Mortgage Servicing Right Valuations Under Stress

June 04, 2026

Mortgage Servicing Right Valuations Under Stress¹

Ronel Elul, Karen Pence, Ben Ranish, and Michael Suher

Executive Summary

Mortgage servicing right (MSR) valuations decrease when mortgage default and prepayment rates increase, as is generally the case when the economy enters into recession. To estimate how large these MSR valuation declines could be for the banking sector in a severe economic downturn, we project the potential increase in default and prepayment rates under the supervisory stress test models and scenarios for mortgages serviced by large banks. We then project the resulting decline in MSR valuations using the sensitivity of MSR values to increased mortgage defaults and prepayments. While banks' current servicing books have low exposure to mortgage defaults, we also consider scenarios that may better reflect the risk of MSRs under stress in the future and scenarios in which banks' servicing books mirror the broader servicing market, as some of the already proposed targeted changes to capital requirements may bring some mortgage-related activity back to the banking system.

We estimate that the decrease in MSR valuations could range from 5% to 13%, depending on the scenario, as a result of the higher mortgage defaults. With respect to prepayments, we estimate that MSR valuations would fall by around 4% for each one percentage point increase in the prepayment rate. While this implied drop in MSR valuations is large, offsetting factors may make a prepayment-driven drop in MSR valuations less consequential for a bank than a default-driven drop.

Introduction

A mortgage servicing right (also known as a mortgage servicing asset) is the right to receive the servicing income from a mortgage as compensation for performing a variety of servicing obligations. MSRs are created when the right to service the loan is contractually separated from the loan itself. This separation typically occurs as part of the securitization process, although MSRs may also be created as part of whole-loan sales. For mortgages that a bank holds in portfolio, the servicing is generally kept with the loan and no MSR is created.

MSR valuations are based on the present value of the expected net income associated with servicing the mortgages. Net income, in turn, depends on borrower default and prepayment rates. When borrowers default more than expected, MSR valuations fall because servicing delinquent and defaulted loans is considerably more expensive than servicing performing loans. Mortgage default can also result in the mortgage being removed from the pool, in which case the servicer loses servicing income more quickly than anticipated. Likewise, when borrowers prepay and refinance their mortgages more quickly than expected, generally because interest rates have fallen unexpectedly, MSR valuations fall because servicers lose the income they would have received if the mortgage loans had not been prepaid.

We consider a stress scenario in which bank-owned MSRs decrease in value due to both increased mortgage defaults and increased prepayments. However, while the macroeconomic factors driving defaults and prepayments are inter-linked, the risks posed to banks from default-driven MSR declines are larger than those from prepayment-driven declines. Mortgage defaults rise when unemployment rises and house prices decline. At such a time, a bank is likely to be under strain more broadly because its other financial assets may also fall in value. Further, servicers who need funds may find that they are unable to sell their MSRs for their carrying values. Potential purchasers may require a sharp discount as compensation for the risk that loans purchased from another servicer during a time of rising defaults may have other defects that make default servicing even more costly.

In contrast, when decreases in interest rates spur an increase in mortgage prepayments (e.g. refinancing), other factors may offset the resulting decrease in MSR valuations. When rates fall, mortgage servicers may benefit explicitly from their interest-rate hedges or implicitly through additional gain-on-sale income from increased mortgage refinancing activity. This activity will also lead to new MSRs being created. Further, MSRs whose values decrease because of prepayments will likely be able to find a buyer at a price near the MSR carrying cost, unlike in the case of default-driven drops in valuation. That said, banks still have some risks in this situation: hedges do not always behave in accordance with historical relationships, origination income and new MSRs may not materialize, and the unhedged changes in MSR values resulting from prepayments can be large.

