MBS ratings and the mortgage credit boom

1. MBS ratings and the mortgage credit boom Adam Ashcraft, Paul Goldsmith-Pinkham, and James Vickery (NY Fed) 2009 Federal Reserve Bank of Chicago Bank Structure Conference May 7, 2009 Views expressed in this presentation are our own, and do not reflect the opinions of the Federal Reserve Bank of New York or the Federal Reserve System.

2. Ratings downgrades: average number of notches Subprime Alt-A Source: ABSNet

3. MBS prices down sharply, even for AAA securities

4. This paper • Goal: evaluate the informational content of initial credit ratings on subprime and Alt-A MBS deals issued in the period leading up to the crisis (2001-07). Key questions: • How informative were credit ratings? Did ratings reflect risk (measured ex-ante) of underlying mortgages? • Did ratings standards decline during the mortgage credit boom?

5. The rating agency defense “In response to the increase in the riskiness of loans made during the last few years and the changing economic environment, Moody’s steadily increased its loss expectations and subsequent levels of credit protection on pools of subprime loans. Our loss expectations and enhancement levels rose by about 30% over the 2003 to 2006 time period…” “Along with most other market participants, however, we did not anticipate the magnitude and speed of the deterioration in mortgage quality (particularly for certain originators) or the rapid transition to restrictive lending.” Michael Kanef, Moodys Group MD Senate testimony, 9/26/07

6. Preview of main findings • Credit ratings insufficiently sensitive to risk. • Projected forecast loss rates from a simple default model strongly forecast worse ex-post deal performance, after controlling for the rating. • CRAs particularly over-rated deals with high share of low-documentation loans and investor loans. • Low-doc finding consistent with claims that rating agencies relied excessively on information from issuers. • Some evidence of deterioration in ratings standards at peak of mortgage credit boom (2005-07)

7. Stylized structure of an RMBS deal Individual Mortgages RMBS Bonds Credit rating measured as fraction of claims below a particular rating notch. Example: AAA subordination = B / [A+B] ‘AAA’ RMBS M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19 M20 M21 M22 M23 M24 M25 M26 M27 M28 M29 M30 A M31 M32 M33 M34 M35 M36 M37 M38 M39 M40 M41 M42 M43 M44 M45 M46 M47 M48 M49 M50 M51 M52 M53 M54 M55 M56 M57 M58 M59 M60 M61 M62 M63 M64 M65 M66 M67 M68 M69 M70 ‘AA’ RMBS M71 M72 M73 M74 M75 M76 M77 M78 . . . M 2000 ‘A’ RMBS M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19 M20 ‘BBB’ RMBS B M21 M22 M23 M24 M25 M26 M27 M28 M29 M30 ‘BBB-’ RMBS M31 M32 M33 M34 M35 M36 M37 M38 . . . Residual M 1000 Source: Kupiec (2008)

8. Our data Security-level data from ABSNet and Bloomberg on the characteristics of each tranche. Original credit ratings, current credit ratings, face value, coupon, insurance, payment features (e.g. IO, PO), etc. Loan-level data on individual mortgages underlying each deal, from LoanPerformance. Loan and borrower characteristics (LTV, FICO, DTI, location, identity of lender etc.) Matched with OFHEO house price data based on location. Performance: Record each month of whether borrower made payment.

9. Nonagency MBS issuance

10. Number of credit ratings by deal type

11. Credit ratings over time (% subordination below AAA)

12. Informativeness of ratings • Credit rating is a summary statistic for the level of credit risk of the deal. (Summarizes the CRAs information set). • Cross-sectional predictions: • Credit ratings should forecast deal performance (defaults, losses, downgrades etc.). • Controlling for the rating, initial risk variables should not systematically forecast deal performance (since to extent relevant, should already be incorporated in rating).

13. Step 1: Loan-level default model • First: produce simple benchmark expected default rate for each MBS deal, using loan level data. • Approach: • Estimate historical default model, as function of underwriting variables (LTV, FICO, loan type etc.) • Substitute each mortgage into model to obtain projected default rate, and aggregate to deal level. • Model estimated recursively for each half-year vintage between 2001-07. • Note: Projected deal default rate based only on “real time” data as at time of deal issuance.

14. Model specification • Dependent variable: 90+ delinquent after 12 months • Key underwriting variables: • Trailing house price appreciation (OFHEO, past 12 months) • Borrower FICO score and debt-to income (DTI) ratio • Combined loan to valuation (CLTV) ratio • Loan type variables (ARM, FRM, interest only, balloon loan) • Documentation of borrower income (full, low, no doc) • Year x quarter dummies • Others: lender dummies, origination channel, investor etc. • Linear probability model. Two specifications: • Simple model: Similar to Demyanyk & Van Hemert (2009) • Complex model: More covariates, interaction terms.

15. 90+ delinquency after 12 months, by deal vintage

16. Predictors of ex-post default: Subprime Dependent variable: % deal 90+ delinquent months after issuance

17. Predictors of rating downgrades: Subprime Dependent variable: Rating downgrade (notches, weighted average)

18. Deal-level covariates Dependent variable: % deal 90+ delinquent after 12 months

19. Other analysis • Also performed all analysis separately for Alt-A deals • Results are similar (omitted given time constraints). • Found similar results studying default at longer horizons • 90+ delinquency, prepayment with loss or REO after 24 months. • Also estimated determinants of initial ratings. • Ratings related to fundamentals (predicted delinquency, excess spread, insurance etc.) in expected ways. • Next slide: Estimate this regression over the pre-boom period, and compare actual subordination to predicted subordination from this regression.

20. Actual and predicted subordination

21. Summary of main findings • Our evidence suggests MBS credit ratings were noisy measures of credit risk of the deal. • Projections from a simple default model significantly outperform ratings as predictors of future deal performance. • Deals with a high fraction of low-doc and investor loans, and loans from areas with high HPA, perform worse ex-post, conditional on the historical data. • Combination of CRAs that rate deal systematically related to performance. Not clear evidence of rating shopping, however. • Some time-series evidence of deterioration in ratings at end of the boom (2006-07)

22. Additional slides

23. Predictors of ex-post default: Alt-A Dependent variable: % deal 90+ delinquent months after issuance

24. Predictors of rating downgrades: Alt-A Dependent variable: Weighted average downgrade (notches)

25. Summary statistics for loans underlying deals

26. Loan-level summary statistics (cont…)

27. Alt-A deal characteristics over time

28. Subprime deal characteristics over time

29. Excess spread, bond insurance by asset class over time

30. Credit rating agency market share by asset class

31. Loan-level models

32. Modelling errors / lack of competence view “We ran our staffing model assuming the analysts are working 60 hours a week and we are short on resources…. The analysts on average are working longer than this and we are burning them out. We have had a couple of resignations and expect more”

33. Motivation and related literature • Role of rating agencies in the subprime crisis: • Bad luck, bad models or bad incentives? • Entry into the credit rating industry restricted by SEC. • Argument is that franchise value improves informational content of ratings and reduces rating shopping. • Failure of “mechanical” models based on historical data

34. Original and current ratings: GSAMP 2006 NC-2