What do Savings-Investment Correlations tell us about the International Capital Mobility of Less Developed Countries?

Transcription

1 Journal of Economc Inegraon 20(3), Sepember 2005; Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal Mobly of Less Developed Counres? Mark J. Holmes Wakao Unversy Managemen School Absrac Ths paper nvesgaes he exen of capal mobly wh respec o less developed counres over he sudy perod For hs purpose, he Feldsen-Horoka equaon lnkng domesc savngs and nvesmen s esmaed. However, here s a novel emprcal approach n hs sudy based on he employmen of panel daa mehods of conegraon esng and esmaon of he long-run savng reenon coeffcen. There s srong evdence of conegraon beween domesc savngs and nvesmen. Panel esmaon based on fully modfed ordnary leas squares ndcaes ha capal s mperfecly moble wh a long-run savng reenon coeffcen of abou one-hrd. Ths low value suggess ha capal mobly, hough mperfec, s noneheless que hgh. Furher esmaon based on sub-groups comprsng Asan and Lan Amercan counres pons owards a smlar degrees of capal mobly. JEL Classfcaons: C5, F3, F4, O5 Key words: LDCs, Capal mobly, Savngs, Invesmen, Panel daa, Conegraon I. Inroducon For a varey of reasons, he exen of capal mobly can have profound macro- and mcroeconomc mplcaons for he well beng of boh developed and less Correspondng address: Professor Mark J. Holmes, Deparmen of Economcs, Wakao Unversy Managemen School, Prvae Bag 3150, Hamlon, New Zealand, Tel: , Fax: , E-mal: holmesmj@wakao.ac.nz 2005-Cener for Inernaonal Economcs, Sejong Insuon, All Rghs Reserved.

2 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 591 developed counres (LDCs). Measurng capal mobly, however, has proved o be problemac. On he one hand, one mgh follow Frankel (1992) and use covered neres pary (CIP) as he mos approprae ndcaor of he degree of fnancal negraon and herefore capal mobly across naonal boundares. Essenally, hs drec approach s based on esng he law of one prce n he conex of dencal fnancal asses where he prce of asses, denomnaed n dfferen currences, wh smlar rsk and maury characerscs end o equaly easly and quckly. For LDCs however, he presence of markes ha are llqud and dffcules n asse comparably conrbue owards daa lmaons ha nhb he formal esng of CIP. An alernave way forward s o consder an ndrec approach ha concenraes on he effecs of capal mobly on macroeconomc aggregaes such as he relaonshp beween domesc savngs and nvesmen [Feldsen and Horoka (1980)]. Here s argued ha here s no need for domesc savngs and nvesmen o be correlaed wh each oher f perfec capal mobly enables arbrage o equalze yelds o nvesors. Domesc savngs reac o nernaonal raes of reurn and so nvesmen s funded from he world capal marke hrough a curren accoun defc. If, however, capal were perfecly mmoble hen one would expec domesc savngs and nvesmen o be characerzed by a correlaon coeffcen of uny. Obanng a correlaon of savng and nvesmen close o one n her cross-secon analyss for sxeen ndusralzed OECD counres for he perod, led Feldsen and Horoka (1980) o rejec he perfec capal mobly assumpon. A number of subsequen analyses usng cross-secon or mes seres daa have confrmed Feldsen and Horoka s (F- H) resuls and have aemped o reconcle hem wh he capal mobly hypohess (see, ner ala, Dooley e al. (1987), Vos (1988), Corbn (2001), Ho (2002)). The majory of exsng work on he savngs-nvesmen correlaon concerns OECD counres wh more lmed work ha addresses he case of LDCs. Moreover, he sudes of OECD counres have generally found a large savng reenon coeffcen compared o a much weaker assocaon beween domesc savngs and nvesmen n he case of LDCs (Coakely e al. (1999), Ho (2002)). The purpose of hs paper s o assess he exen of capal mobly for a large sample of LDCs over he sudy perod of I s possble ha low es power s responsble for denfyng a weak assocaon beween LDC domesc savngs and nvesmen. As argued below, he possbly of low es power jusfes he employmen of panel daa conegraon ess. For LDCs, an nvesgaon no

3 592 Mark J. Holmes he exen of capal mobly s of neres for a number of reasons. Frs, n he conex of foregn capal-led economc growh, capal mobly means ha a savng/nvesmen consran on economc growh can be lfed. Moreover, under he neoclasscal growh model, a relavely poor economy wh a low capal-labour rao wll be characersed by a relave hgh margnal produc of capal and hs can rgger greaer savngs and nvesmen. Even f domesc savngs are nsuffcen o respond o he hgh raes of reurn, he flow of foregn capal wll ensure ha he rae nvesmen wll reman hgh. 1 Of course, he converse case of capal flgh can mean ha nvesmen s rearded. Second, under he Mundell-Flemng (MF) model, perfec capal mobly means ha fscal (moneary) polcy s mos effecve under a floang (fxed) exchange rae. From he pon of vew of polcy-makers, he exen of capal mobly may play a role n he desgn of macroeconomc polcy. Thrd, here are ssues of general macroeconomc nsably assocaed wh capal mobly. There have a mes been massve ncreases n he flows of capal o LDCs (see, ner ala, Dooley e al. (1996)). Much of hs capal s shor-erm n naure and brngs wh many problems. For example, ncreased capal nflows can cause a real apprecaon of he domesc currency and negae he mpac of a nomnal devaluaon. The recevng counry s currency may be subjec o a speculave aack f he shor-erm capal nflows are volale. When he capal accoun drves he curren accoun defc, large capal nflows can lead o an unsusanable curren accoun defc, parcularly f he capal nflow s n response o a consumpon boom. Capal flgh can also have a desablsng effec on neres raes, exchange raes, foregn exchange reserves and money demand (as a resul of currency subsuon beng lnked wh capal flgh). Gven he mporance of measurng LDCs capal mobly and he lack of exsng research n hs area, hs sudy apples panel daa conegraon echnques agans a background of lmed daa. A lmed me seres sample sze reduces he power of conegraon ess hereby makng harder o rejec he null hypohess of non-conegraon beween domesc savngs and nvesmen (perfec capal mobly) agans conegraon alernaves. Creang a larger panel daa se from he ndvdual me seres ncreases he power of es and lessens he lkelhood ha conegraon has been rejeced because he sample sze s small. As well as usng more observaons, he panel daa conegrang echnques explo he cross- 1 In he spr of hs logc, he basc Harrod-Domar growh model can be exended o he open economy.

4 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 593 counry varaons of he daa n esmaon. Ths s he frs sudy o address capal mobly n he case of LDCs hrough he applcaon of panel daa conegraon echnques. No only can we es he null of non-conegraon beween domesc savng and nvesmen (perfec capal mobly) agans he null of conegraon (capal mobly), we are also able o formally es he null hypohess of a longrun uny resrcon on he savng reenon coeffcen ha mples perfec capal mmobly. The paper s srucured as follows. The followng secon dscusses he mehodologcal ssues ha are relevan o hs sudy. We consder he nerpreaon of he basc F-H equaon and oulne he procedure used for he panel conegraon ess and fully modfed ordnary leas squares (FMOLS) esmaon. The parcular panel daa conegraon mehodology ha we employ s based on Pedron (1999, 2001, 2004). The hrd secon dscusses he daa and resuls. We fnd ha LDC capal s mperfecly moble and ha he degree of capal mobly s he same for boh Asan and Lan Amercan counres. The fnal secon concludes. II. Mehodologcal Issues The sandard F-H equaon may be wren as follows: I = α + βs + µ (1) where I s nvesmen and S s domesc savngs (boh expressed as a percenage of GDP). The savng reenon coeffcen s he proporon of ncremenal savngs ha s nvesed domescally and s denoed by β where β=0 (β=1) mples perfec capal mobly (mmobly) whereas 0<β<1 mples mperfec capal mobly. There are, however, several challenges o he F-H nerpreaon of nernaonal capal mobly. Frs, he perssen correlaon beween savng and nvesmen may be no due so much o mperfec capal mobly han o he pro-cyclcal characer of savng and nvesmen n a real busness cycle model. Whle a number of auhors have underaken cross-seconal analyss on sample averages n he perod hus elmnang he nfluence of hese cycles n he savng nvesmen correlaon, permanen nvesmen mus equal permanen savng plus some consan n he long run. If perod averagng s long enough, here mgh be a msspecfcaon because one s esmang an deny n whch he savng reenon coeffcen s equal o uny. Second, proceedng o esmae he F-H equaon hrough me-seres approaches can be subjec o smulaneous equaons bas.

5 594 Mark J. Holmes Thrd, a hgh correlaon beween savngs and nvesmen may acually be ndcave of hgh capal mobly. For example, Murphy (1984) and Baxer and Crucn (1993) sugges ha a hgh domesc savng nvesmen correlaon reflecs he counry s fnancal sze n he world economy. When he counry s fnancal sysem s hghly developed n nernaonal erms, exogenous varaons n domesc savng and nvesmen raes affec world neres raes and nduce jon movemens n domesc savng and nvesmen raes. In hs conex, he F-H es can be nerpreed as a jon es of he hypohess of capal mobly and he sze of a counry s fnancal sysem. In he case of an LDC wh a relavely underdeveloped fnancal sysem, however, hs argumen s lkely o be of less relevance han s he case for developed economes. Fourh, a new nerpreaon of he hgh savngnvesmen correlaon beng suggesed s hrough he acknowledgemen of a solvency consran (no Ponz fnancng). In he long erm, he neremporal budge consran (IBC) s an ndcaor of a counry s solvency expressed n erms of curren accoun consancy, whch can be nerpreed n he Feldsen-Horoka approach as evdence of mperfec capal mobly (Coakley e al., 1996). Moreover, sudes ha denfy a saonary curren accoun balance reflec a savng reenon rao of 1. Thus a hgh correlaon beween savng and nvesmen may be a reflecon of a counry sasfyng s IBC raher han evdence of zero capal mobly. 2 Ths sudy ess for conegraon beween nvesmen and domesc savngs. A major obsacle o dong hs wh respec o me-seres daa s he lack of observaons on S and I. As argued above, one way forward s o conduc he nvesgaon on he bass of a panel daa se. For he purpose of conegraon esng whn he panel of LDCs, we follow Pedron (1999) who proposes a range of sascs ha can be used o deermne he presence of conegraon n heerogeneous panels. 3 These ess do no consran he esmaed slope coeffcens o be same across he panel and are applcable where regressors are fully 2 Usng daa for LDCs over a sample perod, Coakely e al. (1999) denfy curren saonary n 14/44 cases hereby provdng evdence ha less han one-hrd of LDCs conform o he long-run solvency consran. In addon o hs, hey confrm curren accoun saonary n 12/23 OECD economes. 3 Corbn (2001) hghlghs he mporance of conrollng for he heerogeney of counres n a crosssecon analyss of he savng nvesmen correlaon for a group of counres usng panel daa. I s argued ha he ndvdual and emporal dmensons of he daa enables he esmaon of he coeffcen of he savng nvesmen correlaon. Moreover, obanng a hgh coeffcen of correlaon n he crosssecon analyss, may be less due o he exsence of a common characersc affecng all he counres n he sample n he same way n a gven perod (mperfec capal mobly) han o he exsence of specfc ndvdual counry effecs.

6 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 595 endogenous. 4 Whn a panel seng, he es sascs are consruced usng he resduals from he followng hypoheszed conegrang regresson based on equaon (1), I = α + β S + µ (2) where α allows he conegrang regresson o nclude counry-specfc fxed effecs. The procedure for compung he es sascs nvolves esmang he hypoheszed conegraon regresson descrbed n (2) and usng he resduals µ o esmae he approprae auoregresson. From hs, one may compue he panel ADF sasc whch s a paramerc sasc and analogous o he Levn and Ln (1993) panel daa un roo es appled o he esmaed resduals of conegrang regresson. 5 Ths sasc s referred o as a whn-dmenson sasc ha effecvely pools he auoregressve coeffcens across dfferen counres durng he un roo es. A common value for he auoregressve coeffcen s specfed under he alernave hypohess of conegraon. The second sasc s based on a group mean approach. The group ADF sasc s a paramerc sasc and analogous o he Im e al. (1997, 2003) es for a un roo panel ha s appled o he esmaed resduals of a conegrang regresson. Ths sasc s referred o as a beweendmenson sasc ha averages he esmaed auoregressve coeffcens for each counry. Under he alernave hypohess of conegraon, he auoregressve coeffcen s allowed o vary across counres. Ths allows one o model an addonal source of poenal heerogeney across counres. Followng an approprae sandardzaon, boh of hese sascs wll be dsrbued as sandard normal as boh he number of observaons (N) and number of LDCs n he panel (T) grow large. Boh of hese sascs dverge o negave nfny under he alernave hypohess and consequenly he lef al of he normal dsrbuon s used o rejec he null hypohess of non-conegraon. Havng esed for conegraon, he second sage of he nvesgaon s o analyse equaon (2). Pedron (2001, 2004) descrbes how FMOLS procedures can be employed o oban he panel daa esmaes for β. Usng a dynamc modellng 4 See Pedron (2004). 5 In he case of he panel ADF sasc, (2) s also run n frs dfference form where he esmaed resduals are saved. The long run varance of s compued and used as a nusance parameer esmaor n he compuaon of he es sasc.

7 596 Mark J. Holmes procedure resuls n a more powerful es for conegraon as well as gvng generally unbased esmaes of he long-run relaonshp and sandard -sascs. FMOLS amouns o he applcaon of non-paramerc adjusmen o he OLS esmaes of boh he long-run parameer β and assocaed -sasc, on accoun of any bas due o auocorrelaon or endogeney bas ha shows up n he OLS resduals [Phllps and Hansen (1990)]. Followng on from (2), le be a saonary vecor comprs- ng he esmaed resduals and he dfferences n savngs. Also, le Ω = lm T E T 1 T T ξ be he long-run covarance for hs vecor process whch = 1 T ξ = 1 T 0 0 can be decomposed no Ω = Ω + Γ + Γ where Ω s he conemporaneous covarance and Γ ξ = µˆ, S ( ) s a weghed sum of auocovarances. Pedron shows ha he group mean panel FMOLS esmaor s gven as βˆ GFM = N T 1 T ( S S ) 2 ( S S )I Tγˆ N 1 = 1 = 1 = 1 (3) 21 where ( ) and ( Γˆ 22 + ). The I Ωˆ 21 0 Ωˆ = I I S γˆ 21 Γˆ Ωˆ Ωˆ beween-dmenson esmaor s calculaed as βˆ GFM s he convenonal FMOLS esmaor appled o he assocaed -sascs are calculaed as h Ωˆ 22 Ωˆ 22 N 1 N = βˆ FM, where βˆ FM, = 1 member of he panel. The 0 βˆ = N 0.5 βˆ GFM N = 1 FM, (4) where. βˆ = βˆ FM, β 0 FM, T ( ) Ωˆ 11 ( S S ) 2. = 1 In he emprcal analyss, we focus on he beween-dmenson panel FMOLS ess. There are several advanages over he whn-dmenson approach. Frs, he beween-dmenson approach allows for greaer flexbly n he presence of heerogeney across he conegrang vecors where β s allowed o vary. Under he whn-dmenson approach, β would be consraned o be he same value for each counry under he alernave hypohess. Second, he pon esmaes of he beween dmenson esmaor can be nerpreed as he mean value of he conegrang vecors. Ths s helpful n he nerpreng he resuls. Thrd, he