Our approach draws upon data submitted by large bank holding companies (subsequently, "banks") to the Federal Reserve as part of the stress testing process.² Our analysis focuses on mortgages serviced for the government-sponsored enterprises (GSEs, i.e. Freddie Mac and Fannie Mae) and Ginnie Mae (GNMA), which we collectively refer to as the "Agencies," as MSRs corresponding to non-agency MBS and to whole-loan sales represent only a small part of the overall mortgage servicing market. In the severely adverse stress-test scenario that we employ, unemployment rises and house prices fall. Interest rates also fall as investors anticipate reduced economic growth and more accommodative monetary policy. We estimate how defaults and prepayments in these large banks' servicing portfolios would increase in this environment, and then estimate the effect of these higher default and prepayment rates on MSR valuations.

Based on large banks' current servicing portfolios, we project that, under this stress scenario, elevated mortgage defaults would cause aggregate MSR values to decline by around 5%. However, this loss rate may understate banks' losses in the future for two reasons. First, banks' current servicing portfolios, in the aggregate, contain a large share of seasoned loans that benefited from outsized home price appreciation during and after the COVID pandemic. In the absence of that buildup in home equity, we project loss rates on current MSRs could be around 8% under stress. Second, in recent years the new loans entering bank servicing portfolios have primarily been in GSE rather than GNMA pools. We project that, if bank servicing portfolios instead mirrored the total Agency servicing market, MSR loss rates under the stress scenario would be 13%.

We then consider the additional decrease in MSR valuations that could stem from the assumed elevated prepayment rates. We find that the large reductions in mortgage rates under the stress scenario would generate a significant volume of prepayment and refinancing activity. The average annual prepayment rate would rise by around 5 percentage points for the typical servicing portfolio, leading to a large decrease in MSR valuations. However, as discussed earlier, the decrease in MSR valuations in this case may pose less risk than the default-driven declines because of the offsetting gains on hedging positions and the new originations and MSRs generated by the refinancing activity.

Data Sources

Our bank estimates are based on the data that large banks are required to submit to the Federal Reserve as part of the stress testing process (the FR Y-14A/Q/M Capital Assessments and Stress Testing information collection).³ Our estimates for the entire market are drawn from loan-level data from ICE, eMBS (henceforth eMBS).

Bank Mortgage Data. The Y-14M data include detailed monthly loan-level data on all mortgages in securitized pools serviced by the large banks from 2015 to 2025.⁴ In December 2025, the data included 9.9 million loans serviced by 18 large banks on behalf of the GSEs and 1.4 million loans serviced on behalf of GNMA. In dollar terms, these banks are servicing roughly $1.9 trillion in unpaid principal balance for the GSEs and $186 billion for GNMA, or about one third of loans outstanding in GSE pools and roughly 7% of loans in GNMA pools.

Mortgage Servicing Rights Valuations. From 2013 to 2019, large banks reported characteristics of their MSRs on a quarterly basis in Schedule I of the Y-14Q, including basic information on the servicing portfolio and estimates of the sensitivity of their MSR valuations to a variety of shocks. From 2013 to present, the large banks whose Schedule I data we use account for between 75% and 83% of the banking system's mortgage servicing assets, but a significantly smaller share of the overall servicing market. A detailed description of the information collected on Schedule I can be found in the Y-14Q instructions.⁵

Representativeness of Bank Servicing Portfolios

The loans that banks service for the Agencies are not representative of the broader Agency market. In particular, the Agency mortgages serviced by banks have lower loan-to-value (LTV) ratios and are less likely to be guaranteed by the Department of Veterans Affairs (VA). These characteristics mean that the mortgages in bank servicing pools are less likely to default and less likely to lead to large losses for the servicer in the event of default.

The discrepancy stems, in part, from the fact that banks service disproportionately older (seasoned) mortgages. Banks pulled back from originating and servicing mortgages for the Agencies after the Global Financial Crisis. The bank share of the mortgage servicing market fell from 65% in January 2014 to 36% in May 2026 for GSE servicing (Figure 1) and from 66% to 11% over the same period for GNMA servicing.⁶ As a result of this pullback, in 2025 the median age of a mortgage was 5.6 years in a bank-serviced GSE pool and 4.7 years for all loans in GSE pools (Table 1). The discrepancy was even more pronounced for mortgages in GNMA pools. The median age in 2025 was 11.2 years for bank-serviced pools and 4.5 years for all loans in GNMA pools.

