Research Paper No. 2006/137. Juliano Assunção*

Research Paper No. 2006/137 Land Reform and Landholdngs n Brazl Julano Assunção* November 2006 Abstract Land and wealth are closely related n rural Brazl, a country characterzed by hgh levels of nequalty n terms of ncome or landholdngs. After presentng a hstorcal retrospectve of land concentraton and land reform n Brazl, ths study evaluates the mpact of the land reform programme undertaken n the 1990s on land ownershp and land dstrbuton. It s shown that the programme ncreased landownershp among poor rural famles and those wth less educated household heads, reducng the fracton of the other famles wth landholdng. Also, the land reform programme ncreased land nequalty among landowners. Keywords: land reform, land dstrbuton, wealth dstrbuton JEL classfcaton: D31, Q15 Copyrght UNU-WIDER 2006 *Department of Economcs, PUC-Ro Ths study has been prepared wthn the UNU-WIDER project on Personal Assets from a Global Perspectve, drected by Jm Daves. UNU-WIDER acknowledges wth thanks the fnancal contrbutons to ts research programmeme by the governments of Denmark (Royal Mnstry of Foregn Affars), Fnland (Mnstry for Foregn Affars), Norway (Royal Mnstry of Foregn Affars), Sweden (Swedsh Internatonal Development Cooperaton Agency Sda) and the Unted Kngdom (Department for Internatonal Development). ISSN 1810-2611 ISBN 92-9190-921-1 ISBN-13 978-92-9190-921-6

Acknowledgements I gratefully acknowledge the comments from James Daves, Carmen Deere, Patrck Honohan, Branko Mlanovc, John Muellbauer, Danel Waldenstrom and other partcpants of the WIDER research project Personal Assets from a Global Perspectve. Also, I would lke to thank Mchel Azula and Flava Feres for provdng excellent research assstance. All errors are my own. The World Insttute for Development Economcs Research (WIDER) was establshed by the Unted Natons Unversty (UNU) as ts frst research and tranng centre and started work n Helsnk, Fnland n 1985. The Insttute undertakes appled research and polcy analyss on structural changes affectng the developng and transtonal economes, provdes a forum for the advocacy of polces leadng to robust, equtable and envronmentally sustanable growth, and promotes capacty strengthenng and tranng n the feld of economc and socal polcy makng. Work s carred out by staff researchers and vstng scholars n Helsnk and through networks of collaboratng scholars and nsttutons around the world. www.wder.unu.edu publcatons@wder.unu.edu UNU World Insttute for Development Economcs Research (UNU-WIDER) Katajanokanlatur 6 B, 00160 Helsnk, Fnland Camera-ready typescrpt prepared by Lorrane Telfer-Tavanen at UNU-WIDER The vews expressed n ths publcaton are those of the author(s). Publcaton does not mply endorsement by the Insttute or the Unted Natons Unversty, nor by the programme/project sponsors, of any of the vews expressed.

1 Introducton In Brazl, wealth has been largely assocated wth land. From the very begnnng of the colonzaton, only 30 years after the dscovery, the Portuguese Crown dvded the huge terrtory n 15 large tracts of land whch were donated to grantees wth heredtary successon. Today, Brazl has one of the most skewed land dstrbutons n the world and, not surprsngly, t s one of the most unequal countres n terms of the ncome dstrbuton. Facng the challenge of reducng nequalty of land ownershp and ntensfyng land use, the Brazlan government mplemented a land reform programme n 1966, wth the enactment of the Land Act. Snce ts creaton, there are mportant dfferences n the mplementaton of the programme through tme and space. Ths study studes the structure of land ownershp and land dstrbuton n Brazl, nvestgatng the consequences of the land reform programme mplemented n the 1990s. The emprcal strategy s based on the use of tme and space varaton of the land reform programme as a means of dentfyng a causal relatonshp of land reform on land ownershp and nequalty. Ths strategy s mplemented wth household-level data from the Natonal Household Survey (PNAD), coverng the perod from 1992-2002 (except 1994 and 2000), and land dsappropratons publshed by the Brazlan Insttute for Land Reform (INCRA). The man fndngs of the study are the followng. The nvestgaton of the effect of the land reform on landownershp suggests that () there s no ncrease on the access to land of the typcal Brazlan rural household; and () the effect s dfferentated wth respect to household ncome and the educatonal level of the household head there s an ncrease n land ownershp of poorest households and those wth the least educated heads, and a decrease for the other classes of rural households. Concentratng on the famles wth landholdngs, the analyss of the land reform effect on the dstrbuton of land provdes evdence of an ncrease of the land nequalty. Ths result s obtaned both wth the decomposton of the effect of land reform nto household ncome groups and wth the quantle regresson analyss. Land reform seems to reduce the sze of small landowners (poor households) and ncrease the sze of those above the medan (rcher households). These results contrbute for a better understandng of the mpact of redstrbutve land reform n Latn Amerca. Although there s a vast lterature addressng land reform and agraran organzaton, there s lttle evdence about the Latn Amercan experence (Bnswanger and Dennger 1997; Carter and Zegarra 2000; and Dennger and Feder 2000). Although some authors, such as Connng (2001) and Connng and Robnson (2001), have constructed models that exhbt features often observed n Latn Amerca to analyze agraran organzaton and land reform, most of the lterature consders 1