8 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 597 beween-dmenson esmaor suffers from lower small-sample sze dsorons han s he case wh he whn-dmenson esmaor. III. Daa and Resuls The sudy nally employs annual daa on nvesmen and domesc savngs (expressed as a percenage of GDP) for weny-four LDCs over he sudy perod nclusve. The sudy perod s dcaed by daa avalably across hs large sample ha comprses Barbados, Brazl, Chle, Colomba, Cosa Rca, Ecuador, Egyp, El Salvador, Guaemala, Honduras, Inda, Jamaca, Kenya, Mexco, Morocco, Ngera, Paksan, Phlppnes, Sngapore, Souh Afrca, Sr Lanka, Thaland, Uruguay and Venezuela. All daa are aken from he World Developmen Indcaors va he World Bank daabase. Frs of all, he saonary of hese seres s nvesgaed usng unvarae ADF un roo esng. Table 1 repors ha non-saonary n he nvesmen and/or savngs raes s rejeced a he 5 per cen sgnfcance level or sronger n egh cases namely, Brazl, Ecuador, Guaemala, Inda, Kenya, Ngera, Uruguay and Venezuela. 6 These counres are excluded from he man conegraon analyss ha should be based on long-run equlbrum relaonshps beween non-saonary seres. Ths sll leaves more han half he sample characerzed by non-saonary nvesmen and savng raes namely, Barbados, Chle, Colomba, Cosa Rca, Egyp, El Salvador, Honduras, Jamaca, Mexco, Morocco, Paksan, Phlppnes, Sngapore, Souh Afrca, Sr Lanka and Thaland. These sxeen counres are herefore ncluded n he panel daa conegraon analyss. Table 2 repors he Pedron conegraon ess based on equaon (2). These ess nclude me-specfc dummes o allow for he possbly ha resduals are correlaed across counres. The null of non-conegraon s srongly rejeced a he 1 per cen sgnfcance level by boh he panel ADF and group ADF ess respecvely. The former es s relavely more resrcve n ha a common value for he auoregressve coeffcen s specfed under he alernave hypohess of conegraon. The group ADF es allows for he auoregressve coeffcen o vary across counres under he alernave hypohess. Table 3 repors he FMOLS panel esmaes of he conegrang relaonshps beween nvesmen and domesc savng. For he ndvdual counry esmaes, 6 The lag lenghs n hese ADF regressons were deermned by he AIC. There was no affec on he qualave conclusons drawn n hs sudy f alernave mehods for lag selecon such as he BIC were employed.

9 598 Mark J. Holmes Table 1. ADF Un Roo Tess for Indvdual Counres S I ADF ADF (rend) ADF ADF (rend) Barbados Brazl Chle Colomba Cosa Rca Ecuador Egyp El Salvador Guaemala Honduras Inda Jamaca Kenya Mexco Morocco Ngera Paksan Phlppnes Sngapore Souh Afrca Sr Lanka Thaland Uruguay Venezuela Noes for Table 1. These ess employ annual daa on domesc nvesmen and savngs expressed as a percenage of GDP for he sudy perod In all cases, he lag lengh s seleced accordng o he AIC., and respecvely denoe rejecon of he non-saonary null a he 1, 5 and 10 per cen sgnfcance level respecvely. The 1, 5 and 10 per cen crcal values are -3.75, and respecvely for he nonrended ADF ess and -4.38, and respecvely n he rended ADF ess. Table 2. Panel Daa Conegraon Tess Tes Tes Sasc panel ADF group ADF Noes for Table 2. These are he Pedron ess for panel non-conegraon [dscussed n Pedron (1999)] beween nvesmen and savng raes. The panel ADF sasc s a whn-dmenson sascs, he group ADF sasc s a beween-dmenson sasc. Boh ess are asympocally normal. denoes rejecon of he non-conegraon null a he 1 per cen sgnfcance level wh a crcal value of Table 3 repors β > 0 n hreen cases bu here exss consderable varaon n he counry-by-counry experences. The null of a zero slope coeffcen s rejeced a

