Interest Rate Surprises and Stock Prices

Transcription

1 Ineres Rae Surprises and Sock Prices Beno J. Lobo The Universiy of Louisiana a Lafayee Dep. of Economics and Finance P.O. Box Lafayee, LA B_Lobo@Louisiana.edu This version: Spring 00 The auhor hanks Dan Thornon, Gerald Whiney and Lewis Gale for helpful commens on an earlier draf.

2 Ineres Rae Surprises and Sock Prices Absrac This paper examines he impac of unexpeced changes in he federal funds arge on sock prices from 988 o 00. Measures of ineres rae surprises are consruced from survey daa and changes in he 3-monh T-bill yield. I find ha surprises associaed wih decreases in he arge cause sock prices o rise significanly. Surprises associaed wih increases in he arge increase sock marke volailiy on he announcemen day, wih volailiy revering o pre-surprise levels on he day afer he announcemen. This volailiy paern is only eviden since 994. An implicaion is ha concerns abou immediae disclosure causing persisen and heighened sock marke volailiy migh be misplaced. Keywords: Moneary policy; Fed funds rae arge; ineres rae surprises; survey daa; sock marke volailiy; asymmeric reacions; EGARCH model JEL Classificaions: E44/G/E65/G8/C/C5

3 Ineres Rae Surprises and Sock Prices. Inroducion I is widely believed ha moneary policy has a significan impac on asse prices. There is also subsanial casual evidence ha he acions of he Federal Reserve (Fed) affec he expecaions of sock marke paricipans. Since Alan Greenspan was appoined Chairman in 987, he Fed has implemened moneary policy by making discree adjusmens o is arge for he federal funds rae. Given ha a change in he funds rae arge is a moneary policy acion, i migh be expeced ha sock prices would reac o new informaion embodied in such changes. Jensen, Johnson and Bauman (997) and Thorbecke and Alami (994), cie several sudies ha have shown ha changes in Fed policy influence equiy prices. Furhermore, porfolio allocaion rules are ofen predicaed on moneary policy, and prominen invesmen advisors rely on Fed policy as an inpu in heir porfolio selecion process. This paper aemps o quanify he marginal impac of surprise changes in he fed funds arge on sock prices and on sock marke volailiy from January 988 hrough January 00. Ineres rae changes affec asse prices o he exen ha such changes convey new informaion abou eiher shor- or long-run moneary policy objecives. In paricular, ineres rae changes can impac equiy prices hrough wo channels, i.e. by affecing he rae a which he firm s expeced fuure cash flows will be capialized, and by alering expecaions abou fuure cash flows (Smirlock and Yawiz, 985). Furher, one would expec ha if boh capializaion raes and expecaions abou fuure cash flows are impaced by ineres raes, hese effecs would work in he same direcion. In oher words, an increase

4 in ineres raes will cause sock prices o decline and a decline in ineres raes will cause sock prices o rise. An implicaion of he efficien markes/raional expecaions lieraure is ha asse prices reac only o new fundamenal informaion. Pearce and Roley (985) and Hardouvelis (987) were among he early researchers o find ha sock prices reac significanly (and someimes, primarily) o moneary news or unanicipaed moneary policy. However, while he former focused on money supply surprises and announced discoun rae and surcharge rae changes, he laer examined unexpeced changes in free bank reserves. Much of he previous research on he effecs of moneary policy on sock prices has used changes in he discoun rae as a proxy for changes in moneary policy. Smirlock and Yawiz (985) and Chen e al. (999) decomposed discoun rae changes ino echnical (expeced) and non-echnical (unexpeced) changes based on press releases from he Fed afer he rae change. Boh sudies concluded ha unexpeced discoun rae changes generaed a significan announcemen effec, and ha all of he sock price adjusmen o he rae change ook place rapidly, by he nex day afer he announcemen a he laes. Chen e al. (999) also examined he effec of discoun rae changes on he volailiy of sock prices and on rading volume. They found ha unexpeced discoun rae changes conribued o higher, hough shor-lived, volailiy and rading volume. During Greenspan s enure as Chairman, he Fed has adoped a policy of argeing inflaion by making discree adjusmens o he fed funds rae arge. Thus, moneary policy Hardouvelis (987) poined ou ha he inverse relaionship beween sock prices and ineres rae changes could also be raionalized in erms of money supply surprises. The negaive (posiive) reacion of sock prices (ineres raes) o money supply surprises can be explained in erms of he following wo hypoheses. The Expeced Real Ineres Rae hypohesis claims ha sock prices decline because he real componen of nominal ineres raes is expeced o increase, hereby increasing he discoun rae a which fuure cash flows are capialized and also because higher ineres raes affec real oupu adversely, hereby reducing fuure operaing 3

5 is more aply racked by changes in his arge. Thorbecke (997) sudied he effec of acual changes in he fed funds rae arge on he Dow Jones Indusrial and Composie averages over he 4 hours brackeing he news of he arge change. He found ha fed funds arge changes were inversely relaed o sock prices and ha moneary policy exered large effecs on ex ane and ex pos sock reurns. Thorbecke and Alami (994) go similar resuls using a sample of fed funds arge changes from 974 o 979. Neiher sudy aemped o isolae he unexpeced componen of he arge change, nor examine he impac of arge changes on sock marke volailiy. This sudy adds o he lieraure in several ways. To begin wih, he Thorbecke (997) and Thorbecke and Alami (994) sample is exended in examining he impac of changes in he fed funds rae arge on sock prices from January 988 hrough January 00. The anicipaed and unanicipaed componens of he arge change are separaed using wo differen echniques. The firs approach is based on survey daa on marke paricipans expecaions of he funds rae arge. The unexpeced componen (or surprise) is measured as he difference beween he acual/announced change in he arge and he survey median. This approach, as has been demonsraed in several previous sudies, offers a direc assessmen of he impac of news conained in he arge change announcemen on sock prices. The use of survey daa circumvens many of he problems associaed wih he exan approaches o isolaing he unanicipaed componens of ineres rae changes made by cenral banks (Hardy, 998). The second echnique used o disil he surprise componen of arge changes uilizes informaion from he shor-end of he Treasury securiies marke o cash flows. The Expeced Inflaion hypohesis claims ha sock prices decline because he inflaion premium in nominal ineres raes increases, which decreases he afer-ax real dividends. 4

