACR M2 is highly binary Home Price Down and Crisis scenarios follow experience, percentage is peak to trough STACR2 M2 Subprime LCF Optimistic Base Stress Down 10% Down 15% Down 20% Crisis Source: Credit Suisse The CS CRT Compendium 14

15 Proabability of Housing Crisis 2 December 2015 We probability weight the paths for the two bonds to solve for the implied probability of a housing crisis. To do so, we solve for risk-neutral pricing by equating the yields of STACR M2s and the Subprime LCF. We take three paths (optimistic, base, and crisis) and weight the paths to equate the expected yield on the two bonds given current market pricing. Firstly, we vary the optimistic path weight in order to solve for the crisis path. Next, we solve for the crisis path weight where the yield on the Subprime LCF equals the yield on STACR M2s. Figure 20: Market-implied probability of a housing crisis 10% 9% 8% 7% 6% 5% 4% 3% 2% 1% STACR M2 Outperforms Subprime LCF Outperforms 0% 0% 10% 20% 30% 40% Optimistic Home Price Path Weight Source: Credit Suisse As we increase the weight on the optimistic home price path, the corresponding weight on the crisis scenario declines because the expected yield on the Subprime LCF rises but the STACR yield is unaffected. To reach the risk-neutral result of equivalent expected yields, the weight on the crisis path must be incrementally lower. Current market pricing implies three key takeaways: The maximum market-implied probability of a housing crisis is about 9.4%. Above a weight of 38% on the optimistic path, the STACR M2 yield cannot match the Subprime LCF because the implied home price downturn probability is floored at 0%. Investors assigning a higher than 9.4% probability of a housing crisis or a greater than 38% probability of an optimistic HPA path should favor Subprime LCF over STACR M2s. STACR M2s are attractive to investors who believe in a more moderate range of outcomes. Our approach has a few limitations. Firstly, we use a risk-neutral framework to determine the path weights. If the market has any risk aversion to the STACR exposure, it would move the weights lower at each point. Secondly, we only use three home price paths, but given the return profile of the two cash flows, we believe these paths are adequate to discuss the market-implied risk. Thirdly, we assume the housing crisis scenario begins today; as we have discussed previously, STACR M2s are generally insulated from a housing crisis starting much later in the future. We believe this valuation framework is critical to evaluating the GSE credit-risk-sharing transactions versus other home price-levered instruments. After almost five years of mortgage credit trading at a deep discount dollar price, investors now have a liquid par dollar option. We believe our risk metric provides investors with a simplified approach to compare STACR/CAS versus legacy RMBS. The CS CRT Compendium 15

16 This is an exact excerpt from the Global Securitized Products Weekly, published 23 July An IOS framework for STACR/CAS With the issuance of CAS 2014-C03, the GSEs have issued eight credit risk transfer deals to date. Other than a small variance in the structure (notably, the 2014 STACR deals threetranche structure) and the different severity schedule, the deals are nearly identical. The largest difference between the deals, in our view, is the coupon difference between the deals. Comparing the last cash flow bonds to each other, given the highly similar collateral characteristics (save the high LTV groups in the most recent CAS deals), becomes an exercise in pricing the coupon differences. We see many parallels between these differences and those in the IOS market. We believe that coupon multiple valuations offer a convenient way to compare the various last cash flows. We benchmark valuations on credit risk transfer (CRT) last cash flow bonds to IOS multiples of similar vintage and coupon collateral. Under this framework, we believe that the STACR 2013-DN1 M2 is cheap relative to the rest of the CRT last cash flow stack. Overall, most CRT last cash flows appear rich compared to IOS benchmarks, in our view. Figure 21: With credit exposures relatively similar across the different deals CS Model projections The framework a coupon swap of sorts To compare the different last cash flows, we simply take the difference in coupons to create an IO-like security. We then take the difference in price and calculate the implied multiple difference between the securities. Given the low expected defaults on these transactions (Figure 21), we believe that most, if not all, of the dollar price difference between the LCFs is the value assigned to the incremental coupon. Figure 22: The material difference between the deals is the coupon CAS M2 is the swap coupon Cumulative Defaults (%) Swaps LCF Swap Deal name Base Optimistic Stress STACR131/CAS STACR 2013-DN STACR131/STACR CAS141/ STACR 2013-DN STACR141/ STACR 2014-DN CAS131/CAS STACR 2014-DN CAS131/CAS CAS 2013-C STACR131/CAS CAS 2014-C STACR131/CAS CAS 2014-C02 (G1) STACR132/CAS CAS 2014-C02 (G2) CAS131/STACR Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac Next, we bring in the IOS component as a point of comparison. Using the WAC and origination date, we identify the indices that best correlate with each tranche. These will provide the benchmarks for our coupon swaps either using the lower coupon or the higher coupon. Figure 23: IOS-implied multiples based on WAC and origination year Blends are 50/50 simple average of the two multiples (as applicable), July 22 close prices Source: Credit Suisse Deal WAC Origination Year IOS Coupon/Blend IOS Multiple STACR 2013-DN STACR 2013-DN STACR 2014-DN STACR 2014-DN CAS 2013-C CAS 2014-C CAS 2014-C02 (G1) The CS CRT Compendium 16

17 Current valuations and relative value At current valuations, we believe that the STACR 2013-DN1 M2s are the cheapest bonds in the stack. While the 130 dollar price might cause investors concern, the coupon multiple approach reveals its relative cheapness given both the underlying collateral and its relative value versus other bonds in the stack. We believe the STACR 2013-DN1 M2, in this approach, should be about 1-2 points higher. Figure 24: Implied multiples on CRT last cash flow swaps versus IOS comps Indicative levels, July 22 close LCF Swap Source: Credit Suisse Top Bond Price Bottom Bond Price Multiple % of top bond IOS Mult % of bottom bond IOS Mult STACR131/CAS % 91% STACR131/STACR % 101% CAS141/CAS % 112% STACR141/STACR % 115% CAS131/CAS % 103% CAS131/CAS % 108% STACR131/CAS % 93% STACR131/CAS % 98% STACR132/CAS % 100% CAS131/STACR % 125% On the other hand, we believe that most other swaps are modestly overvalued. We believe there is an argument to be made that these coupon swaps should trade lower than IOS multiples given their slightly lower liquidity and lower leverage. In addition, the collateral blend that underlies IOS is modestly different, as it includes both HARP collateral and sub-60 LTV loans as well. Furthermore, when the IOS multiples are lower than CRT multiples, investors can create synthetic premiums at a cheaper dollar price using IOS tranches. While current valuations on most other multiples in the LCF stack are higher than most IOS tranches, we believe that the compression should be small between the tranches. These valuation differences could potentially stem from a difference in the investor bases between CRT bonds and IOS. While using IOS to synthetically replicate higher coupon CRT last cash flows looks attractive here, it may not be in every investors mandate to purchase both. However, we still see IOS multiples as a reasonable benchmark to compare the CRT market to IOS because many investors can participate in both sectors. Conclusions As the GSEs issue more CRT transactions, the coupon multiple approach, in our view, could gain increased traction in valuing last cash flow tranches relative to one another. While the multiples may matter most to the first deals in today s landscape, we believe this approach allows investors to compare tranches on a more normalized basis. The CS CRT Compendium 17

18 Roll Rate (bps) Share of 30 day DQ population (MBA) Cumulative Defaults (bps) 2 December 2015 This is an exact excerpt from the Global Securitized Products Weekly, published 09 October Collateral Performance CRT, revisited deal performance and high LTV tiering At over a year since the first CRT deal priced, we revisit collateral trends in the agency CRT space. We believe credit performance of the transactions is in line with market expectations, with very low cumulative defaults and very little in the way of delinquencies. In addition, we believe the market should begin differentiating the high LTV deals, given the significant current LTV differences between the four original LTV pools. Collateral performance strong with limited drift The first transaction, STACR 2013-DN1, has just over a year of seasoning to date. The transaction has seen just under 2.7 bps of defaults in the first year, exclusive of repurchase activity. The other transactions with buyout data are relatively in line with this performance, with no other transactions reaching 2 bps of total defaults. Figure 25: Cumulative defaults on the CRT transactions are below 3 bps Exclusive of repurchase activity DN1 13-DN2 14-DN1 14-DN2 13-C01 14-C Deal Age (months) Source: Credit Suisse, LoanPerformance Furthermore, delinquencies remain well contained. While the current to 30 rolls (on an MBA basis) seem outsized versus the default performance, the current to 30 rolls (on an OTS basis) are much more in line with default performance. In fact, the majority of the 30- day borrowers (on an MBA basis) cure after a month. Figure 26: MBA current to 30 rolls far outpace OTS current to 30 rolls Source: Credit Suisse, LoanPerformance C->30 MBA C->30 OTS Figure 27: MBA 30-day delinquent borrowers tend to cure the next month 80% 70% 60% 50% 40% 30% 20% 10% 0% Cure Source: Credit Suisse, LoanPerformance Remain 30 Days DQ Worsen The CS CRT Compendium 18

