Export Dynamics in Turkey. Çağrı Sarıkaya

Transcription

1 Cenral Bank Review ISSN prin / online 2004 Cenral Bank of he Republic of Turkey hp:// Expor Dynamics in Turkey Çağrı Sarıkaya Research and Moneary Policy Deparmen Cenral Bank of he Republic of Turkey İsiklal Caddesi No: Ankara, Turkey Cagri.Sarikaya@cmb.gov.r Phone: Absrac This paper presens a srucural vecor auoregression model o explore he expor dynamics in Turkey. Given he noable expor performance afer 2002, albei high-raed real appreciaion of Turkish lira, we invesigae he role of uni wages in explaining he high expor growh. We observe ha, hrough hisorical decomposiion of expors, real uni wage, no he real exchange rae, has been he main deerminan of Turkish expors afer Moreover, he impulse response analysis suggess ha he shor-erm impac of a real uni wage shock on expors is larger compared o ha of he real exchange rae. The same conclusion applies even for he long-run effecs, provided ha he confidence in he economy is mainained. We also demonsrae he imporance of real uni wages by esimaing an errorcorrecion model, which provides consisen resuls wih he impulse response analysis. The analysis poins ou ha he real exchange rae is no he sole deerminan of he expor behavior in Turkey. The main poin of he sudy is ha, expor growh can be susained, even when he real exchange rae is appreciaing, if he improvemen in labor produciviy can be susained. JEL Classificaion Codes: C32, C51, and F17. Key Words: Srucural Vecor Auoregression, Expor Dynamics, Hisorical Decomposiion. The views expressed are hose of he auhor and should no be aribued o he Cenral Bank of he Republic of Turkey. I would like o hank Hakan Kara for his guidance and valuable suggesions. I also hank seminar paricipans a he MEEA sessions a he Allied Social Science Associaion (ASSA) in Philadelphia, January 7-9, 2005, for useful commens.

2 42 1. Inroducion Çağrı Sarıkaya / Cenral Bank Review 2 (2004) There has always been an ongoing debae and criics abou he level of he real exchange rae and is poenial effecs on foreign rade in Turkey. Exporer firms generally complain abou overvaluaion of domesic currency, as if i is he only variable ha deermines he degree of compeiiveness in he inernaional goods markes. This urns ou o be a misbelief considering he enhanced expor performance especially prevailed in he recen years, during which expor developmens canno be explained by he movemens of he real exchange rae: In 2003, he real expors had grown by 28.9 percen while he real exchange rae appreciaion had been 21.4 percen compared o he year The real exchange rae is he mos widely used indicaor since i is an index ha can easily be consruced and allows one o measure compeiiveness bilaerally or agains a group of counries. While comparabiliy is a major advanage, he use of aggregaed price indices due o incomparable daa for sub-secor prices migh make i unrealisic in some cases. Uni labor cos in a cerain indusry emerges as an alernaive indicaor of compeiiveness as i reflecs he comparaive advanage in producion. Alhough his indicaor s role in he deerminaion of a counry s compeiiveness is generally disregarded and underesimaed by he exporer firms, we argue ha i can be significan in explaining he movemens in expors. On he oher hand, recen lieraure mosly focuses on he role of relaive prices in he deerminaion of rade flows. Mos of he empirical works on he deerminans of rade flows are based on he imperfec subsiues model, which was described by Goldsein and Khan (1985). The sudies rying o esimae rade elasiciies, make use of relaive expor price and real world income as explanaory variables in an expor equaion. Khan (1974) esimaes such an expor demand funcion by OLS for a se of developing counries and concludes ha relaive price is a major deerminan of expors. Bahmani-Oskooee and Niroomand (1998) and Senhadji and Monenegro (1999) also employ similar expor demand specificaions. 1 There are oher sudies exending he variable se by including an exchange rae variable in he expor demand funcion. Bahmani-Oskooee (1998) adds a nominal effecive exchange rae in he expor equaion. Warner and Kreinin (1983) employ 1 Senhadji and Monenegro (1999) include a lagged expor volume in addiion o relaive price and world income variables.

