The Effect of Inequality on Growth: Theory and Evidence from the Indian States

Transcription

1 Review of Development Economics, 8(1), , 2004 The Effect of Inequlity on Growth: Theory nd Evidence from the Indin Sttes Sugt Ghosh nd Srmisth Pl* Abstrct The pper exmines the effect of inequlity on growth mong the subntionl sttes in Indi. Theoreticlly, growth of the regionl economy is driven by productive public investment in the provision of helth nd eduction services finnced by liner output tx, nd the optimum tx rte is determined by the medin voter. In contrst to existing results, the uthors obtin n mbiguous reltionship between initil inequlity nd subsequent economic growth. Anlysis of the Indin stte-level dt suggests tht rurl inequlity influences growth of totl output more thn urbn inequlity, nd does so negtively. The indictor of intersectorl inequlity is more importnt in explining sectorl output growth. 1. Introduction Existing studies for developing countries (e.g., Rvllion, 1995; Rvllion nd Dtt, 1996) hve often emphsized the need to boost economic growth in n ttempt to reduce poverty. The reltionship between growth nd poverty is complex nd depends, to lrge extent, on the reltionship between growth nd inequlity (Dtt nd Rvllion, 1992). If there is rise in inequlity while the economy is growing, this my not only offset the poverty-reducing effects of growth, but my lso retrd subsequent growth through redistribution in fvor of nonccumulble fctors. The literture on the effect of inequlity on growth 1 hs gined momentum since the influentil ppers by Alesin nd Rodrik (1994), nd Persson nd Tbellini (1994) (herefter we bbrevite these uthors s AR nd PT, respectively). In both ppers there is redistributive role for the government to combt inequlity within democrtic setup. In AR, government investment in productive services finnced primrily through txtion of cpitl will interct with the growth-enhncing policies. With tx on cpitl, given the fmilir incentive nd disincentive effects on cpitl income, pure cpitlist (who hs no lbor income) prefers the growth-mximizing tx rte. Higher inequlity (defined in terms of the distribution of lbor endowment reltive to cpitl in the cross-section of popultion) will, however, induce the medin voter (who hs some lbor income) to prefer tx rte tht is greter thn the growth-mximizing tx rte, thus lowering growth. Similr results re obtined in the overlppinggenertions (two-period) model of PT, where txes re used only for redistribution; thus higher rte of cpitl tx unmbiguously depresses the incentive for privte investment nd growth. Both ppers nlyze the effects of the politicl outcome 2 (by * Ghosh, Pl: Crdiff Business School, Colum Drive, Crdiff CF10 3EU, UK. Ghosh: Tel: ; Fx: ; E-mil: ghosh@crdiff.c.uk. Pl: Tel: ; Fx: ; E-mil: pl@crdiff.c.uk. We would like to thnk Berk Ozler for providing us with the dt, nd Debsis Bndyopdhyy, Prntp Bsu, Jen Drèze, Gerry Mkepece, Andrew McKy, Ptrick Minford, Dni Rodrik, Udyn Roy, prticipnts of the Econometric Society 8th World Congress, Settle, 2000, nd n nonymous referee for helpful comments nd suggestions. Any remining errors re ours., 9600 Grsington Rod, Oxford OX4 2DQ, UK nd 350 Min Street, Mlden, MA 02148, USA

2 THEORY AND EVIDENCE FROM THE INDIAN STATES 165 ssuming voting process on the level of the tx rte) generted by given income distribution, nd suggest tht countries with greter economic inequlity experience lower future economic growth. Prtridge (1997) empiriclly exmines the nture of the inequlity growth reltionship t subntionl level for the sttes within the United Sttes nd finds tht there is positive reltionship between initil inequlity nd subsequent growth. Fctors like free interregionl mobility of physicl nd humn cpitl were conjectured to be responsible for this positive reltionship, though nothing ws formlly tested. Another interesting pper on the inequlity growth issue t subntionl level is by Rvllion (1998) for 131 counties of rurl Chin for the period , where he rgues tht ggregtion ffects conventionl tests of whether inequlity impedes growth. The present pper exmines the inequlity growth reltionship for developing country (Indi) nd modifies the theoreticl model developed by AR or PT to include fetures tht resemble the subntionl Indin sttes. Among the fetures of the theoreticl model re: model of overlpping fmilies of infinitely-lived gents à l Weil (1989), nd more importntly, n output tx (insted of cpitl tx s in AR or PT). Output txes closely resemble the txes on sles nd purchses which re very importnt source of stte revenues in Indi. We lso consider the provision of public services in helth nd eduction t the stte-level which is entirely under the jurisdiction of the Indin sttes, s enumerted in the stte list of the Indin constitution. In section 2, we consider n endogenous growth model where growth is driven by productive investment in the provision of such public services finnced by n output tx, nd the optimum tx rte is determined by the politicl process (medin voter rule) s in democrtic setup like Indi. As the tx is on output, it ffects both lbor nd cpitl income. Since the medin voter hs both kinds of income, the effect of n output tx on the medin individul s utility is mbiguous. Consequently, the medin voter will choose n optimum tx tht my be higher or lower thn the growth-mximizing tx rte depending on whether the redistribution rises or lowers his/her net lbor income. Using the stte-level dt for the period , the second prt of the pper empiriclly exmines the nture of the inequlity growth reltionship for the Indin sttes: section 3 describes the dt while section 4 reports the econometric nlyses. We perceive tht there is some vlue-dded in our empiricl exercise: first, most existing empiricl studies for developing countries exmine the effect of growth on ensuing inequlity (e.g., Rvllion, 1995; Rvllion nd Dtt, 1996). The only study tht considers the two-wy reltionship between growth nd inequlity is tht by Deininger nd Squire (1998). Using dt from cross-section of countries, they find little support for the Kuznets hypothesis while there is strong negtive reltionship between growth nd inequlity. There re obvious problems with the cross-country comprisons becuse of heterogeneity mong ntionl economies. The problem is minimized if one considers the subntionl sttes insted. Indi is n interesting cse in point, s the importnce of economic growth hs long been emphsized in its fight ginst poverty. Within the federl setup, Indin sttes re lso sufficiently diverse in terms of geogrphic, demogrphic, nd economic chrcteristics. Existing empiricl studies consider intercountry comprisons (see AR, PT, Deininger nd Squire) or interregionl comprison for developed country (Prtridge), while we focus on the interstte comprison for the developing economy of Indi. Also given the pronounced dichotomy between rurl nd urbn sectors of n Indin stte, we exmine the sectorl nture of this reltionship, something tht hs not been explored before.

