CECL Workshop Vintage Method John J. Doherty, CPA MEMBER OF ALLINIAL GLOBAL, AN ASSOCIATION OF LEGALLY INDEPENDENT FIRMS 2017 Wolf & Company, P.C.
Introduction John J. Doherty Member of the Firm jdoherty@wolfandco.com 617-261-8172 2
Overview Vintage analysis measures losses based on the origination date and the historical performances of loans with similar risk characteristics. Vintage methodology works well with loans that follow patterns that are similar and predicative for subsequent generations of loans (homogeneous). Vintage analysis requires segmentation and stratification of the loan portfolio, with the additional requirement that loans be stratified by origination period. 3
Pros & Cons Pros Forecasting ability can improve as more data is collected, allowing more precise qualitative and quantitative adjustments to be made at the vintage level Adequately segmented data eliminates qualitative changes in portfolio growth/mix Can be used to isolate changes in economic environment, collateral value and underwriting to a given year Easier to understand Consistent with disclosure requirements and expectations of life of loan estimate Flexible to add new information for new loans Cons Data mining can be extensive based on the level of disaggregation does your loan system provide enough data to efficiently pull together the required data? Monitoring of prepayments is required to ensure that baseline data that drives the calculation is reasonable. Doesn t work well with revolvers or loans subject to frequent renewal (i.e. commercial) 4
Example: Residential Real Estate 30 year, first position lien, fixed rate residential real estate Consider separate calculations for variable versus fixed rate due to prepayment speeds Generally should apply vintage to a homogeneous portfolio where underwriting standards and loan terms and behavior are generally consistent Loans conforming to secondary market standards are likely homogeneous with respect to underwriting standards 5
Example: Residential Real Estate (in 000 s) Vintage Principal Collections Year Originations 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2011 40,000 6,667 6,125 6,548 6,978 7,152 6,530 - - - - - 2012 42,400-7,067 6,941 7,397 7,581 6,922 6,493 - - - - 2013 44,944 - - 7,491 7,840 8,036 7,337 6,882 7,357 - - - 2014 47,641 - - - 7,940 8,518 7,778 7,295 7,799 8,311 - - 2015 50,499 - - - - 8,417 8,244 7,733 8,267 8,810 9,029-2016 53,529 - - - - - 8,922 8,197 8,763 9,338 9,571 8,739 Totals 6,667 13,192 20,980 30,155 39,704 45,734 36,599 32,186 26,459 18,600 8,739 Period End Loan Balances 33,333 62,542 86,506 103,991 114,787 122,582 85,983 53,798 27,339 8,739 - (Note: shaded regions are future estimates) Data through 12/31/16 is known Estimated life of loan is 6 years, but will vary based on rate environment and prepayment speeds Loans stratified by year of origination and type of loan Above is fairly linear, results will vary significantly by rate environment 6
Example: Residential Real Estate (in 000 s) Origination Charge-offs by Origination Year ($) Year 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Total 2011 52 84 112 124 64 36 - - - - - - 472 1.18% 2012-51 85 114 127 64 38 - - - - - 479 1.13% 2013 - - 49 85 117 130 59 39 - - - - 480 1.07% 2014 - - - 48 86 119 107 61 42 - - - 463 0.97% 2015 - - - - 45 86 90 112 66 47 - - 446 0.88% 2016 - - - - - 43 63 94 120 73 56-450 0.84% Totals 52 135 246 372 439 478 357 306 228 120 56-2,789 Origination Losses by Vintage By Year year Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Total 2011 0.13% 0.21% 0.28% 0.31% 0.16% 0.09% 1.18% 2012 0.12% 0.20% 0.27% 0.30% 0.15% 0.09% 1.13% 2013 0.11% 0.19% 0.26% 0.29% 0.13% 0.09% 1.07% 2014 0.10% 0.18% 0.25% 0.23% 0.13% 0.09% 0.97% 2015 0.09% 0.17% 0.18% 0.22% 0.13% 0.09% 0.88% 2016 0.08% 0.12% 0.18% 0.23% 0.14% 0.11% 0.84% Loss rates here correspond to actual charge-off history and loan balance data as previously presented, for 2011-2016. Blue highlights represent expected credit losses over the life of the loan vintages, qualitatively adjusted for reasonable supportable forecasted items (illustrated on next slide). 7
Example: Residential Real Estate Loss Rates by Vintage Q Factor by Vintage - Eg. MA Unemployment Origination Y1 Y2 Y3 Y4 Y5 Y6 Origination Y1 Y2 Y3 Y4 Y5 Y6 2011 0.13% 0.21% 0.28% 0.31% 0.16% 0.09% 2011 6.70% 6.70% 6.10% 5.10% 4.30% 3.20% 2012 0.12% 0.20% 0.27% 0.30% 0.15% 2012 6.70% 6.10% 5.10% 4.30% 3.20% 2013 0.11% 0.19% 0.26% 0.29% 2013 6.10% 5.10% 4.30% 3.20% 2014 0.10% 0.18% 0.25% 2014 5.10% 4.30% 3.20% 2015 0.09% 0.17% 2015 4.30% 3.20% 2016 0.08% 2016 3.20% 2017 2017 Average 0.11% 0.19% 0.27% 0.30% 0.16% 0.09% Average 5.35% 5.08% 4.68% 4.20% 3.75% 3.20% Loss/Q factor 1.96% 3.74% 5.67% 7.14% 4.13% 2.81% Estimate of expected losses: Calculating the correlation between loss factor and qualitative (Q) factor (MA unemployment). The correlation is established by measuring historical losses and tying to the Q Factor. The idea being that you can predict the Q factor and use this to adjust future loss ratios. Above: the average loss for Y2 is 0.19% which is 3.74% of the Q factor. 8
Example: Residential Real Estate (in 000 s) Loss Rates by Vintage Q Factor by Vintage - Eg. MA Unemployment Origination Y1 Y2 Y3 Y4 Y5 Y6 Origination Y1 Y2 Y3 Y4 Y5 Y6 2011 0.13% 0.21% 0.28% 0.31% 0.16% 0.09% 2011 6.70% 6.70% 6.10% 5.10% 4.30% 3.20% 2012 0.12% 0.20% 0.27% 0.30% 0.15% 2012 6.70% 6.10% 5.10% 4.30% 3.20% 2013 0.11% 0.19% 0.26% 0.29% 2013 6.10% 5.10% 4.30% 3.20% 2014 0.10% 0.18% 0.25% 2014 5.10% 4.30% 3.20% 2015 0.09% 0.17% 2015 4.30% 3.20% 2016 0.08% 2016 3.20% Average 0.11% 0.19% 0.27% 0.30% 0.16% 0.09% Average 5.35% 5.08% 4.68% 4.20% 3.75% 3.20% Loss/Q factor 1.96% 3.74% 5.67% 7.14% 4.13% 2.81% Loss Rates by Vintage Reasonable Supportable Forecast Origination Y1 Y2 Y3 Y4 Y5 Y6 Origination Y1 Y2 Y3 Y4 Y5 Y6 2011 0.13% 0.21% 0.28% 0.31% 0.16% 0.09% 2011 6.70% 6.70% 6.10% 5.10% 4.30% 3.20% 2012 0.12% 0.20% 0.27% 0.30% 0.15% 0.09% 2012 6.70% 6.10% 5.10% 4.30% 3.20% 3.15% 2013 0.11% 0.19% 0.26% 0.29% 0.13% 0.09% 2013 6.10% 5.10% 4.30% 3.20% 3.15% 3.10% 2014 0.10% 0.18% 0.25% 0.23% 0.13% 0.09% 2014 5.10% 4.30% 3.20% 3.15% 3.10% 3.15% 2015 0.09% 0.17% 0.18% 0.22% 0.13% 0.09% 2015 4.30% 3.20% 3.15% 3.10% 3.15% 3.30% 2016 0.08% 0.12% 0.18% 0.23% 0.14% 0.11% 2016 3.20% 3.15% 3.10% 3.15% 3.30% 3.75% Average 0.11% 0.18% 0.24% 0.26% 0.14% 0.09% Average 5.35% 4.76% 4.16% 3.67% 3.37% 3.28% Loss/Q factor 1.96% 3.74% 5.67% 7.14% 4.13% 2.81% Arrive at reasonable supportable forecast for Q factor (future unemployment) Multiply the average loss factor times Q ratio for each period to arrive at an estimated future loss 9
Additional Considerations Vintage analysis can also be applied without tying to a Q factor. The average loss by vintage is useful in itself and can be qualitatively adjusted for new information. Qualitative adjustments can be evaluated in a similar fashion to how they are arrived at now, with the inclusion of reasonable supportable forecasts. In our previous example, one specific Q factor (unemployment), was tied directly to loss rates used as the starting point for calculating expected losses. Additional qualitative adjustments can be made based on the economic current environment (delinquency, management etc.) and other forecasted items (Schiller index, foreclosure rates, interest rates, LTV, other economic data). Qualitative adjustments can be made by vintage or evaluated at the pool level. Vintage analysis identifies the loss emergence period (LEP), which may be relevant information for other methods. For example, a discontinued loan segment that is seasoned would require less reserves if the LEP is known. 10
