The Effect of Bond Rating Agency

2024年2月13日发(作者：小学拼音字母)

THE JOURNAL OF FINANCE • VOL. XLVII. NO 2 . JUNE 1992The Effect of Bond Rating AgencyAnnouncements on Bond and StockPricesJOHN R. M, HAND, ROBERT W. HOLTHAUSEN, andRICHARD W. LEFTWICH*ABSTRACTThis paper examines daily excess bond returns associated with announcements ofadditions to Standard and Poor's Credit Watch List, and to rating changes byMoody's and Standard and Poor's, Reliably nonzero average excess bond returns areobrved for additions to Standard and Poor's Credit Watch List when an expecta-tions model is ud to classify additions as either expected or unexpected. Bondprice effects are also obrved for actual downgrade and upgrade announcements byrating agencies. Excluding announcements with concurrent disclosures weakensthe results for downgrades, but not upgrades. The stock price effects of ratingagency announcements are also examined and contrasted with the bond PAPER EXAMINES DAILY excess bond and stock returns associated with twotypes of bond rating agency announcements; warnings of possible ratingchanges via additions to Standard and Poor's Credit Watch List between1981 and 1983, and actual rating changes announced by Moody's andStandard and Poor's between 1977 and 1982. Prior work that has ud bondprice data to examine the effect of rating changes has been mixed. Weinstein(1977) (monthly bond returns), and Wakeman (1978) (monthly stock andweekly bond returns) do not find a price reaction at the time of ratingchanges. Other studies, such as Katz (1974) (monthly changes in bondyields), Grier and Katz (1976) (average monthly bond prices) and Ingram,Brooks and Copeland (1983) (monthly changes in municipal bond yields), findsignificant bond price reactions. One potential advantage of our study is the* Hand and Leftwich are from the Graduate School of Business at the University of un is from the Wharton School at the University of Pennsylvania. We wish to thankJohn Elliott, Paul Healy, Gene Imboff, Wayne Mikkelson, Dale Mor, Pat O'Brien, ScottSticke!, Tom Stober, an anonymous referee and David Mayers (editor) for their comments ona previous version of this paper, as well as workshop participants at Cornell University,Massachutts Institute of Technology, the University of Michigan, Ohio State University, theUniversity of Pennsylvania, and the University of Rochester, We wish to thank the InteractiveData Corporation for a rearch grant allowing access to the daily bond prices ud in tbis un and Leftwicb gratefully acknowledge the financial support of tbe Institute ofProfessional Accounting of the University of Chicago and the Center for Rearch in SecurityPrices.733

The Journal of Financeu of daily data to isolate the announcement effect on bond prices, althoughnontrading of bonds does not always allow the u of short windows forestimating the price respon. Another potential advantage is the parateexamination of warnings of possible rating changes via additions to theStandard and Poor's Credit Watch List, and of actual rating changes that arenot preceded by such warnings. Standard and Poor's began the Credit WatchList in November 1981, and companies are added to the list when Standardand Poor's believes a rating change is likely, with additions designated as"indicated upgrades" or "indicated downgrades", or "developing" if a ratingchange of unknown direction is likely. As a result, a rating change announce-ment may occur after a firm's debt is placed on the Credit Watch List, or itmay occur without being preceded by a Credit Watch designation. We alsodevelop an expectations model of rating changes to distinguish between thorating changes that are expected and tho that are l results on the announcement effects of additions to the CreditWatch List do not indicate significant average excess bond returns. However,when we study tho additions classified as unexpected additions by ourexpectations model, a significant negative average excess bond return of-1.39% is associated with indicated downgrades and a significant positiveaverage excess bond return of 2.25% is associated with indicated icant average excess stock returns are obrved at the time of indicateddowngrades but not at the time of indicated average excess bond return on the announcement of actual ratingdowngrades by Moody's and Standard and Poor's is -1.27% and the medianexcess return is -0.45%. Investment grade bonds experience a mean excessreturn of -0.55%, with a similar median excess return. Below investmentgrade bonds on average lo 3.82%, but the median bond los only 0.62%.Downgrade announcements also affect the equity price of the rerated firm,and the negative average effects on the deht and equity are similar, thoughthe effects on equity are somewhat more negative than the effect on the evidence on upgrades is weaker; the mean and median announcementeffects are hoth approximately 0.35%, and little difference is obrved be-tween above and below investment grade bonds. Finally, there is littleevidence of positive excess returns for the equity of upgraded determine whether our results are the product of the rating announce-ment or of simultaneously announced news from other known sources, weclassify ohrvations as either "contaminated" or "noncontaminated'' byother news. In general, the excess hond returns associated with Credit Watchannouncements are even more supportive of price effects when the noncon-taminated obrvations are examined. However, the significant negativeaverage excess hond returns associated with actual rating downgrades disap-pear when we examine noncontaminated obrvations, though evidence ofnegative average excess stock returns is still apparent. For upgrades, elimi-nation of contaminated events caus the measured average excess hondreturns to become more positive, but stocks still evidence little impactassociated with the rating change.