19 Mark to Market LTV 2 December 2015 High LTV deals differentiating collateral One space where ongoing loan-level analysis can provide material insight is the subset of high LTV deals. The deals were issued within a few months of each other, with very similar FICO scores, DTIs, and origination LTVs (Figure 28). However, the deals also have some dispersion in WALA; STACR 2014-HQ1 had less seasoning at issuance than the other three high-ltv pools. Figure 28: The pools look relatively similar, save the HQ1 Data as of issuance (origination where noted) STACR CAS 2014-HQ HQ C02 (G2) 2014-C03 (G2) FICO DTI Orig LTV Orig CLTV FICO<720 & DTI>43 3.0% 1.8% 2.5% 2.8% WAC WALA Source: Credit Suisse, LoanPerformance, the BLOOMBERG PROFESSIONAL service As we mark to market these borrowers LTVs, however, we find the high-ltv pools have begun to diverge in terms of current LTV position. Although these pools all originated within one point of each other, the current LTVs diverge by almost ten points (Figure 29). The CAS LTVs have dropped significantly since their issuance date, while STACR LTVs remain above 80. CAS 2014-C02, in particular, has seen almost a five-point decline in LTV since issuance (Figure 29). Figure 29: While the high-ltv deals have similar origination LTVs, current LTVs have somewhat diverged Loan Origination Deal Issuance Today HQ HQ C02 (G2) 2014-C03 (G2) Source: Credit Suisse, LoanPerformance, CoreLogic Currently, the market exhibits preference between the high LTV pools versus the LTV pools, while treating the high LTV pools similarly. With these LTV drifts, however, we believe the market should start to tier these high LTV deals accordingly. At current valuations, we prefer the two CAS deals, in particular the CAS 2014-C02 2M2, versus the two STACR deals. The CS CRT Compendium 19

20 A look at CRT prepayment exposure This is an exact excerpt from the Global Securitized Products Weekly, published 15 February The focus in the CRT space, for the most part, has been the credit aspects of the pools and their subsequent performance. However, with 15 deals now priced at fairly different coupons (both bonds and collateral), we can turn our attention to comparing prepayment sensitivities of the different deals. Examining the prepay sensitivity of the borrowers lends itself to comparing and contrasting the various M1s and IG-rated middle mezzanine M2s, as well as the last cash flows. The IG bonds are slightly more insulated from credit events, but are just as exposed to prepayments than their last cash flow counterparts. In turn, we tend to favor the higher WAC bonds with a modest discount, such as the STACR 2014-DN2 M1 and M2, over higher WAC premiums in deals such as STACR 2014-DN4. Refinancing exposure different across deals With the variance in mortgage rates over the last 24 months, the 15 CRT transactions carry materially different WACs. They range from as low as 3.58 in CAS 2014-C02 (G2) to as high as 4.57 in STACR 2014-DN4. We see this impact prepayment speeds, with three-month CPRs ranging from 5 to over 20. In addition, a good number of deals, mostly mid- to late issuance, have a large amount of borrowers who currently have rate incentive. Figure 30: GWAC and recent prepayment speeds for CRT transactions Deal/Group GWAC 1mo vcpr 3mo vcpr >4 GWAC (% of deal) CAS 2014-C02 (G2) % CAS 2014-C02 (G1) % STACR 2013-DN % CAS 2014-C % STACR 2014-DN % CAS 2014-C03 (G2) % CAS 2014-C03 (G1) % STACR 2014-HQ % STACR 2013-DN % CAS 2013-C % CAS 2014-C04 (G1) % STACR 2014-DN % CAS 2014-C04 (G2) % STACR 2014-DN % STACR 2014-HQ % STACR 2014-HQ % STACR 2014-DN % Source: Credit Suisse, LoanPerformance, Index However, the WAC alone does not tell the whole story since these deals can include materially different distributions. For example, both CAS 2014-C04 (G1) and STACR 2014-DN2 have average WACs around 4, but have a fairly wide dispersion of coupons. Conversely, CAS 2013-C01 has a 3.84 WAC but much less dispersion in the pool. The CS CRT Compendium 20

21 Share of Deal (%) 2 December 2015 Figure 31: The coupon dispersion differs largely across similar WAC pools 40% 30% 20% 10% CAS 2013-C01 STACR 2014-DN2 CAS 2014-C04 (G1) 0% <=3.25 >3.25 & <=3.5 >3.5 & <=3.75 >3.75 & <=4 >4 & <=4.25 >4.25 & <=4.5 >4.5 & <=4.75 >4.75 Source: Credit Suisse, LoanPerformance Relative value impact The spread impact of these differences in rate incentives can be seen across the stack, but is mostly dependent on the bond s deviation from par. For the IG cashflows, even modest premiums or discounts can lead to relatively large swings in DM based on prepayment speed in large part a function of their relatively short WALs. To benchmark the impact of these collateral differences, we ran select deals using our agency prepayment model. We used STACR 2014-DN2, which indicatively prices entirely at a discount, STACR 2014-DN4, which prices entirely at a premium, and CAS 2014-C03, which indicatively trades at a discount and with a slightly lower IG rating. Figure 32: The IG rated bonds in STACR 2014-DN2 look more attractive than their 2014-DN4 counterparts from a prepayment perspective CS6.10, Indicative prices as of 2/18/15, run to base case losses Indicative DM (1M LIBOR) WAL LT CPR 1 year CPR Bond Price Forward Static Forward Static Forward Static Forward Static STACR 2014-DN2 M STACR 2014-DN2 M STACR 2014-DN2 M STACR 2014-DN4 M STACR 2014-DN4 M STACR 2014-DN4 M CAS 2014-C03 1M CAS 2014-C03 1M Source: Credit Suisse, Credit Suisse Locus At current levels for the front-pay bonds, we believe STACR 2014-DN2 M1 is relatively attractive, given the prepayment exposure and modest discount. On the other hand, we believe the modest premium baked into STACR 2014-DN4 M1 makes these cashflows slightly less attractive. However, we acknowledge that the DN4 M1s look modestly more attractive in a 75bp selloff from today s rate levels. In addition, CAS 2014-C03 1M1, at a BBB-/BBBH rating (versus A ratings on the STACR M1s), looks attractive relative to the STACR front pays, albeit at a slightly lower IG rating. The CS CRT Compendium 21

22 Aug-13 Sep-13 Oct-13 Nov-13 Dec-13 Jan-14 Feb-14 Mar-14 Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 GWAC % with incentive (30 yr fixed) DM (bps) 2 December 2015 Figure 33: STACR 2014-DN4 M1 would outperform in a 75bp selloff CS 6.10, immediate rate shock, indicative prices from above (2/18), run to base case losses STACR 2014-DN4 M1 STACR 2014-DN2 M Rate Shock (bps) Source: Credit Suisse, Credit Suisse Locus The middle mezzanine BBB tranches, folding in the prepayment exposure, present an interesting investment profile. The discount M2, on a DM basis to forwards, looks more attractive than its premium counterpart. Additionally, STACR 2014-DN4 M2 compares, on a rating and WAL basis, to CAS 2014-C03 1M1. We prefer the CAS front pay, given its stability and current cashflow (versus the locked out STACR 2014-DN4 M2). Conclusions The first handful of CRT transactions were mostly backed by borrowers with little to no rate incentive to refinance. Later issuance, as well as the rally in mortgage rates in 2014, brought another added variable to the CRT market prepayment exposure. In addition, the various coupons, and subsequent premium and discount bonds, create opportunities to express prepayment views, particularly in the IG-rated bonds. We believe this approach is an important step in determining relative value in CRT IG-rated bonds. Figure 34: Rising levels of collateral are in the money GWAC (LHS) % with incentive (RHS) 45% 40% 35% 30% 25% 20% 15% 10% 5% 0% Source: Credit Suisse, LoanPerformance The CS CRT Compendium 22

