A numerical simulation analysis of (Hukou) labour mobility restrictions in China

r.com/locate/econbase
Journal of Development Economics 83 (2007) 392–410
www.elsevie

A numerical simulation analysis of (Hukou) labourmobility restrictions in China☆

John Whalley a,b,c,⁎, Shunming Zhang d

a Department of Economics, University of Western Ontario, Canadab National Bureau of Economic Research, Inc. (NBER, Cambridge), United States

c Center for Economic Studies — Institute for Economic Research (CES-ifo, Munich), Germanyd School of Economics, Xiamen University, China

Received 2 March 2004; received in revised form 1 August 2006; accepted 21 August 2006

Abstract

We use numerical simulationmethods to analyze the Hukou system of permanent registration in China whichmany believe has supported growing relative inequality over the last 20 years by restraining labourmigration bothbetween the countryside and urban areas and between regions and cities. Our aim is to inject economic modellinginto the debate on sources of inequality in China which thus far has been largely statistical. We first use a modelwith homogeneous labour in which wage inequality across various geographical divides in China is supportedsolely by quantity basedmigration restrictions (urban–rural areas, rich–poor regions, eastern-central and western(non-coastal) zones, eastern and central-western development zones, eastern–central–western zones, moredisaggregated 6 regional classifications, and an all 31 provincial classification). We calibrate this model to basecase data and when we remove migration restrictions all wage and most income inequality disappears. Resultsfrom this model structure points to a significant role for Hukou restrictions in supporting inequality in China. Wethen present a furthermodel extension inwhich urban house price rises retard rural–urbanmigration. The impactsof removing Hukou restrictions onmigration are smaller, but are still significant. Finally, wemodify the model tocapture labour productivity differences across regions, calibrating the modified model to estimates of bothnational and regional Gini coefficients. Removal of migration barriers is again inequality improving but less so.© 2006 Elsevier B.V. All rights reserved.

JEL classification: J6; D5; R23Keywords: Efficiency gains; Migration; Distribution

☆ We thank Jim Davies, Stephen McGurk, Randy Spence and Terry Sicular and seminar group at the University ofWestern Ontario for their comments. We are grateful for support under IDRC-MIMAP programme. We also acknowledgehelpful comments from a seminar group MIMAP meeting in Manila (February 2003).⁎ Corresponding author. Department of Economics, University of Western Ontario, London, Ontario, Canada N6A 5C2.E-mail address: [email protected] (J. Whalley).

0304-3878/$ - see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.jdeveco.2006.08.003

mailto:[email protected]

http://dx.doi.org/10.1016/j.jdeveco.2006.08.003

393J. Whalley, S. Zhang / Journal of Development Economics 83 (2007) 392–410

1. Introduction

The statistical literature on inequality in China (Bramall and Jones, 1993; Chen, 1996; Hareand West, 1999; Jalan and Ravallion, 1998; Kanbur and Zhang, 1999; Lyons, 1991; Rozelle,1994; Tsui, 1991, 1993, 1996, 1998a,b) is widely interpreted as pointing to growing nationalrelative inequality in recent years. According to received wisdom based on some of this literature,the national Gini coefficient from China on income inequality has increased to over 0.4 from 0.3over the last 15 years or so. This change coexists with more slowly growing inequality within theurban and rural segments of the economy (as measured by Gini coefficients), and also withincoastal zones and inland segments of the economy. A sharp increase in the income/capita gapacross these divides is usually taken to account for increased national relative inequality. Absolutepoverty as measured by head count ratios and other measures consistently falls in Chinese datareflecting strong GDP growth in recent years.

A number of attempts have been made to account for this inequality change profile usingstatistical techniques as in the literature listed above. Econometric literature including Zhao (1999)and Shi, Sicular, and Zhao (2002) focuses mainly on the determinants of migration decisions, andnot on the size of the efficiency gains involved nor impacts on wage rates in a consistent microbased structure model. Our aim here is to use numerical simulation methods to provide freshinsights on these dimensions of the issue.

We focus on the system of Hukou in China, or registered permanent residence, which islocation specific. Not having Hukou in urban areas means that migrants receive no education orhealth benefits and cannot purchase housing, since title to it cannot be registered by them.Effectively, Hukou operates as a barrier to urban/rural migration in China and supports largeregional wage differentials which labour markets do not compete away. We ask how muchinequality there would have been in China without the Hukou system.

Our starting point is the literature on the global consequences of immigration restrictions (seeHamilton and Whalley, 1984). In this literature, differences in both wage rates and GDP/capitaacross countries are assumed to be supported by immigration restrictions in a world with countryspecific factor inputs and downward sloping marginal product of labour schedules for otherwisepotentially mobile labour. Parameters for an assumed underlying technology are calibrated so asto be consistent with observed data on wage differentials, labour shares of income, and GDP andpopulation by country, and counterfactuals are performed to analyze the impacts of immigrationbarrier removal. Assumptions that there is homogenous labour across countries or that there areefficiency differences across countries are used as alternatives in these exercises.

Herewemake calculations for China as towhat the impacts of removing internal (Hukou) barriersto regional labour mobility on inequality could be using data on both aggregate and regional GDP/capita using similar methods. In so doing we elaborate on the earlier methodology used to analyzeglobal migration restrictions by using a simple basic model which we sequentially modify in furthermodel elaborations. We first ignore inequality within regions and treat labour as homogenous bothacross regions and across individuals. In this model wage rates across regions are equalized whenmigration restrictions are moved. With region specific fixed factors, regional differences in GDP/capita do not disappear with barrier removal although they fall sharply and national inequality ismuch reduced. Significant efficiency gains also accrue from barrier removal in this model. We alsodiscuss the implications of relaxing the assumption of region specific immobile capital for results.

We then introduce region specific house price effects and capture their dampening impacts onmigration. This is motivated by the desire to also capture the impacts of increases in urban houseprices and housing rents in China on urban–rural mobility over 10 years. We develop a two good

394 J. Whalley, S. Zhang / Journal of Development Economics 83 (2007) 392–410

general equilibrium model with goods and housing, in which location specific housing stockssupport differing urban and rural house prices. In equilibrium migration equalizes the real value ofwage rates and we incorporate differences in house prices across regions through region specifictrue cost of living indices. Removing Hukou restrictions in this model again generates labourflows into urban areas which are now smaller, and significant redistribution occurs between urbandwellers whose house prices rise and rural dwellers whose house prices fall.

Finally, we extend the basic model by also assuming a distribution of labour productivities (orefficiency types) within each region and calibrate an extended model to both regional and nationalestimates of Gini coefficients before barrier removal, and also regional and national data on GDP/capita. Here, to simplify things, we assume that when mobility occurs across regions a represen-tative segment or slice of the distribution of productivities moves and adds result to the dis-tribution within the receiving region proportional to that distribution. We again remove migrationbarriers and calculate national inequality in the absence of Hukou restrictions. Significantefficiency impacts again occur but reductions in national inequality are smaller.

