Efficiency and Equity Impacts of Rural Land Rental Restrictions:

Transcription

1 Publc Dsclosure Authorzed Publc Dsclosure Authorzed Publc Dsclosure Authorzed Publc Dsclosure Authorzed Polcy Research Workng Paper 434 Effcency and Equty Impacts of Rural Land Rental Restrctons: Evdence from Inda Klaus Dennger Songqng Jn Har K. Nagarajan The World Bank Development Research Group Sustanable Rural and Urban Development Team August 007 WPS434

2 Polcy Research Workng Paper 434 Abstract Recognton of the potentally deleterous mplcatons of nequalty n opportunty orgnatng n a skewed asset dstrbuton has spawned consderable nterest n land reforms. However, lttle attenton has been devoted to fact that, n the longer term, the measures used to mplement land reforms could negatvely affect productvty. Use of state level data on rental restrctons, together wth a natonally representatve survey from Inda, suggests that, contrary to orgnal ntentons, rental restrctons negatvely affect productvty and equty. The restrctons reduce the scope for effcency-enhancng rental transactons that beneft poor producers. Smulatons suggest that, by doublng the number of producers wth access to land through rental, from about 5 mllon currently, lberalzaton of rental markets could have far-reachng mpacts. Ths paper a product of the Sustanable Rural and Urban Development Team, Development Research Group s part of a larger effort n the group to assess the mpact of land polces on poverty and economc growth. Polcy Research Workng Papers are also posted on the Web at The author may be contacted at kdennger@ worldbank.org. The Polcy Research Workng Paper Seres dssemnates the fndngs of work n progress to encourage the exchange of deas about development ssues. An objectve of the seres s to get the fndngs out quckly, even f the presentatons are less than fully polshed. The papers carry the names of the authors and should be cted accordngly. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors. They do not necessarly represent the vews of the Internatonal Bank for Reconstructon and Development/World Bank and ts afflated organzatons, or those of the Executve Drectors of the World Bank or the governments they represent. Produced by the Research Support Team

3 Effcency and equty mpacts of rural land rental restrctons: Evdence from Inda Klaus Dennger, Songqng Jn, and Har K. Nagarajan World Bank, Washngton DC, USA Natonal Councl for Appled Economc Research, New Delh, Inda We would lke to thank A. Kaushk, T. Olsen, and S. Vats for excellent research assstance and S. Bery, D. Bradley, R. Buckley, E. Cook, J. Farrngton, G. Feder, T. Hanstad, T. Haque, D. Lal, I. Lavadenz, K. Otsuka, S. Rozelle, two anonymous referees, and the edtor managng ths submsson for nsghtful comments and suggestons. Fnancal support from the collaboratve DFID- World Bank program on land polces and rural development s gratefully acknowledged. The vews expressed n ths paper are those of the authors and do not necessarly reflect those of the World Bank, NCAER, or DFID.

4 . Introducton Researchers and polcy-makers alke have come to realze that unequal access to opportuntes arsng from a skewed dstrbuton of assets can be harmful for sustaned long-term growth and thus of concern Aghon et al. 999, World Bank 005. Not surprsngly n vew of the mportant role and long-lastng mpacts of hstorcal land tenure arrangements Banerjee and Iyer 005, land reforms have occuped center stage n Inda s polcy debates for a long tme. Followng successful abolton of ntermedares mmedately after ndependence, award of property rghts to sttng tenants through tenancy laws and expropraton wth subsequent transfer of above-celng land from large land owners to small farmers were man mechansms to mprove operatonal and ownershp dstrbuton of land. After a slow start, these polces alone transferred rghts to almost 0 mn ha of land, more than three tmes the land dstrbuted n land reforms n Japan, Korea, and Tawan together Kng 977. Stll, despte consderable progress n the past, such legslaton has almost ceased to provde new land access and there s growng concern that, n today s changed context, mantanng restrctons on land rental could cause large effcency losses by precludng effcency-enhancng transfers of land to complement rural dversfcaton. Gven that, despte ncreased opportuntes for rental, the share of Indan households partcpatng n rental markets decreased from 6% n 97 to % n 00 and that n other countres rental markets have been found to sgnfcantly ncrease productvty World Bank 007, the mpact of such restrctons could potentally be very large. Ths s countered by arguments that rental markets could, through reverse tenancy, gve rse to land concentraton that s not only neffcent but would also dsempower the poor by forcng them to offer ther labor to monopsonstc landlords n local markets for casual labor that dscrmnate heavly on the bass of gender and caste. To assess the functonng of land rental markets and explore effcency and equty mpacts of land rental restrctons, we use a model of producers who dffer n endowments and sklls and who face mperfect labor markets and transacton costs -further ncreased by polcy-nduced restrctons- n the land market. The model would lead us to expect three dstnct effects, namely a factor equalzaton effect whereby land markets wll transfer land to farmers wth lower land and hgher labor endowments and wth a hgher level of ablty; a dversfcaton effect whereby ncreases n the non-agrcultural wage rate wll lead to an ncrease n land rental actvty and a decrease n the equlbrum rental rate; and a transacton cost effect whereby hgher levels of transacton costs -whch may be polcy-nduced- wll ncrease the share of producers remanng n autarky thus reducng the number of effcency-enhancng land transfers.

5 We use a natonally representatve household panel spannng the 98 to 999 perod to emprcally test these predctons. As land s a state subject, we use cross-state varaton n land reform legslaton and ts mplementaton to explore mpacts of rental restrctons on land market functonng and outcomes A panel producton functon s used to obtan a measure of producers ablty to assess the productvty-mpact of land rental. Results hghlght that, contrary to what s often assumed, land rental markets help mprove productvty and equty by transferrng land to more productve and landless or land-poor households who, n the process, are able to mprove ther economc status. More mportantly, the pro-poor nature of land markets has mproved over tme as wealth bas that had characterzed such markets earler has been elmnated. The fact that land rental s more actve n locatons wth hgher levels of non-farm actvty supports the noton that t makes an mportant contrbuton to dversfcaton of ncome sources n rural areas -n fact, the opportuntes opened up through dversfcaton could be a key factor underlyng to help elmnate wealth-related barrers to rental market partcpaton. Our analyss demonstrates that tenancy restrctons have not only sgnfcantly ncreased the share of producers who reman n autarky but also prevent land access by more effcent producers. In addton to the results beng very robust, we also fnd that the magntudes nvolved are very large. Smulatons suggest that elmnaton of these restrctons could prompt an addtonal 40%-70% of producers to offer land for rental and, due to the smaller sze of land rented n, double the number of those who are able to access land through rental. Usng the mean dfference between rental partners to make nferences on potental productvty mpacts suggests that such mpacts could be large as well. Thus, whle mplementng such laws has helped to obtan socal gans n the past, mantanng them s an ncreasngly potent obstacle to realzng greater land access by those who are more productve and land-poor or landless and ways to mprove on ths may need to be explored. The paper s structured as follows. Secton two descrbes the hstory of land reforms and tenancy regulatons n rural Inda, revews evdence on the mpact of land rental market restrctons and key dfferences between rural and urban land markets, and uses ths to lay out a conceptual model and emprcal strategy. Secton three presents the data used and revews descrptve statstcs for polcy varables and household characterstcs. Secton four contans econometrc results for the producton functon and dfferent models of land rental market partcpaton and dscusses the results and ther robustness. Secton fve concludes by puttng results nto context and drawng out polcy mplcatons.. Background and conceptual framework Land reform polcy, through abolton of ntermedares, mposton of land celngs, and regulaton of tenancy contracts, played a key role n Inda from the moment t started ts ndependent exstence. A revew key elements of land rental and celng laws, ther mplementaton, and possble lnks between

6 legslaton and land market outcomes s followed by a conceptual model for households land market partcpaton and dervaton of hypotheses that can be tested wth the data at hand.. Orgns and nature of rural tenancy restrctons n Inda Under colonal rule, the man goal of Inda s land admnstraton system was to obtan government revenue. The de facto award of land rghts to revenue collectors zamndars n large parts of the country has consequences that affect development to ths day Banerjee and Iyer 005. Agraran reform was thus at the top of the mmedate post-ndependence agenda and the fact that land was put under the competence of states rather than the Center led to consderable dversty n the tmng, substance, and mplementaton of reforms across states. Abolton of rent-collectng ntermedares was tackled swftly and successfully vrtually everywhere. However, celng legslaton -amng to legslate a maxmum land holdng and force owners to dspose of all that was owned beyond ths lmt- and tenancy reform -whch had the goal of lmtng the rent to be pad for land and prohbtng tenant evctons- took a long tme. The fact that mplementaton started n earnest only after 97 allowed landlords to prepare by resumng self cultvaton, evctng tenants or transformng them nto wage workers, or mplement spurous subdvsons, and enthusasm for such reforms quckly dsappeared after 980 Appu 997. The exstence of wde varatons n legslaton across states provdes ample scope to analyze the mpact of such polces on observed outcomes. However, capturng the fne dfferences n tmng, applcablty, modaltes of mplementaton, and defntons nherent n such legslaton appears dffcult f not mpossble. Rather than tryng to do so, we use the share of households who benefted from key polces as an ndcator for polcy-nduced constrants to the operaton of rental markets. Specfcally, we construct for each state the share of households who were awarded tenancy rghts and the share of celng surplus area that was actually transferred to benefcares. As none of the Indan states permt sub-leasng of lands to whch tenants had receved permanent rghts and most states also mpose restrctons on transfers of land receved n the course of mplementng celng legslaton, ths s proxes for drect restrctons on the operaton of land rental markets. Both fgures also provde an approxmaton for a state government s level of mplementaton effort, a varable that s exogenous to households decsons but that was shown to be of great mportance n earler studes Banerjee et al. 00. Usng census fgures, Appu 997 estmates that, to avod havng to gve rghts to tenants, landlords evcted about 30 mn tenants or about one thrd of the total agrculturally actve populaton. Ths s smlar to evdence from other countres where landlords often succeeded- to evct tenants n antcpaton of legslaton to protect tenants aganst evcton or lmt the rents they would have had to paydennger 003. We use area rather than benefcares because n some cases celng surplus land was dstrbuted to a collectve entty such as a cooperatve so that the number of benefcares would be msleadng. Also, the exstence of large dscrepances between the amount of land exproprated and actually dstrbuted -whch s due to the fact that n some cases land that had been dstrbuted could not occuped by benefcares or was taken back after some tme- led us to focus on land actually dstrbuted. 3

7 Table presents summary statstcs by state for total agrcultural area and 980 populaton columns and and our measures of land reform mplementaton around 980 columns 3 and 4. 3 Relatve emphass across types of nterventon, and the extent of mplementaton vary across states. More than 0% of households receved tenancy rghts n Kerala, Gujarat, West Bengal, and Maharashtra and more than 5% of area was dstrbuted under celng surplus legslaton n West Bengal, Andhra, Maharashtra, Rajasthan, and UP. Some states e.g. West Bengal, Maharashtra heavly reled on both measures, others e.g. Andhra, Rajasthan and UP focused exclusvely on celng surplus, and some e.g. Kerala and Gujarat emphaszed tenancy laws. Although -wthout detaled knowledge on content and mplementaton effort- the number of tenancy laws enacted s at best an mperfect proxy for the number and severty of restrctons on land rental, t s a useful pont of reference. We thus nclude n the mean number of tenancy laws n each state column 5 from Besley and Burgess 000. One notes that the mean number of laws at any pont amounts to.5, from none n Haryana to more than 4 n Taml Nadu. The correlaton between the number of laws and the number of tenants who receved rghts s low ρ = 0.8, supportng the noton that legal provsons alone had lmted mpact and mplementaton effort was requred. Whle we do not separate t out n the table, a more detaled look at the tme dmenson of these measures allows a number of conclusons Kaushk 005. Frst, land reform has been a major effort; up to 000, land reform laws resulted n the transfer of almost 0 mn ha,.5 mn ha under programs to redstrbute of celng surplus land and 7.35 mn ha under tenancy legslaton. 4 Second, after a spurt of land transfers n the 970s and 980s, progress has slowed down consderably; n fact between 995/96 and 003/04,.e. for almost a decade, progress n awardng land rghts to tenants had come to a complete standstll. 5. Tenancy legslaton and rent controls: Internatonal evdence Even though emprcal evdence on the mpact of rent celngs and other forms of tenancy control n rural areas s lmted, the ssue has been analyzed n urban contexts where rent control s a textbook example for polces that can be effectve to transfer resources n the short term but wll be assocated wth neffcences n the medum to long run Arnott 003. The key reason, whch also formed the bass for analytcal approaches, s that, by fxng rents below ther equlbrum level, controls reduce the supply of 3 Wherever avalable, the level of the respectve varables n 980 and 998, respectvely, s used as a rght hand sde varable n the regressons. 4 The amount of land nvolved s much larger than what was redstrbuted n other Asan land reforms such as Japan mn has, Korea 0.58 mn has and Tawan 0.4 mn has. In terms of total area dstrbuted, ths puts Inda on par wth Mexco whch, n a much more land-abundant settng, and startng n 97, managed to dstrbute slghtly more than 3 mn ha Dennger The ncrement n celng surplus land transferred durng the perod amounted to only 0,800 ha whch s only about one tenth of the land declared celng surplus whch had not been dstrbuted. The fact that all the remander remans ted up n ltgaton suggests that further progress n achevng redstrbuton of celng land could be slow -t would take almost 90 years to dspose of remanng celng surplus cases f the current pace s mantaned- and that, by cloggng up the court system and preventng t from quckly dspensng justce n other urgent matters, the celng legslaton may mpose external effects beyond land rental markets Moog

8 new housng or mantenance of exstng unts by landlords who face an artfcally reduced prce Gyourko and Lnneman 990. Thus, although they transfer resources from landlords to sttng tenants at the tme of mposton, they make access to rental property more dffcult thereafter Basu and Emerson 000. Wth a constant or decreasng number of benefcares and an ncreasng number of new entrants who need to access to land n dstorted markets, socal cost of mantanng land rental restrctons wll ncrease over tme Glaeser 00. Whle rent controls may be useful to deal wth emergences n the short term, other polces may be more effectve Malpezz and Ball 99, have fewer undesrable sdeeffects ncludng reduced tenant moblty Munch and Svarer 00, and can be better targeted. Whle there s lttle emprcal evdence on the mpact of rental restrctons n rural areas, a number of reasons would lead one to expect that t wll go far beyond the prce effects on whch the urban lterature has focused. Frst, due to labor market mperfectons, the way n whch rural land s used wll have a clear mpact on productve effcency Bnswanger et al Second, supply of housng to urban markets wll be less elastc as owners can not revert to own- or wage labor-based cultvaton, an ssue that has been hghly relevant n Inda Appu 997. Thrd, as rural rents are generally defned as an n-knd output share, contract terms wll be less flexble than urban ones, lmtng the scope for crcumventng them by adjustng rental rates Basu and Emerson 003. Fnally, rghts gven to tenants are hertable but non-transferable and stll requre rent payment to the landlord, thus reducng both partes ncentves for land-related nvestments and the scope to ncrease allocatve effcency through sub-leasng. 6 Whle all of these ssues could sgnfcantly add to the long-term cost of land reforms, they have not featured promnently n the large lterature on Indan land reforms Warrner 969, Thorner 976, Besley and Burgess 000, Banerjee et al. 00. Explorng ths by assessng how land rental markets work and to what extent they are affected by tenancy restrctons wll allow us to get an ndcaton of such cost and help put the polcy debate on land markets n Inda on a more robust emprcal footng..3 Conceptual framework A key ratonale for producers to engage n land markets s the desre to adjust for dfferences n ther exstng endowments of land and famly labor. Followng smlar models n the lterature Carter and Yao 00, let household be endowed wth fxed amounts of labor L and land A, and a gven level of agrcultural ablty. Agrcultural producton follows a producton functon f,, l,a,a wth standard 6 The tenant wll be unlkely to nvest as dong so wll result n an mmedate ncrease of the rent whereas nvestment by the landlord s unlkely as part of the benefts wll go to the tenant. 5

9 propertes,.e. f >0, f <0 wth respect to all arguments and f la >0. 7 Relatve land scarcty, together wth the cost of supervsng labor Frsvold 994 makes wage-labor based cultvaton undesrable n equlbrum Bnswanger et al. 995, mplyng that households allocate ther labor endowment between farmng ther own land l,a and off-farm employment l,o at an exogenous wage ncurs transacton costs TC n w. Rentng of land for rentng-n and TC out for rentng-out because of the need to obtan nformaton on market condtons, to negotate and enforce payments, and the presence of regulatons that restrct transferablty or completely outlaw certan contract types. We assume transacton costs to be proportonal to the sze of land transferred allow households to structure rental contracts n a way that allows those lackng lqudty to enter nto arrangements to allow them to defer rental payments untl the harvest. Wth ths, household s decson problem s to choose A, l,a and l,o to solve l, a Max, l, o, A pf, l, A + wl I [ A A r + TC ] + I [ A A r TC ], a, o n n out out s.t. l,a +l,o L a l,a, l,o, A 0 where p s the prce of agrcultural goods, r s the rental rate, A s the operatonal land sze, b I n s a out ndcator varable for rent-n = for rent-n, 0 otherwse, I s an ndcator for rent-out = for rentout, and 0 otherwse, TC n and TC out are transacton costs, and all other varables are as defned above. Assumng that the restrctons n a hold wth equalty, the optmal choces of l *,a, l *,o and A * wll solve the frst order condtons FOC of problem,.e. pfl, a, l, a, A = w and for households who rent n A * > n A, pf, l, A = r + TC 3 A, a and for households who rent out A * < out A, pf A, l, a, A = r TC 4 and for autarkc households A * = out n A, r TC < pf, l, A < r + TC 5 A, a The frst order condtons allow dervaton of three emprcally testable propostons as follows wth more detaled proofs to be found n the appendx. 7 Note that, f,, l,a,a s subject to constant return to scale. For example, n the Cobb-Douglass case, f, l A = l A. As ablty, a s not tradable, f l A = l A s subject to decreasng return to scale n land and labor Connng and Robnson 007., a, a, a 6