Table 1. Characteristics of Agency Servicing Pools

Because of this greater seasoning for mortgages in bank-serviced GSE or GNMA pools compared to all GSE or GNMA pools, borrowers in bank-serviced pools have paid down more principal and benefited from more house price appreciation than borrowers overall. As shown in Table 1, the average LTV at time of origination was about the same in banks' books as in the market overall in both 2016 and 2025, at around 70% for mortgages in GSE pools and 95% for mortgages in GNMA pools. The average current LTV, however, was lower in the bank-serviced pools than the market overall in 2025. For GSE pools, the average current LTV was 47% for bank-serviced pools and 55% for the overall market. For GNMA pools, the average current LTV was 56% in bank-serviced pools and 71% for the overall market.

Mortgages serviced for GNMA consist primarily of loans insured by the Federal Housing Administration (FHA) or guaranteed by the VA.⁷ The share of VA-guaranteed loans in all outstanding GNMA pools has risen over time, representing 24 percent in 2016 and 32 percent in 2025 (Table 1). In contrast, because banks tend to service more seasoned pools, the share of VA-guaranteed loans in bank-serviced GNMA pools has remained essentially flat, increasing from 22 percent in 2016 to 23 percent in 2025.

As a result, banks have less exposure than the GNMA servicing market overall to the downside risks of servicing defaulted loans guaranteed by the VA. These risks stem from the fact that the VA guaranty is capped at roughly 25% of the original loan balance.⁸ Suppose house prices decline significantly, such as the 36% decline assumed in the stress test scenario. If a servicer forecloses on a property in this scenario, the proceeds from the foreclosure sale may be considerably less than the outstanding mortgage balance. This unrecovered mortgage balance, plus other foreclosure costs, may exceed the VA guarantee, and servicers will need to absorb the remainder from their own capital. These potential losses, which can run into tens of thousands of dollars if house prices decline significantly, result in lower MSR valuations.

Defaults and Prepayments on Mortgages Serviced by Banks in a Severe Stress Scenario

The first step in our analysis is to apply the Domestic First Lien Mortgage Model from the 2024 supervisory stress tests⁹ to the Y-14M data submitted by each bank. The model was developed to estimate default and prepayment rates on first-lien mortgages that a bank holds in its portfolio. We note below when our estimates based on portfolio mortgages might differ from those based on mortgages in GSE and GNMA mortgage pools.

Our analysis projects default and (voluntary) prepayment rates under the baseline and severely adverse scenarios from the stress test.¹⁰ The baseline scenario follows a profile similar to that of average projections from a survey of economic forecasters. In the severely adverse scenario, the unemployment rate increases from 3.7% to 10% over the course of two years, while the mortgage rate falls from 7.3% to 3.1% and house prices decline by 36%. The default and prepayment components of the model project the probability that a mortgage transitions to a different state (i.e., current, delinquent, default, voluntary prepayment, and paid off) from one quarter to the next, given the scenario and the risk characteristics of the mortgage, most notably the origination credit score of the borrower, the updated LTV ratio, and the interest rate on the mortgage.

Throughout the analysis, we include separate results for GSE and GNMA pools. This distinction is crucial for the default analysis. Mortgages in GNMA pools are more likely to default because of the characteristics of those pools such as their higher LTVs and lower credit scores. Default servicing costs are also higher on GNMA pools because the FHA, VA and GNMA expose the servicers to higher potential default costs than the GSEs, such as the risk of servicing costs exceeding the VA maximum guaranty amount.¹¹ Ideally we would illustrate the potential magnitude of the VA guaranty risk by splitting the FHA and VA mortgages in the GNMA pools, but the Y-14 mortgage servicing rights valuations data are not granular enough to allow this split.

