CASH TRANSFER PROGRAMS WITH INCOME MULTIPLIERS: PROCAMPO IN MEXICO

FCND DP No. 99 FCND DISCUSSION PAPER NO. 99 CASH TRANSFER PROGRAMS WITH INCOME MULTIPLIERS: PROCAMPO IN MEXICO Elsabeth Sadoulet, Alan de Janvry, and Benjamn Davs Food Consumpton and Nutrton Dvson Internatonal Food Polcy Research Insttute 2033 K Street, N.W. Washngton, D.C. 20006 U.S.A. (202) 862 5600 Fax: (202) 467 4439 January 2001 FCND Dscusson Papers contan prelmnary materal and research results, and are crculated pror to a full peer revew n order to stmulate dscusson and crtcal comment. It s expected that most Dscusson Papers wll eventually be publshed n some other form, and that ther content may also be revsed.

ABSTRACT Cash transfer programs nduce multpler effects when recpents put the money they receve to work to generate addtonal ncome. The ultmate ncome effects are multples of the amounts transferred. Ths paper analyzes the PROCAMPO program n Mexco, whch was ntroduced to compensate farmers for the antcpated negatve effect of the North Amercan Free Trade Agreement (NAFTA) on the prce of basc crops. The transfer rules and the tmng of the panel data collected allow unque control of bases n ths mpact analyss. We fnd that the multpler among ejdo sector recpents s n the range of 1.5 to 2.6. Multplers are hgher for medum and large farm households, low numbers of adults n the household, nonndgenous backgrounds, and households located n the Center and Gulf regons. Hgh multplers reflect margnal ncome opportuntes that were unrealzed due to lqudty constrants that the transfers eased. Opportuntes came from the asset endowments that these households have, partcularly rrgated land, and these opportuntes were enhanced by access to techncal assstance.

CONTENTS Acknowledgments...v 1. Prmary and Secondary Effects of Socal Assstance Programs...1 2. The PROCAMPO Program...5 3. The Ejdo Sector and the Data...9 4. Estmatng PROCAMPO Income Multplers...12 The Econometrc Procedure... 12 The Overall PROCAMPO Multpler... 15 Senstvty Analyss... 18 Whch Households Are More Effectve n Generatng Income From PROCAMPO Transfers?... 20 5. PROCAMPO Transfers and the Intensfcaton of Agrculture...21 6. The True Contrbuton of PROCAMPO to Income Mantenance...24 7. Concluson...25 Tables...29 References...38 TABLES 1. PROCAMPO n household ncome, ejdo households, 1997...30 2. Sources of ncome, ejdo households, 1994 and 1997...31 3. Change n total household ncome between 1994 and 1997...32 4. PROCAMPO multplers...33 5. PROCAMPO multplers for selected groups of recpents...34 6. Effect of PROCAMPO on agrcultural actvtes, partal results...35

v 7. Effect of PROCAMPO on labor market partcpaton and self-employment...36 8. Contrbuton of PROCAMPO n ncome changes, 1994-1997...37

v ACKNOWLEDGMENTS The authors are grateful for comments receved from Nora Lustg, Juan Carlos Martnez, and Reed Hertford. All omssons and errors are solely the authors responsblty. Elsabeth Sadoulet and Alan de Janvry Unversty of Calforna at Berkeley and the World Bank Benjamn Davs Internatonal Food Polcy Research Insttute

1 1. PRIMARY AND SECONDARY EFFECTS OF SOCIAL ASSISTANCE PROGRAMS Socal assstance programs are ntended to nduce behavoral responses among benefcares, and n most programs the degree of these responses are fundamental to how well the prmary objectves of the programs are satsfed. Behavoral responses also create a wde range of ndrect effects that need to be factored nto the evaluaton of the programs, as they may contrbute sgnfcantly to the overall welfare mpacts (Subbarao 1997a). These ndrect effects can be qute dffcult to dentfy and measure, partcularly when they are dffused over a wde range of undertakngs and f they occur over long perods of tme. Some of the ndrect effects of the man types of socal assstance programs currently used by governments and nternatonal development agences nclude the followng. 1. Food subsdy programs. The prmary objectve of these programs s to ncrease the nutrtonal status of benefcares. The ncrease n food ntake s typcally sgnfcantly less than the food receved f transfers are nframargnal and, when supramargnal, f there are secondary markets for the food receved. However, these programs are also desgned wth a perspectve on the ndrect effects that hgher food consumpton has on the health of household members, the schoolng achevements of chldren, and the labor productvty of adults (Behrman 1996). There s an extensve lterature on the measurement of such drect and ndrect

2 effects (Pnstrup-Andersen 1988). There are, for nstance, concerns about the potental negatve ndrect effects of these programs, such as the dsncentve to work that they create for recpent households (see, for nstance, Sahn and Alderman (1995) on the Sr Lanka food stamps program). 2. Employment generatng programs. Publc works programs have been used extensvely as countercyclcal nterventons, e.g., to provde employment n the agrcultural off-season or to compensate for negatve ncome shocks such as drought or recesson (Grosh 1994; Subbarao 1997b). The prmary ntended beneft s to provde recpents wth a means of consumpton smoothng. The ndrect effects are usually of a publc goods nature, and the dstrbuton of the benefts depends on the type and qualty of the asset created. For example, the constructon of schools and clncs, rrgaton nfrastructure for small farmers, and roads all beneft dfferent groups. In programs lke food for work, where the work conssts of ntroducng sol conservaton practces on land owned by the benefcares (e.g., Plan Serra n the Domncan Republc; see de Janvry, Sadoulet, and Santos 1995), there s a prvate ndrect effect through ncreased (future) ncome. 3. Credt programs. The prmary objectve of these programs s to ncrease ncome for borrowers after loans have been repad (Morduch 1998; Ptt and Khandker 1998). Indrect effects of credt programs derve from factor reallocatons nduced by the relaxaton of lqudty constrants such as mproved chldren's educaton f they are releved from farm work that competes wth gong to school. Other

