Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
MotivationEmpirical Design
Data and Main Empirical Results
Did Securitization Lead to Lax Screening?Evidence from Subprime Loans
Vikrant Vigwith
Benjamin Keys, Tanmoy Mukherjee and Amit Seru
May 14, 2008
Securitization and Screening 1
MotivationEmpirical Design
Data and Main Empirical Results
Motivation
Framework
I Banks serve as delegated monitors
• Remove duplication in monitoring: Diamond [1984]
I However, must be given incentives to do so• Who monitors the monitors?: Holmstrom and Tirole [1997]
− Illiquidity of loans provides incentives: Diamond and Rajan [2003]
Securitization: Some Facts
I Explosive growth in the last decade or so
• Involves converting illiquid assets to liquid securities
I Changes the business model of financial intermediaries• “risk warehousing” to “originating and distributing”
− “buy and hold” to “buy and sell”
Securitization and Screening 2
MotivationEmpirical Design
Data and Main Empirical Results
Motivation
Several Benefits...
I Improved risk sharing in the economy
• Lower cost of capital
I Banks better at withstanding negative shocks
• Kashyap and Stein [2000]; Loutskiana [2006]; Loutskiana and Strahan[2007]
• “...makes banks more flexible and resilient”: Greenspan at ABAC in2004
Securitization and Screening 3
MotivationEmpirical Design
Data and Main Empirical Results
Motivation
...but What About Costs?
I Banks at arm’s length no longer screen and monitor risks theyoriginate
• Parlour and Plantin [2007]
I Classic liquidity vs.incentives tradeoffI Maug (1997), Bhide (1993), Coffee (1991), Aghion et al. (2004)
I View has gained prominence since the outburst of subprime crisis
• “...securitization contributed to bad lending: in the old days, banksthat originated bad loans bore the consequences; in the new world ofsecuritization, the originators could pass the loans onto others”:Stiglitz [2007]
I Reputation or guarantees from lenders may prevent moral hazard:Gorton and Pennacchi [1995]
I Ultimately an empirical question
• Related to literature on bank sales: Gorton and Pennacchi [1995];Drucker and Puri [2007]
Securitization and Screening 4
MotivationEmpirical Design
Data and Main Empirical Results
Motivation
I Does securitization lead to lax screening by lenders?
I Loans more likely to be securitized default 20% more than similar riskprofile loans with lower likelihood of securitization
Securitization and Screening 5
MotivationEmpirical Design
Data and Main Empirical ResultsStrategy
Identification Strategy
Main Complications
I Endogeneity of securitization makes causal claims difficult
• Use adhoc threshold in lending market to generate exogenousvariation in securitization likelihood of a loan as compared toanother loan with similar risk characteristics
I Conditional on securitization, wide variation possible in loancontracts
• Use detailed data on subprime loans contracts to control forvarious loan characteristics
Securitization and Screening 6
MotivationEmpirical Design
Data and Main Empirical ResultsStrategy
Adhoc Rule Of Lending
Background On Credit Scores (FICO)
I FICO score (350-800) reflects the credit quality of the borrowers
• A scaled probability score with a higher FICO ⇒ borrower withbetter credit quality
• Fair Isaac: “FICO gives ranking of potential borrowers by theprobability of having some negative credit event in the next twoyears”
• Generated via software licensed by Fair Isaac to three independentrepositories (TransUnion, Experian, and Equifax)
I Most reliable measure used by the lender, rating agencies andinvestors: Gramlich [2007]
• High predictability even for low income borrowers• Median score used by lenders
Securitization and Screening 7
MotivationEmpirical Design
Data and Main Empirical ResultsStrategy
Adhoc Rule Of Lending
Threshold of 620 FICO
I Threshold in mid 1990s by Fannie Mae and Freddie Mac intheir guidelines on what loans would be purchased by them
• Fair Isaac: “... those agencies [Fannie Mae and Freddie Mac],have indicated to lenders that any consumer with a FICO scoreabove 620 is good...”
• Guidelines by Freddie Mac: “... a score below 620 is a strongindication that the borrower’s credit reputation is notacceptable...”
