TERM STRUCTURE OF INTEREST RATE AND MACROECONOMIC VARIABLES: THE TURKISH CASE

Transcription

1 TERM STRUCTURE OF INTEREST RATE AND MACROECONOMIC VARIABLES: THE TURKISH CASE Huseyin KAYA Bahcesehir Universiy Ciragan Cad. Besikas/Isanbul-Turkey Absrac This sudy includes very iniial analyses of ongoing research which invesigae he relaionship beween erm srucure of ineres rae and macro variables in Turkey. Iniial.ndings indicae ha corresponding relaion has srucural break around 2002 which coincides wih new mon- eary policy namey in.aion argeing. In pre2002 period role of macro- economic variables in yield curve is limied however in pos 2002 period macro variables play very crucial role in erm srucure of ineres rae. We found ha in.aion and exchange rae are wo majar macro variables ha deermine he shape of yield curve. Key Words: erm srucure, ineres rae, laen facor, yield curve JEL Classificaion: G1, E4, C5 1. INTRODUCTION One of he mos influenial works ha bring new breah o erm srucure modeling was proposed by Ang and Piazzesi (2003) (AP hereafer). AP proposed a erm srucure model wih inflaion and economic growh facors ogeher wih laen variables. In he model i is no allowed o bidirecional link i.e. here is no feedback for macro variables and conemporaneous correlaion of macro and laen facors are zero. Afer consrucing an affine erm srucure model wih boh observed and unobserved facors, AP esimae he model by using wo-sep consisen esimaion procedure. In he firs sep he macro dynamics and coefficiens of shor rae are esimaed by OLS. Holding esimaed parameer fixed, in second sep he res of all parameers are esimaed by numeric maximizaion. AP found ha macro variables namely inflaion and growh explain significan porion of (%85) movemens in shor and middle par of he yield curve bu explain only abou %40 percen of movemens in long end of he yield. Comparing he laen facors from previous lieraure, significan par of he level and slope facor are aribued o macro facors especially o inflaion. However AP's wo sage esimaion mehod relies on he assumpion ha shor rae does no affec he macro variables. Hördahl e al. (2006) redress shorcomings of bidirecional link beween erm srucure of ineres rae and macro economy and consruc dynamic erm srucure model enirely based on hree macroeconomic facors namely inflaion, he oupu gap and shor-erm policy ineres rae. The main assumpion is ha aggregae macroeconomic relaionships can be described using a linear framework. Using German daa and model is esimaed by maximum likelihood esimaion. They found ha esimaed macro variable parameers which are parly deermined by he erm srucure are consisen wih hose esimaed only macro variables. On he oher hand model has significan explanaory power for he erm srucure. On he oher hand i is found ha yields don' seem o 77

2 provide useful addiional informaion in forecasing macro variables however model performs very well in forecasing yield curve. Diebold e al (2006) examine he correlaions beween Nelson Siegel yield facors and macroeconomic variables. The basic model framework for yield curve is a laen facor model bu in dynamics fashion. Diebold e al (2006) characerize he relaionship among Level (L), Slope (S) and Curvaure (C) facors and he macro economy. They found srong evidence of macroeconomic effecs on fuure yield curve and somewha weaker evidence ha yield curve effecs fuure macroeconomic variables. I is found ha marke yields conain imporan predicive informaion abou federal funds rae. In addiion o hese sudies, Redebusch and Wu (2003), Ang e al. (2006) consruc join models and hey boh find ha here is bidirecional link beween yield curve and macroeconomic variables. This sudy includes very iniial analyses for he relaionship beween erm srucure of ineres rae and macro variables in Turkey. Following pars are including facor analysis and regression analysis. Iniial findings indicae ha corresponding relaion has srucural break around 2002 which coincides wih new moneary policy namely inflaion argeing dae. In pre2002 period role of macroeconomic variables in yield curve is limied however in pos 2002 period macro variables play very crucial role in erm srucure of ineres rae. We found ha inflaion and exchange rae are wo major macro variables ha deermine he shape of yield curve. 2. FIRST LOOK AT THE DATA AND INITIAL ANALYSIS This secion provides an overview of daa descripion, summary saisics and some simple analysis. We use monhly daa covering he period 1993M1-2009M2 for 1, 2,3,4,6,9 and 12 monhs mauriies ineres raes; r1, r2, r3, r4, r6, r9, r12. Ineres raes daa has been obained by Riskurk Official bond marke daa has been colleced from Isanbul Sock Exchange and spo yields are solved hen by a simple inerpolaion scheme daily yield curve is consruced. We calculae monhly ineres raes by aking monhly averages We use numbers of macro daa, including inflaion, growh, capaciy uilizaion, USD exchange rae growh, real effecive exchange rae, ineres rae of CBRT, change in Isanbul sock exchange rae, change in budge defici, amoun of domesic borrowing, capial accoun, curren accoun balance. Macro variables are obained from Inernaional Financial Saisics and Cenral Bank of Republic of Turkey. Figure 1 plos he lowes (1 monh) and highes (12 monhs) mauriies ineres raes in he sudy and Figure 2 plos he 1 monh ineres rae and one monh inflaion rae (calculaed from CPI). Shor and long raes moves very closely and boh level and variaion of ineres raes are gradually decreasing afer Decreasing characerisic of he yields seem o be very relaed o he level of inflaion in Turkey (Figure 2). 78