3 166 Sugt Ghosh nd Srmisth Pl 2. Theoreticl Frmework In this section we develop theoreticl model of growth with redistribution t regionl level. The ntionl economy consists of n similr subntionl regions (or sttes within country). A single representtive region is denoted by. Producers in region comprise lrge number of individuls. The output of individul i in region is y i A k i g1 l i g2 G l i 1-g1-g2 = 1( ) ( ) ( K ), A1 > 0, 0< g 1+ g 2 < 1 (1) where k i is his cpitl stock endowment nd l i is the number of units of lbor supplied. G is the region-wide public investment in helth nd eductionl services, nd K is the cpitl stock of the regionl economy. The inclusion of G nd K in the production function represents externlity effects rising out of public investment in the region à l Brro (1990) nd knowledge effects from the cpitl stock of ll firms in the regionl economy à l Romer (1986), respectively. Aggregting over ll individuls in region, given CRS, we hve: -g Y = A K G 1 2 g2 1. In deriving (2), the ggregte lbor endowment L (= L 0 e nt ) is normlized to unity by suitble choice of units. Here n is the rte of growth of the lbor force. The government blnces its budget in every period, but unlike Alesin nd Rodrik who consider txtion of cpitl only, we consider the txtion of overll fctor income (or output), given by t : G = T =t Y. (3) Defining w nd r s the competitively determined privte mrginl products of lbor nd cpitl, respectively, we hve (2) w r y i A k i 1 Ê ˆ = ( 1-t ) = ( 1-t )( 1- g 1) 1Á t l Ë l i y i A k i 1 Ê ˆ = ( 1-t ) = ( 1-t ) g 1 1Á t k Ë l i g g2 g2 1-g1-g2 Y K i g -1 g2 g2 1-g1-g2 Y K i., (4) (5) The consumption side of our model is vrint of the Weil (1989) model where the economy of region consists of mny infinitely-lived fmilies which re not linked to preexisting cohorts through intergenertionl trnsfers. The utility function of n gent i (representtive of the cohort born t time s) s of time t in region is logrithmic: [ ] - - u i = Ú ln c i ( s, q v) e ( v t) dv. t This is mximized subject to dynmic budget constrint: k i = ( 1 -t ) y i - c i = [ rk i + wl i ]- c i. The optimiztion for the representtive individul then gives d ( c i )=( r -q) c i. dt Integrting (7) nd (8) nd ppropritely combining them yields (6) (7) (8)

