Euphoria fueled by rapid house price acceleration may hav

1
Securitization of Subprime MortgagesAn
Empirical Analysis of the Loan Sale Decisions of
Depository Institutions

• Paul Calem and Jonathan Liles
• Freddie Mac
• and
• Christopher Henderson
• Federal Reserve Bank of Philadelphia
• August 20, 2008
• The views expressed are those of the authors and
  do not represent official views of Freddie Mac,
  the Federal Reserve Bank of Philadelphia, or the
  Federal Reserve System
•  

2
Securitization of Subprime Mortgages

• Banks have an incentive to securitize loans to
  reduce regulatory capital requirements, diversify
  funding sources, and limit concentrations of risk
  in the asset portfolio
• The prevailing view is that regulatory capital
  requirements are too high for most mortgage
  loans, giving financial institutions incentive to
  sell less risky loans to investors
• However, agency problems may impede the
  functioning of securitization markets
• Lenders may have private information enabling
  them to sell riskier loans into public markets

3
Securitization of Subprime Mortgages

• This classic lemons problem may have been a
  factor affecting subprime (ABS) securitization
• The lemons problem, exacerbated by overly
  optimistic assessments by investors of the
  outlook for house prices and of credit risk,
  likely contributed to the recent subprime market
  collapse
• It may be a key factor influencing the extent and
  form of future recovery of the subprime market
• Such problems are mitigated in the prime (MBS)
  market due to
• Standardized products
• Lower risk, less heterogeneous borrowers

4
Research Objectives

• Explore the decision of depository institutions
  to sell or retain subprime mortgages during
  2004-2006
• Use HMDA data merged with data from
  LoanPerformance on characteristics of the
  subprime market at the neighborhood (ZIP code)
  level
• Seek evidence on the extent to which regulatory
  arbitrage was a driver of subprime mortgage
  securitization
• Did banking institutions tend to sell
  higher-yield subprime mortgages, contrary to the
  conventional view?
• Seek evidence on euphoria in the ABS market
  fueled by rapid house price appreciation in 2005
  and 2006
• Did loan sale behavior change?

5
Research Objectives

• Seek evidence on information asymmetries in the
  subprime securitization market
• Did banks sell higher risk subprime mortgages
  without fully signaling increased risk through
  higher asset yields?
• Explore the dimensions along which securitization
  transactions costs vary
• Are certain loan categories or types of
  institutions associated with greater likelihood
  of loan sale?

6
Preview of Findings

• During 2004, banks were less likely to sell the
  higher risk subprime loans they originated (those
  with larger APR spread), consistent with
  regulatory capital arbitrage
• Subsequently, we observe a reversal of this
  relationship, consistent with euphoria-driven
  undervaluing of credit risk in ABS markets
• We obtain evidence suggestive of informational
  advantages of depository institutions
• For example, banks were more likely to sell loans
  associated with neighborhoods with high future
  default rates

7
Preview of Findings

• Loan categories associated with higher
  transactions costs, such as very small loan
  sizes, exhibit reduced likelihood of loan sale
• HOEPA loans were comparatively likely to be
  retained

8
Relevant Prior Literature

• Securitization and Agency Problems
• Akerlof (1970)
• Hill (1996)
• Demarzo and Duffie (1999)
• DeMarzo (2005)
• Loan Sales
• Samolyk (1989)
• Drucker and Puri (2007)

9
Relevant Prior Literature

• Regulatory Capital Arbitrage and Mortgage
  Securitization
• Picker (1996)
• Jones (2000)
• Calem and LaCour-Little (2004)
• Ambrose, LaCour-Little, and Sanders (2004)
• Calem and Follain (2007)
• Subprime Mortgage Lending General
• Apgar and Herbert (2006)
• Avery, Brevoort, and Canner (2006)
• Chomsisengphet and Pennington-Cross (2006)
• Smith (2007)

