Environmental Economy and Policy Research

UNIVERSITY OF CAMBRIDGE Department of Land Economy Envronmental Economy and Polcy Research Dscusson Paper Seres Land Tenure Arrangements and Rural-Urban Mgraton n Chna by Katrna Mullan, Paulne Grosjean and Andreas Kontoleon 2008 Number: 37.2008 1

Land Tenure Arrangements and Rural-Urban Mgraton n Chna Katrna Mullan Paulne Grosjean Andreas Kontoleon Correspondng Author Unversty of Cambrdge, Department of Land Economy, 19 Slver Street, Cambrdge CB3 9EP; Unted Kngdom; Tel: +44 1223337157 Fax:+44 1223 337130 E-mal: klm31@cam.ac.uk Department of Agrcultural and Resource Economcs, 210 Gannn Hall, Unversty of Calforna Berkeley, Berkeley, CA 94720-3310, USA. Tel: +1 510 207 2037 E-mal: paulnegrosjean@yahoo.fr Unversty of Cambrdge, Department of Land Economy, 19 Slver Street, Cambrdge CB3 9EP; Unted Kngdom; Tel: +44 1223 339773 Fax:+44 1223 337130 E-mal: ak219@cam.ac.uk Abstract Ths paper examnes the mpacts of the Chnese Household Responsblty System, whch governs rural land tenure, on rural-urban mgraton. Mgraton n Chna has tradtonally been lmted by the hukou system of household regstraton, under whch ndvduals who wsh to change ther place of resdence must gan approval from government authortes. Ths system s currently beng relaxed n an attempt to reduce nequaltes between rural and urban areas. However, mgraton wll not ncrease f addtonal constrants reman for potental mgrants. Usng a model of the relatonshp between land tenure arrangements and mgraton of household members, we examne whether those wth greater tenure securty and formal rental rghts for agrcultural or forest land are more lkely to partcpate n labour markets outsde the vllage. The fndng that greater tenure securty ncreases mgraton suggests that the current system of property rghts, n whch land s perodcally reallocated, acts as a constrant on mgraton. Ths strengthens the case for further tenure reform for agrcultural and forest land. JEL codes: J61, O15, P32 Keywords: Land tenure securty; land rental rghts, rural-urban mgraton, Chna. 2

1 Introducton Despte the vast scale of mgraton n Chna 1, there s evdence that constrants on mgraton reman, whch contrbute towards both rural-urban and regonal nequalty and reduce overall labour productvty at a natonal level. As an ndcaton of the extent of such dspartes, Yang and Ca (2003) estmate that whle the rato of non-agrcultural to agrcultural ncomes for a sample of 36 other countres s usually around 1.5, the same rato n Chna has vared between 2 and 3 snce the early 1980s. To descrbe the source of obstacles to mgraton, the lterature has manly focused on the recevng pont perspectve. The household regstraton (hukou) system has ndeed been descrbed as the major mpedment to mgraton, by preventng rural mgrants from accessng all the benefts assocated wth legal resdence n ctes. Whalley and Zhang (2007) argue that removng hukou restrctons would sgnfcantly ncrease mgraton flows and reduce regonal ncome dspartes 2. Another wdely descrbed obstacle to mgraton reles on the extent of labour market segmentaton and dscrmnaton aganst rural mgrants. For example, Meng and Zhang (2001) fnd sgnfcant evdence for occupatonal segregaton, some of whch may also be traced back to the hukou system (Lu and Song 2006). As a result of awareness of the mpacts that hukou-related constrants on labour moblty have on nequalty and economc effcency, there have been reforms amed at reducng these constrants 3. These reforms wll remove restrctons on mgraton, and should therefore ncrease flows of labour and reduce wage nequaltes over tme. However, as hukou constrants are eased, labour moblty wll not ncrease f other lmts on mgraton reman. Ths paper specfcally consders whether rural land tenure arrangements act as a futher constrant on mgraton. By focusng on such push factors at the emttng pont of rural mgraton, ths paper flls a gap n the lterature on mgraton, whch, as explaned above, has manly focused on pull factors 4. 1 Estmatons reported by the Natonal Bureau of Statstcs amount to 132 mllon rural workers n ctes n 2006 (Natonal Bureau of Statstcs, 2007, n Demurger et al., 2008) 2 The Gn coeffcent s predcted to fall from 0.46 to 0.37. 3 There were chefly two types of reform related to the hukou system. The formal award of permanent resdency rghts was made easer and non-hukou mgrants were enabled to access many publc servces from whch they were prevously excluded. 4 Wth notable exceptons, such as Zhao (1999), Yang (1997), who also focus on land tenure; and Uchda et al (2008) and Groom et al (2006), who examne how the SLCP may allevate lqudy constrants on mgraton. 3

