An Application of Spatial-Panel Analysis:

Transcription

1 CAADIA SOCIAL SCIECE Vol. 6, o. 3, 00, pp ISS An Applcaon of Spaal-Panel Analss: Provncal Economc Growh and Logscs n Chna UE APPLICAIO DE L'AALYSE DE PAEL SPAIAL: LA CROISSACE ÉCOOMIQUE PROVICIALE E LA LOGISIQUE E CHIE SHAO Yang Absrac: hs paper nroduces he spaal panel auocorrelaon model, ulzes C-D producon funcons, consrucs he spaal economerc model and fnall sudes he spaal correlav beween provncal economc growh and logscs. B usng he spaal package of Malab sofware, verfes he possbl f here s he remarkable auocorrelaon of he Chnese provncal economc growh and local logscs. On he base of buldng he spaal panel model, we research he spaal quanave auocorrelaon of he Chnese provncal economc growh and local logscs. Kewords: economc growh; logscs; spaal panel auocorrelaon Résumé: Ce arcle présene le modèle d'auocorrélaon de panel spaal, ulse les foncons de producon de C-D, consru le modèle économérque spaal e enfn éude la corrélavé spaale enre la crossance économque provncale e la logsque. En ulsan le paque spaal de logcel Malab, l vérfe la possblé de l'exsence d'une auocorrélaon remarquable de la crossance économque provncale chnose e la logsque locale. Sur la base de la consrucon d'un modèle de panel spaal, nous éudons l'auocorrélaon spaale quanave de la crossance économque provncale chnose e la logsque locale. Mos-clés: crossance économque; logsque; auocorrélaon de panel spaal Modern economc growh depends srongl on logscs. Logscs has become one of he mos mporan facors o promoe economc growh, adjus ndusral laou and drve he evoluon of economc spaal srucure. Prevous sudes of he relaonshp beween economc growh and logscs, lmed n me seres, whch gnored he dfferences beween locaons. hs paper nroduces he spaal facor no a unfed analcal framework, consders no onl he spaal heerogene bu also spaal correlaon beween economc growh and logscs. hs paper uses ndvdual fxed-effec model as he basc, School of Economcs and Managemen, Changchun Unvers of Scence and echnolog, 7989 Wexng Road, Changchun 300, Chna. Managemen College, Donghua Unvers, 88 Yan an Wes Road, Shangha 005, Chna. E-mal:shaoanglanglang@ahoo.com.cn * Receved on March 0, 00; acceped on Ma 3, 00 83

2 SHAO Yang/Canadan Socal Scence Vol.6 o.3, 00 panel-daa model, and uses laes spaal panel-daa model o sud he correlaon beween provncal economc growh and local logscs n Chna.. SPAIAL-PAEL MODEL AD CORRELAIO ES. Spaal-panel Models Spaal effecs of he spaal economercs nclude spaal auocorrelaon and spaal dfferences. he former s he correlaon of he observaons beween a regonal sample and oher regonal samples. he laer s he spaal-effec non-unform a he regonal level caused b he heerogene of spaal uns (Anseln, 988a. Spaal auocorrelaon n he spaal auoregressve model s refleced n he error erm and he lagged em of dependen varable. herefore here are wo basc spaal economerc models, one s Spaal Auo Regressve Model (SAR, he oher s Spaal Error Model (SEM, and he basc formulas of wo models are: Spaal Auo Regressve Model (SAR: Spaal Error Model (SEM: W X ( X W ( s he dependen varable, X s he vecor of ndependen varables (ncludng consan erm, β s varable facors, s spaal regresson coeffcens, s spaal auocorrelaon coeffcens, s he error componens obeng he normal dsrbuon, W s he spaal marx of n n (n s he number of regon, he wegh coeffcen can defned on acual condons. he above-menoned model s a model for he cross-seconal daa. In order o appl o panel daa, we need o change he model o mee he basc formula of panel daa model. hs paper uses ndvdual fxed-effec model (Elhors 003. he model conrols wo knds of non-observable effecs: spaal fxed-effec and me fxed-effec, he former s he effec of background varables whch changed wh he locaon, bu no changed wh me (such as economc srucure and naural endowmens, ec. on sead-sae level; he laer s he effec of background varables whch changed wh me, bu no changed wh locaon (such as he busness ccle and emporar shock, ec. on sead-sae level. o assume sf s -dmensonal column vecor of spaal fxed-effec; F s -dmensonal column vecors of me fxed-effec, he form s showng as follows: sf (, F (,,,,,, he column vecors of spaal and me fxed-effec of each observaon are showng as follows: s F, F Where s -dmensonal column vecor and s -dmensonal column vecor, all elemens of hese wo column vecors are. hen he equaon ( and ( can be ransformed no he followng model (3 and (4: (I W X (3 X In he one-dmensonal error decomposon model, error decomposon model,, ~ IID(0, ( I W (4 or ; In he wo-dmensonal ~ IID(0, and ~ IID(0,. s he me 84