3 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) OLS echnique o explain expors by a measure of world income, expor price, effecive exchange rae, a proxy for he expeced exchange rae, and a weighed average expor price of foreign counries. They found imporan role for he exchange rae and compeiors expor price in explaining expor dynamics for seleced developed economies. Sahinbeyoglu and Ulasan (1999) esimae error-correcion models for expor supply and demand funcions for Turkey and uses he wo-sep Engle-Granger mehodology. The variables involved are he real exchange rae and some measures of real income. 2 On he oher hand, recognizing he simulaneiy problem in esimaing single equaion models, Behar and Edwards (2004) use a vecor-errorcorrecion model o esimae he expor supply and demand equaions for Souh Africa. The variable se, which is consruced for boh home and compeior counries, consiss of a measure of aggregaed prices, expor prices, and income. Relaive price (or real exchange rae) is found o be a significan deerminan of Turkish expors in a number of sudies for Turkey. 3 However, he noable expor performance in recen years, albei high raed real appreciaion of Turkish lira, creaes room for invesigaing he role of uni wage in explaining he expor behavior in Turkey. 4 I is, herefore, imporan o assess he sensiiviy of expors o changes in above-menioned indicaors of compeiiveness in order o evaluae wheher real exchange raes are he sole driving force of expors. The main purpose of his paper is o examine he expor dynamics in Turkey hrough analyzing he effecs of he real exchange rae, he real uni wage, and he real income. In his respec, effecs of hese variables on real expors are quesioned hrough esimaing he corresponding elasiciies and analyzing he response of real expors o various shocks. In doing so, we make use of economeric evidence in he conex of srucural vecor auoregression (SVAR) and error-correcion mechanism (ECM) models. Secion 2 consiss of idenificaion of a srucural VAR, hisorical decomposiion of real expors and examinaion of he impulse responses of real expors o demand, real exchange rae and real uni wage shocks. Secion 3 inroduces an ECM model hrough which shor-run and long-run elasiciies of expors wih respec o corresponding variables are obained. Secion 4 concludes. 2 Expor supply and demand equaions include domesic income and foreign income respecively. 3 See Saygili e al. (1998), Sahinbeyoglu and Ulasan (1999), Aydin e al. (2004). 4 Aydin e al. (2004) also considers he uni wage in an expor supply equaion.

4 44 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) A Srucural VAR Analysis of Expor Dynamics There has been a considerable ineres in using VAR models o analyze he dynamic ineracion of variables. Given he fac ha single equaion models are inadequae in he presence of feedback relaions, VAR models come ino he scene o handle he issue of exogeneiy. In his conex, Sims (1980) suggesion of reaing all variables as endogenous and esimaing large-scale macro models as unresriced reduced forms, provides he basis for VAR models. However, Sims VAR approach has been criicized, as i does no resor o economic heory. Even hough one can avoid imposing incredible resricions by reaing all variables symmerically, he procedure has been subjec o criicisms as being mechanical. On he oher hand, SVARs add an economic conen as aking he conemporaneous relaions beween he variables ino accoun. I also allows for he idenificaion of he parameers based on an economic model and srucural shocks. In his sudy, he ineracion beween real expors, real income, real exchange rae, and real uni wage are analyzed in a SVAR seing. Therefore, once he decision on wheher he SVAR will be in level or difference form is made, he model will be inroduced. Sims (1980) recommends agains differencing, in he purpose of keeping he relaionships among he variables wihou loss of informaion. In he ligh of he exising lieraure, he quesion of wheher he variables in a VAR need o be saionary is considered in he conex of coinegraion. The fac ha non-saionary variables can lead o spurious regressions in he presence of long-run comovemens requires a formal es for coinegraion. Therefore, once he order of inegraion of he variables in our mulivariae sysem is deermined, Johansen (1988) echnique will be carried ou o es for coinegraion. This procedure requires he selecion of an appropriae lag-lengh of he VAR model, idenificaion of he deerminisic componens (a consan and rend) and he deerminaion of he coinegraion rank. Daa The models are esimaed wih quarerly daa covering he period 1989:1-2003:3. The expor and gross domesic produc (GDP) figures are from he naional accouns published by he Sae Insiue of Saisics (SIS). The real exchange rae is he CPI based rade-weighed index calculaed by he Cenral Bank of he Republic of Turkey (CBRT). The real uni wage in he manufacuring secor is calculaed as

5 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) w. L w = (1) Q Q L where w is he real wage per working hour index in he manufacuring secor, L is he working hours index in he manufacuring secor, and Q is he manufacuring indusry producion, all of which are announced by he SIS. All daa is obained from he elecronic daa delivery sysem (EDDS) of he CBRT. 5 Uni Roo Tess The saionariy of he variables is esed using he Augmened Dickey-Fuller (ADF) procedure. The es resuls sugges ha he null hypohesis of a uni roo canno be rejeced a 5 percen significance level for he series in levels (Table 1). On he conrary, he null of non-saionariy is srongly rejeced for he firs differenced series (Table 2). Therefore, all variables appear o be inegraed of order 1. Table 1 Augmened Dickey-Fuller Tes Resuls for Levels Variables in Logarihms Deerminisic Componen Lag Lengh ADF saisic 95% Criical Value Order of Inegraion gdp C, T I(1) exp C, T I(1) rer C I(1) ruw C, T I(1) * The 5 percen criical values for he ADF ess are compued using he response surface esimaes given in MacKinnon (1991, Table 1). Table 2 Augmened Dickey-Fuller Tes Resuls for Firs Differences Variables in Logarihms Deerminisic Componen Lag Lengh ADF saisic 95% Criical Value Order of Inegraion gdp C I(0) exp C I(0) rer None I(0) ruw C I(0) * The 5 percen criical values for he ADF ess are compued using he response surface esimaes given in MacKinnon (1991, Table 1). 5 The daa is available a hp://cmbf40.cmb.gov.r/fame/webfacory/evdpw/yeni/cb-uk.hml.

6 46 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Deerminaion of Coinegraion Rank Defining z as he vecor of n endogenous variables, an unresriced vecor auoregression involving k-lags of z can be wrien as z = Α1 z 1 + K+ Α k z k + u (2) where z is ( n 1) and Α i is he ( n n) marix of parameers. When reformulaed, he unresriced VAR model akes a vecor error-correcion (VECM) form. z = Γ1 z 1+ K+ Γk 1 z k + 1+Πz k + u (3) where Γ = Ι Α K Α ), ( i = 1, K, k 1), and i ( 1 i Π = ( Ι Α1 K Α k ). In his conex, Johansen (1988) approach o esing for coinegraion is equivalen o finding r, he number of linearly independen columns, namely he rank of Π, which conains informaion on long-run relaionships among he variables consrucing z. The saring poin o es for coinegraion in a mulivariae sysem requires an appropriae lag lengh selecion for he unresriced VAR. The opimal lag is aken as 3 according o he resuls suggesed by he likelihood raio es and Akaike and Schwarz informaion crieria, which are presened in he Appendix (Table 6). The wo widely used ess for he deerminaion of coinegraion rank are he maximal-eigenvalue and race ess. The es resuls, which allow us o decide upon he configuraion of he deerminisic componens a he same ime, are presened in he Appendix (Table 7). Boh es saisics sugges ha here is one coinegraing relaionship among he variables in an unresriced inerceps and no rends model, in which here is one inercep in he shor-run par of he model. Hence, we consruc a VAR model wih he level series in he ligh of Sims argumen ha differencing hrows away informaion concerning he comovemens in he daa in he presence of coinegraion. The Model Leing lower-case leers denoe logarihmic values, he basic VAR specificaion can be wrien as L z = Φ z + u s= 1 s s (4)

7 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) where z is a vecor of real GDP, real expors, real exchange rae and real uni wage. Quarerly daa is used for he sample period 1989:1-2003:3. Our VAR specificaion conains a consan and a cenered seasonal dummy for he hird quarer o capure he effec of higher economic aciviy in ha period. [ y x, q w ] u,, is he corresponding innovaions vecor from he reduced-form VAR. These residuals are serially uncorrelaed bu in general conemporaneously correlaed, i.e. heir variance-covariance marix is no diagonal. In his respec, we will carry ou a srucural decomposiion considering economically meaningful conemporaneous relaionships among he variables, o idenify uncorrelaed shocks. A SVAR(3) is esimaed using quarerly daa for he sample period 1989:1-2003:3. The sysem is formulaed as θ d y = a12 x + b14v + v (5) x x = a 21 y + v (6) where [ ] d x q θ q q = v (7) θ w = a41 y + a43q + v (8) v v, v, v, v is he vecor of uncorrelaed srucural shocks o real demand, real expors, real exchange rae, and real uni wage respecively. Equaion (5) saes ha beyond he effecs of lagged variables, here is a direc impac of real expors on real oupu since oal producion wihin a quarer includes expored goods as well. The movemens of real oupu can also be as a response o srucural shocks o real uni wage and shocks o demand. Equaion (6) reflecs he fac ha some porion of he oal producion wihin a period is direced owards inernaional markes, so here is a conemporaneous effec of real oupu on real expors. Equaion (7) implies ha real exchange rae appears as he mos exogenously deermined variable as here is no direc feedback from he oher variables in he sysem. The fac ha real oupu eners direcly o he calculaion of real uni wage variable is refleced as he conemporaneous effec of real oupu in Equaion (8). Whenever here is a change in he overall aciviy, by definiion here will be an auomaic change in real uni wage. In order o idenify he sysem we imposed wo resricions on he real GDP equaion. Firsly, he simulaneiy problem in (5) and (6) is solved by calibraing a 12. The real expor elasiciy of real oupu, a 12, is resriced o be 0.2. In seing his elasiciy, he coefficien from he linear regression of he percenage annual

8 48 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) change of real GDP o ha of real expors as well as he hisorical average share of real expors in real oupu is aken ino accoun. Secondly, assuming ha oupu is demand driven in he shor-run, b 14 is resriced o be zero. Hisorical Decomposiion The hisorical decomposiion of a ime series is based on he following pariion of he moving average represenaion of he srucural model. z j 1 + j = C sv + j s + s= 0 s= j C s v + j s The sum componen a he far righ is he base forecas of z + j based on he informaion available a ime. However, he base projecion of z + j canno be equal o is realized value a ime + j, provided ha new srucural innovaions from + 1 o + j ake place. The firs sum componen represens he forecas error of in z + j z + j, which allows us o idenify he sources of he unexpeced movemens. In his way, we can figure ou wheher he flucuaions in Turkish expors were he resul of shocks o oher variables in he sysem. Hisorical decomposiion of real expors for he period 1996:1-2003:3 shows ha he expor shocks have he mos explanaory power for he gap beween acual expors and he base forecas unil Afer 1999, almos he enire difference beween acual expors and he base forecas can be aribued o real uni wage shocks. On he oher hand, real exchange rae, which is he mos widely used compeiiveness indicaor, does no explain he deviaion of ex-pos expors from he base forecass in his period (Figure 7). Therefore, real uni wages raher han real exchange rae developmens has been playing a major role in causing movemens in he expor performance since In a labor-inensive small economy, labor and impored inermediae and capial goods emerges as he principal inpus for producion. An appreciaed domesic currency creaes incenive for firms o subsiue labor for capial and hus conribues o he improvemen of labor produciviy. Hence, low real uni wages and appreciaed domesic currency would cause producion coss o fall significanly making exporer firms more compeiive in inernaional markes. In he Turkish economy hrough 1999 o 2003, he real uni wages declined by almos 40% resuling from depressed real wages and improved produciviy. Hisorical decomposiion resuls sugges ha his has been he major driving force in he (9)

9 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) enhanced expor performance especially afer 1999, which is also verified by he figures below. Fig. 1. Expors and Real Uni Wage (Annual % Change) Fig. 2. Real Exchange Rae (1995=100) Expors Real Uni Wage Real Exchange Rae 90Q1 91Q1 92Q1 93Q1 94Q1 95Q1 96Q1 97Q1 98Q1 99Q1 00Q1 01Q1 02Q1 03Q1 90Q1 91Q1 92Q1 93Q1 94Q1 95Q1 96Q1 97Q1 98Q1 99Q1 00Q1 01Q1 02Q1 03Q1 Source: SIS Source: CBRT Impulse Response Analysis The impulse responses are used o invesigae he dynamic effecs of srucural shocks on real expors. In his conex, he elasiciy of expors wih respec o income, price and cos variables can also be analyzed. The responses of real expors o he shocks are consisen wih he economic heory. In ha sense, a demand shock resuls in an expansion in he producion of expored goods, while shocks o real exchange rae and real uni wage have adverse effecs on real expors hrough higher inpu prices and hus producion coss. A 10% rise in demand induces a 0.6% increase in real expors conemporaneously, while he effec rises o 6% in he following period as he producion plans are modified. Afer full adjusmen has aken place he real expors expands approximaely by 11% afer weny quarers. Therefore, mos of he adjusmen in response o new demand condiions akes one quarer (Table 8, Figure 7). A shock o he real exchange rae sars o show is effecs on expors wih a lag of one quarer. A 10% appreciaion in Turkish lira a ime gives rise o a 1.7% reducion in real expors a ime + 1. The effec of he shock srenghens rapidly a each year so ha he cumulaive effec is a 6% fall in real expors afer weny quarers (Table 8, Figure 7). The effec of a 10% real uni wage shock is a 2.5% expansion in real expors

10 50 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) wih a one- quarer lag afer he shock akes place. The full adjusmen occurs in en quarers and he effecs of he shock die ou hereafer. The impac of he shock increases gradually bringing ou a 3.8% drop in real expors afer weny quarers (Table 8, Figure 7). The impulse responses o real exchange rae and real uni wage shocks, being major deerminans of he movemens in expors, have imporan implicaions beyond heir magniudes. Firsly, above figures indicae ha a large porion of he cumulaive response o a real uni wage shock is realized one period afer he shock akes place. On he oher hand, he iniial responses of expors o a change in he real exchange rae are smaller. These differeniaed effecs are consisen wih economic heory. A shock o real uni wage shows is effecs on real expors quickly in such a case ha i works as a direc shock o a real variable, oupu, which becomes available for sale in domesic or foreign markes a he same period. On he oher hand, i akes a longer ime for he oal effec of a real exchange rae shock o be revealed. The goods marke adjusmen in response o a change in he real exchange rae akes ime unil he producers perceive ha change and adjus heir producion plans accordingly. Secondly, he rae of increase of expors in response o real exchange rae depreciaion is greaer compared o he case of a decline in he real uni wage. This makes he cumulaive effec of he former one on real expors larger as opposed o relaive iniial effecs. A similar explanaion may be ha he gains from enhanced produciviy, hus reduced real uni wage, are consumed more quickly so ha a sable level of real expors is observed afer en quarers. On he oher hand, he slow adjusmen in he goods marke in erms of he producers percepion abou he change in heir compeiiveness and adopion of new producion plans makes he effecs of he real exchange rae shock spread over a longer ime period. An Alernaive Model Following he financial liberalizaion in 1989, he Turkish economy has exhibied a boom-bus cycle characerisic, which in urn made he capial flows a significan deerminan of economic growh. The fac ha in he periods of macroeconomic sabiliy, foreign capial has been araced, which gave rise o a srenghening of domesic currency. The dependence of he economy on shor-erm capial flows creaed an environmen, in which a real appreciaion of Turkish lira coincided wih high economic growh periods whereas a high real depreciaion was

11 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) consisen wih crisis periods. Therefore, apar from is direc effec on relaive price of home goods agains heir foreign counerpars, he real exchange has also an indirec effec on growh dynamics hrough a confidence channel depending on he qualiy of macroeconomic fundamenals in Turkey. This argumen leads o a modificaion in he oupu equaion while he oher equaions of he sysem remain he same. q d y = a12 x + b13v + v (10) q where v is assumed o be he confidence shock, which capures he indirec effec of he real exchange rae on oupu. Therefore, a posiive confidence shock is expeced o generae higher oupu and hus expors. Impulse response analysis confirms his anicipaion. A 10 % real appreciaion leads o a 2.8% in oupu and 0.2% in expors conemporaneously. In spie of his iniial posiive impac, he cumulaive effecs on oupu and expors are negaive as expeced. However, he negaive effec of a real exchange rae appreciaion on oupu and expors is more limied wih he new specificaion. In he previous case, he cumulaive effec of he real exchange rae on real expors was found o be larger han ha of he real uni wage as opposed o heir relaive iniial effecs. When confidence shock is included, real uni wage more han compensaes for he negaive effec of real exchange rae, even in he long run (Table 9, Figure 8). In general, impulse response and hisorical decomposiion resuls are robus o his alernaive model specificaion. Afer 1999, real uni wage shocks, raher han he real exchange rae, seem o be responsible for he deviaions of base forecas from acual expors (Figure 8). 3. Esimaing an ECM Model for Real Expors The procedures followed up o now were based on he fac ha all variables are inegraed of order 1. However, non-saionariy of he real exchange rae may leave a doub in ones mind, given he well-known purchasing power pariy (PPP) hypohesis, which assers ha he deviaions from PPP is ransiory, or equivalenly real exchange rae does no conain a uni roo. An alernaive approach o es for coinegraion is he auoregressive disribued lag (ARDL) procedure advanced in Pesaran and Shin (1995) and Pesaran e al. (1996). The main advanage of his echnique is ha i can be applied irrespecive of wheher he regressors are I(0) or I(1). By his way, one can avoid preesing problems associaed wih sandard coinegraion analysis. Therefore, his procedure