3 Figure 1 : Sor and Long Rae Figure 2 : Shor rae and Inflaion IN F 1 TR1 IN F 1 TR1 Figure 3: Term Spread TR12-TR1 Figure 3 plos he spread beween long and shor raes namely welve monh and one monh The yield curve usually upward sloping during he In lieraure i is well documened ha yield spreads are successfully predic he recession (Esrella and Mishkin, 1998, Inova e al. 2000, Chauve and Poer, 2002). I is found ha negaively sloped yield curve is always followed by a recession (Ang e al.2003). According o Durmus (2009), during Turkey has experienced 3 recessions which are in 1994M4-1995M3, 1998M and 2001M2-2002M2. On Figure 3 we can see ha downward sloping yield curve coincides wih hese recession imes. As an iniial analysis, we ry o invesigae wheher he yield curve predic recession in Turkey by using a simple probi model. The model is following; 79

4 y = β ' s k + ε 1, if y < 0 R 0, if y > 0 Esimaed equaion is PR ( = 1) = F( β ' s ) k where y represens he occurrence of recession a ime and s is he vecor including consan and spread as independen variables. We use monhly GDP 1 growh (loggdp-loggdp ₁₂), and capaciy uilizaion as dependen variable. When he monhly growh is negaive hen in esimaed equaion R akes values 1 and 0 oherwise. When we use consruced GDP growh as dependen variable, we canno find any evidence ha spread predic recession. However use of capaciy uilizaion change o deermine recession ime (i indicaes similar recession imes wih Durmus (2009) shows ha negaively sloped yield curve has some predicion power on recession. We find ha coefficien of spread for k=3, is and significan a 10 percen level. For any oher lag srucure coefficiens are insignifican. These findings indicae ha when spread beween long and shor yield urns in negaive, i can be read as a signal for coming recession. In he lieraure i is documened he insabiliy of yield curve and unsable relaionship beween macroeconomic variables and yield curve (Sock and Wason, 2003). The main source of insabiliy is regarded as moneary policy. In Turkey here is moneary policy shif in 2002 in which inflaion argeing is sared. A deail look a he ineres raes and macroeconomic variables, indicaes a srucural change of series around This is mainly because of differen ype of economic and poliical condiions including moneary regimes 2. In order o clarify his argumen we employ srucural break es on ineres raes and macroeconomic variables. By employing Quand-Andrews endogenous srucural break es on he regression y=c+ε where every series are regressed on only consan, we found level srucural break around 2002 for all variables 3. In he ligh of hese findings, we divide sample period as 1993: :12 (pre 2002) and 2002: :01 (pos 2002) o ge more accurae resuls and see he effecs of moneary policy regimes. 1 Monhly GDP is calculaed from quarerly GDP by using qubic spline mehod 2 During period Turkey experienced hree financial crises and a grea earhquake. During CBRT mosly did no serilize he capial inflow however in period CBRT choose o serilized invenory policy. On he oher hand, over he period fixed exchange rae regime is used. Also in his period govermen was changed nine imes. 3 We also apply anoher endogenous srucural break es developed by Bai and Perron (1998,2003) and find very similar break daes. 80

5 Table 1: Break Tes 3. FACTOR ANALYSIS One common way of analyzing yield curve is facor model approach which enables o express a large se of variables as a funcion of small se of unobserved facor. Dai and Singleoon (2000) show ha yield curve can be very well expressed by hree facors. Among pracioner Nelson and Siegel (1997) represenaion of yield curve is very popular. Nelson and Siegel represenaion is: λτ λτ 1 e 1 e λτ y( τ) = β1 + β2 + β3 e λτ λτ where β and λ are parameers, y( τ ) is yield wih mauriy τ.nelson-siegel yield curve is derived from a consan plus Laugerre funcion, which is a popular mahemaical approximaing funcion, ype forward rae curve. Diebold and Li (2006) describe ha his model is a dynamic laen hree-facor model in which β 1, β 2, and β 3 ime varying level, slope and curvaure. Diebold and Li (2006) shows ha Nelson-Siegel yield curve represenaion is a dynamic laen hree-facor model in which β 1 2 and β 3 ime varying level, slope and curvaure facors. Parameer λ governs he exponenial decay rae. Small values of λ produces slow decay and large values of λ produces fas decay and small and large value beer fis he curve a he shor and a he long respecively. Addiionally λ deermine a where β reaches is maximum. Loadings on β 1 is 1 which means ha i does no decay o zero in he limi and hence can be regarded as long erm facor. The loadings on β 2 sars a 1 bu quickly and monoonically decay o zero so i can be inerpreed as shor erm facor. On he oher hand loadings on β3 sars a zero, increases and hen decays o zero: Thus i can be viewed as medium erm facor. Following lieraure, hese facors also may be inerpreed in erms of level, slope and curvaure facors respecively (Diebold and Li, 2006). 3 81

6 Figure 4: Facor Loadings Figure 4 plos he facor loadings. We can easily show ha an increase in β1 increases enire yield curve equally hence change he level of yield curve. β 2 closely relaed o yield curve slope which is defined as long minus shor rae (usually en years minus hree monhs yield). I is defined he yield curve slope as y( )-y(0) which exacly equal o - β 2. Shor raes loads on β 2 is much more, an increase in β 2 increases shor yield more han long yields. Lasly β_{3} is closely relaed o curvaure because an increase in β 3 increases medium erm very much bu shor and long yield very small (Diebold and Lee,2006). One way of esimaing parameers of θ={ β 1 2 3, λ } is use of nonlinear leas squares for each monh. Employing nonlinear leas square produces numbers of challenging numerical opimizaions and ready o suck in local maximums. Insead of doing his, we can esimae parameer by ordinary leas square if we can fix he λ. How can one find a fixed value for λ? We can find an appropriae value for λ, if we consider ha λ deermines he mauriy a which curvaure facor namely β3 reach is maximum. In he lieraure wo or hree years are commonly used in ha regard. However in our sudy longes mauriy is 12 monhs. By aking he mauriy scale ino consideraion hree or four monh can be properly used, and we use he average of hese and picked he 3.5 monhs. We found ha value ha maximizes he loadings on β3 a mauriy 3.5 (i.e. τ=3.5) monhs is Using τ=3.5 and λ= we esimae β 1 2 and β 3 by employing leas squares. In lieraure, level, slope and curvaure facors have empirical counerpars. As level facor is he mos persisen one, long rae is regarded as empirical level facor, in our sudy i is y(12). Empirical slope facor was defined above, in here i corresponds o y(1)-y(12). Empirical curvaure facor is usually defined as 2y(24)-y(120)-y(3). However in his sudy highes mauriy is 12 and by aking scale ino consideraion, we define empirical curvaure facor as [y(3)+y(4)]-y(12)-y(3). 82

7 In pre 2002 period, correlaion beween β 1 2 3,and heir empirical counerpars are 0.82, 0.96 and In pos 2002 period correlaions are 0.99, 0.99 and 0.97 respecively. In boh period, esimaed facors and heir empirical counerpars are highly correlaed, and correlaion is increasing in pos 2002 period..figure plos he facors and empirical counerpars during whole period. Table 2: Descripive Saisics and Uni Roo Tes * indicaes significan a 0.05 level 4. ROLE OF MACROECONOMIC VARIABLES Wha abou role of macroeconomic variables in his picure? By now we can analyze he relaion beween macro variables and esimaed laen facors o ge clues abou he role of macroeconomic variables in yield curve dynamics. Table 2 documened he descripive saaisics and Augmened Dickey Fuller uni roo es resuls for facors. Only for β 1 in pos 2002 period null of uni roo canno be rejeced. Table 3: Correlaions (Pre 2002) In pre 2002 period 1 has he highes correlaion wih β 3. From Fisher equaion one can expec a link beween he level of he yield curve and inflaion expecaions hence 0.35 correlaion beween β 1 and inflaion is consisen wih his. Including Diebold e.al. (2006), Hordahl e al (2006), and Rudebusch and Wu (2003), in macro finance lieraure his relaion is common heme. However correlaion beween usd dollar growh and level of yield curve is a bi higher han correlaion of inflaion. Previous sudies show ha exchange rae has imporan role in risk premium, price level and moneary policy (Diboğlu and Kibriçioğlu, 2004) in Turkey. This link is no documened before in corresponding lieraure and indicaes ha exchange rae should no be rule ou in yield curve sudies in an emerging marke like Turkey. As we documened before - β 2 is highly correlaed wih slope of yield curve y(12)-y(1). In lieraure i is well documened ha slope of yield curve is a good predicor of fuure economic growh. Correlaion beween β and growh shows ha here is no expeced link. bu correlaion 2 83