3 ndrect effects that are not easly measurable derve from ncreased busness selfconfdence for partcpatng women, the nducement of greater nteractons and the creaton of socal captal among recpents, and the development of new actvtes on a collectve bass by members of credt groups. Negatve effects can nclude the removal of chldren from school f loans are used to acqure captal equpment wth hgh rsk of moral hazards when used by hred labor (see Wydck 1999 for the response of weavers to credt programs n Guatemala). 4. Cash transfer programs. For these programs, the prmary objectve s smply to rase ncome through the cash receved. There are, however, many derved effects from the transfers that are dependent on behavoral response. Several studes have, for nstance, focused on the potental negatve ndrect effects of cash transfers on work effort, and on the declne n prvate transfers receved by the targeted households f these transfers are crowded-out by publc transfers (Cox and Jmenez 1992; Cox, Eser, and Jmenez 1998). To our knowledge, there are no studes of the postve ndrect effects of cash transfer programs va ncome generaton by puttng the cash transferred to work. Yet, f the household s lqudty constraned and hence has underemployed and ll-allocated productve assets relatve to an unconstraned stuaton, the cash transfer should generate benefts at least smlar to a credt program and expectedly hgher, snce there s no rsk of falure to repay. The money transferred can be used to purchase current nputs or to nvest n physcal and human captal. Whether short-term ncome effects can be observed depends on the maturaton tme of the expenses. Expenses

4 on nputs for agrcultural, commercal, or mcro-manufacturng actvtes wll have effects vsble n the short run, as opposed to nvestments n equpment and especally n human captal that requre longer maturaton perods. In ths paper, we analyze the drect and ndrect ncome effects of cash transfers to Mexcan farm households n the context of PROCAMPO (Program for Drect Assstance n Agrculture). Ths program was ntroduced to compensate for the antcpated negatve prce effects of trade lberalzaton on basc crops. We focus on the ejdo sector, a large sector of generally mpovershed households that receved access to land through the extensve land reform program. These households are n the unusual poston of beng endowed wth productve assets whle at the same tme severely starved for access to credt due to the ncomplete nature of property rghts n the ejdo that prevents them from usng the land as collateral to access credt. In ths context, cash transfer programs can be expected to have partcularly large multpler effects on ncome. In addton, they can be expected to lead to vsble labor reallocaton effects toward the actvtes where the cash receved s used f household labor was partally allocated as a strategy to overcome credt market falures. We calculate the magntude of the ncome multplers created by these transfers. We also dentfy under what condtons and for what types of households these multplers were largest. Ths, n turn, provdes gudelnes for the management of transfer programs to rural poor households n order to maxmze multpler effects.

5 Assessng the mpact of a program s usually plagued wth the dffculty of controllng for bases arsng from unobserved ndvdual or regonal characterstcs that are correlated wth program placement and program partcpaton. In the case of PROCAMPO, systematc natonal coverage elmnates the standard program placement bas. Partcpaton of elgble households s almost unversal. However, elgblty tself and the amount that households receve are explctly related to ther croppng patterns. There s no doubt that unobserved household attrbutes affect both the croppng pattern, and hence the PROCAMPO transfers, and the outcome of nterest, namely household ncome. Yet, we are able to explot a truly exceptonal stuaton where the bass for PROCAMPO transfers are the 1993 croppng patterns, whle the program tself started only n the Fall of 1994, and households n the panel were surveyed n the Sprngs of 1994 and 1997. As the 1994 survey took place pror to the PROCAMPO transfers, but after the rules were set, a household fxed-effects estmaton crcumvented the problem of household unobservables that could bas the estmated mpact of the PROCAMPO program. Varablty n the amount receved by the dfferent households then allowed an estmate of the margnal effect of one unt of transfer. 2. THE PROCAMPO PROGRAM As a consequence of the North Amercan Free Trade Agreement (NAFTA), trade lberalzaton for basc crops compettve wth U.S. and Canadan exports was antcpated to create a sharp declne n domestc prces for Mexcan producers as prces for these

6 crops had been supported above border prces. The PROCAMPO program was ntroduced when NAFTA started (the Wnter of 1994) as a compensatory ncome transfer targeted to these crop producers. The objectves were poltcal (to manage the poltcal acceptablty of the free trade agreement among farmers), economc (to provde farmers wth lqudty to adjust producton to the new set of relatve prces), and socal (to prevent an ncrease n already extensve levels of poverty among smallholders and a rapd process of outmgraton to the ctes and the border n the North). The program was desgned as a 15-year transton toward free trade. Transfers are on a per-hectare bass, decoupled from current land use, and fxed across the whole country. Hectares that qualfy were planted n any of nne basc crops (corn, beans, rce, wheat, sorghum, barley, soybeans, cotton, and cardamom) n one of the three agrcultural years precedng August 1993. Snce there are two agrcultural cycles per year (Fall Wnter and Sprng Summer), payments are made twce a year for the area that had been planted n the correspondng cycle. Payments are hence qute dfferent across households, but exogenous to current behavor. The only restrcton s that land must currently be used n crops, lvestock, or forestry, or be part of an approved envronmental program (as opposed to beng left dle), wth freedom to choose among these optons. Elgblty, both at ncepton as well as annually, s verfed by local SAGAR (Mnstry of Agrculture, Lvestock, and Rural Development) offcals, most often n conjuncton wth muncpal or ejdo authortes. PROCAMPO s a cash transfer program of sgnfcant magntude. Snce ts ncepton n 1994, t has covered, on average, 14 mllon hectares a year, ncludng more

7 than 95 percent of the area that had been planted n corn, beans, sorghum, and wheat. Payments are made to approxmately 3 mllon producers a year, for a total expendture n 1998 of US$919 mllon (SAGAR 1998). 1 The compensatory payments are regressvely dstrbuted n the farm sector, as they are proportonal to the area that had been planted n these crops. The 45 percent of producers wth farms smaller that fve hectares thus receve only 10 percent of the total PROCAMPO transfer (SAGAR 1998). However, transfers are progressvely dstrbuted on a per-hectare bass, snce they are unform per hectare, unrelated to the yelds that were acheved and to whether households were sellng basc crops before NAFTA, and hence were to be negatvely affected by the expected declne n prces or not. Transfers thus reach producers who had never benefted from pre-nafta prce support programs due to lack of marketed surplus (Martnez 1999). In 1997, transfers represented, on average, US$329 per recpent and US$68 per hectare. Ths represents 46 percent of the gross maze ncome for a farmer who obtaned the average yeld of 1.06 tons per hectare and the average prce of US$140 per ton observed n the ejdo. These payments were to reman constant n real terms for the frst 10ten years, then phased out over the remanng fve years of the program. However, the real value of payments was not fully mantaned, as t was left to erode from US$102 to US$68 per hectare between 1994 and 1997. In spte of ths, transfers stll represent a very sgnfcant cash contrbuton to farm households, partcularly the poor, wth the potental 1 PROCAMPO s supported by a loan from the Inter-Amercan Development Bank.