10 Proposton : The amount of land rented n out s strctly ncreasng decreasng n households agrcultural ablty,, and strctly decreasng ncreasng n the land endowment A. Land rental wll transfer land to effcent, but land-poor producers, thereby contrbutng to hgher levels of productvty and more effcent factor use n the economy. For households rentng n or out land, ths proposton can be derved by totally dfferentatng both sdes of equatons and 3 or and 4 wth respect to or A. Manpulatng terms yelds A * / >0 or A * / A <0. Snce a n or a out, the amount rented n or rented out, respectvely s defned as ether A * - follows. A or A -A *, the result for a n out and a Proposton : The presence of transacton costs defnes two crtcal ablty levels l TC out,.. and u TC n,.. such that households wth ablty [ l, u ] wll reman n autarky. Any ncrease n TC n or TC out wll expand the autarky range, thus reducng the number of producers partcpatng n rental markets and the number of effcency-enhancng land transactons. Compared to a stuaton wth no transacton cost, ths wll decrease productvty and socal welfare. To see ths, note that the cutoff ponts l and u can be obtaned from 4 and 3 by settng A *= A and l,a = where s autarkc household s optmal amount of labor n agrcultural producton whch can be derved from at A * l,a * l,a * = A. Defnng f, l, a, A = l A, we can solve for l and u explctly as, a out *, r TC l = and l a A n r + TC u = * l, a A. Dfferentaton of l or u wth respect to the correspondng transacton cost varable TC out or TC n then yelds the result of nterest. Proposton 3: Increases of the exogenously gven wage for off-farm employment wll mply that hgher amounts of land are transacted n rental markets as households wth low agrcultural ablty who jon the off-farm labor market wll supply more land. Ths leads to a decrease n the equlbrum rental rate whch wll prompt hgh-ablty workers specalzng n agrcultural producton to rent n more land. To derve ths proposton, assume wthout loss of generalty that the economy contans n households who are endowed wth dentcal amounts of land and labor A and L but dfferent levels of farmng ablty and that the latter s unformly dstrbuted between and. We also assume that, wth fxed setup costs, households gve up agrcultural producton completely and rent out all ther land f the amount of labor to be allocated optmally to agrcultural producton s less than l c a. To be able to derve a closed form soluton, we abstract from transacton costs.e. lettng TC n =TC out =0 and use an explct 7

11 functonal form f, l, a, A = l A, a. Wth an exogenous wage w and prces normalzed to, the rental rate r wll be determned endogenously and frst order condtons droppng for notatonal clarty smplfy to f ' = A l l a a = w 6 ' A a = and f = A l r 7. As A * s proportonal to households farmng ablty, the market clearng condton can be expressed as A * d = na, whch together wth 6 and 7 can be used to solve for A *, l * a and r * as * na A =, + * * l a = w and r = w na na Settng l * a =l c a allows to solve for the crtcal level of ablty c below whch households wll rent out all ther land endowment. For households who contnue farm producton, we solve for the new optmal operatonal land sze A ** and new rental rate r ** based on 6, 7 and the new market clearng condton A ** d = na. Settng ** A = A allows us to solve for the farmng ablty by household who c reman autarkc au. Takng dervatves of c, A **, au and r ** wth respect to w yelds: c / w>0, A ** / w>0, au / w<0 and r ** / w<0, suggestng that, as off-farm opportuntes ncrease, a larger number of households wll drop farm producton and rent out all ther endowment, the equlbrum rental rate wll decrease, households who reman n agrcultural producton cultvate more land, and a greater number of households wll rent n land..4 Estmaton strategy Equatons -5 ndcate that producers decson to enter land rental markets depends on ther margnal productvty n autarky, MP A as compared to the rental rate to be pad r n T or receved r out T whch s a functon of transacton costs. Formally, the three regmes are characterzed by I. Rent - out regmea II.Autarky regmea > A : = A : * III.Rent - n regmea < A : * * MPA rtc out MPA + + < ε < rtc MPA + ε > rtc out n < rtc n 8 A producer s margnal product MP A, wll depend on hs or her ablty, endowment wth land A, famly labor L, assets K, and the opportunty cost of labor whch wll be affected by the level of educaton E and the presence of opportuntes n the local off-farm labor market O. Defnng a well- 8