Another Approach Qualitative Adjustments to Historical Losses Current conditions - 2016 vintage Refer to qualitative memo for detail analysis of metrics. Delinquency ratio is consistent year to year but higher than custom peer group. Net losses were relatively elevated during the years 2009 to 2012 and have since decreased and are favorable to peer. The recent trend is positive and the annual loss rate will decrease as lower loss years are added to the historical period. The Bank tracks average FICO and LTV to identify changes in credit risk and there is no change in portfolio metrics. Based on current real estate valuations the average LTV of the portfolio should be improving and providing more collateral support. There have been no significant changes in lending policies, underwriting or management during 2016. There are no indications that the average annual loss rate for 2016 should be adjusted for credit quality concerns. Management will make a qualitative adjustment to increase the historical loss 3 basis points for lack of historical loss consistency across the different vintages. 11
Forecast 2016 vintage Another Approach Qualitative Adjustments to Historical Losses Credit risk drivers are unemployment, local real estate values, and interest rates. (General consensus centers around a 2 year forecast). Unemployment Trend for state unemployment rate is positive at 3.2%, decreasing from 4.3% in 2015. Regional unemployment is 3.6% at December 2016. Fed outlook over the next 6 years is marginal declines over the next two years, followed by increases through 2022 up to 4%. The Bank's loan committee assesses employment factor as stable. An adjustment of 5 basis points will be made based on expected future increases in unemployment. Real estate Per Sept 2016, FRB Boston's quarterly publication NEPPC: Home prices continued to grow both nationally and regionally, with national growth rates continuing to exceed regional rates. All six New England states reported positive house price growth year-over-year, but these gains all trailed the national rate. The Bank's loan committee assesses this factor as stable. No forecast adjustment is necessary for real estate values. 12
Another Approach Qualitative Adjustments to Historical Losses Forecast (continued) Interest rates Fed increased rates during 2016 and effect is reflected in year end prepayment speed assumption. Bloomberg median factor for 30 year FNMA MBS with same terms is 239% (or 5.75 yr life) at 12/31/16. Generally the bank s prepayment speeds lag secondary market speeds. The prepayment speed assumption for this estimate has been adjusted to a 6 year life based on historical performance and already reflects extension due to the 2016 rate hike. Management is conservative in the determination of prepayment risk and no adjustment has been made for future rate hikes as management cannot forecast this factor. 13
Another Approach Qualitative Adjustments to Historical Losses Vintage loss factor and Adjustments - 2016 Vintage average loss (previous 5 year expected vintage loss plus 2016 actual) 1.03% Adjustments Current Conditions Forecasts 0.03% 0.05% Total 2016 Expected Loss Factor - Vintage 1.11% 14
Qualitative Factors Schedule out economic factors by vintage and analyze for trends that should warrant additional consideration. Economic Factor Summary by Vintage Year 2012 2013 2014 2015 2016 Micro Data: Delinquencies 0.76% 0.93% 1.05% 1.09% 0.00% Non-accrual rates 0.82% 0.78% 0.82% 0.71% 0.69% Underwriting stable stable stable stable stable Added 5 resi. Management stable stable stable lenders stable Macro Data: Unemployment 6.70% 6.10% 5.10% 4.30% 3.20% Interest Rates Shiller Price Index 145.53 161.11 168.28 176.98 186.54 Change 6.50% 10.71% 4.45% 5.17% 5.40% Others to consider if available Average LTV Average FICO scores 15
Observations 1. Data mining by segment and vintage is critical 2. Vintage analysis requires a lot of data, but may result in a more precise estimate of expected credit losses. Loss rates decrease over time as borrower obtains equity in collateral, but are also impacted by qualitative considerations, including reasonable supportable forecasts. 3. Different approaches to applying qualitative adjustments 16
Questions? John J. Doherty Member of the Firm jdoherty@wolfandco.com 617-261-8172 17