10
Relevant Prior Literature

• Recent Developments in the Subprime Mortgage
  Market
• DiMartino and Duca (2007)
• Doms, Furlong, and Krainer (2007)
• Edmiston and Zalneraitis (2007)
• Federal Reserve Bank of San Francisco (2007)
• Gerardi, Shapiro, and Willen (2007)
• Presidents Working Group on Financial Markets
  (2008)

11
Relevant Prior Literature

• Subprime Mortgage Lending Agency and Information
  Problems
• Alexander, Grimshaw, McQueen, and Slade (2002)
• Courchane, Surette, and Zorn (2004)
• Crews-Cutts and Van Order (2005)
• An and Bostic (2006)
• Golding, Green, and McManus (2008)
• Securitization of Subprime Mortgages
• DiMartino, Duca, and Rosenblum (2007)
• Ashcraft and Schuermann (2008)
• Keys, Mukherjee, Seru, and Vig (2008)
• Mian and Sufi (2008)

12
Economic Factors Influencing the Loan
Sale/Securitization Decision

• Regulatory capital
• Funding and liquidity
• Informational advantages of depository
  institutions
• Transactions costs
• Risk diversification
• Predatory lending risk

13
Regulatory Capital Arbitrage

• Required regulatory capital exceeds economic
  capital for the credit risk of prime mortgages
• This creates an incentive to retain higher risk,
  subprime mortgages
• Regulatory supervision may bring some restraint
  to this incentive

14
Funding and Liquidity

• A liquid securities market is a dependable source
  of funds for loan originations
• Euphoria among investors may bring too much
  liquidity into the market
• Particular risks may be unrecognized or
  undervalued by the market, generating increased
  loan sales by banks
• Euphoria fueled by rapid house price acceleration
  may have contributed to growth in subprime
  mortgage securitization during 2005-2006

15
Informational Advantages of Depository
Institutions

• Ties to local markets and relationships with
  borrowers
• Community banks are attentive to neighborhood
  conditions that could influence mortgage credit
  risk
• Even large, centralized organizations may develop
  such ties
• They monitor the profitability of branch networks
  and correspondent relationships
• They engage in community reinvestment activities
• More efficient use of information that is
  potentially also available to the securities
  market
• Banks may incorporate more local and regional
  economic data into their decision processes than
  aggregators/asset managers/investors

16
Informational Advantages of Depository
Institutions

• Superior controls around information processes
  due to better aligned incentives and regulatory
  supervision
• Model risk and other operational risks may
  receive greater attention from banks
• Banks may more consistently apply overrides to
  models in fast-growing segments for which the
  models may not be representative
• Hybrid ARM loans with initial, teaser margins
• Non-traditional products
• Loans with piggyback second liens
• ABS investors generally rely on rating agencies
  or other third parties, who may face conflicting
  incentives

17
Transactions Costs

• Transactions costs include due diligence and
  legal expenses costs associated with structuring
  and negotiating contract terms and costs
  associated with sale of the security
• Fixed costs per loan included in an issue and
  fixed costs per issue imply higher transactions
  costs for smaller loans and smaller institutions

18
Other Factors Influencing the Loan
Sale/Securitization Decision

• Risk diversification
• Banking institutions, particularly smaller
  institutions, may sell loans to reduce
  concentrations of credit risk in the asset
  portfolio
• Banks may sell loans with particular duration
  characteristics as an interest rate risk
  management strategy
• Predatory lending risk
• Investors may be reluctant to touch HOEPA loans
  or other loans with very high interest rates or
  fees due to legal risks

19
Data and Empirical Approach

• We analyze the disposition (retention vs. sale)
  of subprime loans originated during 2004-2006,
  using loan-level HMDA data
• The population is limited to loans originated by
  commercial banks, savings banks, and credit
  unions (depository institutions)
• Loans sold to an affiliated institution are
  excluded
• The population is limited to first lien, home
  purchase or refinance mortgages (amounts lt
  1,000,000)
• Subprime loans are defined to be high cost
  (having a HMDA-reported APR spread), or
  originated or purchased by a HUD-identified
  subprime specialist

