The Composition of Public Expenditure and Economic Growth in Developing Countries

Transcription

1 The Composton of Publc Expendture and Economc Growth n Developng Countres by M Emranul Haque Leverhulme Centre for Research on Globalsaton and Economc Polcy, School of Economcs, Unversty of Nottngham, Nottngham NG7 2RD Tel: ; Fax: ; Emal: m.haque@nottngham.ac.uk and Centre for Growth and Busness Cycle Research, School of Economc Studes, Unversty of Manchester, Manchester M13 9PL, Revsed: December 2003 Keywords: Economc Growth, Publc Expendture, Publc Sector, Panel and Cross-country Regressons. JEL Classfcaton: O4; E62 ; H6 The author would lke to thank Jonathan Temple for hs nsghtful comments n relaton to ths research. The permsson to use World Bank Archves at Washngton D.C. for data collecton s gratefully acknowledged.

2 Abstract We have evaluated the effect of changng the composton of publc expendture on economc growth n developng countres for cross-secton and panel data. As we have found the evdences contradctory to each other whle cross-secton analyss supportng the conventonal wsdom, we test for non-statonarty n the panel data. After correctng for nonstatonarty found n the data for shares of current and captal expendture n total publc expendture of developng countres, the results confrm the conventonal wsdom and suggest that swtchng resources from consumpton to nvestment type s growth enhancng n developng countres where governments face resource constrants. 2

3 The Composton of Publc Expendture and Economc Growth n Developng Countres 1. Introducton: In the recent years, a szable amount of research has been drected towards evaluatng the role of fscal polcy n economc development wth partcular emphass on publc expendtures (Barro, 1990; Barro and Sala--Martn, 1992; Barro and Sala--Martn, 1995; Glomm and Ravkumar, 1997). Emprcal attempts to lnk aggregate measures of fscal polcy wth average per capta growth rates n cross-country studes have tended to use one or more of the followng fscal varables: () measures of the overall sze of the government (Landau, 1983, 1985, 1986; Kormend and Megure, 1985; Levne and Renelt, 1992); () dsaggregated measures of publc expendtures, the dsaggregaton ether between consumpton and nvestment expendtures or between sectoral expendtures (Barro, 1990, 1991; Easterly and Rebelo, 1993); () growth of publc expendtures (Ram, 1986); (v) composton of government expendtures (Deverajan et. al., 1996; Gupta et.al., 2002); (v) government surplus / defct (Fscher, 1993); (v) varous government taxes (Easterly and Rebelo, 1993) and (v) dfferent expendtures n the presence of government budget constrant (Kneller et. al., 1999). Whle the focus of these studes was mostly to dentfy the level effects of publc expendtures on growth, lttle effort was spent on examnng the pure composton effects of publc expendtures (.e., to examne the extent to whch a resource swtch between consumpton and nvestment expendture could enhance growth performance of the economy). Such analyss s of vtal mportance to many economes where resources are lmted for rasng current levels of publc spendng. Two recent exceptons n ths drecton are by Deverajan et. al. (1996) and Gupta et.al., (2002). The study by Deverajan et. al. (1996) suggests that, for developng countres, swtchng publc spendng from nvestment to consumpton type s growth enhancng. Usng a panel data for 43 developng countres over 20 years, they fnd that an ncrease n the share of current (captal) expendture has postve (negatve) and statstcally sgnfcant growth effects. The explanaton gven by Deverajan et. al. (1996) for ther counterntutve results s that f productve expendture s already excessve, a further ncrease n ts share n total expendture would have a negatve mpact on growth. A counterntutve polcy mplcaton follows from here suggestng that developng 3

4 country governments have been msallocatng publc expendtures n favor of captal expendtures at the expense of current expendtures and they should allocate ther resources to pay ther bureaucrats rather than buldng nfrastructure. The most recent study by Gupta et.al. (2002) uses the share of dfferent types of spendng n total publc spendng as one of the explanatory varables n growth regressons for thrty-nne low-ncome countres and found the result opposte to Deverajan et. al. (1996). One weakness of ther study, however, s that they have not controlled for total spendng wthn the regresson specfcaton, whch would be dscussed later. Intutvely, nvestment (or, productve) expendture by the government s supposed to rase prvate captal accumulaton, whch n turn wll rase economc growth n the long run. And the mpact of government consumpton should be opposte to ts nvestment counterpart. An nterpretaton gven by Barro (1990) for government consumpton s that government consumpton ntroduces dstortons, such as hgh tax rates, but does not provde an offsettng stmulus to nvestment and growth. Alternatvely, the effect of an ncrease n government consumpton should be nl f we vew t as leavng the productvty of the prvate sector unaffected. In contrast, the effect of publc nvestment should be postve snce ths type of actvty s lkely to enhance the productvty of the prvate sector (see Aschauer, 1989; Barro, 1990). If we consder swtchng expendture from one component to another, especally n the context of developng countres where there s a tremendous scarcty of resources and nfrastructure, standard theory predcts that swtchng from current to captal expendtures would gve hgher growth n the economy. Ths s because developng countres typcally lack nfrastructures lke transport and communcatons that help promote prvate captal accumulaton. The paper evaluates ths ssue gvng due consderaton to propertes of the tme seres data (especally, non-statonarty) for a panel of thrty-three developng countres 1. Throughout our analyss, we use the specfcaton of Deverajan et. al. (1996) to capture the growth effects of resource swtch from current to captal expendtures or vce versa when the economy has the resource constrant. We fnd the results n panel and cross-secton contradctory to each other, where the panel result (same as Deverajan et. al., ) beles the theoretcal ntuton. 1 Our sample comes from the data set collected under a project on sectoral current and captal expendtures by the government after consultng World Bank Country Economc Reports and Publc Expendture Revews. The project ended up collectng the data for only thrty-three countres. 2 Ths confrms that our data set s qualtatvely not dfferent than that have been used by Deverajan et. al. (1996) even though our sample dffers from thers. 4

