Structured Debt Ratings: Evidence on Conflicts of Interest 1 Matthias Efing University of Geneva and SFI Harald Hau University of Geneva and SFI http:// www.haraldhau.com
Feb 25, 2016
1
Structured Debt Ratings: Evidence on Conflicts of Interest
Matthias EfingUniversity of Geneva and SFI
Harald HauUniversity of Geneva and
SFIhttp://www.haraldhau.com
2
Did CRAs grant rating favors to issuers in which they had a large business interest?
US Justice Department files lawsuit against S&P S&P memorandum, July 1, 2004:
Recognizes role of Client Value Managers in “criteria discussion” consideration of "market perspective" and "rating implications“ polling of "three to five investors in the product" and
"an appropriate number of issuers and investment bankers”
Empirical evidence for conflict of interest? Is the poor rating performance in 2007–08 the result of
market considerations that distort rating processes? Simultaneous downgrades of thousands of structured securities in 2007/08 ABX index of AAA-rated MBS dropped by 70% between Jan ‘07 and Dec ‘08
Research Question
3
Literature
Theoretical literature: Strong bargaining power of issuers due to “issuer pays model”
(Pagano & Volpin, 2010; White, 2010) Rated firm can “shop for better ratings”
(e.g. Skreta & Veldkamp, 2009; Faure-Grimaud et al., 2009) CRAs might respond to lobbying for rating favors to attract / maintain
rating business (Bolton et al., 2012; Mathis et al., 2009)
Rating contingent regulation creates incentives to sell regulatory relief in the form of rating favors (Efing, 2012; Opp et al., 2013)
4
Literature
Empirical literature: Rating favors in corporate bank ratings
(Hau, Langfield, & Marquez 2013) Investors require higher yields for MBS sold by large issuers
(He, Qian, & Strahan, 2011, 2012) Subjective rating adjustments, and competitive pressure
(Griffin & Tang, 2012; Griffin, Nickerson, & Tang, 2013) Launch vs. follow-up ratings
(Griffin & Tang 2011) Issuer-pays vs. investor-pays CRA
(Cornaggia & Cornaggia, 2013), Revolving Doors on Wall Street
(Cornaggia, Cornaggia, & Xian, 2013)
5
Data – Structured Debt by Deal Type
6
Data – Structured Debt by Origin of Collateral
7
Data – Boom-Bust Pattern of Structured Debt
8
Collateral Pool
AAA
AAA
AA
BEquity
Complexity of Deal Structures
Credit risk allocated to deal tranches according to seniority
Cash flow cascade further refined(triggers regulating amortization pro rata vs. in order of seniority; decoupling of principle and interest payments; etc.)
Varying tranche access to liquidity reserves or debt insurance; etc.
Risk allocation to deal tranches intractable for large samples with different asset/collateral types
Conduct deal-level analysis
9
Deal-Level Measures of Credit Risk
Ignore complex intra-deal allocation of credit risk Measure credit risk at deal level:
Collateral quality:90plus delinquency rate 9 months after deal closure
Credit enhancement: overcollateralization, debt guarantees, liquidity reserves
Deal complexity:number of tranches
Moody’s PDS: 620 EMEA & 52 North-American deals with 6,514 tranche ratings.
10
Methodology – Rating Implied Spreads
11
Estimation of RIS (Rating-Implied Spreads)
DCM Analytics and Bloomberg 10,625 floaters issued at par
with Euribor/Libor as base rate
Rating dummies explain 48% of variation in launch spreads
Dummies for unrated tranches
Fixed effects and time-interact. for collateral origin, asset type, currency and issuance half-year
Controls for liquidity, maturity and term structure at issuance
12
Methodology – Deal Level Aggregation of RIS
13
Aggregation of RIS to Deal Rating-Implied Spread
Correlation: 0.55
14
Methodology – Determinants of deal ratings
15
Hypothesis
H1: Conflicts of Interest and Ratings InflationIssuers who generate more rating business (high ASSB) (i) receive better ratings and (ii) benefit from lower rating-implied spreads
H1: Evidence from Subordination Levels
16
17
H1: Evidence from Deal Rating Implied Spreads
1,404 deal-CRA pairs
Robust std. errors clustered by deal as well as by issuer
One std. dev. of Log ASSB (2.24)
=> DRIS = -7 bp
for avg. deal with DRIS = 12 bp.