4 THEORY AND EVIDENCE FROM THE INDIAN STATES 167 c ( ) = q k + w, i i i (9) where w i is the present discounted vlue of future lbor income ccruing to those currently live; i.e., w t - Ú [ r ( m) ] dm i 0 Ú t wl i st, wl e dt=, r ( )= ( ) (10) We next consider the politicl process of mjority voting through which the government chooses tht tx rte which mximizes the welfre of the medin voter. Substituting (8) nd (9) into the utility function (6), nd then integrting by prts, we hve n expression for the indirect utility function for the ith individul (u i ) which cn be written s follows: r u i wl i = Ê 1ˆ -q [ lnq + ln( )- ln r ]+. Ë 2 q q (11) The optiml tx t i * stisfies du i /dt = 0, nd the second-order condition. Note tht u i is given by (11), nd w nd r in (11) re given by (4) nd (5). This enbles us to obtin expressions for w / t nd r / t. We cn thus obtin the optiml tx rte t i * which will depend on s i,where s i is defined s y Y Li Ki = y y y Y = Y Y. s i Li L Ki Li Ki Here y nd y re individul i s income from lbor nd cpitl, respectively; nd Y L, Y K re ggregte incomes of region from lbor nd cpitl, respectively. Thus the choice of t i * will depend on the vlue of s i. Given n initil distribution of lbor nd cpitl income (corresponding to certin vlue of t ), higher t ought to chnge the individul s lbor/cpitl income rtio nd lso tht of the region (i.e., both the numertor nd denomintor of the bove expression will chnge). It is therefore not cler how s i will chnge with respect to t, nd consequently the medin voter s optimum tx rte t* will be mbiguously relted to s i. The intuition for this comes from the fct tht here the tx is on output, so there is n incentive s well s disincentive effect on both lbor nd cpitl income. This is in contrst to AR where lbor income responds positively to higher tx rte (which is exclusively on cpitl). Consequently, in our cse, the medin voter who hs both lbor nd cpitl income responds mbiguously to higher t.the pure cpitlist (s in AR) prefers the growth-mximizing tx rte ( ˆt ), but here the medin voter s idel t (t* ) could be greter or less thn ˆt. Therefore, the effect of initil inequlity on growth could be negtive or positive. In order to derive the growth-mximizing tx rte, we use the goods mrket equilibrium condition for region in per cpit terms: k y -c -g e = = - n, (12) k k K L K where e is the per cpit (endogenous) growth rte for the region. Also, k = K /L, y = Y /L, c = C /L, g = G /L = t /L, nd n = /L. L ȧ

5 168 Sugt Ghosh nd Srmisth Pl The evolution of ggregte consumption cn be shown to stisfy Ċ = (r + n - q) C - nqk, from which it follows tht e = c r q nq k c = - - c. (13) In the stedy stte, dk /dt = 0, nd lso dc /dt = 0 by definition. Combining equtions (12) nd (13) by eliminting c, nd then using the blnced-budget constrint (3) in per cpit terms, gives us n eqution (not shown) linking e nd t, from which the Brro (1990)-type inverse U-shped reltion 3 between e nd t emerges here with overlpping fmilies of infinitely-lived gents. The growth-mximizing tx rte ( ˆt ) stisfies de /dt = 0. We next consider the importnt cse tht is typicl of mny developing countries, where proportion (m) of the totl tx revenues T is used by the government s redistributive trnsfers to the lbor component of individul income. This mens tht the proceeds of the tx revenue ugment n individul s lbor income by the mount mt y. i Lbor income fter the trnsfer is therefore w l i + mt y, i nd cpitl income is r k. i Also, the government spends (1 - m)t y i on helth nd eduction schemes, which mens tht we now hve the government budget constrint s G = (1 - m)t Y. By retrcing the steps s before, we cn find the tx rte tht mximizes the utility of the medin individul. Once gin, it is uncler whether the medin voter prefers tx rte (t* ) higher thn the growth-mximizing tx rte ( ˆt ). Wht we cn sy, however, is tht with redistributive trnsfers from the output tx revenue, the medin voter s utility-mximizing tx rte ought to be higher thn without the redistribution scheme. In this sense the possibility of initil inequlity hving negtive impct on growth (s in AR) is incresed, but it is still not certin tht this would be the cse. 3. A Cse Study of Indin Sttes Sections 3 nd 4 will exmine the nture of the reltionship between inequlity nd growth, using stte-level dt from Indi over The Indin constitution gve strong economic/finncil powers to the ntionl government (center) with respect to the core sectors, including industry, defence, rilwys, post nd telegrph, nd tomic energy; while sttes hve primry control in helth nd eduction. The most elstic sources of tx revenue, nmely income, excise nd customs txes, re levied by nd ccrue to the center, while txes on properties, purchses, nd sles (most importnt source of stte tx revenue) re levied by the sttes. However, most expenditures re incurred t the stte level. Given this imblnce in finncil powers between the center nd the sttes, there re different mechnisms of trnsfer of resources from the center to the sttes. Compred to sttes shre in centrl txes, proportions of stte txes re generlly higher in totl stte revenues, which finnce vrious developmentl nd nondevelopmentl expenditure. The bulk of the stte revenues is spent on the development ccount, which includes expenditure on helth, eduction, nd community services; n interstte vrition is noted here. Our theoreticl model ttempts to cpture the importnt chrcteristics of the Indin federl sttes. For exmple, proportionl output tx closely resembles txes on sles nd purchses, which re one of the min sources of stte txes in Indi. Also, public development expenditure in the provision of socil nd economic services fetures prominently in the Indin sttes, nd its inclusion in the production function thus cptures n importnt chrcteristic of the Indin subntionl economy. Finlly, we