6 gauge he marke s expecaions of he likely change in he fed funds rae arge, drawing on he finding ha he 3-monh Treasury bill yield fully reflecs funds rae arge changes (Bonser-Neal e al., 998). Esimaes from using he enire arge change as in Bonser-Neal e al. (998) and Thornon (998) are also repored for purposes of comparison. I examine he impac on sock prices of he change in he Fed s disclosure policy concerning arge changes. Unil i s February 994 meeing, he Federal Open Marke Commiee (FOMC) followed he pracice of issuing a policy direcive several weeks afer he meeing in which a arge change had been approved. This delayed announcemen policy was based on he idea ha immediae disclosure would commi he Fed o a course of acion, hereby impairing moneary policy flexibiliy. Addiionally, delayed disclosure, i was argued, would preven undesirable volailiy-generaing announcemen effecs in financial markes. Criics counered ha he Fed s failure o convey is inenions o he marke in a imely fashion increased uncerainy and volailiy. Financial marke paricipans aemped o decipher he Fed s inenions by monioring he federal funds rae, he Fed s daily open marke operaions, and Fed leaks o he press. Thornon (996) found ha while an announcemen effec was eviden in marke ineres raes wih or wihou immediae disclosure, he magniude of he announcemen effec appeared o be smaller wih immediae disclosure. Moreover, he announcemen effec was srung ou over several days when he marke was lef o decipher policy changes on is own (i.e. pre-994). I es he impac of unexpeced arge changes on he sock marke and he speed of adjusmen of sock prices o such surprises in he wo disclosure regimes. Previous sudies using discoun rae changes (Smirlock and Yawiz (985); Chen e al. (999)) aemped o isolae news associaed wih he rae change announcemen based on ex pos press releases which raders would no have had privy o ex ane. 5

7 There is evidence ha sock prices reac differenly o bad and good news (Koumos, 998, 999). This asymmeric reacion is supporive of he view ha invesors have a higher aversion o downside risk, and herefore reac faser o bad news. Unexpeced changes in he funds rae arge, being measured as he difference beween he announced value of he arge change and he expeced change in he arge, generae posiive (negaive) surprises eiher when he Fed raised raes more (less) han expeced or lowered raes less (more) han expeced. Consequenly, posiive surprises could be consrued as bad news and negaive surprises could be viewed as good news for sock prices. This sudy examines, for he firs ime, he asymmeric sock price responses o posiive and negaive ineres rae surprises. In recen imes, he source of marke volailiy has araced a grea deal of aenion from academic researchers, praciioners and regulaors. No much is known abou how funds rae arge changes in general affec sock marke volailiy. This sudy aemps o quanify he impac of FOMC disclosure policy and differenial (posiive/negaive) ineres rae surprises on sock marke volailiy and enables us o commen on he effec of Fed policy acions on shor-erm sock marke volailiy. To do so, an Exponenial GARCH (EGARCH) model is employed ha capures he peculiar feaures of high frequency sock price daa, namely ime-varying volailiy, volailiy clusering, he leverage effec, excess kurosis and skewness. This empirical approach faciliaes he simulaneous analysis of he effecs of ineres rae changes on he mean and variance of sock prices. The res of his paper is organized in hree secions and a conclusion. Secion conains a descripion of he daa; I describe he empirical sraegy in secion 3; Secion 4 conains an analysis of he empirical findings. 6

8 . Daa Daes and changes in he federal funds rae arge are from Bonser-Neal e al. (998) for , and from he Federal Reserve Bank of New York from 990 hrough January 00. There were 60 changes in he federal funds rae arge from January, 988 o January 3, 00 (see he Appendix for deails). Where wo days were cied for a arge change dae, e.g. November 3-4, 990, he laer of he wo was used as he announcemen day. Of he 60 arge changes, 6 were increases and 34 were cus. The expecaions daa were obained from MMS Inernaional. 3 For he sample sudied here, surveys were conduced every week (ypically, on Fridays). Till July 3, 997, survey paricipans were asked o forecas he level of he Fed funds arge (among oher variables) during he upcoming wo-week Fed saemen period. Thus, here were ypically wo surveys conduced for each saemen period, i.e. he firs survey a he beginning of he saemen period and he second survey one-week ino he saemen period. Since July 997, surveys have been conduced relaive o forhcoming FOMC meeing daes. There were anywhere from five o eigh surveys corresponding o each FOMC meeing dae. The median response from he survey neares he arge change dae was used o gauge marke expecaions concerning arge changes. Through o he February 994 arge change, here were 3 occasions on which survey daes coincided wih arge change daes. To he exen ha prior o 994, he Fed neiher disclosed he arge change immediaely, nor regularly enaced arge changes a scheduled FOMC meeings, i is unlikely ha survey respondens could have known wih cerainy he naure or iming of he arge change. Consequenly, in hose cases where he survey dae coincided wih he arge change dae 3 MMS Inernaional is a subsidiary of Sandard and Poor s and conducs surveys of he views of marke paricipans on a wide variey of daa releases, boh domesic and inernaional. 7

