U.S. mortgage market – Painting a complete picture

Introducing:

Michael Bradley Freddie Mac

Michele Raneri Experian

©Experian 3 4/12/2017 Experian Public Vision 2017

1. Macro industry trends, recently and since the recession

2. A tale of two populations: Foreclosure vs. bankruptcy analysis

3. How Freddie Mac is reducing cost and creating efficiency

Contents

©Experian 4 4/12/2017 Experian Public Vision 2017

Macro industry trends, recently and since the recession

©Experian 5 4/12/2017 Experian Public Vision 2017

Macroeconomic trends

©Experian 6 4/12/2017 Experian Public Vision 2017

0

20

40

60

80

100

120

140

160

Ind

ex

Consumer confidence index = Recession

Consumer confidence

Source: Conference Board

Peak 3Q 2000

Low 1Q-2009

December 2016: 113.7

Dec 2016: Up 18% YoY

©Experian 7 4/12/2017 Experian Public Vision 2017

0

5,000

10,000

15,000

20,000

25,000

Po

ints

Dow Jones Industrial Average

DOW

Sep 2007

13,896 pts

Feb 2009

7,063 pts

Jan 2017

Up 21% YoY

January 24: 19,938

©Experian 8 4/12/2017 Experian Public Vision 2017

-9.0

-6.0

-3.0

0.0

3.0

6.0

9.0

12.0

15.0

Perc

en

t ch

an

ge

Real Gross Domestic Product (percent change from preceding period)

Real GDP change

3Q16: 5.0%

FOMC forecast

2017: 1.9-2.3%

2018: 1.8-2.2%

2019: 1.8-2.0%

Longer: 1.8-2.0%

4Q08 = -8.3%

Source: Bureau of Economic Analysis, Economic Projections of Federal Reserve Board Members and Federal Reserve Bank Presidents

©Experian 9 4/12/2017 Experian Public Vision 2017

U.S. jobs gained or lost through December 2016

0%

2%

4%

6%

8%

10%

12%

-1,000

-800

-600

-400

-200

0

200

400

600

Un

em

plo

ym

en

t rate

M

on

thly

jo

b c

han

ge (

tho

usan

ds)

One-month net change (thousands)

Source: Bureau of Labor Statistics

Unemployment rate 4.7%

FOMC forecast 2017: 4.5-4.6% 2018: 4.3-4.7% 2019: 4.3-4.8%

Longer: 4.7-5.0%

8.6M jobs lost in recession, regained by April 2014

©Experian 10 4/12/2017 Experian Public Vision 2017

Interest rates and home sales

0

1

2

3

4

5

6

7

8

9

10

0%

1%

2%

3%

4%

5%

6%

7%

8%

Ho

mes s

ale

s (m

illion

s)

Mo

rtg

ag

e r

ate

Interest rate Recession Mortgage rate Home sales

December 2016

Mortgage: 4.20%

Home sales: 5.49M

Source: Realtor.org

©Experian 11 4/12/2017 Experian Public Vision 2017

$100

$150

$200

$250

$300

$350

$400

Th

ou

san

ds

Sales price of existing homes U.S. Northeast Midwest South West

Home prices up year-over-year

U.S. +4.0%

YoY (Dec)

December 2016: $232,200 US

Source: National Association of Realtors

©Experian 12 4/12/2017 Experian Public Vision 2017

Credit trends

©Experian 13 4/12/2017 Experian Public Vision 2017

Predicts risk of borrower

• Likelihood of future serious delinquencies (90 days later or greater)