6 THEORY AND EVIDENCE FROM THE INDIAN STATES 169 consider the politicl process of mjority voting to determine the optimum tx rte, which reflects the democrtic nture of the process in line with the Indin prctice. Testble Hypotheses For given vlues of the rte of growth of popultion (n) nd the rte of time preference (q) tken to be identicl for ll regions under considertion growth of output in ny stte (= 1, 2,...,j) over period of time is function of initil output, Y 0, cpitl, k 0, nd some index of initil distribution, l 0 : e = fy ( 0, k0, l0). (14) Inclusion of Y 0 s the level of output or income in the initil period llows us to test the vlidity of the hypothesis of growth convergence s dvocted by Brro (1991). In endogenous growth models with externlities, convergence holds only if mesures of initil humn cpitl re held constnt (Brro, 1991). Thus for given level of initil cpitl, Y 0 is expected to hve negtive influence on growth in the cross-sectionl nlysis. k 0 refers to indices of both physicl nd humn cpitl, both of which re expected to exert positive influence on growth (lso see discussion in section 4). The centrl hypothesis of our concern is tht, other things remining unchnged, economic growth e in ny stte depends on initil inequlity l 0 in the stte: initil inequlity my hve negtive or positive effect on growth, depending on whether the medin voter desires tx rte tht is higher or lower thn the growth-mximizing rte. In the Indin context, the medin voter is clerly one who hs more lbor income thn cpitl income, which follows from the fct tht Indi is lbor-surplus economy. It is lso true tht populrly elected governments undertke redistributive trnsfers in fvor of the poor (whose income is lrgely lbor income). These two fctors re more likely to result in negtive reltion between inequlity nd growth, s the medin voter s preferred tx rte is more likely to be higher thn the (regionl) economy s growth-mximizing tx rte. 4 Indin sttes re often chrcterized by the pronounced dichotomy between rurl nd urbn sectors within stte. Although in the theoreticl section we do not model the rurl nd urbn sectors explicitly, we cn still obtin testble hypotheses regrding the link between initil inequlity nd growth from sectorl perspective. If we now mke the plusible nd quite relistic ssumptions tht (1) the rurl sector comprises individuls hving mostly lbor income, nd (2) the government levies txes to redistribute incomes in fvor of the rurl sector, then the medin voter (who in the Indin context hs more lbor thn cpitl income nd thus cn be ssumed without much loss of generlity to belong to the rurl economy) would fvor higher txes. And given tht the growth rte hs n inverse U-shped dependence on the tx rte, it is more likely tht the medin voter s idel tx rte will lie to the right of the growth-mximizing rte. In other words, rurl inequlity mtters more in the Indin context, nd it is lso more likely tht higher initil rurl inequlity would lower growth. Description of Dt The dt used for our purposes re obtined primrily from vrious government sources like the Ntionl Smple Survey (NSS), government ccounts, nd dtset compiled by the World Bnk (Ozler et l., 1996). This ltter is unique dtset tht contins, mong other things: informtion on net sown re (for ll crops); net stte

7 170 Sugt Ghosh nd Srmisth Pl domestic product (SDP) including sectorl SDPs for griculture, mnufcturing, etc.; popultion, rurl, nd urbn Gini coefficients in the distribution of consumer expenditure; nd stte-level developmentl expenditure on the public provision of socil nd economic services, including helth nd eduction, for 16 mjor sttes in Indi. For the nlysis of this pper, we study the period for which the dtset is complete. This bsic dtset hs been supplemented by the informtion on litercy (source: Reports of the Census nd the Eduction Deprtment, Government of Indi, vrious issues) nd lso tht on stte-level txes nd expenses (source: Reserve Bnk of Indi) for these sttes. An nlysis of rurl nd urbn inequlity Gini indices in 1960 nd output growth per cpit over does not suggest ny specific pttern (positive or negtive) in the reltionship between initil inequlity nd subsequent growth. There re sttes with low (high) initil inequlity nd high (low) subsequent growth, s well s those with low (high) initil inequlity nd low (high) subsequent growth. For exmple, mong the six most initilly unequl rurl regions (Rjsthn, MP, Krntk, Kerl, AP, nd Gujrt), the nnul rte of growth ws low (less thn 1%) in four (the exceptions being AP nd Gujrt). Among the sttes with low rurl inequlity in 1960, Bihr, Jmmu nd Kshmir (J&K), nd West Bengl hd low growth (below 1%), while Mhrshtr nd Tmil Ndu hd high nnul growth rtes round 4%. Punjb witnessed one of the highest rtes of growth mong the Indin sttes, nd levels of initil rurl nd urbn inequlity hve been seventh lrgest mong the selected sttes. Tble 1 shows the bivrite correltion coefficients between growth of output per cpit nd rurl nd urbn Gini indices. Annul growth rte of per cpit output is negtively nd sttisticlly significntly relted to both rurl nd urbn inequlity Gini indices. There is significnt nd positive correltion between per cpit development expenditure nd stte-level tx rtes, nd between Gini indices nd stte-level tx rtes, s ssumed in our theoreticl model. The bivrite correltion nlysis ssumes tht the sttes under considertion re identicl in ll respects other thn growth nd inequlity. There is, however, significnt interstte vrition in the initil levels of income per cpit, litercy rtes, net sown re, nd per cpit stte development expenditure. This, in turn, necessittes n ssessment of the inequlity growth reltionship in multiple regression frmework, fter controlling for ll possible fctors ffecting the reltionship. Tble 1. Bivrite Correltion Coefficients RGINI60 UGINI60 PCDEXP60 TAXRT60 GRPC6094 GRAG6088 GRMF6088 RGINI ** 0.638** 0.247* ** ** UGINI ** * 0.304* * ** PCDEXP ** * TAXRT * 0.304* 0.490* GRPC ** * * 0.641** GRAG * ** GRMF ** * * 0.673** Notes: *nd ** denote the level of significnce t 10% nd 1%, respectively. RGINI60, UGINI60, PCDEXP60, TAXRT60: initil vlues of rurl Gini, urbn Gini, per cpit development expenditure, nd tx rte in 1960, respectively. GRPC6094: nnul rte of growth of SDP per cpit during ; GRAG6088, GRMF6088: nnul rte of growth of griculturl nd mnufcturing SDP per cpit during