5 We nvestgate whether the reason for above les n the tme-seres property of the panel. After correctng for the non-statonarty found n the data, the fnal result suggests that swtchng resources from consumpton to nvestment type s growth enhancng rather than growth reducng. We also conduct varous senstvty analyses to ensure the robustness of the result obtaned. The structure of the paper s as follows. A smple descrptve analyss s provded n secton 2 to show the drecton of government s current and captal expendtures shares n GDP and n total expendtures and growth n GDP per capta. In secton 3, we report the results of base regressons for a panel of developng countres. Secton 4 contans the cross-secton evdence that contradcts the panel (as n Deverajan et. al., 1996) fndngs. Secton 5 tests for statonarty of data on expendtures shares and fnd the results that support the cross-secton evdence. Secton 6 concludes. 2. A Smple Analyss: The reason put forward by Deverajan et. al. (1996) for the negatve co-effcent for captal expendture s share was that perhaps developng countres were rasng captal expendture whle t was already excessve. But n realty developng countres often cannot grow as they tremendously lack nfrastructure lke roads, hghways, gas, electrcty, etc. Due to the largescale nature, there occurs market falure n buldng these nfrastructures. Government has to come forward n order to develop these effectvely through government captal expendture. Our hypothess s that gven the resources to be spent n the publc sector fxed over the years, resource swtch from consumpton (e.g., payng the bureaucrats) to nvestment (e.g., buldng nfrastructure) type expendture would lead to hgher growth. Before conductng a formal test, we begn our prelmnary nvestgaton by smply lookng at the data. We fnd that captal expendture as a percentage of GDP and ts share n total government expendture are low n comparson to ts consumpton (current expendture) counterpart. Table 1 shows the average GDP growth, the average shares of government current and captal expendtures n GDP, and average shares of current and captal expendtures n total government expendture n the world, low-ncome, mddle-ncome, 5

6 developng and hgh ncome countres for avalable years durng the perods and Table 1 s about here Whle n all categores of countres, low, mddle, developng and hgh, both growth and captal expendture s share n GDP has declned, except for low-ncome countres current expendture s share n GDP has ncreased. The world s average GDP growth rate has declned from 3.1 percent per year over to 2.3 percent over , whle ths fgure has margnally reduced from 4.3 to 4.2 n low-ncome countres, from 2.8 to 2.5 n mddle-ncome countres, from 3.0 to 2.8 n developng countres, and most markedly from 3.2 to 2.1 n hgh-ncome countres. Current expendture s share n GDP has rsen durng the same perod from 21.9 percent to 25.4 percent n the world, from 20.1 percent to 23.0 percent n mddle ncome countres, from 18.8 percent to 21.7 percent n developng countres and from 30.7 percent to 34.3 percent n hgh ncome countres, except for low ncome countres where t has not moved n ether of drecton,.e., a neglgble declne from 17.2 percent to 16.9 percent. However, for captal expendtures all categores of countres experenced the same drecton n movement (.e., declned as for GDP growth) for captal expendture s share n GDP; from 5.1 percent to 3.4 percent n the world, from 5.8 percent to 4.6 percent n low ncome countres, from 6.2 percent to 3.7 percent n mddle ncome countres, from 6.0 to 3.9 percent n developng countres, and from 2.6 percent to 2.2 percent n hgh ncome countres. If we consder the composton of total government expendture, we can see from the same table that current expendture s share n total government expendture ncreased over these two perods whle captal s share decreased for all categores of countres. Especally, the ncreasng rate for current expendture s share s much hgher n all categores of developng countres (low and mddle ncome, and the developng as a whole). It may suggest that () both the level and the share of captal expendture n total expendture are postvely assocated wth growth, whle those of current expendture have a negatve assocaton wth growth; () all countres are swtchng on average ther expendtures from the captal to the current component of ther total expendtures over tme, whle ths swtch s bgger n the case of low ncome and developng countres than n hgh ncome countres. Ths analyss 3 Source: World Development Report, The classfcaton of Low-ncome, Mddle-ncome, Developng and Hgh-ncome countres s taken from World Bank. Developng countres comprse low and mddle ncome countres. 6