9 (excluding he February 994 arge change), ha survey was deemed o be he mos proximae o he arge change. The unexpeced (or surprise) componen of he arge change was compued as he difference beween he acual change in he arge and he MMS survey median. In he pos- July 997 period, any arge changes made ouside of scheduled FOMC meeing daes (i.e. iner-meeing) were reaed as surprises. This is because since 994 he Fed rouinely adjused he arge a scheduled FOMC meeings. There were hree excepions: April 8, 994, Ocober 5, 998 and January 3, 00. These were included as surprises on he basis ha he marke had come o expec arge changes o be made a FOMC meeings, and hence hese iner-meeing changes were surprises a leas in erms of iming, if no in magniude as well. The survey-based echnique yielded 34 oal surprises. 4 The proxy for he sock marke used in his sudy is he S&P 500 index, which is a broader collecion of socks relaive o he 30-sock Dow Jones Indusrial Average used by previous auhors. Daa on he 3-monh Treasury bill yield were sourced from he Federal Reserve Board of Governors websie. Table conains preliminary saisics on he raw series of daily percenage changes in he S&P 500 Index. The summary saisics sugges ha while he sample mean is saisically differen from zero, he series is skewed and lepokuroic relaive o he normal disribuion. The rejecion of normaliy can be parially aribued o ineremporal dependencies in he momens of he series. There is evidence of serial dependence in he condiional mean and of condiional heeroskedasiciy, i.e. he Ljung-Box (LB) saisic is highly significan for boh he price change and squared price change series. In fac, he LB 4 For he period afer July 997, survey responses around FOMC meeing daes were examined for surprises resuling from FOMC inacion. No addiional surprises were found in he daase. 8

10 saisic for he squared series poins o significan higher momen dependencies in he daa. This is consisen wih he auoregressive condiional heeroskedasiciy and volailiyclusering characerisic of high-frequency asse price daa (Bollerslev, Chou and Kroner, 99). The summary saisics sugges ha a GARCH-class of model would be appropriae, along wih an error disribuion ha allows for greaer kurosis han he normal. ********** Inser Table *********** 3. Mehodology esimaed: Under he mainained hypohesis of efficien markes, he following general model is P = α + i= 0 β i i + ε ln( h ) = θ + i= 0 ω i i + m j = γ j ln( h j ) + n k = φ g ( z k k )......( ) where, P is he percenage change in he closing values of he S&P 500 index from day - o day 5, are he federal funds rae arge surprises, measured in percenage poins, and ε is a random error. I use wo echniques o separae expeced and unexpeced componens of he arge changes. The firs uilizes survey daa o measure marke expecaions, wih surprises compued as he difference beween he announced change in he arge and he MMS Inernaional survey median. The second measure of arge surprises, following Bonser-Neal e al. (998), uilizes informaion from he shor-end of he Treasury marke o gauge he marke s expecaion 9

11 concerning likely changes in he fed funds rae arge. Accordingly, he expeced componen of arge changes is measured as he change in he 3-monh T-bill yield from he day afer he previous arge change o he day before he curren arge change. The surprise componen was consruced as he difference beween he acual arge change and his measure of he expeced change in he arge. The echnique yielded 60 oal surprises. I bears noing ha using he change in he 3-monh T-bill yield as a measure of marke expecaions is an indirec approach compared o he more direc echnique of surveying marke paricipans abou heir expecaions. The 3-monh T-bill yield has been shown o reflec fully he change in he arge afer he change occurred (Bonser-Neal e al. 998). However, he yield iself migh be affeced by a variey of facors peculiar o he Treasury securiies marke (e.g. order flow, new aucions, ec.) ha could make i an imperfec gauge of marke expecaions regarding an upcoming arge change. This inuiion is borne ou by he fac ha all of he (60) arge changes were deemed o be surprises using his mehodology, compared o 34 when using he survey-based echnique. The condiional variance (h ) is an EGARCH process ha allows for condiional heeroskedasiciy, volailiy clusering, and he leverage effec, i.e. he phenomenon whereby pas negaive shocks increase curren volailiy more han do pas posiive shocks. The parameer γ capures he persisence in volailiy. Sandard likelihood raio lag selecion mehods suggesed an EGARCH(,) process was appropriae (i.e. m=n=). Saionariy of he volailiy process requires ha γ <. The sandardized innovaion, z, is defined as ε /h such ha a posiive (negaive) z implies an unexpeced increase (decrease) in sock prices 5 We use he percenage change in he index o be consisen wih Thorbecke (997) and Thorbecke and Alami (994). Resuls from using he logarihmic difference in he index (as in Chen e al., 999) yielded similar resuls. 0

12 and, g z ) = δz + [ z E z ] ( is an asymmeric funcion of z. The erm in brackes relaes lagged sandardized innovaions o volailiy in a symmeric fashion, while δ relaes sandardized innovaions o volailiy in an asymmeric manner. This specificaion permis he volailiy a ime o depend boh on he size and sign of he pas sandardized residuals or shocks. 6 The general specificaion in () enables us o es for any announcemen day effecs of unexpeced arge change announcemens. Theory suggess ha sock prices should be inversely relaed o ineres rae changes. Accordingly, one would expec β 0 o be negaively signed. The model also allows us o examine wheher he arrival of new public informaion, affecs sock marke volailiy, via a significance es of he parameer ω 0. To he exen ha any news migh reasonably be expeced o increase volailiy, he sign on ω 0 is expeced o be posiive. I also examine he speed wih which prices incorporae new informaion in he arge change announcemen by assessing he pos-announcemen reacion of sock prices. Previous sudies have shown ha sock prices reac very quickly o news abou moneary policy. Consequenly, only a one-day pos-announcemen window (i.e. a one-period lag srucure) is used o examine he speed of adjusmen of sock prices. The added advanage of using a narrow ime window is ha i maximizes our abiliy o capure he full sock marke response o he arge surprise, while reducing he possible effecs of confounding facors lef ou of he regression. A failure o rejec he null hypohesis ha he coefficiens of he lags of are no differen from zero via sandard likelihood raio ess, would be 6 The EGARCH model is more general han he sandard GARCH model in ha i allows innovaions of differen signs o have a differenial impac on volailiy and allows bigger shocks o have a larger impac on