12 behaved net earnng functon g, A, L,K,E,O wth frst dervatve g.,we can wrte a lnear verson of the latter as MP A =g, A, L,K,E,O= A + 3 L + 4 K + 5 E + 6 O. Transacton costs are expected to depend on polcy varables S, household characterstcs Z, and a tme dummy D 99. Defnng an ndex varable y such that y = f A*< A ; y = f A*= A ; y = 3 f A*> A, we can rewrte 8 as an ordered probt model that can be estmated usng maxmum lkelhood methods. 99 Proby = = Φ{ ε < η0 + ηs + ηz + η3d 0 A3L 4K 5E 6O} 99 Proby = = Φ{ η0 + ηs + ηz + η3d 0 A 3L 4K 5E 6O< ε 99 < δ0 + δs + δz + δ3d 0 A 3L 4K 5E 6O} 99 Proby = 3 =Φ{ε > δ0 + δs + δz + δ3d 0 A3L 4K 5E 6O} 9 Varables we expect to affect margnal productvty are agrcultural ablty, the dervaton of whch wll be dscussed below, a dummy for landlessness and the log of the land endowment to represent A, the number of members n the 4-60 and below 4-year age group to represent L, the value of assets and the share of agrcultural assets lvestock, mplements, and agrcultural structures for K, the head s age as a proxy for experence and a dummy for prmary educaton to represent human captal E, and mean vllage ncome O to represent wage labor opportuntes n off-farm labor markets. Transacton cost of land rental partcpaton are affected by producer s caste status, Z, a tme dummy D 99, and land polcy S whch s proxed by ether the share of households who were recognzed under tenancy reform, the share of area dstrbuted under celng legslaton, or the number of tenancy laws enacted as dscussed earler. The propostons from our model allow makng predctons on the sgns of ndvdual coeffcents. The factor equalzaton from proposton mples that rental markets wll transfer land to more productve producers >0 wth lower levels of land endowments <0 and more famly labor 3 >0. The hypothess of wealth bas n rental markets, possbly due to credt market mperfectons, translates nto 4 >0. Dversfcaton effects mpled by proposton 3 suggest that producers wth hgher levels of educaton have better off farm opportuntes and wll be less lkely to rent n land 5 <0 and that hgher levels of non-agrcultural wages, proxed by O, wll make rentng n less lkely 6 <0. Proposton mples that, by movng the cut-off ponts where producers shft from rentng out to autarky and from autarky to rentng n, respectvely, rental market restrctons expand the range of autarky but do not affect producers margnal product. 8 We thus expect η >0 and δ <0, respectvely. By the same logc, hgher transacton costs for producers from scheduled and backward castes mply η >0, and δ <0 8 It s ntutve that rental restrctons wll drectly affect whether or not households partcpate but, unless there are selectvty ssue, are unlkely to affect producers margnal productvty. Indeed testng for selectvty of rental market partcpaton, by ncludng rental restrctons n the margnal product equaton as well, does not produce conclusve results. 9

13 whle a reducton over tme n transacton costs due to better access to nformaton mples η 3 <0 and δ 3 >0. A key element of the regresson 8 s households agrcultural ablty. As the data avalable are a panel of households and ther offsprng who were observed n 98 and agan n 999, we can recover ths parameter from a panel producton functon usng household or dynasty fxed effects to proxy for ablty Dennger and Jn 007. Let technology be represented by the Cobb-Douglas producton functon Q jt θ jt θ jt θ jt exp 3 θ4 = + A L K X exp φ t 0 j jt where Q jt s the value of agrcultural output produced by household n vllage j n year t; A jt, L jt and K jt, X jt are total cultvated area, labor for crop producton, value of agrcultural assets, and amounts of chemcal fertlzer, organc manure, pestcdes, and seeds, θ, θ, θ 3, and θ 4 are techncal coeffcents, j s a tme nvarant vllage level parameter reflectng, among others, access to markets, nfrastructure, and other tme nvarant factors such as clmate, s the tme nvarant household fxed effect whch we use to measure of ablty, and t s a tme dummy so that expφt measures productvty changes over tme. To estmate ths, we let j = + j, take logarthms of both sdes, and add an d error term to obtan q jt = j +θ a jt + θ l jt + θ 3 k jt + θ 4 x jt + φt + ε jt where lower case letters are logarthms. Wth multple observatons per household, we can subtract means q jt - q j = j - + θ Z jt - Z j + φ t-t + ε jt - ε j where Z jt s a vector ncludng a, l, k, x wth coeffcents θ. As j - = 0, ths can be smplfed to q jt - q j = θ Z jt - Z + φ t- t + ε jt - ε 3 Ths can be used to obtan j ˆ j j j j, composed of a producer s dosyncratc ablty and unobserved vllage attrbutes j. Lettng the latter be the average of household fxed effect n the vllage ˆ = j ˆ. 9 j ˆ / n j j Mundlak 96 allows to obtan 3. Data sources and descrptve evdence ˆ, the producer-specfc effect by subtractng ˆ from j 9 An alternatve approach s to use a stochastc fronter producton functon to determne producers level of techncal effcency n each of the perods. Ths assumes that the dsturbance term s composed of two addtve components v and u where v s pure whte nose and u ~N + 0, δ u captures producers level of techncal neffcency TE = exp-u Coell 995. Whle the strong dstrbutonal assumpton and the fact that u captures other shocks mply that ths s nferor to the panel approach, the fact that a large number of households dropped out of the panel makes ths an attractve alternatve whch we use as a robustness check for our results below. 0