7 suggests that partcularly developng countres are not swtchng ther expendtures from current (consumpton) to captal (nvestment) components over tme as has been suggested by Deverajan et. al. (1996) as an explanaton of ther result, rather the flow s the reverse whle havng the experence of lower growth. Hence, swtchng the expendture from current to captal should affect growth postvely and the affect should be hgher for developng countres, whle swtchng n the opposte drecton should have a negatve effect. 3. Emprcal Analyss: Our emprcal analyss focuses on the lnk between the share of the components of publc expendture n total expendture and economc growth n developng countres. Aschauer and Greenwood (1985), Barro (1990, 1991) and others emphasze the dstncton between publc goods and servces that enter nto the household s utlty functon and those that complement prvate sector nputs n producton. Government consumpton s lkely to have a negatve effect on growth, as the reducton n the returns on and ncentves for nvestment due to hgher tax-fnance outwegh the utlty derved from t. Grer and Tullock (1989) confrm ths negatve effect of government consumpton by usng pooled cross-secton/tme-seres data for 115 countres. By contrast, government nvestment expendture, lke nfrastructure servces, s thought to help promote nvestment and growth. Aschauer (1989) fnds the core nfrastructure streets, hghways, arports, mass transt, and other publc captal - has the most explanatory power for prvate sector productvty n the Unted States over the perod Easterly and Rebelo (1993) show that publc nvestment n transport and communcaton rases growth by ncreasng the socal return to prvate nvestment, not by rasng prvate nvestment tself Equaton Specfcaton: The specfcaton used by Gupta et. al. (2002) automatcally takes care of non-statonarty n data as they have taken the frst dfference of the shares of dfferent expendture components n total publc spendng. But they have not controlled for total publc spendng wthn the regresson specfcaton. Hence ther analyss does not exactly show the growth effect of resource swtch from one expendture component to another whle the total publc spendng 7

8 s fxed. Throughout our analyss, we use the followng specfcaton taken from Deverajan et. al. (1996) as t s the most sutable specfcaton for our purpose: 6 ( t+ 1, t+ 5) = j D j + α7 TE / GDP) t j= 1 GRPGDP α ( + α8bmp t Where the varables are, ( G TE ) + α SHOCK + α / + µ 9 t k k t t (1) () GRPGDP ( t+ 1, t+ 5) = Fve-year forward movng average of per-capta real GDP growth for country. () D j = Contnental dummy varables; j = 1, 2,, 6 correspondng to Central and North Amerca, East Asa and the Pacfc, Latn Amerca and the Carbbean, Mddle East and North Afrca, South Asa, Sub-Saharan Afrca 4. t () ( TE / GDP) : Share of total government expendture n GDP for country at tme t. (v) BMP t = Premum n the black-market for foregn exchange n country at tme t, calculated as t [( BER OER ) OER ] BMP = / where BMP = Black Market Premum, OER = t Offcal Exchange Rate and BER = Black Market Exchange Rate 5. (v) SHOCK t t t : The shock varable for country at tme t, s a weghted average of changes n the world real nterest rate (R) and the export prce ndex (PX) and mport prce ndex (PM) for each country. The export and mport prces are ndex numbers expressed n U.S. dollars converted at the annual average of the country s offcal exchange rate. The weghts are the ratos to GDP of debt, export (X), and mport (M), respectvely. By changes we mean the dfference n the average value of these varables between t to t+4 and t+1 to t+5. In symbols t s defned as, Shock t = ( R R )( Debt ) + ( PX PX )( X ) + ( PM PM )( M ) t+ 1, t+ 5 t, t+ 4 GDP t+ 1, t+ 5 t, t+ 4 GDP t+ 1, t+ 5 t, t+ 4 GDP ( r dp ) and P TotalInterestPayments R = ; r = ; R = World real nterest rate. ( 1 + dp ) TotalDebt P dp/p = World nflaton rate measured by the percentage change n the USA GDP deflator. 4 Classfcaton of Contnents s taken from Global Development Fnance, Source: Fscher (1993) for the perod and varous World Currency Yearbooks for the observatons onward. 8

