Introduction Data Predicting Securitization and Loan Performance Model Conclusions Advantages and Disadvantages of Securitization: Evidence from Commercial Mortgages Andra C. Ghent (ASU) and Rossen Valkanov (UCSD) BEROC Third Annual International Economics Conference December 18th, 2012
33
Embed
Advantages and Disadvantages of Securitization: … Ghent.pdfIntroductionData Predicting Securitization and Loan PerformanceModelConclusions Advantages and Disadvantages of Securitization:
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Advantages and Disadvantages ofSecuritization: Evidence from Commercial
Mortgages
Andra C. Ghent (ASU)and
Rossen Valkanov (UCSD)
BEROCThird Annual International Economics Conference
December 18th, 2012
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Securitization
Securitization: The process of creating securities from other assets(usually loans)
Example: Create a security that consists of claims to the cashflows of 200 underlying residential mortgages
• Mortgages can be originated by different institutions
• A sponsor buys the loans from the originators and then issuesa mortgage-backed security (MBS)
• MBS can then be sold to various investors (banks, insurers,hedge funds, pensions, etc...)
• MBS holder receives cash flows from all the mortgages in thepool
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Prior to 2008-2009 financial crisis, securitization
• usual structure for financing residential mortgages in US
• common for commercial mortgages, bank loans, auto loans,student loans, and credit card receivables
• in 2007, approximately 25% of US non-real estate consumerdebt was securitized
Although securitization is far from new (widespread securitizationdates from 1850s in US), it has come under scrutiny since thefinancial crisis
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
MotivationConcerns regarding securitization:
1. Regulatory abritrage: e.g., Acharya, Schnabl, and Suarez(forthcoming)
2. Monitoring and renegotiating securitized loans may be moredifficult
3. How to deal with “securitized-banking” runs? Gorton andMetrick (2012)
4. Potential for adverse selection / reduced sceening• possibility that securitized loans are of lower quality based on
unobservables or• possibility that originators of securitized loans do not acquire
‘soft’ information on loan quality
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Motivation
Rulemaking pursuant to Dodd-Frank Act deals with the concernregarding adverse selection by forcing originators to retain some ofthe risk of any securitized loan on balance sheet (“skin in thegame”)
• effective January 2013
• “qualified residential mortgages” exempt
• Dodd-Frank required originators to retain 5% of the risk ofsecuritized loans
• most likely, implementation will be vertical retention of risk(i.e., retain 5% of every tranche)
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
This Paper
Compare securitized and balance sheet commercial mortgages tosee which loans are more likely to be securitized
First paper with dataset that contains both securitized and balancesheet commercial mortgages from multiple lenders
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
This PaperAdvantages of commercial mortgage market
• Most research focused on residential mortgage market wherepresence of GSEs makes it difficult to extrapolate findings toother asset classes
• No requirement that CMBS issuers or loan sellers have anyexplicit “skin in the game” unlike in Collateralized LoanObligation (CLO) market (Benmelch, Dlugosz, and Ivashina,2012)
• Unique conduit model in commercial mortgage market:securitization status determined at origination rather thanafter origination
• Large loans with high property-specific (i.e., idiosyncratic) riskwhich may make it easier to detect risk sharing motives
This Paper: Findings
1. Main difference between CMBS and balance sheet loans issize
• CMBS loans are much larger than balance sheet loans• diversification being a key motivation for securitization
2. Loans that require more monitoring are less likely to besecuritized
3. Overall, CMBS loans do not perform worse than balance sheetloans after controlling for observables
• no difference in likelihood of default or recovery rates
4. Defaulted CMBS loans slower to get resolved in the sense ofproperty being disposed of by lender
• agency problems with servicing CMBS mortgages
5. Some evidence consistent with adverse selection in subsamples
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Data• Data from Real Capital Analytics (RCA)
• Single-property purchase mortgages on office buildings inBoston, LA, Las Vegas, and NYC metro areas
• culled from property sales• originated Jan. 2005 - Apr. 2012• cities account for about 35% of US office property market• securitization status at origination
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Predicting Securitization
• Estimate probit model of securitization
• Dependent variable takes a value of 1 if loan is securitized, 0otherwise
• Independent variables:• log loan amount• proxy for development / redevelopment loan (LTV≥ 1)• Loan to Value (LTV)• price per square foot• Central Business District (CBD) dummy• multi-building dummy• dummy for depository US institution• year of origination dummies (2005, 2006, and 2007)• MSA dummies• property age dummies• borrower type dummies
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Probit Estimation of SecuritizationDependent Variable is CMBS (=1 if Securitized)