While findings from these exercises are data and parameter sensitive, they nonetheless jointlypoint to significiant impacts from Hukou system in supporting growing overall relative inequalityin China, and significant efficiency gains from its removal. They also show how economicmodelling as well as statistical techniques can be used to analyze inequality change from doingrefer in not only in China but also for the other countries.

2. Analysis of Hukou restriction removal using a basic model

The objective of our paper is to assess how inequality in China might behave if not for the Hukousystem of permanent registration acting as set of restrictions on labour mobility.We calibrate modelsof inter-regional labour mobility in China to base case data in the presence of Hukou restrictions andeliminate these restrictions.We then use themodel generated counterfactural equilibria to assess howmany people might migrate, and what the economy wide efficiency gains might be. In a later modelextension we calibrate both on national and regional Gini coefficients and calculate how thesechange when individuals differ with regions.

We first use a simple model in which each region produces a single good Y according to aregion specific production function in which labour enters as a homogeneous factor input which ismobile across all regions. We first assume region specific fixed factors (rents) which yielddecreasing returns to scale and so in the fully mobile labour case even if wage rates are equalizedacross regions, GDP per capita across regions will not be.

In this simplest version labour is the only factor of production which is mobile across regionsand capital is a fixed factor immobile across regions. We later relax this assumption and considermobile capital as well as labour. In this simplest model form the regional production function is

Ys ¼ fsðLsÞ; s ¼ 1; N ; S ð1Þwhere Ys is production in region s, Ls is labour used in region s, fs′N0 and fs b0. We assume that∑s Ls=L (full employment), and labour market clearing across regions determines the commonwage W=Ws (s=1,…,S ), where there are no barriers to labour mobility. Labour receives itsmarginal product in all regions in this case and

W ¼ fsVðLsÞ; s ¼ 1; N ; S: ð2ÞIf there are barriers to mobility of labour across regions, then the inter-regional labour

allocation no longer corresponds to that supporting a market clearing wage. Suppose this

Fig. 1. Gains from eliminating labour mobility restrictions in a 2 region model.


allocation is given as Ls0 (s=1,…,S) such that ∑s Ls

0 =L, then the marginal products of labouracross regions will differ and wage rates Ws (s=1,…,S ) will also differ, i.e. fs′(Ls

0) differs across s.Removing inter-regional barriers to labour mobility in this case implies an efficiency gain for theeconomy if marginal products of labour are equalized across regions.

Fig. 1 illustrates the case of a 2 region urban–rural economy. Here, initial migration restrictionssupport wage rate differences across regions while the common wage W=WU=WR applies whenmigration restrictions are removed. Income per capita across the two regions still differs with nomigration restrictions as region specific rents differ. Income inequality is much reduced but noteliminated in this case as wage converges across regions. An efficiency gain accrues, as shown, tothe whole economy.

If we assume a diminishing marginal product production function in each region of the form

Ys ¼ AsLass ; s ¼ 1; N ; S ð3Þ

then with decreasing marginal productivity of labour the regional wage Ws is given as

Ws ¼ AYsALs

¼ asAsLas−1s ; s ¼ 1; N ; S ð4Þ

and regional rents, Rs, are1

Rs ¼ Ys−WsLs; s ¼ 1; N ; S: ð5ÞIncome in region s, Is, is given by

Is ¼ Ys; s ¼ 1; N ; S: ð6Þand, assuming equal proportional shares in rents within the region, income per worker in eachregion, Js, is given by the average product of labour Js ¼ Is

Ls¼ Ys

Ls; s ¼ 1; N ; S:

In equilibriumXs

Ls ¼ L ð7Þ

where L is the national endowment of labour.
1 We assume here that labour migrates only in response to its regional wage, not also its proportional share in rents. The
model can be recast in this form and different numerical results will apply.


Typically, the size of the work force and population by region will differ. If the total nationalpopulation is N, and the population in region s is Ns, then N=∑s Ns, and the average income ineach region s, Īs, is given by

I s ¼ IsNs

¼ YsNs

¼ Rs

NsþWsLs

Ns; s ¼ 1; N ; S ð8Þ

National income I=∑sIs=∑s Ys, and national average income, Ī, is

I ¼ IN

¼P

s IsPs Ns

¼P

s YsPs Ns

: ð9Þ

The average wage per individual in region s is

W s ¼ WsLsNs

; s ¼ 1; N ; S ð10Þ

in which case I s ¼ Rs

Nsþ W s:

We can extend this model to cases where capital is mobile across regions. In the extreme casewhere all capital is mobile, specialization equilibria can easily occur under elimination of Hukoulabour mobility restrictions in which there is no production in some regions. Given that in realitycapital refers both to regionally immobile land, and other capital such as equipment and ma-chinery that is mobile in the intermediate term, we use a model extension in which there is bothregionally fixed (specific) capital and regionally mobile capital. If we model capital in region s aspart fixed and part mobile, then the product production function in each region is of the form

Ys ¼ AsKbss Lass ; s ¼ 1; N ; S ð11Þ

where Ks refers to mobile capital used in region s. Specific factors, including immobile capital,are represented by As, and βs+αsb1. In this case, the return of mobile capital will be the same inall regions and equal the marginal value product of capital in the region. This return B is given as

B ¼ AYsAKs

¼ bsAsKbs−1s Lass ; s ¼ 1; N ; S ð12Þ

and in equilibrium the market in mobile capital will clear, ∑Ss=1 Ks=K, where K is the economy

wide endowment of mobile capital.If there are no barriers to labour mobility, the common wage rate W is

Ws ¼ AYsALs

¼ asAsKbss Las−1s ; s ¼ 1; N ; S ð13Þ

and labour market clearing again determines the wage. As before, if there are barriers to mobilitythen the inter-regional labour allocation no longer corresponds to that supporting a marketclearing wage. Removing inter-regional barriers to labour again implies an efficiency for theeconomy as marginal products of labour are equalized across regions, but the size of the gain willdiffer from the fully capital immobile case. In calibrating such a model we set the total return toboth mobile and immobile capital equal to the same total return across the capital under anassumption of fully fixed capital. We then vary the assumed portions of mobile and immobilecapital in repeated calculations to explore sensitivity of model results.