I Confirmed in several papers/ rating agency guidelines/articles/ origination matrices of lenders/ anecdotes
Securitization and Screening 8
MotivationEmpirical Design
Data and Main Empirical ResultsStrategy
Identification Strategy
Using adhoc cutoff as a measure of ease of securitization
I Analogous to Fuzzy RD design
• Make causal inferences on lender’s behavior by comparingperformance of loans to borrowers with scores of 619 (620-)vs. 621 (620+)
Securitization and Screening 9
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Large Dataset on Subprime Mortgages
I Loan Performance database: All securities issues in secondarynon-agency market
• Loans in more than 8000 non prime loan pools• Borrower characteristics: Credit score (FICO), debt to income
ratio, documentation (full, limited, no)• Loan characteristics: LTV, loan amount, term and interest rate
type (ARM vs. FRM), type of property (owner occupied, vacation,investor)
I Restrict sample for reasonable comparison
• New purchases of owner-occupied, single family residences• Not FHA or VA insured or Alt-A• Sample period 2001-2006
Securitization and Screening 10
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Overall Market TrendsSummary Statistics
Panel A: Entire Sample
Year Number of % Low Mean MeanLoans Documentation Loan-To-Value FICO
2001 136,483 26.0% 84.6 6112002 162,501 32.8% 85.6 6242003 318,866 38.9% 87.0 6372004 610,753 40.8% 86.6 6392005 793,725 43.4% 86.3 6392006 614,820 44.0% 87.0 636
I Steady growth in number of loans securitizedI ↑ in % low documentation, LTV ratioI Loans with higher credit score securitized
Securitization and Screening 11
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Adhoc Rule in LendingNumber of Loans (in’00) at each FICO score: Low Documentation
05
1015
500 600 700 800fico
2003
I Large jump in number of loans at 620
Securitization and Screening 12
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Adhoc Rule in LendingNumber of Loans (in ’00) at each FICO score: Low Documentation
01
23
4
500 600 700 800fico
2001
02
46
500 600 700 800fico
2002
05
1015
500 600 700 800fico
20030
1020
30
500 600 700 800fico
20040
1020
3040
500 600 700 800fico
2005
010
2030
40500 600 700 800
fico
2006
I Similar trend across years
Securitization and Screening 13
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Adhoc Rule in LendingEstimating Discontinuity in Low Documentation Loans
Yi =(
α + βTi + θf(FICO(i)) + δTi ∗ f(FICO(i)) + εi
)Low Documentation Loans
Year β t-stat Observations R2 Mean2001 36.83 (2.10) 299 0.96 1172002 124.41 (6.31) 299 0.98 1772003 354.75 (8.61) 299 0.98 4132004 737.01 (7.30) 299 0.98 8312005 1,721.64 (11.78) 299 0.99 1,1482006 1,716.49 (6.69) 299 0.97 903
I Large jump at 620+ relative to 620- in number of lowdocumentation loans post 2001
Securitization and Screening 14
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Delinquencies of LoansDelinquencies: Low Documentation
0.0
5.1
.15
500 550 600 650 700 750fico
2003
I Default rates jump around the 620 threshold
Securitization and Screening 15
MotivationEmpirical Design
Data and Main Empirical Results
DataSummary StatisticsNumber of LoansDelinquencies of Loans
Performance of Loans Around Thresholds60+ Delinquency: Low Documentation
0%
2%
4%
6%
8%
10%
12%
14%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Loan Age ( Months)
Del
inqu
ency
(%)
615-619 (620-)620-624 (620+)
I Loans at 620+ default more relative to loans at 620−
I Large magnitudes relative to mean default rates – around 20% more
Securitization and Screening 16
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Alternative Explanation
What about...
I Selection on Observables
• Borrowers• Investors/Issuer
Securitization and Screening 17
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Loan Characteristics Around ThresholdsLoan To Value Ratio: Low Documentation
050
100
500 600 700 800fico
2003
I No jump in loan to value at 620
Securitization and Screening 18
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Loan Characteristics Around ThresholdsLoan To Value Ratio: Low Documentation
050
100
500 600 700 800fico
2001
050
100
500 600 700 800fico
2002
050
100
500 600 700 800fico
2003
050
100
500 600 700 800fico
2004
050
100
500 600 700 800fico
2005
050
100
500 600 700 800fico
2006
Securitization and Screening 19
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Loan Characteristics Around ThresholdsInterest Rates: Low Documentation
510
15
500 600 700 800fico
2003
I No jump in interest rates at 620.
