Art or Science? Modeling and Challenges in the Post-Financial Crisis Economy Emre Sahingur, Ph.D. Chief Risk Officer for Model Risk Fannie Mae May 2015 2011 Fannie Mae. Trademarks of Fannie Mae. 2015 Fannie Mae. Trademarks of Fannie Mae. 1
About Fannie Mae: The leading source of residential mortgage credit in the U.S. secondary market Provides access to mortgage credit to enable families to buy, refinance or rent homes Securitizes mortgage loans originated by lenders into Fannie Mae mortgage backed securities (MBS) that it guarantees In 2014, provided $434 Billion in liquidity to mortgage market, enabling borrowers to complete 937,000 refis, 887,000 home purchases and provided financing for 446,000 units of multifamily housing Also provided 165,000 loan workouts to help homeowners stay in their homes or otherwise avoid foreclosure Total assets of roughly $3.2 Trillion on the consolidated balance sheet Source: 2014 10-K 2
So what is modeling anyway? 3
What I will try to cover today about modeling: Role of modeling and analytics in consumer lending Key elements of modeling & analytics The art and science in modeling The impact of the recent financial crisis on modeling Some challenges and opportunities in today s economy, and Development of new data sources and their impact on lending 4
Modeling and Analytics The Buzz Words Maximize return on marketing efforts through segmentation/microsegmentation Effectively assess risks to arrive at optimal decisions regarding new exposures Optimize product offerings, up-sell and down-sell offers Maximize the value of relationships through right offers across the customer lifecycle Assess trends in transactions to identify risks Develop effective loss mitigation programs for high-risk exposures Understand portfolio dynamics and sensitivities Allocate capital effectively across investments, Maximize risk-return tradeoffs and manage exposures When applied effectively, data-driven approaches and models are key enablers in achieving key goals for financial institutions. 5
Modeling and Analytics (M&A) in Consumer Lending M&A through the loan lifecycle: Marketing Underwriting & Pricing Servicing & Account Management Loss Mitigation Segmentation Response Modeling Product Development Credit Scoring Collateral Valuation Pricing & Loan Terms Fraud Controls Line Management Cross-sell Product Utilization Fraud Prevention Collections Strategies Loan Modifications Recoveries Modeling and data analytics help decide who to target, how to target, what to offer, how to manage relationships and risks. 6
A Closer Look at Key Data and Modeling Elements Data Models Decisions Business Usage Internal Data External Data New Data Macro Factors Loan performance Relationships Transaction-level Credit bureaus Public records Nontraditional Experiments New sources and partnerships Home Prices Interest Rates Unemployment Modeling & Analytics Behavior Response Utilization Default Prepayment Cash Flows Revenues Losses NPV ROC / ROI Portfolio Concentrations Correlations Stress Testing Decision making Marketing Campaigns Product Design Approval Pricing Account Management Portfolio Management Customer Engagement Business P&L Strategy Development Risk Management Financial Reporting Forecasting Modeling occurs at multiple levels, including loan, customer and portfolio levels, serving different stakeholders with distinct needs and sensitivities. 7
The Art and Science in Modeling Models are inherently imperfect predictors, they are based on historical data and assumptions Management judgment should be applied at every stage, from model design decisions to core underlying assumptions, to interpreting and applying model results Models can not be black boxes. Decision-makers should be able to effectively challenge model assumptions, outputs, sensitivities, and understand limitations and blind spots. Inputs Data inputs Assumptions Model Fitting Process Calibration Outputs Predictions Sensitivities Firms that can successfully marry cutting-edge modeling & analytics with sound judgment have been successful in avoiding traps in decision making. 8
It's tough to make predictions, especially about the future. The future ain t what it used to be. Yogi Berra 9
1975-01-01 1977-01-01 1979-01-01 1981-01-01 1983-01-01 1985-01-01 1987-01-01 1989-01-01 1991-01-01 1993-01-01 1995-01-01 1997-01-01 1999-01-01 2001-01-01 2003-01-01 2005-01-01 2007-01-01 2009-01-01 2011-01-01 2013-01-01 2015-01-01 A good example 180 140 100 S&P/Case-Shiller U.S. National Home Price Index January 2000 = 100 60 20 July 2006 February 2012-27% Sitting in July 2006, how could the housing market crash be predicted? Is having 30 years (actually longer, dating back to the Depression era) of no home price declines indicative of the future? What other data is there to help with projections? What about judgement? 10
Some more hard questions for modelers What happens when markets change quickly and in unforeseen ways? Are the changes we are seeing temporary or permanent? How quickly can models be updated and deployed? Should they be? How meaningful are estimates when data includes a major regime change? Does the past tell us anything about the future anymore? How much weight should we put on data from the crisis period? How do we model with limited history in the new regime? How can we predict the next downturn? How do we separate correlation from causality? Are leading indicators reliable? Can we predict the predictors? Management judgment should be applied in conjunction with model outputs where model uncertainty is high 11
A side note on correlations Source: Tyler Vigen's Spurious Correlations Blog, www.tylervigen.com 12
Source: Tyler Vigen's Spurious Correlations Blog, www.tylervigen.com 13
Key take-away: Question conclusions, challenge model outputs and apply management judgment and domain expertise Source: Tyler Vigen's Spurious Correlations Blog, www.tylervigen.com 14
Trends observed following the financial crisis Customer Behavioral Shifts Changes in payment hierarchy Changes in savings rates and household debt levels Rise of the millennials and use of technology and social networking Changes in Lenders Tightening credit standards Increasing capital levels Online lenders Peer-to-Peer lending Changes in Regulations CCAR / Dodd-Frank / Basel III requirements CFPB Banks are competing with new forms of lenders to meet the demands of a changing consumer base while tightening regulations challenge margins 15
Lenders are increasingly leveraging non-traditional data Traditional Non-traditional Small Summary Credit Bureau Savings accounts Tax returns Paystubs Public records Utility payments Cell phone service Residence stability Social network size Online sales activities Customer ratings Website activity Large lenders face some challenges in fully leveraging (big) data: Unclear data strategy or value proposition Disparate data sources, inherited data Big Transaction-level bureau data Merchant Profile/transactions Geospatial data Call transcripts Social media content structures Lack of data standards, retention policies Translating results into actionable insights New forms of data represent significant opportunities in consumer lending while capabilities are still being developed 16
Conclusions Data-driven approaches and models are heavily utilized in financial services Combination of cutting-edge modeling & analytics with a healthy dose of sound judgment is the right combination for success Changes in consumer behavior, lenders risk appetites, regulations following the financial crisis pose challenges for modelers Evolving technologies and new forms of data present new opportunities as well as new competition for lenders 17
Questions? 18