12 52 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) would no only provide a crosscheck for he coinegraion resuls suggesed by maximal eigenvalue and race saisics, bu also allow us o es and exrac a coinegraing relaionship in he form of an expor funcion. Moreover, while he esimaion of an error correcion equaion for he real expors would provide insighs abou he shor-run dynamics, obaining he long-run elasiciies wih respec o real exchange rae and real uni wage would allow us o make comparisons wih he resuls of he previous analysis. The esimaion resuls indicae ha an increase in domesic economic aciviy affecs he real expors posiively, whereas boh a real appreciaion of he Turkish lira and an increase in real uni wages has adverse effecs on real expors. However, i should be noed ha i is he growh raes of real expor and real exchange rae ha is negaively relaed, no he levels. In oher words, as he appreciaion rae increases, he rae of increase in real expors falls. The growh rae of he real exchange rae has a lagged effec while he uni wage growh has a conemporaneous effec on he rae of increase of real expors. The ECM coefficien in he model implies ha, a each period, fify percen of a deviaion from he longrun equilibrium is correced (Table 3). Table 3 Error Correcion Represenaion for ARDL(2,3,0,2) ± exp exp(-1) rer rer(-1) rer(-2) ruw gdp gdp(-1) ecm(-1) (-2.334) (0.087) (-2.895) (-1.983) (-3.767) (2.418) (2.430) (-3.962) R 2 Adj. R ± T-raios in he parenhesis. The model also conains he firs differences of a consan and hree cenered seasonal dummies, which are no presened in he able for visual ease. The shor-run elasiciies of real expors wih respec o real exchange rae and real uni wage are esimaed o be 0.7 and 0.4 in absolue values, respecively. The shor-run income elasiciy of real expors is esimaed as 1.3 (Table 3). As menioned above, hese elasiciies should be inerpreed as he response of he growh rae of real expors o he changes in he growh raes of he corresponding variables.

13 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Table 4 Long-Run Coefficiens from he ARDL Model ± ln(exp) ln(rer) ln(ruw) ln(gdp) consan sc1 sc2 sc (2.694) (-6.944) (5.163) (-1.676) (2.261) (2.466) (0.831) ± T-raios in he parenhesis. Sc1, sc2, and sc3 are he cenered seasonal dummies for he firs, second, and hird quarers respecively. The long-run elasiciies of real expors wih respec o real exchange rae and real uni wage are esimaed o be 0.6 and 0.8 in absolue values, respecively. The long-run income elasiciy of real expors is esimaed as 1.4 (Table 4). Here, he posiive coefficien of he real exchange rae emerges as an unexpeced resul ha requires aenion. Firsly, he economic crises experienced by Turkey in he las en years gave rise o an unsable economy wih high inflaion and real ineres raes and no susained growh. The mechanism peculiar o he Turkish economy works in such a way ha economic growh has been financed hrough capial inflows. The dependence of he economy on capial flows creaed an environmen, in which he periods of real appreciaion of home currency coincided wih high economic growh periods whereas high real depreciaion periods were associaed wih crisis periods. Moreover, expor performance srongly depends on domesic aciviy and hus, general economic condiions. Given his fac, he negaive relaionship beween real expors and he real exchange rae may be due o counry-specific condiions. However, keeping his explanaion in mind, he heory-inconsisen resul is sill surprising since he real GDP variable was expeced o ac as a conrolling variable for he facs menioned above. Secondly, he esimaed model can be inerpreed as an expor supply funcion raher han an expor demand funcion. Due o he srong dependence of producion on impored inermediae goods, consiuing 75% of oal impors, a real appreciaion of domesic currency creaes an incenive o expand producion by reducing he inpu coss. In his respec, he posiive coefficien of he real exchange rae may be seen as reflecing he simulaed oupu, and hus expors, as a resul of a decline in producion coss, in spie of a loss in compeiiveness. In a previous sudy by Sahinbeyoğlu and Ulasan (1999), Engle-Granger wo-sep procedure is followed and an error-correcion model is esimaed. They esimaed he shor-run elasiciies of real expors wih respec o real GDP and real exchange rae as -0.2 and 0.6 respecively. The corresponding long-run elasiciies are

14 54 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) esimaed o be -0.3 and These esimaed coefficiens significanly differ from our resuls. The esimaed income elasiciies are negaive and low in absolue values, whereas we esimaed hem o be posiive and greaer han 1 for boh shor and long erms. The shor-run price elasiciies are inconsisen wih our resuls in erms of signs or magniudes. In a panel daa sudy, Neyapi e al. (2003) quesions he effec of Turkey s paricipaion in he European Cusoms Union. They esimaed a posiive income elasiciy above 2 and negaive real exchange rae elasiciy as -0.7 wihou a shorrun and long-run differeniaion. The esimaed real exchange rae elasiciy is very close o ours for shor-run in erms of sign and magniude. The income elasiciy is greaer han our esimae bu consisen wih our finding ha real expors are elasic wih respec o income. Table 5 The Effec of a 10 % Change on Real Expors (%) Variables IR Analysis (no confidence shock) IR Analysis (wih confidence shock) ECM gdp rer ruw * Impulse responses cover a period of 12-quarers. As a final evaluaion, he shor-run esimaes of income, price and cos elasiciies of real expors from he ECM analysis and he 12-quarer cumulaive responses of real expors o shocks of hese variables from he previous impulse response analysis, are consisen wih each oher o a wide exen (Table 5). 4. Conclusion This paper uses srucural VAR and error-correcion models o invesigae he role of real GDP, real exchange rae and real uni wage in explaining he expor dynamics in Turkey. The SVAR analysis suggess ha he response of real expors o a demand shock is large and almos fify percen of he adjusmen is compleed in he nex quarer. The response o a real exchange rae shock is lower han ha o a real uni wage shock in he shor-run, bu cumulaive effecs show ha he effec of he real exchange rae is greaer han ha of uni wage in a longer ime period. This migh be reflecing he fac ha he former is an indirec shock o a producion process and akes a longer ime for he producion plans o be adjused once he producers perceive he gain in compeiiveness. On he oher hand, he laer is a

