Breaking up the Collective Farm:

Transcription

1 Breakng up the Collectve Farm: Welfare Outcomes of Vetnam s Massve Land Prvatzaton Martn Ravallon and Domnque van de Walle 1 World Bank, 1818 H Street NW, Washngton DC, 20433, USA 12 November, 2001 Abstract: The de-collectvzaton of Vetnamese agrculture was a crucal step n the country s transton to a market economy. But the assgnment of land-use rghts had to be decentralzed, and local cadres ostensbly had the power to capture ths process. We assess the realzed land allocaton aganst explct counter-factuals, ncludng the smulated allocaton mpled by a compettve market-based prvatzaton. Dependng on the regon, we fnd that 95-99% of maxmum aggregate consumpton was realzed by a land allocaton that generated lower nequalty overall, wth the poorest absolutely better off. We attrbute ths outcome to ntal condtons at the tme of reform and actons by the center to curtal the power of local eltes. Key words: Decentralzaton, land reform, prvatzaton, equty-effcency trade-off, Vetnam JEL codes: D60, P21, Q15 1 Correspondence: mravallon@worldbank.org, dvandewalle@worldbank.org. These are the vews of the authors, and need not reflect those of the World Bank or any afflated organzaton. Helpful comments and assstance were receved from Dorothyjean Cratty, Erc Edmonds, Karla Hoff, Jean- Jacques Laffont, Mchael Lpton, Dlp Mookherjee, Vjayendra Rao, Tomom Tanaka, Carre Turk, Chrs Udry and partcpants at presentatons at Unversty of Mchgan, Mchgan State Unversty, Yale Unversty and the World Bank. The support of the World Bank s Research Commttee s gratefully acknowledged.

2 1. Introducton Vetnam s land reform of 1988 abandoned the collectve farmng system that had been ntroduced n the 1960s. The 1988 Land Law and ts key mplementaton drectve Resoluton 10 gave ndvdual households long-term use rghts over the collectves land and other resources. 2 Four mllon hectares of land were thus scheduled for effectve prvatzaton. The economc sgnfcance of ths new land law s obvous, gven that (around 1990) three-quarters of the country s workforce depended drectly on farmng. Clearly, ths land prvatzaton was hugely mportant to lvng standards and ther dstrbuton n Vetnam. Implementaton of Resoluton 10 was decentralzed to commune level; there was lttle choce, snce the center could not control the local authortes, who were (naturally) much better nformed about local condtons. So the center faced an accountablty problem n ths decentralzed reform. 3 Malarney (1997, p.900) descrbes the problem faced by the reformers:..gven the nsttutonal domnance of the Communst Party, local poltcans wth party backgrounds, whch s to say all, are compelled by the party to be mpartal and commtted to offcal polces; yet, as poltcans drawn from local kn and communty, they are also pressured to nurture nterpersonal relatons, selectvely avod offcal dctates, and use ther postons to brng advantages to kn and/or co-resdents. Ths echoes concerns n recent lterature and polcy dscusson about decentralzed development programs (Bardhan and Mookherjee, 2000; Galasso and Ravallon, 2001). In developng country settngs, the center often faces hgh costs of acqurng the nformaton needed to control outcomes locally, and local agents may well have lttle sympathy wth the center s ams. So there s a real rsk that a decentralzed program wll be captured by a local elte that sees the nterventon as an opportunty to enhance ts own poston. 2 What s wdely known as the 1988 Land Law, was n actual fact enacted December We wll follow the conventon of callng t the 1988 Land Law. Resoluton 10 followed n Aprl Local accountablty problems were apparently also common n the 1960s collectvzaton program (Fforde, 1989) 2

3 Ths paper assesses the assgnment of land-use rghts acheved by Vetnam s decollectvzaton. We use two counter-factuals for assessng mpact. One s an equal allocaton of (qualty-adjusted) land and the other s the allocaton that would have maxmzed aggregate consumpton, as would have been acheved by a compettve market-based prvatzaton under deal condtons. Comparng ths wth the actual allocaton allows us to estmate the mplct value that was placed on effcency versus dstrbutonal goals n the allocaton of land. We also characterze the specfc dstrbutonal outcomes of the realzed land allocaton; possbly effcency was sacrfced, but the poor would have been better off f t had not been. The followng secton descrbes Vetnam s de-collectvzaton, and the factors leadng up to t. Secton 3 outlnes our approach n theory, whle secton 4 descrbes our emprcal specfcaton. Secton 5 dscusses our survey data, collected 2-3 years after the reform was completed. Secton 6 presents the regressons and secton 7 dscusses ther welfare mplcatons. Secton 8 concludes. 2. Prvatzng the collectve s land Wth the am of rasng agrcultural productvty, Vetnam s 1988 Land Law and Resoluton 10 abandoned collectve farmng and granted households long-term use rghts over land and the freedom to cultvate t as they wshed. 4 Land remaned the property of the State, revertng back to the authortes when a household moved or stopped farmng. 5 The decollectvzaton was vrtually complete by 1990 (Ngo 1993). 4 Use rghts for crop land were granted for 10 to 15 years; longer perods appled to tree crops. Some flexblty was allowed n that 10-15% of the cooperatve s land could be kept asde for new households and demoblzed solders, and avalable for hre by households n the meantme (Tran 1997). 5 Although Resoluton 10 affrms the rght to transfer land use and legate t to one s offsprng, such rghts were not fully guaranteed legally (Bloch and Oesterberg 1989). It dd not recognze the rght to exchange, lease, or mortgage land. These rghts were only extended n the 1993 Land Law. Land polces have evolved snce, but t s the mpact of the 1988 Land Laws that s our man nterest n ths paper. 3