The Effeet of Bond Rating Agency Announcements 735Results in previous studies on the effect of rating agency announcementson daily common and preferred stock prices are generally consistent with theresults in this paper. Holthaun and Leftwich (1986) examine the commonstock price respon to Credit Watch announcements and bond rating changes,and fmd evidence consistent with a stock price respon to all of thoannouncements except actual rating upgrades. Stickel (1986) examines theeffect of preferred stock rating changes on preferred stock returns and findvidence consistent with price effects for both upgrades and n I of the paper describes the sample of additions to the Credit WatchList, and of actual rating changes by Moody's and Standard and Poor'n II describes the bond and stock price tests, and Section III reportsresults. A cross-ctional analysis of the excess bond returns associated withrating changes is given in Section IV. Concluding remarks are in Section V.I. Data: Moody's and Standard and Poor's Rating ChangesWe analyze two types of hond rating agency announcements. First, weexamine a sample of approximately 250 additions to the Standard and Poor'sCredit Watch List. Second, we examine a sample of approximately 1,100bond rating change announcements by Moody's and Standard and Poor'rd and Poor's began the Credit Watch List in November ies are added to the list when Standard and Poor's believes a ratingcbange is likely, with additions designated as "indicated upgrades", "indi-cated downgrades", or "developing" if a rating change of unknown directionis likely (for example, due to a potential merger). Additions to the CreditWatch List are published every Friday/ We sampled obrvations from theinception of the Credit Watch List to December 31, 1983. Obrvations areincluded if they relate to taxable straight-debt other than commercial paperand if the firm's common stock was publicly traded on the New York orAmerican Stock Exchanges at the time of the sample of actual rating change announcements consists of bond ratingchanges announced by Moody's and Standard and Poor's during the period1977-1982. Moody's rating changes are obtained from the ction "TaxableCorporate Securities—Ratings Reviewed and Revid" of Moody's BondSurvey. Standard and Poor's rating changes are obtained from a listingentitled "Standard and Poor's Corp. Corporate Finance Rating Changes"supplied by officials of the Standard and Poor's Corporation. Only ratingchanges for fully taxable corporate straight-debt bonds are included in thestudy (i.e., tax-free industrial revenue bonds, convertible debt, and floating^ The announcements usually take place during trading. If Friday is a holiday, the announce-ment is made on Thursday. The Wall Street Journal story about Credit Watch additions typicallyappears on Monday. We define Friday as Day 0 unless the Wall Street Journal story precedesMonday, in which ca Day 0 is the day before the Wall Street Journal story,"•^All Moody's rating changes on April 20, 1982 are eliminated. On that day, Moody's an-nounced a large number of rating changes to reflect a new rating system incorporating threerating grades within each class between AA and B.

