THE WILLIAM DAVIDSON INSTITUTE AT THE UNIVERSITY OF MICHIGAN BUSINESS SCHOOL

Transcription

1 THE WILLIAM DAVIDSON INSTITUTE AT THE UNIVERSITY OF MICHIGAN BUSINESS SCHOOL The sources of Real Exchange Flucuaions in Developing Counries : an Economeric Invesigaion By: Imed Drine and Chrisophe Raul William Davidson Insiue Working Paper Number 653 February 2004

2 The sources of real exchange flucuaions in developing counries : an economeric invesigaion Imed Drine (EUREQua, Sorbonne Universiy 1 ) Chrisophe Raul (EPEE, Universiy of Evry-Val D Essonne 2 ) March 12, 2004 Absrac In his paper we address he wo following quesions: (1) wha are he major sources of real exchange rae flucuaions in developing counries? (2) do economic policy makers have room o face possible real exchange rae flucuaions? To answer hese quesions, we esimae a srucural VAR model for 3 developing counries (Morocco, The Philippines, Uruguay) and carry ou he convenional exercises of impulse response funcions and of variance decomposiion of forecas error. Our invesigaaion sugges ha domesic shocks dominae real exchange rae flucuaions and ha he conribuion of exernal shocks is relaively low. Besides, he low conribuion of he nominal shock pu ino quesion moneary policies which seek o promoe compeiiveness hrough a currency devaluaion. Moreover, our esimaions confirm ha he real exchange rae also depends on shocks on foreign ineres rae and/or on he erms of exchange which can make i move from is equilibrium level. The budgeary ool herefore remains efficien o sabilize he real exchange rae wih respec o possible exernal shocks. Keywords : real exchange rae, developing counries, srucural VAR, impulse response funcions. JEL Classificaion: E31, F0, F31, C15. 1 Paris I, Maison des Sciences de l'economie, bd. de L'Hôpial, Paris Cedex 13, France. drine@univ-paris1.fr 2 Déparemen d'économie, Boulevard François Mierrand, Évry Cedex, France. chraul@homail.com, web-sie : hp://

3 1. Inroducion Real exchange rae flucuaions have a cenral place in he discussions over he choices of economic policies in developing economies. Indeed, hese flucuaions have some repercussions on he economic performances and srong variaions of he real exchange rae enail high economic coss. I is essenially he dependence wih respec o impors and he specializaion in expors which accoun for real exchange rae flucuaions on he economic performances of developing counries. The accessibiliy o he world financial marke which helps o smooh ou consumpion in financing rade balance disequilibrium also plays an imporan role. Real exchange rae movemens and heir relaive imporance are from a heoreical and empirical poin of view one of he major challenges macro-economiss and policy makers have o cope wih. Indeed, heoreical works confirm ha real exchange rae flucuaions depend on he relaive imporance of differen sources of impulsion and ha he relaionship beween he real exchange rae and economic aciviy is raher complex, paricularly for an economy confroned wih various shocks. Previous works, ha generally deal wih developed counries, have ofen focused on domesic shocks and have ignored exernal shocks as possible sources of flucuaions. However, he imporance of exernal shocks seems obvious for developing counries given heir srong dependence wih respec o he world economy. A small developing economy differs from a developed one in differen ways. Firs of all, he expors of developing economies ofen include producs wih low added value whose prices are no sable. Having a low power marke, developing economies are ofen confroned wih srong flucuaions of heir expor prices. Besides, a large par of he incomes provided by heir expors serves o repay heir exernal deb. Secondly, developing counries srongly depend on foreign capial and on inermediae inpus impored for heir producions. On he basis of hese specificiies, i is sraighforward o conclude ha inernaional price flucuaions (primary produc prices, inermediae produc prices, and foreign ineres rae) can have imporan effecs on he cyclic flucuaions and noably on real exchange rae flucuaions 3. 3 Agnor and Meniel (1996) analyzed he relaionships beween inernaional price flucuaions and economic aciviy

4 The ineres o he analysis of he sources of real exchange rae flucuaions is mainly jusified by he cenral role played by he real exchange rae in developing counries and by he necessiy o deermine he room economic policy makers have a heir disposal o sabilize he exchange rae. Indeed, i is generally admied ha a negaive shock of demand (a resricive budgeary policy) leads o a real exchange rae depreciaion whereas a currency offer decrease (a resricive moneary policy) enails a real exchange rae appreciaion. Hence confroned wih exogenous variaions of he real exchange rae and according o economic policy purposes, he governmen can reac by budgeary and moneary policies. The efficiency of his kind of inervenion of course depends on he relaive conribuion of he shocks on public spending and of nominal shocks (moneary shock and shock on he nominal exchange rae) o real exchange rae flucuaions. Idenifying he sources of real exchange rae flucuaions enables o measure, on he one hand, he consequences of economic policies implemened by he governmen on he real exchange raes, and on he oher, he room policy makers have a heir disposal o deal wih possible real exchange rae movemens harmful o economic aciviy. In his perspecive we address in his paper he wo following quesions: wha are he major sources of real exchange rae flucuaions in developing counries? do economic policy makers have room o face possible real exchange rae flucuaions? To answer hese quesions, we esimae a srucural Vecor Auoregressive Model (VAR) for 3 developing counries (Morocco, The Philippines, Uruguay) and carry ou he convenional exercises of impulse response funcions and of variance decomposiion of forecas error in order o quanify he relaive conribuion of he differen shocks o real exchange rae flucuaions. We assume here he exisence of four possible sources of impulsion 4, namely: 1. domesic real shocks which affec he offer side such as produciviy shocks; 2. domesic real shocks which affec he demand side, such as consumpion or invesmen variaions, public spending and he shocks on preferences; in some developing counries and showed ha foreign ineres rae variaions significanly conribued o he cyclic movemens.