Table 1Data used to calibrate 2, 3 and 6 region formulations of Chinese inter-regional labour mobility model for alternativeregional groupings (All data for 20011)

Regional classification2 used in model variants

Urban–rural Rich–poor EC–CW EC–WD E–C–W 6 regions3

GDP per capita by region(103 RMB)

NC 9.22246U 12.38835 R 13.29862 EC 12.08961 EC 9.77881 E 12.08961 NEC 9.93508

EC 11.03664C 6.36075 CSC 7.78470

SWC 4.71466R 6.05665 P 5.79754 CW 5.79994 WD 5.03428 W 4.94210 NWC 5.43888

Wage rate by region(103 RMB)

NC 9.61953U 16.63805 R 11.30238 EC 10.40758 EC 9.24304 E 10.40758 NEC 8.55412

EC 9.82212C 7.28020 CSC 8.01942

SWC 6.04259R 5.73975 P 6.71116 CW 6.77753 WD 5.99797 W 6.03203 NWC 6.00533

Work force by region(106 individuals)

NC 68.458U 142.236 R 218.465 EC 264.782 EC 447.666 E 264.782 NEC 45.216

EC 187.020C 218.451 CSC 182.539

SWC 105.539R 482.291 P 412.062 CW 365.745 WD 182.861 W 147.294 NWC 41.755

Population by region(106 individuals)

NC 147.35U 472.20 R 442.38 EC 527.10 EC 903.36 E 527.10 NEC 106.96

EC 365.77C 447.91 CSC 354.39

SWC 200.86R 795.63 P 825.45 CW 740.73 WD 364.47 W 292.82 NWC 91.96

1. All data are from the Chinese Statistical Yearbook (2002).2. The regional classifications are presented in more details in footnote 3 (page 7).3. NC, NEC, EC, CSC, SWC, and NWC denote Northern China (5 provinces), Northern China (3 provinces), EasternChina (7 provinces), Central and Southern China (6 provinces), Southwestern China (5 provinces), and NorthwesternChina (5 provinces), respectively.


If we assume S=2 we can group China in a number of ways which mirror regional clas-sifications of data from Chinese Statistical Yearbooks. There are: urban and rural; rich and poor,eastern coastal and central and western (non-coastal) zones; eastern and central and westerndevelopment zones. If S=3, we group by eastern, central and western zones. We also consider a6 region case of Northern China, Northeastern China, Eastern China, Central and Southern China,Southwestern China, and Northwestern China. There are 31 provinces, centrally administeredmunicipalities and autonomous regions in China in total2 and we also explore a 31 region variantof the model that captures all of these. Differences in results across these cases reflect the degreeand form of regional disaggregation used in alternative empirical applications of the model.

The models set out above with the two different treatments of capital mobility can be usednumerically to assess the distributional impacts of removing Hukou migration restrictions by firstcalibrating observed wage GDP/capita differences across the various two region divides set outabove for China and then recomputing a counterfactual equilibrium with removal of restrictions.

2 We treat all the provinces, centrally administered municipalities and autonomous regions in China as separate anddistinct entitles (provinces) in more detailed regional reported later.


To do this we use a benchmark data set for a given initial year in the presence of Hukourestrictions on labour mobility, and assess the implications of removing mobility restrictionsgiven the observed initial differences in GDP/capita and wage inequality. In the capital immobilecase, if Ys, Ls, and Ws are observed and given by data, we can use the model in calibration modeand solve the system of 2S Eqs. (3) and (4) for the 2S unknowns As and αs. We can then compute acounterfactual equilibrium for this simple model in which W=Ws (s=1,…,S ) and labour mobilityrestrictions are eliminated. Comparing the original data to the solution generated as thecounterfactual model equilibrium then yields as an evaluation of the impacts of Hukou on bothnational and regional income inequality in this simple case. A similar approach can be used inmore detailed multi-region calculations and with mobility of a portion of capital where the rate ofreturn across regions is equalized and there is full employment of capital.

In Table 1 we reported the 2001 base year data used in calibration for the 2, 3, and 6 regions.These are GDP/capita by region, where rate differentials across regions work force and populationby region. We assume the ratio of the work force to the population in any region is fixed and whena worker moves between regions, a corresponding multiple of the population also moves. Data forthe full 31 regions version is available on request. Using these data as an input, throughcalibration we determine the model parameters As and αs, as well as regional outputs Ys

0, rentsRs0, and other variables Is

0, Īs0, I 0, Ī 0, Js0, J 0, and Ws0. We assess the impacts of eliminating

Hukou restrictions on all endogenous model variable Ys, Ls, Ws, Rs, Is, Īs, I, Ī, Js, J, Ws, andsolve the model in counterfactual model for the case where wage rates are equalized acrossregions and thus determine the migration impacts on work force and population by region.

For the construction of the base year data we have used when calibrating alternative versions ofthe model we first rank provinces from rich to poor (on an income (GDP)/capita basis), taking thetop 10 provinces (approximately 35% of the population) as the rich group. The remainingprovinces make up the poor group. We next group provinces into two eastern coastal and centraland southern (non-coastal) provincial groupings, and two eastern and central and westerndevelopment zones. We also divide the 31 provinces into 3 zones and 6 regions. We finallyconsider all Chinese provinces in a 31 region version of the model. 3 Since data on Chinese GDP/capita (average income) and wage differentials are only readily available on a provincial basis, weuse data for all 31 provinces, centrally administered municipalities, and autonomous regions togroup regions in China in various ways. We consider seven different regional divides groupingdata on population, output, work force, and wage rate for each.

3 There are 31 provinces, centrally administered municipalities and autonomous regions in Chinese mainland. The 31provinces, centrally administered municipalities and autonomous regions ranked from rich to poor by GDP per capita areShanghai, Beijing, Tianjin, Zhengjiang, Guangdong, Jiangsu, Fujian, Liaoning, Shanong, Heilongjiang, Heilongjiang, Hebei,Xinjiang, Hubei, Jilin, Hainan, Inner mongolia, Hunan, Quinghai, Henan, Chongqing, Shanxi, Ningxia, Tibet, Anhui, Jiangxi,Sichuan, Shaanxi, Guangxa, Yunnan, Gansu, Guizhou. The first 10 provinces are grouped as rich. The 21 provinces whichfollow are group as poor. Mainland China is divided into 6 regions as follows; Northern China includes 5 provinces: Beijing,Tianjin, Hebei, Shanxi, Inner Mongolia; Northwestern China includes 3 provinces: Liaoning, Jilin, Heilongjiang; EasternChina in includes 7 provinces: Shanghai; Jiangsu, Zhejiang, Anhui, Fujian, Jingxi, Shandong; Central and Southern Chinaincludes 6 provinces: Henan, Hubei, Hunan, Guangdong, Guangxi, Hainan; Southwestern Chian includes 5 provinces:Chongqing, Sichuan, Guizhou, Yunnan, Tibet; Northwestern Chian inludes 5 provinces: Shaanxi, Gansu, Qinghai, Ningxia,Xinjiang. The Eastern Coastal zone includes 12 provinces: Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang,Fujian, Shandong, Guangdong, Guangxi, Hainan. The Central zone includes 9 provinces: Shanxi, inner Mongolia, Jilin,Heiongjiang, Anhui, Jiangxi, Henan, Hubei, Hunan. TheWestern zone includes 10 provinces: Chongqing, Sichuan, Guizhou,Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang. The Western Development zone includes 12 provinces as InnerMongolia, Guangxi, and western zone. In Tables 2 and 3 and in the discussion that follows, EC–CW zones denote EasternCoastal zone–Central andWestern zones, EC–WDzones denote Eastern andCentral zone–WesternDevelopment zone, E–C–W zones denote Eastern zone–Central zone–Western zone.

Table 2Impacts of Hukou elimination in alternative regional divides in fixed capital version of inter-regional labour mobilitymodel for China (Changes relative to base case data for 2001)

Urban–rural Rich–poor EC–CW EC–WD E–C–W 6 regions

Change in valueof output(106 RMB)

U 3682.041 R 1812.917 EC 1243.425 EC 884.753 E 1242.262 NC 209.063R −2087.854 P −1170.085 CW −1003.264 WD −701.386 C -412.046 NEC 21.482

W −565.689 EC 627.506CSC −91.040SWC −450.122NWC −137.214

N 1414.187 N 342.831 N 240.161 N 183.267 N 324.537 N 179.574Change in average

income per capita(103 RMB)

U −3.49215 R −4.88382 EC −2.55705 EC −0.98651 E −2.54224 NC −1.27485R 6.55073 P1.33713 CW 1.02269 WD 1.94103 C 0.82027 NEC −1.30699

W 1.79173 EC −1.72180CSC 0.26247SWC 1.75336NWC 2.06898

N1.11544 N0.27041 N0.18943 N0.14455 N0.20865 N0.1464Change in income

per worker(103 RMB)

U −18.72939 R −7.30444 EC −5.02016 EC −2.02573 E −5.01631 NC −2.80248R 5.22525 P 2.63985 CW 2.52885 WD 3.84133 C 1.71637 NEC −0.80406

W 3.59331 EC −3.42974CSC 0.45684SWC 0.29488NWC 4.50013

N2.24287 N0.54372 N0.38089 N0.29066 N 0.41955 N0.28480Change in wage

rate (103 RMB)U −7.89761 R −2.56094 EC −1.66614 EC −0.50160 E −1.66614 NC −0.87809R 1.00169 P 2.03028 CW 1.96391 WD 2.74347 C 1.46124 NEC 0.18732

W 2.70943 EC −1.08268CSC 0.72202SWC 2.69885NWC 2.73611

Change in workforce by region(migration millionsof individuals)

U 302.81913 R 158.40700 EC 135.07144 EC 101.17314 E 134.93051 NC 23.51675R −302.81913 P −158.40700CW -135.07144 WD −101.17314 C −53.32496 NEC 2.54381

W −81.60556 EC 69.89272CSC −11.18622SWC −65.03726NWC −19.72979

Change in populationby region(migration millionsof individuals)

U 579.01525 R 318.69952 EC 271.82831 EC 201.97491 E 270.46951 NC 49.94098R −579.01525 P −318.69952CW −271.82831 WD −201.97491C −108.54383 NEC 5.63122

W −161.92568 EC 134.97694CSC −22.88965SWC−124.04133NWC −43.61816


Table 2 reports results for numerical simulations showing the impacts of Hukou eliminationin model variants using alternative Chinese regional divides for the regional specific fixedcapital case. The changes reported imply that for the urban–rural 2 region case the per capitaincome differential falls from 2:1 to 7:10, while in case two (rich–poor) it falls from over 7:3 to4:3. In the urban–rural case, approximately 48% of the work force and 45% of the populationmove from rural to urban areas after Hukou removal. Around 17% of the population remains inrural areas. They become richer, their average income (GDP per capita) being 1.42 times higherthan that in urban areas. Total output increase by about 13%, and GDP per capita and income

Table 3National gain and migration under Hukou elimination for more detailed regional groupings in the model

2 region EC –CW model

2 region EC –WD model

2 region E–C–W model

6 regionsmodel

31provinces model

National gain from removal of Hukourestrictions as % of GDP

2.25110 1.71781 3.04198 1.68320 6.92509

Work force migration (million) 135.07144 101.17314 134.93051 134.93051 175.60454


per worker both increase. In rich–poor cases, there are smaller migration effects after Hukouremoval (approximately 25% of the work force and the population move), and total outputincreases only by about 3.2%. In the other 2 region cases (EC–CW and EC–WD) there aresmaller migration effects from Hukou removal, while total output and GDP per capita andincome per worker again increase. These smaller effects occur because there is an urban– ruraldivide within each region, and the main migration effect is urban–rural rather thangeographical. In the 3 region E–C–W case there are large migration effects out of Central-Western and Eastern zones after Hukou removal. Total output increases by about 3.0%. In the 6region case, migration effects between regions are small, and total output increases by about1.7%.

The picture that emerges from Table 2 is that Hukou registration is a significant policyimpediment preventing the achievement of a more equal distribution of incomes and wages inChina. The number of people who would migrate across regions after its removal in the first 4cases is 600 million people (1/2 of the population of China). Unlike the international migrationcase where Hamilton and Whalley (1984) found very large efficiency effects due to largerinitial wage dispersion, the efficiency effects gains are more modest but still significant. Thegain is 13% of national output in the first case, and considerably smaller at 1.7–32% in theother three geographical divide cases. In each case wage differentials across regional dividesare eliminated with the removal of Hukou restrictions, and average incomes across regions aremore closely equalized. The differences that remain average are due to differences in rentsacross regions.

Table 3 reports a more detailed 31 region calculation using the model. Compared to the smalldimensional geographical divides results showing gain of 1.7 to 3% of GDP, an increase inoutput of around 7% results, and differentials in both GDP/capita and income per worker acrossregions are sharply narrowed. The direction of migration is from poor to rich provinces, that is,from Henan, Guizhou, Sichuan, Anhui, Guangxi, Yunnan, Shaanxi, Gansu,Hunan, Chongqing,Jiangxi, Hebei, Shanxi, to Guangdong, Shanghai, Beijing, Zhejiang, Jiangsu, Fujian, Tianjin.The overall picture conveyed by these results is that removal of Hukou registration plays a moremajor role in reducing relative income inequality in China in a more disaggregated modelcalculation.

Table 4 reports results both for versions of the model incorporating differing degree of capitalmobility. As explained above, fully capital mobility is not consumed since in this casespecialization equilibrium can occur in the model. As the portion of capital assumed mobileincrease from 0 (same as Table 3) to 75%, the national gain increases and more labour movesbetween regions. Within 75% of capital mobile the gain rises to 24% GDP; some 70% larger thefixed capital case. These results occur because as labour leaves rural areas Hukou restrictions areremoved, this raises the marginal product of capital in the receiving region and capital also moves.This in turn raises further the marginal product of labour in receiving region and more labour

Table 4Impacts of Hukou elimination with partial capital mobility between regions in 2 Region (urban–rural divide) model

Portion of capital in each regionassumed mobile

National gain from removal of Hukourestriction as % of GDP

Urban–rural change work forcemigration (millions)

0.00 13.25557 302.819130.25 15.13038 343.566090.30 15.63167 353.919160.35 16.19088 365.147060.40 16.81720 377.285460.45 17.52169 390.331980.50 18.31692 404.216560.55 19.21732 418.752420.60 20.23816 433.538910.65 21.39473 448.031900.70 22.69865 461.177960.75 24.15291 471.76949


moves. Factor flows between regions in this model are thus complementary to one another, eventhough factors are substitutes in production. These results as a set thus indicate that estimates ofgains from elimination of Hukou restrictions while sensitive to assumptions on capital mobilitystill remain large, and that elimination of Hukou restrictions will have a significant and positiveeffect on national inequality.

3. Extending the model to capture house price effects on migration across regions

A key feature of the urban–rural divide in modern day China missing in the model variantsdiscussed above is large differences in housing (apartment) prices between the larger cities andvillage communities. Apartments in Beijing routinely sell for US $100,000, and with averageannual household incomes of around US $2000 in Beijing area, even at 5% the imputed incomeon apartments owned outright can be larger than annual cash income. Thus, households whoreceived apartments at low (or zero) prices in urban areas in the late 1980's or the early 1990'shave a large advantage over rural dwellers who might wish to move to the cities following aremoval of Hukou restrictions. In this case the effect of added inward migration will be to drive upurban house prices further and this house price effect will act to dampen additional migration.Less migration following Hukou elimination appears in models incorporating house price effectsin contrast to models which do not incorporate them; but working with them involves the use of amore complex analytical structure.

We have modified the basic capital immobile model set out in Section 2 to also incorporatehouse price effects stemming from additional urban–rural migration. We consider the 2 regionurban–rural divide. We capture impacts of urban house prices on urban–rural mobility decisionsby using a general equilibrium model with goods and housing separately identified, and withhousing prices in urban and rural differing reflecting market segmentation due to location. In thiscase, removing Hukou restrictions generates labour flows from rural to urban areas, but whenthese are included increases in urban house prices retard additional migration. Labour flows underHukou removal are smaller and significant further redistribution occurs between urban dwellerswhose house prices rise and rural dwellers whose house prices fall. Efficiency gains from Hukouremoval will tend to be smaller in models with house price effects since the number of migrantswill be smaller.


We can present this model variant more formally as follows. We once again denote Ls as thelabour in region s before Hukou removal. The regional production functions are again

Ys ¼ AsLass ; s ¼ U ;R ð14Þ

and the regional wage rates Ws, are given as

Ws ¼ PGAYsALs

¼ PGasAsLas−1s ; s ¼ U ;R ð15Þ

where PG is the goods price, and regional rents, Rs, are

Rs ¼ PGYs−WsLs; s ¼ U ;R: ð16ÞIncome in region s is Is

Is ¼ PGYs; s ¼ U ;R: ð17ÞNational income I=∑sIs. Full employment of labour in this model again implies thatX

s

Ls ¼ L ð18Þ

where L is the national endowment of labour.We assume that there is continuum of individuals who vary in terms of their preferences

toward goods and housing and who are uniformly distributed over a unidimensional interval [tR,tU] for a parameter t. t is a Cobb–Douglas preference function parameter varying across allindividuals. t is the critical value of the preference parameter such that urban individuals all have tvalues above t and lie on the interval [t, tU] and rural individuals all have t values below t and lieon the interval index [tR, t ]. The continuum of individuals [tR, tU] are assumed to differ inpreference shares parameters for Cobb–Douglas preferences defined as

VtðG;HÞ ¼ G1−tHt; ta½tR; tU� ð19Þwhere G is consumption of goods, and H is consumption of housing.

We assume for simplicity that all individuals in urban and rural areas have endowments ofhousing EU and ER that are similar in structure, but differ in size. Their incomes are

IUt ¼ RUXUt þWULUX

Ut þ PUEUX

Ut ¼ ðPGYU þ PUEUÞXU

t ; ta½ t; tU� ð20Þ

IRt ¼ RRXRt þWRLUX

Rt þ PRERX

Rt ¼ ðPGYR þ PRERÞX R

t ; ta½tR; t � ð21Þwhere XU and XR are uniform distribution random variable on [t, tU] and [tR, t ].

Utility maximization subject to a budget constraint in each case implies

GUt ¼ 1−t

PGIUt and HU

t ¼ t

PUIUt ; ta½ t; tU� ð22Þ

GRt ¼ 1−t

PGIRt and HR

t ¼ tPR

IRt ; ta½tR; t� ð23Þ


A general equilibrium in this model involves an equilibrium value tˆ, and the division of thepopulation into urban and rural components is endogenously determined. Before Hukou removalan equilibrium is characterized by wage rates Ws, a goods price PG, urban and rural house pricesPU and PR such that the following conditions hold.[1] (Good Market Clearing) YU þ YR ¼ R

ttUGU

t dt þR ttR

ˆGRt dt;

[2] (House Market Clearing)Rt

tUHUt dt ¼ EU and

R ttR

ˆHRt dt ¼ ER;

[3] (Labour Market Clearing)LU þ LR ¼ L;

If we consider the case of Hukou removal, under the assumption that all individuals haveidentical amounts of labour in urban and rural areas, then labour is allocated by region such that

LU ¼ tU− ttU−tR

L and LR ¼ t−tRtU−tR

L: ð24Þ

After Hukou removal we need to add a migration equilibrium condition involving a moneymetric measure of the relative valuation of a unit of income across the two regions TCLðU ; t Þ

TCLðR; tÞ , whereTCL refers to the true cost of living index. This term reflects the different price levels across thetwo regions due to different housing prices. As TCL indices enter when making migrationdecisions, individuals take into account not only wage rate differences across regions, but also thedifferent cost of housing.

To construct this metric we use the indirect utility functions

VtðIUt Þ ¼1−tPG

� �1−t tPU

� �t

IUt ; ta½ t; tU� ð25Þ

VtðIRt Þ ¼1−tPG

� �1−t t

PR

� �t

IRt ; ta½tR; t� ð26Þ

for which the TCL indices are

TCLðU ; tÞ ¼ 1−tPG

� �1−t tPU

� �t" #−1

; ta½ t; tU� ð27Þ

TCLðR; tÞ ¼ 1−tPG

� �1−t tPR

� �t" #−1

; ta½tR; t� ð28Þ

The equilibrium condition that migration must satisfy changes from the model with houseprice effects is

WU

WR¼ TCLðU ; tÞ

TCLðR; t Þ ¼PU

PR

� � t: ð29Þ

A general equilibrium after Hukou removal is thus given by the critical value of the shareparameter t, wage ratesWs, goods price PG, and urban and rural house prices PU and PR such thatthe following conditions hold.


[1] (Good Market Clearing) YU þ YR ¼ RttUGU

t dt þR ttR

ˆGRt dt;

[2] (House Market Clearing)RttUHU

t dt ¼ EU andR ttR

ˆHRt dt ¼ ER;

[3] (Labour Market Clearing) LU þ LR ¼ L;[4] (Migration Condition) WU

WR¼ PU

PR

h it:

The general equilibrium condition [2] can be written as

EU ¼Zt

tUHUt dt ¼

Zt

tU tPU

IUt dt ¼Zt

tU tPU

ðPGYU þ PUEUÞXUt dt

¼ PGYU þ PUEU

PU

Zt

tUtXU

t dtPUEU ¼ 12ðtU þ tÞðPGYU þ PUEUÞ and PUEU

¼12 ðtU þ tÞ1− 1

2 ðtU þ t ÞPGYU ð30Þ

and

ER ¼Z t

tR

ˆHR

t dt ¼Z t

tR

ˆ tPR

IRt dt ¼Z t

tR

ˆ tPR

ðPGYR þ PRERÞX Rt dt

¼ PGYR þ PRER

PR

Z t

tR

ˆtX R

t dtPRER ¼ 12ð t þ tRÞðPGYR þ PRERÞ and PRER

¼12 ð t þ tRÞ

1− 12 ð t þ tRÞ

PGYU ð31Þ

The general equilibrium condition [1] holds from Eqs. (30) and (31).We use this model to explore the dampening effect of house price rises on migration by

comparing the induced migration from Hukou elimination across the model variants. One of thesecaptures house price effects, and the second does not. The latter model is structurally the same asthe basic model in the preceding section, but is calibrated to different data which removes housingexpenditures from GDP (and hence is smaller than GDP data used in the previous section) tomake the data used in the with and without house price effect models comparable.