The Journal of Financerate notes are excluded).^ The announcement date, Day 0, for each obrva-tion is the press relea date indicated by Moody's and Standard and Poor's.^If veral bonds of the same firm are affected simultaneously, we lect amaximum of five bonds with the longest time to maturity for both thesamples of Credit Watch additions and actual rating changes. Long maturi-ties are sampled to increa the likelihood of detecting excess bond returns. Ifbonds of the same firm in two different rating class are affected, we samplefrom botb class. No firm in the sample has bonds in more than two bond prices and volume, coupon amounts, payment dates and matu-rity dates were collected from the Interactive Data Corporation (IDC) Corpo-rate Bond Data Ba from 11 days before to 60 days after the announcementof an addition to the Credit Watch List or a rating change. Only transactionsprices are ud in this study. Many corporate bonds trade infrequently, andoften only bid prices are reported. We performed extensive filtering on theprice data, visually inspected every ries, and decided to exclude all biddata. Large reversals are common in the price ries if both bid and transac-tion prices are included. Even if transaction prices adjust to news announce-ments or interest rate shocks, many bid prices adjust only with a uently, for some ries the bid prices are consistently either above orbelow transaction prices (by more than half any plausible bid-ask spread) forone or two I provides information on the trading frequency for the 1,548 bondsin the sample of bonds experiencing rating changes. At the extremes, notrades occur in the sample period (11 days before to 60 days after the ratingchange announcement) for 6.9% of tbe sample, and 9.7% of the sample tradeson at least 61 of the 72 days in the sample period. Table I reveals that thetrading frequency is considerably higher for bonds below investment grade(below BAA) than for investment grade bonds (BAA and above). For exam-ple, 25.2% of the 238 bonds below investment grade trade on at least 61 ofthe 72 sample days, while only 6.9% of the investment grade bonds trade thatfrequently. Further inspection reveals that 50% of the bonds below invest-ment grade trade on at least half of the sample days while only 22.8% of theinvestment grade bonds trade that II summarizes the distribution of differences between the originaland tbe revid rating categories for the 1,133 bonds with at least onetransaction price in the 11 days before the rating change and at least onetransaction price in the 60 days after the rating change. Relative to Table I,the sample size declines from 1,548 to 1,133 becau a bond is only includedin Table II if it has a transaction price both before and after the ratingchange announcement. There are 841 downgrades and 292 upgrades. Theimbalance in the number of downgrades and upgrades is probably due to thetime period chon. For the vast majority of obrvations, the difference^ We are grateful for the ofTicials of Moody's and Standard and Poor's for providing thisinformation.

The Effect of Bond Rating Agency Announcements737itive—:i ^J Oc-OSOaiCD.-H00<-^o1ooiCOdTftnOlorHfp05-aCO*jcUu^-as Ct--fN^^T-HocoOlCOCiCOCOc— (N in totn T-H Ol T-HTH OJ iM inre .5rto oii fcH -o t.t^ *^ H4 ^^c^ oC~ CDrtgQ« -oo S01 10 00 01 tq tN q(d CO c4 in in c^ orH (N CO in OJ Ooto [-- QOort H3O 00 CD TfCO Tf 1:0 (N

The Journal of FinanceTable IIThe grading of the rating categories is bad on the rating scheme introduced April 20, 1982Grades are a cardinal variable measured on a scale of 28 (for rating AAA) to 1 (for rating D)Since some of the 10 class (AAA to D) have three grades, each class is assigned three numbersClass without grades (D, C, CC, CCC, AAA, and all Moody's class before April 20, 1982) areassigned the midpoint of the three of RatingsCategories Changed1TotalDowngrades34525821613401Distribution of Differences between Original and RevidRating Categories for a Sample of 841 Downgrades and 292Upgrades of Corporate Taxable Debt Ratings Revid byMoody's and Standard and Poor's during 1977-1982^Upgrades10Total46134231133' The sample includes only bonds that have prices both before and after the rating n the original and the revid rating category is only either onecategory (461 obrvations) or two categories (342 obrvations). Neverthe-less, 330 obrvations change three or more categories. Though not shown inTable II, 50 bonds change from above investment grade (rating categoriesBBB - and above) to below investment grade (rating categories BB + andbelow) and 17 are upgraded to an investment grade category from a below-investment grade of the subquent excess returns tests depend on whether the ratingagency announcement is accompanied by one or more concurrent identify concurrent disclosures for each announcement, we arched theWall Street Journal index for potential new stories in the window spanningthe first trading day with an available bond price prior to the rating changeannouncement through the first trading day with an available bond priceafter the rating change announcement. For each news item found, we readthe story to determine if it discusd the rating agency announcement and ifit contained information from any source other than the rating agency. If astory containing information other than the rating agency announcementappeared in the Wall Street Journal during the period ud to measure theprice effect (defined later), tbe obrvation is classified as contaminated. Allremaining obrvations are classified as noncontaminated. Of cour, in-