23 This is an exact excerpt from the Global Securitized Products Weekly, published 15 April Fat tails CRT prepays driven by higher loan balances and WAC dispersion We have previously discussed the prepayment impact on CRT structures, given the WAC dispersion among the various transactions. However, the last few months of prepayment speeds seem, on the surface, to have diverged meaningfully from comparable benchmarks. In particular, the higher WAC STACR transactions, such as 2014-DN4 and 2015-DN1, have prepaid over 50% faster than their high-level comparable cohort. Figure 35: Prepay speeds for CRT groups and comparable cohorts As of 3/15, IOS speeds lagged one month Deal (Group) WAC WALA Comparable IOS Collateral Comparable IOS 1m vcpr 3m vcpr 1m vcpr 3m vcpr STACR 2013-DN STACR 2013-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-HQ STACR 2014-HQ STACR 2014-HQ STACR 2015-DN NA STACR 2015-HQ NA NA CAS 2013-C CAS 2014-C CAS 2014-C02 (G1) CAS 2014-C02 (G2) CAS 2014-C03 (G1) CAS 2014-C03 (G2) CAS 2014-C04 (G1) CAS 2014-C04 (G2) CAS 2015-C01 (G1) NA CAS 2015-C01 (G2) NA Source: Credit Suisse, LoanPerformance, the BLOOMBERG PROFESSIONAL service We believe that the wider collateral dispersion seen in CRT transactions lends itself to higher prepayment speeds versus high-level comparable cohorts. Two factors, in our view, contribute to this divergence the WAC dispersion and the concentration of higher balance loans. Looking forward, we expect prepayment speeds to remain high in STACR transactions in the short term. Higher WAC CAS transactions, particularly CAS 2015-C01, will likely see higher prepayment speeds next month in large part due to the additional reporting lag. The difference lies in the balance We believe one of the largest drivers of the higher speeds in recent months is the distribution of loan balances in STACR transactions, particularly towards the higher end. In most recent STACR transactions, about 20% of the transaction has a loan balance greater than $400K. Controlling for loan balance and coupon, the more recent STACR speeds (14-DN2, 14-DN3, 14-DN4, and 15-DN1) are roughly in line with comparable agency The CS CRT Compendium 23

24 Figure 36: CPRs for recent STACR transactions by WAC and loan size Figure 37: CPRs for comparable agency pools by WAC and loan size 3 mo. averages, STACR 14-DN2,14-DN3, 14-DN4, and 15-DN1 Corresponding 3 mo. averages, LTV, 6-18 WALA WAC Balance ($K) < > WAC Balance ($K) < > Source: Credit Suisse, LoanPerformance Source: Credit Suisse, CPRCDR We had previously looked at the WAC distribution in given transactions. As a refresher, CRT transactions bucket based on origination quarter; if rates fluctuate during the quarter, a given deal can have a wide dispersion of loan rates. As seen below, some deals have materially wider distributions than others which can cause speed differences at the same average WAC. Figure 38: WAC distribution can differ materially between CRT transactions Pct. of each group s balance in each WAC bucket, as of 3/ >5.0 STACR 2013-DN STACR 2013-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-HQ STACR 2014-HQ STACR 2014-HQ STACR 2015-DN STACR 2015-HQ CAS 2013-C CAS 2014-C CAS 2014-C02 (G1) CAS 2014-C02 (G2) CAS 2014-C03 (G1) CAS 2014-C03 (G2) CAS 2014-C04 (G1) CAS 2014-C04 (G2) CAS 2015-C01 (G1) CAS 2015-C01 (G2) Source: Credit Suisse, LoanPerformance Looking forward We believe higher speeds will continue in STACR transactions and are likely to arise in CAS 2015-C01. With mortgage rates hovering below 4%, substantial amounts of the more recent CRT transactions are both in the money and in the larger balance cohort. While we believe longer-term prepayment speeds will normalize as the higher-balance loans pay off, speeds in the shorter term can remain elevated. The CS CRT Compendium 24

25 Figure 39: Share of group balances that are both $400K+ balance and 4%+ WAC As of 3/2015 Source: Credit Suisse, LoanPerformance STACR 2013-DN1 2.96% STACR 2013-DN2 0.92% STACR 2014-DN1 2.40% STACR 2014-DN2 7.93% STACR 2014-DN % STACR 2014-DN % STACR 2014-HQ1 7.43% STACR 2014-HQ2 2.03% STACR 2014-HQ3 8.04% STACR 2015-DN % STACR 2015-HQ1 9.14% CAS 2013-C % CAS 2014-C % CAS 2014-C02 (G1) 0.86% CAS 2014-C02 (G2) 0.41% CAS 2014-C03 (G1) 2.11% CAS 2014-C03 (G2) 0.98% CAS 2014-C04 (G1) 7.81% CAS 2014-C04 (G2) 4.05% CAS 2015-C01 (G1) 16.65% CAS 2015-C01 (G2) 7.29% Another item of note is the seemingly slower speeds in CAS transactions; in particular CAS 2015-C01, where higher WACs have only materialized into slightly elevated prepayment speeds. We believe this is almost entirely due to the additional month reporting lag in CAS versus STACR. CAS 2015-C01 looks very similar to STACR DN4, and STACR 2014-DN4 saw similar prepayment speeds the month prior. In turn, we expect speeds in CAS 2015-C01 to pick up next month. Figure 40: We expect CAS 2015-C01 (G1) to print higher speeds in April As of 3/2015 WAC WALA Avg Loan Bal ($K) % Bal > 400k Mar vcpr Feb vcpr STACR 2014-DN CAS 2015-C01 (G1) Source: Credit Suisse, LoanPerformance The CS CRT Compendium 25

26 Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 CPR Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 CPR 2 December 2015 CRT prepays flip the script This is an exact excerpt from the Global Securitized Products Weekly, published 30 July When mortgage rates hit their local lows in January, a number of CRT transactions saw significant prepayment increases. We found that larger balance loans with higher coupons were driving prepayment speeds higher. Today s story flips the script; higher turnover speeds, coupled with higher mortgage rates, have caused some prepay speed compression among CRT transactions. We find that turnover speeds have increased, particularly in greater than 80 OLTV transactions, driven by loans experiencing some HPA. On the other hand, the decline in both the concentrations and speeds of higher balance, higher WAC collateral has brought down speeds for in-the-money deals. We believe this trend is likely to continue for some time. Figure 41: CRT prepay speeds have compressed in recent months Slowest 3 Pools Fastest 3 Pools CRT Aggregate Source: Credit Suisse, LoanPerformance Out-of-the-money prepayment speeds have recently picked up In recent months, the slower prepayment deals have seen a pickup in speeds, particularly for out-of-the-money collateral. This is most pronounced in the high-ltv transactions, where loans with sub-80 mark-to-market LTVs are seeing prepayment speeds of over 10 CPR. Figure 42: Out-of-the-money prepayment speeds have picked up <80 mark-to-market LTV, <3.75% rate OLTV 80+ OLTV Source: Credit Suisse, LoanPerformance The CS CRT Compendium 26

27 Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 CPR CPR Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 CPR 2 December 2015 These LTVs are estimated using MSA-level indices, which we recently found can provide fairly different marks than those from more granular approaches (22 July 2015, Global Securitized Products Weekly). In turn, we look at greater than 80 OLTV loans that have an mark-to-market LTV, and find that these have also sped up in recent months. We believe that HPI variances within MSAs could contribute to this. Figure 43: Speeds for out-of-the-money LTV loans have also increased >80 OLTV, current LTV, <3.75% rate Source: Credit Suisse, LoanPerformance In-the-money speeds have slowed in the past few months On the other side of these moves, with the recent backup in mortgage rates, the higher balance (>400K), higher WAC (>4%) loans have slowed down. These loans prepaid over 50 CPR at their peak, and speeds for these have now fallen by almost 50%. Both the and greater than 80 OLTV deals have seen similar slowdowns for these loans. Although slowing, the greater than 80 OLTV loans that are sub 80 LTV on a mark-tomarket basis remain faster than their counterparts in this cohort as well. Figure 44: High balance, high WAC prepay speeds have recently slowed >$400K balance, >4% rate Figure 45: 80+ OLTV loans prepay faster than OLTV loans when at sub-80 mark-to-market LTV >$400K balance, >4% rate, <80 current LTV OLTV 80+ OLTV OLTV 80+ OLTV Source: Credit Suisse, LoanPerformance Source: Credit Suisse, LoanPerformance The CS CRT Compendium 27

28 Apr-14 May-14 Jun-14 Jul-14 Aug-14 Sep-14 Oct-14 Nov-14 Dec-14 Jan-15 Feb-15 Mar-15 Apr-15 May-15 CPR CPR 2 December 2015 Figure 46: At the money collateral vcprs for 80+ OLTV deals have recently picked up more At-the-money speeds remain higher for 80+ OLTV deals In the past few months, the prepayment difference between and greater than 80 OLTV deals became particularly large in at-the-money collateral, which we define as 3.75%-4.00% WAC. For borrowers with greater than 80 OLTVs and sub-80 mark-tomarket LTVs, prepayment speeds have picked up faster than those for their OLTV counterparts. We find that this effect becomes larger in the higher loan balances. Figure 47: The and 80+ OLTV prepay difference is wider for larger loans 3.75%-4.00% rate, <80 current LTV 6-month CPR, 3.75%-4.00% rate, <80 current LTV OLTV 80+ OLTV OLTV 80+ OLTV OLTV 80+ OLTV <$400K >$400K Source: Credit Suisse, LoanPerformance Source: Credit Suisse, LoanPerformance We expect these trends to continue for the CRT transactions through at least the end of the year. In a rally, the deals that previously saw high prepays could see similarly high speeds, in our view. We believe the 80+ OLTV transactions will prepay slightly faster as at-the-money borrowers prepay faster than their OLTV counterparts. The loan balance distributions across these deals are fairly consistent, save for slightly higher CK/T6 concentrations in some groups. The CS CRT Compendium 28