13 anamoun o concluding ha all of he sock price adjusmen o he arge change occurs on he announcemen day. The parameric specificaion for he condiional probabiliy densiy used here is he power exponenial or Generalized Error Disribuion (GED), which includes he normal disribuion as a special case. The funcional form of he GED densiy funcion can be wrien as:.5.5 υ / f ( µ, h, υ) = υ / [ Γ(3/ υ)] [ Γ(/ υ)] (/ h )exp{ [ Γ(3/ υ) / Γ(/ υ)] e / h 0 υ where, µ represens he condiional mean, and h he condiional variance of sock prices. (.) is he gamma funcion and v is a scale parameer conrolling he shape of he disribuion and is esimaed endogenously. For v=, he densiy funcion is ha of he normal disribuion. The double exponenial (v=) and he uniform disribuion (v64) are also nesed wihin his specificaion. For values of v <, he GED is more kuroic han he normal disribuion. Maximum likelihood esimaes of he parameers were obained using he Bernd e al. (974) algorihm. } 4. Empirical Resuls All models were esimaed wih a Newey-Wes correcion for auocorrelaion using en lags. Each model was also esimaed wih conrol variables for he day-of-he-week anomaly. However, hese erms were consisenly insignifican and were dropped from he final model. Oher model specificaion experimens included he use of MA(), AR() and ARMA(,) erms. None was saisically significan. volailiy han does he sandard GARCH model. Also, by modeling he logarihm of he condiional variance, i

14 4.. Impac of federal funds rae arge surprises on sock prices The esimaes in Table are of () using hree differen measures of ineres rae surprises. The firs row (Model A) conains esimaes of he impac on sock values of arge surprises, based on he separaion of unexpeced and expeced arge changes using survey daa. I can be seen ha unexpeced ineres rae changes generae a significan announcemen day reacion in he sock marke. In paricular, esimaes of β 0 sugges ha a % surprise in he fed funds arge resuls in a sock price reacion of almos.5%. These findings are qualiaively consisen wih previous work in he area (Thorbecke (997); Thorbecke and Alami (994); Chen e al. (999); Smirlock and Yawiz (985)). The resricion ha he lag of in he mean equaion (β ) is saisically insignifican canno be rejeced, suggesing ha he sock price reacion o he surprise is compleed on he announcemen day. I presen new evidence concerning he impac of arge change surprises on sock marke volailiy. Likelihood raio ess rejec he resricion ha he lag of in he volailiy equaion (ω ) is insignifican. Esimaes of ω 0 and ω sugges ha volailiy increases on he announcemen day, and hen revers o pre-surprise levels on he day afer he announcemen. This finding is consisen wih he idea ha ineres rae surprises increase volailiy iniially, bu ha he informaion in such news is quickly assimilaed ino prices. Alernaive measures of he dispersion of he survey forecass, namely he range beween he maximum and minimum survey response, he range beween he 75 h and 5 h percenile response, and he survey sandard deviaion, were included as addiional is no necessary o resric parameer values o avoid negaive variances. 3

15 regressors in (). In each case, he measure of dispersion was saisically insignifican in boh he condiional mean and volailiy equaions. In he second row of Table (Model B), he esimaes are based on he separaion of unexpeced and expeced changes in he funds rae arge using changes in he 3-monh T- bill yield. While he volailiy esimaes are qualiaively similar o hose repored in he firs row, he measured announcemen day effec on he condiional mean of sock price changes is smaller han when using survey-based surprises. The sriking difference here, however, is ha he lag of (β ) is significan, suggesing ha he sock price reacion o his measure of surprise is no compleed on he announcemen day. This could be aribued o he fac ha changes in he T-bill yield do no measure marke expecaions abou arge changes as well as surveys do. By way of comparison, esimaes from using he enire arge change as he measure of arge surprises as in Bonser-Neal e al. (998) and Thornon (998) are presened in he hird row of Table (Model C). While he findings are qualiaively similar o hose repored in he firs wo rows, he esimaes of β 0, ω 0 and ω are again much smaller compared o hose in he firs row (i.e. where arge surprises are measured wih expecaions daa). This is supporive of he noion behind he efficien markes hypohesis ha sock prices reac primarily o news. By clubbing ogeher he expeced and unexpeced componens of he arge change, we fail o capure he full sock marke response o he ineres rae surprise. The esimaes for volailiy persisence/clusering (γ) and he leverage effec (φ,δ) are consisenly significan indicaive of he fac ha he EGARCH model suiably accommodaes hose feaures of he daa. Model diagnosics, available on reques, are based on he sandardized residuals (z). The residuals are i.i.d. - hey are linearly and nonlinearly 4

16 independen. Also, χ -ess rejec he hypohesis ha he esimaed degrees of freedom parameer, v, is equal o, suggesing ha he residuals are condiionally non-normally disribued, and ha deparures from normaliy in he raw daa canno be aribued solely o emporal firs and second momen dependencies. ******** Inser Table ********* 4.. Change in Disclosure policy and he impac of arge surprises on sock prices As previously noed, he Fed alered i s disclosure policy wih respec o arge changes a he February 994 FOMC meeing. To examine he impac of he change in he Fed s disclosure policy, he following model is esimaed: P ln( h = α + ) = θ + B ( D i W ( D i ) i ) + i + β D i ω D i i + ε i + γ ln( h ) + φ ([ z E z ] + δz )...( ) where, D is a binary dummy variable ha akes a value of one when arge surprises occurred in he new (immediae) disclosure period (pos-993), and zero oherwise. The B and W coefficiens measure he sock price response o ineres rae surprises from 988 hrough 993 (i.e. in he old, delayed disclosure period), whereas he β and ω coefficiens measure he sock price response o ineres rae surprises from 994 hrough January 3, 00 (i.e. in he new, immediae disclosure period). Table 3 conains only esimaes of he parameers of ineres, namely he B s, β s, W s and ω s. The consans and EGARCH erms (γ, φ, δ) were very similar o hose repored in Table, and were suppressed for purposes of clariy. The esimaes of () provide ambiguous insighs ino how ineres rae surprises have impaced mean changes in 5