9 (v) (G/TE) t : A vector of government expendture ratos to GDP for country at tme t. Expendture shares are by economc classfcaton: Ncur/Te = rato of current expendture to total expendture and Cap/Te = rato of captal expendture to total expendture. Note that the data for all the varables are annual. In constructng the shock varable we take the dfference n the average value of the real nterest rate, mport prce ndex and export prce ndex between perods t to t+4 and t+1 to t+5. In ths way, we capture a smlar type of dstorton to the shock varable of DSZ and get more tme seres data 6. The growth rate of GDP per capta, exports, mports, export prce ndex and mport prce ndex are taken from the World Bank CDROM. Debt and nterest rate are taken from World Debt Tables. The data on the expendture varables for the regressons are taken from Haque (2002) 7. In the base case, we use the contnent dummes to capture the uneven growth experences across contnents. The other two control varables, the black market premum and the composte shock varable, are used to control for the effects of external shocks and domestc polces Regresson Results: As the dependent varable s a fve-year forward movng average of per capta real GDP growth (%), we correct the standard errors by usng the extended method of Deverajan et. al., (1996) for seral correlaton based on Hansen and Hodrck (1980). Table 2 contans the estmates of equaton shows a postve (0.035) and sgnfcant (p-value = 0.05) relatonshp between the fve-year forward movng average of per capta real GDP growth and the rato of current expendture to total expendture 8. It suggests that a unt ncrease n ths rato would ncrease the growth rate by percentage ponts. Clearly ths s an unusual fndng, as Deverajan et. al. (1996) acknowledge. As already mentoned, t s unusual, because standard publc fnance theory predcts that government consumpton should have ether a negatve or nl mpact on producton as t can not provde suffcent stmulus for prvate captal 6 Ths s a slght change from Deverajan et. al. (1996) n order to rase the number of observatons n the sample. 7 The data are avalable upon request. 8 Devarajan, et. al. (1996) estmate the coeffcent for current expendture s share wth sgnfcance at 5 percent and for captal expendture s share wth sgnfcance at 10 percent. 9

10 accumulaton, rather t creates dstorton through taxaton for ts fnancng. Barro (1990, 1991), Levne and Renelt (1992) and others have confrmed ths ntuton. The relatonshp between the captal component s share n government expendture and per capta growth s negatve (-0.036) and sgnfcant (p-value = 0.05) as shown n Once agan ths beles the standard hypothess. Intuton suggests that the resultng stock of nfrastructure from captal expendture would complement prvate sector productvty, and hence, has favorable growth effects. Table 2 s about here As the dummy varable coeffcents for Central and North Amerca, Mddle East and North Afrca and South Asa are almost the same; we use two alternatve specfcatons (models ). For models , we use only 3 contnent dummes for East Asa and the Pacfc, Latn Amerca and the Carbbean and sub-saharan Afrca. The coeffcents of regonal dummes n models suggest that the growth rate n sub-saharan Afrca and Latn Amerca are below the mean whle that for East Asa and the Pacfc s above the mean 10. For models , we exclude the dummy for East Asa and the Pacfc 11. The nterpretaton of the estmated coeffcents on these varables s somewhat problematc, however, because the choce of whch regons to assgn dummes s endogenous. In any case, f we have already ncluded enough explanatory varables to explan why growth was below expectatons n sub- Saharan Afrca and Latn Amerca and the Carbbean and above expectatons n East Asa and the Pacfc, then the estmated coeffcents of the dummy varables would dffer nsgnfcantly from zero. However, even after dong the exercse of ncludng and excludng the regonal dummes we fnd almost the same coeffcent for the shares of current and captal expendtures n total expendtures. The rato of current to total expendture has the coeffcent for the frst two specfcatons whch s margnally sgnfcant (p-value = 0.05). But when we regress wth dummes for only Latn Amerca and the Carbbean and sub-saharan Afrca, we get a slghtly hgher postve coeffcent (0.04) for current expendture s share wth strong statstcal sgnfcance (p-value = 0.01). The result s reversed for the rato of captal to total expendture. Whle the coeffcent for the frst two specfcatons s (p-value = 0.05), 9 It may be noted that the coeffcent s not exactly the negatve of that for current expendture because the latter s net of nterest payments (so that the two shares do not sum to exactly one). 10 It confrms the fndngs of Barro and Sala--Martn (1995). 10

11 the coeffcent for ths wth contnent dummes for only Latn Amerca and the Carbbean and sub-saharan Afrca s (p-value = 0.01). Ths analyss suggests a counterntutve polcy that a resource swtch towards current (captal) expendtures s growth-enhancng (growthreducng). 4. Cross-Secton Evdence: As the result we obtaned from the panel regresson s somewhat puzzlng, we examne the relatonshp further usng the same data and the same specfcaton n a cross-sectonal context. Specfcally, we have taken the average over for the data on all varables. Snce we are left wth one data pont for each country n a cross-secton study compared wth several data ponts n a panel, contnental dummes would be weak explanatory varables. Instead, followng standard cross-secton models, poneered by Barro (1991), we use the country-specfc control (/envronmental) varables, whch are ntal GDP per capta, ntal human captal (proxed by prmary and secondary school enrolments) and poltcal nstablty (proxed by number of assassnatons and number of revolutons and coups) 12. In summary, we fnd that for the cross-secton the results are reversed from the panel case and confrm the standard fndngs of earler studes (Barro, 1991, 1998; Barro and Sala--artn, 1995, 1999;Easterly and Rebelo, 1993). Table 3 s about here Table 3 shows that after controllng for the ntal envronment of the respectve countres n terms of ntal GDP per capta, ntal human captal, and poltcal nstablty, we get negatve (-0.06) and sgnfcant (at 5% level) coeffcent for current expendture s share but postve (0.07) and sgnfcant (at 5% level) coeffcent for captal expendture s share 13. We have also tested for robustness by controllng for the black market premum (BMP) as an external shock varable and the ntal trade share and M2 (as rato to GDP) as polcy shocks, and we get smlar results suggestng negatve and sgnfcant coeffcents (rangng from 0.08 to 11 In these last two specfcatons, we selected regons for whch prevous researchers have observed that growth rates are surprsngly low. 12 The data for GDP per capta, ntal trade share (export plus mport as percentage of GDP) and M2 are collected from World Bank CDROM and other ntal condtonng varables are from Barro and Lee (1994). 13 Agan we need to menton that the coeffcent s not exactly the negatve of current expendture because the latter s net of nterest payments (so that the two shares do not sum to exactly one). 11