4 The new Land Law made recommendatons on how the land was to be allocated across households. It recommended that allocatons take account of the avalablty of land resources, on households labor force and the land that households had been cultvatng. It also placed certan lmts on how much land could go to any one household. 6 Resoluton 10 acknowledges clams to the land farmed pror to collectvzaton. It also entreats the cooperatves to provde approprate jobs and good arable land to the famles of war heroes and martyrs, to those who sgnfcantly contrbuted to the revoluton, to the njured and unable-boded and to others facng consderable dffcultes. But t then dlutes ths request by addng that the well-beng of these groups s really the responsblty of the local Peoples Commttees and that the Mnstry of Labor, War Invalds and Socal Affars and the Mnstry of Fnance wll devse polces on socal assstance to them (Vetnam Communst Party 1988). Whle the new land law extended some gudelnes, t left local cadres wth consderable power over land allocaton and the condtons of contracts. The center s drectves were dssemnated by Provncal Peoples Commttees, who n turn reled on the local authortes, apparently allowng them wde berth n adaptng the gudelnes to local condtons, prortes and customs. One can expect foot draggng on ther part, and the pursuance of qute dfferent objectves n mplementng the central drectves. Those who were makng the decsons locally were essentally the same cadres who had postons of relatve prvlege as the managers of the cooperatves, and relatvely hgh lvng standards under the collectve mode of agrcultural producton (Selden, 1993). The reform threatened to undermne ther power and prvlege. 6 Artcle 27 of the 1988 Land Law stpulates that the land allocaton to each household should not exceed ten percent of the total farm land area of each concerned vllage. It further decrees regonal per capta land celngs for those contractng land for long term use from state operated farms. It has been clamed that celngs were offcally set at two hectares n the fertle but densely populated Red Rver Delta and three hectares n the South (ANZDEC 2000), though we fnd no menton of ths n Resoluton 10 or the 1988 Land Law. 4

5 So there was a real rsk here that the benefts of reform would be captured by selfnterested local cadres, potentally undermnng the center s ams. There s anecdotal evdence of abuse of local power, aganst the center s nterests. Kolko (1997, p. 92) argues that: From ts ncepton, the land redstrbuton was marred by conflct, ambguty and corrupton. Cadres n many vllages mmedately began to dstrbute the best land to ther famles and relatves, and abuse was rfe. There were numerous publc dsputes at the tme, stemmng from (amongst other thngs) conflctng hstorcal clams over land, dsputes over vllage and commune boundares and complants about corrupt party cadres (Nguyen 1992; Pngal and Xuan 1992; Kolko 1997). It has also been argued that those wth the weakest pror clams on plots dd poorly n the land allocaton. For example, Ngo (1993) argues that war veterans, demoblzed solders and ther famles were short-changed n the land allocatons and were over-represented as protagonsts n dsputes. It s unmagnable that such an enormous land reform was free from corrupton. However, the nterpretaton of the exstng qualtatve evdence on ths ssue s unclear. Cases of extreme abuse of power by local eltes were vsble when they boled up n local protests Vetnam s hot spots (Kolko 1997, ctes many examples). However, the fact that local protests were possble can also be nterpreted as evdence that there were constrants on local abuse of power. The possblty for bas n the hstorcal-qualtatve account s clear; the cases of abuse may well have been uncommon but far more vsble. Objectve vllage-level assessments were rare; n the only vllage study we know of to address ths ssue, Tanaka (2001) descrbes the elaborate efforts of the land allocaton commttee n a North Vetnamese vllage to equalze land allocaton. Such efforts are unlkely to have attracted much publcty at the tme. Whle one would not want to generalze from one vllage study, t s no less hazardous to nfer from the avalable evdence that capture by local eltes was the norm. 5

6 There clearly were constrants on the power of the cadres, n part due to actons by the center. Formally, Artcle 54 of the Land Law extends the threat of punshment for offcals found to have abused ther power n the allocaton process. Enforcement s, of course, another matter. However, t s mportant to note that Vetnam s peasants had sacrfced heavly through long perods of war, n the hope of a better lfe after reunfcaton. The outcomes for ther lvng standards had been dsappontng. Peasant resstance to the collectve system was common n the 1980s, and has been dentfed as a factor motvatng the center s de-collectvzaton reforms (Beresford 1985, 1993; Selden, 1993; Kerkvlet 1995). Wth the center s support, the Vetnam Peasant Unon (VPU) was created n 1988 wth the explct am of gvng peasants a stronger voce n reform polces and mplctly at least promotng the center s reforms locally. As wth past peasant unons, t seems that the VPU was eventually captured by local eltes; Wurfel (1993, p.32) argues that by 1990 the VPU had been tamed by local party cadre, who had nterests to protect. But for a crtcal perod the VPU appears to have acted as a counter-weght to the cadres (Wurfel, 1993). Durng the reform perod, the center also gave greater freedom to the press; the press subsequently carred much crtcsm of the bureaucracy, agan helpng the reform process (Wurfel 1993). The reform movement was clearly drven by more than the center s concerns about the welfare of peasants. The same neffcences of the collectve farmng system constraned the resources avalable to the center for ts ndustralzaton plans, and created food shortages n urban areas (Beresford 1993; Kerkvlet 1995). Arguably then, the reforms were only possble through an mplct coalton between the peasants and reformers at the center a coalton that clearly amed to constran the power of local cadres to capture the process. Recent hstory provded a reference pont n decdng how the land should be allocated. 6

