Asset Indivisibility, Security Design and Asset Quality

1
Asset Indivisibility, Security Design and
Asset Quality

• Chris Downing, Dwight Jaffee and Nancy
  Wallace
• Rice University and University of California,
  Berkeley
• CQA Conference 2006

2
Purpose of the Paper

• Model and provide empirical evidence for
• The quality of assets held by special purpose
  vehicles (SPVs).
• Model prediction
• Indivisible assets sold to SPVs are lemons.
• Empirical findings support prediction for one
  asset class
• Freddie Mac Participation Certificates (PC) sold
  to REMIC SPVs are significantly lower quality
  than PCs not sold to REMICs.
• Provide reduced form empirical evidence.
• Provide structural modelling evidence REMIC PCs
  exhibit systematically more efficient option
  exercise.
• Provide implied valuation differentials REMIC
  PCs are lower value.

3
What is a low quality PC?
4
Prior Theoretical Literature

• Market for Lemons (Akerloff, 1970)
• Agents possessing private information about true
  asset quality face problems of market breakdown.
• Signaling literature (Leland and Pyle,1997
  Nachman and Noe, 1994 Myers and Majluf, 1984
  Demarzo and Duffie, 1999 Demarzo, 2005)
• Informed agent should design a security
  representing a fractional claim on the asset
  retain residual claim as a credible but costly
  signal of true asset quality.
• Assume assets are divisible.
• Focus on equilibria for interior solutions, where
  fractional asset sales are greater than zero or
  less than one.

5
Problem for Applications to Securitized Asset
Markets

• Asset divisibility is at odds with legal
  restrictions on securitized asset markets asset
  sales to SPVs must be bankruptcy remote from
  asset sellers.
• Bankruptcy remoteness requires true sales (FAS
  140)
• Future claims back to asset sellers must be
  legally extinguished.
• Implications
• Assets sold to SPVs are not divisible.
• Corner solution equilibria are relevant where
  the fraction of assets sold must be one or zero.

6
Demarzo (2005) Signaling Model

• Assumes
• Assets are infinitely divisible share q is sold,
  where
• Originator is informed and faces a discount rate,
  d , that exceeds the market rate, 1 gt 1/(1r) gt
  d
• Payoff on asset i
• Wi is private information and Zi is idiosyncratic
  risk, EZi Wi 0.
• Worst case outcome for Wi given by wi0.
• Expected payoff to the informed intermediary

7
Demarzo (2005) Signaling Model

• Suppose originator anticipates a market demand
  curve given by P(q)
• Rational inference on the part of investors leads
  to an equilibrium demand function that is
  downward sloping a higher quantity is expected
  to be sold when the outcome is poor.
• Originator chooses a quantity to sell to maximize
  profit
• As long as wi0 gt 0, a unique separating
  equilibrium exists where the market clearing
  price is

8
But Securitized Assets are Indivisible

• Bankruptcy remoteness implies asset
  indivisibility.
• Asset indivisibility implies that assets are
  either entirely sold to an SPV (q1), or entirely
  held on the originators balance sheet (q0).
• Thus, the equilibrium price is given by the
  worst-case outcome
• We are back to the Akerlof prediction assets
  sold to SPVs are lemons.

9
Why this is an important issue?

• Issuers of mortgage-backed securities (MBS) and
  other asset-backed securities (ABS) account for a
  rapidly growing share of the credit market assets
  held in the financial sector.
• At the end of 1980Q1, these special-purpose
  vehicles (SPVs) accounted for just 3 of credit
  market assets held in the financial sector.
• By the end of 2005Q3, their share had risen to
  22.
• ABS issuers have been growing particularly
  rapidly since about 1995.
• Are asset sales to SPVs a market response to the
  problem of providing liquidity to originators of
  indivisible assets of low quality?

10
U.S. Credit Market Asset Holdings
Source Federal Reserve Board, Flow of Funds
Accounts of the United States, Table L.1, Credit
Market Debt Outstanding, Lines 35, 54, 55.
11
Why focus on Freddie Mac PCs and REMIC?

• In general, it is difficult to test theories of
  markets for lemons because the causative factor
  is asymmetric information.
• Thus, cannot observe the key quality variable
  that determines price.
• Advantages of securitized residential mortgage
  market
• Private information held by mortgage originator
  is unobservable ex ante
• Private information related to borrower choices
  on mortgage menu points (horizon) and
  transactions costs.
• Borrower option exercise behavior is revealed ex
  post.
• Freddie Mac monitoring generates
• Detailed performance data
• Means to track whether a given PC pool is sold to
  REMIC

12
Value of Mortgage Liability and Option Exercise
Efficiency

• The mortgage liability is the value of the
  underlying bond less the value of the joint
  termination options
• Prepayment is optimal if
• Ft remaining principal balance
• Xp prepayment transaction costs
• Default is optimal if
• Ht house prices,
• Xd default transaction costs

13
Testable Model Predictions

• Is option exercise behavior more efficient for
  REMIC PCs than for PCs not sold to REMIC?
• Are REMIC PCs lemons?
• Does the fair market value of REMIC PCs reflect
  these relative efficiencies?
• Are these statistically and economically
  important differences?