12 0.04) for current expendture s share and the equvalent opposte result for captal expendture s share (rangng from to 0.08). Our fndngs suggest that the reason for the puzzlng result above may le wth the tme seres for each cross-secton,.e., whether the data for the correspondng expendture s shares s statonary. Because, f a tme-seres for a varable s non-statonary (.e., havng unt roots), then the regresson wth ts level would gve spurous results. 5. Regresson Results wth Statonary data In order to address the ssue of non-statonarty we examne whether the data for the expendture s share of dfferent components n government expendture are statonary. We know from Nelson and Plosser (1982) that generally real macroeconomc varables are nonstatonary and ntegrated of order one,.e., I(1). Ths means that these varables should be statonary f we take ther frst dfference. Table 4 s about here Table 4 provdes us wth tests for the order of the ntegraton of dfferent components share n total publc expendture. Out of 33 countres we can test for statonarty only for 24 countres due to nsuffcent tme seres data for the rest of the countres. Of these, 22 countres are found to have non-statonary data for current and captal expendtures shares, among whch data are judged to be I(1) for 21 countres. Inda seems to have I(2) data for these varables, whch s rather strange. It may be a reflecton of the fact that the number of observatons s only 14 for whch explanatory power of the test may be reduced. Wthn our sample we have only 2 countres (Serra Leone and Thaland) havng statonary or trend statonary data accordng to the augmented Dcky-Fuller (ADF) test. From ths exercse, we can conclude n general that the shares of dfferent components n total expendture are I(1) varables. As dscussed above, the presence of I(1) varables can gve spurous results. In order to correct for non-statonarty we can take the frst dfference of these varables. Ths approach has got another added advantage n that t explctly looks at the effect of swtchng resources (government expendture) from one type to another, as suggested by Deverajan et.al. (1996). 12

13 Usng the frst dfference of components shares n total government expendture, rather than takng the current year levels, may be more approprate n capturng the swtchng effect. We use the same specfcaton as n equaton 1 wth the frst dfference of the explanatory varables 14. Wth the error term ut = µ + ε ( = 1,., N countres; t = 1,, T tme), there t are three alternatve econometrc specfcatons of the model, each dfferng n ts treatment of µ I (see Hsao, 1992): pooled OLS, one-way (country) fxed effects, two-way (country and tme) fxed effects and random effects regressons, keepng the same specfcaton as before. The pooled OLS (frst two columns n Table 6) estmator assumes the error term as ndependent of the cross-sectonal unts (countres) and d normal (that s, the panel s estmated as f t were a cross-country regresson). The one way fxed effects estmator (see thrd and fourth columns n Table 6) treats µ as a fxed (tme nvarant) but unknown constant dfferng across ndvdual countres. ut = µ + ε (7) t where µ s a country-specfc constant and ε t s whte nose. In the fxed effects approach t s assumed that dfferences across countres can be captured n dfferences n the constant term, and so each µ s an unknown parameter to be estmated. In effect, dfferences between countres are vewed as parametrc shfts of the regresson functon. The two way fxed effects (see ffth and sxth columns n Table 6) show the estmated β under the assumpton of fxed country and tme effects, that s, ut = µ + λ + ε (8) t t where µ and ε t are defned as above, and λ t s an exclusvely tme-dependent constant effect. Fnally, the random effects or varance components specfcaton assumes that countryspecfc constant terms are randomly dstrbuted across cross-sectonal unts (that s, countres). In ths approach, each µ ~ N (0, σ 2 µ ), gven that µ = µ + η, where µ s the group mean ntercept and η s the error assocated wth the country (whch are random devatons from ths common mean, and constant over tme). The error term becomes, ut = ε + η (9) t t 14 We have also checked the whole analyss usng the level of other varables, total expendture s share n GDP, black market premum and the shock varable. The results are essentally the same. 13