7 Collectvzaton came soon after the completon of land reform programs that had gone a long way toward redressng the hgh nequalty of land ownershp under French Colonal rule (Beresford 1985; Pngal and Xuan 1992). The pre-collectvzaton allocaton may have nfluenced land allocaton at de-collectvzaton. There are reports that some households smply went back to farmng the land they had orgnally handed over to the cooperatve or collectve, or land they had some hstorcal clam to. 7 Whle there was no legal commtment to restore the precollectvzaton land allocaton, that was an opton for the local authortes. There were mportant dfferences between the North and the South at the tme of the reform. In the South's Mekong Delta, farmers had ressted collectvzaton, and by the tme Resoluton 10 was ntroduced less than 10 percent of all the regon s farmers had been organzed nto collectves. In contrast, vrtually all of the crop land n the North and n the South's Central Coastal provnces where jonng the collectves was seen as a means of rebuldng after the war was collectvzed by the tme of the reform (Pngal and Xuan 1992; Ngo 1993). Southern Vetnamese farm households who partcpated dd so for a much shorter perod, whle many never partcpated n the collectves, notably n the Mekong Delta. However, the land allocaton n the South was stll admnstratvely determned and perodcally re-allocated (Pngal and Xuan 1992); the dfference wth the more collectvzed North s that n the South (especally the Mekong Delta) farmers contnued to farm ndvdually rather than collectvely. Pror to reunfcaton, agrcultural land n the South had also undergone a seres of land reforms. 8 Resoluton 10 allowed farmers n the South to recover land owned pror to 1975, though former 7 Smth and Bnh (1994) quote a number of Son La households n the North as professng, n 1994, to be farmng the same land they had at the tme of the departure of the French. Tran (1997) clams that land was redstrbuted accordng to household orgnal contrbutons to the cooperatves n some areas. 8 The South s land reform programs pror to reunfcaton had ntally conssted of lease prce control and ownershp celngs, but were followed n 1970 by substantal land redstrbuton and ttlng under a land-to-the tller program (Callson 1983; Pngal and Xuan 1992). 7

8 landlords" were explctly barred from dong so (Pngal and Xuan 1992). There are reports that n the Mekong Delta the mplementaton of Resoluton 10 often entaled restorng the land allocaton that prevaled pror to reunfcaton (Hayam, 1993; ANZDEC, 2000). The collectves had also owned and controlled the farm captal stock (tools, machnery, draft anmals) that also had to be allocated among farm households. It s sometmes clamed that ths process more easly allowed cooperatve offcals to favor themselves, ther famles and frends than the more vsble land allocaton process. Because of data lmtatons we can only focus on land allocaton, though we wll look for sgns of dfferng returns to land assocated wth poltcal connectons. 3. Theoretcal model of the actual and counter-factual land allocatons Motvated by the above observatons we shall test whether the local mplementaton of decollectvzaton served dstrbutonal goals possbly reflectng capture by local eltes at some loss to aggregate consumpton. We construct a model that allows us to estmate that loss, and to compare the observed allocaton aganst explct counter-factuals. One of those wll be an equal allocaton of land per capta. Ths s easly calculated. The other s the allocaton that maxmzes aggregate consumpton of the commune; ths requres a model of consumpton. The actual decson-makng process mght be anythng from admnstratve fat (accordng to the cadre s personal preferences) to a complex barganng game. We only assume that the outcome (however t s reached) s Pareto optmal, n that no commune member s utlty can be ncreased wthout someone else beng worse off. To characterze all possble solutons, we represent the problem as maxmzng a weghted sum of welfare levels across all farmhouseholds. The Pareto weght attached to the utlty of household s w = w X ) where X s a vector of exogenous household characterstcs. Naturally, dfferent weghtng functons mply ( 8

9 dfferent dstrbutons of land and utlty. If the weghts tend to be negatvely (postvely) correlated wth welfare (to be defned) then one can say that the outcome wll tend to be propoor ( pro-rch ). The utlty of the th farm-household s assumed to depend solely on ts consumpton of a composte commodty. The household receves L of land, whch yelds an output of F ( L, X ). (For now we treat land as homogeneous; n the emprcal work we allow for observable heterogenety, and we consder the consequences of latent heterogenety n the next secton.) The household also has (postve or negatve) non- farm ncome, Y X ). 9 (At the tme of the reform, and snce, agrcultural labor markets were vrtually non-exstent n Vetnam, so to smplfy the exposton we close off ths market n our model.) The household s consumpton s then: C = C L, X ) = F ( L, X ) + Y ( X ) (1) ( We assume that the functon F s ncreasng and strctly concave n L. Utlty s n turn an ncreasng concave functon of consumpton, U = U C ). ( ( The commune selects an allocaton of the total avalable land wth mean L. The observed land allocaton s: n L across n households, n ( L1,.., Ln ) = arg max[ w( X ) U[ F( L, X ) + Y( X )] L = nl] = 1 n = 1 (2) whch solves: w X ) U ( C ) F ( L, X ) = µ for = 1,.., n (3) ( L where F L, X ) s the margnal product of land and µ s the shadow prce of land n the L ( 9 The functons F and Y may vary between households n dfferent communes. We wll deal wth ths possblty n the emprcal work by ncludng a complete set of commune effects n the regressons. 9

10 commune (the Lagrange multpler on the aggregate land constrant n equaton 2). It s readly verfed that the land allocaton s strctly ncreasng (decreasng) n X j (the j th element of the vector X ) f the sum of the elastctes of w ( X ) and U ( C) FL( L, X ) to X j s postve (negatve). Compare ths to the allocaton that maxmzes the commune s aggregate consumpton: * * n ( L1,.., Ln ) = arg max[ C( L, X ) L = nl] = 1 n = 1 (4) We call ths the consumpton-effcent allocaton. Ths equates C ( L, X ) = F ( L, X ) wth the multpler λ on aggregate land n (4), gvng * X Mean consumpton s then: L = L(, λ) for =1,..,n (5) C * n = C( L, X ) / n (6) = 1 * L * L * The consumpton loss from the actual allocaton s then * C C where C s the actual mean. The consumpton-effcent allocaton s also the compettve equlbrum gven ( X 1,.., X n). In a market-based land allocaton, each household s consumpton wll be F( L, X ) + Y ( X ) λl where λ s the market prce of land. Demands equate F L, X ) = λ L( over all, whch s the allocaton that maxmzes aggregate consumpton under the non-market allocaton. Naturally the market soluton wll also vary wth the jont dstrbuton of the Xs. There are some caveats to ths nterpretaton of the consumpton effcent allocaton. A compettve market s unlkely to have been a feasble opton at the tme n Vetnam. For one thng, agrcultural land markets were vrtually non-exstent. And other markets (notably for credt) and nsttutons (for property rghts enforcement) were probably not functonng well 10