8 THEORY AND EVIDENCE FROM THE INDIAN STATES Econometric Anlysis Bsed on multiple regression frmework, this section exmines the effect of initil inequlity on subsequent economic growth in the Indin sttes. Single Cross-sectionl Estimtes The simplest wy to proceed is to mke use of the single cross-sectionl frmework where the growth rte of output over the smple period is determined in terms of vlues of explntory vribles (including indices of inequlity) previling in the initil yer, The resultnt smple is rther smll s we hve informtion on only 16 mjor Indin sttes observed during this period. Nevertheless, one my rgue tht single cross-sectionl frmework is more suitble to nlyze the effect of initil inequlity on economic growth since it tkes time for the politicl process to ffect distribution nd subsequent growth. Hence, we strt with the single cross-sectionl nlysis of the complete model nd then delete the insignificnt vribles to obtin prsimonious model. The dependent vrible of our interest is the rte of growth of per cpit stte domestic product (GRPCINC) over the period Choice of the explntory vribles is guided by eqution (14). However, given the nture of the dt t our disposl, we need to mke some modifictions to the growth eqution we estimte. We express stte domestic product, its growth, nd lso cpitl s proportion of totl popultion in per cpit terms. For the Indin economy, the size distribution of income is not redily vilble; wht we observe is the size distribution of per cpit monthly consumer expenditure 6 (vilble from the NSS reports) which in turn genertes the estimtes of inequlity Gini indices used in our nlysis. However, we observe seprte indices only of the rurl nd urbn inequlity Gini indices, but not n index of totl stte-level inequlity. Totl inequlity cn, however, be thought of s n ggregte of rurl inequlity (RGINI), urbn inequlity (UGINI), nd the inequlity between rurl nd urbn res. In the bsence of better mesure of the intersectorl inequlity component, we construct the rtio of rurl monthly men consumption to urbn monthly men consumption nd use this rtio (RRUMEAN) s n indictor of intersectorl inequlity. 7 The men rurl consumption is reltively higher thn men urbn consumption if RRUMEAN is greter thn one, nd reltively lower if it is less thn one. If the distribution of consumption is positively skewed, the medin is lwys lower thn the corresponding men. Thus, we include UGINI, RGINI, nd RRUMEAN s the three components of totl inequlity in the growth regression. One my lso rgue for the inclusion of the stte tx rte in our regression. However, there is significnt sttisticl correltion between the tx rte nd inequlity indices. Accordingly, we include only inequlity indices (nd not the tx rte) in our regressions. The set of explntory vribles includes per cpit stte domestic product (PCSDP60), rurl Gini (RGINI60), nd urbn Gini (UGINI60) coefficients, nd the intersectorl inequlity component (RRUMEAN60) previling in the initil period, In view of the difficulty in obtining n overll index of ggregte cpitl t the Indin stte level (Loh, 1995), we include three indictors of cpitl litercy rte (LITRT60) s n index of humn cpitl, per cpit sown re (PCAREA60), nd per cpit stte expenditure in the provision of socil nd economic services (PCDEXP60) in the yer 1960 s indices of physicl cpitl. Mens nd stndrd devitions of explntory vribles re given in column 1 of Tble 2. We first estimte the complete model (column 2) nd then delete the insigni-