3 SHAO Yang/Canadan Socal Scence Vol.6 o.3, 00 dmenson, s cross-secon dmenson, I s an un marx of -dmensonal me marx.. Spaal Correlaon es Spaal correlaon es bases manl on he hpohess esng of he maxmum lkelhood esmae, Wald, LR and LM sascs and spaal-relaed ndces Moran'S I, he null hpohess H 0 :ρ=0 orλ=0. However, Moran'S I (Moran,948, LMerr (Burrdge,980, LMsar, Lraos, Walds (Anseln,988b and oher spaal-relaed ess are appled for a sngle cross-secon regresson model, and can no be drecl used b panel-daa model. In hs paper, block-dagonal marx C I W replace spaal-wegh marx of Moran'S I sascs, ec. So we can easl exend hese ess o panel-daa analss. In he selecon of model, we use frsl he LSDV (Leas Square Dumm Varables mehod esmaon, do no consder he bound model of spaal correlaon, and hen carr ou he spaal-relaed es. If LMsar (or LMerr esmaon s more sgnfcan han LMerr (or Lmsar esmaon, hen he spaal lag model (or spaal error model s more approprae han he spaal error model (or spaal lag model. Anseln and Re (99 use Mone Carlo expermens mehod o show ha hs mehod can provde a good gudance on he selecon of spaal economerc models..3 Parameer Esmaon Usuall we use he maxmum lkelhood mehod (ML o esmae spaal economerc models (Anseln, 988a; Anseln and Hud99. ML esmaon program can no be used drecl b he spaal panel-daa model, because s appled o he cross-secon regresson model. In addon, when he dmenson of spaal-wegh marx s large, here s a problem (Kelejan and Prucha, 999 n he usual ML esmaon procedures n spaal economercs. A presen, a soluon can be used, s he Mone Carlo mehod o approxmae he log-lkelhood funcon, he Jacoban deermnan of naural logarhm (Barr & Pace, 999. hs mehod can be mplemened n he spaal package of Malab, and can be used o esmae model (3, (4.. MEASUREME MODEL he local producon funcon can be denoed b he homogeneous equaon of Cobb-Douglas: Y f K, L AK L (5 Logscs become more and more mporan facor n he process of producon, logscs has been seen as "he hrd prof source", same as he facors of capal and labor, o promoe economc growh. herefore, he C-D producon funcon s mproved. As he logscs(w s ndependen of capal and labor, on Solow producon funcon, he producon funcon whch nclude he elemens of logscs can be descrbed as: Y f(k, L, W AK L W (6 Y s he oupu, A s combnaon of echnologcal advance, K s capal nvesmen, L s labor npu, W s logscs, are elasc coeffcen of capal, labor, logscs on economc growh, respecvel. Consdered he comparson of daa and economc sgnfcance, all varables should be logarhmc, as follows: lny lna lnk lnl lnw (7 Subscrp s he provnce name, s me seres, s he random dsurbance. 3 EMPIRICAL AALYSIS 85