Ths paper also contrbutes more wdely to the lterature on the mpact of land rghts on household decson makng. Chna has a dstnct system of land tenure for agrcultural and forest land, n whch vllage collectves offcally own the land, but ndvdual households have fxed term contracts to use the land for ther own producton actvtes. Over tme, these land use rghts have become closer n nature to prvate property, wth longer contracts, reduced frequency of land reallocatons, and ncreased opportunty to rent land to others. However, ths process s not complete, and there s ongong debate about whether ncreased development of land rghts n the drecton of fully prvate property should contnue. So far ths has manly focused on the mpact of rural land tenure on nvestment and productvty (e.g. Jacoby et al, 2002; Dennger and Jn 2003; Carter and Yao 1999). Jacoby et al (2002) estmate the socal benefts that would results from mproved nvestment n land due to greater tenure securty. They fnd that these benefts are lkely to be small. Brandt et al (2002) fnd that, whle statc neffcency from unequal margnal returns to land and labour due to the absence of rental markets amounts to around 8% of medan household ncome, ths neffcency s reduced by perodc reallocatons. These papers ndcate that the case for land tenure reform n Chna s not as strong as the theoretcal lterature would suggest. Gven ths, there s a queston of whether the lkely productvty gans are suffcent to justfy the transactons costs nvolved n nsttutonal reform, and the potental negatve outcomes for some households, gven that land consttutes an mportant safety net n rural Chna (Burgess, 2001). However, ths s based on neffcences resultng from varaton n returns to land, and dsncentves to nvest n land. Another contrbuton of ths paper s to contrbute to the ongong debate about tenure reform by sheddng lght on another way n whch land tenure may affect household welfare and economc productvty. Ths paper consders the mpact of the current systems of rural property rghts on household mgraton decsons. Followng Besley s (1995) descrpton of the postve lnk between land rghts and nvestment decsons, we can dentfy three arguments accordng to whch land management arrangements would nfluence mgraton decsons. Frstly, mgraton s assocated wth a rsk of expropraton. Indeed, mgraton entals a decrease n the household sze. Egaltaran, but also effcency consderatons (because of the drop n the land to labour rato), and nterest protecton (n order to fulfll quotas for example), would then nduce redstrbuton of some of the household land (Rozelle and L 1998). Secondly, mgraton and property rghts mght also be lnked through the credt market. The ablty to pledge or mortgage land mproves access to fnance and facltates fnancng of mgraton. However, under Chnese law, t s not possble to mortgage land, and ths second lnk s gnored n ths paper. Thrdly, mgraton s encouraged by the development of land exchange rghts. The ablty to rent land whle away reduces the opportunty cost of lost agrcultural labor. Ths paper llustrates the mpact of land tenure arrangements and rental rghts on farmers occupatonal choces and emprcally evaluates the mpacts of property rghts on 4

mgraton n rural Chna. We develop a theoretcal model n whch we show that movng away from exstng property rghts arrangements towards those charactersng a market economy has countervalng effects on mgraton. Ths s then tested econometrcally usng household survey data on mgraton decsons and property rghts to agrcultural and forest land. The rest of the paper s organzed as follows. We begn by revewng the lterature on the determnants of, and contrants on, mgraton n Chna, followed by a dscusson of the current land tenure arrangements relatng to agrcultural and forest land. Secton 3 develops a theoretcal model that explores the lnks between the development of rural land rghts and mgraton by members of farm households. Secton 4 estmates the predcted relatonshps emprcally, and Secton 5 concludes. 2 Drvers of mgraton n Chna In order to understand how changes n rural land tenure may affect the propensty of rural households to mgrate, we begn by revewng the exstng lterature on mgraton n Chna, ncludng patterns of mgraton, constrants on mgraton, and the fndngs of prevous studes on the determnants of mgraton at the household level. In the Harrs-Todaro model (Harrs and Todaro 1970), mgraton s drven by the relatve levels of rural ncome and expected urban ncome. Ths s n turn based on the dfference between rural and urban ncomes and the probablty of employment f an ndvdual mgrates to an urban area. Ths basc model can be adjusted wth other factors that affect the costs or benefts of mgraton such as transport costs or lvng costs n urban areas. It can also be modfed to take account of rsk averson among potental mgrants. Another way of vewng mgraton has been suggested by the New Economcs of Labour Mgraton (NELM) lterature (Stark 1978; Stark and Bloom 1985), whch treats mgraton as a household decson rather than an ndvdual one. Wthn ths framework, the household decson to send out a mgrant s made n the context of the allocaton of all household labour between farm, off-farm, and domestc actvtes. As well as makng predctons about the mpacts of mgraton, these models also provde hypotheses about the factors that are lkely to determne patterns of mgraton. Taylor et al (2003) apply ths model to the Chnese context, and fnd that, as n other countres, although the agrcultural ncome of the source household falls, ths s compensated for by remttances from mgrant members and per capta household ncomes are on average ncreased by 16-43%. They also fnd that mgraton of one or more household members relaxes credt and lqudty contrants for those remanng. The model n ths paper also treats mgraton as a household decson, comparng the returns to total household labour from on- and offfarm actvtes. 5

Although the mpact of land tenure on mgraton has receved lttle attenton n the nternatonal lterature, t has been rased as a potentally mportant ssue n Chna specfcally. Avalable statstcal evdence suggests that rural-urban mgrants n Chna are more lkely than those n other countres to mgrate temporarly rather than permantly, and then return to ther home vllage (Yang 1997). Ths s supported by Zhao (1999), who tests the permanent ncome hypothess usng household survey data, and fnds that households treat remttances from mgrants as temporary, and consume only a small proporton of the addtonal ncome. Yang (1997) explans temporary mgraton, as well as the growth of non-agrcultural rural employment, on the bass of the agrcultural land tenure system. He proposes that because households wll lose ther rghts to agrcultural land f they mgrate permanently, ths creates an addtonal cost to mgraton n the form of lost future ncome. Yang does not test ths hypothess emprcally, but we use household survey data to test a smlar model n the followng sectons of ths paper. In addton to the current land tenure arrangements, there are a number of other potental constrants on mgraton operatng at the orgn locaton n the Chnese context. As wth land tenure constrants, these may also become ncreasngly relevant as the hukou constrants n the destnaton locatons are eased. There are a number of studes that focus on how access to credt or suffcent lqudty may lmt mgraton out of rural areas, n relaton to the role of the Sloppng Land Converson Programme (SLCP) n allevatng those lqudty constrants (Groom et al. 2006; Uchda 2008). It has also been suggested that requrements to meet agrcultural producton quotas or taxes may prevent households from partcpatng n mgraton (Rozelle et al. 1999; Zhao 1999; Flesher and Yang 2006). As well as the theoretcal lterature on rural-urban mgraton, there are a number of emprcal papers that look at a wde range of potental determnants of mgraton n Chna. Some general patterns are that younger adults are more lkely to mgrate, men mgrate more frequently than women, and sngle people wthout dependents are more lkely to mgrate (Rozelle et al. 1999; Zhao 2005). It s also wdely found that those wth more agrcultural land are less lkely to mgrate (Zhao 1999; Zhao 2005). The mpact of ncome on mgraton s less clear, as poor farmers would be expected to have more to gan from mgraton, but may be less able to mgrate f t requres an ntal captal outlay. However, Rozelle et al (1999) fnd that farmers from poor vllages are more lkely to mgrate, and Zhao (1999) fnds that a hgher ntal cash holdng reduces the lkelhood of mgraton. Rozelle et al (1999) also fnd that transport costs do not sgnfcantly affect mgraton, whle the presence of a chan of prevous mgrants from the vllage has a postve effect. There are mxed results n the lterature on educaton (Zhao 2005), whch may be because formal educaton n rural areas s not a good proxy for the sklls requred by mgrant workers, or because the work carred out by mgrants s unsklled. Fnally, Gles and Mu (2007) fnd that due to a lack of socal securty provson n rural area, ll health of elderly parents reduces the probablty of mgraton for adult chldren. 6