• Any type of account

24-month performance

Score range of 300–850

• Higher scores represent a lower likelihood of risk

• Lower scores are higher risk

VantageScore® 3.0 Model overview

A = Super-prime 781–850

B = Prime 661–780

C =Near prime 601–660

D = Sub-prime 500–600

F = Deep sub-prime 300–499

©Experian 14 4/12/2017 Experian Public Vision 2017

Mortgage +54%

HE Loan +24%

Bankcard +15%

Auto Loan

and Lease +10%

HELOC +10%

Retail +6%

Personal Loan -6%

Source: Experian IntelliViewSM

YoY Change 4Q15 to 4Q16

Originations by lending product

0

100

200

300

400

500

600

700

800

0

20

40

60

80

100

120

140

160

180

2011Q 1

2011Q 3

2012Q 1

2012Q 3

2013Q 1

2013Q 3

2014Q 1

2014Q 3

2015Q 1

2015Q 3

2016Q 1

2016Q 3

Mo

rtga

ge

in b

illion

s

Bil

lio

ns

Auto Loan & Lease Bank Card HE Loan

HELOC Personal Loan Retail

©Experian 15 4/12/2017 Experian Public Vision 2017

15.2% 15.3%

14.8%

14.2%

13.8%

13.3%

12.8%

11.7% 11.7% 11.9%

11.0%

11.5%

12.0%

12.5%

13.0%

13.5%

14.0%

14.5%

15.0%

15.5%

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Avera

ge U

S D

eb

t-to

-In

co

me

Debt-to-Income Insights℠ trends

©Experian 16 4/12/2017 Experian Public Vision 2017

0

100

200

300

400

500

600

700

1Q 07 1Q 08 1Q 09 1Q 10 1Q 11 1Q 12 1Q 13 1Q 14 1Q 15 1Q 16

Dolla

rs o

f lo

an

s o

rigin

ate

d

(bill

ion

s)

Source: Experian IntelliView℠

-100%

-50%

0%

50%

100%

1Q 13 1Q 14 1Q 15 1Q 16

Percent change YoY 2013-2016

4Q16:

$695 BN

54.07% YoY

Mortgage originations

©Experian 17 4/12/2017 Experian Public Vision 2017

729 729 729 729

300

400

500

600

700

800

0%

10%

20%

30%

40%

50%

60%

70%

2013 Q4 2014 Q4 2015 Q4 2016 Q4

Super Prime Prime Near Prime Subprime Deep Subprime Average

Source: Experian IntelliViewSM

Distribution of mortgage loan originations by VantageScore® band

©Experian 18 4/12/2017 Experian Public Vision 2017

0.0%

0.5%

1.0%

1.5%

2.0%

2.5%

3.0%

3.5%Percent balances 60-89 DPD 90-180 DPD

Source: Experian IntelliViewSM

4Q16 Data

60-89: 0.54%

90-180: 0.65%

Mortgage late stage delinquency rates

©Experian 19 4/12/2017 Experian Public Vision 2017

Mortgage delinquency rates By state

Best/worst five states*

4Q16 national mortgage

60-89 DPD

1. Alaska 0.24%

2. Oregon 0.27%

3. Washington 0.28%

4. Hawaii 0.31%

5. Colorado 0.31%

···

47. Alabama 0.73%

48. New Jersey 0.75%

49. Louisiana 0.82%

50. West Virginia 0.83%

51. Mississippi 1.41%

*Includes District of Columbia

4Q16

Source: Experian IntelliViewSM

High Low

©Experian 20 4/12/2017 Experian Public Vision 2017

Mortgage trades

45

50

55

60Outstanding loans

80%

82%

84%

86%

88%

90%

92%Balance (% to original loan)

4Q16: 52.63M

4Q16: 82.1%

©Experian 21 4/12/2017 Experian Public Vision 2017

A tale of two populations: Foreclosure vs. bankruptcy analysis

©Experian 22 4/12/2017 Experian Public Vision 2017

Current market situation:

• Consumers locked out for a new mortgage by GSEs

– 7 years after foreclosure

– 4 years after bankruptcy

Analysis will review:

• Is there a significant difference between these two populations

– Time to next new mortgage trade

– Performance of these trades

• Segmentation and profiling of future bad performance

Analysis objective

©Experian 23 4/12/2017 Experian Public Vision 2017

Analysis design

Foreclosure and bankruptcy populations

Jun 2007 2016 Dec 2007

New mortgage trade open window for

analysis population and performance window

Consumers with a mortgage that

was foreclosed or bankrupt

©Experian 24 4/12/2017 Experian Public Vision 2017

New trades by year (vintages)

0

5,000

10,000

15,000

2008 2009 2010 2011 2012 2013 2014 2015 2016

All Loans Bankruptcy Foreclosure

0%

25%

50%

2008 2009 2010 2011 2012 2013 2014 2015 2016

%GSE loans Bankruptcy Foreclosure

©Experian 25 4/12/2017 Experian Public Vision 2017

Post event origination mix

0%

20%

40%

60%

80%

100%

2008 2009 2010 2011 2012 2013 2014 2015 2016

FHA vs. VA loan type BK-FHA BK-VA BK-Other FC-FHA FC-VA FC-Other

©Experian 26 4/12/2017 Experian Public Vision 2017

New mortgage trade performance (%90+ DPD)

