Monday, June 9, 2008

Estimating valuation losses on subprime MBS with the ABX HE index — some potential pitfalls

(Ingo Fender and Peter Hördahl / extracted from June 2008 BIS Quarterly Review) Repeated large-scale writedowns of exposures to the US mortgage market and continuing deterioration of the US housing sector have given rise to strong public and private sector interest in estimates of overall subprime-related losses. In this context, particular attention has been devoted to estimated market value changes for subprime mortgage-backed securities (MBS) and how these compare to disclosed writedowns by banks and other investors.1 A key source of data for such estimates has been the ABX HE series of indices based on credit default swaps (CDS) with subprime exposure. This box conducts a simple analysis of valuation losses on subprime MBS on the basis of ABX prices and highlights a number of possible limitations of such estimates. In particular, it is argued that past estimates of total valuation losses at the AAA level may have been inflated by more than 60%.

The ABX HE index. Trading in the first ABX index series started in January 2006. Each index consists of a group of equally weighted, static portfolios of CDS referencing 20 subprime MBS transactions. Following the example of other major CDS indices, new "on-the-run" index series are being introduced every six months. Each of these ABX series references 20 completely new subprime MBS deals issued during a six-month period prior to index initiation. Each index series, in turn, consists of five subindices, each referencing tranche exposures to the same 20 underlying MBS deals, though at different levels (AAA, AA, A, BBB and BBB-) of the capital structure.2 Index prices reflect the willingness of investors to buy or sell protection on the basis of their views about the risk of the underlying subprime loans, and are quoted as a percentage of par.

Mark to market losses on subprime MBS. There are various ways to measure losses on subprime MBS of which none is inherently superior. Approaches will differ according to loss concepts and data needs, with valuation (ie mark to market) losses arguably the most straightforward ones to calculate.3 This is because of the reliance of the mark to market concept on observed prices, which obviates the need to make assumptions about parametric inputs or historical relationships. To obtain estimates of mark to market losses for subprime MBS, ABX prices, by rating and vintage, can simply be applied to outstanding volumes of these securities. Graph A illustrates the results (centre panel) of such an exercise, based on outstanding volumes (left-hand panel) by rating category for each vintage of subprime MBS issued between 2004 and 2007.4 According to this measure, ABX prices put the value of the outstanding subprime MBS inventory at around 59 cents on the dollar as of end-May 2008. This, in turn, would imply total valuation losses of some $250 billion, of which $119 billion (about 47%) is incurred at the AAA level.

Pitfalls in using the ABX. Estimated mark to market losses and actual writedowns made by banks and other investors can differ for a variety of reasons. Analysts, depending on their objective, thus have to be mindful of potential sources of bias. At least three such sources can be identified, of which two are specific to the ABX index:

  • Accounting treatment. Subprime MBS are held by a variety of investors and for different purposes. While large amounts of outstanding subprime MBS are known to reside in banks’ trading books, banks and other investors may also hold these securities to maturity. This can result in different accounting treatments, which would tend to deflate actual writedowns and impairment charges relative to estimates of mark to market losses on the basis of market indices, such as the ABX. The size of this effect, however, is difficult to determine. Further complexities are added once securities cease to be traded in active markets, implying the use of valuation techniques, which may differ across investors, in establishing fair value.5
  • Market coverage. ABX prices may not be representative of the total subprime universe, due to limited index coverage of the overall market. Original balance across all four series has averaged about $31 billion. This compares to average monthly MBS issuance of some $36 billion over the 10 quarters up to mid-2007, ie almost a month’s worth of subprime MBS supply per index series. Similarly, with 2004-07 vintage subprime MBS volumes estimated at around $600 billion in outstanding amounts, each series represents some 5% of the overall universe on average. At the same time, ABX deal composition is known to be quite similar in terms of collateral attributes (such as FICO scores and loan-to-value ratios) to the overall market (by vintage).6 Therefore, despite somewhat limited coverage, this particular source of bias may not be large.
  • Deal-level coverage. Similarly, ABX prices may not be representative because each index series covers only part of the capital structure of the 20 deals included in the index (see Graph A, right-hand panel, for an illustration).7 In particular, tranches referenced by the AAA indices are not the most senior pieces in the capital structure, but those with the longest duration (expected average life) - the so-called "last cash flow bonds". These claims will receive any cash flow allocations sequentially after all other AAA tranches have been paid; and tend to switch to pro rata pay only when the highest mezzanine bond has been written down. It follows that AAA ABX index prices are going to reflect durations that are longer, and effective subordinations that are lower, than those of the remaining AAA subprime MBS universe. As a result, using newly available data for MBS tranches with shorter durations, the $119 billion of losses implied by the ABX AAA indices as of end-May would be some 62% larger than those implied under more realistic assumptions.

1 See, for example, International Monetary Fund, Global Financial Stability Report, April 2008, pp 46-52, and Box 1 in Bank of England, Financial Stability Report, April 2008. 2 Supplementary indices, called ABX HE PENAAA, were introduced in May 2008 to provide additional pricing information for all four existing vintages. 3 An alternative approach, likely to lead to very different results, would estimate future default-related cash flow shortfalls on the basis of deal-level or aggregate data for subprime securities. To obtain these estimates, such methodologies rely on information about collateral performance and require the analyst to make assumptions about structural relationships and model parameters. Typical subprime loss projections, for example, use delinquency data and assumptions about factors such as delinquency-to-default transitions, default timing, and losses-given-default. See Box 1 in the Overview section of the December 2007 BIS Quarterly Review for an example on the basis of an approach devised by UBS. 4 Mark to market losses (relative to par) are calculated assuming that unrated tranches are written down completely; ABX prices for the BBB- indices are used to mark BB collateral; rated tranches from the 2004 vintage are assumed unimpaired; outstanding amounts remain static. 5 For details, see Global Public Policy Committee, Determining fair value of financial instruments under IFRS in current market conditions, December 2007. 6 See, for example, UBS, Mortgage Strategist, 17 October 2006. 7 Incomplete coverage at the deal level further reduces effective market coverage: typical subprime MBS structures have some 15 tranches per deal, of which only five were originally included in the ABX indices. As a result, each series references less than 15% of the underlying deal volume at issuance. 8 Duration effects at the AAA level are bound to be significant for overall loss estimates as the AAA classes account for the lion’s share of MBS capital structures. Using prices for the newly instituted PENAAA indices, which reference "second to last" AAA bonds, to calculate AAA mark to market losses generates an estimate of $73 billion. This, in turn, translates into an overall valuation loss of $205 billion (ie some 18% below the unadjusted estimate of $250 billion).

No comments: