Land Allocaton n Vetnam s Agraran Transton Martn Ravallon Development Research Group World Bank, 1818 H Street, N.W. Washngton, DC 20433 mravallon@worldbank.org Domnque van de Walle Development Research Group World Bank, 1818 H Street, N.W. Washngton, DC 20433 dvandewalle@worldbank.org World Bank Polcy Research Workng Paper 2951, January 2003 The Polcy Research Workng Paper Seres dssemnates the fndngs of work n progress to encourage the exchange of deas about development ssues. An objectve of the seres s to get the fndngs out quckly, even f the presentatons are less than fully polshed. The papers carry the names of the authors and should be cted accordngly. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors. They do not necessarly represent the vew of the World Bank, ts Executve Drectors, or the countres they represent. Polcy Research Workng Papers are avalable onlne at http://econ.worldbank.org. For comments on ths paper the authors are grateful to Bob Baulch, Quy-Toan Do, Andrew Foster, Emanuela Galasso and Rnku Murga. Useful dscussons on ths topc were had wth Ma Lan Lam, Quang Bnh, Luc Duc Kha, Dnh Duc Snh and partcpants at semnars n Hano (sponsored by the Natonal Economcs Unversty and the Insttute of Socal Studes, Holland), the North-Eastern Unverstes Development Conference at Wllams College and the Unversty of Massachusetts. The support of the World Bank s Research Commttee s gratefully acknowledged as s the assstance of Dorothyjean Cratty and Tomom Tanaka.
Abstract Whle lberalzng key factor markets s a crucal step n the transton from a socalst control-economy to a market economy, the process can be stalled by mperfect nformaton, hgh transacton costs, and covert resstance from entrenched nterests. The paper studes land-market adjustment n the wake of Vetnam s reforms amng to establsh a free market n land-use rghts followng de-collectvzaton. Ineffcences n the ntal admnstratve allocaton are measured aganst an explct counter-factual market soluton. Our tests usng a farm-household panel data set spannng the reforms suggest that land allocaton responded postvely but slowly to the neffcences of the admnstratve allocaton. We fnd no sgn that the transton favored the land rch or that t was thwarted by the contnung power over land held by local offcals. Key words: Land reform, decentralzaton, land markets, Vetnam JEL codes: D60, P21, Q15
1. Introducton Vetnam s agraran transton n the 1990s has closely followed a now classc polcy scenaro for economes n transton. Frst one prvatzes the man productve assets n ths case agrcultural land-use rghts then one legalzes ther free exchange. In the frst step, the de-collectvzaton of agrculture meant that the land that had been farmed collectvely was to be allocated by admnstratve means wthn each commune. Naturally ths left neffcences n land allocaton, wth some households havng too much land relatve to a compettve market allocaton, whle some had too lttle. The second step was reformng land laws so as to create the framework for a free market n agrcultural land-use rghts. Whle land remaned the property of the state, Vetnam reformed land laws n 1993 to ntroduce offcal land ttles and permt land transactons for the frst tme. Havng removed legal obstacles to buyng and sellng landuse rghts, the expectaton was that land would be re-allocated to elmnate the ntal neffcences n the admnstratve assgnment. However, the outcomes are far from clear on a pror grounds. Land was not the only nput for whch the market was mssng or mperfect. Indeed, as a stylzed fact, other factor markets are stll poorly developed n rural areas, whch s lkely to lmt the effcency gans from freeng up land transactons. Pervasve market falures fuelled by mperfect nformaton and hgh transacton costs could well have stalled the process of effcency-enhancng land re-allocatons durng Vetnam s agraran transton. The local state contnued to play an actve role. However, t s unclear whether the contnung exercse of communal control over land was synergstc wth market forces or opposed to them. Possbly the local poltcal economy operated to encourage otherwse sluggsh land re-allocaton to more effcent users.1 Or t may have worked aganst effcent agraran transton, gven pervasve rsk-market falures and lmtatons on the set of redstrbutve nstruments; resstance to the transton may then be an endogenous safety net, recognzng the welfare rsks that a free market n land mght ental. Or t mght be expected that the frctons to agraran transton stemmng from the local poltcal economy worked aganst both greater equty and effcency; whle socalsm may have left n-graned preferences for dstrbutve justce, the new possbltes for 1 In the context of rural Chna, Benjamn and Brandt (2002b) argue that admnstratve land re - allocatons served an effcency role gven other market fa lures. 1
capture by buddng local eltes well connected to the local state authortes would not presumably have gone unnotced. The ex post outcomes of ths reform n Vetnam are also of nterest to neghborng Chna, whch s plannng to lberalze the exchange of agrcultural land-use rghts from 2003 (McGregor and Kynge, 2002). As n Vetnam, the hope s that land wll be reallocated to more effcent users, and that neffcent farmers wll swtch to (rural or urban) nonfarm actvtes. And, as n Vetnam, there are concerns n Chna that local offcals and eltes wll subvert the process. Ths paper offers what we beleve to be the frst emprcal test of whether the classc polcy scenaro of prvatzaton followed by lberalzed exchange has actually worked n a developng transton economy. In partcular, the paper assesses whether the post-reform allocaton of annual agrcultural land-use rghts n Vetnam redressed the neffcences of the ntal admnstratve allocaton. We frst measure the extent of neffcences n the pre-reform admnstratve allocaton, judged relatve to an explct counter factual. We then see to what extent those neffcences can explan the subsequent land re-allocatons n a panel of farm households, wth controls for other non-market factors bearng on land allocaton. The followng secton descrbes key features of the settng. Secton 3 descrbes our approach to testng whether the post-reform land re-allocaton responded to the household-specfc effcency losses from the pre-reform admnstratve allocaton. Our data are descrbed n secton 4. We then present and nterpret our results n secton 5. Secton 6 concludes. 2. Land allocaton n Vetnam s agraran transton In the late 1980s, Vetnam abandoned socalst agrculture, whereby rural workers had been organzed nto brgades that jontly farmed the commune s land. The central government gave local authortes the power to allocate the agrcultural land that had been farmed collectvely to ndvdual households. De-collectvzaton was followed n 1993 by a new land law that ntroduced offcal land ttles and permtted land transactons for the frst tme snce communst rule began. Land remaned the property of the state, but usage rghts were extended (typcally from 15 to 20 years for annual crop-land) and could (for the frst tme) be legally transferred and exchanged, mortgaged and nherted (Cuc and Skor, 1998). 2