8 of not only addng mportantly to ther ncomes but also affectng ther behavor as farm producers, workers, and entrepreneurs n other ncome-generatng actvtes. PROCAMPO s all the more mportant, gven the severe scarcty of formal credt n the agrcultural sector. Access to formal sources of credt dropped drastcally from 1994 to 1997. The percentage of ejdataro households that used formal credt fell from 25 to 11 percent. Furthermore, the amount avalable for ejdo agrculture from formal sources fell over ths perod. Whle n 1994 formal sources granted 134 pesos per hectare, by 1997 ths had fallen to 40 pesos per hectare overall (n 1994 pesos). Overall, average loan sze fell from 534 pesos per hectare n 1994 to 377 pesos per hectare n 1997. Current partcpaton n PROCAMPO s lmted to the households that were ncorporated nto the program when t was ntroduced n 1994. At that tme, farmers had to show that they had planted at least one of the nne staple crops durng the 1991 93 agrcultural cycles. Under PROCAMPO, elgble farmers must go at each agrcultural cycle to one of more than 700 CADER (Centro de Ape al Desarrollo Rural) offces around the country and solct ther PROCAMPO payments. The maxmum quantty of land for whch they may receve transfers s equal to or less that the area they had regstered n 1994. Payments are, n most cases, dstrbuted as checks from CADER offces. PROCAMPO qualfcaton certfcates can also be used as collateral aganst whch to borrow from commercal banks or nput retalers, gvng benefcares flexblty n the tmng when cash s avalable aganst the cost of the nterest charged.

9 3. THE EJIDO SECTOR AND THE DATA In ths paper, we analyze the mpact of the PROCAMPO program on households n the ejdo sector. The ejdo sector was the product of the sweepng land reform that followed the peasant-led revoluton of 1910. It contans approxmately 60 percent of the Mexcan rural populaton, half the country s agrcultural land, and half ts rrgated land (Lamartne Yates 1981). In terms of socal welfare, t s a major reservor of rural poverty and an mportant source of mgrants to the Unted States. Ths sector has been affected by mportant reforms snce 1990 (DeWalt and Rees 1994). They nclude both global reforms affectng the context where ejdataro households operate (trade lberalzaton and NAFTA; generalzed scalng down of subsdes), and reforms drectly targeted at the sector (ntroducton of ndvdual property rghts over land plots formerly n usufruct; scalng down of offcal credt, marketng, and techncal assstance servces provded to the ejdo by specalzed state agences; devoluton of control over ejdo affars to the communty; and greater freedoms for ndvdual ejdataros n makng decsons about ncome strateges). The data we use are derved from a natonwde panel survey of ejdo communtes and ejdataro households wthn these communtes. The data were collected n 1994 by the Mexcan Mnstry of Agraran Reform and the Unversty of Calforna at Berkeley (see de Janvry, Gordllo, and Sadoulet 1997) and n 1997 by the Mexcan Mnstry of Agraran Reform and the World Bank (see World Bank 1998). The data characterze

10 resource use and ncome formaton by households. The sample wth complete panel nformaton on ncome ncludes 958 households. 2 The vast majorty of ejdo households have access to PROCAMPO. As the data n Table 1 show, over 86 percent of the households n the survey had receved PROCAMPO transfers n 1997. Transfers reach equally small and large landholders. More dfferences emerge when vewed by regon. The lower shares of households recevng PROCAMPO n the North Pacfc and n the South are due to hstorcal croppng patterns outsde the nne basc crops covered by the program. The drect value of PROCAMPO transfers represents, on average, almost 8 percent of 1997 ncome for all households n the survey. Whle PROCAMPO transfers show some regressvty wth respect to farm sze, relatve mportance of transfers s reversed n the other asset endowments, representng a hgher share of ncome for the ndgenous populaton than for nonndgenous households, and for households wth low levels of labor, educaton, and mgraton assets. Note that these values underestmate the total effect of PROCAMPO on ncomes as they neglect the ndrect effect of PROCAMPO transfers acheved through the ncome multplers that we wll analyze below. Durng the perod under study, total household ncome ncreased by 14 percent. Hence, drect PROCAMPO transfers represent more than 60 percent of the regstered ncrease n ncome. PROCAMPO transfers served as an mportant compensatng 2 Data for 1994 and 1997 were constructed n a smlar fashon, and nclude wage and other off-farm actvtes, agrcultural and lvestock actvtes, remttances, ejdo ncome, rentals, and government programs such as PROCAMPO. Farm producton not sold was valued at an average shadow prce. Some adjustments were made n order to compensate for such problems as mssng producton cost data, for example.