general aspects or case studes from Asa (Grossman 1994; Horowtz 1993; Besley and Burgess 2000; and Banerjee et al. 2002). The study s presented n sx sectons. Secton 1 presents the hstorcal determnants of land concentraton n Brazl. Secton 2 descrbes the nsttutonal background regardng the Brazlan land reform programme. Data s depcted n Secton 3. Secton 4 nvestgates the correlaton between land and wealth ndcators. Sectons 5 and 6 evaluate the mpact of the land reform on land ownershp and land nequalty, respectvely. Then, a summary of the results and fnal remarks are presented n the concluson secton. 2 Hstorcal determnants of the land dstrbuton and land access n Brazl 2.1 Early determnants of the land concentraton The hghly concentrated Brazlan land dstrbuton s deeply rooted n the colonzaton process. In the 1530s, nspred by the success of land settlements n the Madera Islands, Portugal s Kng João III dvded Brazl nto 15 terrtores called captanas heredtáras (heredtary captances) areas donated to Portuguese grantees (captans) wth heredtary successon. Each captan had complete authorty over hs land. However, due to a seres of obstacles, only a few captanas have succeeded, and sx of the captans never took possesson of ther clams (Bueno 1999). Another wave of settlements occurred n the seventeenth century, wth the ncrease of the global demand for sugar. In another land concentratng ntatve, the crown offered large tracts of land (sesmaras) n order to encourage settlement and producton. The holders of sesmaras experenced complete property rghts over ther holdngs whenever land was kept under cultvaton. It s worth notng that such a condton regardng land use remaned throughout Brazl s hstory and was reassured n the consttuton of 1988. The sesmara system fnshed n 1822 wth the Brazlan ndependence (Alston and Mueller 2003). From 1822 to 1850, no land polcy took place and settlers obtaned land by squattng, enforcng ther clams by socal norms. In 1850, the landowners of the coffee plantatons passed the Land Act, whch set the pattern for modern landholdng. The Land Act of 1850 forbade the colonal practce of obtanng land through squattng, lmtng the acquston to purchase. All exstng squatters were legalzed and, surprsngly, all sesmaras were revaldated (Alston and Mueller 2003). Concentraton of land was the rule and the great majorty of the people (especally after the 1888 abolton of slavery) were forced to work on large plantatons and farms wthout any hope of acqurng a small farm of ther own. 2

2.2 Modern determnants of land concentraton In addton to the hertage from the colonal perod, the macroeconomc envronment n the last 40 years has played a key role as a determnant of land dstrbuton n Brazl. The followng analyss consders landholdngs as hedge aganst nflaton and macroeconomc nstabltes, both of whch tend to set a wedge between the prce of land and the captalzed value of the ncome stream generated from agrculture. Especally n perods of hgh macroeconomc nstablty, people demand land as a mechansm of protecton aganst aggregate uncertanty. Assunção (2005) argues that ths feature, coupled wth mperfectons n the land rental market, leads to an neffcent hgh concentraton of landholdngs. The exstence of a non-agrcultural component n the demand for land can be dentfed through the comparson between land prces and rental rates of croplands and pastures. The emprcal strategy s based on the fact that whle an ncrease on macroeconomc nstablty rases the land prces, the same s less lkely to occur wth the rental rates. The emprcal test s based on a sequence of polces adopted to contan the nflatonary process durng the 1980s and 1990s. Those heterodox polces have resulted n unexpected (and exogenous, from the pont of vew of agrcultural producers) ncreases n the uncertanty of the economy, led by notons concernng nertal nflaton. The mplementaton of economc plans amed specfcally at contanng nflatonary nerta through a set of measures ncludng the dendexaton of the economy, temporary prce freezes, and a freeze on fnancal assets to reduce the economy s lqudty and generate resources for the budget. The uncertanty ntroduced nto the economy by those drastc measures generated a large shft n the demand for safe assets, ncludng land. The non-agrcultural component of land demand, as mentoned above, s dentfed by comparng the effects of the mplementaton of such polces on the land prces for sale and rental. Two dummy varables were bult; one ndcatng the ntroducton of a new economc plan n the current semester, and another one related to the nsttuton of an economc plan n the prevous semester. The followng plans are consdered: Cruzado Plan (February 1986), Bresser Plan (June 1987), Summer Plan (January 1989), Collor Plan (Aprl 1990) and Real Plan (June 1994). Regressons for the whole Brazlan sample were estmated and reported n Table 1. The dependent varables were detrended by usng the Hodrck-Prescott flter and centered around the orgnal mean. Therefore, the ntercept of each regresson can be nterpreted as the average land prce n the absence of new economc plans n the current and prevous semester. The coeffcents of the dummy varables represent the average changes n prces at the mplementaton of the economc plans and n the subsequent semester. The data consst of sem-annual observatons of land prces for sale and rental coverng the perod 1966-2000. They are collected by the Getúlo Vargas Foundaton, whch gathers nformaton from more than 3,600 local agences scattered among many 3

t t t 1 1 Brazlan dstrcts. Observatons refer to actual transactons, collected wthn the dstrcts at the end of each semester. Table 1: Effect of Brazlan economc plans on land prces and rental rates, 1966-2000 Pastures Cropland Estmates sale rental sale rental (A) Constant 1520.1 118.2 2643.8 196.4 (0.000) (0.000) (0.000) (0.000) (B) Dummy varable: new economc plan n 591.1 24 877.8 11.1 the current semester (0.012) (0.005) (0.018) (0.298) (B) / (A) % 38.90 20.30 33.20 5.70 (C) Dummy varable: new economc plan n 733.1 19.2 1095.1 2.85 the prevous semester (0.002) (0.022) (0.004) (0.788) (C) / (A) % 48.20 16.20 41.40 1.50 Observatons 69 69 69 69 R-squared 0.17 0.14 0.15 0.02 Source: Assunção (2005). Table 1 shows that the economc plans promoted sgnfcant ncreases n land prces for sales of both meadows and cropland n the current and next semester, accountng for more than 15 per cent of the total varablty of these varables from 1966 to 2000. The effects upon rental rates are much smaller and statstcally nsgnfcant for cropland. Ths suggests that the response of land prces for sale to an exogenous ncrease n macroeconomc nstablty s larger compared wth rental rates, whch s consstent wth the exstence of a non-agrcultural purpose of landholdng. For pastures, the economc plans have determned an ncrease of almost 40 per cent n land prces of sales n the current semester and up to 50 per cent n the next one. The rental rates have experenced a much lower ncrease, around 20 per cent. For cropland, Table 1 shows sgnfcant ncrements only for land prces, both n the current and subsequent semesters. The dfference between meadows and cropland mght be a result of the demand for lvestock as another source of hedge. 2.3 Latn Amercan land markets In order to put the Brazlan case n a more general perspectve, ths secton presents some evdence on the organzaton of agrculture sector n Latn Amercan countres. Land markets n Latn Amerca present two dstngushng types of mperfectons (Dennger and Feder 2000). Frst, land s used not only as a productve asset but also as a source of other benefts; as a hedge aganst nflaton, as an asset that can be lqudated to smooth consumpton n the face of rsk, as collateral for access to loans, as a tax shelter, or as a means of launderng llct funds (de Janvry et al. 1997). Second, land rental markets n Latn Amerca are underdeveloped. A large body of lterature has been 4

