Why is Chinese Provincial Oupu Diverging? Joakim Weserlund, Universiy of Gohenburg David Edgeron, Lund Universiy Sonja Opper, Lund Universiy
Purpose of his paper. We re-examine he resul of Pedroni and Yao (2006) ha Chinese provincial oupu is diverging using more daa (952-2007) o increase power using economeric ess ha allow for dependence beween provinces examining he possibiliy of "convergence clubs" 2. We find he increased per-capia oupu inequaliies o be due o boh province specific dispariies and o dispariies beween clubs of provinces
A Convenional Divergence Tes y i = log real per capia oupu for province i a ime yij = yi yj = pairwise differences i =,,, j = i+,, A formal es of divergence is hus DIVERGECE H : 0 yij non-saionary for some i, j COVERGECE H y i j * : ij saionary,
Evans (998) suggesed using insead H : 0 yi y non-saionary for some i This is asympoically equivalen o he formal es if he uni roo process can be wrien y f = ρ( y f ) + e y = ρ y + e i i i ij ij ij where f is a single common sochasic rend. In his case we can wrie s i = + ρ is = + i s= y f e f u and f can be consisenly esimaed by y
Pedroni and Yao (2006) Use yearly daa from 952-997 Assume y i has one common facor and ha e i is crosssecionally independen Tes ρ = agains ρ < using he MW (Maddala and Wu) and IPS (Im, Pesaran and Shin) panel uni roo ess. They canno rejec he null hypohesis of divergence. This resul could, however, be due o lack of power, he presence of more han one common facor or crosssecional dependence.
An Alernaive Represenaion Assume he more general represenaion y = λ f + u y = ( λ λ ) f + u i i i ij i j ij where f is a vecor of common facors and λ i are vecors of facor loadings. For y ij o be saionary he following mus be saisfied (i) idiosyncraic componen u ij is saionary (ii) he common facor ( λ λ ) f is saionary i j f saionary, f coinegraed or λ i λ j = 0 i, j
Pesaran's (2007) es All 2 ( ) oupu pairs y ij are uni roo (ADF) esed Le R ij if he null is rejeced for yij a level = 0 if he null is no rejeced α, and R = 2 ( ) i= j= i+ R ij Denoe by δ = he proporion of pairs ha are saionary and le δ = lim δ
I can be shown ha lim ER ( ) = δ + ( δ ) α δ (and also α) T oe ha ( ) he null ER if δ = and ha ER ( ) H0 : δ = 0, all oupu pairs are diverging R can hus be used o es H 0 agains α under H: δ > 0, a posiive fracion of converging pairs Since he disribuion of R is unknown we use a sieve boosrap o calculae he p-value of he es
Esimaing he common facors If he uni roo null is acceped (p-value > α) for y ij hen we will wan o examine if he nonsaionariy is due o he common facors, he idiosyncraic componen or boh. The common facors can be esimaed using principal componens on he saionary relaion Δ y = λ Δ f +Δu ˆ λ and Δ fˆ i i i i fˆ ˆ = Δfs and u = y λ ˆˆ f s= 2 ˆi i i
Resuls for China Oupu Figure : Log per-capia oupu
Figure 2: Mean deviaions of log per-capia oupu
Measures of Convergence Δ = 2 ( ) i= j= i+ s y y 2 2 = ( i ) i= Mean = 2 ( ) y ij i= j= i+ y ij
Figure 3: Overall convergence measures
Table : Pair-wise divergence ess Pre-reform: 952-977; Pos-reform: 978-2007 A large proporion of he oupu gaps seem o be diverging (all p-values > α)
Common Componen Six principal componens are esimaed, which accoun for over 90% of he variance. Five or six of hese are esed o have uni roos using he Bai-g rank ess and he Johansen race es Facor : Inerior facor Facor 2: Urban growh facor Facor 3: Lagging markeizaion facor Facor 4 6: Inerior facors, medium/low markeizaion
Figure 4: Fracion of he oal variance explained by he common componen.
Figure 5: Marginal R 2 for he esimaed facors.
Idiosyncraic Componen: Diverging (see Table ) Robusness Deerminisic rends: Taking possible deerminisic rends ino accoun does no affec our resuls Srucural breaks: The mehod of oulier deecion indicaes five possible province-specific breaks, in addiion o he general reform break. Condiioning on hese breaks does no affec our conclusions
Our pair-wise panel approach indicaes ha he null of a uni roo can only be rejeced a he 5% level in abou 5% of cases, leading o he conclusion ha per-capia oupu is diverging for China as a whole. More imporanly, we find ha his resul canno be aribued merely o he presence of provincespecific dispariies, bu ha here are also clubs of provinces wih separae growh pahs ha hinder he convergence a he aggregae counry level. This has imporan implicaions, boh for policy making and empirical research.