10 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 599 Table 3. FMOLS Esmaon of he Conegraed Panel βˆ FM, Barbados Chle Colomba Cosa Rca Egyp El Salvador Honduras Jamaca Mexco Morocco Paksan Phlppnes Sngapore Souh Afrca Sr Lanka Thaland βˆ GFM Group Noes for Table 3. Ths able repors FMOLS panel daa esmaes of β n equaon (2) usng he Pedron panel daa conegraon mehodology. These esmaes nclude common me dummes. The boom row refers o he group-mean esmaes. Each esmae of β s accompaned by wo -sascs. Column 3 repors -sascs for he null β=0, whle column 4 repors -sascs for he null β=1. All -sascs are asympocally normal. The -sascs are asympocally normal wh 1, 5 and 10 per cen crcal values of +/-2.57, +/-1.96 and +/-1.64 respecvely. βˆ FM, H 0 :β FM, βˆ GFM H 0 :β GFM = 0 = 0 βˆ FM, H 0 :β FM, βˆ GFM H 0 :β GFM = 1 = 1 he 5 per cen sgnfcance level n all cases excep Barbados, Colomba, Cosa Rca, Egyp, Mexco, Morocco, Paksan, Phlppnes, Sngapore and Sr Lanka. A he 5 per cen sgnfcance level, he null of a uny slope coeffcen and herefore perfec capal mmobly s rejeced n he all cases excep Phlppnes, Souh Afrca and Thaland. Accordng o he group mean esmaes, a long-run relaonshp beween S and I s confrmed wh β= A he 5 per cen sgnfcance level, hs coeffcen s boh sgnfcanly dfferen from zero and uny wh -sascs of and respecvely (see columns 2 and 3). Ths resul ndcaes ha capal mobly s mperfec where wh a savngs reenon rao ha s closer o zero han uny. In comparng hs resuls wh exsng panel conegraon sudes of capal mobly, one may reflec on he sudy by Ho (2002)

11 600 Mark J. Holmes who uses panel conegraon echnques o esmae he F-H equaon n he case of welve OECD economes. 7 Usng daa for , Ho produces a panel esmae of β=0.786 suggesng ha capal s acually less moble for he OECD economes han for LDCs over he perod. Also, one mgh noe he sudy by Coakely e al. (1999) who employ daa for fory-four LDCs over a sudy perod ha runs upo Whle panel daa conegraon echnques are no employed, hey fnd ha he average me-seres esmae for β 1 based on he Johansen maxmum lkelhood procedure s 0.44 whle he cross-seconal esmae of β s In he case of he cross secon of OECD economes, he cross-seconal esmae of β s also suggesng ha capal mobly s greaer for LDCs. In hs sudy, he null hypohess of β=1 s rejeced for he group mean esmae hereby rejecng long-run curren accoun susanably hrough mplcaon. 8 A non-saonary curren accoun s nconssen wh he susanably of exernal debs and mgh ndcae here s an ncenve for he counry o defaul on s exernal debs. Furhermore, he non-saonary of he curren accoun s nconssen wh he neremporal model of he curren accoun, and hence refues s valdy. 9 The modern neremporal model of curren accoun deermnaon uses consumpon-smoohng behavor o predc ha he curren accoun acs as a buffer o smooh consumpon n he face of shocks. Ths mples ha he curren accoun should n fac be a saonary raher han a non-saonary seres. A furher neresng ssue s wheher here exss evdence ha capal mobly dffers sgnfcanly across connens. From he full panel of sxeen LDCs, Paksan, Phlppnes, Sngapore, Sr Lanka and Thaland are used o form an Asan panel, whle Chle, Colomba, Cosa Rca, El Salvador, Honduras and Mexco are used o form a Lan Amercan panel. Tables 4 and 5 presen panel conegraon ess and FMOLS group esmaes of β for hese sub-groups. Boh he panel ADF and group ADF sascs repored n Table 4 ndcae ha he null of nonconegraon s srongly rejeced for boh regonal groupngs. Table 5 repors 7 In hs sudy, Ho employs he earler panel daa conegraon es advocaed by Kao (1999) whch s a resdual-based es ha mposes a common slope coeffcen. By conras, he Pedron group ADF es allows he esmaed slope parameers o vary across ndvdual members of he panel (Pedron (2004)). 8 Compared o exsng work on OECD counres, he examnaon of he susanably of LDC curren accoun defcs s a relavely unexplored area. Oher recen sudes of LDCs nclude ner ala Pachs and Kanaan (2001), who fnd ha he rade balance s saonary and herefore susanably n he case of Lebanon, and Coakley and Kulas (1997) who confrm lkewse for Inda bu no Korea and Tawan. 9 See, for example, Hused (1992) and references heren.

