ZMdesk VERSION UPDATE JANUARY 2015 ZM Unified Total Prepayment Model Practical Solutions to Complex Financial Problems
ZM Financial System s Unified Total Prepayment Model (ZMUTP) INTRODUCTION ZM Financial Systems Unified Total Prepayment (ZMUTP) model treats different collateral types equally under a unified framework and projects both types of prepayments (voluntary and involuntary) simultaneously (hence the name Unified Total Prepay). The current ZMUTP model suite supports the following collateral types: 1. Conventional 30-Year, 15-Year 2. Conventional Hybrid Arm 3. GNMA 30-Year, 15-Year 4. GNMA Hybrid Arm 5. Conforming JUMBO 30-Year, 15-Year PREPAYMENT MODEL OVERVIEW A mortgage s loan schedule may be paid in advance of its scheduled pay-off date. Therefore, the ZMUTP model addresses three major factors that cause mortgage prepayments: 1. Home Sales; 2. Refinance; and 3. Default. Expressed as a percentage, total Conditional Prepayment Rate (CPR) projections derived from the ZMUTP model are aggregate projections from three separate sub-models: 1. Housing Turnover Prepayments resulting from home sales are impacted by various factors such as loan age, home price appreciation, seasonality and current mortgage rates. 2. Refinancing Prepayments due to refinancing occur when current mortgage rates are below the rate of a borrower s existing mortgage. Current and past interest rates, credit scores and loan-to-value (LTV) ratio, as well as additional factors, can influence refinancing activity. Generally, when prepayment speeds are very high, refinancing activity is the biggest component. 3. Defaults Prepayments through Defaults occur when delinquent loans are bought out (or prepaid) by an agency or servicer. For specific loan types, or during times of financial stress (economic situations, etc.), Defaults may account for a very large percentage of prepayments. Most often, however, Defaults play a minor part in aggregate prepayments. Page 2
ZMUTP MODEL: ELEMENT SPECIFICATIONS Housing Turnover, Refinancing and Defaults are modeled as individual sub-models in ZMUTP. This independence provides the ability to adapt and calibrate to changing economic and regulatory environments, which increases the ZMUTP model s flexibility and accuracy. For example, the refinancing component is modeled with its own parameters to capture prepayment behavior due to Seasoning, Burnout, Media-Effect, etc. Our three components are described using the 30-year conventional ZMUTP model as an example. 1. Housing Turnover Element One of the most important elements of out-of-the-money speeds are prepayments that occur due to existing home sales also known as Housing Turnover. Housing Turnover has significant impact during the early stages of the mortgage pool. The ZMUTP model calculates this component as: Turnover Speed Projection = Base Housing Turnover * Seasoning * HPA Adjustment * Lock-In * Mobility * Seasonality Base Housing Turnover is calculated by dividing the total existing home sales by total U.S. housing stock. This number is usually expressed as a percentage, and is a calculation of the average regularity of time in which a house is sold. The ZMUTP model utilizes a base housing turnover rate resembling rates used in the mid-1990s. See Figure 1. Figure 1. Housing Turnover, 1990-2013 Seasoning reflects the possibility of homeowners increased likelihood of moving as their new mortgage matures. The ZMUTP seasoning component considers the weighted average maturity of the housing pool, loan type and coupon spread at origination (SATO) for more robust calculations. Page 3
Regression analysis shows that the Home Price Appreciation (HPA) and home sales are highly correlated. For example, homeowners can use built-up equity to purchase a larger home, increasing a homeowner s ability to move. This increase in equity, along with HPA, increases the likelihood of prepayments due to housing turnover. The ZMUTP model addresses this by adding a HPA adjustment multiplier. Contrary to housing turnover is the Lock-In Effect. This reflects a homeowner s unwillingness to move for various financial reasons, such as current mortgage rates being higher than the present rate on their mortgage. The ZMUTP model considers this difference between the weighted average coupon of a pool and mortgage rates on a lagged basis when modeling the Lock-In Effect. The Mobility function calculates housing turnover as moves due to homeowner choice. For example, ARM borrowers tend to be more mobile than fixed rate borrowers: many are motivated to select a hybrid ARM because they expect to move before the initial fixed-rate period expires. In the ZMUTP model, we have elected to use a mobility multiplier for different collateral types in the housing turnover module. The Seasonality function takes into account the observed seasonal variation of home sales for each month: home sales drop during winter months and are at their peak during summer months. 2. Refinancing Element Refinancing is the most volatile part among all prepayment components. In the ZMUTP model, the refinance incentive is measured by how far in-the-money or out-of-the-money the homeowner s rate is compared to the New Mortgage Rate. Here New Mortgage Rate estimation is an involved computation with a number of moving parts. New Mortgage Rate Not everyone can qualify for the market mortgage rate. In the ZMUTP model, the refinance mortgage rate is determined by the primary market mortgage rate and Rate Adjustments. Rate Adjustments are made based on LTV, SATO and loan size. It also includes the credit curing effect borrowers with an above-prevailing market rate might gradually be eligible for a lower rate as their credit score improves and/or LTV falls over time. SATO is used as an initial estimate of the Rate Adjustment. A combination of the rate spread computed in this