8 wih inflaion is abou Including Mishkin (1990), Esrella and Mishkin (1998), Esrella (2004) among ohers showed ha slope of yield curve predics fuure inflaion. So correlaion beween slope facor and inflaion is consisen wih hese sudies. Table 4: Correlaions(Pos 2002) The same correlaion able for pos 2002 period draws a differen picure. Fis of all, he correlaion beween level and slope facor is very high which indicae ha boh of hem effeced by common facor/facors. Correlaion of inflaion wih level facor is 0.90 and slope facor is On he oher hand growh has high correlaion wih boh level and slope facor. Also USD growh's correlaion is no low wih hese facors. From his picure we can say ha boh level and slope facor mainly driven by inflaion and growh. While in pre2002 period, curvaure facor is no correlaed any macro variables, in pos 2002 period correlaion wih macro variables is considerable high. This findings indicaes ha in pos 2002 period yield curve is mainly driven by he macro variables. 5. CONCLUSION This sudy includes very iniial analyses for he relaionship beween erm srucure of ineres rae and macro variables in Turkey. Iniial findings indicae ha corresponding relaion has srucural break around 2002 which coincides wih new moneary policy namely inflaion argeing dae. In pre2002 period role of macroeconomic variables in yield curve is limied however in pos 2002 period macro variables play very crucial role in erm srucure of ineres rae. We found ha inflaion and exchange rae are wo major macro variables ha deermine he shape of yield curve. REFERENCES Ang, Andrew, Piazzesi, Monika. and Wei, M Wha does he yield curve ell us abou GDP growh?, Journal of Economerics, 131, Ang, Andrew., Piazzesi, Monika., A no-arbirage vecor auoregression of erm srucure dynamics wih macroeconomic and laen variables. Journal of Moneary Economics, 50, Bai, J. and Perron, P. (2003). Compuaion and Analysis of Muliple Srucural Change Models, Journal of Applied Economerics, Vol.18, pp Chauve, M. and Poer, S "Redicing a recession: evidence from he yield curve in he presence of srucural breaks, Economics Leers, 77, Dai, Q.; Singleon, K.; "Specificaion analysis of affine erm srucure models", The journal of finance, 55(5),

9 Diebold, F. X., Rudebusch, G. D., and, Aruoba, S.B The macroeconomy and he yield curve: a dynamic laen facor approach, Journal of Economerics,131, Diebold, F.X., Li, C., Forecasing he erm srucure of governmen bond yields. Journal of Economerics, 130, Durmus, D, Recessions in Turkey Las For 11 Monhs" published in Deneim Turkey Business Review, July-Augus. Diboglu, S and Kibriçioğlu, A Inflaion, oupu growh, and sabilizaion in Turkey, , Journal of Economics and Business 56, Esrella, A Why does he yield curve predic oupu and inflaion?, Economic Journal, 115, Esrella, A. and Mishkin,. F. S The predicive power of he erm srucure of ineres raes in Europe and he Unied Saes: Implicaions for he European Cenral Bank, European Economic Review, 41, Esrella, A., Mishkin, F.S., Predicing US recessions: financial variables as leading indicaors, Review of Economics and Saisics, 80, Hördahl, P.; Trisani, O.;Vesin, D.; A join economeric model of macroeconomic and erm srucuure of ineres rae, Journal of Economerics, 131, Ivanova, D., Lahiri, K. and Seiz, F Ineres rae spreads as predicors of German inflaion and business cycles, Inernaional Journal of Forecasing, Elsevier, 6, Mishkin, F. S Wha does he erm srucure ell us abou fuure inflaion? Journal of Moneary Economics, 25, Nelson, C.R., Siegel, A.F Parsimonious modeling of yield curves. Journal of Business 60, Rudebusch, G.D., Wu, T., A macro-finance model of he erm srucure, moneary policy, and he economy. Manuscrip, Federal Reserve Bank of San Francisco. Sock, J.H., and Wason, M.V Forecasing oupu and inflaion: The role of asse prices, Journal of Economic Lieraure, 41,