The Effect of Bond Rating Agency Announcements 739vestors may obtain information from sources other than the Wall StreetJournal, and our tests will not detect that "contaminating" news. However,conditional on employing only one source for news releas, our test isparticularly stringent becau we exclude obrvations even if the contami-nating disclosure is triggered by the bond rating agency's announcement. Forexample, we classify an obrvation as contaminated if a rating changecaus a firm to announce postponement of a new debt issue or reduction in acapital expenditure . Measurement of Excess Bond and Stock ReturnsA. Estimation of Excess Bond Returns'Window-spanning' excess returns are employed becau bonds trade infre-quently, and bond prices are often not available on Days 0 and -I-l, thepreferred window. The window-spanning raw return for each bond is mea-sured from the last transaction price in the period -11 to -1 to the firsttransaction price on or after Day +1. For example, if a bond trades on Days-10, -3, -1-2, and -1-5, the window spanning return is calculated fromtransactions prices on Day -3 and Day +2. Accrued interest is added to theprice change to calculate the bond's return over the period. If there are morethan 20 days between the two prices the obrvation is eliminated.'^ For mostof the following tests, the median number of days in the window spanningreturn is five or six days. Excess returns for each bond for a given firm areequally weighted and the average is treated as a single obrvation.'^We measure excess bond returns as raw bond returns minus the return ona risk free bond (a long-term U.S. Treasury bond) for the window-spanningperiod. The benchmark controls for shifts in interest rates due to changes inreal rates or expectations of inflation, but it does not control for variations interm or default premiums since we u one relatively long-term riskless bondas the benchmark for all bonds, regardless of maturity. Since default premiaare positive, this measure overstates (understates) the impact for upgrades(downgrades). For short estimation periods, this bias is negligible.^Estimation of the return variances and covariances between individualbonds is hampered by the overall paucity of trades. Conquently, we urelatively simple test statistics to determine whether the average excess hondThe results reported are not nsitive to reducing the maximum number of days allowed in awindow spanning return. For example, we obtain similar results with smaller sample sizes if werestrict the sample to firms that have window spans of no greater than five results are similar if tests are conducted on individual bonds, or on a weighted-leastsquares estimate of mean abnormal performance. The weighted-least squares estimate assumesthat each obrvation is independent with the same mean and a variance proportional to thelength of the window. Each obrvation is weighted inverly according to the window obtain similar results with the mean-adjusted excess return model ud by Handjin-icolaou and Kalay (1984). In that model, abnormal performance is defined as the bond's returnless the return on a long-term U.S. Treasury bond, less the bond's default premium estimatedusing data after Day 0.

740 The Journal of Financereturn differs from zero. We calculate a ^statistic bad on the cross-ctionalstandard deviation of window-spanning excess returns, assuming that excesswindow-spanning returns of the firms are cross-ctionally independent andidentically distributed. This ^statistic has one degree of freedom fewer thanthe number of firms in the estimated excess returns for bonds are slightly skewed so we alsoreport both median excess bond returns and the proportion of positivewindow-spanning excess hond returns. A binomial test {a Z-statistic) indi-cates whether the obrved proportion of positive window-spanning excesshond returns is reliahly different from 50%.^B. Estimation of Excess Stock ReturnsExcess stock returns are defined as stock prediction errors calculated froma market model, summed over event Days 0 and +1. Market model parame-ters are estimated using the 300 day post-rating change period from Day +62to +361, hecau there is evidence in previous studies that downgrades(upgrades) are preceded by negative (positive) average excess returns. Theindex is the CRSP equally weighted New York and American Stock Ex-change Index. Median excess stock returns and the proportion of positiveexcess stock returns are also reported. The ^test on the mean excess stockreturn is bad on the cross-ctional standard deviation.^C. Expectations Model for Bond Rating Agency AnnouncementsWe develop an expectations model of bond rating changes bad on yields-to-maturity, and u that model to increa the likelihood of detectingannouncement effects associated with Credit Watch additions and bondrating changes. We measure the expectation of a bond rating change hycomparing the yield-to-maturity on a hond of interest (Credit Watch or actualrating change), estimated from the price available just prior to the ratingagency announcement, with the yield-to-maturity of a benchmark, namelythe median yield to maturity of other bonds with the same bond rating.^Since our measure of the excess bond return does not control for variation in term or defaultpremia, and default premia are positive, the expected proportion of positive window-spanningexcess bond returns is actually slightly higher than 50%.^ We also estimated, but do not report, i-statistics bad on the time-ries standard deviationof the common stock prediction errors of the portfolio. The (-statistics bad on the time-riesstandard deviation always exceed (in absolute vaiue) the f-statistics bad on the cross-ctionalstandard deviation.^ The yield-to-maturity of a rating class at a point in time is estimated from the yields-to-maturity of our sample of rerated bonds in the sixty trading days after they are rerated. A bondis ud to estimate the yield-to-maturity of a rating class if its post-rating change rating classmatches the pre rating change rating class of the bond of interest, and if it trades within 10trading days of the date on which the yield-to-maturity of the bond of interest is estimated. Thisprocedure does not allow us to form an expectation for every announcement, hence the samplesize is reduced for results bad on the expectations model.