11 enough to assure an effcent market-based prvatzaton of land. However, aganst these observatons, t should also be noted that under communsm very lttle moblty had been allowed between communes. People may well have been suffcently well nformed wthn each vllage to know f one famly attached an apprecably hgher value to extra land than another, even though a market dd not exst. It should also be noted that f holdng land gves utlty ndependently of consumpton then the compettve market allocaton of land wll dffer from the consumpton-maxmzng one. For example, f land provdes nsurance aganst rsk then t wll have value ndependently of current consumpton. Then our nterpretaton of the consumpton-maxmzng allocaton as the market soluton would also requre that rsk markets worked perfectly. 10 Snce we have no bass for assgnng a value to land ndependently of the current consumpton t generates we cannot calculate a condtonal market soluton (condtonal on other market falures). Nonetheless, the consumpton-maxmzng allocaton remans a natural benchmark for assessng the realzed allocaton. It wll be of nterest to see how close the non-market allocaton s to ths benchmark. 4. Emprcal model In our emprcal mplementaton of the above model we make the followng assumptons: Assumpton 1: Utlty s gven by log consumpton: U C ) = ln C( L, X ) (7) ( Assumpton 2: Log consumpton s gven by: ln C = α + β ln L + X γ + ε (8) where 0 < β < 1 and ε s a zero-mean..d. error term uncorrelated wth ln L and X. 10 To gve another example, Burgess (2000) argues that food market falures exst n rural Chna such that holdng more land reduces the shadow prce of food. 11

12 Assumpton 3: The welfare weghts take the form: ln w = X b + υ (9) where υ s a zero- mean error term uncorrelated wth Assumptons 1-3 mply that the land allocaton satsfyng equaton (4) can be wrtten n explct form as the regresson model: 11 X. ln L = ln( β / µ ) + X b + υ (10) Ths dentfes drectly the parameters of the mplct welfare weghts of the local land-allocaton authorty. Substtutng (10) nto (8) generates the reduced form equaton for consumpton: ln C = α + β ln( β / µ ) + X ( βb + γ ) + ε + βυ (11) The consumpton-maxmzng allocaton by contrast s gven by: 12 α + ln( β / λ) γ ε ln * X L = + + (12) 1 β 1 β 1 β Comparng (10) and (12), t can be seen that f γ /( 1 β ) = b then the actual allocaton responds to changes n X the same way as the consumpton-effcent allocaton. So f the two allocatons are essentally the same then we should be able to accept the restrcton that b = β b + γ when mposed on the reduced form equatons, (10) and (11). If we cannot accept ths restrcton then t * * * 1 2 n s of nterest to calculate the consumpton-effcent land allocaton, ( L, L,..., L ), from whch we can then measure the dstrbuton of consumpton losses mpled by the actual allocaton, usng the fact that the proportonate consumpton loss for household s ( L * / ) 1. L β 11 To verfy the followng equaton, take logs through (3) and note that note that (7) and (8) mply that U (.) F L = β / L. Equaton (10) then follows usng (9). 12 Gven (8), the consumpton-effcent allocaton to household solves ln L = ln( β / λ) + ln C * where ln C = α + β ln L + X γ + ε. * * * 12

13 Whle allocated land s endogenous n ths model, t s taken to be exogenous to consumpton (.e., Cov(υ,ε )=0). Ths s a standard assumpton n past emprcal work for Vetnam and n other settngs n whch land allocaton s done admnstratvely rather than through markets. 13 The assumpton can also be defended on the grounds that the land allocaton preceded the survey-based consumpton measure by 3-4 years. Our estmates of the parameters of equaton (8) wll be based f there are omtted varables that jontly nfluence the welfare weghts and consumpton levels. The most serous concern n ths respect s heterogenety n land qualty. Hgher land qualty wll probably result n hgher consumpton at gven land quantty. Assumng that the qualty dfferences are publc knowledge wthn the commune, the admnstratve land allocaton wll take them nto account, wth more land beng used to compensate for lower qualty. We wll nclude avalable controls for dfferences n the average qualty of land holdngs. However, latent heterogenety wll create a negatve correlaton between the error terms n the estmated consumpton equaton and the land allocaton equaton (Cov( υ,ε )<0). Notce that our test for systematc dfferences between the effcent and actual land allocatons s robust to heterogenety n land qualty. Our test s based on the reduced form coeffcents n (10) and (11); t does not requre the (potentally based) parameters of (8). Nonetheless, our estmates of the parameters of the mplct equaton for the effcent allocaton (equaton 12) do requre the parameters of the structural model n (8). So bas due to latent heterogenety n land qualty wll contamnate our estmates of the effcent allocaton. In prncple ths could be dealt wth by ntroducng an nstrumental varable that 13 See for example Wens (1998) and van de Walle (1998). The assumpton s also a standard one n emprcal work for Chna, where a large share of crop land s also allocated admnstratvely; see for example, Burgess (2000) and Jalan and Ravallon (2001). 13