0%

20%

40%

2008 2009 2010 2011 2012 2013 2014 2015 2016

All trades Bankruptcy Foreclosure

-10%

10%

30%

2008 2009 2010 2011 2012 2013 2014 2015 2016

GSE trades Bankruptcy Foreclosure

©Experian 27 4/12/2017 Experian Public Vision 2017

Predicting risk Probability 90+DPD

Super prime/Prime

Near prime

Subprime/Deep

©Experian 28 4/12/2017 Experian Public Vision 2017

Deep subprime

Prime

51% of the file 3% bad rate 6% of bads

Prime

Predicting risk Probability 90+DPD in Foreclosure

Near prime VantageScore®

differentiates the highest

and lowest scoring population

19% of the file 15% bad rate 43% of bads

Subprime

©Experian 29 4/12/2017 Experian Public Vision 2017

Predicting risk Probability 90+DPD in foreclosure

Subprime

Prime Differentiation gets harder in the middle

Overlay scores and premiers through a chi-square

Income Insight W2SM

>$40k

1.9% bad rate

<$40k

7.2% bad rate

30% of the file 7.6% bad rate 34% of bads

Near prime

Derogatory credit amount

>$34k

11.4% bad rate

<$34k

3.7% bad rate

Income Insight W2SM

>$60k

7.9% bad rate <$60k

14.1% bad rate

High Risk Low Risk

©Experian 30 4/12/2017 Experian Public Vision 2017

Deep subprime

Prime 62% of the file 4% bad rate 36% of bads

Prime

Predicting risk Probability 90+DPD in bankruptcy

Near prime

13.5% of the file 17% bad rate 32% of bads

Subprime

©Experian 31 4/12/2017 Experian Public Vision 2017

Predicting risk Probability 90+DPD in bankruptcy

Subprime

Prime

24.5% of the file 8.9% bad rate 32% of bads

Near prime

Income Insight W2SM

>$53k

8.9% bad rate

<$53k

19.4% bad rate

#trades opened in L24 months ever DQ

15.5% > 2 trades

12.1% bad rate

9% <2 trades

3.6% bad rate

30+

Low Risk

High Risk

©Experian 32 4/12/2017 Experian Public Vision 2017

• Segmentation and profiling through overlaying premiers and scores further separates consumers with higher and lower risk

• Conservative analysis methods were used here and clearly demonstrates opportunities

• A retrospective test of your applicant population will quantify the swap-in expansion benefit

Key points to remember

Big Data and Advanced Analytics

May 8, 2017

Reducing Costs and Speed to Close

© Freddie Mac 34

Freddie Mac’s Technology Objectives

Ensuring success through the development of an innovative suite of solutions designed to

deliver the Certainty, Usability, Reliability and Efficiency our customers expect.

Certainty

Give you confidence that the loans you originate meet

the requirements for delivery and sale to

Freddie Mac.

Usability

Provide intuitive and easy-to-use tools, with clear and actionable

feedback.

Reliability

Instill trust that our tools will be available, and perform,

when you need them.

Efficiency

Lower cost to originate via automated data validation and focusing your attention

where it needs to be.

Customer Success

Loan Advisor Suite

© Freddie Mac 35

Loan Advisor Suite:

Component Overview

Application Processing / Underwriting

Pre-Closing Closing Loan

Delivery Servicing

Lo

an

Ad

vis

or

Su

ite

The Loan Advisor Suite builds the manufacturing quality story required for

greater purchase certainty.

Loan Product Advisor

Loan Collateral Advisor

Loan Quality Advisor

Loan Closing Advisor

Selling System

Loan Coverage Advisor

Business Intelligence

© Freddie Mac 36

Rethinking Valuation Practices to Meet Customer

Needs

“Our challenge then is to reconsider the appraisal – the process of

developing the opinion of value – and the content of the appraisal report

to serve the future underwriting, pricing, and management of mortgage

credit risk.”

“If credit investors can obtain the same or equivalent information from

other sources – we live in a big data world today, after all – then what is

the value proposition of an appraisal?”

There is a shortage of appraisers and few new entrants into the

profession: The cost of obtaining an appraisal has increased to $700-

$900 in many markets and the time to complete the assignment can take

as long as six weeks.

Freddie Mac’s internal research indicates collateral valuation can be reengineered.

© Freddie Mac

37

Group A: “Chesterfield MO is a

nice suburb of St. Louis.”

Group B: “Chesterfield MO is a

nice suburb of St. Louis.

Do you think the median sale

price in Chesterfield is more or

less than $900,000?”

Group B’s estimate was ~ 40 to 52% higher!