14
PC Sales to REMIC SPVs

• Step 1
• Mortgage originator devises a contract menu to
  screen for credit quality and prepayment
  efficiency.
• Step 2
• Originator securitizes whole mortgages for
  Freddie Mac PCs with same collateral lender
  selects lowest quality subject to the GSE-imposed
  limit.
• Step 3
• Originator selects whole PC pools for sale
  through the TBA market.
• Are indivisible sales (we verified in data)
• Prepayment quality known to originators but not
  to forward contract investor
• TBA is anonymous market repeated game is
  irrelevant.
• Step 4
• PCs sold to REMIC by TBA investors (Investment
  and commercial banks)
• Key prediction PCs sold to REMIC SPVs will be
  lemons relative to PCs not sold to REMIC (this we
  observe).

15
The Freddie Mac PC Time-line
Regressions
Mortgage Origination
PC Origination
TBA Delivery
REMIC Origination
Median 2 mths IQ Range 2 mths
Median 1 month IQ Range 2 mths
16
Testing the Prediction

• Consider whether the REMIC status of a PC pool
  affects terminations.
• Data
• Track all Freddie Mac PC pools originated between
  1991-2002 have pool-specific performance and
  origination data (approx. 70,000 pools).
• Know identity of PCs sold to REMIC SPVs.
• Interact REMIC PC indicator with measures of
  interest rate movements and other controls.

17
Approximate path of interest rate changes

• Summed Treasury Deviations variable cumulates
  monthly deviations in 10-year Treasury rates from
  rate at end of 3rd month after pool origination.

Summed Treasury Deviations variable sums these
two areas in this case.
Interest Rate
R(3)R(T)
Time
T
18
Empirical Results
Dependent Variable is the cumulative unscheduled
terminations starting from PC origination date
(Standard errors in parentheses)
Other variables not shown Coupon, quarterly
time dummies, originator dummies, seasoning
dummy, cumulative unscheduled terminations from
mortgage origination to PC origination,
cumulative house price changes.
19
Does PC REMIC Status matter for Pricing A
Structural Analysis

• Solve a structural two-factor valuation model
  (Downing, Stanton, and Wallace (2005)) with
  embedded option exercise policies for the
  borrowers prepayment and default options.
• Controls for term structure, house price
  dynamics, transaction costs, heterogeneity.
• Condition the borrowers optimal exercise
  policies using an empirical optimization scheme
  to obtain for REMIC and nonREMIC PCs
• Relative background hazard rates,
• Relative seasoning component for prepayment and
  default,
• Relative distributions for transaction costs
• Strike prices for prepayment and default options
  are scaled by a transaction cost distribution.

20
Relative Prepayment Transaction Costs, Xip

• Assume
• Borrowers are heterogeneous different borrowers
  face different transaction costs.
• Prepayment optimal when
• Assume Xip is drawn from a beta distribution,
  R REMIC indicator

21
Relative Hazard Functions

• nonREMIC PCs
• DPt dummy variable that is one when the
  prepayment option is in-the-money, zero
  otherwise.
• DDt dummy variable that is one when the default
  option is in-the-money, zero otherwise.
• ß1, background hazard,
• ß2,ß6 seasoning component when Dt and Pt are
  optimal.
• REMIC PCs
• ß3 coefficient on indicator variable, R1, for
  REMIC Pools

22
Fixed Default Transaction Cost, XiD

• Default optimal when
• Assumed to be fixed at 5 of house value, H(t).
• Needed for computational tractability
• Defaults are rare events of lt 1.

23
Summary of Structural Model Results REMIC PC
More Efficient

• Background termination rates .1 of pool balance
  per month.
• Seasoning effects on prepayment REMIC PC more
  efficient.
• 1 Year 5 Years
  10 Years
• REMIC 6.2 6.9
  7.0
• Non-REMIC 6.3 6.8
  6.8
• Transaction cost differentials
• Mean
  Variance
• REMIC 14.12
  1.1
• Non-REMIC 16.39
  1.3

24
Relative Pricing Results REMIC PCs Lemon
Discount

• Do market prices for PCs reflect the systematic
  differences between REMIC and Non-REMIC pools?
• In other words, do PCs destined for REMIC pools
  carry a discount relative to other PCs?
• We examine the average speed difference between
  REMIC and non-REMIC pools over the period.
• Use coupon and 10-Year Rate matched REMIC and
  NonREMIC PCs
• Valuation discount for REMIC PCs
• 0.39/100 discount, on average.
• About 3-5bp in terms of YTM.

25
Conclusions

• Theoretical prediction that SPVs finance lemons
  is not rejected by analysis of REMIC PC
  performance.
• Reduced form models of relative termination
  speeds.
• Structural model estimates of prepayment option
  efficiency.
• Lemon-ness of REMIC PCs produces significant
  price differentials according to our structural
  model.
• But exactly how efficiently REMIC lemons are
  priced in the capital market remains an open
  question.
• Next step Examination of link between SPV
  financial structure and asset quality.