5 3. nominal shocks which reflec he relaive variaions of money demand wih regard o offer and nominal exchange rae variaions; 4. exernal real shocks, for insance a foreign ineres rae variaion. Having in mind he idenificaion problem of he various shocks, we expose in a second secion a heoreical model which will be used aferwards as a benchmark o specify he dynamics of he various variables wih respec o he four shocks. In a hird secion, we presen he economeric mehodology as well as he idenificaion scheme of he shocks. More precisely, he economeric invesigaion of he sources of real exchange rae flucuaions is carried ou in a srucural VAR model composed of four variables (producion, prices, real exchange rae and foreign ineres rae), which respecs hree heoreical consrains deduced from he previous heoreical model. The originaliy of our sudy is ha we deal wih 3 developing counries (Morocco, he Philippines, Uruguay) 5 and ha we ake exernal shocks as possible sources of real exchange rae flucuaions ino accoun in addiion o domesic shocks. We use quarerly daa covering he 1979:1 o 1998:1 period. The fourh secion deals wih he economeric evaluaion of he relaive imporance of domesic and foreign ineres rae shocks o economic and real exchange rae flucuaions, as well as o he economic inerpreaion and he comparison of our resul o previous works. A final secion synhesizes our findings and concludes abou he sources of real exchange rae flucuaions in developing counries. 2. The heoreical model of a small open economy This secion presens a simple heoreical benchmark model inspired by Hoffmaiser and Roldos (1997), ha clarifies he long-run deerminans of our variables of ineres. Some economic consrains deduced from his model will be used in he hird secion o specify he 4 These shocks are classified according o wo crieria: domesic-foreign and permanen-ransiory. 5 Le us noice ha in order minimize he disorions relaed o regime changes, our empirical invesigaion has been limied o he counries for which economeric ess confirmed he absence of significan breaks in he daa. This has led us o keep only 3 counries (Morocco, he Philippines, Uruguay) o analyze he dynamic relaionships among our 4 variables, whereas our iniial daabase conained 7 developing counries (Korea, Malaysia, Morocco, Mexico, he

6 idenificaion scheme of he four srucural shocks (exernal shock, offer and demand shocks, nominal shock). These economic consrains will enail resricions on he long-run parameers of he VAR model. Consequenly, he shor-run dynamic which is ofen subjec o conroversies will no be consrained here, bu will be compleely deermined by daa. The economy produces radable and non-radable goods. The radable secor requires capial ( K ) and labour ( L ) o produce a quaniy of oupu defined by: e Q = Y = A K L (1) 1 α e e e e where A e represens he echnological level of he secor and denoes ime. The producion of non-radable goods is assumed o require only labour ( L n) and is defined as follows: : Qn = Yn = AnL β n (2) Alhough his specificaion of he non-radable goods producion echnology is somehow resricive, i permis o reproduce boh he labour inensiy which is relaively higher wih regard o he radable secor and he low conens in inermediae goods. The oal oupu expressed in erms of radable goods of he economy is given by: Y = Ye + py n (3) where, p is he relaive price of he non-radable goods in erms of radable ones, which represens for us he real exchange rae. Philippines, Tunisia, Uruguay).

7 The equilibrium on he labour marke allows us o deermine he following relaionship beween he labour produciviy raio of he wo secors and he real exchange rae, ha is : α 1 α Al e e p = β A ( L l K) n e β 1 (4) where l e = L / K and L denoe respecively he opposie of he per capia capial in he radable e secor and he labour oal offer in he economy. Privae agens have access o he inernaional capial marke where hey can borrow a quaniy D a he world ineres rae r. A he saionary sae he sock of foreign asses is consan, which implies he following equilibrium relaionship: Y r D= C (5) e e where C e is he radable goods consumpion. A he equilibrium, radable goods producion excluding deb services is equal o he consumpion of radable ones. Perfec capial mobiliy a he world level permis o equalize capial marginal produciviy o he world ineres rae, ha is: α (1 α) A e l e = r (6) An increase of foreign ineres rae will have a downward effec on producion since i will enail a capial fligh and hence a decrease in radable goods producion. To ake demand shocks ino accoun we assume ha public spending only concerns radable goods; Hence equilibrium on he non-radable goods marke implies ha: Cn = Yn(1 g) (7) where C n and g denoe respecively non-radable goods privae consumpion and he share of public consumpions in he non-radable goods global oupu.

8 In his model he effec of a fiscal policy is o modify he composiion of demand and producion in favour of non-radable goods, wih an ambiguous effec on global oupu. We assume ha he long-run effec of an expansionis fiscal policy on he oupu is low and is no significanly differen from zero 6. If we denoe by s n, he share of non-radable producion in oal oupu, and by λ n he share of labour in he non-radable secor, he logarihmic expression of oal oupu is given by : y =Ω+ a + ( α s / λ )log( l ) + (1 s / λ )log( K ) (8) e n n e n n This equaion enables us o deermine he facors which are likely o affec oupu in he long run. A echnological shock in he radable secor leads o a radable producion increase and enails a rise in oal oupu. A posiive shock on he ineres rae leads o a capial fligh and hence o a producion decrease. The real exchange rae equilibrium level can be expressed as : q =Ω q+ ae an+ (1 β)((1 λ )/ λ )log( K ) ((1 α) (1 β)((1 λ )/ λ )log( l ) (9) n n n n e A posiive offer shock resuling from a echnical progress in he radable secor enails a real exchange rae appreciaion. The posiive wealh effec which accompanies produciviy earnings leads o an increase in he non-radable goods demand and hence o a rise in he nonradable goods relaive price. The excess of non-radable goods demand enails a labour reallocaion owards he non-radable secor and hence an offer increase. A public spending rise also enails a real exchange rae appreciaion, public spending being generally largely composed of non-exchangeable goods. An expansionis fiscal policy leads o a labour facor reallocaion owards he non-radable secor, and he equilibrium on he capial markes will be mainained by a decrease of he capial sock. An increase of he foreign ineres rae enails a capial fligh, and hence a labour produciviy decrease in he radable secor;he real exchange rae will end o depreciae. 6 Blanchard and Quah (1989) showed ha idenifying public-spending shock as he one ha has no long-run effec on oupu is robus if is long-run effec is relaively low wih regard o ha of oher shocks.