14 Before dscussng the econometrc results, we descrbe the data by presentng evdence on socoeconomc characterstcs as well as land market partcpaton and changes n these varables over tme. Ths allows us to explore changes of the economc structure, growth, and asset accumulaton by ndvduals and at the vllage level. Concernng rental markets, our nterest s to fnd out whether our hypotheses are supported by descrptve evdence or whether, possbly as a result of reverse tenancy rental markets lead to transfers from small to larger producers rather than to land-poor ones. 3. Household characterstcs The data used n the analyss come from two rounds of NCAER s ARIS/REDS survey conducted n 98 and 999, respectvely. Ths survey, the frst rounds of whch were mplemented n to evaluate the mpact of an agrcultural development program, covers all of Inda s major states. Even though the frst round sample, stratfed by farm sze and wealth class, was lmted to project areas, the survey was expanded n 98 to make t more representatve at the natonal level, coverng slghtly less than 5,000 households Foster and Rosenzweg 996. The 999 sample contans the 98 sample and replacements for those no longer present. If the orgnal household had splt, all of the households belongng to the same dynasty n the orgnal vllage plus a sub-sample of successor households outsde the vllage were ntervewed, brngng the total to about 7,500 households Foster and Rosenzweg 004. Table reports descrptve statstcs for the whole country and ts four man regons n both perods. 0 We note that there has been a marked ncrease n educatonal attanment, as llustrated by the fact that the share of household heads wth prmary schoolng completed ncreased from about 6% n 98 to 50% n 999. The gap that had earler separated Northern and Southern states also narrowed consderably. At the same tme, populaton growth led to a declne n the average land endowment, from 3.3 ha n 98 to. ha n 999, and a small ncrease n landlessness, from % to 4%. Average household sze decreased, from 6.9 to 6.0, wth about 4 n the 4 to 60 age category, 0.4 aged above 60, and.4 as compared to.8 below the age of 4. The share of female headed households s, wth 6.8%, almost the same n both perods, although female headshp s more pronounced n the South than n other regons. Survey results also pont towards an annual ncrease of per capta ncome of 3.% durng the perod under concern. Ths aggregate masks pronounced dfferences across regons wth the South havng caught up and even replaced the North as the regon wth the hghest ncome n the second perod. Use of the nformaton from the lstng exercse allows us to compute the Gn coeffcent for self-assessed ncome n a way that ncludes all households n the sampled vllages. Wth a Gn coeffcent of 0.3 n 0 We group states nto four regons as follows: The North ncludes the states of Haryana, Hmachal Pradesh, Punjab, and Uttar Pradesh; the West ncludes Gujarat, Maharashtra, Madhya Pradesh, and Rajasthan; the East ncludes Assam, Bhar, Orssa, and West Bengal; and the South ncludes Andhra Pradesh, Karnataka, Kerala, and Taml Nadu.

15 the frst and 0.3 n the second perod, nequaltes n ncome reman modest and have not apprecably ncreased between the two perods. The overall mprovement n lvng standards as llustrated by ncreased ncome s mrrored by a sgnfcant rse n asset values of approxmately 6% per year. It s worth notng that, wth a Gn coeffcent of 0.60 n 98, nequalty n assets s hgher than ncome nequalty, as found n other parts n the world as well. However, ths coeffcent has actually decreased to 0.56 n the second perod. Whle overall asset endowments have ncreased, the broad composton of households asset portfolo shows greater stablty; the house and consumpton durables make up the largest share n both perods 57.5% and 56.9%, respectvely n 98 and 999, followed by fnancal and off-farm assets 6.7% and 3.%, and farm assets ncludng lvestock 5.8% and 9.9%. The bottom panel llustrates the dversfcaton of ncome sources acheved durng the perod by hghlghtng the share of households who partcpated n varous actvtes, notng that households generally rely on more than just one ncome source. Three observatons are of partcular nterest. Frst, despte a drop by about 7 and 5 percentage ponts n the share of households engagng n crop and lvestock producton, respectvely from 70% to 63% for crop and from 78% to 63% for lvestock, self employment n agrculture remans the most mportant source of employment n Inda s rural economy. The fact that ths s followed by agrcultural wage employment, the share of whch has actually ncreased over the perod from 37.6% to 43.3% ponts to a contnued mportance of agrculture and the mportance of agrcultural wage employment for the landless. A second fndng of nterest s that the share of households engagng n self-employment n rural Inda has been more or less constant, mplyng that growth of the rural non-farm sector has been just large enough to absorb populaton growth. Ths contrasts wth other countres where the rural non-farm economy developed as a result of out-mgraton of labor or due to households takng up non-farm self employment. In fact, accordng to our data, the man source of ncome dversfcaton was growth of non-farm wage employment, partcpaton n whch ncreased from 8.6% n 98 to 0% n 999. Ths fact, whch has also been observed elsewhere Foster and Rosenzweg 004, could result from the government s emphass on creatng rural employment through mechansms rangng from drect subsdes for frms to set up n remote areas to ndustral and labor market regulaton such as small scale reservaton acts Besley and Burgess Operaton of land rental markets To provde evdence on the extent to whch our hypotheses on land markets are borne out descrptvely, Table 3 presents key varables by households land rental partcpaton status rent n, rent out, or autarky. It ponts towards a large ncrease n the level of land market actvty over the perod; from 5.3% Whle part of ths ncrease may be due to mprovements n the survey nstrument that resulted n better measurement of assets n the second perod, t s mpossble to test ths hypothess as dsaggregated data for 98 were not avalable.

16 and % for rentng out and rentng n, respectvely, n 98, the share of market partcpants has ncreased to 0.7% and 4.%, n 999. Descrptve fgures also support the propostons from our model and the noton of better rental market functonng n the second, as compared to the frst perod. Comparng the per capta land endowment for land owners who ether remaned n autarky 0.5 ha and 0.36 ha n 98 and 999, respectvely, rented n 0.8 ha and 0.0 ha, or rented out 0.68 ha and 0.64 ha n the two perods llustrates that, n both perods, rental provded opportuntes for relatvely land-scarce and labor-abundant households to gan access to land. Land markets transferred land from households wth more educated and female heads to male headed ones wth lower educatonal attanment. We also note that the share of landless who had ganed access to land through rental markets ncreased from % n the frst to 37% n the second perod, suggestng a marked expanson of outreach towards ths group over tme. Notng that our sample represents about 30 mn rural households, n 999 about 5 mn households -a quarter of them landless- were able to use markets as a means to get access to land. Ths fgure s not only much larger than the number of those who got access to land through land reform but also that, gven the magntudes nvolved, even polces wth modest mpact on the functonng of land rental markets could have mplcatons for a large number of households. Comparng levels of consumpton and assets for households who dffer n the nature of ther land market partcpaton renforces the noton that rental provdes opportuntes for poor segments of the populaton to access productve resources and thereby mprove ther well-beng, especally n the second perod. The value of all assets owned by rent-n households n 999 s, wth Rs. 33,839, more than 5% below the average, compared to asset ownershp that s smlar to the mean for autarkc households and about 33% hgher than the mean for those rentng out, supportng the noton that t s the asset-poor who beneft from the access to land whch rental markets provde. The narrowng of the gap between rent-n and average households wth respect to per capta expendture supports the hypothess of land markets makng a postve contrbuton to the lvelhood of partcpants. Fndng sgnfcant dfferences n the composton of the asset portfolo between rent-n and rent-out households, wth the former havng relatvely more of ther wealth n farmng and lvestock and the latter n off-farm and fnancal assets s not too surprsng. Whle hgher partcpaton n crop and lvestock producton by those rentng n should not come as a surprse, the hgh share of renters engagng n agrcultural wage employment suggests that land rental provdes wage laborers wth opportuntes to earn addtonal ncome. At the same tme, the fact that Whle ths s a large change, the level of rental market actvty ncreased more rapdly, and n a shorter perod, n other Asan countres such as Chna or Vetnam, despte the fact that the more egaltaran land ownershp dstrbuton n these countres would put greater lmts on the potental of land markets to equalze operatonal holdngs than n Inda. In Vetnam, the share of households rentng n ncreased form 3.8% to 5.8% n the 5-year perod between 993 and 998 Dennger and Jn 007. In Chna, the same fgure ncreased from.3% n 996 to 9.4 n 00 Dennger and Jn