The central government s explct am n ntroducng ths new land law was to promote greater effcency n producton by creatng a market n land-use rghts (see, for example, de Mauny and Vu, 1998). (Ths was one element of a set of reforms to ncrease agrcultural output; other reforms nclude relaxng trade restrctons, whch mproved farmers terms of trade; see Benjamn and Brandt, 2002a.) The expectaton was that, after these legal changes, land would be re-allocated to assure hgher agrcultural output, takng account of such factors as farmers abltes, supervson costs of hrng labor and the mcro-geographc organzaton of land plots. Despte the center s am of creatng a free market n land-use rghts, local authortes retaned a degree of power over land. Local cadres oversee ttlng, land-use restrctons and land appropraton for nfrastructure projects. Skor and Truong (2000) descrbe well how the reforms were medated by vllage nsttutons n Son La, a northern uplands provnce: Local cadres were located at the ntersecton of the state and vllages. A large majorty of them came from local vllages and mantaned close tes wth ther kn and fellow vllages. The close tes between local cadres and vllages nfluenced the actvtes of the local state. Local cadres attempted to accommodate vllagers nterests, sometmes even when they contradcted natonal polcy. (Skor and Truong, 2000, p.33). In these crcumstances, t would be wrong to vew the land-market reform as undermnng the power of the local state over land allocaton. Indeed, staff of one NGO argued that the reforms enhanced the power of the state over land usage (Smth and Bnh, 1994). Although both the 1988 and 1993 land laws extended land use rghts for stable and long-term use there are reports that some local authortes contnue to re-allocate land perodcally by admnstratve means, such as n response to demographc changes and new famly formatons. There s anecdotal evdence that the contnung power of the local state stalled the reforms n some parts of Vetnam. Wrtng a few years after the 1993 Land Law, Smth (1997) reports that n one northern provnce (Ha Tnh) the major commercal bank lendng for agrcultural purposes had not yet accepted a sngle land-usage certfcate as collateral for a loan. The resstance of local offcals to have the land sold to an outsder was one of the reasons gven by the bank; another was that the bank was unsure t would ever fnd a buyer for the land should t foreclose on the loan. However, ths should not be generalzed; ndeed, the same study reported cases of land certfcates beng accepted as collateral n another provnce. 3
Just how much the local state has nhbted the development of a land market s unclear. It appears that land transactons can by-pass state control. There have been reports of land transactons wthout ttles (Smth, 1997; de Mauny and Vu, 1998). Possbly a quas-market has emerged despte the contnung ntrusons of the local state. There have also been concerns about rsng nequalty stemmng from the reforms. A report by ActonAd staff exemplfes these concerns; whle presentng no supportve evdence, the report predcted that the reforms would lead to:..a greater concentraton of land ownershp, a greater dsparty n wealth throughout the rural communty and a possble ncrease n the phenomenon of landlessness and full-tme agrcultural wage labour. (Smth and Bnh, 1994, p.17.) There have been reports of rsng landlessness, notably n the south (de Mauny and Vu, 1998; Lam, 2001b). However, there s lttle sgn of sharply rsng ncome or consumpton nequalty. 2 Some of the efforts made to avod rsng nequalty may well have had perverse effects. There are reports that, n response to central Communst Party concerns about rsng landlessness n the late 1990s, some local offcals n the south tred to stop poor famles sellng ther land (de Mauny and Vu, 1998). The consequent devaluaton of ther man non-labor asset would presumably make the poor worse off. It s lkely that transfers stll happened despte such polces, though the transactons would become nformal, and possbly on less favorable terms for those forced to sell ther land because of adverse shocks. There were dfferences between the north and the south that are lkely to have mattered to the pace of the agraran transton. After re-unfcaton n the md-1970s, farmers n the south s Mekong Delta had ressted collectvzaton, and by the tme the country de-collectvzed 13 years later, less than 10 percent of all of the regon s farmers had been organzed nto collectves. By contrast, vrtually all of the crop land n the north and the south s Central Coastal provnces was collectvzed by that tme (Pngal and Xuan 1992; Ngo 1993). The market economy was thus more developed n the Mekong Delta at the begnnng of the transton. It mght be expected that ths hstorcal dfference would 2 Analyses of household survey data for 1992/93 and 1997/98 ndcate a sgnfcant drop n ncome nequalty n the south (from a Gn of 0.46 to 0.42), though there was a slght ncrease n the North (from 0.37 to 0.39) and a slght ncrease n consumpton nequalty n both north and south (Benjamn and Brandt 2002a, Glewwe et al. 2001), though the statstcal sgnfcance of these changes s a moot pont. 4