29 Figure 48: Loan balance concentrations are fairly similar across deals As of today, % share of each deal s balance Pool LLB MLB HLB HLB2 TBA CK/T6 STACR 2013-DN STACR 2013-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-DN STACR 2014-HQ STACR 2014-HQ STACR 2014-HQ STACR 2015-DN STACR 2015-DNA STACR 2015-DNA STACR 2015-HQ STACR 2015-HQ CAS 2013-C CAS 2014-C CAS 2014-C02 (G1) CAS 2014-C02 (G2) CAS 2014-C03 (G1) CAS 2014-C03 (G2) CAS 2014-C04 (G1) CAS 2014-C04 (G2) CAS 2015-C01 (G1) CAS 2015-C01 (G2) CAS 2015-C02 (G1) CAS 2015-C02 (G2) CAS 2015-C03 (G1) CAS 2015-C03 (G2) Source: Credit Suisse, LoanPerformance The CS CRT Compendium 29

30 LTV 2 December 2015 Fast and furious drift the impact of HPA on high- LTV CRT transactions This is an exact excerpt from the Global Securitized Products Weekly, published 18 June The high-ltv (80-97 OLTV) CRT transactions have, through differentiated exposure to home price appreciation, begun to diverge in terms of mark-to-market LTVs. We find that many of the loans today have materially lower LTVs than at origination in the case of the first two CAS deals, LTVs are below 80. The loan level information provided by these transactions allows us to delve deeper into a critical issue for the CRT market: how has HPA impacted the high-ltv transactions? For loans that start in the high-ltv cohort, we find that lower mark-to-market LTVs bring two effects. Firstly, lower LTVs contribute to lower roll rates, both on a MBA and OTS basis. Secondly, the greater than 80 OLTV loans that are now sub-80 LTV have incrementally steeper S-curves steeper, in fact, than their sub-80 OLTV counterparts. Given this, we continue to favor CAS 2014-C02 and 2014-C03 2M2s, given their materially lower mark-to-market LTVs and discount dollar prices. These bonds look attractive with both potential credit gains and prepayment upside, given the lower LTVs. STACR HQ2 and CAS 2014-C04 (G2) also stand to benefit from modest HPA, with large shares hovering above and below 80 LTV. The current state of the high-ltv transactions The higher LTV transactions have seen significant declines in LTV since origination at least a few points, and in some cases over ten points of declines. While HPA has lowered LTVs, over 70% of the loans in these deals still have LTVs above 80. This is in part due to the continued addition of high-ltv deals, not all of which have experienced the same HPA. Figure 49: While the high-ltv deals have similar OLTVs, they have somewhat different mark-to-market LTVs OLTV LTV HQ1 14-HQ2 14-HQ3 15-HQ1 14-C02 (G2) 14-C03 (G2) 14-C04 (G2) 15-C01 (G2) 15-C02 (G2) Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac The impact of HPA on high-ltv CRT deals High-OLTV loans that have realized HPA provide the opportunity to see if HPA can compress at least some of the differences between greater than and less than 80 OLTV loans. We find that high-oltv loans with sub-80 mark-to-market LTVs have current to 30 roll rates in line with OLTV loans, on both a MBA and OTS basis. The impact of declining LTVs on high-oltv loans continues past the 80 LTV barrier, with sub-70 markto-market LTV loans outperforming mark-to-market LTV loans. The CS CRT Compendium 30

31 vcpr C->30 (MBA, bps) C->30 (OTS, bps) 2 December 2015 Figure 50: C->30 (MBA) rolls are similar for sub-80 mark-to-market LTV loans and sub-80 OLTV loans Mark-to-market LTV for OLTV loans Figure 51: C->30 (OTS) rolls show a nearly identical likeness Mark-to-market LTV for OLTV loans < > OLTV Mark-to-market LTV 0.0 < > OLTV Mark-to-market LTV Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac HPA s impact is not limited to credit performance, but also affects prepays. High-OLTV loans with sub-80 mark-to-market LTVs have prepaid faster than sub-80 OLTV loans, at almost all points of moneyness. This effect amplifies the further high-oltv loans decline in mark-to-market LTV. Figure 52: High-OLTV loans reaching lower mark-to-market LTVs prepay faster than OLTV loans Mark-to-market LTV for OLTV loans < > OLTV 5 0 <=-75 >-75 & <=-50 >-50 & <=-25 >-25 & <=0 >0 & <=25 >25 & <=50 >50 Incentive (bps) Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac We note that this dataset only covers about a year of this experience, and the high OLTV loans are susceptible to burnout after maintaining sub-80 mark-to-market LTVs. We found a similar effect in legacy RMBS, where loans that had recently fallen under 80 LTV prepaid faster than loans with over two years of sub-80 LTVs. However, the majority of high-oltv loans in CRT transactions are still above 80 LTV, which could keep the high-oltv cohort speeds elevated, as these loans continue fall below 80 LTV. The CS CRT Compendium 31

32 Share of deal (%) vcpr 2 December 2015 Figure 53: In 2014, legacy RMBS borrowers were sensitive to LTV changes 2014 performance, always current OTS, non-io fixed-rate borrowers, <80 LTV, benchmarked to LTV 2 years prior <80 LTV (2 years prior) >95 LTV (2 years prior) Prime Alt-A Subprime Source: Credit Suisse, LoanPerformance With this in mind, we continue to favor CAS 2014-C02 and C03 2M2s over other high OLTV deals. These deals have sub-80 mark-to-market LTVs and trade at deep discount dollar prices, allowing for gains from both credit improvement and prepayments. STACR 2014-HQ2 and CAS 2014-C04 (G2) also stand to benefit from modest HPA, with large shares hovering above and below 80 LTV. Figure 54: Mark-to-market LTV distributions by deal 60% 50% 40% < >90 30% 20% 10% 0% 14-HQ1 14-HQ2 14-HQ3 15-HQ1 14-C02 (G2) 14-C03 (G2) 14-C04 (G2) 15-C01 (G2) 15-C02 (G2) Source: Credit Suisse, LoanPerformance, Fannie Mae, Freddie Mac The CS CRT Compendium 32

33 This is an exact excerpt from Global Securitized Products Weekly published 22 August Historical pieces/introductions STACR Implications for g-fees and the Future Mortgage Credit Market Summary STACR 2013-DN1 is a landmark transaction in many ways. Its importance as the first fully transparent credit transaction in the non-agency market since the financial crisis cannot be emphasized enough. It is effectively a Freddie Mac corporate exposure synthetically referencing over $22.5B in originations. STACR s stepped severity approach caps the upside on severities, in our view. But it also insulates investors against future policy risks and catastrophic home price declines. Our model, which has been fitted to historical Freddie Mac loan level credit performance, projects 58bps of cumulative defaults over the lifetime in the base case. Given the current limitations, a pro-rata payment structure offers the most efficient execution. Although the ten-year maturity exposes Freddie Mac to tail risk, looking from the perspective of hedging risk for the entire enterprise by regularly issuing such transactions, the vast majority of the entity s risk will be successfully hedged and it will only bear the risk on a small fraction of its book. STACR implied all-in g-fees are about 37-39bps based on our estimates, including 10bps of payroll tax surcharges and another 5-7bps of administrative costs. This is approximately in line with the estimated 42bps of g-fees on the underlying collateral. In comparison, private label RMBS transactions price in a g-fee of 42-49bps, slightly higher than what STACR implies. We estimate that g-fees under a bond guarantor framework could potentially range from 80bps to100bps. Although this is higher than current capital markets execution, it is necessary to have an optimal mix of both avenues to preserve credit availability through economic/housing cycles. The final g-fee will be contingent on the market s ability to absorb the supply, in our view. We think an annual issuance pace of $80B (risk sharing on 10% of annual GSE issuance) under Corker/Warner bill or the PATH Act would lead to a steady state market size of around $520B, potentially outstripping demand for RMBS credit and leading to higher credit costs. However, at a more measured $40B pace, the market reaches a terminal size of $260B and supply-demand technicals are expected to be more evenly balanced. This compares with a 2007 benchmark, where the total size of the cash RMBS subordinate market was slightly higher than $200B. STACR could have ripple effects that may reach beyond its use as just a credit risk sharing mechanism. The potential for creating a mortgage credit index based on a robust and programmatic issuance program cannot be overstated. The CS CRT Compendium 33