The Effect of Bond Rating Agency Announcements 741If the yield-to-maturity of a bond of interest is greater than the benchmark,we argue that investors believe that bond has greater default risk thancomparable bonds. Thus, if the bond is downgi-aded, we classify it as anexpected downgrade. If the bond is upgraded, we would classify it as anunexpected upgrade. If the bond's yield-to-maturity is lower than the bench-mark, a subquent downgrade would be classified as unexpected, while anupgrade would be considered expected. Some tests which follow u therelative size of the expectation error, defined as the yield-to-maturity of thebond of interest less the yield-to-maturity of the benchmark, divided bythe yield-to-maturity of the lected a price-bad expectations model to capture the expectation ofa rating change just prior to its announcement, becau the prior literatureon rating changes indicates that downgrades (upgrades) are preceded bynegative (positive) excess returns for both bonds and stocks. We tested otherprice-bad expectations models, including yield-to-maturity benchmarksbad on matching rating class and duration, and matching rating class andmaturity, but did not discover superior . ResultsA. Standard and Poor's Credit Watch AnnouncementsThe results for the entire sample of Credit Watch additions in Table IIIprovide little evidence of excess bond and stock returns associated with eitherindicated downgrades or indicated upgrades. However, u of the expecta-tions model increas the obrved price effects considerably. For the entiresample of Credit Watch downgrades, mean and median excess bond returnsare -0.35% (^statistic of -1.37) and -0.37%, and 40.4% of the excess bondreturns are positive (Z-statistic of -1.96). The mean excess stock return is-0.79% (^statistic of -1.81). The 75 bonds of 50 firms classified as unex-pected downgrades have a mean excess bond return of -1.39% (^statlstic of-3.15), with only 28% positive (Z-statistic of -3.11). The mean excess stockreturn for the obrvations is -1.78% (^statistic of -2.63). Average excessstock and bond returns for the sample of indicated downgrades classified axpected are not reliably different from zero. Moreover, comparisons of theexpected and unexpected samples suggest reliable differences between thetwo portfolios. For example, for all Credit Watch indicated downgrades, thedifference in mean bond returns of the expected and unexpected portfolios is1.59% with a /-statistic of 2.20. Moreover, regressions of excess bond returnson the magnitude of the expectation error indicate that the coefficient ispositive (4.1) as expected and significantly different from zero with a p-valueof s for the noncontaminated subsample of indicated downgrades areeven more supportive of price effects for unexpected announcements. Meanexcess bond and stock returns are each -0.83% (^statistics of -2.08 and-1.07) for all noncontaminated obrvations. Splitting the sample into ex-

742The Journal of Financefl fl 01c £PeQi•-Hin(-1oO TJ-00O i-iCOoo•—1CU flNlG to I,.2 S 3odcu-O X Oi1* H PHbo i-CSd -^,nnr-"O1CDCO001—COCO00O1o1!M00ooooo1-W OS11|2CArt .Sfl5 35 ca OJ >-a, -a i^ o« m ^ AC o>to O "-i O "-I _,TfOiOOO-H iMOicq-*aOd'-nddooco Qo^inoo° j= OJ 34-1 X3 *Jg J-n

^^:: ^013

CJji}M t-

.e3 '^$s'^^ c rt o2fl ^ -ti•ti CJ T3 ^IX* ^-O Xcn'*- 3 o CO CM Tf OlRetocc ort od01 -^u „ HrtO O "-I CO00 O O O t~CN O O -H (MI I I 1 Ifl 'Od Si .2 g« fl^ .H 5o u«^infl -a = u.t. CD Q- (O3 TD _

-^

^OJl-ln Gx:.G•^

CQflrt«--!-C S F »Ul "3

"1

OJ01*JCJ01ina-dfli2 >i 01 uuQiUnew ti: "S a.I || £>- -ML-. *^BoCUOOJ148411in•a.^u0)0)Ga.«78oooCOooc^o

The Effect of Bond Rating Agency Announcements743OOOOOI I I I Io o o o o o(D A0) O.2 ECO .>COccoactoEnaCOV§•= 63