11 mechansm for the larger landholders and for households wth low labor, educaton, and mgraton assets for whom the observed change n ncome was less than the PROCAMPO transfer. Drect PROCAMPO transfers, however, fell short of compensatng for the fall n ncome n the North-Pacfc regon, where agrculture s more technologcal and dversfed. The data n Table 2 show the structure of household ncome by source for 1994 and 1997. There are several remarkable facts to be noted. One s that even though all households are landed, the share of total ncome that derves from nonfarm actvtes s very hgh, and t rose from 47 percent to 55 percent durng the perod analyzed. Ths ncrease s n part due to the PROCAMPO program, whch dd not exst n 1994, and provded, on average n 1997, 7.7 percent of total household ncome and 14 percent of nonfarm ncome. In the perod, ncome from agrculture declned due to adverse prce ncentves, and wage ncome stagnated as unemployment n Mexco rose wth the aftermath of the peso crss. By contrast, self-employment ncome rose and remttance ncome from the Unted States ncreased due to the double ncentve of poor agrculture and labor market condtons n Mexco enhancng mgraton and a sharp deprecaton of the real exchange rate wth the U.S. dollar that drastcally ncreased the purchasng power of dollar remttances. In what follows, we analyze the ncome multpler effects of the PROCAMPO transfers by dentfyng the determnants of ncome change between 1994 and 1997. We measure the multpler effect of PROCAMPO on total household ncome usng several alternatve econometrc specfcatons. We then calculate ths multpler for specfc

12 subsets of the populaton and for each ncome source. Fnally, we track the orgns of the PROCAMPO multpler n agrculture by analyzng how PROCAMPO has nduced greater use of purchased chemcal nputs. 4. ESTIMATING PROCAMPO INCOME MULTIPLIERS THE ECONOMETRIC PROCEDURE As for any program mpact evaluaton, t s crucal to properly control for bases that could come from endogenous partcpaton n the PROCAMPO program. In ths case, not only s PROCAMPO partcpaton not random but, for a partcpant, the magntude of the cash transfer s drectly determned by the household s hstorcal behavor n ts choce of croppng patterns. Consder the followng ncome equaton that relates ncome y t n year t = 1997 to the household s asset endowment and characterstcs z t, the level of PROCAMPO transfer receved P t, the effect of unobservables φ t, and a random effect ε t : y 97 = z 97 97 97 97 β + αp + φ + ε 97 where β 97 s the vector of margnal return to the assets n 1997 and α the PROCAMPO ncome multpler. Unobservables nclude both household characterstcs such as manageral ablty or land qualty and external factors such as local condtons or government programs other than PROCAMPO. The standard problem n estmatng the

13 PROCAMPO multpler wth such an equaton s the potental bas on α created by a correlaton between any unobservable and the PROCAMPO transfer P. In the mpact assessment lterature, the potental sources of correlaton are classfed under headngs referred to as program placement and household selecton (Ptt, Rosenzweg, and Gbbons 1993; Ravallon and Wodon 1998; Ravallon 1999). A program placement bas may occur f there s any systematc geographcal bas n the way the PROCAMPO program reaches the populaton. The survey data confrm that ths s not the case. All ejdos except 11 have been reached by PROCAMPO, and for eght of these, the reason for not recevng PROCAMPO transfers was nonelgblty of the ndvdual households. Hence only three ejdos, comprsng 11 households (.e., 1.2 percent of the sample), may have been affected by lack of access due to unequal PROCAMPO reach. A household selecton bas occurs f the ndvdual partcpaton to the program or the amount of transfer receved s correlated to unobserved characterstcs, be t through the explct rules of the program tself or from self-selecton of households that do not partcpate despte ther elgblty. Reasons gven by households for not partcpatng were collected by enumerators: 45 percent say they do not qualfy for PROCAMPO transfers, 10 percent that the transacton s too cumbersome to be worth the cost, and 22 percent that they dd not know about the program. One clearly cannot assume that lack of knowledge of the program or complanng about ts functonng s not correlated wth determnants of ncome. Ths self-selecton s thus a potental source of bas. As for the

14 elgblty rule, PROCAMPO transfers are proportonal to the area cultvated n nne basc crops n 1993. The croppng pattern n 1993 s a household decson taken jontly wth all other decsons that determne the ncome of the household. It s therefore the functon of the characterstcs and the assets of the household n 1993 (z 93 ), and of the unobservables φ 93. Hence, any correlaton between unobservables n 1993 and 1997 would create a selecton bas n the estmaton of the PROCAMPO multpler n a smple cross-secton analyss. To elmnate ths potental bas, we wrte the ncome equaton n dfference usng the panel data for 1994 and 1997. Snce there was no PROCAMPO program n 1994, dfferencng gves y 97 y 94 = z 97 β 97 z 94 β 94 + αp 97 + 97 94 97 94 ( φ φ ) + ( ε ε ). 97 94 The dfference n unobservables ( ) φ only captures unobserved events that φ have occurred durng the perod 1994 1997, whle the PROCAMPO transfers are determned on the bass of the nformaton avalable n 1993. Can there be any correlaton between the dfference n unobservables and P? The tmng of the decson ensures that PROCAMPO transfers cannot be nfluenced by these unobserved events. Yet the opposte may happen: as PROCAMPO transfers have been extended snce 1994, some changes n household assets may result from past PROCAMPO transfers. For the analyss to be vald, all productve assets that could have been accumulated wth PROCAMPO transfers between 1994 and 1997 must

15 consequently be observed and not reman among the unobserved φ factor. One mportant asset that s partcularly senstve to wndfall ncome s lvestock. It s therefore essental that lvestock assets at the begnnng of the survey years be ncluded n the set of asset varables z. Incomplete nformaton on lvestock rases some problems n that respect. We wll return to ths pont after we present the basc estmaton results. To summarze, we estmate the equaton, y 97 y 94 = z 97 97 94 94 97 β z β + αp + η, where η combnes dfference n unobservables and error term, and s uncorrelated wth z 97, z 94, and P 97. Note that as the envronment and notably prces has changed between the years of the two surveys, we do not assume constant returns to assets and let the coeffcents vary wth the year. For the household characterstcs that are nvarant over tme, only the dfference between the two parameters β 97 and β 94 s estmated. For the characterstcs that changed over the perod, the estmaton of two separate parameters allows us to dstngush the effect due to the change n the return to any asset z as captured by the dfference between the parameters and the effect due to the change n asset poston (Oaxaca 1994; Bourgugnon, Ferrera, and Lustg 1999). THE OVERALL PROCAMPO MULTIPLIER The set of z varables ncludes land (rrgated and ranfed land, pasture, and the household share of common property land), lvestock, human captal assets (gender and age of household head, number and average educaton of adults), and socal assets