17 sock prices in he wo disclosure regimes. Surprises, measured using survey daa (Model A), appear o have impaced sock prices in boh disclosure periods. However, surprises, measured using changes in he T-bill yield (Model B) and he enire arge change (Model C), appear o have had no impac on sock price changes in he immediae disclosure era relaive o he delayed disclosure regime. There is some evidence ha he longer response ime observed in Table for such surprises (i.e. significan lag coefficien of in he mean equaion) is driven by arge changes in he pre-994 delayed disclosure era, consisen wih he idea ha rae changes in his period migh have aken longer o decipher. The volailiy esimaes of () in Table 3 unambiguously sugges ha ineres rae surprises in he immediae disclosure era have had a much more significan impac on sock marke volailiy, relaive o surprises in he delayed disclosure era. In fac, while sock marke volailiy does no appear o have been significanly affeced by arge changes prior o 994, volailiy in he immediae disclosure period has ended o increase significanly on he announcemen day, and hen rever o pre-surprise levels on he day afer he announcemen. The implicaion of his finding is ha any concerns ha he Fed migh have had abou he disquieing effec of promply disclosing arge changes are probably misplaced. While volailiy picks up on he announcemen day (as migh be expeced when any news arrives), i does no persis and revers o pre-surprise levels on he day afer he arge change. ********** Inser Table 3 ********* 4.3. Impac of posiive and negaive arge surprises on sock prices Esimaes of () imply a symmeric impac of posiive and negaive surprises on sock values. However, here is evidence ha sock prices respond asymmerically o good 6

18 and bad news (Koumos, 998, 999). As previously noed, posiive (negaive) surprises arise eiher when he Fed raised raes more (less) han expeced or lowered raes less (more) han expeced. Consequenly, posiive surprises could be consrued as bad news and negaive surprises could be viewed as good news for sock prices. In (3) below, ineres rae surprises are pariioned ino posiive and negaive surprises o es for asymmeric reacions of sock prices following hese announcemens. P = α + β ([ z E z ] + δz )...( 3) ln( h ) = θ + ω POS i POS i + ω NEG i NEG i + γ ln( h ) + φ POS, i, POS, i +, β NEG, i, NEG, i + ε, The esimaes of (3) are conained in Table 4. The hypohesis ha boh ypes of surprises have a symmeric impac on sock values is easily rejeced. In fac, only negaive surprises (good news) have impaced sock prices, as measured by β NEG,0, while posiive surprises (bad news) have no had a saisically significan effec on mean changes in sock prices. 7 8 Moreover, posiive surprises (ω POS,0 ) have a greaer impac on sock marke volailiy han negaive surprises, consisen wih he idea ha bad news increases volailiy more han good news. The lags of he posiive surprise variable in he volailiy equaion are significan in all hree models, suggesing ha he volailiy reversal noed earlier is driven by posiive surprises. In sum, he resuls sugges ha while good news has a significan 7 Several explanaions of his resul are forhcoming. One, suggesed by Koumos (998, 999) and Lobo (000), is ha sock marke paricipans discoun bad news more vigorously and more quickly (perhaps, in anicipaion of or ahead of he arrival of he news) hereby reducing he observed impac of he news when i does arrive. Anoher possibiliy relaes o he fac ha sock marke paricipans perceive he Fed o be a preempive and effecive inflaion figher (Lobo, 00). Consequenly, increases in he arge are greeed wih a non-negaive reacion, on average. 8 The lags of negaive surprises are significan in Models B and C in line wih he discussion regarding Tables and, making conclusions abou he speed of adjusmen o such surprises ambiguous. 7

19 effec on he condiional mean of sock price changes, bad news has a significan impac on he condiional variance of sock price changes. ********* Inser Table 4 *********** 4.4. Disclosure policy versus ype of surprise: which maers more? The conclusions from he previous wo subsecions, namely ha disclosure policy ambiguously affeced he condiional mean and unambiguously affeced he condiional variance of sock prices, while differen ypes (posiive/negaive) of ineres rae surprises affeced sock prices differenly, lead logically o a es for asymmeric sock price reacions across disclosure regimes. To do so, he following model is esimaed: P ln( h + φ = α + ( D ) ( D ) ) = θ + ω ji j= i= 0 ([ z E z ] + δz )...( 4) j= i= 0 β ji j, i XU + j, i j= i= 0 + β j= i= 0 ji ω D ji D j, i + ε j, i + γ ln( h ) where, D is a dummy variable ha akes a value one in he immediae disclosure period from //94 hrough //0, and zero everywhere else. The j-subscrips index he ype of shock, wih j= being posiive surprises and j= being negaive surprises. The reader is cauioned ha he small samples of posiive and negaive surprises in cerain cases resul in some loss of precision in he esimaions. 9 Noneheless, he esimaes are quie informaive when viewed across he differen measures of surprises used. 9 In he delayed disclosure period, here were increases and 6 cus in he funds rae arge. Posiive and negaive surprises in his period based on survey daa numbered 8 and 7, respecively. The numbers of posiive and negaive surprises based on changes in he T-bill yield were 4 and 4, respecively. In he immediae disclosure period, here were 4 increases and 8 cus in he funds rae arge. Based on survey daa, here were 5 and 4 posiive and negaive surprises, respecively, in his period. Based on changes in he T-bill yield, here were posiive and negaive surprises each. 8

