The information content of the yield spread about future inflation in South Africa

Transcription

1 SA-TIED Souhern Africa Towards Inclusive Economic Developmen The informaion conen of he yield spread abou fuure inflaion in Souh Africa Manqoba Nshakala and Laurence Harris SA-TIED Working Paper 20 June 2018

2 Abou he projec Souhern Africa Towards Inclusive Economic Developmen (SA-TIED) The SA-TIED programme looks a ways o suppor policy-making for inclusive growh and economic ransformaion in he souhern Africa region, hrough original research conceived and produced in collaboraion beween Unied Naions Universiy World Insiue for Developmen Economics Research (UNU-WIDER), Naional Treasury, Inernaional Food Policy Research Insiue (IFPRI), and many oher governmenal and research organizaions in Souh Africa and is sub-region. A key aspec of he programme is o encourage neworking and discussion amongs people involved in policy processes across he paricipaing organizaions and civil sociey aiming o bridge he gap beween research and policy-making. The SA-TIED programme is funded by he Naional Treasury of Souh Africa, he Deparmen of Trade and Indusry of Souh Africa, and he Delegaion of he European Union o Souh Africa as well as by IFPRI and UNU- WIDER hrough he Insiue s conribuions from Finland, Sweden, and he Unied Kingdom o is research programme.

3 WIDER Working Paper 2018/63 The informaion conen of he yield spread abou fuure inflaion in Souh Africa Manqoba Nshakala 1 and Laurence Harris 2 June 2018

4 Absrac: The proposiion ha inflaion expecaions can be exraced as inflaion predicions from he governmen bond yield curve has been esed, wih parially posiive resuls, using daa from he Unied Saes and European counries. Despie he abundance of empirical sudies of he proposiion, relaively few of hese sudies relae o emerging markes, as mos emerging markes lack bond markes wih he liquidiy, breadh, informaion availabiliy, and range of mauriies ha would permi such sudies. Souh Africa s highly developed capial markes do have such characerisics, warraning his sudy s examinaion of he proposiion s validiy for Souh Africa. Using Souh African ime series daa, we find srong evidence for he proposiion ha he slope of he yield curve, measured as a long- o shor-erm spread, conains informaion on he fuure pah of inflaion. Examining he sub-periods separaed by he adopion, in 2000, of inflaion argeing, we find ha he moneary policy regime shif srenghened he relaionship beween he yield spread and fuure inflaion. The resuls sugges ha he yield spread can be used by policy makers and he privae secor o help forecas inflaion in Souh Africa. Keywords: expecaions, inflaion, moneary policy, Souh Africa, erm srucure, yield spread JEL classificaion: E31, E37, E52, G12 1 Naional Treasury of Souh Africa, Preoria, Souh Africa, corresponding auhor: Manqoba.Nshakala@reasury.gov.za; 2 UNU- WIDER, Helsinki, Finland, and SOAS, Universiy of London, London, UK. This sudy has been prepared wihin he UNU-WIDER projec Souhern Africa Towards inclusive economic developmen (SA- TIED). Copyrigh UNU-WIDER 2018 Informaion and requess: publicaions@wider.unu.edu ISSN ISBN Typescrip prepared by Gary Smih. The Unied Naions Universiy World Insiue for Developmen Economics Research provides economic analysis and policy advice wih he aim of promoing susainable and equiable developmen. The Insiue began operaions in 1985 in Helsinki, Finland, as he firs research and raining cenre of he Unied Naions Universiy. Today i is a unique blend of hink ank, research insiue, and UN agency providing a range of services from policy advice o governmens as well as freely available original research. The Insiue is funded hrough income from an endowmen fund wih addiional conribuions o is work programme from Finland, Sweden, and he Unied Kingdom as well as earmarked conribuions for specific projecs from a variey of donors. Kaajanokanlaiuri 6 B, Helsinki, Finland The views expressed in his paper are hose of he auhor(s), and do no necessarily reflec he views of he Insiue or he Unied Naions Universiy, nor he programme/projec donors.

5 1 Inroducion The abiliy o predic fuure economic condiions, such as business cycle variables and an inflaion measure, using presenly available informaion is valuable for firms, households, and policy makers. In paricular, a forward-looking inflaion-argeing regime such as ha he Souh African Reserve Bank (SARB) formally adoped in 2000 requires sysemaically unbiased esimaors of expeced inflaion for Moneary Policy Commiee ineres rae decisions. In principle, he yield spread he spread beween marke yields on shor- and long-erm governmen deb conains informaion on expecaions of fuure inflaion ha can be exraced as an esimae of hem. The ess repored in his sudy suppor he noion ha he yield curve does, in fac, conain useful informaion for inflaion-argeing moneary policy in Souh Africa. The relaionship a any observaion dae beween he marke yields of fixed-ineres governmen securiies of differing erms o mauriy can be represened by a yield curve. Because he insrumens are muli-period asses (liabiliies) and heir yields relae fuure income sreams o he presen, he level, slope, and shape of he yield curve reflec raional agens marke views of fuure economic condiions. Sudies of US and oher markes have demonsraed ha he yield curve does conain informaion on expeced and acual fuure levels of gross domesic produc (GDP) growh, and can be a lead indicaor of business cycle downurns (Esrella and Mishkin 1998; Rudebusch and Williams 2009). Ohers have demonsraed ha i conains informaion on expeced and acual fuure inflaion (iner alia, Kolán 1999; Kozicki 1997; Mishkin 1989). The summary measure of yield curve slope used in exising research is a measure of he yield spread, he difference beween yields on longer- and shorer-daed deb securiies. Alhough many sudies indicae he inflaion-predicing power of he yield curve in developed counries markes, lile is known of he relaionship in emerging economies. The relaively underdeveloped naure of many such economies fixed-income markes may accoun for ha, bu sudies of Souh Africa s yield curve in is highly developed capial marke are no subjec o such limiaions. Souh African governmen bonds wih nominal value oalling more han R1 rillion and a dense range of mauriies are lised on he Johannesburg Sock Exchange, ogeher wih a large volume of corporae and oher deb. Wih nearly R25 billion raded daily, he governmen bond marke, suppored by he marke-making role of appoined primary dealers and a developed ecosysem including markes in derivaive producs, has high liquidiy. 1 This sudy conribues o he exising body of knowledge wihin he Souh African conex by empirically examining he abiliy of he yield spread o provide informaion abou fuure inflaion. Addiionally, we examine he effec of he moneary policy regime shif associaed wih he adopion of inflaion argeing on he relaionship beween he yield spread and fuure inflaion. This paper is srucured as follows: Secion 2 reviews he key relevan research lieraure. Secion 3 describes he mehodology employed in his paper and he derivaion of our model o be esimaed. Secion 4 deails he daa used, he empirical resuls, and inerpreaion of he findings; Secion 5 concludes

6 2 Lieraure review Empirical sudies of he relaion beween he erm srucure and fuure inflaion are predominanly founded on a combinaion of wo heoreical proposiions. The firs is a model of expecaions. The core expecaions model of he erm srucure underpinning modern research has is roos in lieraure of he 1940s (Hicks 1946; Luz 1940). The long-erm rae is deermined by expeced values of fuure ineres raes and a risk premium. In models of a fricionless marke 2 in which defaul-free securiies are regarded as perfec subsiues disinguished only by heir mauriy, long raes are deermined by raional expecaions of shor raes over he erm o mauriy plus a risk premium as he price for erm-relaed uncerainy of fuure shor raes (Cox e al. 1985; Cuaresma e al. 2005; Shiller and McCulloch 1990). The second is he Fisher Effec, which relaes an asse s nominal ineres rae o is real ineres rae plus expeced inflaion (Fisher 1930). Wih real ineres raes assumed consan, he expeced nominal rae a any mauriy is deermined by he expeced inflaion rae during is erm and he erm-relaed risk premium. An exogenous shock o he rae of inflaion expeced o persis over a given horizon will cause an equivalen shock o he nominal yield on bonds of he corresponding mauriy (Sargen e al. 1973). Sudies of US and European markes suppor a consensus ha he yield spread conains predicive power for near-erm economic aciviy, including he rae of growh of real GDP, bu sudies of he relaion beween he yield spread and fuure inflaion have generaed a variey of resuls. While Shiller e al. (1983), Tzavalis and Wickens (1996), and Mankiw and Summers (1984) poin o weak or no evidence of he yield spread s abiliy o predic inflaion, resuls published by Esrella (2005), Fama (1984), Fama and Bliss (1987), Mishkin (1990a, 1991), Mishkin and Posen (1998), Campbell and Shiller (1991), Kozicki (1997), Ang e al. (2008), Engsed and Tanggaard (1995), and Schich (1999) suppor he hypohesis ha he yield spread does conain informaion abou fuure inflaion for he counries sudied. The differences in resuls are parly due o differences in esimaion echniques, bu generally implies ha he inflaion informaion conen of he yield spread is condiional on facors such as definiions of he variables and segmen of he yield curve, he daa sample, he moneary policy regime and, more generally, he naional economy under sudy (Berk 1998; Esrella 2005; Kozicki 1997; Schich 1999; Tabak and Feiosa 2009). The presen sudy adds evidence from Souh Africa o he exising lieraure. Recen sudies provide evidence for he proposiion ha he yield curve does conain informaion abou fuure levels of economic aciviy and is able o sysemaically predic business cycle downswings (Boha and Keeon 2014; Clay and Keeon 2011; Khomo and Aziakpono 2007). Bu here is lile published evidence of a sysemaic relaion beween Souh Africa s yield curve and fuure inflaion. Using Souh African quarerly daa from 1985 o 1999, Wesso s vecor error correcion esimaes indicae ha long-erm bond yields in ha period are largely driven by inflaion expecaions, bu he finds a seeper yield curve does no necessarily signal a rise in acual fuure inflaion. Wesso (2000) inerpres he resuls as implying ha shifs in long-erm yields could 2 In he absence of perfec subsiuabiliy, due o marke fricions or oher reasons for invesors having srong mauriy preferences, shor and long securiies will rade in markes ha are a leas parially segmened (Cox e al. 1985; Culberson 1957; Vayanos and Vila 2009). 2

7 parly reflec shifs in he credibiliy of he cenral bank s commimen o low inflaion, as, for example, could resul, ouside he sample period, from Souh Africa s adopion of inflaion argeing. Deriving implici inflaion expecaions from forward ineres raes relaed o he erm srucure in a Fisher Effec expecaions model, Reid (2009) found ha inflaion expecaions in Souh Africa have indeed been well anchored by he SARB over he period ( ) afer he esablishmen of inflaion argeing. Many emerging markes, including Souh Africa, have seen a shif in heir moneary policy frameworks since he 1990s, adoping inflaion-argeing regimes. In esimaing he inflaion predicive power of he yield curve, his sudy also examines wheher Souh Africa s adopion of inflaion argeing changed he predicive power of he yield spread. Esrella (2005), Esrella and Mishkin (1997), Bernanke (1990), and Tzavalis and Wickens (1996), for insance, did no only confirm ha he yield spread helps in predicing inflaion, bu also revealed ha he prevailing moneary policy regime and moneary policy sance have pivoal roles o play in his regard. The inuiion is ha he sizes of he reacion parameers are imporan if he moneary policy reacs o deviaions of inflaion from he arge (Reid 2009). Wih Souh Africa having adoped an inflaionargeing regime since 2000, we can expec, a priori, ha his has enhanced he forecasing power of he yield spread, in line wih empirical findings of Lauren (1988), Blinder (1999), Bernanke and Blinder (1992), and Mankiw and Miron (1986). 3 Mehodology Our analysis is carried ou using wo building blocks. The firs is he Fisher equaion (1930), which decomposes he nominal ineres rae ino he (ex-ane) real ineres rae and expeced inflaion rae. This hen implies ha if movemens in nominal ineres raes are primarily driven by flucuaions in expeced inflaion rae raher han changes in real ineres raes, he yield spread will help predic he fuure pah of inflaion (see Fama 1975; Kolán 1999). The second hypohesis is he exisence of raional expecaions. As is common in he lieraure, insead of a coninuous yield curve, we use a shor-erm o en-year yield spread as a measure of he slope of he yield curve. 3 We esimae he inflaion-change forecasing equaion, which suggess ha he yield spread has a predicive power for he fuure pah of inflaion. This forecasing equaion is essenially a regression of he change in he fuure m-period inflaion rae from he n-period inflaion rae m n ( π π ) on he slope of he yield spread m n ( i i ). The predicive power of his equaion is dependen on he esimaed value of he coefficien of he spread, which is normally posiive and increases wih he lengh of he spread (Tzavalis and Wickens 1996). This equaion is expressed as: m n m n m, n π π = m, n + βm, n i i + ε (1) 3 Daily zero coupon yield curves have been calculaed since 2003; since 2012 hey have been consruced and published daily as JSE Zero Coupon Yield Curves using an improved mehod (Johannesburg Sock Exchange 2012). 3

8 3.1 The derivaion of he inflaion-change forecasing equaion Following Mishkin (1990) and laer wriers, including Kozicki (1997) and Kolán (1999), he model is based on he Fisher equaion: where: i m = Eπ m + r m (2) m i is he nominal m-period ineres rae a ime ; m π E is he raional expecaions operaor based on informaion available a ime ; is he inflaion rae beween ime and m; and he (ex-ane) real m-period ineres rae a ime. The observed, acual inflaion rae over he nex m-period can be expressed as he expeced rae of inflaion plus he forecas error of inflaion; ha is: π m = E π m + ε m (3) m r is E π Subsiuing in for ( ) m from (2) ino (3) yields he following: π m = i m r m + ε m (4) To obain an expression for he relaionship beween he slope of he yield curve and he change in he inflaion rae, a similar n-period inflaion rae equaion is subraced from (4), (m > n), yielding he slope of he yield curve: where: π m π n = ( i m i n ) ( r m r n ) + ( ε m ε n ) (5) m n m n m, n π π = m, n + βm, n i i + ε (6) n m = r r (6a) m, n β mn, = 1 (6b), ε m n = ( ε m ε n ) ( u m u n ) (6c) u m = r m r m (6d) u n = r n r m (6e) To ensure consisen esimaes, Mishkin (1990) assumes a consan slope for he real yield curve m n m n hroughou ime such ha ( r r ) = m, n (consan). If his condiion holds, he u u erm in (6c) disappears and he error erm mn, ε in equaion (6) is reduced o m n ( ε ε ). Also, urning o he assumpion of raional expecaions, he forecas error canno be forecased given he informaion a ime. Tha is: m n E ε = E ε =0 and he forecas errors m ε and n ε are hen 4

9 orhogonal o he righ-hand side regressors of equaion (6). A violaion of hese assumpions makes he inerpreaion of he yield curve complicaed and reduces is forecasing power for inflaion. The consancy of he slope of he real ineres rae has been subjec o scruiny. For example, Lowe (1992) assers ha if prices insananeously adjus o moneary policy and are fully flexible, he assumpion of a consan slope of he real ineres rae is plausible and bea (β m,n) should equae o 1. This is also suppored by Frankel and Lown (1991), who claim ha even hough he real ineres rae may be variable in he shor run, i converges o a consan in he long run, ensuring a robus forecasing power. These assumpions herefore ensure ha he ordinary leas-squares (OLS) esimaes of equaion (6) produce consisen esimaes of (β m,n). However, if he price flexibiliy assumpion fails and he real yield curve varies over ime, he nominal yield spread will sill conain informaion abou he fuure inflaion pah, bu i is no longer going o be he opimal predicor because m n ( u u ) is no longer zero. m n ( π π ) This asserion herefore leads us o esing wheher he spread,, predics he change in he n-period rae over he life of he (m n) periods rae. We herefore go on o es for he saisical significance of he coefficien on he nominal ineres rae spread (β m,n) and also invesigae wheher i differs from 1 or no. The saisical rejecion of he null hypohesis (β m,n = 0) leads us o conclude ha he slope of he yield spread conains significan informaion abou he change in he fuure m-period inflaion rae from he n-period inflaion rae. This also implies ha he yield spreads of boh he nominal and real ineres raes do no move one-for-one wih each anoher. On he oher hand, he rejecion of he null hypohesis (β m,n = 1) leads o he conclusion ha he slope of he real yield curve is no consan over ime and hence he nominal yield spread is no an opimal predicor of fuure inflaion. This mehod carries a number of drawbacks. (1) I canno be assumed ha he error erms are independen and idenically disribued. This could emanae from he fac ha he error erm is made up of hree componens real ineres rae, risk premium, and inflaion innovaions, which could cause esimaion bias and heeroscedasiciy. (2) The exisence of sicky prices is anoher drawback, wih Frankel and Lown (1991) and Lowe (1992) arguing ha he assumpion of a consan slope of he real yield curve is overly resricive. This implies ha a long-erm ineres rae is more likely o reflec inflaionary expecaions more accuraely han are shor-erm raes. (3) Since he quarerly inerval of daa is shorer han he forecas horizon, he forecass are overlapping. This is herefore likely o resul in serial correlaion wih he moving average (MA) process. To accoun for hese problems of auocorrelaion and heeroscedasiciy, a Newey Wes correcion procedure is employed (Newey and Wes 1987). We now urn o he empirical analysis for his sudy, in which a descripion of he daa used is given. 4 Daa and empirical resuls 4.1 Daa This analysis uses Souh African quarerly daa on inflaion raes (consumer price index), he governmen s 91-day Treasury bill, and en-year governmen bond yields (he difference beween 91-day and en-year yields being our measure of he yield spread). The analysis spans he period ; daa for he en-year governmen bond rae before July 1988 were no available. The Treasury bill, governmen bond daa, and inflaion daa were all obained from he SARB. 5

10 Uni roo ess using he augmened Dickey Fuller (ADF) and he (Kwiakowski Phillips Schmid Shin) KPSS procedures showed ha long- and shor-erm yields are no saionary in levels bu are all inegraed of order one (Harris 1992; Syczewska 2010). I is, however, noeworhy ha he yield spread is saionary in levels, ha is, inegraed of order zero. Inflaion, on he oher hand, is saionary a he 10 per cen level only when we es for saionariy from 1991Q1; i is herefore assumed ha inflaion is saionary. This guaranees he feasibiliy of he usage of OLS regression in levels. The sandard errors of he OLS regression are, however, likely o be incorrec due o serial correlaion caused by he use of overlapping daa. This implies ha he horizon of he inflaion rae and yields is longer han he observaion inerval. To accoun for his, as in Mishkin (1989) and Kolán (1999), we esimae equaion (6) using he Newey Wes correcion procedure, which accouns for any possible auocorrelaion and heeroscedasiciy in residuals. The analysis is firs conduced on he full sample period (1991Q1 2016Q1); he series is hen spli a February 2000, since a shif in he Souh African moneary policy regime occurred a ha ime. Towards he end of he 1980s and prior o February 2000, he SARB moved o an eclecic inflaion-argeing regime (Van der Merwe 2004). This regime is normally pursued by counries wih high credibiliy in mainaining low and sable inflaion wihou he need for being fully ransparen and accounable. We suspec ha his new framework may have provoked a change in he informaion conen of he yield spread on fuure inflaion. Figure 1 shows he SARB s repo rae, inflaion, and he yield spread (difference beween yields on he en-year Souh African governmen bond and 91-day Treasury bill) series over he period 1988Q3 2016Q1. The shaded bands show hisorical recessions as defined by he SARB business cycles. The figure shows ha he spread ends o decline as he repo rae and inflaion increase, and in some cases as he economy eners he downward phase of he cycle. The spread ends o move in he opposie direcion o he moneary policy cycle, which is in line wih economic heory. Higher shor-erm raes imply lower fuure inflaion and lower shor raes in he fuure. Figure 1: Inflaion, yield spread, and he SARB s repo rae Noe: grey shaded area denoes he downward phases of he business cycle as defined by he SARB. Source: auhors, based on daa from he SARB. Figure 2 shows he movemens of he yield spread agains is individual componens (long-erm and shor-erm yields). This graph shows wheher changes in he yield spread are driven eiher by he longer or shorer end of he yield curve. I is clear in Figure 2 ha shor-erm yields generally move faser han long-erm yields during a moneary policy easing/ighening cycle. Recessionary periods are generally associaed wih invered yield curves, wih he spread becoming negaive. 6

11 Figure 2: 91-day Treasury bill, yield spread, and en-year bond Source: auhors, based on daa from he SARB. Figure 3 shows he SARB s repo rae and he end of moneary policy ighening cycles. These cycles are defined as when one of he following condiions is saisfied: (1) he repo rae is higher han a any ime from welve monhs prior o nine monhs afer and i is 50 basis poins higher han a he beginning of he period; (2) he SARB s repo rae is higher han a any ime from six monhs before o six monhs afer and is 150 basis poins higher han he average a hese poins (Adrian e al. 2010). Figure 3: SARB s repo rae and end of moneary policy cycles Source: auhors, based on daa from he SARB. 7

12 4.2 Empirical resuls and inerpreaion The relaionships beween he yield spread and inflaion are esimaed over varied ime frames and horizons. This sudy ess for he predicive power of he yield spread over 4, 8, 12, 24, 30, and 32 quarers, which came as a consequence of he yield spread showing insignifican resuls prior o he 24h quarer and he predicive power dying ou afer he 32nd quarer across all he ime sample periods. The resuls in beween hese quarers proved no o be differen from hose of he quarers chosen; for insance, he resuls obained from lagging by 16 quarers did no give new insigh o hose produced by lagging by 12 quarers. The resuls broadly agree wih inernaional evidence, which shows ha during he period of inflaion argeing, he spread has subsanial predicive power (Engsed and Tanggaard 1995). Resuls are repored in Tables 1 3. Table 1 shows resuls for he enire period (1990Q3 2016Q1); he sub-periods (1990Q1 1999Q4) and (2002Q1 2016Q1) are repored in Tables 2 and 3 respecively. Panel B includes lagged inflaion as one of he explanaory variables, as in he sudy by Kozicki (1997), which showed ha pas inflaion rae does help in predicing curren inflaion. This exercise is done o see if he yield spread sill explains fuure inflaion over and above he inclusion of pas inflaion and also o improve he regression fi of he daa. Esimaion resuls indicae ha he yield spread has predicive power in he Souh African daa, paricularly for he full sample period and for he inflaion-argeing regime sub-period. The yield spread has a subsanial predicive power beween 2000Q1 and 2016Q1, as repored in Table 3. These resuls end o suppor he raional expecaions heory. The resuls for he period 1990Q3 o 1999Q4, however, do no show any predicive power of he spread in forecasing inflaion. Table 1: Esimaes of inflaion change Panel A: = + i i + ; period: Sepember 1990 o March 2016 m m n m, n m, n m, n h Spread Horizon (in quarers) mn, SSE (0.94) (0.97) (0.63) (0.57) (0.50) (0.51) -sa 7.63* 7.55* 10.66* 11.09* 11.20* 10.55* 91-day Treasury bill, en-year bond βm,n SSE (0.37) (0.35) (0.24) (0.21) (0.20) (0.22) -sa ** 3.24* Adjused R Wald es a βm,n = Wald es a βm,n =

13 Panel B: = + i i + + ; period: Sepember 1990 o March 2016 m m n m m, n m, n m, n h 4 Spread Horizon (in quarers) mn, SSE (1.24) (0.94) (0.98) (1.39) (0.96) (0.94) -sa 2.66* 3.32* 3.67* 4.20* 5.80* 6.02* βm,n SSE (0.33) (0.21) (0.19) (0.23) (0.20) (0.23) -sa ** 3.09* Inflaion SSE (0.14) (0.15) (0.15) (0.19) (0.16) (0.17) -sa 3.56* 3.59* 2.84* Adjused R day Treasury bill, en-year bond Wald es a βm,n = Wald es a βm,n = Noe: *, **, and *** denoe significance a he 1, 5, and 10 per cen criical level, respecively. Sandard errors calculaed for Newey Wes adjused covariaion marices. a Wald es: p-values of F-saisic repored Source: auhors, based on SARB daa. The resuls in Table 1 for boh Panel A and B are roughly he same and show ha he spread has no o lile predicive power for he shor o medium erm (ha is, for 4 24 quarers (1 6 years)). In mos cases he β m,n coefficiens have he wrong signs and are insignifican. However, for forecas horizons over 30 quarers (ha is, 7.5 years), he coefficiens of he yield spread are significan and close o 1. The fi of he regression is very poor, as shown by very low adjused R 2, wih Panel B s relaively beer han hose of Panel A. There is a significan improvemen in he adjused R 2 for forecas horizons of over 30 quarers. This sudy also presens resuls of he Wald es. This es assumes he null hypohesis of β m,n = 0 or 1 and fails o rejec a 1, 5, and 10 per cen significance levels. The resuls of he Wald es: β m,n coefficien being significanly differen from 0 is no rejeced for forecass over 32 quarers. For quarer 30, however, he Wald es rejecs boh he cases ha he β m,n coefficien is significanly differen from 1 and 0; his is a borderline scenario. On a broader view, hese resuls can be inerpreed as follows. The yield spread for he Souh African daa does conain useful informaion abou he fuure pah of inflaion for forecas horizons above 30 quarers. Even hough no precisely, he resuls of his paper broadly compare well wih hose of Mishkin (1990a, 1991), Miskin and Posen (1998), Campbell and Shiller (1991), Kozicki (1997), Esrella (2005), Ang e al. (2008), and Schich (1999) in ha he erm srucure of ineres raes provides lile informaion on he shor end of he yield curve abou fuure changes of inflaion. As noed by Kolán (1999), his could be as a resul of grea variabiliy of real ineres raes in he shor run, which could eclipse he inflaion expecaions componen. Figure 4 shows how well he forecas according o 32 quarers racks he acual daa on inflaion in Souh Africa. 9

14 Figure 4: Inflaion forecass according o an eigh-year lag spread for he enire sample period Noe: grey shaded areas denoe he downward phases of he business cycle as defined by he SARB. Source: auhors, based on SARB daa. Swiching o he analysis of he regime swich effec, Table 2 presens resuls for regime 1 (1990Q1 1999Q4), in which he SARB did no formally arge inflaion. This analysis is aimed a uncovering he usefulness of he inflaion-argeing regime in anchoring inflaion expecaions pos Esrella (2005), Gurkaynak e al. (2006), and Reid (2009) all concurred ha he relaionship beween he yield spread and he fuure inflaion evoluion are broadly influenced by he moneary policy regime. In Table 2 Panels A and B, we see ha he β m,n coefficiens are all insignifican across all forecas horizons, and mos of hem even carry wrong signs. Even hough he adjused R 2 are slighly higher compared o Table 1, he Wald es ha hypohesized ha he β coefficiens significanly differ from 0 canno be rejeced for all forecas horizons. As such, hese resuls are in agreemen wih hose of Esrella, Gurkaynak e al., and Reid, ha in he absence of an inflaion-argeing regime, inflaion expecaions are no anchored and hence he yield spread ends o have weak or no predicive power abou fuure inflaion. 10

15 Table 2: Esimaes of inflaion change Panel A: = + i i + ; sub-period: Sepember 1990 o December 1999 sample m m n m, n m, n m, n h Spread 91-day Treasury bill, en-year bond mn, Horizon (in quarers) SSE (1.32) (1.33) (0.93) (0.59) (0.71) (0.86) -sa 7.62* 7.61* 10.23* 13.00* 9.88* 8.32* βm,n SSE (0.54) (0.46) (0.28) (0.31) (0.36) (0.33) -sa Adjused R Wald es a βm,n = Wald es a βm,n = Panel B: sample = + i i + + ; sub-period: Sepember 1990 o December 1999 m m n m m, n m, n m, n h 4 Spread Horizon (in quarers) mn, SSE (2.14) (2.11) (2.81) (2.12) (2.00) (2.58) -sa * 6.29* 5.65* βm,n day Treasury bill, en-year bond SSE (0.37) (0.33) (0.40) (0.35) (0.41) (0.29) -sa Inflaion SSE (0.18) (0.19) (0.27) (0.29) (0.24) (0.35) -sa 3.54* 3.92* 2.25** 1.90* 2.89** 2.82** Adjused R Wald es a βm,n = Wald es a βm,n = Noe: *, **, and *** denoe significance a he 1, 5, and 10 per cen criical level respecively. Sandard errors calculaed for Newey Wes adjused covariaion marices. a Wald es: p-values of F-saisic repored. Source: auhors, based on SARB daa. Table 3 presens he resuls of he inflaion-argeing regime employed by he SARB in February In Panel B we lag inflaion by wo insead of four; his is because curren inflaion wihin an inflaion-argeing regime is more forward-looking and hence if lagged by four quarers i fails o explain curren inflaion and is insignifican. The resuls for Panel A and B differ slighly from each oher in ha he inclusion of he inflaion componen means he spread has a predicive power as 11

16 early as from quarer 24 going forward. The resuls of Table 3 also prove o be more robus relaive o he previous resuls. Table 3: Esimaes of inflaion change Panel A: = + i i + ; sub-period: February 2000 o March 2016 sample m m n m, n m, n m, n h Spread 91-day Treasury bill, en-year bond mn, Horizon (in quarers) SSE (1.04) (0.64) (0.47) (0.70) (0.42) (0.46) -sa 5.60* 8.72* 11.80* 8.36* 11.93* 10.36* βm,n SSE (0.42) (0.21) (0.22) (0.24) (0.17) (0.22) -sa * 4.20* Adjused R Wald es a βm,n = Wald es a βm,n = Panel B: = + i i + + ; sub-period: February 2000 o March 2016 sample m m n m m, n m, n m, n h 2 Spread 91-day Treasury bill, enyear bond mn, Horizon (in quarers) SSE (0.61) (0.58) (0.62) (0.54) (0.48) (0.66) -sa 3.07* 3.96* 3.88* 3.57* 4.94* 4.61* βm,n SSE (0.26) (0.18) (0.21) (0.12) (0.14) (0.22) -sa *** 4.08* 3.01* Inflaion SSE (0.10) (0.12) (0.12) (0.12) (0.10) (0.13) -sa 6.12* 4.85* 4.64* 5.48* 4.96* 2.74* Adjused R Wald es a βm,n = Wald es a βm,n = Noe: *, **, and *** denoe significance a he 1, 5, and 10 per cen criical level respecively. Sandard errors calculaed for Newey Wes adjused covariaion marices. a Wald es: p-values of F-saisic repored Source: auhors, based on SARB daa. The β m,n coefficiens of Table 3 Panel A are insignifican up o quarer 26 (6.5 years); however, from quarer 27, he yield spread coefficien is significan and hus explains he fuure changes of inflaion. Ineresingly, including a wo-lag inflaion erm, he spread sars becoming significan as early as quarer 24 (six years) and has correc signs across all forecas horizons. The adjused R 2 12

17 are very low for Panel A prior o quarer 30, indicaing a poor fi of he regressions. There is, however, a sligh improvemen from quarer 30 going forward. Panel B, on he oher hand, shows relaively higher adjused R 2 across all horizons; his shows ha including a lagged inflaion erm as one of he regressors significanly improves he fi. In Panel A, he hypohesis ha β m,n significanly differs from 0 is no rejeced from quarer 30 onwards. For Panel B, we can conclude ha since we canno rejec he null (β m,n = 0) for quarers 4, 8, and 12, he slope of he real yield curve is no consan over ime and hence he nominal yield spread is no an opimal predicor of fuure inflaion. The resuls of Table 3 can herefore be broadly inerpreed as follows. The Souh African yield spread conains useful informaion abou he fuure evoluion of inflaion only for forecas horizons from 24 quarers ahead. Figure 5 shows ha he forecas using quarer 30 racks he acual daa on inflaion more closely han he one showed in Figure 2, which confirms and complemens he findings of Esrella, Gürkaynak e al., and Reid. Table 3 furher shows ha inflaion expecaions are well anchored under he inflaion-argeing regime, which implies ha long-erm yields provide useful informaion abou he fuure of inflaion. Figure 5: Forecas of 30-quarer yield spread compared o acual inflaion Noe: grey shaded areas denoe he downward phases of he business cycle as defined by he SARB. Source: auhors, based on SARB daa. Table 4 shows he fi of forecas according o 32 quarers; he forecas racks he acual daa of inflaion in Souh Africa quie closely. In Table 4 he comparison of he forecas equaion esimaes using 30 quarers (7.5 years) spread lag for boh he enire period and he inflaion-argeing regime. The resuls for he inflaion-argeing regime are more robus compared o hose of he enire period. For insance, he adjused R 2 for he enire period regression (0.12) is much lower han ha of he inflaion-argeing regime (0.51). This shows ha he regression for he inflaion-argeing regime is a beer fi and produces beer resuls. Table A1 shows he esimaes for all he lags, and he resuls for he inflaion-argeing regime end o ouperform hose of he enire period across horizons. 13

18 Table 4: Forecas comparison beween he enire sample and he inflaion-argeing regime Esimaes 2000Q1 2016Q1 30-quarer lag 1991Q1 2016Q1 βm,n sa 4.08* 2.37** Adjused R AIC SIC Noe: *, **, and *** denoe significance a he 1, 5, and 10 per cen criical level respecively. Source: auhors, based on SARB daa. 5 Conclusion The evidence provided in his paper suggess ha he slope of he yield curve is useful in forecasing he fuure pah of inflaion. These resuls pu forward ha he yield spread should no be used o forecas he near-erm inflaion rae (ha is, 23 quarers or less). The yield spread is, however, useful for predicing changes in fuure inflaion over 24 quarers in he Souh African case. The resuls are much sronger for he inflaion-argeing regime, confirming he credibiliy of moneary policy in anchoring long-erm inflaion. These findings are in harmony wih hose of Reid (2009), affirming ha he SARB has been able o sabilize and manage inflaion expecaions hrough is ransparen and credible moneary policy. The resuls of his paper are consisen wih he heory ha moneary policy has direc effecs on he shor end (real ineres raes) of he yield curve because of price sickiness. Long-erm yields, however, more closely mimic he behaviour of inflaion expecaions han do shor-erm raes as prices are fully flexible in he long run. Berk (1988), however, poins ou ha cauion should be exercised by policy makers when using he yield spread as a ool o forecas inflaion. This is because many facors can shif he ends of he yield curve and a face value may promp moneary auhoriies o respond inappropriaely. 14

19 References Adrian, T., A. Esrella, and H.S. Shin (2010). Moneary Cycles, Financial cycles and he Business Cycle. Saff Repor 421. New York: Federal Reserve Bank of New York. Ang, A., G. Bekaer, and M. Wei (2008). The Term Srucure of Real Raes and Expeced Inflaion. The Journal of Finance, 63(2): Berk, J.M. (1998). The Informaion Conen of he Yield Curve for Moneary Policy: A Survey. De Economis, 146(2): Bernanke, B. (1990). On he Predicive Power of Ineres Raes and Ineres Rae Spreads. Working Paper Cambridge, MA: NBER. Bernanke, B.S., and A.S. Blinder (1992). The Federal Funds Rae and he Channels of Moneary Transmission. The American Economic Review, 82(4): Blinder, A.S. (1999). Cenral Banking in Theory and Pracice. Cambridge, MA: MIT Press. Boha, F., and G. Keeon (2014). A Noe on he (Coninued) Abiliy of he Yield Curve o Forecas Economic Downurns in Souh Africa. Souh African Journal of Economics, 82(3): Campbell, J.Y., and R.J. Shiller (1991). Yield Spreads and Ineres Rae Movemens: A Bird's Eye View. The Review of Economic Sudies, 58(3): Clay, R., and G. Keeon (2011). The Souh African Yield Curve as a Predicor of Economic Downurns: An Updae. African Review of Economics and Finance, 2(2): Cox, J.C., J.E. Ingersoll Jr, and S.A. Ross (1985). A Theory of he Term Srucure of Ineres Raes. Economerica: Journal of he Economeric Sociey, 53(2): Cuaresma, J.C., E. Gnan, and D. Rizberger-Grünwald (2005). The Term Srucure as a Predicor of Real Aciviy and Inflaion in he Euro Area: A Reassessmen. Review of World Economics, 141(2): Culberson, J.M. (1957). The erm srucure of ineres raes. Quarerly Journal of Economics, 71(4): Engsed, T., and C. Tanggaard (1995). The Predicive Power of Yield Spreads for Fuure Ineres Raes: Evidence from he Danish Term Srucure. The Scandinavian Journal of Economics, 97(1): Esrella, A. (2005). Why Does he Yield Curve Predic Oupu and Inflaion?. The Economic Journal, 115(505): Esrella, A., and F.S. Mishkin (1997). The Predicive Power of he Term Srucure of Ineres Raes in Europe and he Unied Saes: Implicaions for he European Cenral Bank. European Economic Review, 41(7): Esrella, A., and F.S. Mishkin (1998). Predicing US Recessions: Financial Variables as Leading Indicaors. Review of Economics and Saisics, 80(1): Fama, E.F. (1975). Shor-Term Ineres Raes as Predicors of Inflaion. The American Economic Review, 65(3): Fama, E.F. (1984). The Informaion in he Term Srucure. Journal of Financial Economics, 13(4):

20 Fama, E.F., and R.R. Bliss (1987). The Informaion in Long-Mauriy Forward Raes. The American Economic Review, 77(4): Fisher, I. (1930). The Theory of Ineres. New York: Macmillan. Frankel, J.A., and C.S. Lown (1991). An Indicaor of Fuure Inflaion Exraced from he Seepness of he Ineres Rae Yield Curve Along is Enire Lengh. Working Paper Cambridge, MA: NBER. Gürkaynak, R., A. Levin, and E. Swanson (2006). Does Inflaion Targeing Anchor Long-Run Inflaion Expecaions? Evidence from Long-Term bond Yields in he US, UK, and Sweden. Working Paper San Francisco, CA: Federal Reserve Bank of San Francisco. Harris, R.I. (1992). Tesing for Uni Roos Using he Augmened Dickey Fuller Tes: Some Issues Relaing o he Size, Power and he Lag Srucure of he Tes. Economics Leers, 38(4): Hicks, J.R. (1946). Value and Capial, 2nd ediion. Oxford: Clarendon Press. Johannesburg Sock Exchange. (2012). The JSE Zero-Coupon Yield Curves. Mehodology Documen. Johannesburg: Johannesburg Sock Exchange. Khomo, M.M., and M.J. Aziakpono (2007). Forecasing Recession in Souh Africa: A Comparison of he Yield Curve and Oher Economic Indicaors. Souh African Journal of Economics, 75(2): Kolán, V. (1999). The Term Srucure of Ineres Raes and Fuure Inflaion. Easern European Economics, 37(5): Kozicki, S. (1997). Predicing Real Growh and Inflaion wih he Yield Spread. Economic Review- Federal Reserve Bank of Kansas Ciy, 82(4): 39. Lauren, R.D. (1988). An Ineres Rae-Based Indicaor of Moneary Policy. Economic Perspecives, 12(1): Lowe, P.W. (1992). The Term Srucure of Ineres Raes, Real Aciviy and Inflaion. Research Discussion Paper Sydney: Economic Research Deparmen, Reserve Bank of Ausralia. Luz, F.A. (1940). The Srucure of Ineres Raes. The Quarerly Journal of Economics, 55(1): Mankiw, N.G., and J.A. Miron (1986). The Changing Behavior of he Term Srucure of Ineres Raes. The Quarerly Journal of Economics, 101(2): Mankiw, N.G., and L.H. Summers (1984). Do Long-Term Ineres Raes Overreac o Shor- Term Ineres Raes?. Working Paper Cambridge, MA: NBER. Mishkin, F.S. (1989). The Informaion in he Longer Mauriy Term Srucure Abou Fuure Inflaion. Working Paper Cambridge, MA: NBER. Mishkin, F.S. (1990). Wha Does he Term Srucure Tell Us Abou Fuure Inflaion?. Journal of Moneary Economics, 25(1): Mishkin, F.S. (1991). A Muli-Counry Sudy of he Informaion in he Shorer Mauriy Term Srucure abou Fuure Inflaion. Journal of Inernaional Money and Finance, 10(1): Mishkin, F.S., and A.S. Posen (1998). Inflaion Targeing: Lessons from Four counries. Working Paper Cambridge, MA: NBER. Newey, W.K., and K.D. Wes (1987). Hypohesis Tesing wih Efficien Mehod of Momens Esimaion. Inernaional Economic Review, 28(3):

21 Reid, M. (2009). The Sensiiviy of Souh African Inflaion Expecaions o Surprises. Souh African Journal of Economics, 77(3): Rudebusch, G.D., and J.C. Williams (2009). Forecasing Recessions: The Puzzle of he Enduring Power of he Yield Curve. Journal of Business & Economic Saisics, 27(4): Sargen, T.J., D. Fand, and S. Goldfeld (1973). Raional Expecaions, he Real Rae of Ineres, and he Naural Rae of Unemploymen. Brookings Papers on Economic Aciviy, 1973(2): Schich, S. (1999). The Informaion Conen of he German Term Srucure Regarding Inflaion. Applied Financial Economics, 9(4): Shiller, R.J., and J.H. McCulloch (1990). The Term Srucure of Ineres Raes. Handbook of Moneary economics, 1: Shiller, R.J., J.Y. Campbell, K.L. Schoenholz, and L. Weiss (1983). Forward Raes and Fuure Policy: Inerpreing he Term Srucure of Ineres Raes. Brookings Papers on Economic Aciviy, 1983(1): Syczewska, E.M. (2010). Empirical Power of he Kwiakowski Phillips Schmid Shin Tes. Working Paper 45. Warsaw: Deparmen of Applied Economerics, Warsaw School of Economics. Tabak, B.M., and M.A. Feiosa (2009). An Analysis of he Yield Spread as a Predicor of Inflaion in Brazil: Evidence from a Waveles Approach. Exper Sysems wih Applicaions, 36(3): Tzavalis, E., and M.R. Wickens (1996). Forecasing Inflaion from he Term Srucure. Journal of Empirical Finance, 3(1): Van der Merwe, E. (2004). Inflaion Targeing in Souh Africa. Occasional Paper 19. Preoria: Souh African Reserve Bank. Vayanos, D., and J. Vila (2009). A Preferred-Habia Model of he Term Srucure of Ineres Raes. Working Paper Cambridge, MA: NBER. Wesso, G.R. (2000). Long-Term Yield Bonds and Fuure Inflaion in Souh Africa: A Vecor Error-Correcion Analysis. Quarerly Bullein: Souh African Reserve Bank, 216:

22 Appendix Table A1: Forecas comparison beween he enire sample and he inflaion-argeing regime: comparing he enire period and he inflaion-argeing period Period Horizon (in quarers) Q1 2016Q1 βm,n sa *** 4.08* 3.01* Adjused R AIC SIC Q1 2016Q1 βm,n sa ** 3.09* Adjused R AIC SIC Noe: *, ** and *** denoes significance a he 1, 5, and 10 per cen criical level respecively. Source: auhors, based on SARB daa. 18