theoretcally addressng the reasons for the mperfectons n the land rental market.1 Many reasons provde explanatons for a reducton n the share of output approprated by the tenants. Specfcally, a factor that s lkely to be mportant n most Latn Amercan countres s the landlord s fear of loss of the land (Macours et al. 2001). Table 2 compares several ndcators of Latn Amerca wth Asa, Europe and the Unted States. Overall, Latn Amercan countres combne a hghly skewed dstrbuton of land wth the predomnance of owner or owner-lke form of land tenure, and agrcultural land s mostly represented by meadows and pastures, remanng under exploted. Table 2: Internatonal Indcators Indcators Latn Amerca Asa Europe Unted States Number of holdngs 10,281,607 143,934,358 7,625,520 1,911,859 Total area (ha.) 705,586,803 268,741,639 125,394,061 377,088,222 Average farm sze 68.6 1.9 16.4 197.2 Land Gn coeffcent 0.83 0.52 0.64 0.75 Agrcultural land % 65.70 91.80 70.60 88.90 cropland % 20.40 99.50 65.60 52.10 meadows/pastures % 79.60 0.50 34.40 47.90 Land tenure (area) owner/owner-lke % 85.50 88.70 61.40 33.90 rented from other % 3.60 2.80 23.30 11.60 mxed/others % 10.70 8.50 15.30 54.50 Below 10 farms % 58.70 97.40 74.50 29.50 area % 2.50 72.60 11.40 1.20 Notes: Latn Amerca Argentna, Brazl, Colomba, Chle, Honduras, Panama, Paraguay, Peru, Puerto Rco, Uruguay, Venezuela; Asa Bangladesh, Cyprus, Inda, Iran, Japan, Nepal, Pakstan, Phlppnes, Thaland, Turkey; Europe Austra, Belgum, France, Italy, Luxembourg, Portugal, Span, Swtzerland, Unted Kngdom. Source: FAO (1990, 2000). Ths regularty can be explaned by the absence of a land rental market and nonagrcultural purposes for landholdng (Assunção 2005). Imperfectons n rental market are reflected n 85 per cent of the total area operatng under owner or owner-lke forms, 1 The basc arguments are: rsk-sharng (Cheung 1969); hdden actons and moral hazard (Stgltz 1974; Ghatak and Pandey 2000; Eswaran and Kotwal 1985); screenng (Hallagan 1978; Allen 1982); and lmted lablty constrants (Shetty 1988; Laffont and Matouss 1995). 5

most of whch are consttuted by large landholdngs. In contrast, 58.7 per cent of the farms have less than 10 hectares. In addton to the nverse relatonshp between farm sze and productvty (Berry and Clne 1979; Feder 1985), ths evdence ndcates that even wth low productvty, large landowners nether lease out nor sell ther plots. They do not lease them out because the rental market has many mperfectons. And snce they obtan non-agrcultural payoffs as a result of land ttles, they are not wllng to sell them. Ths conjugaton makes Latn Amercan agrculture unque, and the fact that only oneffth of the agrcultural land s not covered by pastures and meadows may be an ndcaton of an neffcent organzaton of producton, wth sgnfcant welfare mplcatons. Although Asan countres, for the most part, exhbt the same tenure structure, they have a much more egaltaran dstrbuton of land and a sgnfcantly smaller average farm sze. There s no evdence of a mssng rental market snce t seems that there s no demand for t. The predomnance of cropland ndcates the hgh ntensty of land use. The stuaton n European countres s smlar but less evdent. The rental market works, and only two-thrds of the area s operated by owner or owner-lke forms. As n Latn Amercan countres, the Unted States exhbts a concentrated land dstrbuton, and the average plot sze s large. However, only one-thrd of the area s under owner or ownerlke form of tenure, and less than 30 per cent of the farms have less than 10 hectares. As a result, 52.1 per cent of agrcultural land comprse crop. 3 Brazlan land reform 3.1 A bref hstory A part of land reform hstory n Brazl begns wth the Land Act of 1964, brought about by the mltary regme. The long and comprehensve text consttuted detaled proposals for agraran reform. The law created the Brazlan Insttute for Agraran Reform (IBRA) and the Natonal Insttute for Agrcultural Development (INDA) n order to carry out the Act. In 1971, IBRA and INDA were merged nto the Natonal Insttute for Rural Settlement and Agraran Reform (INCRA). The Act was a means of dssuadng the pressure for land redstrbuton of socal movements, especally the peasant leagues and the emergng actvsm of Catholc prests. The poltcal context n Latn Amerca n early 1960s was characterzed by peasant mltances and threats of agraran rebellons. However, nstead of dstrbutng property, the economc strategy of the mltary regme amed at the modernzaton of the large landholders by means of subsdzed rural credt. Soybean cultvaton the man target of the rural polces generated large surpluses for export and, smultaneously, resulted n the absorpton of small farmers by medumand large-szed propertes, concentratng the land dstrbuton. Wth the return of democracy n 1985, the frst Natonal Agraran Reform Plan (1985-89) was prepared and launched, establshng the unrealstc target of settlng 1.4 mllon 6

famles n fve years. But, as shown n Table 3, Sarney s government dsapproprated less than 5 mllons of hectares, slghtly more than 10 percent of the ntal proposal. On the other hand, Sarney s government determned the frst of two sgnfcant waves of dsappropratons n the recent Brazlan hstory, as shown n Fgures 1 and 2. Fgure 1: Number of expropratons: Brazl, 1980-2002 350 900 Number of Processes (Regonal) 300 250 200 150 100 50 800 700 600 500 400 300 200 100 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 0 South Southeast Central-West Northeast North Brazl In the followng Collor government the programme came to a halt only 15,065 hectares were dsapproprated. Wth the mpeachment of Collor and the substtuton of the presdent, the land reform process was resumed and more than 20,000 famles were settled n almost 1.5 mllon hectares. Durng hs frst run of offce (1995-98), Presdent Fernando Henrque Cardoso accelerated the rhythm of the settlements. As shown n Table 3, more than 7.5 mllons hectares were dsapproprated n the perod.2 Fgures 1 and 2 shows that the dsappropraton wave undertaken n Cardoso s admnstraton s substantally dfferent from the dsappropratons whch occurred durng the Sarney government. Whle the dsapproprated area n each year of the two governments are comparable, the number of processes establshed n the Cardoso s s much hgher, suggestng that the settlements were more decentralzed through the Brazlan terrtores. The perod was also characterzed by conflcts and land nvasons, mostly assocated wth the Landless Workers Movement, whch s the largest socal movement 2 The offcal report ndcates that the frst Cardoso s government settled landless households on 12 mllon hectares. On the other hand, the data on dsappropraton process from INCRA ndcates 7.5 mllon hectares. The dfference of 4.5 mllon hectares (37 per cent) may be due to settlements on publc lands or even to errors n the computaton of the 12 mllon hectares. 7