20 The esimaes of (4), which are repored in Table 5, reveal ha he impac of ineres rae surprises on he condiional mean were driven by he sign or ype of he surprise, and no he disclosure policy adoped by he Fed. In paricular, negaive surprises increased sock prices in boh disclosure regimes. This resul, which is robus across all hree measures of surprises, helps resolve he ambiguiy wih respec o esimaes of he condiional mean in Table 3. However, he impac of arge change surprises on he condiional variance of sock prices appears o have been driven by he Fed s disclosure policy. In paricular, sock marke volailiy was affeced by posiive surprises only in he immediae disclosure regime. Again, his resul is robus across all hree measures of surprises. The overall conclusion hen is ha he ype of ineres rae surprise is a more imporan influence on mean changes in sock prices regardless of he manner by which marke paricipans gaher his informaion, while he naure of disclosure is a more imporan facor affecing sock marke volailiy. ********* Inser Table 5 *********** 5. Conclusion This paper sough o examine he impac of unanicipaed changes in he federal funds rae arge from January 988 o January 00 on he condiional mean and variance of sock prices. The empirical analysis uilized hree differen measures of surprises: wo echniques o separae expeced and unexpeced changes in he funds rae arge, using survey daa and changes in he 3-monh T-bill yield, and he enire arge change. Consisen wih much of he general news lieraure, his sudy finds ha ineres rae surprises have an announcemen day effec on sock values. Moreover, surprises ha are 9

21 consruced from survey responses yield more precise esimaes of he impac of arge changes on sock prices. Alernaive measures of surprises bring some ambiguiy o conclusions regarding he speed wih which sock prices adjus o he ineres rae news. This sudy provides new evidence peraining o he impac of ineres rae surprises on sock marke volailiy. In paricular, such news increases volailiy on he even day (as migh be expeced of any unanicipaed even), bu hen revers o pre-surprise levels on he day afer he ineres rae change. Moreover, his volailiy paern is only associaed wih surprises in he immediae disclosure period beginning in February 994. One implicaion of his finding is ha Fed concerns abou immediae disclosure causing persisen and heighened sock marke volailiy migh be misplaced. Ineresingly, he analysis shows ha posiive and negaive surprises have an asymmeric effec on he mean and volailiy of sock prices. In paricular, negaive arge surprises (good news) impaced he condiional mean of sock prices, while posiive surprises (bad news) impaced he condiional variance of sock prices in a saisically significan manner. A es of hese asymmeric effecs across disclosure regimes revealed ha while he ype of ineres rae surprise influenced mean changes in sock prices, he naure of disclosure was a more imporan facor affecing sock marke volailiy. Several useful exensions of his research are forhcoming. This and oher sudies have shown ha moneary policy has a significan impac on sock values. The issues of how and why such policy acions affec sock prices need o be more direcly addressed. Perhaps, an analysis of how expecaions regarding ineres rae changes evolve could shed ligh on rader behavior and micro-marke dynamics and on he issue of moneary ransmission mechanisms o sock prices. Addiionally, since January 000, he FOMC 0

22 began he pracice of publicly announcing any immediae change in he sance of policy (bias), as well as he commiee s assessmen of he balance of risks beween heighened inflaion pressure and economic weakness over he foreseeable fuure (Thornon and Wheelock, 000). Addiional research ino how such changes in disclosure policy impac asse prices is warraned, perhaps when more daa becomes available. A fruiful area of relaed research could address he issue of he spillover effecs of U.S. ineres rae shocks on foreign markes.

23 Appendix: Changes in he Federal Funds Rae Targe Targe changes and daes hrough o 989 are from Bonser-Neal e al. (998). The remaining arge changes and daes are from he Federal Reserve Bank of New York. FFRT is he Fed Funds Rae Targe. DW is day of he week. T3m yield is he change in he 3-monh T-Bill yield from he day afer he previous arge change o he day before he curren arge change. # Survey daes are coinciden wih arge change daes. Bold are inermeeing arge changes since February 994. Targe Change Dae DW Level of FFRT (%) Change in FFRT (%) T3m Yield (%) Targe Change Dae DW Level of FFRT (%) Change in FFRT (%) T3m Yield (%) /8/88 TH /3/9# F //88 TH /3/9 TH /30/88 W /6/9 W /9/88 M /6/9# F /5/88 W /0/9# F //88 W /9/9 TH //88# F //9# TH /9/88 T /4/9# F /0/88 TH /4/94# F //88 T //94 T /5/88 TH /8/94 M /5/89 TH /7/94 T /4/89 T /6/94 T /4/89# F /5/94 T /6/89 T //95 W /7/89# F /6/95 TH /7/89 TH /9/95 T /9/89 TH /3/96 W /6/89 M /5/97 T /0/89 W /9/98 T /3/90# F /5/98 TH /9/90 M /7/98 T /4/90 W /30/99 W /7/90# F /4/99 T /9/90 W /6/99 T /9/9 T //00 W //9# F //00 T /8/9# F /6/00 T /30/9 T /3/0 W /6/9 T /3/0 W

