ONE SOURCE. POWERFUL SOLUTIONS. : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : Big Data, Presented by: Mark Katibah, Senior Vice President, Director of Data & Analytics Consulting, LPS Strategic Consulting Services Mark Milner, Vice President, LPS Data & Analytics Joel Farrand, Senior Analytics Consultant, LPS Strategic Consulting Services Sponsored by: 1 Big Data Comes From Many Asset Classes and Is Being Used in New Ways Lending Business Example Competitor Analysis Loan Decisions Property & Comps Loss Reserving Data & Analytics Forecasts Portfolio Monitoring Collections Customer Acquisition 2 October 2, 2013 1
Big Data Presents Big Issues and Big Opportunities Importance of data and model validation Developing next generation operational dashboards to drive business processes Combining data with model results to better focus resources Portfolio stress testing requires processing Big Data New CFPB rules mean sellers and buyers of assets need to be Big Data savvy when servicing transfers 3 Data and Model Validation Has Become a Key Issue for Regulators Model validation is the set of processes and activities intended to verify that models are performing as expected, in line with their design objectives and business uses. www.occ.treas.gov/news-issuances/bulletins/2011/bulletin-2011-12a.pdf OCC 4 October 2, 2013 2
Data Validation Processes Need to Be Defined and Utilized Documented uniform data definitions Analysis and reporting Each field - percentage populated Each field s distribution by pre-defined ranges Logic-based exception tracking Logic for dealing with data failing exception tests Transformations required to prevent impact to overall statistics 5 Model Validation Scope & Requirements Are Defined by Business Use Decision regarding individual loan or portfolio? Near-term or long-term projection scenario? 3.0 Millions 6.0 5.0 4.0 3.0 2.0 1.0 National Seriously Delinquent Inventory Optimistic Pessimistic 0.0 2009 2010 2011 2012 2013 2014 2015 2016 Actual Model (Known Rates) Need to minimize model error or forecast error? Focus on ordering or ranking (discrimination) or prediction of outcome (calibration)? Prepayment % % of Loans to PSDD6 2.5 1,200 2.0 1,000 1.5 800 1.0 600 0.5 400 0.0 200 2005 2006 2007 2008 2009 2010 2011 2012 2013 Serious Delinquency Rate per Month Monthly Prepayments Actual Pess 0 2005 2006 2007 2008 2009 2010 2011 2012 2013 100 LPS: PSDD6 90 80 Original FICO 70 1.2 Actual Model (Calibrated) Model (Uncalibrated) Random 60 Actuals Updated FICO 1.0 50 40 Poorly Calibrated Calibrated Model 0.8 Model 30 20 0.6 10 0 0.4 0 10 20 30 40 50 60 70 80 90 100 Decile 0.2 0.0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 6 October 2, 2013 3
Basic Model Validation Approaches 100 90 LPS: PSDD6 Lift curves to measure model ranking or discrimination power % of Loans to PSDD6 80 Original FICO 70 60 Random Updated FICO 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100 Decile Back-test tracking to measure model calibration Serious Delinquency Rate per Month 1.2 1.0 0.8 0.6 0.4 0.2 Actual Model (Calibrated) Model (Uncalibrated) Actuals Poorly Calibrated Calibrated Model Model Stress testing to measure sensitivity to changes in inputs New Monthly Seriously Delinquent Loans 0.0 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Actual Baseline Opt Pess 500 450 400 350 300 250 200 150 100 50 0 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 7 ONE SOURCE. POWERFUL SOLUTIONS. : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : Operational Dashboards Turning Data Into Insights 8 October 2, 2013 4
Turning Data Into Insights Through Dashboards Access Interpret mounds of data using visually engaging dashboards Leverage intelligent drilldowns to provide view of data in actionable context Overlay industry data to drive strategic business decisions Alerts Gear metrics to be early warnings React earlier to key trip-wire conditions Analytics Determine downstream impacts of volume movement View metrics based on conditions across groups Compare results to meaningful criteria Conduct what-if analyses in real-time 9 Looking at ALL Data having a map and compass to make sense of complex data. Level 1 Show all indicators consolidated across multiple organizational, transactional and historical data sources Level 2 Mid-level metrics expose deficiencies and critical business drivers Level 3 Low-level metrics pinpoint specific details/transactions impacting performance 10 October 2, 2013 5