n Latn Amerca wth more than a 1.5 mllon members. In the second run of Cardoso s admnstraton the focus of land reform changed from the dsappropraton model to a new form of negotated land reform (Dennger 1998). Fgure 2: Exproprated area: Brazl, 1980-2002 1,000,000 2,500,000 900,000 Dsapproprated area (Regons) 800,000 700,000 600,000 500,000 400,000 300,000 200,000 2,000,000 1,500,000 1,000,000 500,000 100,000 0 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 South Southeast Central-West Northeast Brazl Brasl Fgure 3 shows the spatal dstrbuton of the dsappropratons durng the Cardoso perod. The process s clearly heterogeneous, restrcted to sub-regons of the country. In other to focus our analyss on the areas where the process was more concentrated, a subsample of selected Brazlan States s bult. Ths study evaluates the consequences of ths modern wave of land redstrbuton based on dsappropratons, coverng the perod 1992-2002 and correspondng to the governments of Itamar Franco and Fernando Henrque Cardoso. 3.2 Land dsappropraton: procedures and costs Introduced by the Land Act, the land dsappropraton legslaton was sgnfcantly changed by Brazl s 1988 Consttuton. Snce then, only unproductve land s under the rsk of dsappropraton for whch the state needs to pay a far prce. After the 1993 amendment, the far prce became the market prce. Therefore, at the same tme there s a permt of confscaton, the government needs to pay the market prce whch, n prncple, s determned by buyers and sellers rather than anythng else. 8

Table 3: Brazlan land reform expropraton processes from 1979 to 2003 Brazlan presdents João Fgueredo (Mar/1979 to Mar/1985) José Sarney (Mar/1985 to Mar/1990) Fernando Collor de Mello (Mar/1990 to Oct/1992) Itamar Franco (Oct/1992 to Jan/1995) Fernando Henrque Cardoso I (Jan/1995 to Jan/1999) Fernando Henrque Cardoso II (Jan/1999 to Jan/2003) Central-West Brazl North Regon Northeast Regon Regon Southeast Regon South Regon no. of no. of no. of no. of events total area events total area events total area events total area no. of no. of events total area events total area 131 2,845,029 21 1,503,700 34 488,966 26 532,296 11 47,557 39 272,510 701 4,811,507 128 1,789,716 258 1,276,426 89 1,290,367 85 281,368 141 173,630 7 15,065 2 5,550 0 0 3 3,041 1 3,584 1 2,890 245 1,365,263 36 402,473 113 476,309 48 419,772 15 30,746 33 35,962 2,323 7,561,048 358 2,181,950 999 2,260,640 431 2,414,377 253 442,025 282 262,056 1,265 2,785,296 186 511,376 633 1,175,412 189 775,182 156 249,238 101 74,089 Source: INCRA (1999). 9

Fgure 3: Land dsappropraton n Cardoso's government Source: Natonal Insttute for Rural Settlement and Agraran Reform (INCRA). 10

Ths conceptual confuson n the legslaton of dsappropraton along wth other nsttutonal falures mposed hgh costs to the land reform. Accordng to INCRA (1999), the fnal cost stpulated by the judcal system n the end of a dsappropraton process s, on average, fve tmes the ntal evaluaton. In the southeastern part of the country the averaged multpler was 14. Reydon (2000) descrbes eght necessary steps of the dsappropraton process. The process begns wth an act sgned by the Brazlan presdent and fnsh, usually, wth a judcal decson. There are three ssues addressed n the judcal demands: the tems to be compensated, the amount of the ndemnty, and the form of payment (publc bonds versus cash). 4 Data The followng analyss s based on combned data from two sources. The frst one, s the database on the dsappropraton processes publcly avalable n the INCRA s webste.3 The data comprses the date, farm s name, area, and muncpalty of each approved process of land dsappropraton snce 1979. The second source of nformaton s the Natonal Household Surveys (PNAD), collected annually4 snce 1981 by the Brazlan Census Bureau (IBGE). Snce 1992, the PNAD survey provdes nformaton on landholdngs and, therefore, the perod consdered n the analyss s 1992 to 2002. The sample conssts n all rural households avalable n the PNAD survey, from 1992 to 2002. The nformaton across years refers to repeated cross-sectons. It s not possble to form a panel wth PNAD data. For each household, there s nformaton on landholdngs, household head characterstcs, spouse characterstcs and household characterstcs. Snce the PNAD survey s representatve at the state level, the nformaton of each household s combned wth nformaton on land dsappropraton n the correspondng state n the prevous year. It s assumed, mplctly, that settlements take about one year after the dsappropraton to be establshed. The results are robust to the use of dfferent lags of the nformaton about land reform. Actually, the dsappropraton n the current year s very correlated wth dsappropraton n the prevous year or two years before. Table 4 reports all the varables consdered n the analyss. We consder two man dependent varables: a bnary varable ndcatng whether the household s landowner or not, and the logarthm of landholdng area. On average, 39 per cent of all 131,775 observatons n the sample hold a postve amount of land. The average farm sze s 41 hectares for those wth landholdngs.5 The man ndependent varables n ths study are those wth nformaton on land reform. There are two varables measurng land dsappropraton n the state n the prevous year. The frst one s a dummy varable 3 www.ncra.gov.br. 4 Except for the years of 1991, 1994 and 2000. 5 If we nclude the households wth no land, the average becomes 16.6 hectares as shown n Table 4. 11