Orig. Yr. FEs Yes Yes Yes Yes YesMSA FEs Yes Yes Yes Yes YesProp. Age FEs Yes Yes Yes Yes YesBorrower Types Yes Yes Yes Yes YesOriginator FEs No No No No YesNumber of Obs. 1962 1962 825 825 514Pseudo-R2 33% 35% 41% 30% 44%
1 2 3 4 5 6 7 8 9 100
0.1
0.2
0.3
0.4
0.5
Loan Size Decile
Pro
ba
bili
ty
Panel A: Probability of Securitization by Loan Size Decile
1 2* 3** 4*** 5*** 6*** 7*** 8*** 9*** 10***−2
−1
0
1
2
3
Loan Size Decile
Co
eff
icie
nts
Panel B: Probit Coefficients on Loan Size Decile
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Predicting Securitization: SummaryDifferences:
• large loans are much more likely to be securitized• robust to including measures of size of institution• effect persists within lender
• development loans are less likely to be securitized
• loans originated depository institutions more likely to besecuritized
• loans originated by large institutions more likely to besecuritized
Sensitivity Analyses:
• different measures of size
• only 2005-2007 originations
• logit instead of probit
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Predicting Default
• Estimate probit model of default
• Dependent variable takes a value of 1 if loan defaults, 0otherwise
• Independent variables:• indicator for CMBS• log loan amount• development loan proxy• LTV• price per square foot• CBD dummy• year of origination dummies (2005, 2006, and 2007)• MSA dummies• property age dummies• borrower type dummies
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Probit Estimation of DefaultDependent Variable is Default (=1 if Default)
Orig. Yr. FEs Yes Yes Yes Yes YesMSA FEs Yes Yes Yes Yes YesProp. Age FEs Yes Yes Yes Yes YesBorrower Types Yes Yes Yes Yes YesOriginator FEs No No No No YesNumber of Obs. 1962 1962 1066 825 514Pseudo-R2 11% 11% 13% 11% 16%
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Predicting Default: Summary of Full Sample Results
• Overall, CMBS loans do not default more than balance sheetloans
• Loans from depository institutions default less
• Loans from larger institutions are less likely to default
• Loans on newer properties more likely to default
• Loans on property in Las Vegas more likely to default
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Predicting Default: Subsample Results
• Within set of higher quality of originators (depositoryinstitutions) where we can control for originator size, someevidence that CMBS loans default more
• Within set of large originators with significant balance sheetand CMBS programs, and for which we can control fororiginator fixed effects, some evidence that CMBS loansdefault more
• Suggests there is substantial heterogeneity in quality of loansfrom various originators
• Many low quality loans held on balance sheet bynon-depository originators may mask differences in defaultbetween CMBS and balance sheet loans in full sample
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Predicting Default: Sensitivity Analysis
• Potential collinearity between size and CMBS:• estimate only with large loans (above median size) omitting
size as control• estimate with size in only four categories• drop size outliers
• Other Sensitivity Analyses:• estimate using only 2005-2007 originations• exclude development loans• proportional hazards model using only information known at
origination• proportional hazards model using characteristics known at
origination and dynamic information on MSA level employmentand office property prices
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Resolution of Defaulted Loans
• Resolved loans: Loans for which balance sheet lender / CMBStrust has disposed of asset
• Recovery rate on resolved CMBS loans: 72%
• Recovery rate on resolved balance sheet loans: 70%
• By end of sample, 55% of defaulted balance sheet loans hadbeen resolved
• By end of sample, only 31% of defaulted CMBS loans hadbeen resolved
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Probit Estimation of ResolutionDependent Variable is Resolved (=1 if Resolved)
Orig. Yr. FEs (2005, 2006, and 2007) Yes No YesMSA FEs Yes Yes YesDefault Yr FEs (2008, 2009, 2010) No Yes YesNumber of Obs. 175 175 175Pseudo-R2 8.3% 14.2% 14.6%
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Disentangling Adverse Selection from Causal Effects ofSecuritization
We observe only default, not distress
Some literature (Gan and Mayer, 2007; Ambrose, Sanders, andYavas, 2009) suggests securitization impedes renegotiation ofdistressed commercial mortgages
How do we interpret the lack of differences in default in terms of amodel of adverse selection based on risk?
Although securitization status usually determined at origination, doloans end up in the securitized market because they have weakersoft information characteristics?
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Simple Model to Disentangle Adverse Selection
Y ∗i ,t = xi ,t−1β + γεDISTRESSi ,t + εCMBS
i ,t . (1)
• Y ∗i ,t is latent variable for securitization (securitized loan ifY ∗i ,t > 0)
• εDISTRESSi ,t and εCMBS
i ,t are orthogonal to each other
• εDISTRESSi ,t represents unobservables that affect the likelihood
that a loan will be in distress and at risk of default
• εDISTRESSi ,t captures soft information
• εDISTRESSi ,t is not directly observable from the data
• γ > 0 indicates adverse selection
Introduction Data Predicting Securitization and Loan Performance Model Conclusions
Distress (not directly observable)
Z ∗i ,t+1 = x̃i ,t−1η + εDISTRESSi ,t+1 (2)
• note that securitization does not affect whether loan getsdistressed
εDISTRESSi ,t+1 = εDISTRESS
i ,t + vi ,t+1 (3)
• vi ,t+1 is i.i.d. zero-mean random variable
Introduction Data Predicting Securitization and Loan Performance Model Conclusions