16 (Mexco and U.S. mgraton assets, ethncty, and access to techncal assstance and to formal credt). A household s mgraton assets characterze both the hstorcal mgraton and the current permanent mgraton of household members. The constructon of ths varable s based on nformaton common to the two surveys. 3 Hstorcal mgraton s measured by the number of household members who had mgrated earler but had returned home at least two years pror to the survey. The current permanent mgraton s measured by the number of chldren of the household head who are permanently establshed away from home. Note that no household members currently n temporary mgraton are ncluded n these assets. Ths s because current temporary mgraton s a household decson jontly taken wth all the other choces that contrbute to the formaton of ncome. Among ejdataros, access to techncal assstance and to formal credt are essentally supply-determned and hence are consdered exogenous to household decson-makng. In addton to these assets, regonal effects are added to control for geographcal characterstcs such as land qualty, weather, and local level of economc development. An mportant ssue wth agrcultural household ncome s ts extreme volatlty, due to large fluctuatons n weather condtons. Ths creates several econometrc problems. The frst s the presence of a large number of observatons that are clear 3 A more complete specfcaton of mgraton assets, ncludng both famly and socal networks, was used and shown to be mportant n the decson of a household to send mgrants and hence n recevng remttances (Wnters, de Janvry, and Sadoulet 1999). However, lack of comparable data on the extended famly n 1994 and 1997 forced us to reduce the varable to the famly members n ths analyss.

17 outlers. The second s the fact that, snce these fluctuatons are weather related, they are lkely to be correlated across observatons from the same geographcal area. The thrd s a more standard problem of heteroscedastcty as the volatlty of ncome s drectly related to the agrcultural ncome tself and hence lkely to land assets. To address these potental problems, we estmate the ncome equaton model wth Robust Regresson and Least Absolute Devatons (LAD) (or medan) estmators. The Robust Regresson screens out or dscounts outlers by weghtng observatons. It s an teratve process n whch the calculaton of weghts s based on the absolute resduals of the prevous teraton. The LAD estmator does not assume any specfc dstrbuton of the resduals η and gves consstent estmates even n the presence of heteroscedastcty and nonndependent resduals. LAD estmators are also less senstve to outlers than OLS, because they mnmze the devatons around the medan rather than the square of the devatons around the mean. Whle LAD estmators seem to perform well n large samples, the standard devatons of the parameters are, however, usually large for small samples, whch s our case. Hence, results from the two estmaton technques have dstnct advantages and nconvenences, and should be looked at as mutually renforcng. The results from the Robust Regresson estmaton, reported n Table 3, show that household ncome s mportantly determned by rrgated and ranfed land assets, number of adults, and access to techncal assstance, and n 1997 by US mgraton assets and adult educaton. Note that, as condtons for agrcultural producton deterorated n 1997, land assets have lost mportance n ncome determnaton compared to 1994. In contrast,

18 human captal assets and mgraton assets, whch are both sources of off-farm ncome, have ganed n mportance. Geographcally, the regon that benefted most durng the perod s the Gulf and the regon that dd worse s the North Pacfc. Access to cash transfers through PROCAMPO creates postve externaltes on ncome change, wth a 1 peso transfer nducng a drect ncrease of 1.97 pesos as estmated wth Robust Regresson. The correspondng 95 percent confdence nterval reported n Table 4 s [1.5 2.6]. Hence, the margnal ncome effect of a 1 peso ncome transfer through PROCAMPO on benefcary households s hgh. Ths s assocated wth PROCAMPO helpng relax the lqudty constrant on farm households. SENSITIVITY ANALYSIS A partcular ssue arses wth the measure of the lvestock asset. Lvestock s both a productve asset and a flexble savngs nstrument. Therefore, the herd sze that generates the lvestock ncome n 1997 may tself have been partly acqured wth the current year PROCAMPO transfer. Hence the estmaton of the equaton, y 97 y 94 = z 97 97 94 94 97 97 94 94 97 β z β + zl βl zl βl + αp + η, where zl and βl refer to the lvestock asset, would tend to underestmate the mpact of PROCAMPO. An alternatve specfcaton s to use the herd sze n the prevous year. However, snce the 1993 stock was not observed, usng nstead the 1994 stock ntroduces a bas, as follows: y 97 y 94 = z 94 93 94 97 ( zl zl ) βl + αp η. 97 97 94 94 96 97 94 94 β z β + zl βl zl βl + +

19 94 93 Is the mssng term ( ) zl,.e., growth of the lvestock herd n 1993, zl correlated wth the PROCAMPO transfer? Descrptve statstcs ndcate postve but low correlaton between PROCAMPO transfers and herd sze or growth n herd sze. An estmaton of the equaton above would thus tend to gve an overestmaton of the mpact of PROCAMPO. Estmatons of both equatons, gvng potental under- and overestmaton of the parameter, respectvely, are reported n Table 4. They show very smlar values for the PROCAMPO parameters, ndcatng that no bas s ntroduced when usng the 1997 and 1994 lvestock herd szes. To check on the robustness of the estmaton of PROCAMPO multplers, we use several alternatve econometrc specfcatons. Table 4 compares estmates for the PROCAMPO multpler under robust regresson, medan regresson, and ordnary least squares (OLS). In each case, we report the 95 percent confdence nterval and the test of whether the parameter s sgnfcantly greater than one at 99 percent,.e., that there s a sgnfcant multpler effect. We see that whle the OLS and the LAD pont estmators are hgher than the robust regresson multpler, they have larger standard devatons. Overall, however, t s the remarkable smlarty of these values across econometrc estmators that gves confdence n the robustness of the large multpler effects observed for the PROCAMPO transfers.