Securitization and Screening 20
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Loan Characteristics Around ThresholdsInterest Rates: Low Documentation
510
15
500 600 700 800fico
2001
510
15
500 600 700 800fico
2002
510
15
500 600 700 800fico
2003
510
15
500 600 700 800fico
2004
510
15
500 600 700 800fico
2005
510
15
500 600 700 800fico
2006
Securitization and Screening 21
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Distribution of Loan Contracts around 620Interest Rates: Low Documentation
0.1
.2.3
.4D
ensi
ty
0 5 10 15initrate
Kernel density estimatekdensity initrate
I No difference in the distributions of interest rates offered at 620+ and 620−
I KS test rejects that the two distributions are not equal at 1%
Securitization and Screening 22
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Distribution of Loan Contracts around 620Loan to Value: Low Documentation
0.0
2.0
4.0
6.0
8D
ensi
ty
20 40 60 80 100origltv
Kernel density estimatekdensity origltv
Securitization and Screening 23
MotivationEmpirical Design
Data and Main Empirical ResultsSelection on Observables
Borrower Demographics Around ThresholdsHousehold Income: Low Documentation
4060
80
500 600 700 800fico
2001
4060
80
500 600 700 800fico
2002
4060
80
500 600 700 800fico
200340
6080
500 600 700 800fico
200440
6080
500 600 700 800fico
2005
4060
80500 600 700 800
fico
2006
I No jump in borrower demographic variables at 620 across years
Securitization and Screening 24
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
Other Tests
What about...
I Manipulation of FICO Scores
• Why manipulate?
I Soft Information
• Do effects attenuate with more hard information?
Securitization and Screening 25
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
A Natural ExperimentNumber of Loans: Low Documentation
010
020
030
040
0
500 600 700 800fico
I Predatory laws passed in Georgia and New Jersey in Oct 2002
I Subsequently reversed Georgia (April 2003) and New Jersey (May 2004)
Securitization and Screening 26
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
Another Adhoc Rule Of Lending
Threshold of 600 FICO
I Threshold appears in advice by Fair Isaac
• Fair Isaac: “...anything below 600 is considered someone whoprobably has credit problems that need to be addressed...”
• Einav, Jenkins and Levin [2007]: “...a FICO score above 600, atypical cut-off for obtaining a standard bank loan”
I Value of soft information is lower for loans that provide fulldocumentation
Securitization and Screening 27
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
Another Adhoc RuleNumber of Loans at each FICO score: Full Documentation
05
1015
20
500 600 700 800fico
2003
I Large jump in number of loans at 600
Securitization and Screening 28
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
Adhoc Rule in LendingNumber of Loans at each FICO score: Full Documentation
05
10
500 600 700 800fico
2001
05
10
500 600 700 800fico
2002
05
1015
20
500 600 700 800fico
20030
1020
3040
500 600 700 800fico
20040
2040
60
500 600 700 800fico
2005
010
2030
4050
500 600 700 800fico
2006
I Large jump in number of loans at 600
Securitization and Screening 29
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
Delinquencies of LoansDelinquencies: Full Documentation
0.0
5.1
.15
.2
500 550 600 650 700 750fico
2001
0.0
5.1
.15
500 550 600 650 700 750fico
2002
0.0
5.1
.15
500 550 600 650 700 750fico
2003
0.0
5.1
.15
500 550 600 650 700 750fico
2004
0.0
5.1
.15
500 550 600 650 700 750fico
2005
.05
.1.1
5.2
.25
500 550 600 650 700 750fico
2006
Securitization and Screening 30
MotivationEmpirical Design
Data and Main Empirical Results
Manipulation of FICO scoreSoft Information
Robustness Checks
Additional Tests
I Variation within:
• Pool• Lenders• States
I Other counterfactual checks
I Other cutoff rules
I Other performance measures (delinquency definitions)
Securitization and Screening 31
MotivationEmpirical Design
Data and Main Empirical Results
Conclusion
I Securitization destroys screening incentives of lenders
• Extrapolation required to assess effects on the entire market
I Cautious on welfare implications of securitization
• Benefits need to be evaluated with these costs
I Implications in general for defaults models and regulationthrough models (BASEL II)
• Default models not invariant to strategic behavior ofparticipants: Lucas [1976]
Securitization and Screening 32
MotivationEmpirical Design
Data and Main Empirical Results
Permutation TestsDelinquencies
010
2030
4050
Den
sity
-.03 -.02 -.01 0 .01 .02beta_01_06
14 points dropped on either side of 500 and 750
I Same pattern for all years
Securitization and Screening 33
MotivationEmpirical Design
Data and Main Empirical Results
Permutation TestsInterest Rate
01
23
45
Den
sity
-.4 -.2 0 .2 .4interest_01_06
I Same pattern for all years
Securitization and Screening 34