To implement these twomodels we again calibrate to base case data for 2001 and consider Hukouremoval. From Chinese Statistical Yearbook (2002) urban and rural non-housing GDP in 2001 are5849.777 million RMB and 4818.849 million RMB. If we assume the shares of housing/apartmentsin GDP in urban and rural areas to be 22.5% and 18.50%, 4 then the base case value of endowmentsof housing are 1316.200 and 891.487. The consumption value of non-housing goods in the urbanand rural areas is 4533.577 (=5849.777−1316.200) and 3927.362 (=4818.849−891.487). Urbanand rural labour are 148.236 and 482.291, and wage rates are 16.638 and 5.740. Rents are 2067.219(=4533.577−16.638×148.236) and 1159.132 (=3927.362−5.740×482.291).

For the model which ignores housing price effects, we use the same as in Section 2 to analyzethe effects of Hukou removal, but only focus on the non-housing portion of the economy which issmaller. Through calibration we compute the scale and share parameters in production asAU=298.8121 and AR=50.4371, αU=0.5440 and αR=0.7049. Equilibrium outcomes afterHukou removal are output: YU=9116.090 and YR=1251.721, work force: LU=535.296 and

4 See the discussion in Wang and Zuo (1999) which supports estimates of this size for the mid-1990's appealing tosurvey evidence from various sources.


LR=95.231, common wage rate: WU=WR=9.265. Labour migration under Hukou removal is535.296−148.236=482.291−95.231=387.060.

To evaluate the impacts of Hukou removal in the presence of house price effects, we use themodel above and compare model results without and with house price effects. We assumeequilibrium prices in the base case are PG=1, PU=1 PR=1. YU=4533.577 and YR=3927.362,EU=1316.200 and ER=891.487, LU=148.236 and LR=482.291, WU=16.638 and WR=5.740,RU=2067.219 and RR=1159.132, IU=4533.577 and IR=3927.362. Calibration yields the scaleand share parameters as AU=298.8121 and AR=50.4371, αU=0.5440 and αR=0.7049.

In this case if we assume tU=0.2344 and tR=0.1544, this implies a value of t=0.2156, whichis consistent with the proportion of the population in urban and rural areas in base case data andgiven by tU− t

tU−tR and t−tRtU−tR. This implies

RttU I tUt dt ¼ 5849:777 and

R ttR

ˆI tRt dt ¼ 4818:49. ThenRttUGU

t dt ¼ 4533:577 andRttUHU

t dt ¼ 1316:200,R ttR

ˆGRt dt ¼ 3927:362 and

R ttR

ˆHRt dt ¼ 891:487.

To analyze Hukou removal in this case, in the base case tU=0.2344 and tR=0.1544, this impliesa new equilibrium value of t=0.1760. The equilibrium prices are PG=1.000, PU=1.647 andPR=0.419. Other variables are YU=8394.802 and YR=1887.051, LU=460.035 and LR=170.492,WU=9.927 and WR=7.802, RU=3827.859 and RR=556.949, IU=8394.802 and IR=1887.051,RttU IUt dt ¼ 10; 562:420 and

R ttRˆ IRt dt ¼ 2260:536. Under Hukou elimination

RttUGU

t dt ¼ 8394:802and

RttUHU

t dt ¼ 2167:617,R ttRˆGR

t dt ¼ 1887:051 andR ttRˆHR

t dt ¼ 373:485. Labour migration underHukou removal in this case is=460.035−148.236=482.291−170.492=311.799 and efficiencygain from elimination of Hukou restrictions is=12,822.955−10,668.626=2154.329. Risinghouse prices thus serve to dampen migration relative to the 387.05 value from the comparable nohouse price rise model. The size of efficiency gains is also reduced.

4. A model with a distribution of productivities within regions

The model set out above can also be elaborated on to also incorporate within region efficiency(or productivity) differences across individuals and hence inequality within regions both beforeand after the elimination of Hukou. This also allows us to calibrate the model not only to wagedifferentials across regions, but also to urban and rural Gini coefficients. With these featurespresent, wage differentials across individuals will not disappear with elimination of Hukourestrictions as in the basic model presented above. We limit our discussion of this case to themodel from where capital is fully immobile across regions.

Table 5 estimates of urban, rural and national household income Gini coefficients from Chang(2002) which show how income inequality has changed China over the 20 or so years. This isrepresentative of data claimed to show the growing income divide in China (but see the earlierfootnote 1 on the implications of alternative data for recent year). At national level there isconsistent worsening of national inequality which is seemingly always more unequal than regionalinequality (as measured by the Gini coefficient). In our base case data we use Gini coefficients forurban–rural, rich–poor, EC–CW, EC–WD. E–C–W, and NC–NEC–ECCSC–CSC–SWCNWCfor 2001 reported in Chinese People Daily (November 4, 2003 and March 30, 2004) which areclose to Chang's (2002) estimates. These are displayed later in Table 6.

To incorporate efficiency differences across individuals within regions into the model we use afunctional form for the distribution of incomes within regions which through calibration implies adistribution of the efficiencies of labour across individuals within region. This approach allows usto specify initial differences in efficiencies across individuals in specific regions of economy so asto also calibrate a modified version of the simple model to within region inequality as measuredby Gini coefficients for each region, as well as to national Gini inequality measure of income

Table 5Regional and National Gini coefficients for household income in China from 1978 to 2000

Year Urban areas Rural areas National

1978 0.160 0.2121979 0.160 0.2411980 0.150 0.241 0.3301981 0.150 0.2321982 0.150 0.2461983 0.160 0.2441984 0.190 0.227 0.3001985 0.190 0.3041986 0.200 0.3051987 0.230 0.3031988 0.230 0.3101989 0.230 0.3101990 0.240 0.3071991 0.250 0.3131992 0.270 0.3291993 0.300 0.3211994 0.280 0.342 0.4001995 0.280 0.323 0.4151996 0.290 0.329 0.4241997 0.300 0.337 0.4251998 0.295 0.336 0.4561999 0.4572000 0.320 0.458

Source: Chang (2002).


inequality. In this model variant removing labour mobility restrictions once again equalizes wagerates per efficiency unit labour across regions, but with differing endowments of efficiency unitsof labour across individuals wage rate inequality will remain both across individuals and acrossregions when Hukou restrictions are removed. An elimination of Hukou restrictions will beequalizing, but compete equality of wage rates across individuals within regions will not beachieved as efficiency differences remain.