The only study that has prevously tested the mpact of land tenure on mgraton n Chna s Rozelle et al (1999). The authors explore the mpacts on the probablty of mgraton of a set vllage level of nsttutons ncludng securty of property rghts and rghts to transfer land, although formal hypotheses about the relatonshps are not formulated. The other nsttutons are access to credt, and whether farmers are subject to crop producton quotas. The study fnds that producton quotas and tenure securty are not mportant, whle the ablty to rent land and access credt both have postve mpacts on mgraton. Ths work dffers from the analyss n ths paper because t estmates the determnants of mgraton at the vllage level, specfcally, the proporton of people leavng the vllage as mgrants. Furthermore, t mplctly treats mgraton as the decson of the ndvdual mgrant. Followng the NELM lterature, we consder mgraton to be a household decson, and drectly examne whether land tenure arrangements affect the allocaton of labour between farm and non-farm actvtes for all famly members. An addtonal dfference s that Rozelle et al (1999) only consder the mpacts of land tenure for agrcultural land, not forest land. To sum up, from the evdence dscussed so far, t appears that the hukou system has been an mportant constrant on mgraton. However, as that constrant appears to be gradually relaxed over tme, t s mportant to consder whether other constrants reman. One partcular ssue that has been rased, for example by Yang (1997), s that the system of land tenure n Chna may prevent some households from sendng members to work outsde the vllage because they rsk losng ther rghts to agrcultural land. However, although Rozelle et al (1999) have tested ths at the vllage level, t has not prevously been nvestgated emprcally at the household level or snce the hukou restrctons have started to be relaxed. 3 A theoretcal framework lnkng land tenure arrangements and mgraton decsons Ths secton dscusses the characterstcs of land tenure arrangements n Chna and develops a theoretcal model of the household mgraton decson n ths context. Ths wll provde the necessary analytcal gudelnes to formulate the emprcal work presented n Secton 4. 3.1. Rural land tenure n Chna Agrcultural land n Chna s subject to what may be descrbed as quas-prvate property rghts (Kung 2002). Between the 1950s and the late-1970s, land was collectvely owned and managed. However, after 1978, land was allocated to ndvdual households under the Household Responsblty System (HRS). Under ths system, land offcally remans under collectve ownershp, but s allocated among vllage households to cultvate as they choose. 7

The extent to whch these rghts approxmate prvate property rghts n practce has vared over tme and space. An mportant feature of the HRS s perodc reallocaton of land at the dscreton of the vllage leader to account for changes n populaton and the formaton of new households, whch lmts the securty of land tenure. In the late 1990s, the government drected that the length of tme over whch households have contracts to land should be ncreased from an ntal 15 years to 30 years, wth the objectve of mproved nvestment n land (Brandt et al. 2002). However, there s evdence that reallocaton of land has contnued to be wdespread even where such contracts exst. For example, Kung and Lu (1997) fnd that 70% of ther surveyed sample of 800 households had experenced reallocatons snce the 15 year contracts were ntroduced, and Dennger and Jn (2003) fnd that 80% of vllages experenced one or more reallocatons between 1983 and 1990. Even where 30 year contracts have been ntroduced, Schwarzwalder et al (2002) fnd that whle 60% of households have contracts, only 13% rule out land reallocaton n a survey of 17 provnces. The authors also fnd that 25% of these 30 year contracts explctly allow for land readjustments and the rest are unclear. Another aspect of the current system s the extent to whch rentng land s permtted. Intally t was forbdden n all cases, but ncreasngly t s accepted ether outrght or subject to the permsson of the vllage leader. However, the land rental market remans thn n most places. Accordng to Brandt et al (2002), n a 1995 survey only 3% of land was rented out. They suggest that ths s because reallocaton of land acts as a partal substtute for a rental market because land can be reallocated from those who would otherwse have rented out land to those who would have rented n land. They also argue that vllage leaders prefer reallocatons because they provde nfluence over households, whch s needed for the vllage leaders to meet quotas and collect taxes, and that for ths reason they do not encourage land rentals. Lu et al (1998) fnd smlarly low rates of land rental n a survey of four provnces. Fnally, although households have land use rghts, and n some cases have tenure securty and the ablty to rent land, t s stll not possble to buy and sell agrcultural land or convert t to alternatve uses. Land tenure arrangements relatng to collectve forest land n the south and southwest of Chna are smlar to those concernng agrcultural land, whle forest land n the northeast of Chna s almost entrely state-owned. The Household Responsblty System was appled to collectve forest land n the md-1980s, a few years after t was ntroduced n the agrcultural sector. Households were allocated famly plots of forest land or waste land on whch trees could be planted, as well as areas of forested responsblty mountan land. Both of these land types are held under contract, and ndvdual households have the rghts to manage the land for tmber. Any trees planted by the household belong to them, but trees planted prevously by the collectve do not, and the revenue from harvestng the latter s shared between the household and the collectve (Lu 2001). 8