12 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 601 Table 4. Panel Daa Conegraon Tess for Sub-groups Tes Asa Lan Amerca panel ADF group ADF Noes for Table 4. These are he Pedron ess for conegraon [dscussed n Pedron (1999)] beween nvesmen and savng raes. The panel ADF sasc s a whn-dmenson sascs, he group ADF sasc s a beween-dmenson sasc. Boh ess are asympocally normal. and denoe rejecon of he non-conegraon null a he 1 per cen and 5 per cen sgnfcance levels wh respecve crcal values of and Table 5. FMOLS Esmaon of he Conegraed Panel for Sub-groups βˆ GFM βˆ FM, H 0 :β GFM Asa Lan Amerca Noes for Table 5. Ths able repors FMOLS panel daa esmaes of β 1 n equaon (2) usng he Pedron panel daa conegraon mehodology. These esmaes nclude common me dummes. The boom row refers o he group-mean esmaes. The esmae of β s accompaned by hree -sascs. Column 3 repors -sascs for he null β=0, column 4 repors -sascs for he null β=1 and column 5 repors - sascs for he null β= The -sascs are asympocally normal wh 1, 5 and 10 per cen crcal values of +/-2.57, +/-1.96 and +/-1.64 respecvely. = 0 βˆ FM, H 0 :β FM, = 1 βˆ FM, H 0 :β FM, = FMOLS group esmaes of β=0.340 and β=0.330 for he Asan and Lan Amercan groups respecvely. In boh cases, he respecve nulls of zero and uny are rejeced confrmng mperfec capal mobly for hese LDCs bu he esmaed savng reenon raos are very slgher lower han for he full panel. Ths suggess ha capal mobly may be slghly greaer for he Asan and Lan Amercan LDCs, however Table 5 also ndcaes ha he null β=0.341, whch s based on he savng reenon raon across he enre sample of LDCs, s easly acceped a he 5 per cen sgnfcance level for boh sub-groups. IV. Summary and Concluson Ths s he frs sudy of long-run capal mobly wh respec o LDCs usng panel daa conegraon mehods. The assessmen s based on esmang he longrun relaonshp beween nvesmen and savngs as orgnally proposed by Feldsen and Horoka and ohers n he measuremen of OECD capal mobly. Usng annual daa for a panel of sxeen LDCs over he sudy perod , a long-run conegrang relaonshp beween domesc savngs and nvesmen s