mean that land allocaton would adjust more rapdly n the Mekong after the reforms. However, there are other factors to consder. Rural per capta ncome growth was hgher n the south over ths perod, fuelled n part by mprovements n farmers terms of trade arsng from external trade reforms; Benjamn and Brandt (2002a) report a 95 percent ncrease n real ncome per person n the south over 1993-98, versus 55 percent n the north. Such rapd growth n real ncomes may well have dampened the pressure to secure the effcency gans from land re-allocaton n the south. There were other pre-reform dfferences between the north and south. The dstrbuton of land was more equal n the north. 3 The collectvzaton of agrculture n the north over roughly a generaton fostered a more equtable allocaton at the tme of decollectvzaton. In the south, the fall back poston was the land allocaton preunfcaton, and the realzed allocaton was more unequal than n the north (Ravallon and van de Walle, 2001). Lower nequalty n the north may well have made t easer to acheve cooperatve outcomes, ncludng more effcent assgnments of land-use rghts. 4 A related manfestaton of ths dfference can be found n the performance of (formal and nformal) nsttutons that deal wth rsk and are also lkely to matter to land allocaton. The safety net n rural areas of Vetnam s largely communty-based; central and provncal programs have weak coverage (van de Walle, 2002). It s wdely beleved that vllages n the north are better organzed socally than n the south, so that when a farm household n the north suffers a negatve shock (such as crop damage or ll-health) t wll almost never need to sell land to cope. For example, wrtng about Son La provnce, Smth reports that:..there s a tendency for the local authortes to seek to protect households from the dangers of a market n land, despte the provsons of the 1993 Law. Ths consttutes an attempt to protect poor households who may be tempted to sell ther land for short term gan and lose ther prncpal means of subsstence. (Smth, 1997, p.11.) By contrast, an Oxfam team n the provnce of Tra Vnh n the Mekong Delta (n whch the NGO had been workng for a few years) reported that: 3 Ths dfference shows up n the results from the VLSS of 1992/93. The coeffcent of varaton n the log of allocated annual agrcultural land was 8.3% n the North s Red Rver Delta, versus 15.3% n the south s Mekong Delta (Ravallon and van de Walle 2001). (Among the fve regons for whch the sample sze was deemed adequate, these were the regons wth lowest and hghest land nequalty respectvely.) 4 For an excellent revew of the theoretcal arguments as to why hgh nequalty can mpede effcency see Bardhan et al., (1999). 5
The crucal problem s that there are no safety nets for helpng households who encounter temporary crses. It s no surprse that many famles resort to transferrng or mortgagng ther land, dscountng the future to cope wth the current crss (de Mauny and Vu, 1998, p.23). Ths dfference between the north and the south s no doubt n part a legacy of the longer perod of collectve organzaton n the north. However, the more equal land allocaton n the north after breakng up the collectves could well have facltated ths, by makng t easer to contnue to acheve quas-cooperatve arrangements wthn communtes. Better nsurance n the north s lkely to have also made t easer for land transactons to be made on effcency grounds. Landholdngs n the south, by contrast, are lkely to have been less flexble, snce land would be more lkely to be held as nsurance than n the north. These observatons suggest that t would be naïve to thnk that smply legslatng the pre-requstes for a compettve land market n ths settng would make t happen. The realty s more complex and uncertan, gven the nsttutonal/hstorcal context. In prncple, the contnung (and possbly enhanced) power of local cadres could ether undermne the expected effcency gans from the center s reforms or help secure those gans. The dstrbutonal outcomes are equally unclear; the local state had the power to ether magnfy any adverse dstrbutonal mpacts of the reforms, or dampen them. The outcome s lkely to depend n large part on the outcomes of a power struggle at local level, whch can be taken to determne the (explct or mplct) dstrbutonal goals of the local land allocaton process. Capture of ths allocaton process by local eltes could lead to even worse dstrbutonal outcomes. 5 On the other hand, a desre to protect the poor could soften the mpact. These same features of the Vetnamese rural economy that could nhbt the effcency gans from ntroducng land ttles and other trappngs of the market economy lead one to queston any presumpton that effcency gans from the land law would necessarly come wth a cost to equty. Local nsttutons would have been capable of both stallng the market and protectng the poor from any polarzng forces t generated. In the rest of ths paper we wll study the outcomes of ths process of post-reform land re-allocaton, gven ts nsttutonal and hstorcal context. 5 Ths has been a concern n recent analyses of the case for communty-based welfare programs more generally (Bardhan and Mookherjee, 2000; Galasso and Ravallon, 2001). 6
3. Modelng land allocaton The man hypothess to be tested s that land re-allocaton durng the agraran transton helped offset pror neffcences n the admnstratve allocaton. To test ths, we need to explctly characterze the extent of neffcency n the ntal allocaton. Then we wll see how subsequent re-allocatons of land responded. 3.1 Ganers and losers from the ntal admnstratve allocaton An ntal admnstratve allocaton of land was made as part of de- A collectvzaton, gvng an amount L of land to household for =1,..,n. The admnstratve allocaton need not be effcent n the specfc sense of maxmzng aggregate output or consumpton. To characterze the effcent allocaton, suppose that holdng L of land yelds an output of F ( L, X ) for household where X s a vector of exogenous household characterstcs. We assume that the functon F s ncreasng and strctly concave n L. The household also has (postve or negatve) non-farm ncome, Y ( X). The household consumes ts current ncome: 6 C = C L, X ) = F( L, X ) + Y( X ) (1) ( The allocaton that maxmzes the commune s aggregate current consumpton s: n * * ( L1,.., Ln ) = arg max[ C( L, X) L = nl] = 1 n = 1 (2) The soluton equates F L, X ) wth the multpler λ on aggregate land n (2), gvng: ( * L * L = L(, λ) (=1,..,n) (3) X We call ths the consumpton-effcent allocaton. Ths s also the compettve equlbrum assumng that utlty depends solely on consumpton. In the market allocaton, each household s consumpton wll be market prce of land. Demands then equate F( L, X ) + Y( X ) λl where λ s the F L, X ) = λ over all, whch s the L( allocaton that maxmzes aggregate consumpton. In our emprcal mplementaton, we assume that (1) takes the specfc form: ln C = a + bln L + cx + ν (4) 7
where a, b and c are parameters and ν s a whte nose error process. Gven estmates of the parameters and error term and data on X, we then calculate the consumpton effcent allocaton to each household. For 0<b<1 the soluton s * L = exp[(ln( b / λ ) + X c + ν ) /(1 b)]. The effcency loss from the admnstratve allocaton s measured by τ * A * A = τ( L, L ) = φ( L ) φ( L ) (5) for some strctly ncreasng functon φ ; we adopt ths functonal form to assure that τ ( L, L) = 0. We can embrace a reasonably wde range of possble emprcal measures by restrctng attenton to the class of functons: extreme cases are () proportonate dfferences, n whch η = 0 * A ln( L / ) (notng that lm ( L 1)/ η = ln L L whereby τ = L * A L. η 0 3.2 Modelng post-reform land re-allocaton η η φ( L ) = ( L 1) / η where η [0,1]. The two, mplyng that τ = ); and () absolute dfferences ( η = 1) We only observe a sngle tme nterval n the process of land re-allocaton after legalzng market transactons and we do not, of course, assume that the process has reached ts long-run soluton by the end of the perod of observaton. However, we do assume that the dynamc process wll eventually converge to a unque long-run equlbrum, whch depends on the compettve market allocaton of land to that household but can also be nfluenced by the household s weght n local decson makng about the allocaton of use rghts. R R R The new allocaton at a date after the reform s L, L,..., L ). Let R A ( 1 2 ρ = ρ( L, L ) denote a measure of the extent of land re-allocaton. We clearly want R A ρ ( L, L ) to be strctly ncreasng n L and decreasng n R A want to assure that f ρ ( L, L ) = τ L, L ) then R ( * A R * A n L wth ρ ( L, L) = 0. We also L = L ; f land re-allocaton for household exactly matches the ntal effcency loss then the household must have reached the market soluton. These condtons requre that ρ and τ have the same functonal form.e., ρ = φ( L ) φ( L ). R A To see how land allocaton responded to ntal neffcences we begn by studyng the non-parametrc regresson: ρ ( + (6) = f τ ) ε 6 We gnore savng/dssavng and borrowng/lendng; ncorporatng these features would complcate the model n unmportant ways for our purposes. 8
where f τ ) E [ ρ τ ]. In the extreme case wth f ( 0) = 0 and f ( τ ) = 1, there are no ( ε systematc non-market constrants on land re-allocaton, so R * L = L n expectaton. Adjustment to the market soluton s then complete wthn the perod of observaton. More generally one can allow 0 f ( τ ) 1 n whch case we have a (nonlnear) partal adjustment model by whch land holdngs adjust to any dscrepances between the admnstratve allocaton and the market soluton, though the process need not be * complete n the perod of observaton. Wth repeated observatons, L wll be reached whatever the ntal start value of the process (n ths case, the admnstratve allocaton at de-collectvzaton). The slope, f τ ), s the partal adjustment coeffcent for ( household gvng the speed at whch ntal neffcences are elmnated. The smple partal adjustment model s questonable from a number of ponts of vew. One concern s the possblty of measurement error n the data for the ntal land A allocaton. Classcal measurement error n L wll bas the Ordnary Least Squares (OLS) estmate of the lnear partal adjustment coeffcent, though the drecton of bas s ambguous n ths case. (The usual attenuaton bas wll be at least partly offset by the fact that the measurement error also appears postvely n the dependent varable.) Wth an extra pre-reform survey round one could correct for ths usng an Instrumental Varables Estmator, but that s not an opton. However, land allocaton appears to be well known at farm-household and commune level, and so we do not expect szeable bas for ths reason. A second concern s that the process may not be homogeneous n that the ntal land allocaton may nfluence land re-allocaton ndependently of the gans and losses from the ntal admnstratve allocaton. Imposng homogenety when t does not hold wll bas upward (downward) the OLS partal adjustment coeffcent f there s convergence (dvergence) at a gven land defct relatve to the effcent allocaton. By addng error n A A L as an addtonal regressor, we can test homogenety. Agan, any measurement L may nduce some bas, whch wll tend towards showng convergence. A thrd concern s that the effcent allocaton of land may have changed over tme. For example, demographc shocks wll no doubt shft the consumpton-effcent allocaton. Ths can be thought of as measurement error n our estmate of the loss from the admnstratve allocaton. We address ths ssue by addng controls for observed changes n household characterstcs that are lkely to nfluence the effcent allocaton. Latent measurement error wll leave some bas. 9
A fnal concern s that the local poltcal economy nfluenced land re-allocaton, R* as dscussed n secton 2. We can postulate nstead a soluton, L, such that the hgher * R* τ ( L, L ), the hgher the weght that a gven household has n local decson makng R* about land. We assume that L depends on assets (educaton and other types of land), connectons (such as havng a government job and beng a long-standng resdent) and possble dscrmnatng varables (such as gender of head and ethncty). We then augment the partal adjustment model for these household characterstcs. Notce that the ntal admnstratve allocaton may tself be one such factor; f a hgher ntal admnstratve allocaton gave one the power to acqure more land then we wll see sgns of a dvergent (non-statonary) process. Combnng these consderatons, we shall estmate a parametrc model: ρ = α + βτ + γ ln L + πz + ε (7) A n whch Z denotes a vector of other controls for other (market and non-market) factors, ncludng demographc shocks, nfluencng land allocaton. It s readly verfed that the long-run soluton to (7) (when R A R* L = L = L and S = 0 ) s: R* 1 * α γ A π ε L = φ [ φ( L ) + + ln L + Z + ] (8) β β β β We can also allow the partal regresson coeffcent of ρ on τ to vary between ndvduals accordng to ther characterstcs, by testng for approprate nteracton terms to equaton (7). In augmentng the uncondtonal partal adjustment model for these controls, we wll not be able to cleanly separate market from non-market forces on land allocaton. In ths settng t s hard to magne any household characterstc that could be unambguously nterpreted as one rather than the other. For example, fndng a sgnfcant effect of gender or ethncty s suggestve of a non-market force at work, but we cannot know n whch market t operates; possbly the dscrmnaton s n access to credt rather than land. However, we wll be able to see whether the controls renforce or offset the adjustment process. We wll say that the controls are cooperant ( noncooperant ) wth the market forces arsng from neffcences n the ntal admnstratve allocaton f the uncondtonal adjustment coeffcent (settng γ = π = 0) s found to be based upward (downward). 10
4. Data We use the household panel data from the 1992/93 and 1997/98 Vetnam Lvng Standard Surveys (VLSS). The frst survey preceded the change n the land laws n 1993. These are natonally representatve, hgh qualty surveys wth comprehensve and carefully collected data on a wde range of household characterstcs ncludng consumpton expendtures, producton and land holdngs (World Bank 1995 and 2000). The surveys contan a balanced panel of 4308 households. We lmt our sample to the 2559 rural farmng households n the panel who had allocated annual agrcultural land n 1993. The 1992/93 VLSS s self weghted so that expanson factors are not needed. Both surveys spanned 12 months. Perennal, forest and water surface land have also been allocated to households. However, we focus on allocated annual agrcultural land because of ts mportance n producton and total area, and because ts allocaton began earler and has progressed more rapdly than for other land types. 7 (Annual agrcultural land s for annual crops such as rce or groundnuts.) Annual agrcultural land can be rrgated or non-rrgated. To facltate the analyss we convert all allocated annual agrcultural land nto an allocated rrgated land equvalent amount for each household. Non-rrgated land amounts are weghted by the rato of the coeffcents on non-rrgated to that of rrgated land estmated from regonspecfc regressons of farm profts on allocated rrgated and non-rrgated annual land and all other land cultvated by households, household characterstcs and commune dummes. The weghts are estmated usng the 1992/93 VLSS and used to create the allocated rrgated land equvalents n both 1992/93 and 1997/98. 8 A household s cultvated land can dffer from ts allocated land. Rural households typcally have ther own prvate resdental land wth ts garden area. We consder ths type of land as beng a well-known and longstandng asset assocated wth each household and hence we control for t n our analyss. The rental market s thn. Rented-n land represented 6.2 percent of annual crop land n 1993 and 5.1 percent n 1998. A more actve rental market has clearly not emerged snce the reforms. Our mpresson s that rentals tend to be temporary arrangements, such as when a famly worker s sck or temporarly absent. There s also a small amount of aucton land that s effectvely rented from the commune. (Ths accounted for 2.1 percent of all cultvated land n 1993, 7 We wll hereafter refer to allocated annual agrcultural land smply as allocated land. 8 See Ravallon and van de Walle (2001) on constructon of the allocated land equvalent. 11
and 2.2 percent n 1998.) We do not control for land obtaned though rental arrangements, gven the possble endogenety concerns. The land stuaton has been evolvng durng the 1990s reflectng changng offcal atttudes towards the market economy and the role of land, and consequent polcy and legal reforms. Ths s apparent n the surveys. Land categores and defntons changed between the 1992/93 and 1997/98 VLSS. Our am here s to study changes n the allocated annual land amounts over tme. Fortunately, ths s straghtforward. In 1993, our allocated land varable comprses the questonnare categores allocated and long-termuse annual land. (Both categores refer to land allocated to households for long-term use. They dffer only n that the allocaton terms are slghtly dfferent wth the frst arrangement more common n the north and the second more so n the south.) By 1998, ths dstncton s no longer enforced. The 1997/98 VLSS refers to allocated land as ether long-term-use or contract land. The latter s also allocated to households for long-term and stable use, but ts land-use ttle s held by a state managed farm or enterprse rather than the household. Ths category of land was subsumed n ether allocated or long-term use land n the 1992/93 survey. We consder ths to be part of the allocated land category n 1998. Fnally, n contrast to the 1992/93 VLSS where allocated annual land amounts nclude any area that was rented out, the latter s recorded separately n 1998 and so must be added n to determne the household s total allocated annual land amount. The measure of consumpton n 1992/93 (used to estmate the consumptoneffcent land allocaton) ncludes the value of consumpton from own producton, mputed housng expendtures and the use value of consumer durables (World Bank 1995). It also takes account of temporal prce varaton across the survey year as well as spatal prce dfferentals and s expressed n real 1993 Dongs. Vetnam s characterzed by marked geographcal varaton, some of whch reflects dfferent hstorcal evolutons. The country s commonly dvded nto seven regons that are relatvely homogeneous. We estmate our regressons natonally as well as for the fve regons for whch there was suffcent data, namely the Northern Uplands, the Red Rver, North Coast (these three are n the north) and the Central Coast and Mekong Delta (the south). In addton, the augmented model ncludes a full set of commune dummy varables to capture geographc dfferences n prces and possbly nsttutonal dfferences. 12