9 A he saionary sae, he balance of he rade balance can be expressed as follows 7 : BC = ( Y C ) =Ω + log( r ) + log( K ) (10) e e bc A posiive shock on he foreign ineres rae enails a higher surplus of rade balance. Indeed, he defici of capial balance induced by a capial fligh is compensaed by a posiive rade balance. A public spending increase leads o a decrease of capial physical sock and hence o a rade balance decrease. As i is usually done in real business cycle lieraure, we assume ha nominal shocks have no long-run effec on real variables. However, nominal variables are affeced in he long run by oher real variables. This effec can be passed on hrough money demand or / and hrough feedback effecs conneced o indexaion pracices and nominal anchoring. To specify beer he naure of nominal shocks and o es he relevance of our resuls we also esimae our model in he economeric par by subsiuing he nominal exchange rae for he price index. 3. Economeric mehodology and shock idenificaion We show in his secion ha by imposing six long-run resricions on our VAR model, we can idenify he four srucural shocks as an exernal shock, a shock on offer, a shock on demand and a nominal shock. In radiional macroeconomic IS-LM models hese shocks can be inerpreed as shocks affecing he goods marke (movemen of he IS curve), shocks affecing he moneary marke (movemen of he LM curve), shocks on he oupu capaciy and shocks on he world marke. The resricions imposed on he long-run parameers of he sysem ha we use o idenify he srucural shocks, are based on he heoreical model presened in he previous secion. The basic assumpion ha we adop here o differeniae he demand shock from he offer shock is ha in he long run oupu is exclusively deermined by offer facors (produciviy and foreign ineres rae shocks). Hence, we suppose ha in he long run oupu is a is full 7 We hence suppose ha a he saionary sae domesic physical capial is perfecly subsiuable o foreign iles.

10 employmen level and hus only depends on offer facors. However in he shor run, given price rigidiy, he four sources of impulsion can have an effec on oupu. For counries open and srongly dependen on foreign economies a foreign ineres rae shock can affec he economy hrough is effec on he real exchange rae. In paricular, a foreign real ineres rae increase enails a capial fligh and hence an exchange rae depreciaion. Furhermore, he negaive wealh effec which follows he decrease in he foreign ineres rae (decrease in he foreign asse reurn) enails an expendiure decrease and hence a real exchange rae depreciaion (via he non-radable goods price decrease). To ake he indirec effec on he economy ino accoun which can lead o foreign ineres rae variaions, we do no impose any resricions on is impac on he real exchange rae. The exernal shock is idenified as being he only shock ha can have a long-run effec on he foreign ineres rae. This resricion is jusified by he fac ha we deal here wih small economies whose marke power is relaively low on he inernaional. I is imporan o noice however ha he exernal shock is no he only one ha can have an effec on he economy hrough he exchange rae. For an open economy, he hree oher shocks can also have an effec via he exchange rae. In order o differeniae he real shock from he nominal one, we suppose ha he laer has no long-run effec on he real exchange rae (Clarida and Gali, 1994). Hence, we suppose ha real exchange rae variaions are boh cyclic movemens and permanen deviaions o PPA. Nominal shock enails a price increase and a proporional depreciaion of he nominal exchange rae. As a resul he real exchange rae remains sable in he long-run as prediced by Dornbush and Mundell-Fleming s models. As he hree real shocks can have a permanen effec on he real exchange rae we do no impose resricions on heir long-run effecs. In paricular, according o he macroeconomic models of an open economy, a posiive demand shock enails a real exchange rae appreciaion (hrough price increase). The effec of a posiive offer shock on he real exchange rae is on he conrary difficul o predic. On he one hand, produciviy increase enails a permanen price decrease and hence a real exchange rae depreciaion. On he oher, he posiive wealh effec which accompanies a produciviy shock leads o a non-radable goods price increase and hence o a real exchange rae appreciaion.

11 3.1. Economeric mehodology To specify he various shocks (exernal shock, offer and demand shocks, nominal shock) we use here a VAR model composed of four variables (foreign ineres rae, GDP, he real exchange rae and prices) aken in firs differences 8. The economy is hence represened wih a vecor of four observable series X = ( r, Y, cr, P) a every dae, resuling from he dynamic combinaion of a vecor of r o g n four pas srucural shocks ε = [ ε, ε, ε, ε ] ; where r ε, ε o, ε g exernal offer, demand and nominal srucural shocks. Le us consider he following model: e n ε denoe respecively he X = θ + H(0) ε + H(1) ε 1... (11) X = θ + H() i ε (12) 0 i For convenience he expression is rewrien as follows: X = θ + H( L) ε (13) where L denoes he lag operaor and where H is a (4 4) marix which deermines he dynamics of he four componens of he X vecor following exogenous shocks. The insananeous effec of ε on X is given by H(0) (i = 0) ; he laged effecs of ε on X are given by H(i), i >0. As X is supposed o be saionary, no shock has a long-run effec on he foreign ineres rae, he GDP, he real exchange rae and he price raes of variaion. On he conrary only exernal shocks have a long-run effec on he ineres rae level, nominal shocks have a long-run effec only on he price level and demand shocks have no long-run effec on he GDP level. The srucural moving average represenaion (13) has a srucural auoregressive represenaion wih p lags defined as follows: 8 The saionariy of each variable as well as he exisence of coinegraion relaionships among hese variables is examined in secion 4. I appears ha all variables are inegraed of order 1 bu no coinegraed, which jusifies he