14 As the coeffcents for one-way fxed effects and random effects models are almost the same, we chose to report both rather than selectng between them usng a Hausman specfcaton test. Table 6 shows that resource swtch from captal expendture (nvestment) to ts current (consumpton) counterpart lowers economc growth n the long run whle the resource swtch n the opposte drecton rases economc growth. In all the regressons for the growth rate n per capta GDP, we have found negatve and sgnfcant coeffcents (rangng from to ) for current expendture s share and postve and sgnfcant coeffcents for captal expendture s share (rangng from 0.12 to 0.19), after controllng for the change n total government expendtures (see the frst part of Table 6). Table 5 s about here For senstvty analyss, we have also used a 5-year forward movng average of per capta real GDP growth rate as the dependent varable wth Hansen and Hodrck (1980) standard error correcton. Although the coeffcents become smaller (rangng from to for current expendture s share and to for captal expendture s share), the sgns (negatve for current expendture s share and postve for captal expendture s share) and the strong sgnfcance reman the same (see the second part of Table 5). 6. Concluson We have re-evaluated and tested the mpact of resource swtch from one type of publc spendng to another whle the total spendng s fxed. We fnd n our ntal analyss that the results n panel and cross-secton contradct wth each other, whle the panel results are counterntutve (as n Deverajan et. al., 1996). When we look at the tme seres property of the data for dfferent countres n the panel, we fnd that the shares of current and captal expendture n total publc expendture are non-statonary [.e., I (1) varables]. After correctng for non-statonarty, takng frst dfferences of the varables, the result we obtan supports the conventonal wsdom and the cross-secton fndngs and suggest that swtchng resources from consumpton to nvestment type s growth enhancng and vce-versa. Our result also confrms the vew of Ram (1986) that t s the growth not the level of government expendture that nfluences growth. The paper also ncludes varous senstvty analyses to ensure the robustness of the obtaned result, applyng dfferent econometrc methods. 14

15 The paper suggests that n the presence of non-statonarty n data, one needs to take frst dfference n order to avod spurous results. The two studes we have re-evaluated are Deverajan et. al. (1996) and Gupta et. al. (2002). Whle the former uses a correct specfcaton to capture the growth effect of resource swtch n a resource constraned economy wthout consderng the ssue of non-statonarty, the later consders nonstatonarty wthout controllng for the resources avalable to spend. Combnng both, we fnd the support for conventonal wsdom. 15

16 Table 1: Publc Expendtures and Economc Growth World Low Income Mddle Income Developng Hgh Income GDP Growth Y (%) GDP Growth Y (%) Current Expendture/GDP 1980 (%) Current Expendture/GDP 1996 (%) Captal Expendture/GDP 1980 (%) Captal Expendture/GDP 1996 (%) (Current, Captal) (-, +) (+, +) (-, +) (-, +) (-, +) Total Expendture/GDP 1980 (%) Total Expendture/GDP 1996 (%) Current/Total 1980 (%) Current/Total 1996 (%) Captal/Total 1980 (%) Captal/Total 1996 (%) (Current/Total, Captal/Total) (-, +) (-, +) (-, +) (-, +) (-, +) Source: calculated from World Development Report Note: + means movement n the same drecton wth growth. - means movement n the drecton opposte to growth. 16

17 Table 2: Composton of government expendture and economc growth, wth t-statstcs n parentheses (Dependent varable = fve-year forward movng average of per capta real GDP growth rate) Constant (0.73) 0.04 (1.00) 0.01 (0.64) 0.05 (4.73) Central and North Amerca (0.42) (8.93) East Asa and the Pacfc (1.07) (5.08) (1.91) (1.90) Latn Amerca and the Carbbean (-2.07) (0.46) (-4.89) (-4.90) (-5.52) (-5.55) Mddle East and North Afrca (0.50) (3.56) South Asa (0.46) (3.71) Sub-Saharan Afrca (-1.57) 0.01 (1.64) (-4.93) (4.91) (-5.99) (-6.01) Te/GDP (0.27) (0.27) (-0.19) (-0.19) (-0.02) (-0.03) Curt (1.97) (1.97) (2.32) Capt (-1.95) (-1.95) (-2.38) Black Market Premum (-4.70) (-4.70) (-4.68) (-4.66) (-4.74) (-4.72) Shock (-2.24) (-2.24) (-2.10) (-2.11) (-2.27) (-2.28) Adj. R-sq No. of Observatons DW Notes: Estmaton s by OLS wth Hansen and Hodrck (1980) error correcton for forward movng average of the dependent varable, used n Deverajan et. al. (1996). Data are for 33 countres for