9 172 Sugt Ghosh nd Srmisth Pl Tble 2. Single Cross-sectionl OLS Estimtes of Growth of Totl Output Dependent vrible: GRPC6094 Explntory Men (SD) Coefficients (t-rtio) Coefficients (t-rtio) vribles (1) (2) (3) Intercept 4.5 (21.694)** 4.5 (22.386)** PCSDP (82.22) (2.989)** (2.969)** RGINI (8.22) (5.113)** (4.927)** UGINI (0.827) RRUMEAN (0.17) (2.304)* (2.273)* LITRT (0.10) 2.27 (6.341)** 2.32 (7.202)** PCAREA (0.19) 1.8 (7.452)** 1.8 (7.423)** PCDEXP (0.09) 1.3 (2.414)* 1.17 (1.976)* R F-sttistic 7.41** 9.52** No. of observtions Notes: * denotes tht the vrible is significnt t the 10% level, ** t the 1% level. PCSDP60, LITRT60, PCAREA60, PCDEXP60: initil vlues of totl output per cpit, litercy rte, net sown re, nd stte development expenditure per cpit in 1960, respectively. RGINI60, UGINI60, RRUMEAN60: initil vlues of rurl, urbn nd intersectorl inequlity indices in 1960, respectively. These estimtes use White s correction for heteroskedsticity. ficnt vribles to obtin more prsimonious specifiction (column 3). These estimtes use White s correction for the presence of heteroskedsticity. R 2 nd F sttistics describe the goodness of fit of ech specifiction. A comprison of these two sets of estimtes suggests tht the prsimonious model yields slightly better results in terms of the F-sttistic. The min inferences from our estimtes re worth noting. The coefficient of PCSDP60 is negtive nd significnt, implying tht sttes with reltively higher level of initil output per cpit hve significntly lower growth rte over the period This in turn lends support to Brro s b convergence hypothesis. It is lso noteworthy tht the rte of convergence mong the Indin sttes hs been rther low, s hs been observed by Cshin nd Shy (1996). Secondly, the effects of PCAREA60, PCDEXP60, nd LITRT on growth per cpit revel tht the coefficients of ll three types of cpitl re positive nd significnt: so both physicl nd humn cpitl per cpit significntly enhnce growth of totl output per cpit, thus confirming our priori expecttion. More importntly, the coefficient estimte of RGINI60 is negtive nd significnt while tht of UGINI60 is positive (but insignificnt): thus initil rurl inequlity hs negtive reltionship with economic growth per cpit in the ensuing period while initil urbn inequlity hs positive but insignificnt impct on growth. In ddition, the intersectorl inequlity index RRUMEAN60 is negtive nd significnt. In terms of the testble hypotheses rising out of our theoreticl model, the result on RGINI seems plusible while tht on RRUMEAN60 seems less plusible. As discussed in section 3, given our ssumptions, we expect initil rurl inequlity (s indicted by the rurl Gini index, RGINI) to hve negtive impct on growth. With regrd to the sign for RRUMEAN60, ssuming tht the medin voter lives in the rurl sector nd tht the populrly elected government uses n output tx to redistribute incomes in fvor of this sector (see the discussion in section 3), the medin voter will desire reltively higher tx (which is expected to led to lower growth) when men

10 THEORY AND EVIDENCE FROM THE INDIAN STATES 173 consumption in the rurl sector is lower, nd vice vers when the men rurl consumption is higher. In other words, we expect the coefficient of RRUMEAN to be positive when regressed on growth. One possible sttisticl reson for this inconsistency is the omitted-vrible bis in the single cross-sectionl regression tht rises from ignoring the stte-specific spect of the ggregte production function. Islm (1995) rgues tht this omitted-vrible bis cn be redressed if one considers pnel-dt pproch insted, which tkes us to the next section where we reexmine this inequlity growth reltionship by considering both the stte-specific effects s well s the vrition over time. Pnel-dt Estimtes Islm (1995) rgued tht the significnt convergence result of cross-country regression nlyses my be revised in pnel-dt nlysis when the differences in the ggregte production function cross economies re llowed for. In our cse, however, the use of pnel dt introduces individul stte-specific effects s hs been indicted by the striking diversity mong the Indin sttes (see the discussion in section 3). The sttespecific spect of the ggregte production function tht is ignored in the single crosssectionl regression is correlted with the included explntory vribles tht cretes omitted-vrible bis. The pnel pproch s discussed below redresses this bis. Let us now modify eqution (14) to consider simple fixed-effects model of the following form for the sth stte, s = 1, 2,...,16: est = b xst-1 + fs + ust, (15) where f s is the fixed effect, x st-1 is the k 1 vector of lgged exogenous regressors, nd u st is the vector of rndom disturbnce term. The fixed-effects f s of (15) will ccount for the unobserved differences in growth due to interstte differences in history nd economic structure, nd will cpture the heterogeneity tht cuses the inconsistency in the OLS regression. We consider fixed-effects rther thn rndom-effects model becuse the set of sttes selected in our dtset is unlikely to be rndom it is rther conscious choice lrgely determined by the vilbility of relevnt informtion. We hypothesize tht the growth of output per cpit for given stte s over ech subperiod t depends on: output per cpit (PCSDP); rurl, urbn, nd intersectorl inequlity indices (RGINI, UGINI, RRUMEAN); litercy rte (LITRT); per cpit totl sown re (PCAREA); nd per cpit stte development expenditure (PCDEXP) t the beginning of tht subperiod. We correct for the presence of heteroskedsticity in the error structure (using White s correction) when stte-level heterogeneity is considered in the fixed-effects model. There re 35 nnul observtions (over ) for ech of 16 mjor sttes in Indi, though there re missing observtions on inequlity Gini indices for the yers when the NSS ws not conducted. Since growth tkes plce over period of time where the short-term disturbnces for nnul observtions my feture prominently, we rerrnge the nnul dt so tht 35 yers of nnul observtions for ech stte is divided into seven subperiods, nmely , , , , , , nd The choice of these subperiods hs been dictted by the yers for which NSS informtion on inequlity Gini indices re vilble. Thus, the totl number of observtions for 16 sttes turns out to be 112. Using this five-yerly pnel dt, we estimte the fixed-effects lest-squres dummy vrible (LSDV) model (15). Mens nd stndrd devitions of the regression vribles re shown in column 1 of Tble 3, while the coefficient estimtes with t-sttistics (shown in prentheses) re