12 B(0) X = a+ B(1) X +... B( P) X + ε (14) 1 p where ε is a whie noise. Muliplying each elemen of (14) by 1 B(0), enables o wrie : X = c+ A(1) X A( p) X + u (15) 1 p X = c+ A() i X + u (16) i 0 ALX ( ) = c+ u (17) where A(0) = I 1 wih c= B(0) a Ai B Bi and 1 () = (0) () 1 u = B(0) ε The VAR Represenaion given by (17) is he reduced form of he general dynamic model (13). As ε is a whie noise, i is also he case for u (since u j is a linear combinaion of srucural shocks). Given ha he dynamics is saionary, he u are defined as canonical innovaions, ha is as he par of X which canno be explained linearly by is own hisory. u is a whie noise wih a variance - covariance marix given by V( u ) = Σ. The vecor of canonical innovaions u is p defined as: u$ = X j j A X ji j1, j = 1234,,, (i is he resul of he insananeous combinaion of i= 1 srucural shocks). A each dae, innovaions are esimaed as he residuals of regressions corresponding o he esimaion of he VAR model. Canonical innovaions are he smalles unpredicable pars of he wo series a dae, given he informaion relaive o he se of he pas values of he X vecor a dae. In his sense, hey are represenaive of surprises which use of a VAR model.

13 resul from shocks [Bruneau and De Brand (1999)]. The moving average represenaion which corresponds o a Wold decomposiion of he dynamics is he resul of he inverse funcion of equaion (17), i.e. : X = AL c+ u (18) 1 ( ) ( ) X = d + C( L) u (19) wih 1 d A( L) c = and AL 1 ( ) = CL ( ) X = d + C() i u (20) i= 0 i X = CLu ( ) (21) where C(0) = I. The responses of wo series o innovaions are deduced from he dynamic mulipliers C. From (13), (17) and (21), we can noice ha: u = B ε = H ε = Pε (22) 1 (0) (0) where P = H(0) = B(0) 1 H(0) represens he insananeous impac of a shock ε on innovaions are linear combinaions of srucural shocks. From wha precedes we obain he following relaions : X. A each dae canonical X X = H( L) ε (23) = CLu ( ) (24) u = Pε (25)

14 Combining hese equaions we ge : H( L) C( L) P = (26) As we can obain an esimaion of C (L) marix (by invering he marix A of he esimaed VAR) and an approximaion of he u vecor of innovaions (wih he residuals of he esimaed VAR), he knowledge of he P marix is sufficien o idenify boh he srucural shocks (by muliplying (25) by P 1 ) and he H(L) marix which describes he dynamic impac of hese shocks on he observable variables (23). From he residual of he r, Y, cr and P equaions we ry o idenify he four shocks (exernal, offer, demand and nominal) Shock idenificaion Our aim here is o idenify he ransformaion marix P. To do so, i is necessary o impose n n, ha is 16 idenifying consrains o be able o esimae he simulaneous equaion sysem corresponding o he srucural VAR model. I is usual o suppose ha srucural shocks ε are uncorrelaed and have a uniary variance. We impose by his hypohesis alone hree bilinear consrains on he elemens of he P marix. Equaion (25) implies he following relaionship beween he variance -covariance marix of he innovaions (V( u ), and ha of srucural shocks V( ε )) : Vu ( ) = PV( ε ) P (27) As he wo srucural shocks are assumed o be uncorrelaed, heir variance-covariance marix is uniary (V( ε ) = I). Equaion (27) can be hen rewrien as: Vu ( ) = PP =Σ (28) 9 Srucural shocks are said o be idenified when we can esimae hem from he observable series for he considered

15 The V( u ) marix being symmeric, equaion (28) includes n (n+1)/2 independen non linear consrains on he componens of he P marix; hese en consrains are of saisical naure 10. Six economic consrains remained hen o be imposed. Blanchard and Quah (1989) proposed ha he idenifying consrains concern he shor and\or long-run effecs of srucural shocks on he various componens of he sysem. When he dynamics is saionary, only shorrun consrains can express he absence of insananeous response o some srucural shocks. On he conrary, long-run consrains can only concern he response of a saionary series in firs difference and in any case he response of a saionary series [Bruneau and De Brand (1999)]. In he presen case he sysem is supposed o be saionary in firs differences (cf. he resuls of he economeric par ); We only consider long-run consrains deduced from he previous heoreical model (18). The firs consrain implies ha inernal shocks do no affec he exernal variable, namely he ineres rae. This resricion implies ha he exernal variable should be exogenous and ha he accumulaed effec of domesic shocks on he ineres rae variaion is equal in zero. This assumpion hus provides us wih hree addiional resricions. The second resricion implies ha nominal shocks do no affec in he long run real variables and ha public spending shocks do no affec he oupu in he long-run. This provides us wih he las hree consrains. These resricions can be expressed as follows: h = h = h = 0 (29) 12i 13i 14i i= 0 i= 0 i= 0 h = h = 0 (30) 23i 24i i= 0 i= 0 h34i = 0 (31) i= 0 period. 10 From an economic poin of view hese consrains res upon a srong assumpion : here is lile reason o hink ha an offer shock is a each dae uncorrelaed wih a demand shock. This is probably one of he weakness of he srucural VAR mehodology [Bruneau and De Brand, 1999].