Appraisers Suffer from Anchoring Bias

© Freddie Mac 38

Assess primary functions performed by the current appraisal process

» Estimate value

» Inspect property condition

Use big data and advanced analytics to automate these same functions

» Improve the timeliness and customer satisfaction of our Single-Family guarantee

business while maintaining credit quality and prudent risk management

» In certain cases, in lieu of a new appraisal, leverage our risk tools and big data

to develop an independent opinion on the collateral condition and value

» Leverage prior experience, our top-tier AVM (HVE), and big data/advanced

analytics

Design Structure

Our design leverages big data and emphasizes function.

© Freddie Mac 39

Examples of the types of data we are using to develop various checks:

» Repeat sales data to assess house price growth of the property compared to the

appreciation rate for other homes in the local market (zip code).

» Prior appraisal data from UCDP:

– Review the last known condition and quality rating

– Confirm there are no historical and unacceptable findings for the subject property

– Confirm that the property is a single-family property

» Public records: quality ratings, property age, and tax assessment data

– Typically, homes of higher quality construction will decline at a slower rate due to longer lasting

materials.

– Also used to determine distressed sale information.

Big Data & Advanced Analytics

Big data serves as the foundation for rule development.

© Freddie Mac 40

Knowledge Graph is a knowledge base that connects millions

of pieces information to enable “smarter” search capabilities.

Google Knowledge Graph was introduced in 2012 using many

information sources such as Search Clicks, Linked Pages, etc. to

extrapolate the edges (relations) between the nodes (entities).

Facebook has the Social Graph whose nodes and edges are

constructed from the social network’s millions of users and their

data.

Knowledge Graphs

© Freddie Mac 41

We can use Graph clustering algorithms to map out neighborhoods programmatically and

determine the right neighborhood boundaries.

Application of Knowledge Graphs to Mortgages!

© Freddie Mac 42

Imagine estimating the value of a subject home (yellow outline) with three comparable homes. We

can use the Home Knowledge Graph to find homes that are similar to the three comps and

increase the number of properties used to model the subject home’s value, decreasing the noise in

the system.

Home Knowledge Graph

© Freddie Mac 43

Map out neighborhoods and determine

the right neighborhood boundaries

Increase the number of properties used

to estimate the subject home’s value

Application of Knowledge Graphs

to Appraisals

Similar

Neighborhood

Estimation

© Freddie Mac

Image processing capabilities detect

appraisals with photo issues

Address: 123 Main Street Address: 210 Maple Avenue

© Freddie Mac 45

The Dawn of Artificial Intelligence

Artificial

Intelligence

Machine Learning Old, tried-and-true statistical tools

don’t get you far in the age of Big

Data.

Need machine learning’s nonlinear

capabilities to model most

phenomena.

‘Learning’ algorithms

Big Data The more data, the more we can

learn!

Sensors and networks (Hyperion)

Think about a HouseFax report

Infinite Computing Ongoing progression from a

scarce and expensive

resource to one that is

plentiful and free.

Encourages experimentation.

Once an interface becomes available, the

technology is about to take off! Think of Watson on the Cloud!

https://www.youtube.com/watch?v=V1eYniJ0Rnk

© Freddie Mac 46

Artificial Intelligence

1047*

10127**

*, ** Source: Showdown: Artificial Intelligence and Go, The Economist,

March, 12, 2016, pages 73-74.

© Freddie Mac 47

A Cautionary Tale

* From: Weapons of Math Destruction: How Big Data Increases Inequality and Threatens

Democracy, by Cathy O’Neil, Crown Publishing, New York, N.Y., (2016).

“The math-powered applications powering the data

economy were based on choices made by fallible human

beings….many of these models encoded human

prejudices, misunderstanding, and bias into the software

systems that increasingly managed our lives”*

Examples:

- Models used to evaluate teachers

- Model that ranks universities and departments

- For-profit universities

- Parole models

© Freddie Mac 48

The Road Ahead

FCRA Regulatory Guidance

Fair Lending

Recent paper by Google on Lattice Regression

http://arxiv.org/pdf/1505.06378.pdf

How to Leverage in the Context

of a Regulated Financial Institution?

Emulation Model

Machine

Learning

Deep

Learning

Entity

Resolution

©Experian 49 4/12/2017 Experian Public Vision 2017

Experian contact:

Michele Raneri michele.raneri@experian.com 714-830-5843

Questions and answers

Michele Raneri

@MLRaneri

©Experian 50 4/12/2017 Experian Public Vision 2017

Share your thoughts about Vision 2017!

Please take the time now to give us your feedback about this session.

You can complete the survey at the kiosk outside.

How would you rate both the Speaker and Content?