24 References Bernd, E.K., H.B. Hall, R.E. Hall and J.A. Hausman, 974. Esimaion and inference in nonlinear srucural models, Annals of Economic and Social Measuremen 4, Bollerslev, T., R. Chou and K. Kroner, 99. ARCH modeling in finance, Journal of Economerics 5, Bonser-Neal, C., V.V. Roley and G.H. Sellon, Jr., 998. Moneary policy acions, inervenion and exchange raes: A reexaminaion of he empirical relaionships using federal funds rae arge daa, Journal of Business 7 (), Chen, C. R., N. Mohan and T. Seiner, 999. Discoun rae changes, sock marke reurns, volailiy, and rading volume: Evidence from inraday daa and implicaions for marke efficiency, Journal of Banking and Finance 3, Hardy, D.C., 998. Anicipaion and surprises in cenral bank ineres rae policy, IMF Saff Papers 45 (4), Hardouvelis, G.A., 987. Macroeconomic informaion and sock prices, Journal of Economics and Business 39, Jensen, G., R. Johnson and W. Bauman, 997. Federal reserve moneary policy and indusry sock reurns, Journal of Business, Finance and Accouning 4(5), Koumos, G., 998. Asymmeries in he condiional mean and he condiional variance: Evidence from nine sock markes, Journal of Economics and Business 50, Koumos, G., 999. Asymmeric price and volailiy adjusmens in emerging Asian sock markes, Journal of Business, Finance and Accouning 6 (&), Lobo, B., 000. Asymmeric effecs of ineres rae changes on sock prices, The Financial Review 35, Lobo, B., 00. The fed and financial markes: The message in federal funds rae arge changes. Working Paper, The Universiy of Louisiana a Lafayee. Pearce, D.K. and V.V. Roley, 985. Sock prices and economic news, Journal of Business 58(), Smirlock, M. and J. Yawiz, 985. Asse reurns, discoun rae changes, and marke efficiency, Journal of Finance Vol. XL No. 4, Thorbecke, W., 997. On sock marke reurns and moneary policy, Journal of Finance Vol. LII No.,

25 Thorbecke, W. and T. Alami, 994. The effec of changes in he federal funds rae arge on sock prices in he 970s, Journal of Economics and Business 46, 3-9. Thornon, D. L. and D.C. Wheelock, 000. A hisory of he asymmeric policy direcive, Federal Reserve of S. Louis Review, Sepember/Ocober 000, -5. Thornon, D. L., 998. Tess of he marke s reacion o federal funds rae arge changes, Federal Reserve of S. Louis Review, November/December 998, Thornon, D. L., 996. Does he fed s new policy of immediae disclosure affec he marke? Federal Reserve of S. Louis Review, November/December 996,

26 Table Descripive saisics on daily percenage changes in he S&P 500 index: January, 988 o February, 00 (N=346) Daily changes ( P ) were compued as percenage changes in he closing values of he S&P 500 index from day - o day. The p-values are for he es ha he esimaes are no differen from zero. LB(n) and LB (n) are he Ljung-Box saisics for P and P, respecively, disribued Π wih n degrees of freedom, where n is he number of lags. Measure Esimae p-value Mean Sd. Deviaion Skewness Kurosis LB(0) LB (0)

27 Table The impac of federal funds rae arge surprises on sock prices Parameer esimaes are from he following model, where P is he percenage change in he closing values of he S&P 500 index, and are ineres rae surprises. The models are disinguished by he measure of ineres rae surprise (XU) used, as follows: Model A: are he unexpeced arge changes compued by subracing from he acual arge change he expeced change in he arge, measured using survey median daa Model B: are he unexpeced arge changes compued by subracing from he acual arge change he difference beween he 3-monh T-Bill yield on he day before he arge change and he yield on he day afer he previous arge change. Model C: are he enire arge change Sandard errors are in parenheses. LRS and LRS are likelihood raio es saisics, each disribued χ wih degree of freedom. P = α + i= 0 β i ([ z E z ] + δz )......( ) ln( h ) = θ + ω i i + γ ln( h ) + φ i= 0 i + ε Model Parameer Esimaes α x 00 β 0 β θ x 00 ω 0 ω γ φ δ LRS Null: β =0 LRS Null: ω =0 A (.3) *** -.46 (0.75) *** -0.6 (0.6) 0.6 (0.09) *.66 (.) ** -3. (.8) *** 0.99 (0.0) *** 0.05 (0.0) *** (0.3) *** *** B (.30) *** -.03 (0.59) * -.3 (0.43) *** 0.08 (0.).3 (0.8) * -.39 (0.8) * 0.99 (0.0) *** 0.06 (0.0) *** -0.5 (0.3) *** 6.37 ** 3.87 ** C (.30) *** (0.44) * -0.8 (0.8) 0.8 (0.) *.53 (0.64) ** -.73 (0.63) *** 0.99 (0.0) *** 0.05 (0.0) *** (0.3) *** *** *** indicaes saisical significance a he 0.0level ** indicaes saisical significance a he 0.05 level * indicaes saisical significance a he 0.0 level 6

28 Table 3 Disclosure policy and he impac of federal funds rae arge surprises on sock prices The parameer esimaes are for he model described below. See also noes o Table. D is a dummy variable ha akes a value in he period from //94 hrough //0, and zero everywhere else. This break in he sample is moivaed by he change in he FOMC s disclosure policy afer 993. Unil i s February 994 meeing, he FOMC followed he pracice of issuing a policy direcive several weeks afer he meeing in which a arge change had been approved. Since February 994, however, he FOMC has publicly announced changes in curren moneary policy immediaely upon making hem. The sub-sample esimaes are designed o capure he differenial impac, if any, of arge surprises on sock prices in he wo disclosure regimes. The esimaes of he B s and W s are for he sub-sample from 988 hrough 993, whereas esimaes of he β s and ω s are for he sub-sample from 994 hrough 00. Sandard errors are in parenheses. LRS and LRS are likelihood raio es saisics, each disribued χ wih degrees of freedom. P ln( h = α + ) = θ + B ( D i W ( D i ) i ) + i + β D i ω D i i + ε i + γ ln( h ) + φ ([ z E z ] + δz ) ( ) The models are disinguished by he measure of ineres rae surprise (XU) used, as follows: Model A: are he unexpeced arge changes compued by subracing from he acual arge change he expeced change in he arge, measured using survey median daa ; Model B: are he unexpeced arge changes compued by subracing from he acual arge change he difference beween he 3-monh T-Bill yield on he day before he arge change and he yield on he day afer he previous arge change; Model C: are he enire arge change. Model Selec Parameer Esimaes B 0 B β 0 β W 0 W ω 0 ω LRS (H 0 : B =β =0) LRS (H 0 : W =ω =0) A -.64 (0.90) * -.37 (0.96) -.78 (.47) ** 0.99 (0.7) 0.07 (.75) (.74) 4.55 (.53) * (.57) ** *** B -.00 (0.58) * -.7 (0.49) ** -.66 (.46) (.) 0.80 (.8) (.7).95 (.6) ** -.96 (.5) ** 6.69 ** 8.5 *** C (0.44) ** -.09 (0.40) *** -. (0.90) 0.47 (0.54) 0. (.0) -0.3 (.07).77 (.0) ** -3. (.) *** 7.8 *** 93. *** *** indicaes saisical significance a he 0.0 level ; ** indicaes saisical significance a he 0.05 level ; * indicaes saisical significance a he 0.0 level 7

