2. 2
Heard It Through The Grapevine
"The actual sensitivity of MSRs to implied volatility
is complex and somewhat controversial”
Ben Golub in "Mark-to-Market Methodology, Mortgage
Servicing Rights, and Hedging Effectiveness“
“The model we use doesn’t even get the sign right
for volatility hedging of MSRs”
A/L management advisor
“The price response to skew adjustment seems
exaggerated”
Hedge fund manager
Why do intuition and model disagree
when it comes to volatility?
3. 3
Observations
Modeling prepayments is only a means to an end
The goal is proper valuation and risk measurement
A mortgage is a callable amortizing bond
Prepayment models should be consistent with callable
bond models
Bonds (mortgages) are refunded (refinanced) when
the call option is worth more dead than alive
Therefore bond and mortgage models should respond
similarly to interest rate levels and volatility changes
4. 4
Dynamic Versus Static Variables
In an MBS model
Interest rate driven prepayments
Dynamically hedged
Modeled using a stochastic interest rate process
Other prepayments, such as turnover and defaults
Either not hedged or statically hedged
Modeled statically in CLEAN
5. 5
A financial engineer homeowner uses an option
valuation model
Refinances optimally
Others refinance too early or too late
Early refinancers are called “leapers”
Rarely occurs
Late refinancers are called “laggards”
AKA analysis shows that the 50 bps rule of thumb
is sensible
Most homeowners refinance near-optimally!
Optimum Option Exercise Provides
Benchmark for Suboptimal Behavior
6. 6
MBS Valuation Using CLEAN™
Two separate yield curves are required
One calibrated to mortgage rates
Other calibrated to MBS yields
Modeled using coupled lattice
Mortgage rates used to determine refis
Using notion of call efficiency
MBS rates used for discounting MBS cash
flows
7. 7
Benchmark yield curve and volatility
USD swap curve and appropriate swaption vol
Prepayment parameters
Laggard distribution
Turnover speed vector
Default/buyout speed and recovery percentage vectors
Refinancing cost
Fixed percentage of original principal
Homeowner credit spread
Analogous to corporate credit spread
MBS price/OAS
For EOD pricing, use OAS calibrated to TBA prices
CLEAN™ Model Input Parameters
8. 8
Calibration of CLEAN™:
Straightforward and Intuitive
Rarely adjusted
Laggard distribution
Turnover speed
Default recovery percentage
Refinancing cost
Occasionally adjusted
Homeowner credit spread
Default/buyout speed
9. 9
Calibrating Homeowner Credit Spread
For Agency Pools
Should be consistent with prevailing mortgage rates
Approximately 120 bps for current coupon pools
Implies refi option premium of approximately 40 bps
Higher credit spread for higher coupon collateral
Implies weaker credit, ceteris paribus
Calibrated to dealer consensus duration/convexity, and specified pool
pay-up grids
Additional factors that can be incorporated:
Fannie/Freddie vs. Ginnie/FHA
LTV, FICO
Year of origination
Loan size
Percent of non-owner occupied (low refi rate, high turnover rate)
Average points paid
Credit migration
15. 15
TBA Price Movement vs. Model Implied
Delta Movement
Actual Change in TBA Market Price vs. Sum of Implied Price
Change Due to Risk Factors (1/2/2009 - 7/1/2010)
-3
-2
-1
0
1
2
3
1/1/2009 4/2/2009 7/2/2009 10/1/2009 12/31/2009 4/1/2010 7/1/2010
%ofpar
FNCL 4.5 Δ price
Σ implied price change due to risk factors
16. 16
Why CLEAN™ Is Ideal for
Trading, Hedging, and Risk Management
Realistic transparent behavior
Based on well established financial and economic principles
Instead of mysterious mathematical formulas and parameters
Consistent with valuation models for callable bonds and
cancelable swaps
Calibration is straightforward and intuitive
Concretely defined model parameters
Easier to simulate
Model behavior always realistic
Based on fundamental financial and economic principles
Not on statistical fitting of historical behavior
And ridiculously fast
Criticial for simulation
17. 17
Modeling prepayments
Turnover and defaults modeled using deterministic speeds
Refinancings modeled using stochastic interest rate model
Modeling a mortgage
As a callable amortizing bond
A financial engineer will refinance when the option is worth more
dead than alive
Others will refinance too early (never really happens) or too late
(“laggards”)
Modeling heterogeneous refinancing behavior
Divide mortgage pool into 10 buckets according to laggard
parameter
Use a standard laggard distribution for a new pool
Modeling seasoned pools
Fastest refinancing buckets disappear first
Automatically accounts for ‘burnout’
The CLEAN™ Way
18. 18
References
Andrew Kalotay & Qi Fu (June 2009), A Financial Analysis of
Consumer Mortgage Decisions, Mortgage Bankers
Association.
Andrew Kalotay & Qi Fu (May 2008), Mortgage servicing rights
and interest rate volatility, Mortgage Risk.
Andrew Kalotay, Deane Yang, & Frank Fabozzi (Vol. 1, 2008),
Optimum refinancing: bringing professional discipline to
household finance, Applied Financial Economics Letters.
Andrew Kalotay, Deane Yang, & Frank Fabozzi (Vol. 3, 2007),
Refunding efficiency: a generalized approach, Applied
Financial Economics Letters.
Andrew Kalotay, Deane Yang, & Frank Fabozzi (December 2004),
An option-theoretic prepayment model for mortgages
and mortgage-backed securities, International Journal of
Theoretical and Applied Finance.
Available from http://www.kalotay.com/research