The model also generates different predictions for the prepayment rates of GSE and GNMA pools. In particular, the higher LTVs and lower credit scores of mortgages in GNMA pools lead to lower projected prepayments than on GSE pools. Actual GNMA prepayments may be higher than the model's prediction, as the model is based on portfolio loans and so does not incorporate the FHA and VA streamlined refinance programs.

Figure 2 shows cumulative mortgage default rates projected by the first-lien mortgage model, separately for large banks' portfolios of mortgages serviced for the GSEs and GNMA, for data as of December 2023. As compared to the baseline scenario, under the severely adverse scenario cumulative defaults rise rapidly in the early quarters, eventually reaching 11% for GSE loans and 21% for GNMA loans after 36 quarters.

Figure 3 shows cumulative mortgage prepayments (excluding defaults) projected by the first lien mortgage model, separately for large banks' portfolios of mortgages serviced for the GSEs and GNMA, for data as of December 2023. As compared to the baseline scenario, under the severely adverse scenario, interest rates decline sharply, leading prepayment rates to roughly double.

We convert the lifetime cumulative default rates that we estimate to conditional default rates (CDR) and similarly convert the lifetime cumulative prepayment rates into conditional prepayment rates (CPR). The CDR and CPR are annualized measures of the share of currently performing loans that are expected to default (CDR) or be prepaid (CPR) in the course of a year. Across the 18 banks that report Y-14M data on their mortgage servicing portfolios as of December 2023, the interquartile range of our estimated stress CDR shock spans 41 to 67 bps. The range reflects differences in the banks' servicing portfolios: some banks service more seasoned loans than others, for example, and some banks service more loans for GNMA. The interquartile range of our estimated stress CPR shock spans 497-628 bps, reflecting variation in the interest rates on the mortgages in bank servicing books as well as other factors.

Defaults under stress could be higher in the future or for other mortgage servicers. For example, house price appreciation may be more moderate in the future, leading to higher loan-to-value ratios and a larger number of defaults under any subsequent economic stress. Another possibility is that the proposed bank capital rules may change how bank servicers engage in the mortgage origination and servicing markets.¹² In particular, the removal of the requirement to deduct MSR amounts beyond a threshold from regulatory capital may cause banks to expand their mortgage originating and servicing activity. In that case, banks' servicing books may become less concentrated in seasoned mortgages.

As preparation for exploring these possibilities later in this note, we calculate the stress CDR shocks under two additional scenarios. First, we roll back the house price appreciation that has occurred since 2020 and assume that house prices were unchanged from March 2020 onward (or from the origination date, if later). Second, we calculate CDRs separately for different combinations of credit score, marked-to-market LTV, and Agency type (GSE/GNMA), and then weight these CDRs in accordance with the distribution of these characteristics in the mortgage market overall.¹³

Estimating the Sensitivity of MSR Valuations to Changes in Mortgage Defaults and Prepayments

Our second step is to estimate the sensitivity of MSR valuations to a rise in mortgage defaults and prepayments, as measured by the CDR and CPR. We have information from 16 large banks that reported their MSR valuations separately for loans serviced on behalf of the GSEs and GNMA from 2013 to 2019.¹⁴ Banks reported the change in their MSR values that would result from 100, 500, and 1000 bp increases in each of the CDR and CPR. We extract sensitivities to CDR and CPR shocks for each bank and year, which we express as the dollar change in the fair value of the MSR per dollar of underlying unpaid principal balance (UPB) that results from a CDR or CPR shock.¹⁵ Scaling by unpaid principal balance rather than MSR value ensures measured sensitivities are not overly influenced by periods when MSR values are depressed.¹⁶ Given that banks report sensitivities to multiple shock sizes, we considered both linear and non-linear responses of MSR value to CDR and CPR shocks.¹⁷ However, as the non-linearity appears relatively unimportant, the analysis presented assumes a linear response of MSR value to CDR and CPR shocks. The sensitivities based on an assumption of a quadratic response are presented in the Appendix.