29 Table 4 The asymmeric impac of posiive and negaive arge surprises on sock prices The parameer esimaes are for he model described below, designed o es for asymmeric sock price responses o posiive and negaive surprises. POS are posiive arge change surprises or bad news, which arise eiher when he Fed raised raes more han expeced or lowered raes less han expeced. NEG are negaive arge change surprises or good news, which arise eiher when he Fed raised raes less han expeced or lowered raes more han expeced. Sandard errors are in parenheses. LRS and LRS are likelihood raio es saisics, each disribued χ wih degrees of freedom. P = α + β ([ z E z ] + δz )...( 3) ln( h ) = θ + ω POS, i POS, i + ω NEG, i NEG, i + γ ln( h ) + φ POS, i POS, i + β NEG, i NEG, i The models are disinguished by he measure of ineres rae surprise () used, as follows: Model A: are he unexpeced arge changes compued by subracing from he acual arge change he expeced change in he arge, measured using survey median daa ; Model B: are he unexpeced arge changes compued by subracing from he acual arge change he difference beween he 3-monh T-Bill yield on he day before he arge change and he yield on he day afer he previous arge change; Model C: are he enire arge change. + ε Selec Parameer Esimaes Model A Model B Model C A. Condiional Mean β POS, (.54) (.0) (0.7) β POS, 0.7 (0.96) (0.74) 0.35 (0.45) β NEG, (.6) ***.34 (0.7) *.5 (0.6) ** β NEG, 0.93 (0.76).4 (0.5) **.4 (0.46) *** LRS (H 0 ): β POS, = β NEG, = *** 7.88 *** B. Condiional Variance ω POS, (.99) **.8 (.6).69 (0.85) ** ω POS, (.03) *** -.73 (.65) * -.00 (0.86) ** ω NEG,0.54 (.67).4 (0.9).4 (.03) ω NEG, -.88 (.64) -.6 (0.93) -.45 (.0) LRS (H 0 ): ω POS, = ω NEG, = 0 5. *** *** *** indicaes saisical significance a he 0.0level ** indicaes saisical significance a he 0.05 level * indicaes saisical significance a he 0.0 level 8

30 Table 5 Disclosure policy versus ype of surprise: wha maers more? The parameer esimaes are for he model described below, designed o es for asymmeric sock price responses o posiive and negaive surprises across disclosure regimes. D is a dummy variable ha akes a value in he immediae disclosure period from //94 hrough //0, and zero everywhere else. The j-subscrips index he ype of shock, wih j= being posiive surprises (POS) and j= being negaive surprises (NEG). Sandard errors are in parenheses. LRS and LRS are likelihood raio es saisics disribued χ wih 4 degrees of freedom, for he es ha he lags in he condiional mean and variance, respecively, are collecively equal o zero. P ln( h = α + + γ ln( h ) = θ + j= i= 0 ) + β ji ( D ) ( D ) ω ji j, i + j= i= 0 j= i= 0 φ ([ z E z ] + δz )...( 4) j, i + j= i= 0 β ji ω D ji D j, i The models are disinguished by he measure of ineres rae surprise () used. See Table 4 for deails. + ε j, i Selec Parameer Esimaes Model A Model B Model C A. Condiional Mean Delayed Disclosure Period β POS, (.56) -.8 (0.95) ** -.9 (0.90) β POS, 0.4 (.63) 0.60 (.33) -0.5 (0.88) β NEG, 0.78 (0.99) * 0.8 (0.67) 0.6 (0.58) β NEG,.74 (.7).58 (0.5) ***.58 (0.49) *** Immediae Disclosure Period β POS, (.58).0 (.85) -.86 (.77) β POS,.8 (.38) -.80 (0.64) *** 0.54 (0.65) β NEG, (.89) *** 4.76 (.86) *.5 (0.78) *** β NEG, -0.8 (0.94) 0.7 (3.7) 0.7 (0.8) LRS **.66 ** B. Condiional Variance Delayed Disclosure Period ω POS,0.06 (3.79) -0.7 (.75) -.4 (.3) ω POS, -.64 (3.7) 0.77 (.69).8 (.0) ω NEG, (.) 0.8 (.38) 0.85 (.05) ω NEG, -0.6 (.9) -.0 (.37) (.04) Immediae Disclosure Period ω POS, (4.08) 5.66 (.6) ** 4. (.40) * ω POS, (4.9) * (.56) *** (.48) ** ω NEG, (7.7).85 (4.68).55 (.36) ω NEG, -.3 (7.7) -. (4.49) -.59 (.38) LRS 7.7 *** 3.6 ** *** ***, ** and * indicae saisical significance a he 0.0, 0.05 and 0.0 levels, respecively. 9