16 The esimaion of he P marix is made possible by he imposiion of 16 consrains (10 saisics and 6 economic ones), wha allows o idenify he srucural shocks and he dynamic impac of hese shocks on he observable variables. I remains o noice ha he assumpion of an absence of correlaion beween he shocks and noably beween he demand and offer shocks is a srong hypohesis. Le us underline, however, ha his assumpion does no consrain he canals by which he wo shocks affec GDP and he real exchange rae. Besides, i is necessary o noe ha a demand shock can also have a permanen impac on GDP: variaions of he realizaion rae or changes in fiscal policy can affec he savings rae and hence he produc in he long erm. Blanchard and Quah (1989) admied hose impacs of demand shocks. However, hey posulaed ha he long-run effecs of demand shocks on GDP are low in comparison o he long-run effecs of offer shocks. From his posulae, hey demonsraed ha he idenificaion procedure is nearly correc insofar as he procedure used (null effec in he long-run of demand shocks on GDP), is close o he one ha should have been used (long-run effec of demand shocks on GDP bu whose size is lower han ha of offer shocks). For our analysis, we classify he shocks ino 3 caegories: nominal and real shocks; domesic and exernal shocks; permanen and ransiory shocks. Permanen shocks are hose which have a permanen effec on he real exchange rae and on oupu, namely offer shocks, demand ones and exernal ones (shocks on foreign ineres rae). The offer shock, which has a long-run effec on oupu, can come from a srucural change such as an economic liberalizaion. The relaive demand shock can have for origin public spending. Finally, he nominal shock which will only have a long-run effec on prices can be considered as a moneary one.

17 4. The economeric invesigaion of he sources of real exchange rae flucuaions We esimae a firs model composed of four variables, namely foreign ineres rae, oupu, he real exchange rae and prices and, a second one where prices are replaced by he nominal exchange rae. The counries chosen for he empirical evaluaion are Morocco, he Philippines and Uruguay. We use quarerly daa covering he 1979:1 o 1998:1 period. Hence he X vecor of daa analyzed for each of he hree counries, is given by : X = ( r, Y, cr, P) (32) where r denoes he log of he real ineres rae in he USA, Y he log of GDP, cr : he log of he real exchange rae, and where P is he log of he consumpion price index. For some counries we approximae GDP by he manufacuring producion given he unavailabiliy of quarerly daa for income. We choose he mulilaeral index expressed as he raio of domesic and foreign consumpion price indices as an indicaor of he real exchange rae. The real exchange rae is defined wih regard o he main rading parners of each counry, and an increase implies an appreciaion. The choice of he mulilaeral exchange rae is jusified by he fac ha i is he bes indicaor allowing o give a good picure of he evoluion of he exernal rue value of he naional currency and economic aciviy (Bankim and Prasad, 1997). We use he general price index o calculae he real exchange rae. The daa are exraced from he Agénor e al (2000) and from he IMF daabase. We examine in a firs sep he univariae properies of macroeconomic series, and we es in a second one he exisence of coinegraing relaionships beween he componens of he vecor for each of he 3 counries Sochasic properies of daa X In his paragraph we invesigae he sochasic properies of each series for he 3

18 counries. To deermine he mos appropriaed economeric specificaion o daa properies, i is necessary o es wheher he variables are saionary or inegraed of order 1. To do so, we implemen he uni-roo ess of Kwiaowski, Phillips and Shin (1992 ), hereafer KPSS, as well as he efficien es proposed by Ellio, Rohenberg and Sock (1996), hereafer ERS. The resuls of hese ess repored in appendix (cf. able 1) clearly indicae ha all our macroeconomic series are inegraed of order 1, since in mos cases, he null hypohesis of uni roo canno be rejeced a he 5 % level of significance. The nex sage consiss in esing he exisence of a coinegraing relaionship beween he four variables aken in level (for each of he 3 counries). As i is generally recognized, if he variables are coinegraed, a VAR model in firs difference is no be he appropriae specificaion o analyze he sources of impulsions for he differen variables, bu a VAR-ECM model mus be used. The procedure developed by Johansen and Juselius (1990) allows o es for he number of coinegraing relaionships beween our sysem of variables, using he well known following wo es saisics: he race and he maximal eigen value ess. For his purpose, he basic represenaion of he n dimensional rewrien as he above error correcion model: X vecor ( X = AX A X + µ + ε ) can easily be k k X = Π X Π X + Π X + µ + ε (33) 1 k+ 1 k+ 1 k wih Π = ( I A1... A k ) and Π i = ( Ai Ak) for i = 1... k where is he differeniaion operaor. The Π represens he long-run dynamics of he X sysem of hree variable while he Π i deermines he shor-run dynamics. ε is a whie noise. The coinegraion hypohesis is formulaed as he rank (number of independen line vecors ) of he Π marix : H () r αβ 1 :Π= (34) where α and β are ( p r) marices of rank r p, and α he marix conaining he coinegraing vecors (r being he number of coinegraing relaionships and p he number of

19 variables). Since he H () 1 r hypohesis implies ha he X process is saionary and as X is, hen he β X combinaion is necessarily saionary. The β X marix conains he long-run relaionships beween he four variables. The rank of he Π marix deermines he number of long-run relaionships. More exacly, his r rank can be invesigaed using he wo following well-known es saisics proposed by Johansen-Juselius-90: Trace es = 2ln QH ( ( r )/ H ) = T ln(1 λ ) 1 0 i= r+ 1 3 r max λ es = 2ln QH ( 1( r 1)/ H1( r)) = Tln(1 λ r ) The race saisic allows o es he null hypohesis ha he rank of he marix is equal o r agains he alernaive ha he rank is p 1 and he maximum eigenvalue saisic ess he null hypohesis ha he rank of he marix is r agains he alernaive ha he rank is r + 1. The asympoic disribuions of hese es saisics are non-sandard and can be found in Johansen (1988) and Johansen and Juselius (1990). The opimal number of lags o inroduce ino he model is seleced according o a sequenial procedure which consiss in esing he null hypohesis of k lags agains he alernaive of q lags, wih q < k (he es saisic having in his case a Chi Square disribuion wih n n ( k q) degrees of freedom, n being he number of variables). For our 3 counries he race and maximum eigenvalue saisics 11 boh indicae a he 5 % level of significance he absence of a coinegraing relaionship beween he r, Y, cr and P variables over he 1979:1 o 1998:1 period (cf. able 2 in appendix). Consequenly, he economeric modelling compaible wih he daa propery is a VAR model for he four variables aken in firs differences, for he 3 counries. 11 In his kind of procedure he choice o inroduce or no deerminis componens ino he coinegraing vecors is raher complicaed. Indeed, he bad specificaion of he long-run relaionship can lead o wrong conclusions on he naure of he dynamic which characerizes he movemen of he differen variables. Hence, a preliminary exercise consiss in idenifying he underlying model. In his opics Johansen (1992) developed a sequenial procedure o idenify he rue model. The implemenaion of his procedure has finally led us no o include any deerminis componen.