We convert our estimates from loss relative to UPB in response to a CDR or CPR shock to percentage changes in MSR value by assuming that MSR values equal one percent of UPB, which is close to the historical median.¹⁸ To construct a consistent time series of the average response in MSR valuations to CDR or CPR shocks, we need to account for the fact that some banks did not provide usable submissions in all years. We do this by regressing percent change in MSR value on a set of bank and year fixed effects. The regression-adjusted time series is constructed by adding the average of the bank fixed-effect coefficients to each year fixed-effect coefficient.

Figure 4 plots these average changes in MSR valuations from a 100 bp increase in the CDR for each year from 2013 to 2019. The series are shown separately for GSE and GNMA portfolios and calculated on both an equal- and UPB-weighted basis. In our discussion of the results, we put more weight on the UPB-weighted sensitivities, as we expect firms with larger servicing portfolios to have more sophisticated models that capture more comprehensively the complexities of MSR modeling. Table 2 shows the average of these time-series values across years.

Table 2. Modelled Decline in MSR Value from a 100bp CDR Shock

Table 2 indicates that GSE portfolios are projected to lose about 6% of their value in response to a 100 bp CDR increase. As shown in Figure 4, this estimate is stable over the sample period and the equal- and UPB-weighted sensitivities are similar. The changes in GNMA MSR valuations in response to the CDR shock, however, vary across time and across the equal- and UPB-weighted averages. Averaging across years, the UPB-weighted sensitivity is a 17% decrease in MSR valuations (Table 2), but the projected decrease is 22% in 2013 and 14% in 2019. The larger decrease in 2013 likely reflects the conditions at the time: foreclosure backlogs meant elevated costs for servicing defaulted loans, and the cumulative declines in house prices since their 2006 peak meant that borrowers were less likely to cure their delinquencies and loan losses were more likely to exceed the VA guaranty.

Given this evidence that the costs that servicers incur when borrowers default on their mortgages vary with the economic and policy environment, we also consider a more conservative figure, which might be more reflective of sensitivities under a severe housing downturn. To calculate this "max" estimate, shown in Table 2, we identify the largest magnitude decline in MSR value ever reported by each bank to a 100 bp CDR shock. We then average these bank-specific maximums. For GSE pools, the "max" estimate is only a bit lower than the average. For GNMA pools, the "max" estimate is a 24% decrease, substantially more than the overall average decline of 17%.

The average sensitivities shown in Figure 4 obscure considerable variation across banks in their estimated sensitivity of MSR valuations to CDR shocks. One indicator of this is the large difference between the Ginnie Mae equal-weighted and UPB-weighted average. Another indicator is the fact that across firms, the modelled sensitivity is about 3 times as large at the 75th percentile as at the 25th percentile for GSE portfolios and 1.5 times as large for GNMA portfolios. While this variability could reflect differences in the characteristics of bank portfolios, differences in modelling assumptions across banks likely also play a role. Most banks report in the Schedule I data that they use their own proprietary default model, with the remainder showing little overlap in default model vendor choice.

Table 3 provides banks' modelled change in MSR value from a 100 bp increase in the CPR, which might commonly result from refinancing activity triggered by a sharp decrease in mortgage rates. As with the CDR estimates, we report this sensitivity separately for GSE and GNMA portfolios, and average across banks alternatively on an equal- or UPB-weighted basis. We show only the averages calculated over all years of the data, as there is little variability over time in this sensitivity. As with the CDR shocks, we focus on the UPB-weighted sensitivities, and convert to percentage changes in MSR value assuming that MSR values equal one percent of UPB.

Table 3. Modelled Decline in MSR Value from a 100bp CPR Shock¹⁹

As compared with the sensitivity to mortgage defaults, the sensitivity to prepayment shocks is more similar between GSE and GNMA portfolios. In addition, there is greater consensus. Across banks, the modelled sensitivity is only about one third larger at the 75th percentile than at the 25th percentile.

Estimating the Change in MSR Valuations Under Stress

Our third step translates the CDR and CPR shocks generated by the stress scenario into MSR valuation changes.