11 Tble 3. Lest-squres Dummy-vrible Pnel Estimtes of Growth GRPCY GRPCAGY GRPCMFGY Men (SD) Coefficient (t-rtio) Men (SD) Coefficient (t-rtio) Coefficient (t-rtio) Vribles (1) (2) (3) (4) (5) PCSDP (137.1) (4.1)** PCAGY (49.6) (2.1)* PCMFGY (30.4) (1.65)* RGINI (6.26) (3.35)** (6.68) (0.551) UGINI (6.62) (7.02) (0.485) RRUMEAN 0.73 (0.16) (1.688)* 0.73 (0.173) 0.2 (1.948)* 0.98 (1.917)* LITRT 0.37 (0.16) 0.23 (5.770)** 0.35 (0.15) 3.2 (2.227)* 16.5 (2.188)* PCAREA 0.27 (0.15) 0.28 (0.16) PCDEXP 0.39 (0.48) 0.34 (0.49) R F-sttistic 2.77* 2.93* 2.67* No. of observtions Sugt Ghosh nd Srmisth Pl Notes: * denotes tht the vrible is significnt t the 10% level, ** t the 1% level. RGINI, UGINI, RRUMEAN: rurl, urbn, nd intersectorl inequlity indices. LITRT, PCAREA, PCDEXP: litercy rte, net sown re (hectres) for ll crops per cpit, nd stte development expenditure (in Rs) per cpit, respectively. GRPCY, GRPCAGY, GRPCMFGY: growth of totl output, griculturl SDP, nd mnufcturing SDP per cpit, respectively. PCSDP, PCAGY, PCMFGY: totl output, griculturl nd mnufcturing SDP per cpit (in Rs), respectively.

12 THEORY AND EVIDENCE FROM THE INDIAN STATES 175 given in column 2 of the tble. All estimtes re corrected for the presence of heteroskedsticity. We compre the LSDV generlized lest-squres estimtes (column 2 of Tble 3) with corresponding single cross-sectionl OLS estimtes with the prsimonious specifiction shown in Tble 2 (column 3). As before, there is evidence of convergence of totl output mong the Indin sttes s the coefficient of initil stte-level output per cpit (PCSDP) is negtive nd significnt though the bsolute vlue of the coefficient is now smller. Among vrious cpitl vribles, only LITRT still significntly enhnces growth per cpit, while the coefficients of PCDEXP nd PCAREA re now insignificnt (nd hence dropped from the prsimonious specifiction). The insignificnce of the two cpitl terms could be ttributed to the limited smple size. Significnt differences rise with respect to the impct of inequlity on growth: while in single cross-sectionl estimtes the coefficients of both RGINI nd RRUMEAN were negtive nd significnt, in the LSDV model the coefficient of RGINI is negtive nd tht of RRUMEAN is positive, both coefficients being significnt. Thus we obtin consistent LSDV estimtes such tht lower rurl or intersectorl inequlity enhnces economic growth per cpit; however, urbn inequlity still fils to hve significnt impct on subsequent growth in the pnel-dt nlysis. In view of the dichotomous effect of rurl nd urbn inequlity on economic growth, finlly we estimte the rte of growth of griculturl nd mnufcturing output per cpit in terms of rurl, urbn, nd intersectorl inequlity, respectively, mong other things. As before, the finl specifiction s summrized in columns 4 nd 5 of Tble 3 is obtined by deleting the insignificnt cpitl vribles. These estimtes re lso corrected for the presence of heteroskedsticity. In this cse, however, only the intersectorl inequlity term is significnt for both rurl nd urbn sectors, nd the coefficient is positive s in the cse of totl output growth per cpit (column 2 of Tble 3). While the coefficient of RGINI is still negtive, this bsolute rurl inequlity does not hve ny perceptible impct on growth in either sector. In other words, these results indicte the importnce of sectorl interlinkges in explining sectorl growth. Tken together, our results indicte tht higher initil rurl inequlity lowers subsequent growth of totl output while urbn inequlity does not hve ny perceptible impct. One plusible explntion of this result is tht bout 70% of the Indin popultion still lives in rurl res so tht populr government is urged to respond more to rurl (bsolute nd/or reltive) inequlity by undertking vrious tx-finnced redistributive progrms. In ddition, we find tht the indictor of the intersectorl inequlity component is more importnt thn urbn inequlity, especilly when we consider sectorl growth regressions. Thus compred to most existing studies like by AR (1994), PT (1994) or Deininger nd Squire (1998), we derive richer set of results. In ddition to the decomposition of totl inequlity into sectorl inequlity components nd the use of the pnel-dt technique, our results my lso be influenced by the fct tht we use consumption inequlity insted of income inequlity. Since the extent of consumption inequlity is generlly less thn tht of income inequlity, it is expected tht the optimum tx of the medin voter would be less thn in the cse of income inequlity, thus lowering the disincentive effect of tx on investment nd subsequent output growth. 5. Conclusion In terms of n endogenous growth frmework this pper hs exmined how initil inequlity ffects economic growth in the ensuing period, where growth of the regionl