H1: Evidence from Deal Rating Implied Spreads
19
Hypothesis
H2: Rating Favors by Deal Quality and ComplexityRating favors are concentrated in those deals for which they are most profitable to issuers and CRAs. Deals of low quality benefit from larger rating favors.
(more profitable than rating favors on already high ratings) More complex deals benefit from larger rating favors.
(rating precision more expensive; external quality verification more difficult)
20
H2: Quantile Regressions
21
H2: Quantile Regressions
MBS (ABS) account for 57% (43%) of observations with DRIS beyond Q90.
… 22
H2: Deal Complexity
23
Hypothesis
H3: Conflicts of Interest over the Credit CycleRating favors are more pronounced during credit booms.
(lower default probabilities & reputational costs; best analysts work for banks rather than for CRAs)
… 24
H3: Conflicts of Interest over the Credit Cycle
25
Hypothesis
H4: Ratings Shopping over the Credit Cycle During credit booms risk aversion and perceived
asymmetric information are low.
Issuers suppress bad ratings so that deals rated by only one CRA have on average better ratings.
In normal times issuers publish multiple ratings to mitigate adverse selection.
Only very risky deals with on average worse ratings are rated by just one CRA.
… 26
H4: Ratings Shopping over the Credit Cycle
27
Reverse Causality
CRA errs in Bad Faith: CRA caters rating favors to secure future business.
CRA errs in Good Faith: CRA would have to err repeatedly in the same direction
across the (many) and potentially heterogeneous deals of an issuer to accumulate a large ASSB.
But Good-Faith errors should be independent.
Favorable Rating Error (d,a)
Rating Mandate (d,a)
Repeated Favorable
Rating Error
Accumulate Large ASSB
28
Causality Issues & Structure of Rating Errors
… 29
Robustness: CRA fixed effects & interactions
…
30
Robustness: Rating Favors across Asset Types
Ca. 1.5% of avg. deal unsecuritized
Base line regress.:Weight unsec. part of deal with dummy for Unrated Junior
Column (1):Weight unsec. deal part with avg. RIS
Column (2):Weight unsec. deal part with RIS(Junk)
31
Robustness: Alternative DRIS Models
32
Robustness: Rating Favors Priced Into Yield Spreads
Yield spreads might contain a premium for the risk that rating of security is inflated.
Estimate new spread model and control for (log) securitization business shared between CRAs and security issuers. (coefficient not significant)
Re-computed all RIS and DRIS and rerun regression for Hypothesis 1.
33
Robustness: Regression based on AAA subordination
E.g. Ashcraft et al. (2010), He et al. (2011) use level of AAA subordination to summarize tranche ratings to deal level.
34
Main findings and policy implications
Statistically and economically large rating favors Rating favors in High Value issuer-agency relationships. Effect difficult to attribute to rating errors made in good faith or to an
omitted variable specific to a particular asset class. Reallocation of resources from disadvantaged to large issuers.
Competitive distortions, bank concentration, too-big-to-fail banks,break-down of rating-contingent regulation, … ?
Rating favors more pronounced for credit risk lemons Rating favors twice as large for the 10% of deals with highest rating-
implied credit risk. Incentive distortion to supply more low quality products causing a
quality degradation during the structured debt boom 2004-06 ?
35
THANK YOU
36
Structure of Rating Errors
Test for error clustering:
1) Use baseline specification to estimate rating errors.2) Compute error correlation between the deals
of a specific issuer for ratings of a specific CRA.3) (Reject) H0 of a random non-directional Good-Faith error.
4) Do 1) – 4) for High and Low Value relationships separately.
5) Do 1) – 4) computing error correlation only between deals belonging to different asset classes.
=> Rule out omitted variable bias specific to an asset class.