manner is used to derive the Rate Adjustment used in the New Mortgage Rate calculation. Eligibility Framework Related to Refinancing is Eligibility Framework, as not all loans in a pool are eligible for refinancing. Factors determining eligibility are primarily loan size, FICO and LTV. Additional adjustments depend on origination year (to account for the quality of collateral issued during a given period). When the market is under stress, high underwriting standards and services capacity constraints must be considered. What s more, the ZMUTP model allows for adjustment of the eligibility based on SATO. For example, Borrower A and Borrower B have similar credit scores and LTVs; however, Borrower A has a higher SATO due to high DTI or the presence of a second lien and therefore will have a downward adjusted eligibility. An eligibility adjustment percentage is calculated for the pool as a whole and applied to the refi S-curve (discussed on page 5). Non-eligible loans may have a second chance at eligibility depending on changing financial environments. Page 4
The S-Curve The Refinancing S-Curve shows graphically the prepayment rate as a function of the refi incentive (loan rate divided by current market rate). Generally, prepayment speeds increase slowly until the refi incentive gets to a meaningful amount, then speeds increase quickly before they slow considerably as the incentive gets past a certain point. See Figure 2. Figure 2. Sample Refinancing S-curve Also applied to this base projection: 1. Seasoning Seasoning reflects the fact that a homeowner who just took out a mortgage is less likely to refinance than a borrower who has had a loan for a period of time. Getting a mortgage on a home involves paperwork, can be time intensive and is sometimes an emotionally stressful event. Someone who just went through the ordeal of getting a loan may not want to relive that process right away if rates suddenly drop. On the other hand, a homeowner who has had a loan for a while and is presented with a refinancing opportunity may not mind going through the loan application process once again. The Seasoning measure goes from a value of zero (0) for brand new loans to one (1) for fully seasoned loans where the homeowner is willing to refinance. The model includes a parameter to cap the seasoning speed. This is needed to stop some collateral from seasoning too quickly. This cap generally begins when the mortgage is originated right before a large drop in the refinance rate or before the rates reach a historical low. 2. Loan Level Characteristics When available, the ZMUTP model utilizes loan level characteristics such as credit score, LTV ratio and loan size to improve prepay projections. Credit Score: Its impact to refinance includes two parts. The higher the FICO score, the more likely the borrower will be eligible to refinance when underwriting standards are high. So the FICO score changes refinance projection through the eligibility framework. At the same time, the FICO score is used together with SATO to adjust the new mortgage rate calculation in the ZMUTP model. Better scores and lower SATO lead to better mortgage rates. Page 5
LTV Ratio: When LTV is high, it is very difficult for the borrower to refinance even when the mortgage rate is at the borrower s favor. The ZMUTP model uses the current-loan-to-value (CLTV) to adjust the probability of refinance. CLTV is used because it is recalibrated with up-to-date home price indices (HPI) and takes into account loan amortization. In the ZMUTP model, we use FHLMC CMHPI (Conventional Mortgage Housing Price Index). Figure 3 displays the refinance S-curve for different LTVs. Figure 3. Refinance S-curve for different LTVs Loan Size: A borrower with a larger loan will save more money through refinancing than a borrower with a smaller loan. The loan size affects refinancing through the mortgage rate adjustment in the ZMUTP model. A standardized year 2000 equivalent loan size is used to determine the mortgage rate adjustment. Figure 4 shows the refinancing profile across various loan size buckets. Figure 4. Refinance S-curve for different loan sizes Page 6
3. Media Effect The Media Effect takes place when the media reports on significant downward movement in mortgage rates relative to historic lows. As more reporting is focused on mortgage rates and more news is generated around the topic, borrowers have an increased awareness of what is going on in the marketplace. This leads to higher refinancing activity. In the ZMUTP model, this effect is based on both the time since mortgage rates were previously this low and a term that approximates the ratio of the historical low to current mortgage rates. The Media Effect is currently less important than in the past; however, the model predicts its importance will gradually increase as time passes. 4. Cash-out Refinancing Numerous factors such as the financial crisis of 2007-2009, equity growth, tight underwriting and lower home price appreciation have led to a significant drop in Cash-out Refinancings. While Cash-out Refinancings include both borrowers taking out larger mortgage loans while lowering the rate on their mortgage, these refinancings are driven by the fact that borrowers were able to lock-in better rates. For the ZMUTP model, Cash-out Refinancings are a subset primarily related to equity rather than regular refinancings. Burnout Burnout occurs when a mortgage pool exhibits a slower response rate to favorable refinance conditions as time passes. This is the result of changes in the population of borrowers in the pool due to the rapid response of some borrowers to favorable refinance opportunities. As these borrowers exit the pool, the remainder become slower to respond (Burnout). This is modeled by tracking the evolution of a slow prepay speed population and fast prepay speed population (with the fast population having a smaller impact on total projected speeds as it becomes a smaller fraction of collateral). The ZMUTP model specifically tailors these two segments ( fast and slow ) to account for the slow segment utilizing a lower-amplitude S-curve. 