Looking Deeper Into Data metrics that drive business decisions and enable action. Tightly align top-level results to mid-level process metrics and use dashboards to drive alerts to avoid failures 11 Looking Horizontally Across Data monitor other volume metrics to determine flow and velocity. Combining multiple data sources that intertwine relevant data so that trends and subtle relationships become apparent, allowing you to stay ahead of and plan for volume-based activities. 12 October 2, 2013 6
Summary Embrace a Broad View Take advantage of information consolidation via dashboards Capture a view of all metrics and their conditions, then decide where to focus Build Alerts Build in notifications when threshold are reaching tolerances BEFORE targets are missed Build in recovery time prior to failure point Actively manage potential failures Leverage the Analytics The data can tell you the story of where your business had been Trending can show where the business is going Leverage the dashboard to trigger proactive resource management by volume monitoring outside normal scope 13 ONE SOURCE. POWERFUL SOLUTIONS. : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : Applying Big Data to Models to Solve Problems or Meet Regulatory Requirements 14 October 2, 2013 7
Behavioral Prepayment & Credit Models Using 45+ data elements from servicing data to predict mortgage loan and borrower behavior Economic Drivers Behavioral Drivers Loan Data Drivers Loss Severity Drivers Interest Rates House Prices Original LTV Current LTV (ZIP/price-tier HPI Calculated) Age of Loan Seasonality Payment Changes Exposure to Refinancing Equity Geography Credit Score Loan Size Loan Type Purpose Property Type Documentation SatO Property Value Lien Position Fixed Costs Variable Costs Lost Interest Good portfolio data can be used to discriminate among loans and predict outcomes 15 Big Data Can Drive a Model to Generate Even More Big Data Economic Drivers Behavioral Drivers Loan Data Indicators To Month 0 (Origination) Cur 0-29 30-59 60-89 90+ FC REO PP To From Cur 91.5 5.41 Month 1 2.98 Cur 0-29 30-59 60-89 90+ FC REO PP 0-29 30.6 To 46.5 18.41 4.09 From Cur 91.5 5.41 Month 2 2.98 30-59 13.0 24.6 Cur 28.0 0-29 28.3 30-59 60-89 90+ FC 3.33 REO PP 0-29 30.6 To 46.5 18.41 4.09 60-89 From Cur 9.53 7.88 91.5 14.5 5.41 Month 3 22.9 30.0 11.6 3.53 2.98 30-59 13.0 24.6 Cur 28.0 0-29 28.3 30-59 60-89 90+ FC 3.33 REO PP 90+ 0-29 1.36 0.86 30.6 To 0.76 46.5 1.93 18.41 85.2 5.50 3.52 4.09 60-89 From 9.53 Cur 7.88 91.5 14.5 5.41 Month 4 22.9 30.0 11.6 3.53 2.98 FC 30-59 3.03 1.01 13.0 0.65 24.6 Cur 0.46 28.0 0-29 6.14 30-59 28.3 80.4 60-89 5.90 90+ 2.38 FC REO 3.33 PP 90+ 1.36 0-29 0.86 30.6 To 0.76 46.5 1.93 18.41 Month 85.2 5 5.50 3.52 4.09 60-89 From Cur 9.53 7.88 91.5 14.5 5.41 22.9 30.0 11.6 3.53 2.98 FC 3.03 30-59 1.01 13.0 0.65 24.6 Cur 0.46 0-29 28.0 6.14 30-59 28.3 80.4 60-89 5.90 90+ 2.38 FC REO 3.33 PP 90+ 0-29 1.36 0.86 30.6 To 0.76 46.5 18.41 1.93Month 85.26 5.50 3.52 4.09 60-89 From Cur 9.53 91.5 7.88 5.41 14.5 22.9 30.0 11.6 3.53 2.98 FC 30-59 3.03 1.01 13.0 0.65 24.6 Cur 0.46 28.0 0-29 6.14 28.3 30-59 80.4 60-89 5.90 90+ 2.38 FC 3.33 REO PP 90+ 0-29 1.36 30.6 0.86 To 46.5 0.76 18.41 1.93Month 85.27 5.50 3.52 4.09 60-89 From Cur 9.53 7.88 91.5 14.5 5.41 22.9 30.0 11.6 3.53 2.98 FC 30-59 3.03 13.0 1.01 24.6 0.65 Cur 28.0 0.46 0-29 28.3 6.14 30-59 80.4 60-89 5.90 90+ 2.38 FC 3.33 REO PP 90+ 0-29 1.36 0.86 30.6 0.76 46.5 1.93 18.41 85.2 5.50 3.52 4.09 60-89 From 9.53 Cur 7.88 91.5 14.5 5.41 22.9 30.0 11.6 3.53 2.98 FC 30-59 3.03 1.01 13.0 0.65 24.6 0.46 28.0 6.14 28.3 80.4 5.90 2.38 3.33 90+ 1.36 0-29 0.86 30.6 0.76 46.5 1.93 18.41 85.2 5.50 3.52 4.09 60-89 9.53 7.88 14.5 22.9 30.0 11.6 3.53 FC 3.03 30-59 1.01 13.0 0.65 24.6 0.46 28.0 6.14 28.3 80.4 5.90 2.38 3.33 90+ 1.36 0.86 0.76 1.93 85.2 5.50 3.52 60-89 9.53 7.88 14.5 22.9 30.0 11.6 3.53 FC 3.03 1.01 0.65 0.46 6.14 80.4 5.90 2.38 90+ 1.36 0.86 0.76 1.93 85.2 5.50 3.52 FC 3.03 1.01 0.65 0.46 6.14 80.4 5.90 2.38 16 October 2, 2013 8
Big Data- and Model-Driven Portfolio Segmentation Can Support Collections LPS Prepayment & Default Model Probability of Prepayment 6, 12, 24 Months Probability of Seriously DQ 6, 12, 24 Months Probability of Default 6, 12, 24 Mos + Life Expected Loss Current Estimated LTV Portfolio Segmentation Risk Focused Sort by Expected Loss Account for Payment Day Segment for Collections and Loss Mitigation 17 17 Big Data is Critical for Stress Testing: the Federal Reserve s CCAR Comprehensive capital analysis and review Mandatory stress tests of bank assets Performed annually in November/December Part of a comprehensive capital planning process Final capital plan requires Fed approval Goal: ensure that banks have sufficient capital to continue operations throughout times of economic and financial stress accounting for each bank s unique risks 18 October 2, 2013 9