20
Data and Empirical Approach

• These data are merged at the tract-level with
  housing and mortgage market variables from
  several sources
• Aggregated tract-level characteristics from HMDA,
  such as percent of loans that are high cost
• Aggregate ZIP code level subprime market
  characteristics from LoanPerformance, such as
  percent of loans that are high LTV
• The data are transformed to tract-level by
  weighting proportionately across ZIP codes that
  overlap each Census tract
• Data are from the LoanPerformance Servicing (not
  Securities) database
• MSA house price appreciation (OFHEO index), and
  MSA housing starts (from economy.com)
• The number of owner-occupied units in the Census
  tract, from the 2000 U.S. Census

21
Data and Empirical Approach

• We analyze the disposition of the loan in
  relation to the size of the APR spread and
  proxies for various dimensions of credit risk,
  along with a number of control variables
• The APR spread provides a direct test of
  regulatory capital arbitrage
• Regulatory capital arbitrage implies an inverse
  relationship between the APR spread (priced risk)
  and likelihood of sale
• A reversal of this relationship during 2005-2006
  would suggest an impact of euphoria in the
  securitization market

22
Data and Empirical Approach

• Along particular dimensions where banks have
  informational advantages, we expect likelihood of
  sale to increase with risk
• Variables used to assess bank informational
  advantages include
• Loan for non-primary residence
• Ratio of loan amount to borrower income
• Local area house price appreciation rate

23
Data and Empirical Approach

• Model 1 we test for bank informational
  advantages using a number of local mortgage and
  housing market variables
• Percent of loans in the Census tract that are
  subprime
• Percent of loans in the Census tract that are
  non-primary residence (investment property or
  second home)
• Percent of the Census tracts subprime loans that
  are originated by subprime specialists
• Housing market depth (log number of tract owner
  occupied units)
• Local area percent change in housing starts
• Model 2 we test for informational advantages
  using the the future (third quarter 2007) default
  rate in the Census tract (in place of the above)

24
Data and Empirical Approach

• We distinguish between home purchase and
  refinance
• Control variables include
• Loan size
• Institution size
• Type of depository institution
• Thrift vs. commercial bank
• HUD-identified subprime specialist
• Metropolitan area dummy variable
• Percent of tract subprime loans that are junior
  lien
• Percent of tract subprime loans that are high LTV
  (LTV ? 90)

25
Overview of the Data Sample Size
26
Overview of the Data Disposition of Home
Purchase Loans
27
Overview of the Data Disposition of Refinance
Loans
28
Overview of the Data Structure of the Loan
Origination Market
29
Overview of the Data Structure of the Loan
Origination Market
30
Overview of the Data Structure of the Loan
Origination Market
31
Overview of the Data APR Spreads
32
Market Context Interest Rate Trends

• Source Federal Reserve Bank of St. Louis

33
Overview of the Data APR Spreads
34
Overview of the Data Product Type
35
Overview of the Data Documentation Level, LTV,
and FICO
36
Market Context Piggyback Lending Trends

• Source HMDA data with matched first/second liens
  obtained from Federal Reserve Board

37
Overview of the Data Ratio of Loan Amount to
Income
38
Overview of the Data Average Loan Size
39
Overview of the Data Investor Loans (non-primary
residence)
40
Overview of the Data Housing Market Conditions
41
Empirical Results

• Model 1 tract-level variables proxy for
  localized informational advantages
• Model 2 future default rate in tract proxies for
  localized informational advantages
• Estimated separately for home purchase and
  refinance in 2004, 2005, and 2006

42
(No Transcript)
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
Empirical Results Shifting Relationship to APR
Spread

• Likelihood of sale declines with APR spread in
  2004, consistent with regulatory capital
  arbitrage
• Likelihood of sale generally increases with APR
  spread in 2005 and 2006, consistent with
  undervaluing of credit risk in ABS markets