In our augmented model below we control for exogenous household level varables that descrbe the household s ntal 1993 stuaton n terms of assets, connectons and possble dscrmnatng varables. These nclude the years of educaton of the head and of other household adults; dummy varables for hs/her relgon (1 f the head practces the Chrstan or Buddhst relgon, 0 otherwse), ethncty (1 f the head belongs to an ethnc group other than the majorty Knh or relatvely wealthy Chnese mnorty) and whether born locally; dummes for whether the household contans one or more handcapped adult members, members who work for the government or for a state owned enterprse, and whether the household s a recpent of socal nsurance fund transfers. The latter are gven to war heroes or martyrs and ther famles households that are often sngled out for preferental polcy treatment by the authortes. The fact of recevng the transfer s the only way of dentfyng them n our data. We run the model wth and wthout ths dummy varable. We also control for the household s prvate land (dscussed above), whether t cultvates swdden land or not, and the share of ts rrgated and non-rrgated land that s consdered of good qualty. In addton, we nclude varables that capture exogenous changes n the household s characterstcs that are lkely to shft the consumpton effcent allocaton namely the change n the number of dsabled adult members, the change n the number of able boded workng age members, the number of new members aged between 8 and 99 n 1998, and whether an adult or elderly member ded between the two surveys. Table 1 provdes summary statstcs for the natonal sample. We also present the data separately for the Mekong Delta and for the natonal sample omttng the Mekong Delta. 13
Table 1: Varable defntons and summary statstcs Mean st.dev. Log change n allocated rrgated land equvalent (m²) 0.142 0.66 Proportonal effcency loss (log effcent allocaton mnus log -0.016 0.78 actual n 1993) Relgon: 1 f h hold head s Buddhst or Chrstan (0 f other, 0.307 0.46 anmst or none) Ethnc: 1 f h hold head s of ethncty other than majorty 0.121 0.33 Knh or Chnese Local born: 1 f head s born locally 0.861 0.35 Gender of household head (male=1) 0.791 0.41 Labor age adult member s handcapped 0.007 0.09 SOE: member has prmary or secondary occupaton n State 0.018 0.14 owned enterprse Gov t job: member works for gov t n prmary/ secondary 0.059 0.25 occupaton or retred from gov t (professonal codes 20 and 21) Socal subsdy: dummy var. for recept of gov t transfers to 0.103 0.30 war heroes, martyrs, dsabled etc Household head s years of educaton 6.107 3.83 Other h hold adults years of educaton 10.648 9.22 H hold s prvate rrgated land (m²) 158.853 658.68 H hold s prvate non-rrgated land (m²) 228.824 955.31 H hold s prvate perennal land (m²) 349.057 1492.13 H hold s prvate water surface land (m²) 55.913 478.74 H hold cultvates swdden land=1 0.108 0.31 Share of good rrgated land 0.304 0.39 Share of good non-rrgated land 0.374 0.46 No. >=16 n 1993 who ded by 1998 0.109 0.33 No. >=50 n 1993 who ded by 1998 0.089 0.30 Change n number of dsabled adults 1993-98 -0.004 0.15 Change n no. of able boded workng age members 1993-98 -0.138 1.19 H hold has new ndvdual aged 8-99 n 1998 0216 0.60 Source: 1992/93 and 1997/98 Vet Nam Lvng Standards Surveys. 2559 observatons except for the change n log allocated land for whch n=2361. 5. Results Recall that n measurng land re-allocaton and the ntal effcency loss we η assume that φ( L ) = ( L 1) / η where η [0,1]. To choose a value of η we regressed ρ on τ across the entre data set for alternatve values of η at 0.1 ntervals over the [0,1] nterval. The best ft (measured by the t-rato on the partal adjustment coeffcent) was obtaned at η = 0, whch gave a partal adjustment coeffcent for proportonate dfferences of 0.33. 9 The coeffcent for absolute dfferences ( η = 1) was 0.17 and 9 All standard errors n ths paper are corrected for both heteroskedastcty and clusterng. 14
between the two, the t-rato declned monotoncally. So we chose the proportonate (log dfference) specfcaton n all further work. However, ths specfcaton has the drawback that we lose some observatons wth zero land allocaton n 1997/98 (snce we cannot take the log of zero); ths apples to slghtly less than 8 percent of the sample. 10 We wll study ths sub-sample wth zero allocated land n the second survey more closely, and test for sample selecton bas, later n ths secton. For the present dscusson we confne attenton to the proportonate case. Fgure 1 plots the proportonate changes (log dfferences) n land allocaton aganst our measure of the ntal loss relatve to the effcent allocaton, measured * A by ln( L / ), for the natonal sample. The emprcal relatonshp suggests a tendency for L land re-allocaton to respond postvely to the ntal neffcency n the admnstratve allocaton. As already noted, the lnear regresson coeffcent s 0.33 (wth a t-rato of 9.8), ndcatng that one thrd of the ntal dsparty between the admnstratve allocaton and the market allocaton was elmnated over ths fve year perod. Fgure 1 also gves the nonparametrc regresson functon (usng Cleveland s, 1979, local regresson method). The slope s postve but less than unty throughout, though t s clear that f ( 0) 0, reflectng an overall expanson n allocated annual land area over ths perod. Fgure 1 s suggestve of partal adjustment toward the market allocaton, though stll leavng two-thrds of the ntal mean proportonate effcency loss after fve years. However, as noted n the prevous secton, there are a number of concerns about bas, whch mght go n ether drecton. One concern s that the relatonshp mght not be A homogeneous, as assumed by equaton (6). On addng ln L to the regresson of R A * A ln( L / L ) on ln( L / ), we could convncngly reject the null hypothess mpled by L homogenety. The regresson coeffcent on partal adjustment coeffcent fell to 0.217 (7.09). A ln L was -0.287 (t-rato of 8.05), whle the 10 We also tred defnng the proportonate dfference as the percentage change rather than log dfference, thus allowng us to keep these observatons; the results were smlar, though (agan) the log dfference specfcaton gave a better ft. 15
Fgure 1: Proportonate land re-allocaton 1993-98 aganst the proportonate loss from the admnstratve allocaton n 1993 4 Change n log land allocaton 1993-98: Natonal 2 0-2 -4-2 0 2 4 Proportonate effcency loss 1993 Table 2 gves the estmated partal adjustment coeffcents when the varous controls are added step-by-step (cumulatvely). We gve natonal results and a breakdown by regon. Let us focus frst on the natonal results. Consstently wth Fgure 1, all of our tests ndcate a hghly sgnfcant postve coeffcent on the ntal effcency loss, mplyng that the land re-allocaton process was n the drecton of a more effcent allocaton. However, as can be seen from Table 2, the partal adjustment coeffcent falls to less than half the value mpled by Fgure 1 when all controls are added. Ths s the combned effect of both relaxng homogenety and addng the controls for shocks and non-market factors, ncludng commune fxed effects. There were also many sgnfcant commune effects. These could reflect prces rather than nsttutonal factors. Of all these changes, relaxng homogenety and addng commune effects does most of the work; wth just these two changes, the partal adjustment coeffcent falls to 0.155 (t=5.18), whle addng the rest of the control varables only brngs t down an extra 0.014 (Table 2). 16