20 WHICH HOUSEHOLDS ARE MORE EFFECTIVE IN GENERATING INCOME FROM PROCAMPO TRANSFERS? We can dentfy who n the heterogeneous ejdo populaton was able to derve greater advantage from the PROCAMPO transfers by comparng the ncome multpler across populaton subsets. Results are presented n Table 5. One expects that the multpler should be greater when a household has more assets and when they are more underused due to greater lqudty constrant. Ths suggests that nether the households wth very low asset endowments, nor the best endowed households who may face less severe lqudty constrants, would beneft as much as a group ntermedate between the two. Ths s exactly what the multpler by farm sze ndcates (classes are defned on land use n 1994, as land use n 1997 s endogenous): the multpler s 0.24 on smaller farms, 2.77 on medum farms, and 2.04 on the larger farms. In terms of human captal assets, the multpler s hgher for households wth a smaller labor force (2.75 versus 0.93), snce the lqudty constrant s more bndng on them as they could engage less n compensatory actvtes that serve as sources of lqudty. The multpler s also hgher for households wth hgher levels of educaton (1.60 versus 1.25), although the dfference n parameters s not statstcally sgnfcant. The PROCAMPO effect s ndependent of the presence of mgraton assets, ndcatng that households wth remttances are not subject to lqudty constrants. Fnally, the PROCAMPO ncome multpler s lower for ndgenous households (0.19 versus 2.27 for nonndgenous), and for households lvng n the North and North-Pacfc (n both cases not sgnfcantly dfferent from 0). These multplers reveal the shadow ncome value of lqudty for the correspondng category of recpent

21 households, reflectng n each case the partcular margnal cost of uncaptured opportuntes due to constranng lqudty. Results thus reveal that the greatest absolute ncome payoff (pesos of ncome per pesos of transfer) from relaxng lqudty constrants s among medum and large farmers, famles wth a small number of adults, nonndgenous households, and, regonally, the Gulf and the Center. 5. PROCAMPO TRANSFERS AND THE INTENSIFICATION OF AGRICULTURE The above results on PROCAMPO multplers by farm sze show that transfers can be productvely used n agrculture. Ths s clearly confrmed by farmers responses to questons about use of ther PROCAMPO recepts. In the survey, 70 percent of the households responded that they use the PROCAMPO money to purchase nputs. In a larger survey done by the Mnstry of Agrculture (SAGAR 1998), 44 percent of respondents sad that PROCAMPO transfers allowed them to ncrease ther nput purchases and another 17 percent to start usng purchased nputs. Ths happened despte the fact that PROCAMPO transfers have often arrved late n the season (93 percent say they arrve after the promsed date). Whle farmers would clearly beneft more from recevng PROCAMPO at the tme they purchase ther nputs, many of them purchase nputs ether by collateralzng ther PROCAMPO rghts, or drectly by obtanng credt from supplers on the bass of the forthcomng transfers. Respondents say, however, that transfers are not suffcent to nduce changes n croppng patterns or n the extent of areas

22 planted. To confrm ths, we analyze the changes n agrcultural ncome and n nput use nduced by PROCAMPO transfers usng the 1994 and 1997 panel data. Results are presented n Table 6. Although explanng the very volatle agrcultural ncome s dffcult, results ndcate a postve multpler effect n agrculture, wth every peso of PROCAMPO transfer generatng, at mean value, 0.33 pesos n agrcultural ncome. As can be seen from the nteracton terms wth land, ths effect s obtaned through ownershp of rrgated land. It s also largely nfluenced by the avalablty of complementary techncal assstance. Settng techncal assstance to zero would reduce the ncome multpler from 0.33 to 0.23. The nteractve term wth credt confrms the role of PROCAMPO as a substtute for credt. Settng access to credt to zero (rather than at the mean value of 0.18) would ncrease the multpler from 0.33 to 0.51. Hence, t s those households that control more rrgated land, have access to techncal assstance, and no access to credt that are able to take greater advantage of the cash transferred n generatng more ncome. Use of chemcals as estmated by a random effect probt equaton also shows a sgnfcant postve response to PROCAMPO transfers. We do not have nformaton on the amount used n 1994 and hence cannot measure the effect reported n ntervews of ncreased applcatons for those who already used some chemcals n 1994. Lvestock ncome also responds to PROCAMPO transfers, wth every peso of transfer generatng 0.28 pesos of lvestock ncome. Note that ths estmaton does not nclude the use of PROCAMPO transfers to purchase lvestock. It therefore lkely underestmates the long-term effect of PROCAMPO on lvestock ncome.

23 As n all cash transfer programs (Subbarao 1997a), t s mportant to assess the mpact transfers have on labor market partcpaton. The labor market ncome effect due to PROCAMPO can derve both from a change n labor market partcpaton and/or from a drect effect of PROCAMPO once partcpaton has been decded. For farm households that partcpate n the labor market wth the objectve of relaxng lqudty constrants or of achevng portfolo dversfcaton n ther sources of ncome, cash transfers can lead to a reallocaton of labor from the labor market to the farm. The partal results presented n Table 7 show that the PROCAMPO effect s ndeed negatve on labor market partcpaton. On average, PROCAMPO transfers reduce labor market partcpaton by 9 percent (from 45.4 to 41.8 percent). No sgnfcant effect s observed on partcpaton to self-employment actvtes. Hence, ths suggests that some households wthdrew from the labor market n order to spend more tme n agrculture once they ganed access to the necessary lqudty. Ths result supports the noton that there was excess partcpaton n the labor market by households endowed n land and other productve resources n response to credt market falures. PROCAMPO thus served ndrectly as a mechansm to allevate the strngency of ths constrant, allowng households to reallocate part of ther labor tme to more proftable agrcultural actvtes.

24 6. THE TRUE CONTRIBUTION OF PROCAMPO TO INCOME MAINTENANCE As ndcated n Secton 3 of ths paper, the true contrbuton of the PROCAMPO cash transfers to ncome mantenance s the sum of the monetary value of the transfers (drect effects) and the ncome generated through ndrect effects measured by multplers n excess of one. We can use the estmated multplers n Table 5 to predct what would have been the change n ncome durng the 1994 1997 perod had there been no PROCAMPO program. Ths s done n Table 8. For all households, the observed ncome change s 14.2 percent. The drect cash transfer represents an ncrease of 8.7 percent over 1994 ncome. Wth a multpler of 2.06, the ndrect effect s a contrbuton to ncome of 9.3 percent over 1994 ncome. Hence, the total PROCAMPO contrbuton s an ncrease n ncome of 18 percent over the 1994 level. Had there been no PROCAMPO program, household ncome would, on average, have declned by 3.9 percent. Contrastng ths result wth the data n Table 1 shows the mportance of accountng for the ndrect ncome effects of cash transfers n assessng the mpact of such programs. In Table 1, t appeared that PROCAMPO transfers would not have been necessary for ncome mantenance snce other sources of ncome would have sustaned an ncome ncrease of 5.4 percent. In fact, these other ncomes were themselves related to the hdden multpler effects of PROCAMPO transfers. If they had not put the cash transfers to work, many categores of households would have suffered ncome declnes n spte of the appearance of rsng ncome after