18 Table 3: Cross-secton regresson for growth of real GDP per capta on Composton of government expendture, wth t-statstcs n parentheses (1.7) Constant 0.03 (1.92) Intal GDP per capta Intal Prmary School Enrollment Intal Secondary School Enrollment (-1.25) 0.02 (0.70) 0.07 (1.12) Number of Assassnatons 0.02 Number of revolutons and coups (0.74) (-1.13) Te/GDP 0.07 (1.31) Ncur/Te (-2.01) (1.8) (-0.21) (-1.21) 0.02 (0.74) 0.07 (1.07) 0.02 (0.67) (-1.06) 0.06 (1.24) Cap/Te (1.9) 0.04 (1.90) (-0.71) 0.01 (0.57) 0.03 (0.55) 0.01 (0.34) (-0.34) 0.07 (1.65) (2.42) (-3.10) M2/GDP (1.10) (-2.17) (-0.66) 0.01 (0.51) 0.03 (0.56) 0.01 (0.24) (-0.30) 0.07 (1.61) (0.80) (1.11) 0.01 (0.79) (-0.01) 0.04 (1.70) (-0.63) 0.01 (0.24) (1.27) 0.14 (2.25) (3.44) 0.03 (0.66) (-2.51) (1.12) (-2.04) (0.01) (1.69) (-0.62) (0.20) (-1.24) (2.22) (0.94) Intal trade share (-2.62) (1.13) 0.02 (1.30) (-0.63) 0.03 (1.23) (-0.04) (0.10) (-0.12) 0.12 (2.19) (2.55) 0.04 (0.86) (-2.53) Black Market Premum (-2.18) (1.14) (-0.79) (-0.62) (1.18) (-0.62) (-0.12) (-1.16) 0.12 (2.21) (0.34) (-2.62) (-2.20) Adj. R-sq Obs (2.19) 0.03 (0.87) (-1.83) (-2.13) 18

19 Table 4: Test for Statonarty of Current and Captal Expendtures shares n Government Expendture Country Descrpton t ADF t 0.05 t 0.01 Observatons 1. Bahamas CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Bangladesh CURT (Constant, Lag = 2) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 2) DCAPT (Constant, Lag = 2) * Congo CURT (Constant, Trend, Lag = 1) DCURT (Constant, Trend, Lag = 1) * CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Trend, Lag = 1) * Ethopa CURT (Constant, Trend, Lag = 1) DCURT (Constant, Trend, Lag = 1) * CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Trend, Lag = 1) * Ghana CURT (Constant, Trend, Lag = 1) DCURT (Constant, Trend, Lag = 1) ** CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Trend, Lag = 1) ** Malysa CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) * CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) * Guatemala CURT (Constant, Trend, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Lag = 1) ** Inda CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) DDCURT (Constant, Lag = 1) * Indonesa CURT (Constant, Lag = 2) DCURT (Constant, Lag = 1) * CAPT (Constant, Lag = 2) DCAPT (Constant, Lag = 1) * Kenya CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Malaw CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Marocco CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1)

20 DCAPT (Constant, Lag = 1) ** Maurtus CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) * CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) * Nepal CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) * CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) * Ngera CURT (Constant, Trend, Lag = 1) DCURT (Constant, Trend, Lag = 1) ** CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Trend, Lag = 1) ** Pakstan CURT (Constant, Trend, Lag = 1) DCURT (Constant, Trend, Lag = 1) * CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Trend, Lag = 1) * Rwanda CURT (Constant, Trend, Lag = 1) DCURT (Constant, Lag = 1) * CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Lag = 1) * Serra Leone CURT (Constant, Trend, Lag = 1) ** CAPT (Constant, Trend, Lag = 1) ** Sr Lanka CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Sudan CURT (Constant, Trend, Lag = 1) DCURT (Constant, Lag = 1) * CAPT (Constant, Trend, Lag = 1) DCAPT (Constant, Lag = 1) * Syra CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Thaland CURT (Constant, Trend, Lag = 1) * CAPT (Constant, Trend, Lag = 1) * Tanzana CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Zamba CURT (Constant, Lag = 1) DCURT (Constant, Lag = 1) ** CAPT (Constant, Lag = 1) DCAPT (Constant, Lag = 1) ** Notes: CURT = Current Expendture s Share n Total Government Expendture; DCURT = Frst Dfference of CURT; CAPT = Captal Expendture s Share n Total Government Expendture; DCAPT = Frst Dfference of CAPT; ** = Sgnfcant at 1% level; * = sgnfcant at 5% level. Constant, lags and trends are chosen based on Akake Informaton Crteron and scatter plots of the varables. 20