4 SHAO Yang/Canadan Socal Scence Vol.6 o.3, he Selecon of Facors and Daa Collecon Panel daa from 978 o 007of 30 provnces n Chna are used o be emprcal analss. Daa s manl from he "ew Chna, Complng Sascal Informaon on Ff-fve Years" and "Chna Sascal Yearbook" ( In order o compare daa and reduce heeroscedasc, all daa should be changed no logarhm.he deal daa of varables are as follows: ( GDP: for he elmnaon of prce change facors, we hnk 95 ear as he base perod, and generae he real GDP accordng o GDP ndex (Un: hundred mllon. ( Logscs level: we use cargo urnover of varous provnce o measure he level of logscs and logscs capac (un: 00 mllon on-km. (3 Labor: we use emplomen numbers of he whole soce (un: en housand. (4 Capal sock: we use a perpeual nvenor mehod (Goldsmh 95, whch s now used wdel b OECD counres, and s basc formula s: I K K, (8 K represens he capal sock of -regon's a -ear, K, - represens he capal sock of -regon's a (--ear, I represens he nvesmen of -regon's a -ear; s he -ear s economc deprecaon rae. We use Zhang Jun s capal sock daa whch menoned n he paper "Chna's Provncal phscal Capal Sock Esmae: ", and capal sock n he oher perod s calculaed b he daa of "Chna Sascal Yearbook". Deprecaon rae 5%. (Un: hundred mllon. 3. he Deermnaon of Economc Spaal-wegh (Wj Spaal-wegh marx (W j embodes he regonal spaal-effec. Obeng he rule of Rook, he adjacen rule, he marx W j s: w j 0 When he regon and he regon jare adjace When he regon and he regon jare no adjacen he man dagonal elemens are 0. w j (=,,,n,j=,,,nshould be sandardzed. here are borders beween neghborng regons, bu he economc es are no dencal beween neghborng regons. Relave o he backward regons, he drvng mpac of backward regons on developed regons s weak, whle he developed regons can generae grea drvng mpac on he backward around regons, whch s nensve spaal nfluence. herefore, we ge economc wegh-marx based on he bnar wegh marx (Ln Guang- Png, 005, he formula s: W* W E, E j, and, (9 0 o W s he wegh-marx of spaal locaon, E s he marx of economc srengh. We calculae he mean of proporon whch s he real GDP of ever regon accouned for real GDP of all regons, wh he resul of hem, measure he regonal economc level. And assumng ha he economc srengh of hs regon s srong, he spaal mpac of on surrounds s srong, conrar o he weak (Xao-png Chen, Guo-png L, 006. Economc spaal-wegh marx ( W j s he dagonal marx whch s he produc of a geo-spaal-wegh( w j and he mean of proporon of regonal GDP, he formula s: n Wj w j dag(,,, (0 and 0 o, n( 0 n o 86

5 SHAO Yang/Canadan Socal Scence Vol.6 o.3, Emprcal Analss Wh hese assumpons of he model and esmaon mehods, usng sub-provncal panel daa, we esablsh he ndvdual fxed-panel regresson model of 30 regons from 978 o 007, analzed b Evews 6.0 sofware, use LSDV mehod o esmae he ndvdual fxed-effecs model, and ge he elasc coeffcens and assocaed es resuls of he regresson model, and esmaes of he ndvdual fxed-effec coeffcens. he resuls are showng n he able : able : he emprcal resuls of ndvdual fxed-effec model of spaal-panel daa of varous provnce from 978 o 007 lnk lnl lnw c Coeffcen Sasc Prob R =0.963,Adjused R =0.986,F-Sa.=5.47,DW-Sa= In able he values of R and Adjus R s hgh n he regresson models, ndcae ha he resul of he smulaon fng of he model daa s ver good. In able judgng from he fxed-effecs esmae of varous regon, we can fnd he sze of he value of fxed-effecs s close n adjacen regons of Bejng and anjn, Jangsu and Zhejang, norheas, souhwes and norhwes provnces, shows ha here s sgnfcan regonal relevance, s necessar o do spaal-relaed es frsl, wh he resul of, we can know f s necessar o do spaal-panel analss furher. Accordng o he regresson resul, DW =0.4665, shows ha here s auocorrelaon beween he varables, and hen we es auocorrelaon of spaal regresson error erms, he followng esmae of he model are used wh Spaal Economerc Modules of Malab7.0, he resuls of esmaon are showng n he able 3. able : he cross-seconal esmae of nfluence coeffcens of varous provnce from 978 o 007 Provnce Provnce Provnce Provnce Bejng 0.55 Shangha Hube 0.35 Yunnan 0.77 anjng 0.36 Jangsu Hunan Shanx -0.0 Hebe Zhejang Guangdong 0.38 Gansu Shanx Anhu 0.70 Guangx Qngha Inner Mongola Fujan Hanan ngxa Laonng Jangx Chongqng Xnjang Jln Shandong Schuan Helongjang Henan Guzhou -0.3 able 3: he spaal correlaon es n =900 Lmerr Lmsar Lraos Moran I Walds value ch(.0 value Prob Wh he es resuls, fve es values (spaal dependence are ver sgnfcan (Prob. =0.0000, prove ha here s a sgnfcan spaal correlaon beween he logscs and regonal. hus he spaal facors mus be aken no accoun n order o show he neracon beween varous regons GDP and logscs. he es value of spaal-panel lag erm s bgger han he es value of spaal-panel error erm, 87

