درباره سازگاری رتبه بندی و بازده بازار اوراق قرضه
کد مقاله | سال انتشار | تعداد صفحات مقاله انگلیسی |
---|---|---|
15287 | 2004 | 20 صفحه PDF |
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Banking & Finance, Volume 28, Issue 11, November 2004, Pages 2769–2788
چکیده انگلیسی
We study the consistency of the credit-risk orderings implicit in ratings and bond market yields. By analyzing errors in term structure estimates for bonds with particular ratings, we show that for significant periods, a quarter of some categories of high credit quality bonds are rated in a manner that is inconsistent with their pricing. Adjusting for economic determinants of spreads (tax, liquidity and risk premiums) and allowing for the dynamic adjustment of ratings and spreads largely eliminates the inconsistencies, however.
مقدمه انگلیسی
Long-term bond ratings produced by agencies like Moody’s and Standard and Poor’s provide financial market participants with judgments, of a standardized nature, on the likelihood that bond issues will be repaid in an orderly manner. The importance of ratings as a source of information to investors has increased in recent years as bond markets have grown more international and come to include a wider range of obligors.1 Ratings have also acquired new roles, as supervisory authorities have made regulatory requirements for financial institutions contingent on ratings.2 Recently, it has been suggested (see Basel Committee on Banking Supervision, 1999) that regulatory capital for G10 banks be based in part on the agency ratings of the bank’s borrowers. In view of the increasing reliance on bond ratings in credit risk markets, it is important to ask how reliable are ratings as indicators of credit standing, both in general and for particular types of obligor. In particular, are ratings consistent, cross-sectionally and over time, with other measures of credit risk? Two recent papers have critically examined ratings as measures of default risk in this way. First, Blume et al. (1998) show that firms with given accounting ratios received a significantly lower rating in the early 1990s than firms with similar accounting ratios would have received in the late 1970s and early 1980s.3 The implication is that rating agencies have changed the way in which they evaluate credit standing. Second, Delianedis and Geske (1998), use equity and liability data for US firms, to construct alternative credit risk indicators and compare their forecasting performance to that of ratings. They conclude that the default probabilities generated by their models increase well in advance of ratings down-grades. They cite this as evidence of “rating stickiness”, i.e., that rating agencies do not immediately change ratings when news affecting an obligor’s credit quality is revealed. In this paper, we study a third aspect of ratings, namely their consistency or otherwise with bond market prices. Altman (1989) shows that, for all years from 1973 to 1987, mean yields to maturity increase monotonically as one descends the ratings scale. However, Altman’s finding only implies that average bond spreads and ratings are consistent. If individual spreads within a particular rating category vary substantially around their mean, it may be that the implicit credit quality ordering attributed to obligors by the rating agencies and the bond market are very different. To investigate this empirically, we extract average yields for different rating categories using Nelson–Siegel techniques as described in Nelson and Siegel (1987). The data we employ consists of ratings and price histories in the period 1988–1998 for a large number of non-callable, dollar-denominated, international bonds, primarily Eurobonds. For each trading day, we calculate yields for different maturities for the three highest credit quality rating categories, AAA, AA, and A. We then compare the bonds’ actual market values with the prices they would have if a claim to the bond’s cash flows were priced with our estimated yields. We say that a bond valuation is inconsistent with its rating if the market price is above (below) the price it would have if it were valued using average term structures corresponding to a higher (lower) rating category. Thus, the price of a AAA bond is inconsistent with its rating if it is lower than the value one obtains using the AA term structure. Similarly, an AA bond price is inconsistent with the bond’s rating if it is higher than the price obtained using the AAA term structure or below that one obtains using the A term structure. We find that, on average, between a fifth and a quarter of AA bonds are priced in a way that is inconsistent with their ratings. Smaller fractions of AAA- and A-rated bonds are inconsistent but only because these can only be reclassified in one direction (down for AAA and up for A since we only consider A and above rated bonds in our study). Some