21 Table 5: Regressons of Growth n Real GDP per capta on Changes n Expendture Shares, wth t-statstcs n parentheses. (1.15) (1.16) (1.17) (1.18) (1.19) (1.20) (1.21) (1.22) Pooled-OLS Pooled-OLS One-way FE One-way FE Two-way Two-way RE RE FE FE Dependent Varable: per capta real GDP growth rate Te/GDP (-1.89) (-1.74) (-1.78) (-1.60) (-1.95) (-1.86) (-2.03) (-1.86) Cur/Te (-3.07) (-4.09) (-2.72) (-4.22) Cap/Te (2.77) (3.69) (2.51) (3.82) BMP (-3.41) (-3.41) (-0.23) (-0.26) (-0.53) (-0.55) (0.80) (0.75) SHOCK (-1.64) (-1.64) (-1.14) (-1.05) (-1.61) (-1.61) (-1.71) (1.62) Adj. R-sq Obs DW Dependent Varable: fve-year forward movng average of per capta real GDP growth rate Te/GDP (-0.97) (-1.04) (-0.05) (-0.06) (-1.36) (-1.41) (-0.10) (-0.13) Cur/Te (-1.96) (-2.33) (-2.07) (-2.37) Cap/Te (2.28) (2.47) (2.26) (2.52) BMP (-4.57) (-4.55) (-3.21) (-3.20) (-2.23) (-2.20) (-2.54) (-2.53) SHOCK (-1.45) (-1.45) (-0.44) (-0.39) (-0.54) (-0.55) (-0.15) (-0.10) Adj. R-sq Obs DW

22 References: Aschauer, D. (1989): Is Government Spendng Productve?, Journal of Monetary Economcs, 23, Aschauer, D. A. and Greenwood, J. (1985): Macroeconomc Effects of Fscal Polcy, Carnege Rochester Conference on Publc Polcy, 23, Barro, R.J. (1990): Government Spendng n a Smple of Endogenous Growth, Journal of Poltcal Economy, 98, 5 (October), Part II, S103-S125. Barro, R.J. (1991): Economc Growth n a Cross Secton of Countres, Quarterly Journal of Economcs, CVI (May 1991), Barro, R. J. (1998): Determnants of Economc Growth: A Cross-Country Emprcal Study, MIT Press. Barro, R.J. and Lee, J-W (1994): Sources of Economc Growth, Carnege-Rochester Conference Seres on Publc Polcy. Barro, R.J. and Sala--Martn, X. (1992): Publc Fnance n s of Growth, Revew of Economc Studes, 59, Barro, R.J. and Xaver Sala--Martn (1995), Economc Growth, McGraw-Hll, Boston, Mass. Barro, R.J. and Xaver Sala--Martn (1999), Economc Growth, Frst MIT Press Edton. Capolupo (1997): Devarajan, S., Swaroop, V., Zou, H. (1996): The Composton of Publc Expendture and Economc Growth, Journal of Monetary Economcs, 37 (1996), p Easterly, W. and Rebelo, S. (1993), Fscal Polcy and Economc Growth: An Emprcal Investgaton, Journal of Monetary Economcs, 32 (December), p Fscher, S. (1993): The Role of Macroeconomc Factors n Growth, Journal of Monetary Economcs, 32 (December), p Glomm, G., Ravkumar, B., (1997): Productve Government Expendtures and Long-run Growth, Journal of Economc Dynamcs and Control, 21 (1), Grer, K.B., and Tullock, G., (1989): An Emprcal Analyss of Cross-Natonal Economc Growth, , Journal of Monetary Economcs 24, Gupta, S., Clements, B., Baldacc, E., and Mulas-Granados, C., (2002): Expendture Compoton, Fscal Adjustment, and Growth n Low-ncome Countres, IMF Workng Paper, WP/02/77. Hansen, L. P. and Hodrck, R. J. (1980): Forward Exchange Rates as Optmal Predctors of Future Spot Rates: An Econometrc Analyss, Journal of Poltcal Economy, 88, Haque, M.E. (2002): Essays on Government Polces and Growth, PhD Dssertaton, School of Economc Studes, Unversty of Manchester. Hsao, C. (1992): Analyss of Panel Data, Econometrc Socety Monograph, Cambrdge, UK: Cambrdge Unversty Press. Kneller, R. Bleany, M.F. and Gemmell, N. (1999): Fscal Polcy and Growth: Evdence from OECD Countres, Journal of Publc Economcs, Vol. 74, pp Kormend, R.C., and Megure, P.G., (1985): Macroeconomc Determnants of Growth: Cross-Country Study, Journal of Monetary Economcs, 16 (September), pp

23 Landau, D. L., (1983): Government Expendture and Economc Growth: A Cross Country Study, Southern Economc Journal, 49 (January 1983), pp Landau, D. L., (1985): Government Expendture and Economc Growth n the Developed Countres, , Publc Choce, No. 3, 1985, 47, Landau, D. L. (1986): Government and Economc Growth n the Less Developed Countres: An Emprcal Study for , Economc Development and Cultural Change, 35, Levne, R and Renelt, D. (1992): A Senstvty Analyss of Cross-Country Growth Regressons, Amercan Economc Revew, 82 (4), Nelson, C. R. and Plosser, C. I. (1982): Trends and Random Walks n Macroeconomc Tme Seres: Some Evdence and Implcatons, Journal of Monetary Economcs, 10 (September), Ram, R. (1986): Government Sze and Economc Growth: A New Framework and Some Evdence from Cross-secton and Tme Seres Data, Amercan Economc Revew, 76, World Bank (1999): World Development Report , Washngton, D.C. 23