20 4.2. Sabiliy ess One of he mos imporan seps o underake before proceeding o he examinaion of he impulse response funcions and he variance decomposiion of forecas error, is o make sure ha he esimaed economeric relaionship is sable over he period of sudy. To do so, we have o examine wheher he esimaed parameers of his relaionships are robus or no. Indeed, hese parameers may change over ime because of economic policy changes (modificaions of exchange rae regimes) or because of inernal or exernal economic modificaions. However, in some cases economic insabiliy does no enail significan exchange rae regimes alering he naure of he relaionship beween variables. To examine he robusness of he relaionship beween he four variables, we carried ou he wo well known sabiliy ess proposed by Chow (he1-sep es, and he N-sep es) and we concluded ha he esimaed parameers of ou rvar models are sable over ime for Morocco, he Philippines and Uruguay Impulse response funcion analysis Figure 1 in appendix provides for each of he hree counries he impulse response funcions of he four variables o he four srucural shocks. I depics he response of each variable aken in level o a 1 % shock as well as he confidence inerval 13. The examinaion of he impulse response funcions enable o evaluae he relevance of he idenificaion scheme. As our variables are expressed in firs difference, he analysis of he effecs of he differen shocks on he variables in level is made on he basis of he accumulaed response funcions. 12 I is imporan o emphasise ha acually our original daabase conained 7 developing counries (Korea, Malaysia, Morocco, Mexico, he Philippines, Tunisia, Uruguay). However on he basis of he Chow ess we found ha in Tunisa, Malaysia, Korea and Mexico he esimaed parameers of he VAR model were subjec o srong variaions according o he sample period considered. This led us o exclude hese counries from our sample and o focus exclusively on Morocco, he Philippines and Uruguay o analyze he dynamic relaionships among our 4 variables. 13 The confidence inerval is calculaed using he boosrap mehod.

21 The effec of a shock on he foreign ineres rae According o heoreical expecaions a posiive shock on he foreign ineres rae enails a producion decrease and a real exchange rae depreciaion. The response of GDP and of he real exchange rae is almos idenical in he hree counries. Indeed, a foreign ineres rae increase leads o a decrease of abou 0.1 % of GDP and of he real exchange rae. However, in Uruguay a small insananeous increase of GDP can be noiced ha vanish in he firs period; he adjusmen o he equilibrium sae is however raher slow. Prices increase in Uruguay and in he Philippines; This rise compleely disappears a he 13h period in Uruguay. For Morocco, on he conrary he long-run effec of he exernal shock on prices is negaive bu no significan. Hence he wealh effec which accompanies he ineres rae increase doesn seem imporan enough o enail a demand increase The effec of a shock of offer The response of he real exchange rae and GDP o a shock of offer confirms heoreical predicions. Indeed, a posiive shock of offer enails a GDP increase and a real exchange rae appreciaion - The offer shock affecs in he same way all secors of he economy; - The offer effec which accompanies he offer shock dominaes he wealh effec (demand effec).. We can noiced ha he long-run response of GDP is relaively smaller han is shor-run response and ha adjusmen o equilibrium is relaively slower han ha of he real exchange rae. A shock of offer of 1 % enails a GDP long-erm increase of 0.8 % in Uruguay, 0.6 % in Morocco and 0.9 % in he Philippines 14. Besides, i seems ha he real exchange rae sensibiliy o a shock of offer is relaively higher in he Philippines han in he oher counries: he real exchange rae increases by % in he Philippines and only abou 0.1 % in Morocco and in Uruguay. The price response o a shock of offer is a priori ambiguous since i is posiive in Uruguay and negaive in Morocco and in he Philippines. The price decrease in he Philippines

22 and Morocco reflecs well he predominance of he direc effec of offer on he effec of demand. However in Uruguay i is he effec of demand ha is dominan The effec of a shock of demand A posiive shock of demand enails a price increase and a real exchange rae appreciaion, which confirms ha i is effecively a shock of public spending. However, he producion decrease in Uruguay and in he Philippines is a priori conradicory o he heoreical predicions and needs consequenly o be accouned for 15. Acually, in he heoreical model explained previously, he long-run effec of a shock of demand on GDP is ambiguous. Two explanaions can be proposed. Firsly, as public spending is generally relaed o non-radable goods, a public spending increase enails a resource reallocaion and a decrease of non-radable goods offer 16. Secondly, he expeced view of fiscal policy approach predics he recessionis effec of an expansionis public fiscal policy (see Giavazzi and Pagano, 1990). Following a posiive shock of demand, prices increase and reach heir long-run level in abou 4 and 20 quarers. Prices seem sicker in Uruguay bu heir response is relaively higher (2 % agains % in he Philippines and 0.4 % in Morocco). Following a posiive shock of demand he real exchange rae appreciaes and he adjusmen o he equilibrium sae is progressive (no overshooing), relaively fas in Morocco and in he Philippines and akes beween wo and four quarers; adjusmen is relaively slower in Uruguay. If we compare he responses of he real exchange rae and prices, we noice ha a posiive shock of demand also leads o a nominal exchange rae appreciaion The effec of a nominal shock The effec of a moneary shock on he real exchange rae, he nominal exchange rae and prices is in accordance wih heoreical predicions. Indeed, a posiive nominal shock enails a 14 Kamin and Roger (1997) obained a similar resul for Mexico. 15 Hoffmaiser and Roldos (2001) obained a similar resul for Brazilian economy. 16 In our sudy he GDP is approximaed by he indusrial producion which is composed of radable goods.