For the risk of mortgage defaults, we interact the CDR shocks with the sensitivities presented in Figure 4, separately for the GSE and GNMA portfolios. Those two loss rates are then combined based on the assumed GSE and GNMA share of aggregate MSR holdings. Table 4 presents the projected loss rates associated with mortgage defaults under a variety of scenarios.²⁰ The "Base" stress scenario takes the CDR shocks observed in the most recently run severely adverse scenario, which were around 50 bps for GSE pools and 100 bps for GNMA pools. The product mix is matched to what large banks reported in the year-end 2025 Y-14M submission, where servicing pools were approximately 90% GSE loans and 10% GNMA loans. This yields a loss in MSR value of 4.8%. If instead we apply the conservative "max" CDR sensitivities that banks reported during the Schedule I collection period the loss rate increases to 6%.

Table 4: Projected reductions in MSR values - from Mortgage Defaults (CDR shocks)

The securitized mortgages that large banks currently service are likely to perform well under a default scenario because they are seasoned and have low LTVs. Future servicing portfolios might not have such favorable default characteristics. For example, house prices might not remain at such elevated levels. The Federal Reserve's Financial Stability Report suggested that as of March 2026 house prices were about 20 percent higher than levels consistent with their historical relationship to fundamentals.²¹ In addition, the proposed changes to the treatment of MSRs in the revised capital rule might induce banks to re-enter the servicing market for securitized mortgages, in which case banks might be servicing more newly originated loans.

To gauge the loss rates that might materialize under stress for a more representative share of the securitized mortgage market, we perform two exercises. First, in the "no equity buildup" scenario we take our sample of bank-serviced mortgages and roll back the house price appreciation that they experienced since 2020. This change notably raises the projected CDR increases under stress to around 90 and 170 bps for GSE and GNMA loans, respectively. Under such a scenario, aggregate losses on MSR values are projected to reach around 8%. Second, in the "entire market" scenario we re-weight the mortgages in the banks' securitized servicing book so that their characteristics match that of securitized mortgages overall, as found in the right-hand side columns of Table 1. Under this scenario, mortgages in both GSE and GNMA pools have higher LTVs, and GNMA pools make up 30 percent of banks' servicing portfolios. These changes raise projected CDR increases under stress relative to the base scenario because they move it towards greater risk. Altogether, aggregate MSR loss rates in this scenario are projected to be near 13%.

MSR losses due to mortgage defaults may differ from these estimates in an actual stress scenario. The scenarios in Table 4 represent a 36% decline in nationwide home prices, a bit larger of a decline than experienced in the Global Financial Crisis. However, as an offsetting consideration, this exercise measures only changes in modelled fair values and not actual changes in market value. Under conditions involving widespread mortgage distress, the market for MSRs may be illiquid, making it difficult to realize full carrying value in MSR sales.

Rate-driven prepayment shocks also have a large effect on MSR values in the severe stress scenario. As shown in Figure 3, the CPR for large banks' GSE and GNMA servicing books increases by approximately 570 and 390 bps in the stress scenario. Our estimated CPR effect for the servicing market as a whole (not shown in Figure 3) is lower, at roughly 480 and 140 bps for GSE and GNMA loans respectively, reflecting the significantly higher LTVs of the broader servicing book. Combining these CPR shocks with the estimates in Table 2 suggests that higher prepayment activity in a stressed scenario could lead to a decrease in MSR valuations of roughly 20 percent for large banks' current servicing book, or about 14 percent for the market as a whole.