6 SHAO Yang/Canadan Socal Scence Vol.6 o.3, 00 ha s, Lmerror = <Lmsar = Lmerror = <Lmsar = Based on he crera descrbed prevousl, he Sar-panel lag model s he opmal model. he spaal-panel lag model s used o esmae he correlaon beween economc growh and he logscs. Resuls are showng n he able 4 and able 5: able 4: he esmaon resuls of sar-panel model parameer from 978 o 007 LnK LnL LnW / Sa Prob R-squared=0.9775,Rbar-squared=0.9758,sgma^=0.0308,log-lkelhood= able 5: he esmae of spaal-fxed nfluence coeffcen of varous regons from 978 o 007 Provnce Provnce Provnce Provnce Bejng Shangha Hube 0.43 Yunnan anjng 0.45 Jangsu Hunan Shanx Hebe Zhejang Guangdong Gansu Shanx Anhu 0.35 Guangx Qngha Inner Mongola Fujan Hanan ngxa Laonng Jangx Chongqng Xnjang Jln 0.53 Shandong Schuan -0.8 Helongjang Henan Guzhou In able 4 and 5, wh he resuls of model esmaon, we can draw he followng conclusons: Frsl, n he esmaon resuls, he esm aon of he parameersρn he spaal-panel lag model sgnfcance es s b %. I ndcaes ha here s a sgnfcan spaal correlaon beween GDP and logscs n 30 provnces. As he logscs has he nework properes, he logscs can connec economc acv no a whole un. hrough he spaal overflow (dffuson benef, he rapd economc growh regons drve he economc developmen of slower economc growh regons. I demonsraes posve spllover effec. Meanwhle he logscs wll have a negave spllover effec, producon facors flow easl no developed regons, he economc growh n a regon s lkel o be on he expense of economc recesson n oher regons. Secondl, he fng of R value whch we nroduced he spaal and me fxed-effecs no he spaal-panel lag regresson model are beer han ha of he radonal fxed-effec model. I proves ha can explan he model beer and show he acual suaon beer afer we nroduce he me and spaal fxed effecs no he model. he elasc coeffcens of GDP wh capal sock and labor are 0.56 and 0.30, respecvel. he sgnfcan level s %, ndcang ha he effec of nvesmen and labor on economc growh s sll he mos mporan facor, he elasc coeffcen of GDP on logscs s 0.08, he sgnfcan level s 0.64%. ndcang ha he logscs has a sgnfcan mpac on GDP, bu he degree of nfluence on economc growh s lmed, whch s showng he curren suaon of Chna..e. modernzaon of Chnese logscs s no hgh, he logscs nework ssem s mperfec, s lack of he applcaon of nformaon echnolog and he level of logscs managemen s low. hrdl, wh he esmae resuls of spaal-fxed nfluence parameer (η of he varous regon, he fxed-effec parameers of dfferen regons are showng sgnfcan dfference. he logscs 88

7 SHAO Yang/Canadan Socal Scence Vol.6 o.3, 00 developmen level s beer n he developed regons, worse n he developng regons. here are hree levels of logscs developmen n Chna. he bes one s developed coasal areas, such as Shangha c, Jangsu provnce, Zhejang provnce and oher developed coasal areas, second level s mddle level whch nclude cenral and norh-easern regon of Chna, such as Jangx provnce, Hunan provnce, Hube provnce, Laonng provnce, he hrd level s he wors one nclude norh-wesern regon of Chna. I s conssen wh he acual developmen suaon n varous regons of Chna. 4. COCLUSIO Based on spaal-panel model, we esmaed he correlaon beween he logscs and economc growh. We can draw he concluson ha here s sgnfcan spaal-correlaon beween GDP and logscs n varous regons, he GDP and logscs has obvous spaal overflow (dffuson benef beween adjacen regons. he logscs has a sgnfcan mpac on local GDP, bu he degree of nfluence on economc growh s lmed, he reasons are low degree of logscs modernzaon n Chna, he mperfec ssem of logscs nework, lack of applcaon of nformaon echnolog and he low level of logscs managemen. Fxed effec parameers of dfferen regons are showng sgnfcan dfferences. he local economc developmen s beer; he local logscs developmen level s hgher. REFERECES Sae Bureau of Surveng and Mappng: he Map of he Republc of Chna. (008.Chna:Sno Map Press. LI Xu-ng, GU Lan. (004. he Spaal Auoregressve Model and he Relevan Esmaon. Journal of Sascal Research, 6, HE Jangp, ZHAG Xn-zh. (006. Chnese Regonal Economc Growh and Convergence: Spaal Panel Daa Analss.Souh Chna Journal of Economcs, 5, Anseln, L.Spaal Economercs. (988. Mehods and Models. Dordrech:Kluwer Academc(Chaper 4. Anseln, L.. (988. Lagrange Mulpler es Dagnoscs for Spaal Dependence and Spaal Heerogene. Journal of Geographcal Analss, 0, -7. Balag,Bad H.Economerc. (00. Analss of Panel Daa. London: John Wle&Sons, Chcheser, Uned Kngdom. Elhors. J. (003. P-Specfcaon and Esmaon of Spaal Panel Daa Models. Journal of Inernaonal Regonal Scence Revew, 6,