14 nfluences land allocaton but not consumpton condtonal on land,.e., at least one element of the parameter vector b n (9) would have to be set to zero, whle leavng the correspondng element of γ unrestrcted. However, there s no theoretcal bass for such an excluson restrcton; anythng that can be ncluded from our data set could presumably have been observed or antcpated by the local authortes. It should also be noted that whle there s lkely to be heterogenety n land qualty across plots wthn communes, the scope for land fragmentaton combnng land from dfferent plots when formng a package for each household means that the varance across households n the average qualty of ther allocatons can be consderably less than the underlyng nter-plot varance. In the only evdence on ths pont that we know of, Tanaka (2001) fnds that plot fragmentaton n North Vetnamese vllages was used to produce land parcels of relatvely even qualty. Then heterogenety n land qualty would not be a problem for our analyss. 5. Data Our data are from the Vetnam Lvng Standards Survey (VNLSS) of 1992/93. Ths s one of the natonal, mult-purpose, surveys sponsored by the World Bank under the Lvng Standards Measurement Study (LSMS). 14 The VNLSS follows establshed LSMS practces (World Bank 1995). Our sample s the 2810 rural farmng households n the VNLSS wth complete data. Some 400 households had to be dropped due to mssng data on key varables. There are also 419 households n the rural farmng sample wthout any allocated rrgated or nonrrgated agrcultural land dentfed n the survey. Our readng of the lterature and casual observatons suggest that t s unlkely that there s genune censorng, such that some farmng 14 The VLNS s publc access, subject to standard condtons. For further nformaton on the LSMS see 14

15 households were delberately left out of the land prvatzaton. Under that assumpton, we focus solely on the sample of farmng households wth complete data. Table 1 gves summary statstcs on the varables we wll use from the data set, by regon. Household consumpton ncludes the value of consumpton from own producton, mputed expendtures on housng and the deprecated value of consumer durables. It s deflated by a monthly prce ndex to allow for temporal varaton n household ntervews and by a spatal prce ndex to take account of regonal prce varaton (World Bank, 1995). Geographc heterogenety across communes s to be expected, f only because of dfferences n the shadow prce of land ( µ n equaton 10). Whle the sample sze does not permt estmaton of a separate model for each commune, all regressons ncluded a complete set of commune dummy varables. And all parameters are allowed to vary regonally. Vetnam s routnely dvded nto seven regons reflectng geographcal and hstorcal smlartes. We conduct the analyss both natonally and separately for the Northern Uplands (NU), Red Rver (RR), North Coast (NC), Central Coast (CC) and Mekong Delta (MD) regons. 15 Wthn annual crop-land, the survey dentfes fve land types: () Allocated land: Ths s the land allocated to households by the cooperatve or productve group under Resoluton 10; ths accounts for the bulk of the North's crop land; () Long term use land: Predomnant n the South, ths dffers from allocated land only n that the farmer owes no contracted output (n addton to oblgatory taxes for all allocated land) to the cooperatve or productve group that allocated the land; () Auctoned land: Ths refers to a part of the cooperatves' land reserved for bddng by households, wth a three to fve year tenure dependng on the regon; (v) Prvate land: Ths 15 In the Central Hghlands regon, land s mostly perennal. In the South East there were too few observatons n the sample; after excludng non-farmng households and those wth mssng data we are left wth a sample of only 99 observatons n the South East. 15

16 conssts of land nherted and used by households as a garden area, as well as an area equal to 5% of the commune s agrcultural land that has been handed to households for ther prvate use. Ths land requres no payment; and (v) Sharecropped or rented land. What we refer to here as allocated land s annual crop land, ether rrgated or unrrgated, whch s defned as ether allocated land by the survey respondents or long-term use land. Ths ncludes all allocated land, ncludng any that s not actually cultvated by the household. There s also an allocaton mechansm for perennal, forest and water surface land. However, these other land types followed a much slower and haphazard allocaton process so that we lmt our analyss here to allocated annual rrgated and non-rrgated cropland. We aggregate rrgated and non-rrgated land usng regon-specfc weghts to obtan rrgated land equvalents. To calculate the weghts, we estmated regon-specfc regressons of farm proft on total rrgated and non-rrgated annual crop land, perennal, forest and other land amounts (ncludng swdden, bald hll and newly cleared land), and commune effects. 16 Controls were also ncluded for household characterstcs (the head's relgon, ethncty, age and age squared and whether born locally; household sze, the share of male adults n the household, the years of prmary schoolng of the head and of other adults and a dummy for whether the household s a socal subsdy benefcary). The rato of the coeffcents on non-rrgated to that on rrgated land was then used as the weght on non-rrgated land to recalculate an allocated rrgated land equvalent quantty for each household. The weghts seemed plausble We exclude water surface land from the farm profts regressons because we are unable to adequately calculate net profts from water surface land. The questonnare does not allow a separaton of expenses ncurred n rasng water products from that of rasng lvestock, and assumptons must also be made about consumpton from own producton. 17 Our estmated weghts for non-rrgated land are for the natonal sample, for the Northern Uplands, for the Red Rver, for the North Coast, for the Central Coast and for the Mekong Delta. On the measure of farm proft see van de Walle (1998). 16

17 The survey asked respondents to assgn ther total annual crop land nto the categores good, medum and poor qualty. Unfortunately, the questonnare desgn does not allow us to separately dentfy qualty for allocated land versus other land types. So we cannot use these qualty assessments n calculatng our measure of allocated rrgated land equvalents. These qualty assessments are problematc from other ponts of vew. The categores are probably qute well-defned wthn communes, but are unlkely to be comparable between communes. Nor can t be assumed that they would account fully for omtted heterogenety n land qualty n our man results. The exogenety of these land qualty varables s also questonable. Aganst these consderatons, excludng these varables adds to the aforementoned concerns about omtted heterogenety n land qualty. So we chose to nclude each household s proportons of good rrgated and non-rrgated land n the consumpton and land allocaton regressons, as controls for qualty. 18 We also tested robustness to droppng these varables. We treat prvate land n a specal way. As can be seen n Table 1, land classfed as prvate s not neglgble and falls under all usages. The category s clearly broader than resdental or garden area. Ths type of land has typcally been wth the household for a long tme and the amounts of ths land were clearly known at land allocaton tme. So t s reasonable to treat the amounts of ths type of land as exogenous explanatory varables. We treat all other land as endogenous, so that t does not appear n the model. Our data were collected fve years after the 1988 Land Law (though pror to the 1993 Land Law). In tryng to explan the allocatons we want to use varables that reflect the stuaton around We have no explct nformaton on the methods for allocatng land use rghts n 18 Very few households reported that they had bad qualty rrgated land or good qualty nonrrgated land. So we aggregated the categores nto two; by good qualty non-rrgated land we mean good or medum qualty. 17