The majorty of vllages also retan some forest land that s collectvely owned and managed, ether for tmber or fuelwood. Vllage households can usually use ths land for collectng non-wood forest products and grazng lvestock, whle the ncome from tmber or fuelwood goes to the collectve (Lu 2001). A varaton on ths ownershp arrangement, that have emerged n recent years, are shareholdng systems. These may be self-ntated f households choose to pool ther plots of forest land and manage them jontly, or appled to collectvely owned land wth all vllage households recevng dvdend-payng shares of the forest land (Song et al. 2004). The stuaton relatng to forest land rental s smlar to agrcultural land, n the sense that t s permtted, but not common. However, n some vllages, ndvdual households or prvate frms can obtan contracts to manage and harvest tmber on relatvely large areas of collectve forest land, usually for ether a fxed payment or a share of the earnngs. The next sub-secton present a theoretcal model of household mgraton decson n ths context of frst, a rsk of land expropraton and second, a constrant on land rentals. 3.2. Theoretcal model of household mgraton decson Consder a rural household that maxmzes ts total labor ncome by allocatng ts fxed labour resource L [0,1] between farm and off farm actvtes l. Let us assume for smplfcaton that off farm actvtes are only avalable through mgraton to urban areas and wage employment. To access mgraton, households have to pay a fxed cost F. The wage rate w s dscounted by the varable costs of mgraton and by the probablty of beng unemployed n urban areas (Harrs and Todaro 1970). The mgraton partcpaton constrant s therefore: wl F 0. The farm producton technology s gven by: f ( L l, n), where l s the amount of household labour allocated to mgraton, L l s the resultng farm labour and n s the land nput. Captal s gnored as an nput n ths smple model. It s assumed that f(.,.) s ncreasng n both ts arguments at a decreasng rate and that farm labour and land are complementary factors of producton so that the cross dervatve s postve. Denote the household land endowment. In the baselne model, land s consdered to be fxed and non transferable. Indeed, agrcultural land s collectvely owned and households receve use rghts only n rural Chna. Subleases, even when authorzed by the central government are rare (Zhao 1999). Land s thus treated as an exogenous and free factor so that: n = N. Ths assumpton s later relaxed n the model. 9

3.1 Expropraton rsk and mgraton Because of egaltaran land dstrbuton, effcency and, possbly, nterest protecton motvatons by vllage leaders, the reducton n household sze resultng from mgraton entals an expropraton rsk. To see how ths mght affect mgraton, let us consder a two perod decson makng framework. In the frst perod, the household decdes how much labor to allocate to mgraton. In the second perod, the household sze havng been reduced by the amount of mgrant labor, the household faces a rsk of expropraton, whch s assumed to ncrease lnearly wth the amount of labor allocated to mgraton. Let us also assume that the rsk of expropraton depends on the development of land rghts n the vllage, whch we denote by. A hgher value of R means property rghts more smlar to those of a market economy, so that the rsk of expropraton h( R) [0,1] s decreasng n R. The probablty of expropraton when the household supples l to mgraton s: lh(r). Ths probablty s comprsed between 0 and 1 snce: l, h( R) [0,1]². We stll consder that there s no possblty of land exchange, whle we do not consder dscount rates. The household decson problem s therefore: max l, n wl + pf ( L l, n) (1) s.t. n N[ 1 lh( R)] (2) wl F 0 (3) Because of the absence of a land exchange market, the household uses all land avalable and constrant (2) holds wth equalty. Let us frst consder the case where the mgraton partcpaton constrant (3) s slack at equlbrum so that the household supples a postve amount of labor to mgraton. The frst order condton wth respect to mgrant labor s then: * * * * w pf ( L l, N[1 l h( R)]) pf ( L l, N[1 l h( R)]) Nh( R) 0 (4) 1 2 = The thrd term of ths expresson shows how the expropraton rsk represents a tax on mgraton. How does mgraton vary wth the development of land rghts? Dfferentaton of (4) gves: 10

* dl dr Nh'( R)[ pf12 (.,.) l pf 2 (.,.) + pf 22 (.,.) Nlh( R)] = (5) pf (.,.) + Nh( R)[ pf (.,.) Nh( R) + pf (.,.) + pf (.,.)] 11 22 12 22 Here, the development of land rghts has two countervalng effects on mgraton 5. A decrease n the probablty of expropraton s akn to a reducton of the tax on mgraton, whch boosts mgraton ncentves. However, ths postve effect s counterbalanced by a negatve effect due to the complementarty between land and farm labor. A decrease n the probablty of expropraton means that the household wll keep more land, whch necesstates farm labor and decrease mgraton ncentves. Hence, n ths smple model, the development of land rghts assocated wth an ncrease n tenure securty does not have the straghtforward effect of boostng mgraton. Our result shows that a decrease n the probablty of expropraton does not necessarly ncrease mgraton ncentves. Regardng the mpact of land rghts on the decson to supply a postve amount of labor to mgraton, the partcpaton constrant (3) s ether relaxed or tghtened dependng on whether land tenure securty R mpacts l postvely or negatvely respectvely. 3.2 Land exchange rghts and mgraton We now gnore the expropraton rsk and consder nstead the possblty of land rentals. Land rentals are allowed wth a probablty s(r) whch s ncreasng n R, and the rental rate s normalsed to 1. The quantty of land that the households rents out s denoted by o o n, and the quantty of land that t rents n by n. Net land rentals are: e = n n. The household revenue maxmzaton problem becomes: max l, n, e wl + pf ( L l, n) + s( R) e (7) s.t. n + e = N (8) wl F 0 (9) Whch can be rewrtten as: max l, n wl + pf ( L l, n) + s( R)( N n) (9) s.t. wl F 0 (10) 5 The frst term of the numerator and the last two terms of the denomnator are postve, ndcatng that an ncrease n decreases mgraton ncentves, whle the other terms are negatve, ndcatng a postve mpact of on mgraton ncentves. 11