25 dscountng the drect transfers. Ths apples to medum farmers, nonndgenous households, and households n the Center regon. There are fve nstances where transfers were not used to generate more ncome: smallholders, households wth a large number of adults, ndgenous households, and households n the North and North-Pacfc, all of whch were found to have ncome multplers of less than one. In spte of ths, however, all households were made better off by the cash transfers compared to the ncome levels they would have acheved wthout the transfers. 4 7. CONCLUSION The ndrect effects of cash transfer programs have receved lttle attenton. Yet, we found that these effects can be hghly sgnfcant and that they deserve full consderaton n the desgn of such programs. We show, n partcular, that PROCAMPO, a cash transfer program to Mexcan farmers ntroduced n compensaton for the antcpated declne n the prce of staple crops as a consequence of trade lberalzaton, created large ndrect effects among ejdataros through multplcaton of the lqudty receved. The multpler for all households s n the range of 1.5 to 2.6. Multplers are hgher for households wth medum and large farms, low numbers of adults n the household, nonndgenous backgrounds, and located n the Center and Gulf regons. Large multplers reflect uncaptured margnal ncome opportuntes due to lqudty constrants that are relaxed by the transfers. Opportuntes come from the asset 4 The only negatve multpler, for households n the North-Pacfc, s not sgnfcantly dfferent from 0.

26 endowments that these households have, partcularly rrgated land, and they are enhanced by access to techncal assstance. Lqudty constrants derve from ncomplete property rghts n the ejdo sector, and from the current dsarray of fnancal nsttutons servcng agrculture followng scalng down of the agrcultural development bank mpled by structural adjustment. Large multplers thus reflect szable gaps between opportuntes and constrants. Households wth mgrants sendng remttances and wth hgher levels of educaton may thus have lower multplers because they were able to work around the lqudty constrants more effectvely than other households. Households wth lttle land and wth ethnc backgrounds may have lower access to lqudty, but also have lower opportuntes to nvest addtonal cash receved, agan resultng n lower multplers. There are two polcy mplcatons that derve from ths analyss. Frst s that f multpler effects are mportant and polcy-responsve (as suggested by heterogenety of multplers across households), then the PROCAMPO program would gan from beng managed as part of a comprehensve effort to maxmze these multplers (snce the ultmate goal s to rase the ncome of targeted households). Ths can be done by ntroducng complementary rural development ntatves that ncrease opportuntes to use the transfers productvely. It s worth emphaszng that the households that beneft from the multpler effect are prmarly from hgher ncome groups. Thus, whle ncreasng ncomes, the ndrect effect does not renforce the mpact of the drect effect on poverty. Thus the cash transfer that maxmzes the multpler effect may not be the most effectve at reducng poverty.

27 We observed that addtonal lqudty serves prncpally to ncrease the use of current nputs. Transfers thus enhance the level of tradtonal actvtes. We found no evdence of technologcal change or of the ntroducton of new actvtes (see SAGAR 1998 for a smlar observaton). In correspondence wth the new set of ncentves ntroduced by NAFTA, transfers should nstead be used for the modernzaton of agrculture and ts dversfcaton toward hgh value actvtes wth comparatve advantage. That ths s not happenng s not surprsng. In the current context of declnng nsttutonal support to agrculture, only 18 percent of the ejdataros have access to formal credt, 13 percent to Alanza para el Campo, the man publc program n support of rural development, and 7 percent to techncal assstance. Multpler effects could thus be sgnfcantly ncreased f the PROCAMPO program were accompaned by a serous effort at nsttutonal reconstructon and technologcal change n support of the modernzaton and dversfcaton of ejdo agrculture. The second polcy mplcaton s that hgh multplers show capacty to borrow even at hgh nterest rates. Wllngness to pay for lqudty s somewhat overrepresented by the magntude of the multplers, snce they need to be dscounted for the rsk of borrowng, whch s not present wth gfted money. However, the results show that there s a clear unmet need for lqudty that can be productvely nvested, and that ejdataros can pay for ths servce at nterest rates that are qute compatble wth current commercal rates. Ths shows the hgh payoff that exsts from constructng an alternatve set of fnancal nsttutons able to replace the parastatals that prevously served the sector. If land s to serve as collateral n accessng loans n these fnancal nsttutons, then the

28 current ttlng program should have a hgh payoff. Ttlng wthout access to credt wll, however, not change the current stuaton. PROCAMPO multplers help reveal the shadow value of lqudty n every category of potental borrowers. They consequently provde a metrc to dentfy where the effectve demand for fnancal servces s the greatest. The magntude of the PROCAMPO multplers should be taken as proof that the ejdo sector s not to be dscounted as a lvely sector for nvestment and growth. To avod ths potental beng wasted, transfers should be complemented wth nvestment opportuntes n new commodtes and new technologes, and the nsttutons that servce the sector, most partcularly for accessng lqudty, should be reconstructed.