18 the communes. As we have noted, Resoluton 10 left ths qute vague. Some observers menton that household sze was taken nto account (Ngo, 1993; Hayam, 1993), whle n other cases t seems that an effort was made to take nto account avalable labor. 19 Our demographc varables nclude household sze and the dependency rato. Household sze s that reported n the 1992/93 survey mnus all members younger than sx years of age. The dependency rato s one mnus the rato of labor age members (between 20 and 65 for men and 20 and 60 for women) to all household members mnus those aged less than sx years. We nclude dummes for the gender of the head, whether he/she was born locally, whether he/she reports practcng the Chrstan or Buddhst relgons as opposed to no relgon, anmsm or other and for whether the head of household belongs to an ethnc group other than the majorty Knh or the relatvely well-off Chnese ethnc groups. We nclude a dummy varable for whether the household reports cultvatng swdden land. Ths ams to capture a ethno-cultural partcularty of those who practce shftng cultvaton. Snce at least the sxtes, the government has pursued polces to sedentarze such groups by apportonng land to them (Bloch and Oesterberg 1989). Resoluton 10 also states that practcal measures should be adopted to promote permanent agrculture and settlement. One mght therefore expect these households to get more allocated land as a result. We also nclude a dummy varable for whether a household contans a handcapped adult of labor age. 20 The latter could nfluence the land allocaton decson negatvely, through effects 19 For example, Tran (1997) descrbes one local allocaton rule as gvng a full share to members of workng age (defned as 16 to 60 for men and 16 to 55 for women), half a share to those above workng age and n the 13 to 15 age range, and one thrd share to the youngest. Also see Hayam (1993). 20 We create ths varable from those ndvduals for men and for women who sad they dd not work durng the last 12 months, or look for work n the last 7 days, and gave beng handcapped as the man reason. Ths varable may not be measured well as t may exclude those who have been cured snce the land allocaton tme and t may well nclude more recently dsabled adults. 18

19 on productvty. Aganst that, the Vetnamese government has had a number of polces bestowng preferental treatment to the dsabled and those ndvduals and ther famles who suffered n the wars. A handcapped adult mght thus be favored. However, ths varable wll not fully capture the possblty that solders and ther famles were treated dfferently to others as decreed by Resoluton 10 (Vetnam Communst Party 1988) and alleged by Ngo (1993). We test for ths by also addng a dummy varable for whether the household or one of ts members s a recpent of socal subsdy transfers from the government. These transfers are targeted to the dsabled, war wounded and the famles of war heroes and martyrs. Recept of ths transfer appears to be the best way to dentfy such households n our data. There are, however, possble concerns about the endogenety of ths varable (notably f the nonpoor select out of the program). So we dd our analyss wth and wthout ths varable. The survey dd not dentfy members of the Communst Party. However, we do know f a household member worked for the cooperatve, a socal organzaton, a State Owned Enterprse (SOE) or the government for fve years or more, ether n ther prmary or secondary jobs. On a pror grounds t s unclear how these varables would nfluence land allocaton. On the one hand, other sources of employment may ental a substtuton effect, wth the commune allocatng less land to such households. On the other hand, t may well come wth a power effect, whereby households wth such employment also have more power over local decsons (nterpretable as an effect on the welfare weghts n equaton 3). As also noted n secton 2, t s possble that the most egregous abuse and corrupton occurred n the dstrbuton of collectvely owned farmng mplements and draft anmals rather than that of land. If so, we would expect to fnd postve mpacts on consumpton through the returns to land for favored households. We wll test ths by ncludng n the consumpton 19

20 equaton an nteracton effect between land and whether a household member worked for a cooperatve at or pror to de-collectvzaton. Ths s an mperfect test as t allows only for favortsm through household member tes, but ths s the best we can do wth the data. 6. Regressons for consumpton and allocated land For the sample as a whole and each regon, we can convncngly reject the null hypothess (wth probablty less than 5) that the observed land allocaton responded the same way to household characterstcs as the consumpton-effcent allocaton that one would have expected from a compettve market-based prvatzaton, under our assumptons. The reduced form regressons for consumpton and test statstcs for the hypothess that the two allocatons are the same can be found n the Appendx. So we proceeded to estmate the effcent allocaton. Table 2 gves the structural model of consumpton (equaton 8). The results are generally unsurprsng. Household consumpton s a rsng functon of household sze, wth an elastcty less than unty. In most regons consumpton s hgher for households wth a government or SOE job. It s ncreased by hgher household educaton. And consumpton rses wth the amount of allocated land n all regons. Table 3 gves the reduced form equaton for the actual land allocaton (equaton 10) and the estmated parameters of the mpled equaton for the consumpton-effcent allocaton (equaton 12). There s dversty between regons n how much the two allocatons dffer, notably between the North (the Northern Uplands, the Red Rver and North Coast) and South (the Central Coast and Mekong Delta). For example, n the North, the actual allocaton s more responsve to household sze than the effcent allocaton would have been. Ths reverses n the South. The dependency rato sgnfcantly negatvely affects the actual allocatons n the North but not n the South (the CC and MD). The negatve coeffcent on the dependency rato 20