Consderng frst that the mgraton partcpaton constrant s slack at equlbrum, the frst order condtons wth respect to mgrant labor and cultvated land are, respectvely: ** w pf ( L l, n) 0 (11) 1 = ** 2 ( L l, n) s( R) = pf 0 (12) The margnal productvty of labor s equalzed to the real off farm wage, whle the margnal productvty of land s equalzed to the land rental rate. Now, how does mgraton vary wth a development of land rental rghts? Dfferentaton of equaton (12) gves: dn s' ( R) = (13) dr pf 22 (.,.) whch s negatve gven the concavty of the producton functon. Hence, the amount of land under cultvaton decreases wth an ncrease n R. As a result, more labor s freed from farmng and mgraton ncreases. Indeed, dfferentaton of (11) gves: ** dl pf12 (.,.) = dn pf11(.,.) (14) The last equaton s negatve whch mples that. a development of land rental rghts ncreases mgraton ncentves. Gong back to the mgraton partcpaton constrant, snce the amount of labour allocated to mgraton ncreases, the partcpaton constrant s relaxed by the development of rental rghts. To sum up, ths theoretcal model of household labor allocaton decson predcts that mproved land rental rghts have an unequvocal postve effect on mgraton ncentves; whle ncreased tenure securty has a countervalng effect on mgraton. Ths s because a lower rsk of expropraton both ncreases mgraton ncentves due to the reducton of an mplct tax on mgrant labor, and reduces mgraton ncentves because f expropraton does not occur, the household wll have more agrcultural or forest land to manage. 4 Emprcal Analyss of mpact of land tenure on mgraton 4.1 Data Ths paper makes use of data from two household surveys carred out by Unversty College London, Cambrdge Unversty, and Pekng Unversty. Both surveys nvolved face-to-face ntervews, wth questons on land tenure; household labour allocaton, ncludng mgraton out of the vllage by household members; and nformaton on the demographc and economc characterstcs of the household. The frst survey, whch 12

contaned questons about property rghts relatng to agrcultural land, covered 131 households from 21 vllages n Guzhou Provnce and 155 households from 23 vllages n Nngxa Provnce. The second survey, whch contaned questons about property rghts relatng to forest land, covered 285 households n Guzhou Provnce. In addton to the household level surveys, a separate questonnare was used to collect vllage level data from vllage leaders. 4.2 Econometrc Methods The survey data are used to estmate the mpacts of land tenure securty and land transfer rghts on mgraton n the sample households. We estmate the effects on whether any household members mgrate, and also on the number of household members mgratng. In order to estmate the mpact of property rghts on whether any household members mgrate, we use a probt model. Ths assumes a latent varable specfcaton n whch the household decdes whether any famly members should mgrate. Mgraton ( y = 1 ) occurs f the expected net utlty from mgratng rather than all members remanng wthn the vllage (y*), s postve. Otherwse, mgraton does not occur ( y = 0 ). The unobserved latent varable s y * = x' β + ε (18) and we observe y = 1 f y* > 0 (19) y = 0 f y* 0 where Pr( y* > 0 x) = Pr( ε > x' β x) = F( x' β ) = Φ( x' β ) (20) We use models for count data to estmate the effect of property rghts on the number of household members mgratng because the dependent varable entals non-negatve nteger values only. The most commonly used model n ths context s the Posson model (Cameron and Trved 1998): λ Yt e λ Pr( Y = y x) = Y! (21) X 'β where λ = e and the expected number of events for a gven perod s: X 'β E[ y x ] = Var[ y x ] = λ = e (22) so that E[ y x x ] = λβ 13

A drawback of the Posson model s that t assumes equalty of the mean and varance of the dstrbuton of y. If the mean and varance are not equal, an alternatve dstrbuton must be used. The most common alternatve s the Negatve Bnomal dstrbuton, as suggested by Cameron and Trved (1986). Ths generalses the Posson model by ntroducng an ndvdual, unobserved effect nto the condtonal mean (Greene 2003) so that: ln μ = x ' β + ε = ln λ + ln u (23) If we condton on x and u, the dstrbuton of y remans the Posson dstrbuton, λ u y e t ( λ ) f ( y x, u ) = (24) y! and the uncondtonal dstrbuton s: λ u y e t ( λ ) f ( y x ) = g( u ) du (25) 0 y! A gamma dstrbuton s normally used for u, so that: θ θ θu θ 1 g( u ) = e u (26) Γ( θ ) Ths gves an uncondtonal dstrbuton for y of: Γ( θ + y ) y θ λ f ( y x ) = r (1 r ), where r = (27) Γ( y + 1) Γ( θ ) λ + θ In the followng secton, we estmate both the Posson and Negatve Bnomal models. Comparng the two specfcatons provdes a test of overdsperson,.e. whether the varance s greater than the mean. Due to the large proporton of zero values n the sample (approxmately half of the surveyed households who were asked about ther rghts to agrcultural land, and 40% of those who were asked about ther rghts to forest land), we also consder whether the Zero-Inflated Posson (ZIP) or Hurdle models provde a better ft wth the data. Both of these types of models assume that the data are generated by two separate processes. The frst process determnes whether a non-zero observaton s possble, and the second determnes the count value of the observaton. In the context of household mgraton decsons, the frst process would determne whether there s a possblty of the household havng any mgrants. Ths mght depend on whether they are constraned n some way, for example f the household does not have suffcent lqud assets to afford the transport costs of sendng any members away from the vllage, or t mght be that they have no wsh to work off-farm, for example because all adult members are elderly. The second process would determne how many household members mgrate. We would 14