17 -n contrast to what was found n 98- non-farm self employment s much hgher among rent-n households than ether the mean or those who remaned n autarky suggests that land rental s not an obstacle to partcpaton n the rural non-farm economy. To the contrary, rentng may provde an opportunty to accumulate experence and captal to replace the agrcultural ladder Spllman 99 wth a general ncrease n occupatonal moblty ncludng the non-farm sector Alston and Ferre Econometrc evdence We fnd that measures to mplement land reform, whether proxed by the number of laws passed, the area transferred under celng legslaton, or the number of benefcares from tenancy acts, constran rental market actvty, thereby reducng both equty and effcency. They make t more dffcult for producers wth hgh ablty, ncludng the landless, to gan access to land whle at the same tme consttutng an obstacle for gvng up farmng by those wth low ablty who want to get out of agrculture. 4. Producton functon To obtan a measure of households or dynastes agrcultural ablty, a producton functon, coeffcents for whch are reported n Table 4, was estmated. Although a sgnfcant number of households for whom producton s observed only n one of the perods are dropped, the specfcaton ft the data very well wth an R of 0.76 for the fxed effect estmaton, and of 0.83 for OLS wth coeffcent estmates from both beng close to each other. Concernng the ndvdual varables, land s estmated to be by far the most mportant nput to crop producton; doublng cultvated land area alone would lead to a 50% to 58% ncrease n total crop producton. Ths s followed by seed expendtures and labor use wth an estmated elastcty of 3% to 7% each. Compared to these, returns to fertlzer, pestcdes, rrgaton and assets are more moderate wth elastctes of about 5%, -3%, -%, and 4% for expendture on fertlzer, pestcdes, rrgaton and others. Whle nether educaton nor the gender of the household head are sgnfcant, land qualty matters and doublng land values, whch we use as a proxy for land qualty, would ncrease total output by -%. Sgnfcant varaton of ablty across households could mply that, even wthout a strong pull from non-agrcultural employment opportuntes, the scope for market-medated transfers to brng about effcency gans could be large. The estmated sze of technologcal change between the two perods s between 4 and 4% Determnants of land market partcpaton 3 The fact that coeffcents for the fronter producton functon are very smlar to those obtaned usng OLS and the panel ncreases our confdence n the robustness of the results. 4

18 Results from ordered probt estmaton of the rental market partcpaton equaton usng the pooled sample for 98 and 999 are reported n Table 5. 4 The three pars of columns correspond to our polcy varables,.e. recognton of tenants, dstrbuton of celng land, and tenancy laws passed. To nterpret them, recall the codng of for the rent-out, for the autarky, and 3 for the rent-n regme, mplyng that a postve coeffcent wll ncrease the probablty of rentng out. As ablty s defned only for households or dynastes who engaged n agrcultural producton n both perods, ncluson of ths varable reduces the sample and we also report regressons wthout ablty n columns, 3 and 5. The hghly sgnfcant coeffcent on ablty mples that, n lne wth expectatons, rental markets mprove productvty of land use by transferrng land from less to more effcent producers. The mpled magntude s large; accordng to the estmates, the probablty for the most effcent household n the sample to rent n s more than double that for the average household as s the probablty of rentng out for the least effcent producer n the sample. Whle lack of data on profts before and after rental partcpaton makes t dffcult to assess the net mpact on productvty, evdence from Chna, where rental helped ncrease productvty gans by some 60% Dennger and Jn 006, suggest that these can be large. We also fnd strong factor equalzaton effects. Hgher endowments wth land and lmted access to famly labor, especally n the 4-60 year age group, are estmated to ncrease the propensty to supply land to the rental market, supportng the noton that, by transferrng land to labor-rch but land-poor households, markets provde scope for ganful employment of rural labor. Sgnfcance and magntude of the coeffcent on the landless dummy suggest that rental s an mportant avenue for landless households to access land. To nterpret the coeffcents, note that, at 5.4 to 8.6 percentage ponts above that for a land ownng household, landless producer s propensty to access land through rental s almost double that of the former. The noton of land rental markets beng based n favor of the wealthy s further undermned by the lack of sgnfcance for the coeffcent on total assets. 5 A strong dversfcaton effect also emerges. The head havng completed at least prmary educaton ncreases the propensty to rent out land, e.g. because more educated ndvduals have more opportuntes outsde of agrculture. At mean values of the other varables, the probablty to rent out rent n land for a household whose head attaned prmary educaton or above s.% hgher.% lower than one wthout prmary educaton. Mean vllage ncome smlarly ncreases the tendency to rent out, 6 mplyng that as, wth overall development, the level of ncome ncreases, households wll be more lkely to move out of agrculture, supply land to the rental market, and allow those wth hgher ablty to 4 As, n most states, land reform mplementaton was essentally completed by 98, mplementaton effort after 98 s not a good measure for states serousness n mplementng land reform. The fact that ths measure would dsproportonately reward latecomer states, together wth the challengng econometrc nature of fxed effect ordered probt estmaton, leads us to use the pooled sample throughout for our estmatons. 5 Incluson of an nteracton between the tme dummy and asset ownershp not reported suggests that land rental markets had been based n favor of the wealthy n 98 but that, presumably due to better credt market access n the study areas, ths bas had dsappeared by The coeffcent on mean vllage ncome loses sgnfcance f ablty s ncluded, presumably due to the much lower number of observatons. 5