21 ndcates that the admnstratve allocaton n the North put hgher weght on household members who were of prme workng age than the consumpton-effcent allocaton would have requred. In the North (except the Uplands), beng n a mnorty group sgnfcantly ncreases the admnstratve allocaton, but decreases the effcent allocaton (though only sgnfcantly so n the RR). In the other two regons there s less dfference n how ethncty affected the two allocatons. The postve and sgnfcant effect of beng a mnorty household n the northern regons probably captures the fact that the mnortes were gven more land as a result of havng contrbuted more to the collectves orgnally, as allowed by Resoluton 10. Havng a household member wth a government job or n a SOE tended to reduce the admnstratve allocaton, though the effect s generally not sgnfcant. But these characterstcs would have resulted n a hgher effcent allocaton suggestve of greater access to credt and/or productve nputs by these households. Agan there are some regonal dfferences n these effects. For example, there s no sgnfcant effect of SOE on the effcent allocaton n the South; the sgnfcant natonal effect stems from the NU and RR. Admnstratve allocatons responded postvely to male household headshp, and much more so than the effcent allocaton. Generally, educaton of the household head had no sgnfcant effect on the actual allocaton (the sole excepton s n the CC, where hgher educaton reduced the allocaton.) The educaton of others n the household was also nsgnfcant n the actual allocaton. However, the consumpton maxmzng allocaton would have favored households wth hgher educaton, presumably reflectng complementartes between educaton and land productvty. The MD s the one excepton. Recept of a socal subsdy s found to have reduced the actual land allocaton natonally, though ths effect was confned solely to the Mekong Delta. Ths provdes some support for the 21

22 clams that war veterans and ther famles were unequally treated n the land allocaton process n the South. In contrast, we found ths varable to be nsgnfcant n the consumpton equaton for all regons (suggestng that the socal transfer compensated fully for the ncome loss due to war dsablty). The effcent allocaton would have gnored whether or not the household receved socal subsdes. All other results were robust to ncludng ths varable. The practce of cultvatng swdden land ncreased the admnstratve allocaton n the RR and the MD, but not elsewhere. The postve effect n these regons can be nterpreted as a polcy effort to dscourage ths form of land usage (on the assumpton that lack of access to regular crop land encouraged swdden farmng.) The effcent allocaton n the MD would also have gven weght to ths characterstc, but consderably less so than the actual allocaton. When we tested an nteracton term between allocated rrgated land equvalents and a dummy for whether a household member worked for a cooperatve, we found no sgn of any effect on consumpton n the natonal or ndvdual regonal samples. However, n testng the nteracton wth prvate land amounts, we fnd a sgnfcant postve effect of water surface land on consumpton n the natonal sample and n the Red Rver and Northern Uplands. There was also a sgnfcant negatve nteracton effect wth prvate perennal land n the North Coast, and a sgnfcant negatve nteracton effect wth non-rrgated prvate land n the Central Coast, though at the same tme there was a postve nteracton wth prvate rrgated land n that regon. On balance, our results suggest that havng a cooperatve job provded no advantage n dervng benefts from a gven land allocaton, though there are sgns of lmted mpact on the productvty of other land types, notably water surface land n some regons. In the aggregate sample, the proporton of good qualty land (rrgated or not) had no sgnfcant effect on ether the actual or effcent allocaton. Ths holds n all regons except NU 22

23 and RR, where there s an ndcaton that households wth hgher qualty non-rrgated land tended to get lower total land allocatons. Other coeffcents n both equatons were lttle affected by droppng these land qualty varables (gven possble endogenety concerns). 7. Welfare comparsons The frst panel of Table 4 gves varous summary statstcs on welfare outcomes for the actual allocaton, namely mean consumpton and measures of nequalty and poverty. The nequalty measure s the Thel ndex (E(0)), gven by the dfference between log mean consumpton per capta and the mean of log consumpton per capta. The poverty measures are the headcount ndex (% below the poverty lne) and the squared poverty gap ndex (Foster et al., 1984) whch penalzes nequalty among the poor. The poverty lne s from Glewwe et al. (2000) and ams to measure the cost of a set of basc food and non-food consumpton needs. The poverty lne was developed on the same survey and agreed to by the government. The second panel n Table 4 gves the same statstcs for the consumpton-effcent allocaton, for whch we gve mean consumpton and nequalty. The thrd panel s for an equal allocaton, n whch the rrgated land equvalent s equalzed on a per capta bass across all households wthn the commune. Recall that the socalst mode of agrcultural producton had been n place for a shorter tme n the South and that the Mekong Delta, n partcular, had been far less collectvzed than the North and the Central Coast (though stll subject to other controls under socalst agrculture). 21 So the land allocaton n the MD at the tme of de-collectvzaton was undoubtedly more nfluenced by the pre-communst allocaton, as determned by hstorcal land 21 Recall that the Central Coast was probably a somewhat specal case gven that t had been a war zone and so collectvzaton was more easly adopted (Pngal and Xuan 1992; Ngo 1993). 23

