Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Feb 2017
Afshin Goodarzi
afshin@1010data.com
What is a mortgage?
A borrowers view
lender and the borrower? P = amount of principal, net of initial payments, meaning "subtract any down-
payments"
i = periodic interest rate
n = total number of payments
Credit score
Origination Loan Appraised Amort Reset Periodic Life
Loan # LTV Rate Product Purpose DTI Term Date upb amount Value FICO prop type PPT MI GOVT Period Period cap cap Ceiling Floor ppt term
1 80% 6%Fixed Purchase 42% 30 25-Jul-05 90000 90000 112,500.00 680SFR NO NO VA 360
2 8% 7%IO Refi-cash out 30% 15 25-Jul-04 75000 750000 900,000.00 780condo yes yes no 180 12 2% 6% 13% 4%3 years
3 90% 6.50%arm purchase 45% 30 15-Jan-05 123000 125000 136,666.67 6002-4 yes yes no 360 6 2% 5% 11.50% 3.50%4 ywars
Delinquency
Measured in 30 day increments, 30, 60, 90.
Prepayment or curtailment
Partial or full payment of principal prior to the term of the loan
Transition Matrix
Our Focus
St
St-1
We will focus on the transition to zero or Prepayment.
( Option to payoff a mortgage at any time in the life of
the loan )
Regional economy
Unemployment
Cost of option
Interest Rates
House prices
Loan Age
M
D
I
What was considered prepay risk in 1989
Prepayment risk = D + R + M + I
D: Default Risk = 0 Backed by the full faith and credit of the US
government hence considered by market participants to be default-free
R: Interest rate risk substantial
M: Relocation (moving) risk 5% to 7% of home owners move per year
I: Idiosyncratic prepay complex
better
source?
S curve
SMM single month mortality
smmi = 100 *
( FS Fi )
( Fi1 FS )
Data
source:
McDash
Servicing
&
FHFA
Interest rate paths: The Vasicek Model
Mean
Variance
Age of loan
Prepayment Speed Assumptions
100% PSA assumes
n prepayment rates of 0.2% CPR
in the first month following
origination of the loans with a
constant 0.2% CPR in each
succeeding month until the 30th
month.
n On and after the 30th month,
100% PSA assumes a fixed
annual prepayment rate of
6.0% CPR
Assumption : early prepayment is
small Not always true! 100% PSA
http://www.beyondbond.com/pdf/2.pdf
House price index
Unemployment
Observations?
Unemployment and foreclosure
Is this real?
Is it always true?
Underwriting effects of pre/post 1990
How should this effect be incorporated?
J. Lounsbury observes:
The unemployment rate for each state was for August, 2010.
1. Below about 9% unemployment there may be a weak linear effect on foreclosure from the level of
unemployment. However the correlation is poor.
2. Above about 9% unemployment there is a greater than linear dependence of foreclosure rate on
unemployment. Although the correlation is good, just looking at the scatter evident on the graph raises the
question of whether there may be additional factors related to foreclosure activity. For example, the four
states with the biggest housing bubbles all fall in the group of states with the largest increase in foreclosure
rates. https://www.creditwritedowns.com/2010/10/unemployment-and-foreclosures.html
What is burnout?
If borrower has
been exposed n
times to low
rates an no refi
then they are
burnt out!
Are there better
measures?
http://www.beyondbond.com/pdf/2.pdf
Data Growth M by N matrix
N number of observations
M number of variables
Other data:
Servicer Text
Servicer Voice
Growth of both M and N is Faster Future
o
Disclosure Packet
U.S.
WellsBancorp
Fargo WAMU
Agencies
UNDERWRITER REVIEW
n
IF counteroffer is
Bank of America Fannie Mae
Accept File From accepted (Change to
Fannie FNMA MBS
s
Int. Rate, Lower LTV, Mortgage Purchases
PennyMac Dealers
Queue
Correpondent
Approved with Conditions Review Approval
etc.) The mortgage
Company will Modify Channel
Mae Issued
PHH Corp.