3. Default/Buyout Element When loans in a pool become seriously delinquent or go into foreclosure, they will be bought out of the pool (or be fully prepaid) according to the process of the agency guarantee. These prepayments are called Defaults or Buyouts. Default/Buyout CPRs are generally determined by borrower credit quality. Defaults are more likely with poorer credit quality borrowers; therefore, Buyouts will be higher. In the ZMUTP model, various inputs are used, such as SATO, credit score, LTV and loan size, along with current rates and home prices. Default/Buyout is calculated as follows and is illustrated in Figure 5. Page 7
Projected Defaults/Buyouts = F(Age, SATO) * H(FICO, LTV, Loan size) Figure 5. Default multiplier by CLTV and FICO On the other hand, delinquency provides a good indication for near-term default probability. By analyzing historical delinquency data, we derived a transition matrix which is used to predict the probability of loans changing between delinquent status. When available, the ZMUTP model integrates the current delinquency information with the formula above to provide an adjusted default projection. The fundamental idea is to start by projecting the default rate via the current delinquency rate and gradually shift to the long term default projection as calculated from the default function above. Page 8
PREPAYMENT MODEL OVERVIEW Ginnie Mae (GNMA) pools are different from Fannie Mae and Freddie Mac pools in numerous ways. They usually are associated with lower credit scores, higher LTVs and smaller loan sizes. GNMA CONVENTIONAL Vintage OLTV FICO Loan Size OLTV FICO Loan Size 2010 94.40% 701 181,000 69.90% 759 223,000 2011 94.70% 697 178,000 71.10% 758 216,000 2012 94.60% 709 189,000 71.40% 760 227,000 2013 94.70% 709 192,000 71.70% 766 235,000 Source: EMBS, PIMCO As a result, the refinancing behavior is quite different for GNMA pools. This leads to different refinance S-curves in the ZMUTP GNMA models. In addition, GN borrowers have to pay mortgage insurance premiums in the form of an upfront fee (UFMIP) and an annual premium (MIP). MIP has the most impact on refinancing as the UFMIP is rolled into the new loan balance. Since 2008, FHA changed its insurance premiums many times. For example, before 2010 loans with an LTV greater than 95 would pay 55 bps in annual MIP; however, the current MIP on these loans is 125 bps. The ZMUTP model take this into account by adding certain extra spread to the mortgage rate for GN pools of different vintage years. This spread gradually decreases as the projected LTV decreases. HYBRID ARMS Since the financial crisis of 2007-2009, Hybrid ARMs make up a relatively small segment of the mortgage market. The ZMUTP hybrid ARM model s template uses our 30-year fixed-rate model as a base then incorporates key ARM behavior characteristics. One important observation on ARMs is that borrowers act fast to lock in a low long-term fixed rate when the spread is narrow and fixed rates are low. The ZMUTP hybrid ARM model captures this by calculating borrower refinance incentives based on a mixture of the 30-year rate and 5x1 mortgage rates. Also, since remaining hybrid borrowers tend to hang onto their low rate as long as possible, and then refinance in large numbers before their interest rate begins to float, the ZMUTP hybrid ARM model also assumes an interest-rate sensitive spike around the reset date. Additionally, hybrid ARM borrowers may be more prone to defaults given the nature of the loan. This gives hybrid loans a higher base line default rate. Page 9
JUMBO The ZMUTP JUMBO model focuses on pools with loan sizes greater than $471,000 ($625,500 in high-cost areas). Coupled with larger loan sizes, JUMBO loans have additional factors such as higher FICOs which increase refinancing sensitivity. At the same time, the differences between slow and fast populations are assumed to be much smaller for the ZMUTP JUMBO model. See Figures 6 and 7. Figure 6. FNCL 4s 2010 Figure 7. FNCK 4s 2010 Page 10
SUMMARY OF FORECASTED SPREADS Page 11
About ZM Financial Systems ZM Financial Systems brings practical solutions to complex financial problems, offering complete solutions in securities and fixed-income analytics, credit-adjusted ALM, liquidity, risk management, budgeting and funds transfer pricing. We also offer large bank solutions to meet the evolving regulatory risk reporting requirements. With nearly 1,000 institutions depending on ZMFS products/analytics to identify, measure and monitor risk and value in their balance sheets, we are one of the fastest growing financial software companies in the U.S. Founded in 2003, ZMFS is a privately-held corporation located in Chapel Hill, N.C. In addition to the 25 percent of our staff who have PhD s in the advanced quantitative field, our development and product support teams all have experience in the finance arena. Because our teams continuously collaborate, we can quickly navigate complex solutions to complete client-requested enhancements in days or weeks, versus months or years. Delivering state-of-the-art risk/reward analysis tools, such as ZMdesk, OnlineALM.com and OnlineBondSwap.com, our clients are empowered to uncover hidden risk while maximizing performance; test lending, investment and funding strategies; and respond to various regulatory requirements while efficiently delivering actionable information. For more information: Email: sales@zmfs.com Phone: 919.493.0029 Web: www.zmfs.com 5915 Farrington Road, Suite 201 Chapel Hill, NC 27517 www.zmfs.com 2015 ZM Financial Systems. All Rights Reserved.