fraction of bond price variation no doubt reflects liquidity, risk premiums and tax effects. To allow for these influences, we regress pricing errors from the Nelson–Siegel fits on variables designed to proxy for economic determinants of spreads, i.e., risk premiums, liquidity and tax. The risk premium variables are based on bond market factor “betas”. Liquidity proxies include the age and face value of the bond issue and the frequency with which it is quoted. The proxy for tax effects is the coupon rate. After subtracting the fitted value of these effects (the regression coefficient times the regressor) from the market price, we once again compare the adjusted market prices with the prices obtained using estimated yields for superior and inferior ratings categories. About a third of the inconsistencies are eliminated by adjusting this way for tax, liquidity and risk premiums. After six months, during which time spreads and ratings have had time to adjust dynamically, around a half of the remaining inconsistencies disappear. Hence, we conclude that ratings and bond market yields suitably adjusted are reasonably consistent, contrary to what one might believe if one looked at the unadjusted data alone. A substantial number of earlier studies have looked at the relationship between ratings and bond prices. West (1973), Liu and Thakor (1984), Kao and Wu (1990) and Ederington et al. (1987) find that, conditional on economic and firm specific variables, ratings do have explanatory power for bond yields. In these studies, the ratings may proxy for (publicly known) omitted variables which affect yield spreads. To avoid this problem of firm-specific omitted variables, several studies have examined whether rating changes coincide with excess returns on either the obligor’s equity or debt values (see Katz, 1974; Weinstein, 1977; Griffin and Sanvicente, 1982; Ingram et al., 1983; Hand et al., 1992; Goh and Ederington, 1993; Kliger and Sarig, 2000). While evidence from the earlier studies was mixed, the more recent contributions suggest that rating changes do impart some new information, not publicly available to the investor. A significant part of our study involves estimation of term structures for corporate bonds. Earlier papers which have extracted such term structure estimates include Sarig and Warga (1989), Schwartz (1998) and Dullmann et al. (2000). None of these has examined bond-spread/rating inconsistencies of the kind we analyze here, however, although both Schwartz (1998) and Dullmann et al. (2000) observe crossings in the mean spreads for different rating categories and Schwartz (1998) discusses trading strategies based on these inconsistencies. Finally, note that an important literature has recently emerged on the determinants of bond market spreads. Notable contributions to this literature include Delianedis and Geske (1999), Elton et al. (2001), Elton et al. (2000), Huang and Huang (2002) and Houweling et al. (2003). These papers model determinants of spreads as we do when we adjust spreads for non-credit pricing factors.4 But they do not share our focus on a comparison of credit risk orderings implicit in ratings. The structure of the paper is as follows. Section 2 describes our bond data set and the Nelson–Siegel techniques we use to extract estimates of average yield curves for different rating categories. Section 3 defines two notions of rating/spread consistency and inconsistency. Section 4 discusses adjustments for tax, liquidity and risk premiums and dynamic effects. Section 5 concludes.
نتیجه گیری انگلیسی
In this study, we identify and document substantial differences in the ordering of credit standing implicit in bond market yields and ratings. For example, one quarter of AA-rated bonds have inconsistent prices and ratings. It is possible that the orderings implicit in yields and credit ratings differ even if ratings agencies and bond market participants are correctly evaluating credit quality. This will be the case if risk premiums, tax and liquidity effects are substantial. To adjust for these influences, we regress residuals from defaultable bond term structure fits on risk premium betas and proxies for tax and liquidity effects. Using the fitted part of the regressions to adjust for these effects, we reduce the number of reclassifications significantly. It is also true that the credit quality orderings implicit in ratings and bond spreads are broadly consistent but that they adjust towards each other over time. We examine the rate at which inconsistencies between ratings and spreads are eliminated over time and find that after six months between half and two-thirds of inconsistencies have disappeared. We conclude that allowing (i) for economic determinants of spreads and (ii) for dynamic adjustments, the apparent and very substantial discrepancies between ratings and bond market spreads can be accounted for.