The underwriter delivers and support docs DENIED/
u
the terms and
conditions that must be COUNTEROFFER
cleared before closing can
Verify Information Resubmit. Countrywide GMAC
is true and current
Citigroup
IF Denied with no
occur Lenders/
Use guidelines to
make decision m
counteroffer, the file
is considered Dead. Mortgage
Street and
Banks Flagstar Life Insurance
e Internet/ Bancorp Freddie Mac
Freddie
Non- Companies
Lender is notified and contacts the borrower to give them the good news, and find the correct Chase FNMC
time to close. The closing is scheduled with the title company and borrower
Retail
Channel Ginnie Mae Ginnie GNM MBS Savings & Loans
The Mortgage Lender delivers a Closing Instructions Fee Sheet with all final charges and
instructions on how to build the HUD-1 Settlement Statement. The title company creates the
Mortgage Purchases Mae Issued
final HUD, and submits it to the lender for review. ONCE APPROVED.
The Title Company prints the Entire closing package (Approximately 120 pages) with all
Branch
Network of Smaller Players
required documents and a copy package for the borrowers records. They review the docs for Retail
accuracy and prepare to close with the borrowers. A notary public is necessary to sign the
Lenders Other
documents. Many times, title agents or representatives have their notary license.
Better algorithms
Marital
Status bias
https://www.bls.gov/opub/btn/volume-2/patterns-of-homeownership.htm
Cost of Refinance
Upto 200 bp in
refi fees and
taxes in Queens
Typically not
modeled
http://ai.stanford.edu/~ronnyk/mortgage.pdf
CCLTV Comparison Charts
Sources:
Freddie Mac Loan Level Data
Equifax Credit Risk Insight Agency
FHFA HPI
Different formulations
Refinance Incentive Measure
Spread..
Spread ratio.
Present Value Adjusted.....
Dollar equivalent...
Remaining $ Interest..
Where:
t is the date
C is the Coupon or Note rate
M is the Market rate as measured by the Freddie Mac Primary Mortgage Market Survey
T is the term of the loan measured in months
A is the age of the loan measured in months
Pic is the Principal and Interest as reported without any escrows.
Pim is the Principal and Interest as computed off of the fixed rate without any escrows.
Confluence of Forces => Better Models?
count
Coarse vs. Fine Models Refinance Incentive
Distribution of the Refi Incentive Beta is long tailed for the Fine
models case!
Average Beta for Refi Incentive is higher in the coarse model 2.0
as compared to the fine models 1.84. Suggesting that the Refi
incentive measure is more important and more positively
correlated in general. Which is simply not true.
Note that the classical s-curve is not as pronounced in this case
Beta
Prepay Rate
-1 Sigma
+1 Sigma
Refi Incentive
Refi Incentive
Fine vs. Coarse: Cash flows
Homework
Data source
https://www.consumerfinance.gov/data-
Download data for
2 states
research/hmda/explore.html
At least one year of data
Questions:
Are minorities discriminated against in lending?
n What metrics did you use in this data to prove or disprove?
n What factors did you control for?
n Did you use other data sets? Why?
n Should you use HMDA data to improve your prepay model? How would you do it?
n Is there a Geographic effect?
n Are there other trends in the data that you see? Are the effects you are describing getting worst or better
over time?
Q&A
Opportunities for expanding/extending the data
Horizontally
Better
formulations of drivers
New data sources
Vertically
More granularity in time
Sources
https://research.stlouisfed.org/wp/
2012/2012-017.pdf
sfg Big data hype
is Over!
Big Data
When the coffee table book arrives the hype has peaked
Volume
i
Velocity M
Big Data- A Definition IMHO
# of tables? 10s 1000s
Variety
Volume
Velocity
# of Rows? Millions - Trillions
# of bytes? Terabyte - Petabyte
Complexity of the ER?
Number of users? 100s 10,000s
How much history? Hours, days, months, years?
Granular? Raw? Filtered? Transformed?
History
Usage
Granularity
The Question
Coarse vs Fine modeling
Are the populations homogeneous or heterogeneous