13 176 Sugt Ghosh nd Srmisth Pl economy is driven by productive public investment finnced by liner output txtion. It is suggested tht initil inequlity in the distribution of income leds to the optimum rte of txtion (determined by the medin voter) being different from the rte tht mximizes the economy s growth rte. However, the precise reltionship remins mbiguous nd depends on the net effect of the output tx on lbor nd cpitl income of the medin voter. Empiricl estimtes from the Indin sttes suggest tht rurl inequlity is more importnt to explin growth of totl output per cpit nd there is negtive reltionship between the two. It is lso cler tht the indictor of intersectorl inequlity is quite importnt in explining sectorl output growth. References Alesin, Alberto nd Dni Rodrik, Distributive Politics nd Economic Growth, Qurterly Journl of Economics 109 (1994): Brro, Robert J., Government Spending in Simple Endogenous Growth Model, Journl of Politicl Economy 98 (1990):S , Economic Growth in Cross-Section of Countries, Qurterly Journl of Economics 106 (1991): Blundell, Richrd nd In Preston, Consumption Inequlity nd Income Uncertinty, Qurterly Journl of Economics 113 (1998): Cshin, Pul nd Rtn Shy, Internl Migrtion, Center Stte Grnts nd Economic Growth in the Sttes of Indi, IMF Stff Ppers 43 (1996): Dtt, Gurv nd Mrtin Rvllion, Growth nd Redistribution Components of Chnges in Poverty Mesures: Decomposition with Applictions to Brzil nd Indi in the 1980s, Journl of Development Economics 38 (1992): Deininger, Klus nd Lyn Squire, New Wys of Looking t Old Issues: Inequlity nd Growth, Journl of Development Economics 57 (1998): Glor, Oded nd Joseph Zeir, Income Distribution nd Mcroeconomics, Review of Economic Studies 60 (1993): Islm, Nzrul, Growth Empirics: Pnel Dt Approch, Qurterly Journl of Economics 110 (1995): Kuznets, Simon, Economic Growth nd Income Inequlity, Americn Economic Review 45 (1955):1 28. Loh, Jcky, Eduction nd Economic Growth in Indi: n Aggregte Production Function Approch, in School Effectiveness nd Lerning Achievement t Primry Stge, New Delhi: Ntionl Council of Eductionl Reserch nd Trining (1995). Ozler, Berk, Gurv Dtt, nd Mrtin Rvllion, A Dtbse on Poverty nd Growth in Indi, Policy Reserch Deprtment, World Bnk (1996). Prtridge, Mrk D., Is Inequlity Hrmful for Growth? Comment, Americn Economic Review 87 (1997): Persson, Torsten nd Guido Tbellini, Is Inequlity Hrmful for Growth? Americn Economic Review 84 (1994): Rvllion, Mrtin, Growth nd Poverty: Evidence for Developing Countries in the 80s, Economics Letters 48 (1995): , Does Aggregtion Hide the Hrmful Effects of Inequlity on Growth? Economics Letters 61 (1998):73 7. Rvllion, Mrtin nd Gurv Dtt, How Importnt to Indi s Poor Is the Sectorl Composition of Economic Growth? World Bnk Economic Review 10 (1996):1 25. Romer, Pul M., Incresing Returns nd Long-Run Growth, Journl of Politicl Economy 94 (1986): Weil, Philippe, Overlpping Fmilies of Infinitely-Lived Agents, Journl of Public Economics 38 (1989):

14 Notes THEORY AND EVIDENCE FROM THE INDIAN STATES While ccording to Kuznets (1955) inverted-u hypothesis, growth cuses higher or lower inequlity depending on the level of development, the direction of this cuslity hs been reversed in the recent endogenous growth literture. 2. There hve lso been studies tht focus on growth with nonpoliticl considertions of redistribution. For exmple, see Glor nd Zeir (1993). 3. This is cler from simultions with different vlues of t, nd using prmeter vlues consistent with the Indin cse. 4. Note from section 2 tht for the cse where m-proportion of tx revenues is used s redistributive trnsfers to the lbor component of income, this is wht we would expect. 5. In the single cross-sectionl nlysis, initil inequlity is considered to be predetermined reltive to growth over the next 35 yers; thus ny direct custion from growth to inequlity is ruled out. 6. Consumption is well insulted from trnsitory movements nd cn thus help us focus on the pure long-run component of income inequlity (Blundell nd Preston, 1998). 7. We thnk referee for suggesting this mesure.