Practice Leader, ERM and Structured Products Advisory, Promontory Financial, New York
The credit risk in US agency mortgages and mortgage-backed securities has always received little attention compared with the interest rate (prepayment) risk in them. It has been the wisdom that as long as each mortgage satisfies the agencies’ specific conditions (including the presumption that there is no systematic fraud in the documentation) and the portfolio is well diversified both in terms of size of each loan relative to the size of the portfolio and in terms of geographic distribution across the US, the portfolio credit risk is small. In the last couple of months, this wisdom has taken a significant beating as evidenced by investors’ worries over Fannie Mae and Freddie Mac, the two very large US government-sponsored agencies (GSE).
While the focus in the popular press has been on whether and by how much the GSE are undercapitalized, whether or not the US government should nationalize the GSE and the steep decline in stock prices of the GSE, everything stems from perceived credit risk in agency mortgages and mortgage-backed securities and consequently the potential losses the GSE face.
The US mortgage loans can broadly be categorized as “agency”, “Alt-A” and “subprime”. Of all US mortgages outstanding, about 12% is subprime, about 50% is agency and the rest is Alt-A. Of the originations in the 2003–07 timeframe, the proportion of agency mortgages is significantly lower than 50% while after August 2007, agency mortgages are the dominant category being originated.
Increasing delinquency and foreclosure in subprime mortgages catalyzed the near demise of the mortgage securitization markets in 2007 and huge marked-to-market losses in CDO books of many banks in the United States and elsewhere in 2007 and 2008. Yet subprime constituted a small fraction of all US mortgages. In fact, the delinquency rates for US subprime mortgages may have reached a plateau, although it is too soon to be conclusive. Investors now fear increasing delinquencies in Alt-A and possibly agency mortgages. Since the size of agency market is much bigger than the subprime market and the originally expected losses are relatively small, any small change in projected delinquencies and losses can have a big effect on the percentage increase in economic capital required to support the GSE portfolios as well as on the mortgage market as a whole.
Investors extrapolated downgrades and losses in mezzanine RMBS and CDO of RMBS tranches observed in mid-2007 to the much larger-sized, super-senior tranches without explicit computation of their relative (credit) risks. Similarly, investors are extrapolating losses in subprime to the much larger sized mortgage markets of Alt-A and agency mortgages without necessarily assessing relative credit risks in securities backed by sub-prime mortgages and by agency mortgages. This coupled with the uncertainty of any government action on matters of restructuring mortgages to prevent foreclosure and of nationalizing the GSE has left the mortgage market in a sorry state, postponing the possibility of a comeback of mortgage-related securitizations.
In this issue, we present two full-length research papers and two technical reports. The first paper “Break on through to the single side” is by Madan and Schoutens. In this paper, the authors consider a structural form model for a firm’s asset value process, with default occurring when, for the first time, asset value breaches a low barrier. They assume that a firm’s value follows a Lévy process with negative jumps and derive the distribution of a firm’s default time. They also calculate the corresponding CDS spread. After that the authors study the calibration quality of their model in the context of a portfolio of 125 representative credits contained in the iTraxx and CDX indexes.
The second paper “Valuing loan credit default swap cancellability” by Benzschawel et al presents a general framework for calculating the additional premium required by protection sellers as compensation for potential cancellation of loan credit default swaps (LCDS). They assume that when high-yield credit becomes investment-grade the issuer firm will repay its loans and contractually the LCDS will be terminated. The method requires as inputs both a probability of default and likelihood of cancellation. While risk-neutral probabilities of default may be inferred from market spreads, the actual probabilities of cancellation are unobservable. Because of this, the authors use historical ratings transitions to estimate likelihoods of cancellation for credits of various ratings and calculate their corresponding cancellation premiums. The authors also provide an approximate formula for LCDS cancellability as the ratio of fee legs for cancellable and noncancellable LCDS.
This issue has two technical reports. A technical report describes a particular practical technique and enumerates situations in which it works well and others in which it does not. Such reports provide extremely useful information to practitioners in terms of saved time and duplication of efforts. The contents of technical reports complement rigorous conceptual and model developments presented in the research papers and provide a lot of value to practitioners.
The first technical report “Development and validation of credit scoring models” is by Glennon et al. The authors develop a suite of credit scoring models using the same data but with three modeling techniques widely adopted in the banking industry, including logistic regression, rank ordering and CHAID. The authors use three approaches in variable selection: stepwise, resampling and the intersection between the two. The performance of the models is then evaluated with two types of measures: discriminatory power (K-S Statistic) and predictive accuracy (Hosmer–Lemeshow Goodness of Fit Test). The statistical validations are performed on both in-time and out-of-time samples, as well as multiple segments of the samples. The conclusions are: 1) the models have discriminative powers; 2) the differences across the models are small and 3) the out-of-time validation shows a substantial loss of accuracy for all models – the loss is mainly from the large changes in the default rates of the high risk components of the population. Practitioners will find this article highly useful in developing industry-standard scorecards. This paper reads like a credit scoring primer and its principles and techniques are accessible to even beginner readers.
The second technical report “Maturity adjustments under asymptotic single risk factor models: a comparative analysis” by McCoy compares three approaches for calculating maturity adjustment for credit exposures of maturity greater than one year within the scope of asymptotic single risk factor (ASRF) model. These approaches are: (i) multi-period probability of default based on the basic Merton framework as derived in Gurtler and Heithecker (2005); (ii) “annualized” probability of default based on the basic survival analysis – a relationship that has been in existence in credit literature for a long time; and (iii) default probability term structure derived from market credit spreads. Actually there is a fourth one lurking in the background: the Basel II maturity adjustment formulation. Maturity adjustment in this context is the adjustment needed on the default-mode economic capital calculated with a one-year horizon and some prespecified level of confidence. The overall contents of the article provide a good survey of most things associated with maturity adjustment in economic capital calculations. To the not-so-sophisticated practitioner, it provides a wealth of simple knowledge about maturity adjustment. Maturity adjustment is an area that has not received much attention in credit risk literature as evidenced by the rarity of papers with coverage of maturity adjustment specifically.