23 emporary (respecively permanen) depreciaion of he real (respecively nominal) exchange rae and a permanen price increase. The price adjusmen o a nominal shock is relaively low in Uruguay and akes more han 20 quarers. However, for Morocco and he Philippines, prices ake 2 o 4 quarers o reach heir long-run level. Furhermore, he response of prices is relaively high in Uruguay, abou 3 % whereas i is only 1.5 % in he Philippines and 1.2 % in Morocco. The emporary depreciaion of he real exchange rae is relaively fas in Morocco and in he Philippines and he nominal shock effec vanishes a he 5h quarer. However, in Uruguay he real exchange rae needs more ime o go back o is equilibrium level. The real exchange rae adjusmen which is relaively slower in Uruguay, is compaible wih he price response o a nominal shock. I can also be explained by he fac ha in Uruguay he nominal exchange rae is no perfecly correlaed o prices as i is he case in Morocco and in he Philippines. Conrary o he heoreical expecaions for a closed economy he GDP response o a posiive nominal shock is negaive in Uruguay, whereas heoreically i is expeced o enail an economic expansion. Acually, he GDP response o a posiive nominal shock confirms he idea of he recessionis effec of an exchange rae depreciaion (Lizondo and Moniel, 1989). 4.4 The variance decomposiion of forecas error The variance decomposiion of forecas error enables o deermine he mos imporan flucuaion sources of he endogenous variables for he period of sudy. In paricular, he variance decomposiion of forecas error permis o measure he par of he anicipaed variance of each endogenous variable explained by he differen shocks for he differen horizons. We calculae he conribuion of he exernal offer, demand and nominal shocks o he variaions of he level of each variable 17. The resuls are repored in he appendix (cf. ables 3, 4 and 5) for a horizon of 25 quarers. The variance decomposiion of forecas error confirms ha for all counries he GDP growh rae is mainly deermined by domesic shocks and ha he exernal shock only plays a 17 The variance decomposiion of forecas error of he variables aken in firs difference can be differen from ha of he variables in level. In fac, he variance decomposiion of forecas error of he variable in level (respecively in firs difference) is based on a non-linear ransformaion of he impulse response funcions of he variables in level (respecively in firs difference) for he differen shocks.

24 minor role. Indeed, he conribuion of domesic shocks o GDP flucuaions is abou 98 % for he hree counries, boh for he shor and long-run. Among he domesic shocks, i is he shock of offer which dominaes GDP flucuaions (97 % in Morocco, 82 % in Uruguay and 84 % in he Philippines). Nominal shock plays a significan role in GDP flucuaions, wih a share close o 14 % in Uruguay and 6 % in Philippines. The conribuion of he shock of demand is raher limied and does no exceed 8 %. For he real exchange rae, he analysis of he variance decomposiion of forecas error indicaes ha for he hree counries he shocks of demand dominae real exchange rae flucuaions, represening 94 % in Morocco, 85 % in he Philippines and 75 % in Uruguay. In second posiion we find he nominal shock which explains 14 % of he real exchange rae variaions in Uruguay and 10 % in he Philippines. However, in Morocco, he conribuion of he nominal shock is very low, wih a share which does no exceed 4 %. The shock of offer has an insignifican effec in Morocco and is limied in Uruguay and in he Philippines, wih respecive shares of 6 % and 10 %. Similarly, he exernal shock conribuion is raher limied wih a par which does no exceed 10 %. As expeced, mos price variaions are explained by nominal shocks, wih a share of 65 % in Morocco, 54 % in Uruguay and 45 % in he Philippines. The shocks of offer play a relaively imporan role in price flucuaions, noably in he Philippines and in Uruguay wih a share of 35 %. The shock of demand explains a relaively low bu significan par of price variaions, represening beween 14 % in he Philippines and 18 % in Morocco. However he exernal shock seems o conribue very lowly o price flucuaions, wih a share which does no exceed 8 %. These resuls are compaible wih hose of Clarida and Gali (1995) and Asley and Garra (2000) who noiced ha for a sample of developed counries real exchange rae flucuaions are generally dominaed by he shocks of demand. Besides, Hoffmaiser and Roldos (1999) found similar resuls for a panel of Asian and Lain American counries. Indeed, Hoffmaiser and Roldos (1999) considered ha GDP flucuaions are compleely explained by domesic shocks. Among domesic shocks, i is he shock of offer which is he major source of GDP flucuaions wih a share close o 90 % for Asia and 60 % for Lain America, boh in he shor and long-run. The shock of foreign ineres rae only explains 5 % of GDP long-run flucuaions; he shor-run effec is low. For Asia hey noiced ha a large par of real exchange rae flucuaions can be explained by he shock of demand and ha he conribuion of he exernal shock is low. The

25 conribuion of exernal shocks on he ineres rae is, on he oher hand, no significan for Lain America. These resuls confirm ha he variaions of he non-radable goods relaive price dominae real exchange rae flucuaions and ha he deviaion o PPA for radable goods only plays a small role. The predominance of real shocks which can be explained by he imporance of srucural changes, implies ha he real exchange rae behavior depends on economic specificiies. I is he inernal dynamics of each economy, he behavior of agens abou consumpion and invesmen as well as he way budgeary policy in conduced which explain real exchange rae variaions. However, hese conclusions do no imply ha exernal and nominal shocks are no poenial sources of real exchange rae flucuaions. Impulse response funcions clearly indicae ha he real exchange rae is also sensiive o nominal and exernal shocks. 5. Conclusion Wha are he major sources of real exchange rae flucuaions in developing counries? I is hose quesions ha we have ackled in his paper in order o underline he relaive imporance of he differen sources of impulsion. Our economic resuls sugges ha domesic shocks dominae real exchange rae flucuaions and ha he conribuion of exernal shocks is raher low. These resuls are in accordance wih hose of Hoffmaiser and Roldos (1999) for developing counries and wih hose of Froo and Rogoff (1991) for developed counries. Among domesic shocks he shocks of demand essenially accoun for real exchange rae variaions. However, ransiory shocks only lowly conribue o real exchange rae variaions boh for he shor and long-run. I is imporan o noice ha he low conribuion of nominal shocks o real exchange rae flucuaions is no alered by he choice of nominal variables. Indeed, we carried ou similar exercises of variance decomposiion of forecas error by replacing prices by he nominal exchange rae and we came o he same conclusion, i.e. ha nominal shocks lowly conribue o real exchange rae variaions. This resul goes agains he models which res upon nominal shocks o accoun for real exchange rae flucuaions in developing counries. Besides, he low conribuion of he nominal shock calls ino quesion moneary policies which seek o promoe compeiiveness hrough currency devaluaion.