We assume that when mobility of labour occurs across regions, labour moves in the form of arepresentative portion or slice of the distribution of productivities. This departs the leaving region,and adding itself proportionally to the distribution of productivities in the receiving region. This isa treatment that simplifies the analysis, and leaves the within region distribution of productivitiesunchanged in both migrating and receiving regions as mobility occurs. The regional incomedistribution changes however, because of the presence of region specific rents.

The model we use for this case can be formally stated as follows. The same Eqs. (3) (4) (5) (6)and (7) characterize the production side of the model, but there are now also distributions ofincome within regions which reflect differences in the efficiency of labour within regions.Denoting Ns as the number of people in region s, and Ls as labour in efficiency units in region s(Ns≠Ls), we have average regional incomes Īs, national income I, average national income Ī, andthe average region wage Ws as in Eqs. (8) (9) and (10) above.

The functional form for the income distribution within each region s can take many forms. Weassume the non-linear form

Ins ¼ Cs þ Dsnþ Esnds; n ¼ 1; N ;Ns; s ¼ 1; N ; S: ð32Þ

Table 6Effects of Hukou elimination on regional and national Gini coefficients and Theil measures of inequality using a modelwith distribution of efficiencies within regions

Regional divide in model variant and data in column headings

1 Regional and national Gini coefficients

Urban–rural Rich–poor EC–CW EC–WD E–C–W

Gini coefficients before Hukou removalGU=0.3200 GR=0.4094 GEC=0.4119 GEC=0.4186 GE=0.4226GR=0.3500 GP=0.2030 GCW=0.2040 GWD=0.1600 GC=0.1440

GW=0.1600G=0.4600 G=0.4600 G=0.4600 G=0.4600 G=0.4600

Gini coefficients after Hukou removalGU=0.357188 GR=0.423638 GEC=0.397921 GEC=0.224439 GE=0.254828GR=0.368747 GP=0.169154 GCW=0.112343 GWD=0.181277 GC=0.189328

GW=0.113556G=0.370538 G=0.373878 G=0.347042 G=0.229139 G=0.259639

2 Theil measures of inequality

Urban–rural Rich–poor EC–CW EC–WD E–C–W

Theil measures before Hukou removalTU=0.171850 TR=0.291932 TEC=0.285837 TEC=0.173458 TE=0.122389TR=0.203112 TP=0.0788384 TCW=0.102791 TWD=−0.118314 TC=−0.075694

TW=−0.070750Tw1 =0.185971 Tw=0.1961614 Tw=0.212126 Tw=0.123277 Tw=0.043293

Tb1 =0.064300 Tb=0.084295 Tb=0.065886 Tb=0.035041 Tb=0.069722

T=0.250270 T=0.280437 T=0.278010 T=0.158318 T=0.113015

Theil measures after Hukou removalTU=0.224532 TR=0.315792 TEC=0.256137 TEC=0.096899 TE=0.136043TR=0.234890 TP=0.063729 TCW=0.025677 TWD=0.077606 TC=0.083320

TW=0.025021Tw1 =0.226873 Tw=0.233030 Tw=0.186570 Tw=0.094884 Tw=0.115340

Tb1 =0.009734 Tb=0.010959 Tb=0.010367 Tb=0.002850 Tb=0.010194

T=0.236607 T=0.243990 T=0.196937 T=0.097735 T=0.125534

Tw refers to the Theil measure for within region inequality, Tb to between region inequality.


where n is an index 1,…, Ns across the individual Ns in regions ranked from poor to rich, and Cs,Ds, Es and δs are parameters of the distribution functions. This provides sufficient free parametersto calibrate the model to both regional and national Gini coeffiecients through a generateddistributed of efficiencies. We could alternatively use a functional form from the class used indistributional literature such as a Pareto distribution, although the implied distribution ofefficiencies would not be Pareto. If a simple linear form were used there would not be sufficientfree parameters for calibration.

Using Eq. (32), we calibrate the model to satisfy 2S+1 conditions reflecting total income andGini coefficient constraints and in addition use the same calibration conditions for the simplemodel above. Since

Is ¼XNs

n¼1

Ins ¼XNs

n¼1

½Cs þ Dsnþ Esnds �; s ¼ 1; N ; S; ð33Þ


Gs ¼ 1

2N2s I s

XNs

n¼1

XNs

n V¼1

jIns −In Vs j ¼ 1−1

N2s I s

XNs

n¼1

XNs

n V¼1

minfIns ; In Vs g; s ¼ 1; N ; S ð34Þ

where Gs is the region s Gini coefficient in incomes 5.The national Gini coefficient, g, is given by

G ¼ 1

2N2 I

XSs¼1

XNs

n¼1

XNs

n V¼1

jIns −In Vs j þ 2Xsps V

XNs

n¼1

XNs V

n V¼1

jIns −In Vs Vj( )

¼ 1−1

N 2 I

XSs¼1

XNs

n¼1

XNs

n V¼1

minfIns −In Vs g þ 2Xsps V

XNs

n¼1

XNsV

n V¼1

minfIns −In Vs Vg( )

ð35Þ

One difficulty in implementing this model variant is that initial data on regional and nationalGini coefficients and estimates of regional and national GDP/capita can be mutually inconsistent.Where this occur, we have adopted a procedure of accepting data on regional and national GDP/capita, and accepting one or more of the Gini coefficient estimates for 2001 and setting remainingGini coefficients at bounds implied by mutual consistency requirements. In the main, these areclose to the available literature estimates, but do depart from these slightly. No other procedureshort of modifying other portions of input data seem to suggest itself.

For the urban–rural case above where S=2, IU=5849.777, IR=4818.847 and I=10668.626from Section 2. If the three Gini coefficients (drawing on Chang, 2002) are set as GU=0.3200,GR=0.3500 and G=0.4600, and these are used as inputs into calibration. This allows us tocalibrate the model for 8 (=4×2) unknown distribution function parameters: Cs, Ds, Es, and δs fors=U, R. If we then remove Hukou restrictions and compute a new model solution the threecomputed Gini coefficients reported on Table 5 are GU=0.3572, GR=0.3687 and G=0.3705.Compared to the case of the simple model, the impact on inequality of Hukou removal is smaller,but it is still significant and especially so for national inequality.

For the rich–poor case above S also equals 2, and IR=5883.043 and IP=4785.583,I=10668.626 (from Section 2). Drawing on Liu, Yao, and Zhang (2001) we set the three Ginicoefficients to be gR=0.4094, gP=0.2030, and g=0.4600. We again calibrate the model andcompute the 8(=4×2) unknown parameters: Cs, Ds, Es and δs for s=R, P, and remove Hukourestrictions computing a new model solution. The three Gini coefficients in this solution areGR=0.4236, GP=0.1692, and G=0.3739.