expect ths to be determned by the net benefts of each addtonal mgrant, takng account of the opportunty costs of labour lost from the farm, potental earnngs off the farm, and other monetary and non-monetary factors. In the Hurdle model (Mullahy 1986), once the ntal hurdle of decdng to partcpate n the mgrant labour force has been crossed, the number of mgrants s assumed to be strctly postve. In the ZIP model (Lambert 1992; Greene 1994) n contrast, the partcpaton process determnes whether the household may partcpate n the mgrant labour market, but even those who can partcpate, and wsh to partcpate, may have no mgrant household members durng the perod of tme covered by the sample. In both cases, the decson about whether any household members wll mgrate can be determned by a bnary probablty model such as the probt or logt model. The number of mgrants s then estmated usng a truncated Posson or negatve bnomal model n the case of the Hurdle specfcaton, or a Posson/negatve bnomal model n the case of the ZIP specfcaton. Versons of these models have been used n a wde range of applcatons. In the specfc context of mgraton decsons, Taylor et al (2003) use a Posson regresson to estmate the determnants of the number of mgrants sent out from a household n Chna, whle Brown and Leeves (2007) use a negatve bnomal regresson for a smlar model of household mgraton decsons n Fj and Tonga, as do Pandey et al (1989) n Inda. As dscussed above, these models assume the same processes determne the decsons of whether to send any mgrants, and how many mgrants to send. Knfu (2006) nstead uses a zero-nflated negatve bnomal regresson to estmate the determnants of nter-regonal labour mgraton by ndgenous Australans, whch allows for two separate decson processes, whle Tsega (2005) estmates the number of mgrants n a household usng a zero-nflated Posson model. Bohara and Kreg (1996) compare models that account for excess zeros n ths way wth those that do not, and fnd that the zero-nflated model has the best ft wth US mgraton data. 4.3 Results The mpacts of ncreased land tenure securty and mproved rghts to rent land on mgraton are estmated usng a Probt model for whether households have any mgrants, and a Posson model, and related varants, for the number of household members who mgrate. All models are estmated usng NLOGIT 3.0. We use data from the two surveys descrbed above to estmate the mpacts of rghts to agrcultural and forest land. In both cases, tenure securty s measured by whether households thnk t s lkely that ther land wll be reallocated, and rental rghts are measured by whether household land can be rented out wthout authorsaton from the vllage leader. Other covarates ncluded n the models are based on the prevous studes of mgraton dscussed n Secton 2. These nclude the number of chldren and elderly 15

people n the household; the level of educaton of the household head; the amount of agrcultural or forest land that the household has rghts to; household assets, proxed by whether the household has a telephone; and the remoteness of the vllage, as measured by the dstance to the man townshp. Ideally the off-farm wages of the mgrants would be ncluded n the model. However, many households do not provde nformaton on offfarm wages, and poorly functonng rural labour markets make the estmaton of a predcted wage equaton unrelable. We therefore use the educton as a household level proxy for wages 6. Ths s supplemented wth the ncluson of county level dummy varables to account for local varaton n wage rates. Other covarates such as the age of the household head, the ethnc background of the famly, and alternatve measures of household assets were ncluded n alternatve models, but dd not mprove the ft wth the data (see Table 1) [INSERT TABLE 1 ABOUT HERE] Lookng at the two samples n Table 1, we can see that the frequency of mgraton s smlar for the two groups, but forest households have on average more household members workng outsde the vllage. There s a notable dfference n the perceved securty of rghts to agrcultural and forest land, wth around ¼ of the agrcultural sample belevng that reallocaton wll not occur, compared wth over ¾ of those asked about ther forest land. Rghts to rent out land are more smlar, wth a large majorty n each group statng that they are permtted to rent out land wthout authorsaton. As far as the other explanatory varables are concerned, agrcultural households have more chldren and fewer elderly members than forest households, and lower ncdence of telephone ownershp. Levels of educaton and dstances to townshps are smlar. Fnally, households have an average of 2.54 mu of agrcultural land, relatve to 24.9 mu of forest land, per adult 7. Table 2 presents the results of the probt estmaton of the mpacts of property rghts over agrcultural and forest land on whether any members of the household choose to mgrate for off-farm employment. [INSERT TABLE 1 ABOUT HERE] 6 For the same reasons, a smlar approach s employed and found to be robust n other studes of the rural labour market n Chna (e.g. Zhang et al 2001; Lohmar et al., 2000). 7 1 mu = 1/15 hectares 16

In the model relatng to agrcultural land, nether the securty of land tenure or the rghts to rent land affect whether households have any mgrants or not. The varables that are mportant determnants of mgraton nclude the number of chldren, whch reduces the lkelhood of any mgraton; and the land to labour rato, whch s also negatvely related to the probablty of mgraton. Most of the county level dummy varables are also sgnfcant, ndcatng regonal varaton n the probablty of mgraton. As Hezhang, Zhjn and Wenng countes are n Guzhou Provnce, whle the other three countes are n Nngxa Provnce, the results also suggest that mgraton s more lkely among households n Nngxa than those n Guzhou. The perceved securty of rghts to forest land s found to have a postve and sgnfcant effect on the probablty of mgraton by household members, although rental rghts to forest land are agan not sgnfcant. As wth the agrcultural land sample, a greater number of chldren n the household reduces the probablty of mgraton, and n ths case, a hgher number of elderly people also reduces the probablty of mgraton. We compare alternatve specfcaton for modellng the mpact of property rghts on the number of household members mgratng to work outsde the vllage. Table 3 shows the results of comparng the standard Posson model wth the alternatve Negatve Bnomal dstrbuton, and wth the ZIP and Hurdle models. In the case of the agrcultural land model, the dsperson parameter n the Negatve Bnomal model was nsgnfcant, suggestng the Posson model dd not suffer from overdsperson. In addton, the Vuong test for comparng the non-nested ZIP model wth the Posson model rejected the ZIP model, and the LM statstc for the Hurdle model vs the Posson model was also nsgnfcant. The concluson s therefore that the agrcultural land model should be estmated usng a standard Posson model. Comparng the alternatve models for the mpact of forest land rghts on mgraton, we observe that there s evdence of overdsperson, whch supports the use of the Negatve Bnomal rather than the Posson dstrbuton. The Vuong test of the zero-nflated model vs. the Negatve Bnomal model supports the use of the zero-nflated Negatve Bnomal model. [INSERT TABLE 3 ABOUT HERE] Table 4 contans the best-ft models for the determnants of the number of household members who choose to mgrate. In the case of the agrcultural households, ths s the standard Posson model, and n the case of the forest households, ths s the Zero-nflated Negatve Bnomal model. [INSERT TABLE 4 ABOUT HERE] 17