24 rghts and pror land reforms (secton 2). So t s notable that, relatve to the consumptoneffcent allocaton, we fnd that the actual allocaton n the MD entaled a greater loss of aggregate consumpton, wth a four percent consumpton loss (Table 4). A seemngly plausble explanaton s that the hstorcal (pre-unfcaton) land allocaton had become less effcent over tme but was nonetheless the more natural fall-back poston n the MD. Ironcally then, t can be argued that the fact that socalst agrculture had been more short-lved n the South meant that the regon could not acheve the potental effcency gans avalable to the North from land reallocaton under de-collectvzaton. The hstory of Vetnam meant that the North was n a somewhat better poston to acheve a relatvely effcent land allocaton. Both the effcent and equal-land allocatons would have resulted n a lower poverty rate than the actual allocaton, though the dfferences are small (two percentage ponts overall). Ths s somewhat deceptve snce we found that the poverty lne turns out to be close to the ntersecton of the cumulatve dstrbuton functons. However, the poverty lnes used here are hgher (n real terms) than the poverty lnes used n Vetnam at the tme of the 1988 allocatons (Dollar and Glewwe, 1998). So t can be argued that poverty ncdence would have been hgher under the effcent allocaton when assessed by the local standards of poverty at the tme. These observatons are renforced by Fgure 1 whch gves scatter plots of the percentage losses from the actual relatve to the consumpton-maxmzng allocaton aganst actual consumpton, and a non-parametrc regresson functon (usng Cleveland s, 1979, local regresson method as programmed n STATA). It can be seen that the losses from the actual allocaton tend to rse wth consumpton, both natonally and wthn each regon. Natonally, mean consumpton gans are about 15% for the poorest, wth losses of about 20% for the rchest (comparng end ponts on the regresson functon n Fgure 1(a)). The mean proportonate gans 24

25 are roughly lnear n log consumpton. The pont where the mean gan s zero s farly close to the poverty lne (ndcated by the vertcal lne). The gans to the poorest are also reflected n the squared poverty gap measures n Table 4, whch are hgher for the consumpton-effcent allocaton. It s evdent from Fgure 1 that there are large dfferences between regons n the condtonal varance of the proportonate losses. In partcular, the relatonshp between welfare losses and consumpton levels s less precse (though stll postve) for the MD, where there are clearly other factors at play n determnng the ncdence of the losses relatve to the consumpton-effcent allocaton. Agan, hstorcal (pre-unfcaton) allocatons are lkely to have had greater nfluence n ths regon. An equal allocaton of land (n terms of ts rrgated equvalent) across all households would have acheved a close approxmaton to the levels of mean consumpton and nequalty observed n the data. There were of course devatons from equal land n practce, but the overall outcomes for the dstrbuton of consumpton were smlar. However, under the equal-land allocaton the poorest are generally better-off relatve to the actual allocatons as evdenced by lower squared poverty gap ndces. It s notable agan that the regon where the equal allocaton dffered most from the actual s the Mekong Delta. It mght be conjectured that the market-based allocaton would have acheved substantally hgher average consumpton f only land could have been redstrbuted between communes. To address ths queston, Table 5 repeats the smulatons reported n Table 4 except that we gnore commune boundares when makng the calculatons. Thus the calculaton entals maxmzng aggregate consumpton over the entre regon subject only to the aggregate amount of (rrgaton-equvalent) land n the regon. In practce ths would of course requre movng 25

26 households between communes, whch was rare n Vetnam. However, ths smulaton gves an dea of how much mmoblty constrans the problem. The maxmum levels of consumpton would of course have been hgher allowng households to be moved between communes, so that only aggregate land endowments at the regonal level matter. The dfference s not large however (comparng Tables 4 and 5). The actual allocaton wthn communes, wthout redstrbuton between them, entaled losses n mean consumpton between one and nne percent as compared to a consumpton maxmzng land allocaton wth redstrbuton allowed. Impacts on poverty also look smlar. The headcount ndex of poverty s lower everywhere but the Red Rver regon, whle the very poorest households would have a worsenng under the effcent allocaton wth moblty across communes. When we compare the outcomes under the actual allocaton wth those resultng from the equalzaton of land at the regonal level, we fnd the losses n consumpton to be slghtly lower rangng from one to eght percent. Ths scenaro shows the largest mpact on poverty. Both the rate and the severty of poverty would be lower under a regon-wde equalland allocaton relatve to the actual land allocaton. Agan the Mekong Delta stands out as havng hgh unrealzed consumpton gans from land re-allocaton. If moblty were possble wthn the regon, the actual land allocaton entals a nne percent loss of aggregate consumpton relatve to the consumpton maxmzng allocaton, and eght percent relatve to an equal allocaton; n both cases ths s about twce the overall mean consumpton loss (Table 5). Lack of moblty under Communsm appears to have come at an unusually large cost n the MD. Ths s consstent wth our casual observatons that household plot szes vary greatly wthn the regon. 26

27 8. Conclusons The heavy relance on decentralzed mplementaton of polcy reforms n developng countres has rased concerns about capture by local eltes whose nterests are not well served by the center s ams. We have tred to see f such concerns are borne out by evdence on how landuse rghts were allocated n practce under the massve reform to land laws ntroduced by Vetnam n Ths reform was arguably the most mportant step n the country s transton to a market-based agrcultural economy after abandonng collectve farmng. We have used a model of household consumpton to assess the dstrbuton of consumpton mpacts relatve to counterfactual allocatons, ncludng the one that would have maxmzed aggregate consumpton, whch would have been the compettve market allocaton under our assumptons. Our results are not consstent wth the pcture that many commentators have panted (based on anecdotal evdence) of an negaltaran land allocaton stemmng from the power of relatvely well-off local cadres to capture the process. In terms of the mpact on average consumpton and consumpton nequalty, the observed allocaton of land n our data was roughly equvalent to gvng every household n the commune the same rrgated-land equvalent. The observed allocaton was sgnfcantly dfferent to what one would have expected from an effcent (consumpton-maxmzng) allocaton, as would be acheved by a compettve prvatzaton at market-clearng prces. The consumpton-effcent allocaton would have put greater weght on educaton (whch rased the margnal utlty of land), and gven less weght to household sze, labor force, mnorty-groups and male heads of household. We fnd no evdence that land allocaton unduly favored households wth government or sem-government jobs; ndeed, the market allocaton would have gven hgher weght to these attrbutes, because such 27