744 The Journal of Financepected and unexpected downgrades has the predicted effect for both bond andstock returns. For the unexpected sample, the average excess bond (stock)return is -1.79% (-2.14%), with a ^statistic of -2.31 (-1.84). The mediansfor excess bond and stock returns suggest similar the indicated upgrades, comprid of 38 bonds reprenting 23 differentfirm-obrvations, average excess bond and stock returns are not reliablydifferent from zero. Splitting the sample into expected and unexpected sub-samples reveals some evidence of positive average excess returns for bonds inthe unexpected sample, but not for the stocks. A regression of excess bondreturns on the magnitude of the expectation error again produces a coeffi-cient that is significantly greater than zero with a p-value of 0.019. Resultsfor the noncontaminated Credit Watch upgrades are marginally significantfor the bonds but the results are bad on a small number of obrvations.B. Moody's and Standard and Poor's Bond Rating ChangesWhen the sample of actual rating changes by Moody's and Standard andPoor's is analyzed, the results are difficult to interpret becau of veralapparent inconsistencies; for example, we obrve significantly negativeaverage excess bond and stock returns for downgrades, but weaker positiveaverage excess bond and stock returns for upgrades. Moreover, becau theexpectations model does not sharpen the results as it did in the ca of CreditWatch announcements, we do not prent subsamples of rating changesbad on expected and unexpected classifications. The results obtained fromanalyzing the rating changes are summarized in Table ed results for downgrades are reported in Table V. Announcementsof rating downgrades are associated with statistically significant negativeaverage excess bond and stock returns. For the entire sample of 785 down-grades reprenting 356 firm-obrvations, the average excess bond return is-1.27% (^statistic of -4.33) for the window spanning return which has amedian length of 4 days. The results are inconsistent with Weinstein(1977), Pinches and Singleton (1978), and Wakeman (1978), who do not findsignificant bond price effects associated with rating changes. Even thoughthe average excess bond return includes some large negative individualTable IVSummary of Excess Return Results for Moody's and Standardand Poor's Rating Change AnnouncementsSample of 841 downgrades and 292 upgrades of corporate taxable debt ratings revid byMoody's and Standard and Poor's during ades

Bonds

Entire sample

Noncontaminated

Negative

Zero

Stocks

Negative

Negative

UpgradesBonds

Positive

Positive

StocksZeroZero

The Effect of Bond Rating Agency Announcements•;Vc3"OJM1Ebu0oA<-•ccsrstaflcflchathecaOlbe-T3ingrao^^5>,,rafl-oa>u0)J=L-urn(NcCOCOlT3OdACOas CO CO CO ot1UlUl<1>CJr-.j=3no,>..OJQa;fl S ^O CO -1' (Nc flCCrH CO rH O^ O •—I^ I '^ I ^ jflCO* caC 2 3 "^a; 0)O ,* " QJ^ 6mIIIa4ofl fl.9 >O CO CO COI I I ICmcflXCB .>saCQ3a)flcooQfl '^fl«flra flra«ICO&=: - mW H CO o

746 The Journal of Financereturns,^" the median excess bond return is still -0.45%, and the percentageof firms with positive returns is 41.1% (Z-statistie of -3.37). Over the periodDay 0 to +1, the mean exeess stock return for the firms is -1.52%(^statiBtic of -4.11), and the median return is -0.75%, with only 39.1% ofthe excess stoek returns positive (Z-statistic of -3.99).The remainder of Table V provides information about certain strata ofdowngrades. The (pre-rating change) investment grade bonds reprent 273firm-obrvations and have mean and median exeess bond returns of -0.55%and -0.45%. The stock of the firms los 0.83% on average. The teststatistics indicate that the negative mean excess returns are reliably differ-ent from zero for both bonds and stocks. The 87 firms below investment gradeexperience mean and median excess bond returns of -3.82% and -0,62%,refiecting large loss experienced by 12 of the 87 obrvations. Only 40.2%of the obrvations have positive excess bond returns. The mean and medianexcess stock returns for the firms are -4.22% and -2.22%. Test statisticsindicate the bond and stock returns are significantly different from zero. Theaverage excess bond and stock returns for below investment grade bonds arereliably more negative than for investment grade bonds driven in part by thenumber of firms with bonds losing more than 10%.While the entire sample of downgrades provides strong evidence of bondand stoek price effects, the results bad on noncontaminated obrvationsprovide little evidence that tho bonds experience negative average exeessreturns (mean of -0.37% with a ^-statistic of -1.16). Paradoxically, for thesame obrvations, the stocks experience an average excess return of -1.12%(^statistic of -1.82). Table V also prents parate evidence for noncontam-inated bond rating changes on bonds above or below investment grade. Of the205 noncontaminated obrvations, 176 are bonds which were above invest-ment grade prior to the rating change. The mean and median excess bondreturns are generally clo to zero for investment grade bonds. Nevertheless,firms with downgraded investment grade bonds classified as noncontami-nated experience mean and median excess stock returns of -1.06% (t- andZ-statistics of -2.17 and -3.76). The be low-investment grade downgradesare associated with exeess stock and bond returns of approximately -1.90%,but tho results are not reliably different from zero due to the largestandard deviation of returns (in excess of 6%) and becau of the smallsample size. The difference in mean returns between noncontaminated in-vestment grade and below-investment grade bonds of -1.82% is insignifi-cantly different from is some evidence of a significant positive mean excess bond returnfor upgrades, but there is no evidence of a significant positive mean excessstock return. Table VI contains results for upgrades of 263 bonds of 141different firms. The mean exeess bond return is 0.35% (^statistic of 1.93),The median return is 0.36%, and 62% are positive (Z-statistic of 2.85), The^° Thirteen firms have bonds with window spanning returns below -10%, including one of-63.3%, and five in the range -30% to -45%,