However, a fall in MSR valuations generated by higher prepayment poses less stress for a bank than a fall in MSR valuations generated by higher defaults. In the higher prepayment case, banks generally recoup much of the MSR-valuation decline immediately through their hedges and in the ensuing weeks through a pickup in mortgage origination activity and the creation of new MSRs. In the higher default case, banks do not receive offsetting revenue. MSRs that have lower values due to prepayment also can generally be sold at prices close to their carrying value if a bank needs to raise funds, but this may not be the case for MSRs whose values fall due to default.²³

For nonbank servicers, the calculation may be somewhat different. On the one hand, nonbanks often recapture more of their refinancing borrowers than banks do, meaning a larger share of their MSRs may be replenished in the weeks after the drop in interest rates. On the other hand, nonbanks, unlike banks, commonly use their MSRs as collateral for borrowing facilities. Nonbanks also are less likely to hedge the rate-driven fluctuations in their MSR valuations. If rates fall, nonbanks may receive margin calls from their lenders on their MSR facilities. In the absence of offsetting hedge revenue, nonbanks may not have the funds to meet these margin calls, leading to substantial financial strain at the nonbank.

Conclusion

Our results indicate that the sensitivity of bank MSR valuations to economic distress depends crucially on the characteristics of the bank servicing books and on the macro environment. In 2025, large banks primarily serviced older loans that had benefited from substantial house price appreciation. Large banks also had minimal exposure to GNMA servicing, which entails greater default servicing costs than GSE servicing. With a servicing book this strong, even under the severely adverse stress scenario of the stress-test model, the average bank would experience only about a 5 percent reduction in its MSR valuations associated with the additional mortgage defaults.

However, the current rosy circumstances may not be a good guide for decreases in bank MSR valuations in the future. House price appreciation may be less robust, and banks may return to their historical role of being the primary servicer for all segments of the mortgage market. Under these circumstances, our estimates suggest that mortgage defaults would lead to larger decreases in MSR valuations.

In addition to defaults, an economic stress scenario may trigger significant mortgage prepayments through a reduction in mortgage rates. While a prepayment shock may lead to large drops in MSR valuations, the risks to the bank overall are different than the risks associated with a default-driven decline in MSR valuations. In the prepayment shock case, the bank usually has hedges in place or anticipates future originations and MSR creation that may partially offset the drop in MSR valuations. The risks in the prepayment scenario center around whether the hedges behave as expected and the origination income and new MSRs materialize as expected. In contrast, no such offsetting factors are in place for a bank when MSR valuations decrease because of default.

References

Benson, David, You Suk Kim, and Karen Pence (2026). "Nonbank Securitizers and Credit Supply." May. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4615551.

Board of Governors of the Federal Reserve System (2024a). "2024 Stress Test Scenarios (PDF)." February.

Board of Governors of the Federal Reserve System (2024b). "2024 Supervisory Stress Test Methodology (PDF)." March.

Board of Governors of the Federal Reserve System (2026). "Financial Stability Report (PDF)." May.

Hamdi, Naser, Erica Xuewei Jiang, Brittany Lewis, and Manisha Padi (2025). "The Rise of Non-Banks in Servicing Household Debt." August. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4550175.

Kim, You Suk, Steven Laufer, Karen Pence, Richard Stanton, and Nancy Wallace (2018). "Liquidity Crises in the Mortgage Market." Brookings Papers on Economic Activity, Spring.

Appendix

This appendix shows the estimated changes in MSR values under the assumption that the relationship between CDR or CPR shocks and changes in MSR valuations is quadratic rather than linear (as assumed in the main text).

Appendix Table A.1 compares the changes in average UPB-weighted MSR valuations under the linear and quadratic extrapolations for CDR shocks of 50 and 200 bps (the range of CDR shocks implied by our default scenario in the main text). The extrapolated quadratic shock curves on average are slightly concave, so value changes are slightly amplified for small CDR shocks and muted for large CDR shocks.

Appendix Table A.1. Effect of CDR Shock on MSR Valuations Under Linear and Quadratic Extrapolation

Appendix Table A.2 repeats the exercise for CPR shocks. The CPR shocks in the main text span a larger range of 100 to 600 bps, so the quadratic extrapolation has a larger damping effect for the CPR results than the CDR results.

Appendix Table A.2. Effect of CPR Shock on MSR Valuations Under Linear and Quadratic Extrapolation