34 0.0% 0.2% 0.4% 0.6% 0.8% 1.0% 1.2% 1.4% 1.6% 1.8% 2.0% 2.2% 2.4% 2.6% 2.8% 3.0% Cumulative Losses 2 December 2015 The STACR transaction STACR is effectively a Freddie Mac corporate exposure synthetically referencing over $22.5B in 2012 originations, with tranched credit exposures. The structure has a ten-year final maturity when Freddie Mac will call the deal. The STACR transaction consists of three major groups of bonds (for reference, any bond that ends in H is retained by Freddie Mac). The top 97% of the structure was retained by Freddie Mac, as was the B- H, which was the 30bp first loss piece. Freddie Mac sold the M1, which was the 1.65%- 3% cut, and the M2, which was the 0.3%-1.65% cut, as par floaters with a spread over one-month LIBOR (Figure 55). The M1 and M2 notes have no claims on the collateral itself; Freddie Mac pays the notes as the pool pays down. The collateral is 2012 originations with a gross WAC of 3.84, LTV between 60 and 80, and an average FICO score of 766. Figure 55: Offered and retained certificates Tranche Intial Notional Amount ($M) Subordination Cpn Offered/ Retained A-H $21, % Retained M-1 $ % 1ML Offered M-1H $ % Retained M-2 $ % 1ML Offered M-2H $ % Retained B-H $ Retained Source: Credit Suisse, Freddie Mac Since the deal has no claims on collateral, STACR has a few idiosyncrasies embedded in the structure. In lieu of true defaults, STACR uses the term credit event, which occurs at 180 days delinquency (a true default is also a credit event ). In case of a credit event, STACR has pre-determined severities that start at 15% with steps up to 25% and 40% if cumulative credit events exceed 1% and 2% respectively (Figure 56 ). In the case of a modification that takes place prior to a credit event, the loan remains in the pool with no impact on STACR. Furthermore, a reps and warrants breach and successful putback effectively reverse the credit event as well. In terms of cash flows, STACR is very similar to a standard senior-subordinate CMO although it doesn t employ a shifting interest structure. The structure has two triggers: a subordination test (the "A-H" has at least 3% support) and a predetermined cumulative credit event test. If the structure passes both tests, both the seniors and subordinates receive a prorata share of both the scheduled and unscheduled principal (among the subordinate certificates, the M1 and M2 classes receive the principal sequentially). If either trigger is failing, the M1 and M2 receive a pro-rata share of the scheduled principal but are locked out of any unscheduled principal. Figure 56: Stepped severity & losses 0.9% 0.8% 0.7% 0.6% 0.5% 0.4% 0.3% 0.2% 0.1% 0.0% Source: Credit Suisse Cumulative "Credit Events" Of note, however, is the final maturity on the structure; Freddie Mac intends to call the deal in ten years without regard to collateral balance, although the notes can be redeemed at any date when the collateral factor is less than or equal to 10%. The CS CRT Compendium 34

35 Stepped severity upside and downside are both capped but shields from policy risk Historically, loss severities on Prime loans have been low. If we look at loans originated prior to the financial crisis and their performance prior to 2006 we find that on an average about 64% of the loans that were liquidated after reaching 180 days of delinquency did not incur any losses. And of the rest, the average loss severity was 23% and were liquidated after spending about 16 months in delinquency (Figure 57). Netting out P&I advances that were made after the initial six months (as is expected within the STACR transaction), the effective loss severity on all liquidated loans comes to about 7%. The average balance on the liquidated Jumbo loans was $277K, slightly higher than the STACR transaction s $235Kbut this is not a meaningful difference to significantly alter severity expectations. The expected severity can go up by a couple of points after taking into account current longer timelines compared with pre-2006 experience, and the servicer s continued responsibility for advancing taxes and insurance on the property. From this analysis, it appears that STACR s stepped severity approach somewhat caps the upside on severities. At the same time, this arrangement also insulates investors against future policy risks and catastrophic home price declines. Figure 57: Historically, Prime Jumbo loss severity experience has been low For liquidations with Losses Avg. bal ($K) Avg. Liq. Timeline Source: Credit Suisse, LoanPerformance % of loans Liquidated Without loss % of Loans Liquidated with loss Severity net of 6 mths of Effective advances severity Vintage Avg LTV Severity % 23% 25% 19% 4% % 31% 21% 15% 5% % 31% 19% 14% 4% % 27% 25% 19% 5% % 41% 21% 14% 6% % 56% 27% 21% 12% Average % 36% 23% 18% 7% Valuation We ran the bonds under four HPA scenarios an optimistic scenario, base case, stress case, and a crisis scenario tracking the home price path during the housing bust. Our credit model has been fitted to historical Freddie Mac loan level collateral credit performance. Based on our model projections using loan level collateral data backing the deal, cumulative defaults (or credit events) in the first ten years amount to 47bps, increasing to 58bps over the course of 30 years (Figure 59). In the stress scenario, where home prices drop by a cumulative 1% in the first two years, defaults increase to 65bps and 77bps in ten and 30 years respectively. In the first three scenarios, the M2 does not take a writedown and offers a stable yield and WAL profile. But in the crisis scenario, 100% of the M2 tranche and 42% of the M1 tranche get written down and cumulative defaults amount to 6.84% over the span of 30 years. To put it in perspective, the estimated default rates in another crisis scenario is only about 60% of the historical default rates on the vintages (Figures 58 and 60) primarily because of much tighter underwriting standards today. In order to assess the sensitivity of the tranches to collateral performance we performed another scenario analysis in which we ran multiples of our base case CDR projection holding other variables constant. We found that the M2 tranche takes its first dollar of loss at 3.43 times our base case default rate assumption. The CS CRT Compendium 35

36 Cum Credit Events 2 December 2015 Figure 58: The M2 tranche does not take a writedown even in our stress scenario Optimistic Base Stress Crisis Yield M1 WAL Principal Writedown 0.0% 0.0% 0.0% 42.4% Yield M2 WAL Principal Writedown 0.0% 0.0% 0.0% 100.0% Liquidation (10 Yrs) % Loss (10 Yrs) % Collateral Liquidation (30 Yrs) % Loss (30 Yrs) % HPA Scenario Year 1 Year 2 Year3 Year 4 Year 5 Optimistic 12% 7% 5% 4% 3% Base 7% 4% 3% 3% 3% Stress 0% -1% 2% 3% 3% Crsis -31.1% in the first 5 yrs followed by 2,4,3,3,3.. Figure 59: Cumulative credit event projections 0.9% 0.8% 0.7% 0.6% 0.5% 0.4% Stress (30 year) 0.3% Base (30 year) 0.2% Optimistic (30 year) Stress (10 year) 0.1% Base (10 year) Optimistic (10 year) 0.0% Period 0.9% 0.8% 0.7% 0.6% 0.5% 0.4% 0.3% 0.2% 0.1% 0.0% Source: Credit Suisse Source: Credit Suisse Locus That said, our default projections do not make adjustments for repurchases and/or modifications (within the first six months of delinquency) which should result in credit events reversals or lowering the cumulative defaults. Based on Freddie Mac s credit data, cumulative repurchases and modifications on pre-2005 vintages have averaged around 8bps and 17bps respectively (Figure 60). Conservatively, assuming that only 40% of the modified loans survive without re-defaulting, the effective net credit event reversals average at about 14bps (8+17*40%). Given that the default expectations on the collateral is low to begin with, if even half of this 14bps comes through, it would result in a meaningful decline in cumulative defaults. The M2 bond is more cuspy, longer in duration, and also more attractive from a valuation perspective, in our view. The Figure 60: Net credit event reversals have averaged at 16% of cumulative defaults for pre-2005 vintages In %. Freddie Mac loan Level credit data for loans with LTV >= 60 & <= 80; 700 < FICO <= 760 Cum Cum Repurchase Cum Net Credit Event Mods Vintage Default reversals Source: Credit Suisse Locus, Freddie Mac collateral corresponds to a discount price pass-through (at 3.84 gwac), is already fully extended, and offers a stable 715DM (9.75% yield) for a 9.7 year bond. As of 14 August, both the M1 and the M2 have tightened and were offered at 291DM and 635DM respectively. A pro-rata payment structure is the most efficient execution By making this transaction a corporate exposure, Freddie Mac has created a credit risk transfer construct that works smoothly with the TBA market (this is not possible with a senior-sub structure) and sidestepped potential market challenges associated with Commodity Futures Trading Commission s (CFTC) commodity pool regulations. Such challenges are a hurdle for implementing this risk transfer as a standard credit-linked note due to an embedded swap (a CLN is essentially a fully funded CDS through which one party buys credit protection on the collateral). The CS CRT Compendium 36