Impacts on inequality in this case are smaller compared to the simple model but still significant.Impacts on the three Gini coefficients differ from the urban–rural case because the initialdispersion in Gini coefficients is larger. For the eastern coastal–central and western case againS=2, IEC=6372.436, ICW=4296.190 and I=10,668.626 from Section 2. Appealing again to Liu,Yao, and Zhang (2001) we now set the three Gini coefficients as GEC=0.4119, GCW=0.2040 andG=0.4600. Using these, we can again calibrate the model and remove Hukou restrictions. Thethree new Gini coefficients GEC=0.3979, GCW=0.1123, and G=0.3470. For the eastern central–western development 2 region case, GEC=0.4186, GWD=0.1600 and G=0.4600. RemovingHukou restrictions gives the three Gini coefficients GEC=0.2244, GWD=0.1813, and G=0.2291.

We have consolidated these results into a single table (Table 6) which reports base case andcounterfactual Gini coefficients for this extended model applied to four alternative regionaldivides (urban–rural, rich–poor, and eastern coastal–central and western provinces). We also

5 See Sen (1973).


report summary statistics of both regional and national inequality based on the Theil measure (seeTheil, 1967) since this is a decomposable measure. Impacts of Hukou removal on inequality asrepresented by the Gini coefficients are significant. Using both Gini coefficients and Theilmeasures, national inequality is reduced, but inequality can either worsen or improve at a regionallevel. In departing region, inequality typically improves due to the equal distribution of. rents. Itcan however change the other way due to wage change which affects the relative size of wageincome and rents. Impacts on inequality are reduced relative to the simple model set out abovesince the dispersion of rents. Impacts on inequality are reduced relative to the simple model set outabove since the dispersion in efficiencies across individuals implies that some inequality in wagerates remains after Hukou removal.

We also report Theil measures for within and between region before and after Hukou removal.The national Theil measure is

T ¼ 1N

XNn¼1

InIlog

InI

ð36Þ

where Ts is the Theil measure in region s. It can also be written in additive form across measuresfor each region

T ¼XSs¼1

TsNs I sN I

þXSs¼1

Ns I sN I

logIsI

ð37Þ

The two terms in the right hand side of Eq. (37) are within and between Theil measures.The majority of inequality is within region inequality, and this increases as Hukou restrictions

are removed. Across region inequality falls sharply reducing national inequality. This is madeclear by the decomposition results reported in Table 6.

5. Conclusions and remarks

This paper presents numerical simulation results on the impacts of the Hukou system ofpermanent registration on income inequality and labour migration in China. Our aim is to usenumerical modelling to help assess the contribution of various policy and other factors in China ineither contributing to or retarding inequality in China. We use three model variants to developresults. The base model is taken from Hamilton andWhalley (1984), who evaluated the impacts ofcross country immigration restrictions on global inequality, and we examine four 2 region cases, a3 region case, a 6 region case, and 31 province case in which alternative groupings and divisions inChinese data are used. A second house model captures the effects of higher urban house prices inretarding rural labour movement in to urban areas. A final extension allows for productivitydifferences across individuals within each region in the model and explores income inequalityimpacts for five model cases.

All model results point towards a significant role for the Hukou system in preventingmovement of labour which moves the Chinese economy towards a more equal distribution ofincome. There effects are smaller in the second two model variants than in the first. We see allmodels as a simplification from a more complex reality, and so we do not aim to provide firmpoint estimates of impact. But the themes of results seem clear, and in addition we offer asimulation approach which can also be used for the analysis of mobility restrictions in othereconomies in ways different from existing econometric literature which focus on the determinantsof mobility more than impacts on overall equality and efficiency.


References

Bramall, Chris, Jones, Marion E., 1993. Rural income inequality in China since 1978. Journal of Peasant Studies 21,41–70.

Chang, Gene H., 2002. The Cause and Cure of China's Widening Income Disparity. Department of Economics, Universityof Toledo. http://unr.edu/homepage/elliottp/cer134chang.pdf.

Chen, Jian, 1996. Regional income inequality and economic growth in China. Journal of Comparative Economics 22,141–164.

China State Statistics Bureau, 2002. China Statistical Yearbook. China Statistical Press, Beijing.Hamilton, Bob, Whalley, John., 1984. Efficiency and distributional implications of global restrictions on labour mobility.

Journal of Development Economics 14, 61–75.Hare, Denise, West, Loraine A., 1999. Spatial patterns in China's rural industrial growth and prospects for alleviation of

regional income inequality. Journal of Comparative Economics 27, 475–497.Jalan, Jyotsna, Ravallion, Martin, 1998. Transient poverty in postreform rural China, 1952–1985. Journal of Comparative

Economics 26, 338–357.Kanbur, Ravi, Zhang, Xiaobo, 1999. Which regional inequality? The evolution of rural–urban and inland–coastal

inequality in China from 1983 to 1995. Journal of Comparative Economics 27, 686–701.Liu, Aying, Yao, Shujie, Zhang, Zongyi, 2001. Convergence of China's Regional Income: 1952–97. Chinese Economic

Review 12, 243–358.Lyons, Thomas P., 1991. Interprovincial disparities in China: output and consumption, 1952–1987. Economic

Development and Cultural Change 39, 471–506.Rozelle, Scott, 1994. Rural industrialization and increasing inequality: emerging patterns in China's reforming economy.

Journal of Comparative Economics 19, 362–391.Sen, Amartya, 1973. On Economic Inequality. Oxford University Press.Shi, Xinzheng, Sicular, Terry, Zhao, Yaohui, 2002. Analyzing urban–rural income inequality in China. Paper Presented at

the International Symposium on Equality and Social Justice in Transitional China, Beijing, July 11–12.Theil, 1967. Economics and Information Theory. North-Holland, Amsterdam.Tsui, Kai-yeun, 1991. China's regional inequality, 1952–1985. Journal of Comparative Economics 15, 1–21.Tsui, Kai-yuen, 1993. Decomposition of China's regional inequalities. Journal of Comparative Economics 17, 600–627.Tsui, Kai-yuen, 1996. Economic reform and interprovincial inequality in China. Journal of Development Economics 50,

353–368.Tsui, Kai-yuen, 1998a. Factor decomposition of Chinese rural income inequality: new methodology, empirical findings,

and policy implications. Journal of Comparative Economics 26, 502–528.Tsui, Kai-yuen, 1998b. Trends and inequalities of rural welfare in China: evidence from rural households in Guangdong

and Sichuan. Journal of Comparative Economics 26, 783–804.Wang, Feng, Zuo, Xuejin, 1999. Inside China's cities: institutional barriers and opportunities for urban migrants. American

Economic Review: AEA Papers and Proceedings 89 (2), 276–280.Zhao, Yaohui, 1999. Leaving the countryside: rural-to-urban migration decisions in China. American Economic Review

890, 281–286.

http://unr.edu/homepage/elliottp/cer134chang.pdf

Documents

A numerical simulation analysis of (Hukou) labour mobility restrictions in China