ndcatng whether or not there was any dsappropraton n the state, untl the prevous year. More than 3/4 of the sample have at least one dsappropraton process on ther state. The second varable ams at capturng the ntensty of the dsappropraton, and t s defned as the rato between the dsapproprated area (measured n hectares) and the number of rural households of each state untl the prevous perod. There are also three sets of control varables regardng characterstcs of the household, household head and spouse. Summary statstcs on all consdered varables are depcted n Table 4. Table 4: Descrpton of the varables Varables Obs Mean Std Dev. Mn Max Dummy (household wth landholdngs=1) 131775 0.3948473 0.48882 0 1 Total area of the landholdngs 131775 16.56629 149.48 0 10000 log(total area of the landholdngs) 52031 11.25869 1.72691 0 18 Dummy (postve dsappropraton untl the prevous year=1) Dsapproprated area per rural household n the state untl the prevous year Log (dsapproprated area per rural household untl the prevous year) Household head characterstcs 131775 0.7775223 0.41591 0 1 102458 0.0001445 0.00034 0 2.56E-03 102458-9.85296 1.27841-13.554-5.967 Gender 131775 0.8706887 0.33555 0 1 Age 131757 46.59258 16.0858 10 106 Years of schoolng 131589 3.44639 2.95841 1 16 Dummy (employer=1) 115555 0.0461598 0.20983 0 1 Dummy (employee=1) 115555 0.4219116 0.49387 0 1 Dummy (self-employed=1) 115555 0.4730648 0.49928 0 1 Income 128465 417.0069 833.528 0.000 43032.780 Spouse characterstcs Dummy (spouse s present =1) 131775 0.7849972 0.41083 0 1 Gender 103443 0.013727 0.16513 0 1 Age 103429 39.84355 14.4284 11 98 Years of schoolng 103179 4.053121 3.06749 1 16 Income 103111 83.39452 277.456 0 22258 Number of household members 131775 4.140178 2.1597 1 24 Number of members above 60 years old 131775 0.3645987 0.65412 0 6 Number of members under 10 years old 131775 1.073899 1.29223 0 10 12

Household characterstcs Per capta ncome 127691 174.228 348.246 0 16749 Dummy (lves on own land=1) 92928 0.8836949 0.32059 0 1 Dummy (access to pped water=1) 131274 0.472226 0.49923 0 1 Dummy (house wth bathroom=1) 131271 0.6342223 0.48165 0 1 Dummy (access to electrcty=1) 131267 0.6907905 0.46217 0 1 Dummy (has water flter=1) 131265 0.4307774 0.49519 0 1 Dummy (has colour TV=1) 131271 0.3636523 0.48105 0 1 Dummy (has refrgerator=1) 131255 0.4637538 0.49869 0 1 Source: IBGE, Natonal Household Surveys (PNAD) (varous). Table 5: Dstrbuton of household landholdngs Year % households Centles wth landholdngs 1% 5% 10% 25% 50% 75% 90% 95% 99% 1992 44.2 0.3 0.6 1.0 2.0 7.3 24.2 68.0 137.9 565.0 1993 42.7 0.3 0.7 1.0 2.1 7.3 24.2 62.9 121.0 484.0 1995 42.7 0.3 0.6 0.9 2.0 6.1 24.0 62.0 121.0 615.0 1996 39.6 0.2 0.6 0.9 2.0 6.0 21.8 67.0 121.0 470.0 1997 40.4 0.3 0.6 0.9 2.0 6.1 22.0 60.5 120.0 426.6 1998 38.4 0.3 0.6 0.9 2.0 6.1 23.0 60.0 106.5 366.0 1999 39.2 0.3 0.6 1.0 2.0 6.0 20.0 58.1 111.3 423.5 2001 42.0 0.1 0.6 1.0 2.0 7.0 24.2 62.0 120.0 484.0 2002 42.2 0.2 0.6 1.0 2.0 7.0 24.2 60.0 106.5 380.0 Source: IBGE, Natonal Household Surveys (PNAD) (varous). Table 5 presents the evoluton of the percentage of households wth landholdngs and the land dstrbuton of the households wth postve area. There s no clear trend nether n the proporton of landowner households nor n landholdng dstrbuton. Table 6 reports the evoluton of the per capta household ncome for the perod under analyss. The only clear pattern depcted n Table 6 s that ncome nequalty s lower when we restrct the ncome dstrbuton to the households wth land. 13

Table 6: Dstrbuton of the per capta household ncome All households Only households wth landholdngs Year 25% 50% 75% 90% 95% 25% 50% 75% 90% 95% 1992 40 84 172 314 457 42 87 169 278 402 1993 40 89 194 373 599 43 92 183 304 474 1995 46 93 185 348 522 54 104 186 328 491 1996 46 98 194 356 531 53 104 184 338 516 1997 45 93 185 346 544 53 104 180 341 498 1998 46 93 187 331 501 58 108 187 345 514 1999 49 98 190 339 516 57 109 188 344 514 2001 48 103 215 367 574 54 107 207 337 478 2002 50 100 200 330 480 55 103 200 320 455 Source: IBGE, Natonal Household Surveys (PNAD) (varous). 5 Wealth and land Ths secton assesses the relatonshp between land and wealth ndcators n Brazl. Snce there s no nformaton on personal assets, the focus s restrcted to an approxmaton of the household wealth. Ths approxmaton s comprsed by three components. The frst component s the per capta household ncome. Under mperfect credt markets, household s expected ncome s an ncreasng functon of wealth (Banerjee and Newman 1993; Galor and Zera 1993). Thus, nformaton on the total household ncome reflects, to some extent, nformaton on famly wealth. The frst column of Table 7 presents a regresson of the logarthm of household landholdngs on the logarthm of per capta ncome. The estmated coeffcent suggests a very tght relatonshp between ncome and land, statstcally sgnfcant at 1 per cent. In the next two columns, Table 7 shows the relatonshp between land and other components of household wealth. The second wealth component s conssted by durable goods: water flter, colour TV and refrgerator. And the thrd wealth component on whch there s nformaton n the PNAD survey s related to the value of the houses and, n partcular, to the housng nfrastructure. The underlyng assumpton of ths exercse s that wealther famles lves n better equpped houses, both n terms of durable goods and n terms of nfrastructure. 14