The Effect of Bond Rating Agency Announcementso tolO O ffi O INin ^^ lo ~' •»,2 £ 3ii td rt^ m CS >-30 K 0)oinG03J3O rH O O O OOOOOOO3 0)^C _OI— ^ "flG ajSt•^ •* O- 1-H Tf(N CJ C^ (N ^ o'43

.2 5«OJ^CQ,2 S 3-U

G I £« cufl J»J!i W K13 'Sai « CJfl iS^,2« fl g-Bfl S|iS H «d ci d5-s I a:fc-ectCOG S ^™ y -^cIO ^ inoIO .-H t-.fHaiCOoG13ElUjaO1-.BuoBCBOT5S'I3•'-'0)>C03u3oa0)oCOoou•G 3to"5 «S JnCO .X:=! -; w

748 The Journal of Financemean excess stock return is 0.24% ((-statistic of 1.13), The differencesbetween investment and below-investment grade bond returns are the 263 upgrades, 66 are classified as noncontaminated. The mean andmedian excess return estimates for the noncontaminated upgrades are 0.60%(^statistic is 2.77) and 0.71%, witb 72.4% excess returns being positive(Z-statistic is 2.41). Thus, unlike downgrades, the noncontaminated upgradesample is more supportive of a positive mean bond price reaction to therating change announcement. There is no evidence of positive mean excessstock returns for upgraded bonds classified as . Cross-Sectional Analysis of Excess Bond ReturnsWe estimate multivariate regressions to try to explain cross-ctionalvariation in the window-spanning excess bond returns. Separate regressionsare estimated for downgrades and upgrades, and for contaminated and non-contaminated obrvations. The regi'essions are bad on individual bonds(not firm-obrvations) becau the independent variables are primarily bondcharacteristics, not firm characteristics. Conquently, there is probablysome cross-ctional dependence in the residuals of the regi-ession. Thefollowing variables are included in the regression:a) a cardinal variable reprenting the number of grades changed (newrating less old rating, measured on a scale of 28 for rating AAA to 1 forrating D);b) the magnitude of the yield-to-maturity expectation error (defined previ-ously);c) a dummy variable t equal to one if the rating change is a resolution ofa Standard and Poor's Credit Watch Listing;d) a dummy variable t equal to one if the rating change moves the bondinto or out of investment grade;e) a dummy variable t equal to one if the rating change occurs afterDecember 31, 1980; andf) the time lap since a similar rating change by the other excess bond return should depend positively on the number of gradeschanged by the rerating, and should be clor to zero for resolutions of aCredit Watch Listing. The yield-to-maturity expectation error is designed tocapture the degree of the market's expectation of the rating change—stronger(meaning larger in absolute value) excess bond returns should be obrvedfor reratings that are more unexpected. If there are parate clienteles forinvestment and below-investment grade bonds, rating changes that cross theinvestment/below-investment grade boundary should result in stronger ex-cess returns. The dummy variable for pre- and post-1980 splits the sampleperiod approximately in half becau the fmancial press conjectures thatcompetition forced rating agencies to begin monitoring companies more