These results show that the perceved securty of property rghts to both agrcultural and forest land has a postve mpact on the number of household members workng outsde the vllage. The theoretcal model gave ambguous predctons regardng ths effect, but the results ndcate that the effect of secure land tenure on reducng the rsk of expropraton f household members leave outweghs the effect on household demand for labour f land s not exproprated. The theoretcal model predcted a postve mpact of rental rghts on mgraton, but these results show that n practce n ths sample, whether or not land can be rented out does not affect the number of mgrants. The other factors determnng the numbers of household members mgratng are farly smlar to those determnng whether any mgraton occurs. The number of chldren n the household has a negatve effect on the numbers of mgrants, n both agrcultural and forest households, and the area of agrcultural land s negatvely related to the number of household members mgratng, although the area of forest land does not have a sgnfcant effect. The wealth or assets of the household dd not have a sgnfcant effect on whether mgraton occurred, but we fnd that n both the forest and agrcultural models there s a postve relatonshp between assets and the number of household members mgratng. In the model of forest household decson makng, ncreasng dstance from the man townshp results n fewer households partcpatng n the mgrant labour force. Fnally, a number of the county level dummes sgnfcantly affect the mgraton decson, ndcatng geographcal varaton n the numbers mgratng as well as the probablty of any mgraton. 5 Conclusons Ths paper has looked at whether rural land tenure arrangements act as constrants on rural-urban mgraton n Chna. Although there has been lttle work elsewhere on the relatonshp between land tenure and mgraton, the unusual structure of property rghts for agrcultural and forest land has led others to suggest that these mght affect labour moblty n Chna (Yang 1997; Zhao 1999). Ths s an mportant tme to consder possble constrants on mgraton because of the relaxaton of prevously very strct lmtatons on where people could move to fnd employment. The ntenton behnd the easng of the hukou regulatons s to reduce dspartes between rural and urban areas, and allow contnued growth of the ndustral sector. However, the removal of restrctons at the destnaton locaton wll only ncrease rates of mgraton f households are not constraned by other factors at the orgnal locaton. These could nclude a lack of access to funds for the ntal stages of the mgraton process, or lack of nformaton about employment opportuntes. Ths paper has focused specfcally on whether tenure nsecurty or lmts on rentng land act as constrants on mgraton. 18

A smple theoretcal model of household labour allocaton between work on the household s agrcultural or forest land and wage employment outsde the vllage predcted that secure land tenure would have an ndetermnate mpact on mgraton. Ths s because t both ncreases the ncentve to mgrate due to the lower rsk of land expropraton, and reduces the ncentve to mgrate because f expropraton does not occur, the household wll have more land to manage. The theoretcal mpact of mproved land transfer rghts was unambguously postve. Testng ths emprcally, n relaton to both forest and agrcultural land for a sample of households n Guzhou and Nngxa Provnces, we found that n practce, greater tenure securty tends to ncrease mgraton, whle restrctons on land rentals dd not have a sgnfcant mpact. The second of these fndngs, relatng to land rentals, s somewhat surprsng because not only dd the labour allocaton model predct a postve mpact, but other studes have also made smlar predctons (Yang 1997; Zhao 1999). However, t may be because property rghts are already evolvng rapdly that ths aspect was not sgnfcant. A large majorty of households n the sample sad that they already had the rght to rent out agrcultural (87%) and forest (73%) land, suggestng that whle ths may have prevously been a constrant (and may stll be n other locatons), t s no longer for these households. The fndng that ncreased land tenure securty has a postve effect on mgraton leads to the mportant concluson that where land s at rsk of reallocaton, rural households may not allocate labour to mgraton to the extent that they otherwse would. Ths type of constrant on household decson makng has the drect effect of reducng household welfare by lmtng ther optons for utlty maxmsaton. In addton, t s lkely to ncrease rural-urban nequalty, and may potentally slow rates of economc growth at the natonal level. Ths s an addtonal reason, beyond the standard arguments relatng to nvestment ncentves, for strengthenng the securty of household rghts to land. 19

6 References Besley, T. (1995). "Property Rghts and Investment Incentves: Theory and Evdence from Ghana." The journal of poltcal economy 103(5): 35. Bohara, A. K. and R. G. Kreg (1996). "A zero-nflated Posson model of mgraton frequency." INTERNATIONAL REGIONAL SCIENCE REVIEW 19(3): 211-22. Brandt, L., J. Huang, G. L and S. Rozelle (2002). "Land rghts n rural Chna: Facts, fctons and ssues." The Chna Quarterly 47(1): 67-97. Brown, R. and G. Leeves (2007). "Impacts of Internatonal Mgraton and Remttances on Source Country Household Incomes n Small Island States: Fj and Tonga." FAO-ESA Workng Paper: 07-13. Burgess, Robn (2001). "Land and Welfare, Theory and Evdence from Chna". Workng paper, London School of Economcs Cameron, A. C. and P. K. Trved (1986). "Econometrc Models Based on Count Data: Comparsons and Applcatons of Some Estmators." Journal of Appled Econometrcs 1: 29 53. Cameron, A. C. and P. K. Trved (1998). Regresson Analyss of Count Data, Cambrdge Unversty Press. Carter, M. R. and Y. Yao (1999). Specalzaton wthout Regret: Transfer Rghts, Agrcultural Productvty, and Investment n an Industralzng Economy. Polcy Research Workng Paper Seres, World Bank. Dennger, K. and S. Jn (2003). "The Impact of Property Rghts on Households' Investment, Rsk Copng, and Polcy Preferences: Evdence from Chna*." Economc Development and Cultural Change 51(4): 851-882. Demsetz, H. (1967). "Toward a Theory of Property Rghts. 57." Amercan Economc Revew Papers and Proceedngs 347: 166. Démurger, Sylve & Marc Gurgand & L Sh & Yue Xmng (2008)."Mgrants as secondclass workers n urban Chna? A decomposton analyss," PSE Workng Papers 2008-20, PSE (Ecole normale supéreure). Dong, X. Y. (1996). "Two-ter land tenure system and sustaned economc growth n post-1978 rural Chna." World Development 24(5): 915-928. Flesher, B. M. and D. T. Yang (2006). "Problems of Chna's Rural Labor Markets and Rural-Urban Mgraton." Chnese Economy 39(3): 6-25. 20

