Residential Mortgage Credit Model June 2016 data made beautiful
Four Major Components to the Credit Model 1. Transition Model: An idealized roll-rate model with three states: i. Performing (Current, 30-DPD) ii. Non-performing (60+ DLQ) iii. Terminated (Voluntary or Involuntary) 2. Liquidation Timing: Time from non-performing to terminated. 3. Loss Severity Model: A function of future home price scenarios as well as other factors 4. Prepayment Model: Separate functions predicting turnover and refinance rate Separate models for each loan type: Subprime, Alt-A, Prime, Second Lien, Agency Page 2
Conceptualizing Credit Transitions Actual Roll Rate Model Too many parameters to estimate Liquidation Current 30 DLQ 60 DLQ 90+DLQ FCL REO Voluntary Payoff Idealized Roll Rate Model A tractable problem Enhanced user control to adjust results Quicker model update and calibration cycle Speed of model runs Increased transparency Current Delinquent Liquidation Voluntary Payoff Page 3
Idealized Roll Rate Model The net roll rate is modeled. Conceptually this combines the probability of moving from current->delinquent and delinquent-> current. Transition to non-performing and Cure transition modeled separately at the loan-level, then combined to generate net roll rate Roll rates are applied to individual loans on a fractional basis. For example, if the Current >Delinquent transition rate is 2%, a current loan becomes 98% current and 2% delinquent in the next time period. Loan Period Loan Status 0 1 2 3 Performing 100.0 98.0 95.8 93.8 Non-Performing 2.0 4.2 6.2 Net Roll Rate Probability 2.00% 2.22% 2.15% Page 4
Key Model Factors The net transition rate from Current to Non-performing was statistically estimated from the data set, using the following parameters: Occupancy Status, Purpose, Loan Term, Loan Size Original FICO, Original DTI Loan Age and Seasonality Current CLTV Modification type (non-agency models) Documentation type (non-agency models) Current loans are separated into Clean and Dirty at the beginning of the simulation based on their delinquency and modification history RiskSpan s proprietary Vintage Quality Index (VQI) is used to control changing underwriting standards, this is used in the Agency sub-model only VQI: Empirically grounded way to control for vintage that will capture future changes in underwriting standards Page 5
Macroeconomic Factors Macro forecasts can be supplied by RiskSpan or by the user input Macro factors considered include: Home price index (zip, MSA, state, or national level) National unemployment rate Interest rates All Macro factors are time vectors Page 6
Net Roll Rate Framework Clean Dirty Non-Performing Starting States Clean Performing Dirty Performing 3 1 less 3 Non Performing (60+, FC) Period 1 Clean Performing Dirty Performing 2 less 3 Non Performing/ Pre-Default (60+, FC, REO) 1 less 3 Period n Clean Performing Dirty Performing 2 less 3 Non Performing/ Pre-Default (60+, FC, REO) 5 5 4 Terminal State Voluntary Prepayment Voluntary Prepayment Default # of transitions/models: 5 1. Clean Performing to Non Performing Model 2. Dirty Performing to Non Performing Model 3. Cure Model 4. Liquidation timing 5. Prepayment Model Page 7
Liquidation Timing Empirically-based timelines specific to each non-performing bucket (60+, 90+) with geography and non performing pipeline seasoning as key drivers Time from delinquent to liquidation is modeled as a Weibull function, based on: Geographic location Judicial versus non-judicial foreclosure process The probability a loan will liquidate is based on the number of months a loan has been non-performing Page 8
Loss Severity Calculation: 2 Options 1. Calculation-based: Expense assumptions plus stressed property discount model driven by non performing pipeline seasoning Deducting expenses accrued during non-performance Modeling residual losses. Based on the following characteristics: Origination vintage State / Geographic Location Loan Balance Length of time in liquidation process 2. Statistical loss severity model Model is Based on the Following Characteristics MTMCLTV Judicial v. Non Judicial State Time in non-performing pipeline Geography Loan Balance Other Characteristics For either option, private mortgage insurance (PMI) is taken into account and deducted from the sale price. Users can override PMI. Page 9
Prepayment Model 1. Turnover function: Models prepayment due to home sale (moving), the main driver is seasonality (high in the summer, low in the winter) 2. Refinance function: Accounts for borrowers propensity to pay off their loan and refinance to a lower rate. Follows the typical S-Curve approach, where refinance incentive is the main driver. Other model factors include: Burnout: Borrowers who miss opportunities to refinance are less likely to refinance in the future Current LTV: A High LTV may prevent a borrower from refinancing. Loan Size: Higher balance loans have a bigger benefit from refinancing FICO Score: Borrowers with higher FICO score prepay faster Page 10
Next: Appendix
Developmental Evidence: Credit Component Agency Model Page 12
Vintage Quality Index 160 140 RiskSpan VQI Historical Trend Max: Jan 2007-139.10 120 100 Reference: Jan 2003-100.00 80 60 Min: May 2012-62.98 40 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Page 13
Team Contact Info: Janet Jozwik Director RiskSpan, Inc. (703) 956-5200 (202) 531-9515 jjozwik@riskspan.com Page 14
TYSONS CORNER, VA - Headquarters RiskSpan, Inc. 8150 Leesburg Pike, Suite 740 Vienna, VA 22182 (703) 956-5200 rs-info@riskspan.com STAMFORD, CT - Tech Innovation Lab RiskSpan, Inc. 281 Tresser Blvd, Suite 1203 Stamford, CT 06901 (203) 355-1510 rs-info@riskspan.com CHARLESTON, SC - Production Analytics RiskSpan, Inc. 49 Immigration St., Suite 203 Charleston, SC 29403 (843) 466-8300 rs-info@riskspan.com