47
Empirical Results Shifting Relationship to APR
Spread
48
Empirical Results Bank Informational Advantages

• Likelihood of sale is inversely related to local
  area house price appreciation and percent change
  in housing starts
• Suggests that banks may incorporate more local
  and regional economic data into their decision
  processes than aggregators/asset
  managers/investors
• Home purchase in 2006 is exception

49
Empirical Results Bank Informational Advantages

• Loans for non-primary residence are less likely
  to be sold
• Consistent with relationship lending or with
  specialized information regarding local
  investment opportunities
• Refinance in 2006 is exception
• Likelihood of sale increases with ratio of loan
  amount to borrower income
• Consistent with more cautious assessment of debt
  payment capacity (model overrides)

50
Empirical Results Bank Informational Advantages

• Likelihood of sale increases with
• Percent of tract loans that are subprime
• Percent of tract loans that are non-primary
  residence
• Percent of tract subprime loans that are
  originated by subprime specialists
• Consistent with asymmetric information concerning
  neighborhood risk factors
• Alternatively, to the extent that borrowers with
  similar risk profiles tend to cluster
  geographically, may reflect asymmetric
  information concerning borrower-specific risks

51
Empirical Results Bank Informational Advantages

• For home purchase loans, likelihood of sale is
  inversely related to number of owner occupied
  units in neighborhood
• Consistent with superior information regarding
  appraisal accuracy
• In model 2, a higher neighborhood delinquency
  rate ex-post (in Q3 2007) is associated with
  increased likelihood of sale

52
Other Empirical Results

• Likelihood of sale increases with percent of
  tract subprime loans that are junior lien
• Likelihood of sale increases with percent of
  tract subprime loans that are high LTV (only for
  home purchase loans)
• Likelihood of sale reduced for HOEPA loans and
  those in the highest APR spread category
• Consistent with legal risks

53
Other Empirical Results

• Large home purchase loans are less likely to be
  sold
• Very small loans are less likely to be sold
• Fixed transactions costs per securitization
• Very small institutions tend not to sell
• Fixed costs associated with access to securities
  market or other loan sale channels
• Loans outside metro areas less likely to be sold

54
Other Empirical Results

• Purchase and refinance loans exhibit somewhat
  different patterns of loan disposition in
  relation to type and size of institution
• May reflect differing diversification
  requirements differing roles of relationship
  lending

55
Robustness

• Results (models 1 and 2) are robust to
• Adding dummy variables for Census Divisions
• Including observations with missing income
  (setting the loan-to-income ratio0 and using a
  dummy variable)
• Excluding loans sold to other depository
  institutions
• Any of these changes has little impact on sign,
  magnitude, or statistical significance of other
  variables

56
Robustness

• Model 1 is robust to inclusion of additional
  ZIP-code level measures of subprime market
  composition
• Product mix
• Percent of tract subprime loans that are to lower
  FICO borrowers (FICO score lt 620)
• Including these has little impact on sign,
  magnitude, or statistical significance of other
  variables
• No consistent relationship to loan disposition
  observed for any of these

57
Robustness

• Model 1 is robust to excluding the measures of
  subprime market composition from LoanPerformance
  data
• Using either the full HMDA sample before merging
  with LoanPerformance data, or the smaller merged
  sample
• Model 2 is robust to excluding the tract-level
  control variables

58
Conclusions

• We obtain evidence suggestive of euphoria-driven
  undervaluing of credit risk in ABS markets during
  2005-06
• In 2004, depository institutions were less likely
  to sell subprime loans with larger APR spreads,
  consistent with regulatory capital arbitrage
• In 2005-06, we observe a reversal of this
  relationship
• We also obtain evidence suggestive of
  informational advantages of depository
  institutions
• Likelihood of sale increases with ratio of loan
  amount to borrower income
• Likelihood of sale related to a variety of
  neighborhood risk factors

59
Conclusions

• Some caveats apply
• Relationship between loan disposition and APR
  spread may have been affected by shifting yield
  curve
• We cannot observe whether banks offered greater
  credit enhancements on riskier loans