The Effect of Bond Rating Agency Announcements 749cloly in the early 198O's.'^ Post-1980 rating change announcements areexpected to result in stronger price effects. The time lap variable testswhether a change by one rating agency has the same impact if it follows asimilar rating change by the other agency. Bond ratings which immediatelyfollow a similar rating by the other agency are not expected to produce strongexcess bond results in Table VII indicate little consistency other than the findingthat the number of grades is significantly related to the excess bond return inthe downgrade and upgrade sample for the contaminated obrvations. In theregression using contaminated downgrade obrvations, the ^statistic on thenumber of grades changed is 16.17 and the coefficient of 1.87 implies thatholding all el constant, the marginal effect of an additional one gradedecline (e.g., from AA + to AA) on average results in a 1.87% decline in thevalue of the bonds. The variable PREV is also significant and has thepredicted sign, indicating less of a price impact when there was a recent priorrating change on the same company by the other agency. The adjusted-ff^ ofthe regression is 37.7%. While the number of grades changed is significant inthe regression using contaminated upgrades, the explanatory power is muchlower (^statistic of 2.32) and no other variable is significant. While theresults on the contaminated obrvations do not reveal whether the excessreturns around the time of bond rating change announcements result fromthe rating change or the contaminating announcement, at a minimum theevidence on # GRADES suggests that rating agencies respond relativelyquickly to some events. In the regressions using noncontaminated obrva-tions, not a single variable is both significant and has the predicted sign.V. Summary and ConclusionExcess bond returns for additions to the Credit Watch List are generallyinsignificant until the expected rating changes are excluded. Examination ofnoncontaminated obrvations which are classified as unexpected, generallyincreas the measured bond and stock price effects of the Credit Watchannouncements. In addition, statistically significant average excess bond andstock returns to announcements of downgi-ades of straight debt are obrved,with less reliable effects for upgrades. Moreover for downgrades, the averageexcess bond returns are stronger for below investment grade bonds than forinvestment grade bonds. When obrvations with concurrent disclosures areeliminated, the negative average excess returns associated with downgradesdisappear, but tho for stocks remain. For upgrades, significant bond priceSee, for example, "Moody's Dominance in Municipals Market is Slowly Being Eroded", WallStreet Journal, November 2, 1981, pp. 1 and 20. In addition, it is likely that the start of Standardand Poor's Credit Watch List was a respon to the long standing charge that rating agenciesreacted slowly to changes in default risk.

750The Journal of Finance

The Effect of Bond Rating Agency Announcements751ftj1dd(1.79)ooo00o(MrHoCOCOdOIOlOCOrfIOo o o oo103o o o oO) ra -r

« co 5:;ra T>.5 ^ g ^ ^^ > -aE ii .y

752 The Journal of Financeeffects are obrved for the noncontaminated sample, but no stock price effectis are some inconsistencies in the results; in particular, we find asym-metric results with respect to rating change downgrades and upgrades. Forexample, we obrve significantly negative average excess bond and stockreturns for downgrades, but weaker positive average excess bond and stockreturns for upgrades. When we examine the noncontaminated samples theasymmetries disappear in the bond results (or if anything, they rever, thatis, there is a greater absolute effect for the upgrades than downgrades). Theasymmetry persists for stocks in the noncontaminated subsample. However,we do find symmetric results for the excess bond returns of indicated down-grades and upgrades for additions to the Credit Watch List, especially whenwe control for prior expectations. Despite the inconsistencies, our overallconclusion is that there are both bond and stock price effects associated withannouncements of additions to tbe Credit Watch list, and with announce-ments of actual rating changes by Moody's and Standard and Poor'NCESGrier, P. and S, Katz, 1976, The differential effects of bond rating changes on industrial andpublic utility bonds by maturity. Journal of Business 49, nicolaou, G. and A, Kalay, 1984, Wealth redistributions or changes in firm value: Ananalysis of returns to bondholders and stockholders around dividend announcements, Jour-nal of Financial Economics 14, un, R. and R. Leftwich, 1986, The effect of bond rating changes on common stock l of Financial Economics 17, , R., L. Brooks and R. Copeland, 1983, The information content of municipal bond ratingchanges: A note. Journal of Finance 38, , S., 1974, The price adjustment process of bonds to rating reclassifications: A test of bondmarket efficiency, Journal of Finance 29, 551-559,Pinches, G. and J- Singleton, 1978, The adjustment of stock prices to bond rating l of Finance 33, i, S., 1986, The effect of preferred stock rating changes on preferred and common stockprices. Journal of Accounting and Economics 8, n, L., 1978, Bond rating agencies and the capital markets. Working paper. University ofRochester, Rochester ein, M-, 1977, The effect of a rating cbange announcement on bond price. Journal ofFinancial Economics 5, 329-350.