Table 7: The relatonshp between wealth ndcators and landholdng Dependent varable: log (area of the household landholdngs) (1) (2) (3) (4) log(per capta ncome) 0.569*** 0.460*** 0.400*** 0.317*** (0.007) (0.008) (0.008) (0.012) Dummy (has water flter=1) 0.293*** 0.270*** 0.165*** (0.015) (0.015) (0.015) Dummy (has colour TV=1) 0.067*** 0.095*** -0.021 (0.020) (0.019) (0.020) Dummy (has refrgerator=1) 0.462*** 0.422*** 0.244*** (0.018) (0.022) (0.023) Dummy (access to pped water=1) 0.508*** 0.402*** (0.021) (0.022) Dummy (house wth bathroom=1) 0.299*** 0.212*** (0.019) (0.020) Dummy (access to electrcty=1) -0.620*** -0.570*** (0.020) (0.021) Household head characterstcs No No No Yes Spouse characterstcs No No No Yes Household characterstcs No No No Yes Constant 8.879*** 9.070*** 9.319*** 7.409*** (0.037) (0.038) (0.038) (0.217) Year dummes Yes Yes Yes Yes Observatons 48957 48860 48859 40860 R-squared 0.13 0.16 0.19 0.26 Notes: Robust standard errors n parentheses. * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1%. Source: IBGE, Natonal Household Surveys (PNAD) (varous). Column (2) of Table 7 reports that the presence of all durable goods are hghly correlated wth land. Moreover, when nformaton on durable goods are ncorporated n the regresson, the coeffcent of ncome s reduced from 0.569 to 0.460. Ths s evdence that ncome n column (1) reflects part of the household wealth whch s ncorporated n column (2). The same seems to occur wth the ntroducton of the nformaton about nfrastructure n column (3). The only excepton s access to electrcty, whch has a statstcally sgnfcant and negatve coeffcent. The results mght reflect the fact that households wth larger tracts of land are located n more solated areas. In ths case, access to electrcty s more related to urbanzaton than to wealth. Fnally n column (4), all avalable nformaton about the characterstcs of the household, head and spouse are ntroduced n the regresson to control for observed heterogenety. Even after controllng for all these characterstcs, land remans hghly correlated to the three wealth components: per capta ncome, durable goods and 15

house s nfrastructure. Thus, the followng analyss about the land dstrbuton can be also nterpreted as a study of the wealth dstrbuton n Brazl. 6 Land reform and land ownershp Ths secton estmates the mpact of land reform on the fracton of the rural households wth landholdngs, through household-level data. The sample comprses all surveyed rural households, whether they hold a postve amount of land or not, for the perod from 1992 to 2002. Households are pooled across years,.e., households of dfferent perods are consdered as dfferent households. The results are estmated consderng the followng lnear probablty model6: Pr { L 0 D, X } = α I{ D > 0} + γ log( D ) I{ D > 0} + β X, > (1) where I D > 0 s a bnary varable ndcatng whether or not there s dsappropraton untl the prevous year n the State where household lves, log ( D ) I{ D > 0} s the logarthm of the dsapproprated area per rural household n State wth postve dsappropraton untl the prevous year, and X s a vector of control varables ncludng household head characterstcs, spouse characterstcs, household characterstcs and year dummes. L stands for the total area owned by household, { } Under the assumpton that, gven the observed characterstcs, the dsappropraton untl the prevous year s not correlated wth the unobserved determnants of landholdng, parameters α and γ measure the effect of land reform on the fracton of rural households wth landholdngs. Manpulatng (1) t s possble to show that: { L > 0 D > 0, X } Pr{ L > 0 D 0, X } α = Pr = (2) and { L > 0 D > 0, X } d log( D ) { L > 0 D 0, X }. d Pr d Pr > γ = = (3) dd D Thus, the parameter α measures the effect of the frst dsapproprated hectare per rural household on the fracton of rural famles wth landholdng, and the parameter α represents the effect of 1 per cent change n the dsapproprated area per rural household on land ownershp. Results from the estmaton of Equaton (1) are presented n Table 8, consderng nested specfcatons for the vector of control varables. In column (1), whch control only for the year dummes, the exstence land dsappropraton n the prevous year has an effect of 31.7 percentage ponts on the fracton of households wth land. The effect of 1 per cent of varaton n the dsapproprated area per rural household 6 For ease of notaton, t s consdered log(0) 0=0 n the nterpretaton of Equaton (1). 16

Table 8: Effect of land reform on the fracton of the rural populaton wth landholdngs Dependent varable: Dummy varable ndcatng whether the household own land Decomposton of the effect wth respect to: Wthout controls Wth controls ncome age schoolng (1) (2) (3) (4) (5) Dummy (postve dsappropraton untl the prevous year=1) Dummy (postve dsappropraton) x dummy (2nd quntle) Dummy (postve dsappropraton) x dummy (3rd quntle) Dummy (postve dsappropraton) x dummy (4th quntle) Dummy (postve dsappropraton) x ummy (5th quntle) log(dsapproprated area per rural household) log(dsapproprated area per rural household) x dummy (2nd quntle) log(dsapproprated area per rural household) x dummy (3rd quntle) log(dsapproprated area per rural household) x dummy (4th quntle) 0.317*** 0.01 0.135*** 0.089*** 0.077*** (0.013) (0.014) (0.024) (0.030) (0.020) -0.104*** -0.076* -0.043 (0.034) (0.040) (0.030) -0.135*** -0.057-0.141*** (0.037) (0.040) (0.029) -0.243*** -0.117*** (0.039) (0.039) -0.173*** -0.178*** (0.038) (0.041) 0.031*** 0.002 0.013*** 0.011*** 0.007*** (0.001) (0.001) (0.002) (0.003) (0.002) -0.010*** -0.009** -0.002 (0.003) (0.004) (0.003) -0.011*** -0.007* -0.012*** (0.004) (0.004) (0.003) -0.022*** -0.013*** (0.004) (0.004) 17

log(dsapproprated area per rural household) x dummy (5th quntle) -0.014*** -0.020*** (0.004) (0.004) Household head, spouse and household characterstcs No Yes Yes Yes Yes Constant 0.407*** -0.063*** -0.055*** -0.041*** -0.083*** (0.003) (0.011) (0.012) (0.015) (0.012) Year dummes Yes Yes Yes Yes Yes Observatons 131775 63562 63562 63562 63562 R-squared 0.00 0.59 0.60 0.59 0.59 Notes: Tercles rather than quntles were consdered for the case of years of schoolng due to the large proporton of heads wth 1 year of schoolng or less. Robust standard errors n parentheses. *sgnfcant at 10%; **sgnfcant at 5%; ***sgnfcant at 1%. Source: IBGE, Natonal Household Surveys (PNAD) (varous). 18