37 By employing a pro-rata payment schedule for the subordinates and not implementing a more typical shifting interest structure, the bonds pay down much faster and have much shorter average lives. This in turn has led to a much better execution for the bonds. Freddie Mac s exposure lies in a scenario in which credit events are back loaded and initial prepayments are fast. Under such a scenario, faster initial prepayments would pay down the subordinates exposing the seniors to back-ended losses. In addition, the tenyear maturity on the bonds leave Freddie Mac exposed to tail losses if credit performance deteriorates after ten years. While this may be thought of as weakness to the current loss sharing transaction, looking from the perspective of hedging risk for the entire enterprise, and not just one isolated transaction, the ten-year maturity on the bonds makes sense to us. If the plan is to regularly issue such risk-sharing deals, then at any point the vast majority of the entity s risk would be successfully hedged. The issuer would only bear the risk on collateral that is more than ten years seasoned and arguably is only a small fraction of the entire book. The implications for guarantee fee pricing The GSEs raised guarantee fees (g-fees) twice in 2012 and they are currently around 50bps. According to FHFA Acting Director Edward DeMarco, one of the primary motivations behind such increases was to bring GSE credit risk pricing in line with where the private market would price such risk. Ideally, the most direct impact of STACR is a market-driven solution for g-fee pricing. However, this transaction pricing, although useful, is only a first step and remains some distance away from providing a full-fledged capital markets-implied g-fee benchmark. Several considerations arise when inferring the fair value of g-fee based on STACR pricing. A non-exhaustive list includes (1) the value of Freddie Mac s retention of the bottom 0.3% needs to be evaluated from a private credit buyer s perspective, (2) the tail risk following the ten-year final maturity on credit exposure, (3) the value of the deterministic severities, (4) the thinness of the credit tranche sold compared with non-agency subordination levels, (5) the value of Freddie Mac s much stronger reps and warranty enforcement capability, and (6) the potential impact of large-scale programmatic risk transfers consistent with the GSEs footprint. We explore some of these considerations in the following sections. Value of the bottom 0.3%: Based on where the bonds priced and assuming that all of the M1 and M2 tranches were sold, the cost of buying credit protection to Freddie Mac is 14.2bps (340bps*1.35% + 715bps*1.35%). But this does not give us the full picture as to what the g-fee is since the bottom 30bps have been retained. If we were to assume that this bottom tranche could be sold at L+1,200bps, then the cost of hedging credit risk, or the implied g-fee, would be about 18bps. If it priced at L+1,500bps, the implied g-fees would have moved by just 1bp to 19bps. At 0.3% thickness, the bottom-retained tranche is so thin that pricing differences on this piece do not move the overall g-fee materially. Value of the ten-year final: Arguably, this cost is only for the first ten years. But the cumulative losses on the pool do not increase by more than 10% between the tenth year and the final maturity of the loans under all our scenarios (Figure 58). Conservatively, if we assume that the M2 tranche would have priced 300bps wider, if the ten-year maturity were to be extended out, total credit cost would still not go up by more than 4bps (=300*1.35%) taking it to 22bps. Payroll tax and GSE operational costs: Including the 10bp payroll tax surcharge, the implied g-fee would be 32bps. Adding another 5-7bps for administrative costs takes the total g-fee to 37-39bps. This is comparable to our estimate of a 42bp g-fee on the underlying collateral. We recognize that this is lower than the current average g-fee of nearly 50bps (on 2013 origination) cited by DeMarco. However, this includes a broad distribution of collateral characteristics, such as HARP, lower FICO, higher LTV loans, which carry higher g-fees unlike the current reference collateral. The CS CRT Compendium 37

38 Reconciling STACR and non-agency credit pricing: Our above analysis points to only a modest difference in credit pricing between the STACR transaction and the non-agency market. Although currently there is no liquid market for new issue Jumbo subordinates, anecdotally the subordinate stack is priced around $80 with a yield of 6-7%. With 7% average subordination, this brings the implied g-fee to 42-49bps for private label transactions slightly above our estimated 37-39bps derived from the STACR transaction. First and foremost, a large part of the difference stems from the relative size of the credit tranches. While this is limited to a 3% slice for the STACR transaction, in the non- Agency market AAA investors demand a far higher level of credit protection currently at 7%. In addition to this, the non-agency market suffers from illiquidity even at the top of the capital stack. Second, the value of the reps and warrants enforcement provided by Freddie Mac and by aligning their interests with those of investors by holding on to a slice of the transaction creates an additional layer of protection otherwise unavailable to mortgage credit investors. Third, the ten-year final maturity (compared with the much longer WAL on non- Agency subs) and fixed severities on STACR takes a good deal of uncertainty out of the equation, making it relatively more attractive. In addition, the potential for a liquid market (albeit sometime in the future) also helped lower the liquidity premium. On the other hand, the lack of any rating may have precluded interested investors from participating in the inaugural deal. A rating provided by a rating agency, particularly for any tranche resembling the M1 (or any slice above it), would bring more ratingsconstrained investors into the fold. In addition, an NAIC rating would make such transactions significantly less onerous on insurance companies, which we believe will be an integral source of capital in a post-gse world. Once remedied, a rating would substantially increase the liquidity of the transactions and may even drive spreads tighter on future transactions. Examining g-fees from a bond guarantor perspective: We estimate that g-fees for a regulated bond guarantor corporation with bank-like capital (that could operate in a Corker/Warner-like framework) would be 75bps to 90bps. This is based on a 4% to 5% capitalization, a 15% ROE, assumed economic cycle average losses of 10bps, and administrative fees of 5bps. The fair value of a government reinsurance wrap at these capitalization levels is minimal, but the government may choose to charge 5-10bps, should Congress decide on such a housing finance model. This would bring the overall g-fee to bps. At 10% ROE, the g-fee would decline by about 25bps. Clearly, these are much higher than current capital-markets execution levels, which would suggest that most credit risk would be sold through the capital markets under current market conditions. However, over an economic cycle, there are likely to be weaker market conditions when the opposite would be the case. Furthermore, under stress conditions it is possible for capital market execution to become virtually unavailable. As a practical matter both avenues of providing housing credit should co-exist. Institutional impact: According to anecdotal reports, a large investor base, both in count and diversity, participated in the first STACR transaction. The housing recovery has drawn quite a bit of investor attention towards mortgage credit, which is a two-sided exposure: default frequency and severities. We believe this transaction particularly catered to investors who wanted to express their view on default expectations going forward without the uncertainty of severities. However, investors who want to express a view on severities and do not want the upside on severities to be capped would likely prefer a more collateralized approach closer to that of non-agency structures. The CS CRT Compendium 38

39 Depending on institutional preferences and their relative size, different forms of credit exposure may produce varying depths for the future credit market. With this in mind, we try to estimate how much issuance this market can potentially handle. How much credit risk can the market absorb? Figure 61: Annual subordinate issuance in non-agency RMBS market As per FHFA s 2013 Scorecard for the GSEs, the aim is to move from a dollar 20,000 target (currently at $30B for each GSE) to 0 a percentage of business target sometime in the future. The question that, in our view, looms largest on the horizon is Source: Credit Suisse, Mortgage Market Statistical Annual how much can clear at the prices achieved in the first issuance if the GSEs were to become programmatic sellers of credit risk. Assuming a $ B annual issuance between the two GSEs, sharing risk on a 3% slice would imply annual issuance of mortgage credit risk transactions to the tune of $21-24B. Add in another $50B of prime Jumbo issuance with 7% CE and the total amount of new-issue credit risk goes up to $28B. In the Corker/Warner bill and the PATH Act, a 10% subordination would be required. This target would require issuance of STACR-like transactions to run at a formidable $70-$80B annual pace. Eventually, in a steady state, assuming a $5-7T conventional single family mortgage securities market, something of the order of $450B to $650B of housing credit risk would have to be placed with private investors, based on these goals. To get an estimate of how much credit risk the market can absorb we look at the size of the market for subordinates at the end of 2007 as one benchmark. At that time the total size of the non-agency market was about $2.1T with about $453B in subordinates outstanding, by our estimates. Of this, about $270B was absorbed by the CDO market (Figure 62). About $34B of the outstanding Mezz and High Grade CDOs at that time was rated BBB and below which we add back to our estimates of the total market size, as these investors were pure credit investors. From this estimate it appears that the market size or total investor demand for cash subordinate investors was just above $200B. ($B) Figure 62: RMBS subordinate market size at the end of 2007 In $B Source: Credit Suisse 120, ,000 80,000 60,000 40,000 Prime Alt-A Option ARM Subprime A Outstanding Non Agencies $ 2,172 B Estimated RMBS Subordinates Outstanding $ 453 C Cash Subordinates backing CDOs $ 270 D BBB and Below CDOs abosorbed by credit investors $ 34 = B-C+D Net Outstanding Subordinates absorbed by end Investors $ 216 The question remains whether there will be sufficient appetite for such large-scale issuance of credit exposure and how much additional compensation investors may demand to absorb the additional supply. To answer this question we look at two scenarios. The first is the Corker/Warner bill which, in its purest form, according to our estimates will require annual issuance of $70-80B. The CS CRT Compendium 39