15 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) direc shock o a real variable, which improves he producion possibiliies wih he same quaniy of inpus. Besides, when a confidence shock erm is included in he oupu equaion, real uni wage more han compensaes for he negaive effec of he real exchange rae, even in a longer erm. Moreover, hisorical decomposiion of real expors for he period indicaes ha, afer 1999, almos he enire difference beween acual expors and he base forecas can be aribued o real uni wage shocks. This implies ha he real exchange rae is no he only and he mos imporan indicaor of compeiiveness, and hus is no he sole deerminan of he expor performance in Turkey. Insead, real exchange rae appreciaion conribues o he high expor performance by reducing he relaive price of capial goods, hus inducing firms o subsiue labor for capial. Increased capial-labor raio in he producion process promoes labor produciviy and hereby causes uni wages o decline. In his respec, las wo years experience in Turkey signifies he dominance of he cos channel over he convenional demand channel in ransmiing he effecs of he real exchange rae. Therefore, expor growh can be susained, even when he real exchange rae is appreciaing, if he decline in he real uni wage hrough eiher a fall in real wages and/or an improvemen in labor produciviy can be achieved. Given he significan impac of real uni wage on expors, an ECM is esimaed. The shor-run elasiciy of real expors wih respec o income, price and cos variables compued using ECM analysis exhibi a srong resemblance wih hose obained from he 12-quarer impulse responses. The income elasiciy of real expors is greaer compared o he elasiciies wih respec o real exchange rae and real uni wage. Moreover, he sensiiviy of real expors o real exchange rae is greaer han ha o real uni wage. On he oher hand, he long-run relaionship shows ha he income elasiciy of real expors is close o he one for he shor-run whereas he long-run elasiciy of real expors wih respec o real uni wage is almos doubled compared o is shor-run elasiciy. An ineresing resul is ha an appreciaion of Turkish lira is associaed wih higher expor performance in he long run, which migh be a consequence of he dependence of expors on impored inermediae inpus.

16 56 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) References Aydin, M.F., Ciplak, U. and Yucel, E.M Expor Supply and Impor Demand Models for he Turkish Economy. CBRT Working Paper No. 04/09, June Bahmani-Oskooee, M Coinegraion Approach o Esimae he Long-run Trade Elasiciies in LDCs. Inernaional Economic Journal 12, Bahmani-Oskooee, M. and Niroomand, F Long-run Price Elasiciies and he Marshall-Lerner Condiion Revisied. Economics Leers 61(1), Behar, A. and Edwards, L Esimaing Elasiciies of Demand and Supply for Souh African Manufacured Expors Using a Vecor Error Correcion Model. CSAE WPS / Goldsein, M. and Khan M.S Income and Price Effecs in Foreign Trade, in Handbook of Inernaional Economics, Vol II: , eds. R.W. Jones and P.B. Kenen, Elsevier Science Publishers B.V. Harris, D Coinegraion in Mulivariae Sysems. Ch.5, in Using Coinegraion Analysis in Economeric Modelling, Prenice Hall. Johansen, S Saisical Analysis of Coinegraion Vecors. Journal of Economic Dynamics and Conrol 12, Khan, M.S Impor and Expor Demand in Developing Counries. IMF Saff Papers, 21, Neyapi, B., Taskin, F. and Ungor, M Has European Cusoms Union Agreemen Really Affeced Turkey s Trade?. Inernaional Conference on Policy Modeling, July 3-5, 2003, Isanbul. MacKinnon, J.G Criical Values for Coinegraion Tess. Ch. 13, in Long-Run Economic Relaionships: Readings in Coinegraion, eds R.F. Engle and C.W.J. Granger, Oxford Universiy Press, Oxford. Pesaran, M.H. and Shin, Y. 1995a. An Auoregressive Disribued Lag Modelling Approach o Coinegraion Analysis. DAE Working Paper No. 9514, Deparmen of Applied Economics, Universiy of Cambridge. Pesaran, M.H., Shin, Y. and Smih, R.J. 1996a. Tesing for he Exisence of a Long-Run Relaionship. DAE Working Paper No. 9622, Deparmen of Applied Economics, Universiy of Cambridge. Sahinbeyoglu, G. and Ulasan, B An Empirical Examinaion of he Srucural Sabiliy of Expor Funcion. CBRT Discussion Paper No. 9907, Sepember Saygili, M., Sahinbeyoglu, G. and Ozbay, P Compeiiveness Indicaors and he Equilibrium Real Exchange Rae Dynamics in Turkey. Macroeconomic Analysis of Turkey: Essays on Curren Issues, ed. E. M. Ucer, Cenral Bank of he Republic of Turkey. Senhadji, S.A. and Monenegro, E.C Time Series Analysis of Expor Demand Equaions: A Cross-Counry Analysis. IMF Saff Papers, 46, Sims, C Macroeconomics and Realiy. Economerica 48, Warner, D. and Kreinin M.E Deerminans of Inernaional Trade Flows. Review of Economics and Saisics 65,