19 ncrease ther holdngs and ncome levels. Ths phenomenon, whch has also been observed n Chna, hghlghts the mportance of rental markets wth economc development and suggests that mposng restrctons on the operaton of such markets could not only nflct effcency losses but also become more dffcult over tme. Regardng the lower bound equaton, regressons suggest that polcy restrctons wll lead to a sgnfcant and quanttatvely large reducton of land supply to rental markets. Estmated effects are weakest for the number of laws columns 5 and 6 and strongest for recognton of tenants col. and, consstent wth the noton that mplementaton s more mportant than mere legslaton and that landlords wll be less wllng to rent out f dong so can attenuate ther property rghts or f there are lmts on ther ablty to negotate the amount of rent. 7 The share of land affected by redstrbuton of celng land s n the mddle between these two, consstent wth expectatons that celng legslaton poses less of a threat than tenancy regulaton -as the latter apples to all market partcpants rrespectve of ther holdng szeand enforcng t s more poltcally controversal and admnstratvely complex than mplementng tenancy legslaton. Coeffcents on other varables suggest that, even after adjustng for polces, scheduled and backward castes are less lkely to rent out land than the remander of the populaton, possbly due to lower levels of socal captal and less opportunty to fnd partners n rental markets or to enforce protecton of property rghts for rented land. The hghly sgnfcant and postve coeffcent on the 999 dummy llustrates that, over tme, land rental supply ncreased sgnfcantly, possbly due to nformaton access. To quantfy the mpact of polcy restrctons we compute, for every household, the predcted probablty to rent out wth actual values for all rght hand sde varables and wth the tenancy restrcton varable takng a value of zero. Takng the dfference between these two values as a measure for the mpact of tenancy restrctons suggests that ther removal could lead to a consderable ncrease n rentng out, by between 40% and 70%. Turnng to the upper bound between autarky and rentng n, the postve coeffcent on all the polcy varables suggests that over and above the reducton n supply they also depressed demand, makng t more dffcult for households to obtan land through rental. Across polcy varables, those relatng to the ntensty of enforcement are agan more sgnfcant; n fact the number of laws, whle of the expected sgn, s not dfferent from zero at conventonal levels. In most equatons, coeffcents are bgger for the upper as compared to the lower bound, suggestng that the mpact of polcy-nduced restrctons wll be larger on the demand sde. Applyng the methodology used earler for the lower bound to assess predcted mpacts of elmnatng polcy restrctons suggests that, wth a coeffcent much larger than for the upper bound, removal of tenancy restrcton could double partcpaton rates by rent-n households. 7 Note that the coeffcents on the number of laws and the amount of land dstrbuted va celng laws lose sgnfcance n the smaller sample. 6

20 Consstent wth what emerged for the upper bound, we also fnd that backward and scheduled castes are more lkely to reman n autarky and that over tme, the sze of the autarky area has decreased,.e. land rental markets have become more actve. Whle ths s encouragng sgn and suggests that tme may partly offset undesrable mpacts of rental regulaton, the magntude of the coeffcents s small; estmated coeffcents mply that almost a century wll be requred to fully offset these effects. Together wth case studes suggestng that crcumventng such legslaton s easer for the rch than for the poor Yugandhar 996, Thangaraj 004, ths mples that passage of tme s unlkely to elmnate the negatve effects of tenancy regulaton or do so n an equtable manner. To test the robustness of our conclusons, we compare results those derved from usng the productvty measure derved from the fronter producton functon nstead of the panel. Dong so ncreases the sample sze by two thrds but leads to very smlar substantve conclusons; n partcular provdng strong support to the factor equalzaton and dversfcaton effects as well as the negatve mpact of polcy restrctons Appendx Table. 4.3 Effcency mpact of land rental restrctons Results thus far hghlght that polcy restrctons negatvely affected effcency by reducng overall rental market and mpedng access to land by the poor. Interactng polcy varables wth producers estmated productve effcency allows more detaled exploraton of rental restrctons mpact on effcency. Results from dong so, wth three columns agan representng three types of polcy varables, are reported n Table 6 where man equaton coeffcents are very consstent wth those reported n Table 5. The postve and sgnfcant coeffcent of the nteracton wth tenancy and celng mplementaton n the upper bound equaton bottom panel suggests that land rental restrctons have prevented land access by the most effcent producers. One explanaton consstent wth ths s that sttng tenants who already own land but are not necessarly the most effcent beneft from tenancy regulaton at the cost of more productve and landpoor producers who are constraned by tenancy restrctons and unable to effectvely express ther demand n the market. 8 Thus even f they had postve dstrbutonal effects, land rental restrctons would be dffcult to justfy as t should be possble to use the productvty gans form ther elmnaton to compensate losers. The fact that, accordng to our analyss, ths s not the case mples that an elmnaton of such restrctons would beneft everybody. The fact that there are lkely to be further dynamc neffcences as the negatve productvty effects from rental restrctons wll accumulate over tme -for example wth generatonal shft, e.g. f chldren do not want to contnue n farmng as n the case of the Phlppnes Dennger et al. 00- makes the case for ther removal more compellng. 8 The fact that nteractons reman nsgnfcant throughout n the lower bound equaton wth the polcy varables even losng sgnfcance for celng land redstrbuton- s not unexpected as the data do not allow us to obtan a measure of productve effcency for all producers who rent out all ther land n the second perod. 7