40 We assume paydowns occur at an annual rate of 10 CPR and a STACR-like transaction, where prepayments are distributed pro-rata. Based on these the size of the outstanding credit need would be about $275B after four years, with a peak at about $520B after ten years consistent with a $6T conventional single family securitization market. Under this scenario, the potential supply outstrips our estimated investor demand and should entail significant spread widening. However, other potential sources of demand may come through, for example, current non-agency investors could be an additional source of demand as the legacy non-agency market would have Figure 63: Projected size of RMBS credit market under different scenarios shrunk by $ B in another four years time. This may make available roughly $ B in credit demand. In the second scenario, we assume a variation of the above plan where only half of the 10% credit slice is issued reducing the annual supply to about $40B. In this case, supply and demand are expected to be more evenly balanced, with the market size peaking at close to $260B. In our view, whether pricing can hold at these levels as issuance volumes rise is key to truly unearthing the price of mortgage credit risk. Creating a mortgage credit Index potential ramifications far beyond STACR STACR, in our view, could have ripple effects that reach much further than simply using it as a risk-sharing mechanism; its potential as an index may, eventually, have an impact on mortgage credit pricing that cannot be overstated. As STACR shifts from inaugural to programmatic issuance, the sheer size of the STACR market would render it ready for a reference index. We believe that such an instrument would be well accepted by the market, both as a cash mortgage credit proxy, as well as an opportunity to either hedge or outright short mortgage credit. However, we believe that the market must tread carefully in creating such an index, as there must be outsized demand for the cash product before the market considers such an index. The secondary impact of a STACR index on banks would, in our view, be an underestimated but integral part of the future mortgage market. An index would provide, on a generic basis, the current market price of mortgage credit. The index would allow banks to move mortgage credit, from non-agency subordinates to any other iteration, from a level 3 asset to level 2 treatment. We believe this impact, while certainly not a consideration in the very near future, could be fundamental to STACR s legacy. ($B) Source: Credit Suisse 80B in Ann. Issuance 40B in Ann. Issuance Yr The CS CRT Compendium 40

41 Price 2 December 2015 Exploring the fair basis between STACR and CAS This is an exact excerpt from the Global Securitized Products Weekly, published 24 October With the official issuance of CAS last week, the non-agency market officially has a true comparable for the STACR transaction. Both deals have caught the MBS credit tailwind of the last few weeks, with both bonds in each deal rallying quite a bit since the start of the month (Figure 64). However, secondary trading of CAS has given rise to more questions than answers on its pricing comparison to STACR and where the fair relative spread between the two deals lies. Figure 64: Prices have rallied significantly in the last few weeks 116 CAS M1 114 CAS M2 112 STACR M1 110 STACR M /5 7/25 8/14 9/3 9/23 10/13 11/2 Source: Credit Suisse, TRACE On a collateral basis, we believe the two should trade at very tight levels to one another. From a credit perspective, the two transactions are nearly identical; the same can be said about the prepay profiles as well. On the other hand, the structural and technical aspects of the two deals could lead to a modest spread between the corresponding bonds in the two deals. We believe the CAS M1 should trade inside STACR by a small margin for two reasons. First, the inclusion of an IG rating affords a more attractive level of funding on the CAS M1, as well as opening up the opportunity for ratings-constrained money managers to participate. Second, the lower fixed severities for the first 2% of credit events for CAS provides further protection in a material housing downturn. The basis between the two, at about bp, is fair, in our view. The case for the CAS M2 trading tighter than the STACR M2 becomes quite a bit murkier. The lower severity schedule will certainly benefit the CAS M2 in a significant, but not crisis-esque home price depreciation scenario; in a home price path that mimics , we believe both M2s would be fully written down, rendering the CAS M2 advantage minimal at best. Furthermore, the locked out nature of the cash flows, as well as the 10-year final maturity, strips almost all of the option cost out of the two M2s. Even at STACR M2 s current premium of about 15, we believe the whipsaw exposure remains well contained. We believe that, from a fundamental perspective, the STACR M2 should trade tighter than the current spread to the CAS M2. The Two Transactions: An Overview The collateral underlying the two transactions is very similar. As seen in Figure 65, the differences between the two pools are relatively minute. STACR has a slight advantage in terms of LTV, FICO, and investor share. At the same time, while STACR has a slightly tighter distribution of FICO scores with a standard deviation of 37.9 points compared to 39.7 for CAS, the DTI and combined LTV distributions are modestly tighter for CAS. Similarly, looking for risk layering, we find that the proportion of borrowers that lie outside one standard deviation both in FICO and DTI are very close between the two transactions, with CAS coming out slightly stronger. The CS CRT Compendium 41

42 Our model projects slightly higher defaults on the STACR (see Figure 66). We think, the higher expected defaults on the STACR transaction are driven by the 10 bp and 0.5 bp of 30 days and 60+ days delinquency the transaction printed this month. Without this additional history on the STACR, our model projected defaults on the two transactions are almost on top of each other. Structurally, there are two main differences between the two transaction. The fixed severities on CAS are slightly lower than that of STACR for the first 2% of defaults. For the first 1% of defaults, CAS uses a 10% severity instead of STACR s 15%; the following 1% sees a 5% lower severity as well in CAS (20% and 25%, respectively). In addition, CAS has no cumulative loss trigger, but it still has the same hard enhancement trigger. The M1s: CAS should trade modestly inside STACR In our view, the CAS M1 is slightly more attractive than the STACR M1 for mostly structural and technical advantages. First, the lower fixed severities would, in a housing crash resembling that starts immediately, lead to much lower losses on the CAS M1 than the STACR M1 in our projections. Although the impact of the fixed severities is only seen on the M1s in a true housing crisis, the incremental safety of CAS should provide some spread differential, however modest that may be, over its STACR counterpart. In addition, the IG rating on the M1 should assist some buyers, in particular money managers with IG rating constraints. Furthermore, the IG rating also affords a lower haircut and more attractive funding levels on the CAS M1 tranche, leading to modestly tighter spreads. Figure 66: Lower severities lead to lower losses on CAS in a housing crisis Home price path follows experience, percentage indicates peak-trough home price experience M1 Writedowns Home Price Path Cum Defaults CAS Loss STACR Loss CAS STACR Down 10% 1.44% 0.19% 0.26% 0% 0% Down 15% 2.20% 0.38% 0.48% 0% 0% Down 20% 3.20% 0.78% 0.88% 0% 0% (Down 31%) 6.56% 2.12% 2.22% 35% 43% Source: Credit Suisse Figure 65: Collateral comparison & default expectation CLTV FICO DTI Source: Credit Suisse, Deal docs STACR CAS WAC LTV Wtd Avg Std Dev LTV Range Wtd Avg Std Dev Wtd Avg Std Dev % with FICO < 1 SD & DTI > 1 SD % with FICO < 1SD & CLTV > SD Average Balance ( Refi Percentage Investor Owner State #1 CA 22.4% CA 26.7% State #2 VA 5.2% TX 5.9% State #3 IL 4.7% NY 4.5% Proj Optimistic Liquidation Base (10 Yrs) Stress The CS CRT Compendium 42

43 Absence of a cumulative loss trigger The lack of the loss trigger in the CAS transaction has very little to nearly no impact on the M1s versus STACR, in our view. The hard enhancement trigger, where payments are only diverted to the M1 and M2 if the enhancement level on the senior is above 3%, will lock out both STACR and CAS in higher loss scenarios. The STACR cumulative loss trigger will only have an impact if there are large-scale early defaults that cease after a short while and allow enhancement to build up over time. In other words, the loss trigger comes into play when we have a housing downturn, followed by a significant increase in home prices, causing defaults to stop altogether or decline dramatically. Such a scenario is highly unlikely, in our view. The M2s: The CAS advantage also muted The M2 difference from a credit perspective, in our view, will only be seen in a significant home price decline, although not quite as deep as the crash. If the home price experience of occurred starting today, both M2s would be written down in full rendering the CAS M2 advantage minimal at best (Figure 67). However, we also consider other home price scenarios where we assume the price declines start immediately and the path follows the same trajectory as that during the experience, and only the peak-to-trough price declines vary across the different scenarios. As seen from Figure 67, in more muted home price declines, the CAS M2 would receive a lower write-down than its STACR counterpart. However, we assign a low probability to a more than 10% home price decline starting immediately. A much more realistic situation would be home prices start to decline with some delay. If we delay the start of a home price crisis by seven years from today, neither M2s take a write-down. Figure 67: The M2 credit exposures only differ in immediate, but moderate home price declines and are identical in a crash years from now Home price path follows experience, percentage indicates peak-trough home price experience M2 Writedowns Home Price Path Cum Defaults CAS Loss STACR Loss CAS STACR Down 10% 1.44% 0.19% 0.26% 0% 0% Down 15% 2.20% 0.38% 0.48% 6% 13% Down 20% 3.20% 0.78% 0.88% 36% 43% (Down 31%) 6.56% 2.12% 2.22% 100% 100% Down 31% (drop begins in 7 years) 0.55% 0.06% 0.08% 0% 0% Source: Credit Suisse Convexity concerns M1 more exposed, M2s shielded (even in STACR) As STACR M2 prices reached in excess of 110, the convexity risk on these transactions drew quite a bit of attention and there are large segments of investors who are uncomfortable buying bonds with any convexity exposure at such high prices. While we think this is true, a closer examination reveals that the convexity exposure is relatively muted on the M2s, even for STACR M2 with a current price of While these borrowers remain out of the money by about 45 bp, concerns over these higher premium bonds in a rally focus mostly on the STACR M2. However, we believe that, given the structure of the two transactions, the M1s are more exposed to a prepayment increase and, in a way, shield the M2s from higher prepays. The relatively long lockout on the M2, buttressed by the 10-year maturity, materially mutes the impact of The CS CRT Compendium 43