Table 2: Effects of addng controls on the partal adjustment coeffcents by regon Northern Uplands Red Rver North Coast Central Coast Mekong Delta Full Sample No controls 0.476 (5.97) 0.294 (6.81) 0.306 (3.35) 0.172 (2.17) 0.350 (4.51) 0.328 (9.82) Addng ntal land allocaton 0.170 (1.61) 0.094 (2.67) 0.129 (1.24) 0.025 (0.37) 0.221 (3.06) 0.218 (7.09) Addng commune effects 0.205 (3.96) 0.123 (2.98) 0.132 (1.52) 0.079 (1.32) 0.171 (1.62) 0.155 (5.18) Addng controls for demographc shocks 0.255 (4.89) 0.150 (4.02) 0.175 (2.24) 0.074 (1.15) 0.215 (2.20) 0.182 (6.46) Addng controls for connectons and assets 0.268 (4.54) 0.071 (1.39) 0.173 (1.68) 0.069 (1.16) 0.074 (0.73) 0.131 (4.09) No. observatons 432 790 459 269 308 2,361 Note: The table gves regresson coeffcents of the change n log annual land allocaton on the estmated proportonate loss from the ntal admnstratve allocaton relatve to the counterfactual market allocaton. The regressons are cumulatve n that as controls are added the prevous controls are kept n. There are regonal dfferences n the estmated adjustment coeffcents, though the pattern of declnng coeffcents as controls are added s smlar across regons. There s lttle sgn of a dfference between the north and the south; whle the hghest coeffcent wthout controls s for the Northern Uplands, the south s Mekong Delta s the second hghest. Whle the separaton of market from non-market forces s clearly problematc n ths settng, these results suggest that any non-market forces beng pcked up by our controls tended to be cooperant wth market forces, as captured by the adjustment coeffcent to ntal losses from the admnstratve allocaton. Ths s evdent from the fact that, on balance, controls that rase (lower) land allocaton tend to be postvely (negatvely) correlated wth the loss due to the ntal allocaton. The only excepton s for the controls for demographc shocks, whch tended to work n the opposte drecton (as s evdent n Table 2), though the effect on the partal adjustment coeffcent s small. In Table 3 we gve the complete results for the most comprehensve model we estmated. For ths we also added nteracton effects between the ntal loss varable and both ntal land allocaton and head s educaton, to allow the adjustment coeffcent to vary wthn regons. The nteracton effect wth educaton was nsgnfcant natonally and n most regons. However, we fnd a sgnfcant nteracton effect between the ntal loss relatve to the effcent allocaton and the ntal land allocaton. The speed of 17
adjustment toward the effcent allocaton was hgher for those who started off wth less land. We fnd a number of other factors that nfluence land re-allocaton. There s a hghly sgnfcant effect of an ncrease over the tme perod n the number of persons of workng age and new people jonng the household. (We also tred droppng the latter varable gven possble endogenety concerns, but other results were affected lttle n the natonal model.) Households wth male heads were also favored n the land re-allocaton process. Havng hgher amounts of other types of land resulted n sgnfcantly hgher access to allocated land. There are some regonal dfferences n the model wth controls. The sgnfcant negatve nteracton effect (such that there s a hgher adjustment coeffcent for households wth less land) s only found n the Mekong. Whether ths s a market response s unclear; t could also reflect the efforts of local offcals n the Mekong to avod rsng landlessness (Secton 2). The mpacts of demographc and labor force changes appear to be generally stronger n the northern provnces. Ths s also where local authortes are more lkely to enforce perodc land re-allocatons. Beng from an ethnc mnorty household helped ncrease annual land holdngs n the north, and (especally) the Central Coastal regon, whle t tended to reduce holdngs n the Mekong Delta; note, however, that the ethnc groups are not the same n these two regons. Ethnc effects also become sgnfcant and postve n the Northern Uplands and North Coast regons when we omt the number of new household members n 1998. Havng a member who works for an SOE has a pronounced negatve mpact on annual land changes n the Northern Uplands and the Central Coast, though t has no mpact elsewhere. In both the Northern Uplands and Central Coast regons a hgher share of good qualty rrgated land reduced the land reallocaton over tme. 11 The tendency to favor male heads of household s strongest n the north. 11 We tested a dummy for beng a socal fund transfer recpent, one of the few ways to dentfy households that may be treated preferentally by local authortes. Ths was nsgnfcant n the natonal model and all regons except the North Coast where t had a postve effect. 18
Table 3: Determnants of changes n allocated annual agrcultural land Northern Uplands Red Rver North Coast Central Coast Mekong Delta Full sample Proportonal loss from admn. allocaton 0.433 (2.65) 0.197 (0.52) 0.501 (1.09) 0.230 (0.67) 1.494 (2.90) 0.700 (4.51) Log ntal land allocaton -0.481 (7.20) -0.434 (6.32) -0.298 (3.47) -0.495 (10.04) -0.394 (4.01) -0.405 (11.78) Interacton of loss wth ntal land -0.024 (1.06) -0.017 (0.34) -0.047 (0.84) -0.022 (0.52) -0.168 (3.02) -0.077 (3.87) Adult member ded 1993-98 0.096 (0.52) 0.110 (1.22) 0.043 (0.18) -0.059 (0.53) 0.170 (1.07) 0.043 (0.53) Elderly member ded 1993-98 -0.150 (0.67) -0.118 (1.18) -0.034 (0.14) -0.143 (0.96) -0.162 (0.99) -0.080 (0.88) Change n no. dsabled 1993-98 0.204 (2.15) 0.240 (1.66) 0.122 (1.77) 0.043 (0.43) -0.008 (0.04) 0.119 (2.03) Change n no. of able boded members 0.119 (5.08) 0.150 (8.70) 0.119 (5.56) 0.052 (1.44) 0.05 (1.72) 0.100 (8.92) New member 8-99 1993-98 0.113 (2.20) 0.189 (4.59) 0.111 (1.73) 0.050 (0.94) 0.205 (3.74) 0.124 (5.00) Relgon 0.151 (2.13) -0.049 (1.12) 0.020 (0.20) -0.054 (0.45) 0.126 (2.61) 0.005 (0.16) Ethncty 0.254 (2.06) -0.128 (3.40) 0.089 (0.75) 1.014 (14.57) -0.288 (1.44) 0.096 (0.93) Born locally 0.159 (1.71) 0.018 (0.25) 0.160 (1.36) 0.178 (2.15) -0.026 (0.22) 0.093 (2.13) Gender of head (male=1) 0.121 (3.93) 0.121 (2.73) 0.097 (1.61) 0.091 (1.27) 0.068 (0.64) 0.123 (4.35) Government job -0.142 (1.01) -0.060 (0.75) -0.142 (1.58) -0.171 (0.86) 0.124 (0.94) -0.090 (1.56) SOE job -0.462 (4.19) 0.104 (0.56) -0.087 (0.37) -0.216 (2.06) 0.174 (1.05) 0.036 (0.28) Educaton of head -0.006 (0.78) 0.011 (2.48) -0.000 (0.05) -0.001 (0.18) 0.028 (1.40) 0.006 (1.58) Educaton of other adults 0.004 (1.52) 0.004 (1.60) -0.001 (0.20) 0.007 (2.79) 0.009 (2.09) 0.004 (2.18) Share of good qualty nonrrgated land -0.032 (0.38) -0.047 (0.81) 0.032 (0.50) -0.058 (0.63) 0.005 (0.06) -0.009 (0.27) Share of good qualty rrgated land -0.256 (2.21) -0.001 (0.01) -0.088 (0.84) 0.118 (1.59) 0.271 (1.94) -0.063 (1.23) Prvate rrgated x 10 3 0.051 (0.61) 0.249 (1.57) 0.275 (1.92) -0.020 (0.18) 0.051 (2.56) 0.058 (2.44) Prvate non-rrgated x 10 3 0.077 (0.78) 0.111 (4.04) 0.195 (2.06) 0.056 (0.92) 0.080 (7.34) 0.042 (1.88) Prvate perennal x 10 3-0.031 (0.063) 0.015 (0.016) -0.139 (1.29) 0.092 (1.11) 0.044 (2.00) 0.024 (2.04) Prvate water surface x 10 3 0.334 (2.72) 0.027 (0.52) -0.043 (0.31) -- 0.041 (5.45) 0.059 (3.86) Swdden land dummy varable -0.149 (2.37) 0.266 (6.75) 0.242 (1.85) 0.122 (0.88) 0.171 (3.09) 0.064 (0.94) Commune dummy varables Yes Yes Yes Yes Yes Yes Constant 2.938 (6.97) 2.793 (5.57) 2.067 (3.68) 4.235 (8.68) 2.165 (2.56) 2.615 (7.82) R² 0.631 0.461 0.435 0.548 0.438 0.490 RMSE 0.472 0.390 0.454 0.420 0.610 0.483 No. observatons 432 790 459 269 308 2,361 Note: The dependent varable s the log change n annual agrcultural allocated land between 1993 and 1998. Absolute t-ratos n parentheses are based on standard errors corrected for heteroskedastcty and clusterng. Unless otherwse noted, all varables are ntal 1993 values. 19
We also tested for effects of the ntal effcency of land allocaton on the probablty of becomng landless (n terms of allocated annual land). Table 4 gves the proporton of the 1997/98 sample that had no allocated land classfed by the estmated ntal loss relatve to the effcent allocaton n 1992/93. The hgher the loss relatve to the effcent allocaton the hgher the probablty of havng no allocated land n 1997/98. Table 4: Dsposal of allocated land % landless n 1998/99 % landless 1 (Ganed relatve to the effcent allocaton) Quntles of households ranked by the loss from admnstratve allocaton of land, 1992/93 4.6 [477] 2 2.6 [537] 3 5.9 [579] 4 10.7 [533] 5 (Lost relatve to the effcent allocaton) 16.4 [433] 7.7 [2559] Note: % of households havng no allocated annual agrcultural land n 1997/98; total number of sampled households n [.] We also estmated probts for landlessness usng the same regressors as n Table 3. We dd ths for both dsposal of allocated annual land and dsposal of all cultvated land. Vrtually the only sgnfcant predctors n any of these regressons was the proportonate effcency loss, whch had a sgnfcant postve coeffcent n most cases, and geographc dummy varables. Becomng landless was more lkely for households who had too lttle land relatve to the effcent allocaton, and t was more lkely n the south than the n north. Our results are suggestve of a land polarzaton process among those who started off wth too lttle land relatve to the effcent allocaton. The bulk of these households traded up, acqurng more land n the more market-orented economy. However, a mnorty smply dsposed of ther allocated land. The results n Table 4 are suggestve of an nterpretaton n whch a subset of those households who started out wth too lttle land (relatve to the effcent allocaton) smply cashed n, possbly to take up other non-farm actvtes or pay off debts. The dfference n behavor of those households who dsposed of ther allocated land rases a concern about the possblty of sample selecton bas n our man 20
regressons for land re-allocaton. 12 In fact there are two possble sources of such bas. The frst stems from the fact that our preferred specfcaton for the functonal form entaled that some observatons had to be dropped; the second s panel attrton, n that some of the orgnal random sample could not be ntervewed n the second survey for varous reasons (they had left ther orgnal address or they chose not to partcpate agan). Motvated by the approach to testng for panel attrton bas n Ftzgerald, Gottschalk and Mofftt (1998), we tested for both sources of bas usng ntal land allocaton as the auxlary endogenous varable n a probt for whether a household dropped out of the sample (for ether reason), wth controls for all other observable exogenous characterstcs n the baselne survey. (We used the same set of controls as n our model of land re-allocaton.) Ths assumes that the ntal land allocaton s correlated wth the selecton-bas error component n the man regressons but does not appear on the RHS of our model of land re-allocaton ndependently of the ntal effcency loss; the latter excluson restrcton s mpled by our theoretcal model (as dscussed n the prevous secton). The ntal land allocaton varable was statstcally nsgnfcant (at the 10 percent level) natonally and for all regons, suggestng that there s lttle or no bas due to sample selecton n our regressons for land re-allocaton. 6. Conclusons The standard polcy prescrpton for transformng a socalst command economy nto a market economy s to prvatze productve assets and then change the law to permt free transactons n those assets. We have put ths model to the test n the context of Vetnam s agraran transton. We fnd some support for the standard model durng a perod that ncluded major lberalzng reforms to land laws. There are sgns that land allocaton responded to the neffcences of the ntal admnstratve assgnment at de-collectvzaton. Households who started wth an neffcently low (hgh) amount of crop land under the admnstratve assgnment tended to ncrease (decrease) ther holdngs over tme, through the process of re-allocaton allowed under the new land laws. The partal adjustment coeffcent was about 1/3 n the aggregate, meanng that one thrd of the ntal gap between the actual allocaton and the effcent allocaton was elmnated wthn fve years. 12 It mght be conjectured that ths explans why we get a better ft usng the log dfference specfcaton; snce the observatons that dsposed of ther allocated land behaved very dfferently to dfferences n the nta l neffcency of ther allocaton, droppng these (because one cannot take the log of zero) mproved the ft. However, we got a better ft wth the log specfcaton across the same (truncated) sample when compared to other values of η (tested at 0.1 ntervals over the [0,1] nterval). 21