CCAR Requirements Are Significant All asset classes Mortgage, HE, auto, commercial, credit card, etc. Data quality is critical Fed is collecting it too Residential mortgage portfolio & home equity Probability of default (PD) and loss given default (LGD) Dollar losses 3 Fed-defined HPI and rate scenarios 2 user-defined scenarios of HPI and rates Must run model in-house Must use same model in ALM framework $10+ billion asset size 19 Defining Outputs and Time Horizons Is Critical Clear definitions get everyone on same page PD and LGD What is default Securities market: liquidate at loss from REO Servicing group: 90 days past due When does loss occur actual vs. accounting 1 st lien: 180 dpd, foreclosure sale, REO sale? HE: 120 dpd, other What is needed to input into your capital model? 20 October 2, 2013 10
Model Calibration and Validation Is a Critical Requirement of CCAR 25.0 25.0 $ Losses per Month 20.0 15.0 10.0 5.0 Monthly Dollar Losses (MMs) 20.0 15.0 10.0 5.0 0.0 2009.5 2010 2010.5 2011 2011.5 2012 2012.5 2013 2013.5 0.0 2009 2010 2011 2012 2013 2014 Good model fitting depends on data quality and availability 21 ONE SOURCE. POWERFUL SOLUTIONS. : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : CFPB Impact on Servicing Transfer: Reporting and Balancing 22 October 2, 2013 11
CFPB Requirements 23 CFPB Requirements 24 October 2, 2013 12
CFPB Requirements Specify which post-transfer audits the transferee servicer conducts to confirm that all data were properly transferred, and whether the transferee servicer corrects any identified errors. 25 Finally A transferor servicer s policies and procedures may provide for transferring documents and information electronically, provided that the transfer is conducted in a manner that is reasonably designed to ensure the accuracy of the information and documents transferred and that enables a transferee servicer to comply with its obligations to the owner or assignee of the loan and with applicable law. 26 October 2, 2013 13
So Where Does All That Leave a Lender Involved in a Transfer? 27 Transferor or Transferee? Can the transferee comply with obligations? What activities would the transferor have taken in the near time period and was the necessary data sent for the transferee to do those activities? Identify critical activities and let that drive data identification. 28 October 2, 2013 14
Did You Plan Your Work? Did You Work Your Plan? Document, document, document! What would you supply in answer to the question, How do you know? What reporting and tracking is available to show all tasks were complete, issues were identified and resolved timely, system transitions were executed and results tracked? Did you update your plan with a new set of eyes? 29 Diligence Things to consider when developing your own policies and procedures: Do you have summary reports to begin your balancing? Do you have a test plan for loan-level balancing? Is the transaction small enough for 100% sampling? Consider critical fields for each functional area Don t just balance internally to your own side of the transaction, balance across servicers as well! Does the other servicer have the capability to provide the necessary information to comply? 30 October 2, 2013 15
Where Did the Loans Come From and Where Did the Loans Go? How many loans did each functional unit contribute? Where in those units did the loans come from? How long had those loans been in those stages? 31 Remediation Does the transferee servicer correct any identified errors and consumer complaints? If an error happens, how quickly is it identified and fixed? What tracking and reporting is part of the transfer plan and how often is it reviewed and by whom? 32 October 2, 2013 16
Mine Public Information for Clues Supervisory Highlights, Summer 2013 Published by CFPB 33 Finally Do you have the backing of your Compliance Department, Operational Risk Department and Legal Department? 34 October 2, 2013 17
Mark Katibah mark.katibah@lpsvcs.com 704-246-6593 Mark Milner mark.milner@lpsvcs.com 415-955-8781 Joel Farrand joel.farrand@lpsvcs.com 704-754-0772 Questions? 35 October 2, 2013 18