44 any prepayment increases. If prepayments increase, the sequential nature of the two bonds exposes the M1 to twice the prepayments than its pro-rata share. On the other hand, if prepays slowdown materially, the 10-year maturity on the M2 protects it from extending. Thus the M2 bond remains relatively well protected from whipsaw risk. To test the sensitivity of the four bonds, we ran our model base case default vectors for a number of prepayment vectors. We use 5 CPR for the first year followed by varied peak speeds for the next two years, followed by a 5 CPR tail speed to test the convexity of the four bonds. In scenarios with sustained higher prepayments, the M1s in both deals are noticeably more reactive to higher prepayment speeds, with a relatively limited difference in sensitivity; the actual impact can be seen below in Figure 68. Figure 68: While M1s are more exposed to prepay risk, M2 remain relatively shielded Prepay vectors used for each scenario: 5 for 12 Peak Speed for 24 5 / Run at constant price. Indicative pricing as of October 23. CAS M1 (at ) STACR M1 (at 103) CAS M2 (at ) STACR M2 (at ) Peak Speed DM WAL DM WAL DM WAL DM WAL Source: Credit Suisse However, the M2s are slightly less reactive than their M1 counterparts. Furthermore, even in large speed-ups, the STACR M2 is only modestly more sensitive even at these premium dollar prices. To put it into perspective, at a 15 CPR peak speed, the CAS M2 DM declines by about 4 bp or about 10 ticks. In the same scenario, the STACR M2 DM declines by about 11 bp, or in other words, it price impact is about 28 ticks. In turn, we believe the concerns over the convexity risk in STACR M2s are potentially overdone; the actual impact is so mitigated by the 10 year maturity that the very large premium only makes a modest difference in a large prepay wave. On the other hand, the CAS M1, at these dollar prices, looks noticeably more stable than STACR s M1. Conclusion In our view, the CAS M1 provides solid relative value over the STACR M1. The credit and convexity advantages, at these levels, lead us to prefer the CAS M1, and we believe the bp spread between the two is justified. While the CAS M2 is slightly less exposed to prepayments and slightly more immune to moderate home price declines, we believe the current spread between the two deals at about bp should be tighter. In turn, we believe the STACR M2 has relative value versus the CAS M2 and expect the basis between the two to tighten. The CS CRT Compendium 44

45 GSE Risk Sharing capital market execution is more efficient than MI Pool Policy This is an exact excerpt from the Global Securitized Products Weekly, published 17 October Fannie Mae released the pricing details of the second credit risk sharing transaction of the GSEs last week. In August, Fannie Mae announced that National MI, in a competitive bidding process, won the opportunity to provide pool-level mortgage insurance on a $5B pool of relatively new origination. The details reveal a structure that allows for some level of comparison with STACR. The transaction is known as 2013 MIRT 01 and fulfills the stated 2013 Conservatorship Scorecard objective of sharing credit risk with the private investors through various channels. We believe that the pricing points of the National MI deal provide some insight into market valuations of credit risk on new GSE originations. Under the pool MI policy premium is bp annually. We find that at current spread levels, a capital market execution leads to a lower annual premium. Unlike the STACR and the CAS transactions, the MI provider effectively owns the upside in severity expectations. However, the severity upside is not large enough to overcome the liquidity advantage enjoyed by the STACR and CAS transactions, in our view. The collateral and structure The National MI collateral and the STACR collateral look relatively similar (Figure 69), save a higher LTV range on the National MI pool. The reference pool for the National MI policy is $5B. Fannie Mae purchased the loans in the fourth quarter of 2012, with all of the loans originated in Every loan has an original LTV between 70 and 80. In comparison both the STACR and the CAS transactions had a LTV range. The average LTV on the MIRT pool is 76.9%, about two points higher than the other two risksharing transactions. National MI will receive 0.92 bp monthly, or an annual premium of bp. Figure 69: MIRT has very similar underlying collateral characteristics to STACR and CAS Source: Credit Suisse, Fannie Mae STACR MIRT CAS WAC LTV CLTV Pool LTV Range ( Max- Min) FICO DTI Average Balance (K) Refi Percentage Owner Occupancy State #1 CA 22.4% CA 28.5% CA 26.7% State #2 VA 5.2% TX 5.5% TX 5.9% State #3 IL 4.7% CO 3.9% NY 4.5% The National MI policy is a hybrid of a standard pool policy and loan level coverage. The policy coverage kicks in after losses on the reference pool exceed 20 bp. Thereafter, National MI provides loan level loss coverage between 50 LTV and the original LTV on the loan. The CS CRT Compendium 45

46 For example (Figure 70), for a $180K loan with a 75 LTV at origination, National MI will provide $61.2K of loss coverage; anything exceeding that will be borne by Fannie Mae. In addition, the policy has an aggregate maximum loss limit of 2%, and, like STACR, expires after ten years. Comparing National MI deal to STACR/CAS transactions To compare the pricing of this deal against where STACR or CAS trades today in the secondary market, we must first highlight the differences between the two Figure 70: Example of loan level coverage A Orginal Value $ 240,000 B Initial Principal Balance $ 180,000 C Initial LTV 75% D Specified Exposure Percentage 50% E = (1- D/C) F= E*B Loan Coverage Percentage (rounded to next highest integer) Maximum Insurance Benefit Source: Credit Suisse, Fannie Mae = (1-50%/75%) = 34% structures. Firstly, the National MI transaction references actual loan level losses, as opposed to STACR s fixed severities on defaults. On the other hand, the National MI policy does not have the liquidity associated with a capital-markets-based approach. Finally, the National MI deal, as a 0.2%-2% slice, covers 90 bp less than STACR, which provided coverage between 0.3% and 3%. However, the differing slices of coverage mean we must adjust the pricing on STACR to provide a proper comparison. To reconcile the pricing of the two deals, we must first use STACR s (or CAS s) price points to recreate where the capital-markets-based approach would price (Figure 71) the MIRT transaction. The first difference comes from the 0.2%- 0.3% slice, for which we do not have pricing (for reference, we will call this piece the M3). As a result, we assume that the M3 tranche prices anywhere between 1000 DM and 1500 DM. The second arises from the 1.65%-2% slice, which, on its own, we believe would price wide of the M1 (for reference, we will call this piece the M1*) as it is the bottom 35 bp of the 1.35% M1 tranche. As a reference point, indicative levels today for STACR M1 and M2 are at 215 bp and 530 bp, respectively. The same for CAS M1 and M2s are at 175 bp and 486 bp, respectively. Using a 530 DM for the M2, we create a pricing grid showing comparable prices on the MIRT transaction, for different DM levels on the M1* and the M3 tranche. As shown in Figure 72, based on a 225 DM and 1000 DM on the M1* and M3 tranches, respectively, the annual MI cost on the MIRT transaction comes out to 8.94 bp. Thus the capitalmarkets-based approach prices the MIRT transaction about 2 bp tighter than the bp on the MIRT. The scenario analysis below leads to the conclusion that although the pools are very similar in credit characteristics, the capital market execution leads to a lower annual premium at most spread levels on the M1* and M3 tranches. The outcome of the open bidding process that led to National MI s agreement with Fannie Mae reveals the implied value of actual loss exposure versus the added liquidity of STACR. The STACR transaction, with its fixed severities, effectively left HPA upside with Freddie Mac. However, the National MI pricing indicates that the value of this upside is not tremendously large. While the liquidity of STACR certainly lowered the price of the transaction versus the National MI policy, the severity upside is apparently not large enough to overcome the liquidity advantage. $ 61,200 The CS CRT Compendium 46

47 Figure 71: Re-purposing the STACR structure to create a true comparison to MIRT Source: Credit Suisse, Fannie Mae Figure 72: Capital market execution of the MIRT transaction under almost any DM scenario, the capital market execution is tighter than annual MI cost of bp M2 DM of 530, measured as basis points across whole collateral MIRT: bp M1* DM (bp) Source: Credit Suisse M3 DM (bp) The CS CRT Compendium 47