Gles, J. and R. E. N. Mu (2007). "Elderly parent health and the mgraton decsons of adult chldren: Evdence from rural Chna." Demography 44(2): 265-288. Greene, W. H. (1994). Accountng for Excess Zeros and Sample Selecton n Posson and Negatve Bnomal Regresson Models, New York Unversty, Stern Busness School. Greene, W. H. (2003). Econometrc analyss. London, Prentce-Hall Internatonal (UK). Groom, B., P. Grosjean, A. Kontoleon, T. Swanson and S. Zhang (2006). Relaxng Rural Constrants: a Wn-Wn Polcy for Poverty and Envronment n Chna? Unversty of Cambrdge, Centre for Law & Economcs for Envronment and Development Workng Paper. Harrs, J. R. and M. P. Todaro (1970). "Mgraton, Unemployment and Development: A Two-Sector Analyss." Amercan Economc Revew 60(1): 126-142. Holden, S. and H. Yohannes (2002). "Land Redstrbuton, Tenure Insecurty, and Intensty of Producton: A Study of Farm Households n Southern Ethopa." Land Economcs 78(4): 573-590. Jacoby, H. G., G. L and S. Rozelle (2002). "Hazards of Expropraton: Tenure Insecurty and Investment n Rural Chna." Amercan Economc Revew 92(5): 1420-1447. Knfu, Y. (2006). Spatal moblty among Indgenous Australans: patterns and determnants, Australan Natonal Unversty, Workng Papers n Demography, No. 97. Kung, J. K. (2002). "Choce of Land Tenure n Chna: The Case of a County wth Quas- Prvate Property Rghts*." Economc Development and Cultural Change 50(4): 793-817. Kung, J. K. and S. Lu (1997). "Farmers Preferences Regardng Ownershp and Land Tenure n Post-Mao Chna: Unexpected Evdence from Eght Countes." The Chna Journal 38(2): 33-63. Lambert, D. (1992). "Zero-nflated Posson regresson, wth an applcaton to defects n manufacturng." Technometrcs 34(1): 1-14. Lohmar, Bryan, Scott Rozelle and Changbao Zhao. 2000. The Rse of Rural Labour Markets n Chna. Workng paper No. 1029, Department of Agrcultural and Resource Economcs, UC Davs. Lu, D. (2001). "Tenure and Management of Non-State Forests n Chna snce 1950: A Hstorcal Revew." Envronmental hstory 6(2): 239-263. Lu, S., M. R. Carter and Y. Yao (1998). "Dmensons and dversty of property rghts n rural Chna: Dlemmas on the road to further reform." World Development 26(10): 1789-1806. 21

Lu, Z. and S. Song (2006). "Rural urban mgraton and wage determnaton: The case of Tanjn, Chna." Chna Economc Revew 17(3): 337-345. Meng, X. and J. Zhang (2001). "The Two-Ter Labor Market n Urban Chna Occupatonal Segregaton and Wage Dfferentals between Urban Resdents and Rural Mgrants n Shangha." Journal of Comparatve Economcs 29(3): 485-504. Mullahy, J. (1986). "Specfcaton and testng of some modfed count data models." Journal of econometrcs 33: 341-365. Pandey, A., H. L. Sharma and K. K. Sngh (1989). Compoundng negatve bnomal and Posson dstrbuton to descrbe rural-urban out-mgraton at household level. Populaton transton n Inda. S. N. Sngh, M. K. Prem, P. S. Bhata and A. Bose. Delh, Inda, B.R. Publshng, B.R. Publshng. Vol. 2. Rozelle, S., L. Guo, M. Shen, A. Hughart and J. Gles (1999). "Leavng Chna's Farms: Survey Results of New Paths and Remanng Hurdles to Rural Mgraton." The Chna Quarterly(158): 367-393. Rozelle, S. and G. L (1998). "Vllage Leaders and Land-Rghts Formaton n Chna." The Amercan economc revew 88(2): 6. Schwarzwalder, B., L. Png, Z. B. S. Yufang and Z. Lchang (2002). Tenure and Management Arrangements for Chna's Forestland Resources: Feldwork Fndngs and Legal and Polcy Recommendatons. CCICED Western Chna Forest Grasslands Taskforce. Song, Y., G. Wang, W. R. Burch and M. A. Rechln (2004). "From nnovaton to adaptaton: lessons from 20 years of the SHIFT forest management system n Sanmng, Chna." Forest Ecology and Management 191(1-3): 225-238. Stark, O. (1978). Economc-Demographc Interactons n Agrcultural Development: The Case of Rural-to-Urban Mgraton, Food and Agrculture Organzaton of the Unted Natons (FAO), Rome. Stark, O. and D. E. Bloom (1985). "The New Economcs of Labor Mgraton." Amercan Economc Revew 75(2): 173-178. Taylor, J. E., S. Rozelle and A. de Brauw (2003). "Mgraton and Incomes n Source Communtes: A New Economcs of Mgraton Perspectve from Chna*." Economc Development and Cultural Change 52(1): 75-101. Tsega, D. (2005). The Economcs of Mgraton n the Volta Basn of Ghana Household and Dstrct-level Analyss, Cuvller Verlag. Uchda, E. (2008). "Conservaton Payments, Lqudty Constrants and Off-Farm Labor: Impact of the Gran for Green Program on Rural Households n Chna." Amercan Journal of Agrcultural Economcs (forthcomng). 22