17 Appendix Table 6 Lag Choice Crieria Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Lag AIC SBC LR Tes p-value Table 7 Johansen Coinegraion Tes Resuls Maximal Eigenvalue Tes Null Alernaive Model 1 Model 2 Model 3 Model 4 Model 5 r = 0 r = r<= 1 r = r<= 2 r = r<= 3 r = % Criical Values for he Maximal Eigenvalue Tes Null Alernaive Model 1 Model 2 Model 3 Model 4 Model 5 r = 0 r = r<= 1 r = r<= 2 r = r<= 3 r = Trace Tes Null Alernaive Model 1 Model 2 Model 3 Model 4 Model 5 r = 0 r>= r<= 1 r>= r<= 2 r>= r<= 3 r = % Criical Values for he Trace Tes Null Alernaive Model 1 Model 2 Model 3 Model 4 Model 5 r = 0 r = r<= 1 r = r<= 2 r = r<= 3 r =

18 58 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Table 8 Impulse Response of Real Expors o 10% Shocks Response of Real Expors o 10% Shocks (%) Period Demand Shock RER Shock ULC Shock Table 9 Impulse Response of Real Expors o 10% Shocks (Alernaive Specificaion) Response of Real Expors o 10% Shocks (%) Period Demand Shock RER Shock ULC Shock

19 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Fig. 3. Hisorical Decomposiion of Real Expors acual expor base forecas effec of gdp acual expor base forecas effec of rer Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q1 1995Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q acual expor base forecas effec of exp acual expor base forecas effec of ulc Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q1 1995Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q1

20 60 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Fig. 4. Hisorical Decomposiion of Real Expors (Alernaive Specificaion) acual expor base forecas effec of gdp acual expor base forecas effec of rer Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q1 1995Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q acual expor base forecas effec of exp acual expor base forecas effec of ulc Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q1 1995Q1 1996Q1 1997Q1 1998Q1 1999Q1 2000Q1 2001Q1 2002Q1 2003Q1

21 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Fig. 5. Response of Real Expors o 10% Shocks 12 Response o a 10 % Demand Shock Response o a 10 % RER Shock Response o a 10 % ULC Shock

22 62 Çağrı Sarıkaya / Cenral Bank Review 2 (2004) Fig. 6. Response of Real Expors o 10% Shocks (Alernaive Specificaion) 12 Response o a 10 % Demand Shock Response o a 10 % RER Shock Response o a 10 % ULC Shock

23 Fig. 7. Responses o One Sandard Deviaion Shocks response of gdp o gdp response of ex o gdp response of rer o gdp response of ulc o gdp response of gdp o ex response of gdp o rer response of gdp o ulc response of ex o ex response of ex o rer response of ex o ulc response of rer o ex response of rer o rer response of rer o ulc response of ulc o ex response of ulc o rer response of ulc o ulc Çağrı Sarıkaya / Cenral Bank Review 2 (2004)

24 Fig. 8. Responses o One Sandard Deviaion Shocks (Alernaive Specificaion) response of gdp o gdp response of ex o gdp response of rer o gdp response of ulc o gdp response of gdp o ex response of ex o ex response of rer o ex response of ulc o ex response of gdp o rer response of ex o rer response of rer o rer response of ulc o rer response of gdp o ul c response of ex o ulc response of rer o ulc response of ulc o ulc Çağrı Sarıkaya / Cenral Bank Review 2 (2004)