ncreases landholdng n 3.1 percentage ponts. However, when the full set of control varables are ntroduced, column (2) shows that the effect vanshes. Thus, on average, results from columns (1) and (2) of the Table 8 suggest that land reform does not ncrease the proporton of rural famles wth landholdngs. On the one hand, these results mght be true n the sense that Brazlan experence wth land reform does not ncrease the access to land. On the other hand, the estmated zero effect mght be the result of countervalng effects. As shown n Secton 2, the Brazlan land reform programme conssts on redstrbutve transfers from large landowners to small farmers and landless peasants. Unmproved and large tracts of land are under the rsk of expropraton, whle small and productve farms cannot be taken. Consequently, the process by tself has dfferentated effects on the rural households. In order to nvestgate possble dfferentated effects, Table 8 presents estmates of Equaton (1) n whch the parameters α and γ are decomposed accordng to the household per capta ncome, age and years of schoolng of the household head. For the cases of ncome and age, the sample was dvded nto quntles and, for the case of schoolng, t was consdered tercles, due to the large number of heads wth 1 year of schoolng or less. The results reported n columns (3) to (5) suggest that the absence of the effect estmated n column (2) s the result of heterogenety, related to ncome and educaton. Land reform ncreases the access to land of low ncome households and those wth less educated head, consderng both the occurrence of land reform ( ) ntensty ( γ ). α and ts Land dsappropraton ncreases n 13.5 percentage ponts the fracton of the rural famles wth landholdng, n the frst quntle of the per capta ncome dstrbuton, as depcted n column (3) of Table 8. To all other groups of ncome, the effect s substantally lower or even negatve. The effect of the ntensty of the land reform, whch s measured by the dsapproprated area per rural household, s also postve and statstcally sgnfcant to low ncome households. A smlar pattern s presented for the educatonal level of the household head n column (5). Only the lowest tercle, whch corresponds to the household head wth 1 year of schoolng or less, s affected postvely by the land reform. 7 Land reform and land dstrbuton The prevous secton nvestgates the effect of land reform on land ownershp. Here, the analyss s restrcted to landowner households, amng at estmatng the effect of the land reform on the land dstrbuton. It s not possble to assert, a pror, whether a redstrbutve land reform n the way were mplemented n Brazl ncreases or reduces the average landholdng sze. It depends on the relatonshp between the holdngs affected and not affected by the reform. If the farm sze of the benefcares s smaller than the average non-affected farm, land reform tend to reduce the typcal farm sze. On 19

the other hand, f the confscated farms are not the largest, t s possble to have an ncrease on the average post-reform landholdngs. The emprcal analyss whch follows s presented n two steps. Frst, t s consdered the effect on the average farm sze. Then, quantle regressons are used to nvestgate the effect of the land reform on each decle of the land dstrbuton. The frst set of results consders the followng lnear specfcaton7 focusng on the average landholdng sze: E ( log( L ) D, X ) φ I{ D > 0} + λ log( D ) I{ D > 0} + δ X. = (4) Agan, f the dsappropraton untl the prevous year s not correlated wth the unobserved determnants of landholdng sze, condtonal on the observed varables X, the parameters φ and λ measure the effect of land reform on the fracton of rural households wth landholdngs. Smple computatons wth (1) show that: ( ( ) ( ) ) D= 1 D = = = 0 D 1 D L L 0 log L log L X E X D 0 L φ (5) and = E = ( log( L ) D > 0, X ) d log( D ) dl de L λ = = E D > 0, X. (6) dd D Thus, the parameter φ represents the percent change of the frst dsapproprated hectare per rural household on the sze of the landholdngs. The parameter λ s the elastcty of the sze of the landholdngs wth respect to the dsapproprated area per rural household, for those States wth postve dsappropraton. Panel () of the Table 9 shows the estmates of Equaton (4), consderng dfferent sets of control varables and decompostons. Column (1) suggests that land reform reduces the average farm sze. However, controllng for all observed characterstcs, the effect becomes postve. The frst dsapproprated hectare per household has an mpact of ncreasng n 57.3 per cent the average farm sze. It s mportant to keep n mnd that the average of ths varable n the sample, accordng to Table 4, s substantally smaller than 1. Smlarly to the analyss of land ownershp, column (3) shows that land reform has dfferentated effects wth respect to the household per capta ncome. There s a reducton on the average farm sze of the 20 per cent poorest households and an ncrease n landholdngs of the others. The decomposton n terms of age does not present a clear pattern, whle there s also some heterogenety wth respect to the head s schoolng. 7As n the prevous secton, t s assumed that log(0) 0=0 n the nterpretaton of Equaton (4) for the sake of smplfcaton. 20

Table 9: Effect of land reform on the sze of landholdngs Dependent varable: log(area of the household landholdngs) Panel (): OLS estmates Decomposton of the effect wth respect to: Wthout controls Wth controls ncome age schoolng Dummy (postve dsappropraton untl the prevous year=1) Dummy (postve dsappropraton) x dummy (2nd quntle) Dummy (postve dsappropraton) x dummy (3rd quntle) Dummy (postve dsappropraton) x dummy (4th quntle) Dummy (postve dsappropraton) x dummy (5th quntle) log(dsapproprated area per rural household) log(dsapproprated area per rural household) x Dummy (2nd quntle) log(dsapproprated area per rural household) x dummy (3rd quntle) (1) (2) (3) (4) (5) -0.300*** 0.573*** -0.458*** 0.192 0.071 (0.077) (0.082) (0.143) (0.198) (0.125) 0.876*** 0.314 0.482*** (0.207) (0.252) (0.186) 1.171*** 0.397 1.175*** (0.226) (0.243) (0.174) 1.464*** 0.597** (0.233) (0.245) 2.118*** 0.452 (0.221) (0.285) -0.002 0.075*** -0.029** 0.033* 0.028** (0.007) (0.008) (0.014) (0.019) (0.012) 0.099*** 0.031 0.047** (0.020) (0.025) (0.018) 0.123*** 0.049** 0.113*** (0.022) (0.024) (0.017) 21

log(dsapproprated area per rural household) x dummy (4th quntle) log(dsapproprated area per rural household) x dummy (5th quntle) Household head, spouse and household characterstcs 0.147*** 0.068*** (0.023) (0.024) 0.201*** 0.044 (0.021) (0.028) no yes yes yes yes Constant 11.389*** 8.424*** 8.286*** 8.484*** 8.466*** (0.016) (0.197) (0.200) (0.209) (0.200) Year dummes yes yes yes yes yes Observatons 52031 35652 35652 35652 35652 R-squared 0.00 0.27 0.29 0.27 0.28 Panel (): Quantle regressons 10% 30% 50% 70% 90% Dummy (postve dsappropraton untl the prevous year=1) log(dsapproprated area per rural household) Household head, spouse and household characterstcs -0.585*** -0.649*** 0.071 1.132*** 1.663*** (0.099) (0.088) (0.085) (0.106) (0.120) -0.034*** -0.041*** 0.028*** 0.126*** 0.174*** (0.009) (0.008) (0.008) (0.010) (0.011) yes yes yes yes yes Year dummes yes yes yes yes yes Observatons 35652 35652 35652 35652 35652 Notes: Tercles rather than quntles were consdered for the case of years of schoolng due to the large proporton of heads wth 1 year of schoolng or less. Robust standard errors n parentheses. *sgnfcant at 10%; **sgnfcant at 5%; ***sgnfcant at 1%. Source: IBGE, Natonal Household Surveys (PNAD) (varous). 22