13 602 Mark J. Holmes srongly confrmed and fully modfed ordnary leas squares esmaon ndcaes ha capal s mperfecly moble. The low savng reenon coeffcen of abou one-hrd suggess ha capal mobly may acually be farly hgh for LDCs. Furher panel esmaon for sub-groups comprsng Asan and Lan Amercan counres offer lle evdence ha he exen of capal mobly s sgnfcanly dfferen beween hese panels. Receved 29 March 2004, Acceped 17 December 2004 References Baxer, M. and Crucn, M., (1993), Explanng savng-nvesmen correlaons, Amercan Economc Revew, 83, Coakely, J., Kulas, F., and Smh, R. (1996), Curren accoun solvency and Feldsen- Horoka puzzle, Economc Journal, 106, Coakley, J., and Kulas, F., (1997), The Conegraon of Long Span Savng and Invesmen, Economcs Leers, 54, 1-6. Coakley, J., and Hasan, F., and Smh, R., (1999), Savng, Invesmen and Capal Mobly n LDCs", Revew of Inernaonal Economcs, 7, 4, Corbn, A. (2001), Counry specfc effec n he Feldsen Horoka paradox: a panel daa analyss, Economcs Leers, 72, Dooley, M., Frankel, J., and Maheson, D. J. (1987), Inernaonal capal mobly: wha do savng-nvesmen correlaons ell us?, IMF Saff Papers, 34, 3, Dooley, M., Fernandez-Aras, E., and K. Klezer (1996), Is deb crss hsory? Recen prvae capal nflows no developng counres, World Bank Economc Revew, 10, 1, Feldsen, M. and C. Horoka (1980), Domesc Savngs and Inernaonal Capal Flows, Economc Journal, 90, 2, Frankel, J. (1992). Measurng nernaonal capal mobly: a revew. Amercan Economc Revew 82, Ho, T. (2002), A Panel Conegraon Approach o he Invesmen-Savng Correlaon, Emprcal Economcs, 27, Hused, S., (1992), The Emergng US Curren Accoun Defc n he 1980s: A Conegraon Analyss, Revew of Economcs and Sascs, 74, Im, K., Pesaran, M. H., and Y. Shn.(1997), Tesng for Un Roos n Heerogeneous Panels. Deparmen of Appled Economcs Workng Paper, Unversy of Cambrdge, December. Im, K., Pesaran, M. H., and Y. Shn.(2003), Tesng for Un Roos n Heerogeneous Panels. Journal of Economercs, 115, Kao, C., (1999), Spurous regresson and resdual-based ess for conegraon n panel daa,

14 Wha do Savngs-Invesmen Correlaons ell us abou he Inernaonal Capal~ 603 Journal of Economercs, 90, Levn, A., and Ln, C., (1993), Un Roo Tess n Panel Daa: Asympoc and Fne Sample Properes. Unpublshed manuscrp, Unversy of Calforna a San Dego. Murphy, R. G. (1984), Capal mobly and he relaonshp beween savng and nvesmen n OECD counres, Journal of Inernaonal Money and Fnance, 3, Pachs, C., and Kanaan, M., (2001), Is Lebanon s rade defc susanable? A conegraon analyss, Economa Inernazonale, 54, 1, Pedron, P. (1999), Crcal values for conegraon ess n heerogeneous panels wh mulple regressors, Economcs Workng Paper, Indana Unversy, Oxford Bullen of Economcs and Sascs, Specal Issue, Pedron, P. (2001), Purchasng power pary n conegraed panels, Revew of Economcs and Sascs, 83, Pedron, P. (2004), Panel Conegraon: Asympoc and Fne Sample Properes of Pooled Tme Seres Tess wh an Applcaon o he PPP Hypohess, Economerc Revews, forhcomng. Phllps, P.C.B., and Hansen, B.E. (1990), Sascal nference n nsrumenal varables wh I(1) processes, Revew of Economc Sudes, 57, Vos, R., (1988), Savng, nvesmen and foregn capal flows: Have capal markes become more negraed?, Journal of Developmen Sudes, 24,