26 If he conribuion of exernal shocks (shocks on he foreign ineres rae and on he erms of exchange) is low, i does no imply however ha exernal facors have no role o play in real exchange rae flucuaions. Indeed, he impulse response funcions o shocks confirm ha he real exchange rae also depends on exernal shocks and ha he shocks on foreign ineres rae and/or on he erms of exchange can make i move from is equilibrium level. The budgeary ool herefore remains efficien o sabilize he real exchange rae wih respec o possible exernal shocks. References [1] Asley M. e A. Garra (2000), Exchange Raes and Prices : Sources of Serling Real Exchange Rae Flucuaions , Inernaional Economic Review, pp [2] Agénor P.R. e P. Moniel, eds (1997), Developmen Macroeconomics, New Jersey, Princeon Universiy Press. [3] Agénor P., J. MC Dermo e E. Pradsad (2000), Macroeconomic Flucuaions in Developing Counries : Some Sylised Facs, World Bank Economic Review, Vol. 14, N o 2, pp [4] Bankim C. e Prasad E. (1997), Real Exchange Rae and he Business Cycle : Evidence from Japan, IMF Saff Paper, Vol. 44, pp Barro (1999), Noes on Growh Accouning, Journal of Economic Growh, vol 4, n o 2. [5] Blanchard O. e D. Quah (1989), The Dynamic Effecs of Aggregae Demand and Supply Disurbances, American Economic Review, Vol. 79, pp [6] Bruneau C. e De Brand O. (1999), La Modélisaion VAR Srucural Applicaion à la Poliique Monéaire Francaise, Economie e Prévision, Vol. 137, pp [7] Clarida R. e J. Gali (1994), Sources of Real Exchange Rae Flucuaions : How Imporan are Nominal Shocks? Carnegie Rocheser Conference Series on Public Policy, Vol.41, pp [8] Ellio B., Rohenberg e Sock J. (1996), Efficien Tess for an Auoregressive Uni Roo, Economerica, Vol. 64, pp [9] Froo, K. e K. Rogoff (1991), Governmen Consumpion and he Real Exchange Rae: The Empirical Evidence, mimeo, Harvard Business School. [10] Giavazzi F. e M. Pagno, (1990), Can Several Fiscal Conracions be Expansionary? Tales of Two Small European Counries, In NBER Macroeconomic Annueal, edié par O; Blanchard e S; Fisher, pp , Cambridge, Mass. [11] Hoffmaiser A. e J. Roldos (1997), Are Business Cycles Differen in Asia and Lain America, IMF Working Papers. [12] Hoffmaiser A.W. e J. Roldos (2001), The Sources of Macroeconomics Flucuaion in Developing Counries : Brasil and Korea, Journal of Moneary Economics, pp [13] Johansen S. e K. Juselius (1990), Maximum Likelihood and Inference on Coinegraion wih Applicaions o he Demand for Money, Oxford Bullein of Economics and Saisics, Vol. 52, pp

27 [14] Johansen S. (1988), Saisical Analysis of Coinegraion Vecors, Journal of Economic Dynamics and Conrol, vol 12, pp [15] Johansen S. (1992), Deerminaion of he Coinegraion Rank in he Presence of a Linear Trend, Oxford Bullein of Economics and Saisics, Vol. 54, pp [16] Kamin S. e J. Rogers (1997), Oupu and he Real Exchange Rae in Developing Counries : An Applicaion o Mexico, Board Governors of he Federal Reserves Sysem, Inernaional Finance, Discussion Papers N o 580. [17] Kwiakoski D., P. Philipps, P. Schmid e Y. Shin (1992), Tesing he Null Hypohesis of Saionariy Agains he Alernaive of a Uni Roo : How Sure are we ha Economic Time Series have a Uni Roo? Journal of Economerics, Vol. 54, pp [18] Lizondo S. e P. Moniel (1989), Concracionary Devaluaion in Developing Counries : An Analyical Overview, IMF Saff Papers, Vol. 36, pp

28 Appendix All daa are quarerly and exraced from Agénor e al (2000) and from he IMF daabase. The esimaion period goes from 1979:1 o 1998:1 period. The variables are he following : TCR : he effecive real exchange rae calculaed using consumpion price index, Y : real GDP, P : consumpion price index, r * :world (foreign) real ineres rae (US). Table 1: Uni-roo ess counry variable Ellio-Roenberg and Sock (ERS) s es Schmid-Phillips s es Kwiakowsky, Phillips, Schmid and Shin s es (KPSS) 18 Morocco TCR Y P r * Wihou consan Wih consan Wih a linear rend Wih a linear 19 Wih a quadraic rend l = 8 20 Characerisaion Characerisaion Characerisaion κ Characerisaion κ Characerisaion ηc Characerisaion η Characerisaion I (0) + C I (0) + T The Philippines TCR Y P r * Uruguay TCR Y P r * The criical values a he 5% level of significance arerespecively and for he η µ and ητ ess. 19 The criical value a he 5% level of significance is l represens he number of lags used in he Barle window W (S, l) o calculae he long run variance esimaor», S2 (l). We have chosen here l =8 as advocaed by he auhors. 25