29 TABLES

30 Table 1 PROCAMPO n household ncome, ejdo households, 1997 Categores of households PROCAMPO Number of observatons Partcpaton* as share of ncome n 1997 Observed total ncome change 1994-97 PROCAMPO Sources of ncome change Other ncomes (%) (%) (%) (% of 1994 ncome) All households 956 86.4 7.7 14.2 8.7 5.4 Farm sze: land used n 1994 Small (< 3 hectares) 322 84.8 6.3 20.9 7.6 13.3 Medum (3 to 7 hectares) 282 90.1* 8.3 29.7 10.7 19.0 Large (> 7 hectares) 352 84.9 7.9 4.8 8.3-3.5 Labor asset: number of adults n the household Low (< 4 adults) 514 85.4 8.4 2.2 8.6-6.4 Hgh ( 4 adults) 442 87.6 7.0 27.3 8.9 18.4 Educaton asset: average adult educaton Low (< 4.5 years) 480 87.3 9.7 9.8 10.7-0.9 Hgh ( 4.5 years) 476 85.5 6.5 16.8 7.6 9.1 Mgraton assets for the Unted States No mgraton asset 526 84.8 8.0 6.2 8.5-2.3 Postve mgraton asset 430 88.4* 7.4 20.8 9.0 11.8 Socal assets: ethncty Indgenous 209 84.7 8.9 30.4 11.6 18.8 Nonndgenous 747 86.9 7.5 12.1 8.4 3.7 Regons: North 207 90.3 9.0 42.6 12.8 29.9 North Pacfc 103 74.8* 5.3-40.2 3.2-43.4 Center 262 89.7 8.7 19.7 10.4 9.2 Gulf 166 90.4 9.5 94.4 18.4 76.0 South 218 81.2* 4.5 15.0 5.2 9.8 Notes: * Indcates sgnfcantly dfferent at 95 percent. For farm sze, comparson s made wth the small farms. For the regons, comparson s made wth the North.

31 Table 2 Sources of ncome, ejdo households, 1994 and 1997 All households 1994 1997 Percent change n ncome Test of dfference n ncome Total household ncome (1994 pesos) 10,828 12,361 14.2 * (shares, n percentage) Farm ncome 53.1 45.1-3.0 Agrculture 38.3 27.7-17.5 Lvestock 14.8 17.4 34.5 ** Farm ncome 46.9 54.9 33.6 ** Off-farm actvtes 36.4 40.2 26.3 ** Wage ncome 27.6 24.2 0.0 Self-employment 6.7 9.8 67.2 ** Remttances 2.1 6.3 242.8 ** Other off-farm ncome 10.6 7.0-23.9 ** PROCAMPO 0.0 7.7 Number of observatons 956 956 Note: * (**) means sgnfcantly dfferent at 95 percent (99 percent).

32 Table 3 Change n total household ncome between 1994 and 1997 Robust regresson Mean value Coeffcent Standard devaton P-value Asset or characterstc n 1994: parameter s β 94 Irrgated area owned (hectares) 0.9 609 175 0.00 Ranfed area owned (hectares) 6.5 140 43 0.00 Pasture area owned (hectares) 3.5 5 31 0.88 Common property land per ejdataro (hectares 24.8 2 10 0.83 Cattle (number of heads) 6.4 265 32 0.00 Number of adults 3.4 500 202 0.01 Average years of educaton among adults 4.5 254 176 0.15 Mexco mgraton assets 0.14-1,011 835 0.23 U.S. mgraton assets 0.42 451 378 0.23 Access to techncal assstance (dummy) 0.10 1,822 1,046 0.08 Access to formal credt (dummy) 0.31 938 662 0.16 Asset or characterstc n 1997: parameter s β 97 Irrgated area owned (hectares) 1.24 230* 111 0.04 Ranfed area owned (hectares) 7.8 30* 34 0.39 Pasture area owned (hectares) 4.2-14 23 0.55 Common property land per ejdataro (hectare 25.3 8 9 0.38 Cattle (number of heads) 7.4 236 28 0.00 Number of adults 3.64 639 196 0.00 Average years of educaton among adults 4.55 547* 184 0.00 Mexco mgraton assets 0.21-950 695 0.17 U.S. mgraton assets 0.74 791 271 0.00 Access to techncal assstance (dummy) 0.07 2,918 1,199 0.02 Access to formal credt (dummy) 0.18-140 849 0.87 PROCAMPO transfer PROCAMPO transfer (pesos) 947 2.1 0.3 0.00 Constant asset or characterstc: parameter s β 97 - β 94 Gender of household head (man = 1) 0.97 474 1,764 0.79 Age of household head 51.8 30 24 0.22 Indgenous (dummy) 0.22-807 805 0.32 Regonal effects (base = North) North Pacfc 0.11-4,555 1,236 0.00 Center 0.27-1,450 879 0.10 Gulf 0.17 1,235 1,124 0.27 South 0.23-414 974 0.67 Intercept -2,539 2,512 0.31 Goodness-of-ft Number of observatons 956 F(30, 925) 10.10 1 Standard errors estmated wth bootstrappng. * Sgnfcantly dfferent from the 1994 parameter at 95 percent.

33 Table 4 PROCAMPO multplers Number of observatons: 956 Wth cattle stock n 1997 Wth cattle stock n 1996 95% conf. 95% conf. Coeffcent nterval Coeffcent nterval Robust regresson 2.06** 1.5-2.6 2.09** 1.5-2.7 Quntle regresson: medan (LAD) a 2.20** 1.3-3.1 2.17 1.1-3.2 OLS 2.24 0-4.5 2.30 0-4.6 Note: ** Sgnfcantly dfferent from 1 at 99 percent. a Standard errors estmated wth bootstrappng.

34 Table 5 PROCAMPO multplers for selected groups of recpents Robust regresson Coeffcent t-statstc All households 2.06 7.1 Farm sze: land used n 1994 Small (< 3 hectares) 0.24 0.4 Medum (3 to 7 hectares) 2.77 3.6 (P-value for test of dfference) (0.01) Large (> 7 hectares) 2.04 4.3 (P-value for test of dfference) (0.42) Number of adults n the household Low (<4 adults) 2.75 9.2 Hgh ( 4 adults) 0.93 2.0 (P-value for test of dfference) (0.02) Average adult educaton Low (<4.5 years) 1.25 3.6 Hgh ( 4.5 years) 1.6 3.3 (P-value for test of dfference) (0.56) U.S. mgraton assets Zero mgraton assets 2.05 6.7 Postve mgraton assets 1.9 3.6 (P-value for test of dfference) (0.79) Socal assets: ethncty Indgenous household 0.19 0.5 Nonndgenous household 2.27 6.6 (P-value for test of dfference) (0.00) Regonal effects North-Pacfc -2.1 0.3 North 0.15 0.8 (P-value for test of dfference) (0.28) Gulf 2.21 5.5 (P-value for test of dfference) (0.01) Center 2.81 4.7 (P-value for test of dfference) (0.41) South 1.09 0.8 (P-value for test of dfference) (0.25)