Fgure 4: Quantle estmates of the effect of land reform on the sze of landholdngs 2.5 2 1.5 1 0.5 0-0.5 10% 20% 30% 40% 50% 60% 70% 80% 90% -1 Fgure 5: Quantle estmates of the effect of the dsapproprated area on the sze of landholdngs 0.25 0.2 0.15 0.1 0.05 0-0.05 10% 20% 30% 40% 50% 60% 70% 80% 90% -0.1 Thus, ths frst set of results suggest that land reform has ncreased the average sze of the landholdngs, but ths effect s not homogeneous wth respect to the household per capta ncome poorer households experenced a reducton whle rcher households experenced an ncrease on the average farm sze. Comparng these results wth those of Secton 4, n whch land s hghly correlated wth ncome, Table 9 ndcates that land reform ncreases the nequalty n the dstrbuton of land. In order to address ths queston n a more systematc way, quantle regressons are estmated. The specfcaton 23

presented n Equaton (4) s estmated to each decle of the land dstrbuton. Results for 10, 30, 50, 70 and 90 percentles are depcted n panel () of Table 9, and the coeffcents φ and λ are plotted n Fgures 4 and 5, respectvely. The results suggest that Brazlan land reform have, surprsngly, ncreased the nequalty of the land dstrbuton. Land reform has negatve effects on holdngs wth sze below the medan and postve effects on holdngs above the medan of the land dstrbuton. 8 Concluson Throughout the Brazl s hstory, wealth s hghly assocated wth land ownershp. In ths sense, ths study looks at the recent Brazlan experence wth redstrbutve land reform n order to shed lght on ts effect on the dstrbuton of wealth n rural areas. After presentng the hstorcal determnants of land concentraton and the nsttutonal background to the land reform n Brazl, the study evaluates the mpact of land dsappropraton on land ownershp and land dstrbuton. Two man conclusons arse from the nvestgaton of the mpact of land dsappropratons on the fracton of the rural famles wth landholdngs. Frst, land reform does not ncrease land ownershp n rural areas, at least from an aggregate perspectve. Second, the decomposton of ths mpact accordng to household ncome and educaton of the household head reveals mportant dfferences. There s an ncrease n landownershp among the poorest households and those for whch the head has no more than one year of schoolng. For all other household classes there s a reducton n the percentage of landowners. Thus, consderng the whole rural populaton, land reform ponts towards a less unequal dstrbuton of assets, snce t ncreases land ownershp among poor households and reduces land ownershp among rch households. Interestngly, the analyss of the effect of land reform on the dstrbuton of land among landowner households suggests the opposte. Both the quantle regressons and the decomposton of the mpact accordng to ncome ndcate an ncrease n the nequalty of the holdngs. Land reform reduces the sze of the holdngs of poor famles and those wth smaller tracts of land. Concomtantly, t ncreases the farm sze of rch famles and those wth larger holdngs. References Allen, F. (1982). On Share Contracts and Screenng, Bell Journal of Economcs 13: 541-77. Alston, L., and B. Mueller (2003). Property Rghts to Land, Oxford Encyclopeda of Economc Hstory, Oxford Unversty Press: Oxford. 24

Assunção, J.J. (2005). Non-Agrcultural Land Use and Land Reform: Theory and Evdence from Brazl, mmeo, Department of Economcs, PUC-Ro: Ro de Janero. Banerjee, A.V., P.J. Gertler, and M. Ghatak (2002). Empowerment and Effcency: Tenancy Reform n West Bengal, Journal of Poltcal Economy 110(2): 239-80. Banerjee, A.V. and A.F. Newman (1993). Occupatonal Choce and the Process of Development, Journal of Poltcal Economy 101(2): 274-98. Berry, R.A., and W.R. Clne (1979). Agraran Structure and Productvty n Developng Countres, Johns Hopkns Unversty Press: Baltmore. Besley, T., and R. Burgess (2000). Land Reform, Poverty Reducton, and Growth: Evdence from Inda, Quarterly Journal of Economcs 115(2): 389-430. Bnswanger, H.P., and K. Dennger (1997). Explanng Agrcultural and Agraran Polces n Developng Countres, Journal of Economc Lterature 35(4): 1958-2005. Bueno, E. (1999). Captães do Brasl: A saga dos Prmeros Colonzadores, Coleção Terra Brasls Vol.3, Objetva: Ro de Janero. Carter, M.R., and E. Zegarra (2000). Land Markets and the Persstence of Rural Poverty n Latn Amerca: Post-Lberalzaton Polcy Optons, n A. Valdes and R. Lopez (eds) Rural Poverty n Latn Amerca: Analytcs, New Emprcal Evdence and Polcy, Macmllan: Basngstoke. Cheung, S.N.S. (1969). The Theory of Share Tenancy, Unversty of Chcago Press: Chcago. Connng, J. (2001). Latfunda Economcs, mmeo, Department of Economcs, Wllams College: Wllamstown MA. Connng, J., and Robnson, J.A. (2001). Land Reform and the Poltcal Organzaton of Agrculture, mmeo, Department of Economcs, Wllams College: Wllamstown MA. de Janvry, A., N. Key, and E. Sadoulet (1997). Agrcultural and Rural Development Polcy n Latn Amerca: New Drectons and New Challenges, FAO Agrcultural Polcy and Economc Development Seres 2, FAO: Rome. Dennger, K. (1998). Makng Negotated Land Reform Work: Intal Experence from Colomba, mmeo, World Bank: Washngton DC. Dennger, K., and G. Feder (2000). Land Insttutons and Land Markets, n B. Gardner and G. Rausser (eds) Handbook of Agrcultural Economcs, North Holland: Amsterdam. Eswaran, M., and A. Kotwal (1985). A Theory of Contractual Structure n Agrculture, Amercan Economc Revew 75(3): 352-67. 25