WPS3994-IE IMPACT EVALUATION SERIES NO. 6 Investing Cash Transfers to Raise Long-Term Living Standards Paul Gertler Sebastian Martinez Marta Rubio-Codina In this paper we test whether poor households use cash transfers to invest in income generating activities that they otherwise would not have been able to do. Using data from a controlled randomized experiment, we find that transfers from the OPORTUNIDADES program to households in rural Mexico resulted in increased investment in micro-enterprise and agricultural activities. For each peso transferred, beneficiary households used 88 cents to purchase consumption goods and services, and invested the rest. The investments improved the household's ability to generate income with an estimated rate of return of 17.55%, suggesting that these households were both liquidity and credit constrained. By investing transfers to raise income, beneficiary households were able to increase their consumption by 34% after five and a half years in the program. These results suggest that cash transfers to the poor may raise long-term living standards, which are maintained after program benefits end. JEL Codes: I38, D12, D13, O12 Keywords: conditional cash transfer programs, randomized experiment, long run living standards, agricultural investments, micro-enterprise activity, liquidity and credit constraints. World Bank Policy Research Working Paper 3994, August 2006 The Impact Evaluation Series has been established in recognition of the importance of impact evaluation studies for World Bank operations and for development in general. The series serves as a vehicle for the dissemination of findings of those studies. Papers in this series are part of the Bank's Policy Research Working Paper Series. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Acknowledgements: We have benefited from comments by Alain de Janvry, Pierre Dubois, Edward Miguel, Elizabeth Sadoulet, John Skinner, Doug Staiger, Chris Udry and seminar participants at UC Berkeley, The World Bank, University of Toulouse, Stanford University, Columbia University, Yale University, the Mexican National Institute of Public Health and the NEUDC meetings at Brown University. We thank the OPORTUNIDADES program for its support, and in particular Rogelio Gomez-Hermosillo, Concepcion Steta and Iliana Yaschine . For the qualitative field work we gratefully acknowledge support from UCMEXUS and assistance from Sarit Beith-Halachmi, Giovanna Limongi, Ryo Shiba, and Jennifer Sturdy. Affiliation and Contact Information: Paul Gertler, Haas School of Business, University of California, Berkeley and World Bank; gertler@haas.berkeley.edu. Sebastian Martinez, World Bank, smartinez@worldbank.org, Marta Rubio-Codina, University of Toulouse (GREMAQ, INRA), marta.rubio@univ-tlse1.fr. 0. Introduction Five years ago when my oldest daughter was in school and we received money from PROGRESA, we saved 600 pesos to buy wood and the other materials for building a chicken coup, and with what was left we bought a few chickens. Since then, we have raised many chickens which we sometimes sell, and we collect 10 to 15 eggs per week that we eat ourselves. OPORTUNIDADES (PROGRESA) beneficiary in rural Mexico (August 2004)1 Cash transfer programs are important policy tools for fighting poverty in both developed and developing countries. A major concern in developed countries has been the extent to which beneficiary populations become dependent on government assistance for maintaining current consumption levels2. In contrast, we argue that beneficiary families in developing countries may invest part of their cash transfer in productive enterprises that boost their income generating ability and consequently raise living standards permanently. We test this hypothesis using a controlled randomized experiment to identify the extent to which beneficiary households from Mexico's OPORTUNIDADES3 program invested cash transfers in income generating activities, and the extent to which those investments increased long-term household consumption. In developing countries there are two primary pathways by which transfers correct market failures that limit investment in productive activities. First, transfers alleviate liquidity and credit constraints that contribute to poverty traps, whereby poor households are able to afford the startup costs associated with entrepreneurial activities4. The households studied here, with under US $1 per day in per capita consumption, are likely to be liquidity constrained, and therefore unlikely to be able to afford the "low" levels of capital investment needed to start micro-entrepreneurial activities5. Second, if transfers are perceived as a secure source of income, risk adverse households will be more willing to increase ownership of risky assets, even in the presence of risk. In these 1 From a qualitative survey of agricultural production and micro-enterprise activity in rural Mexico, conducted by the authors. Interview from August 3, 2004. 2See Moffit (1992) for a review of the US case. 3Skoufias (2005) provides a description of the program. 4Models along these lines are developed by authors including Banerjee and Newman (1993), Aghion and Bolton (1997), Lindh and Ohlsson (1998), Lloyd-Ellis and Bernhardt (2000), and Banerjee (2004). 5McKenzie and Woodruff (2006) report that it takes $100 USD to start a micro-enterprise in Mexico. 1 ways, monetary assistance from OPORTUNIDADES provides a secure income stream that may help alleviate liquidity constraints and grant access to capital, leading to increased investment6. This study uses the randomized experiment from the rural evaluation of the OPORTUNIDADES conditional cash transfer program to estimate the impact of an exogenous increase in unearned income on investment and long-term living standards. We find that beneficiaries invest in production and draft animals, and that previously landless beneficiary households obtain land for agricultural production. Furthermore, there are significant increases in the number of households that operate micro-enterprises. Our estimates indicate that for each peso transferred, beneficiary households consume 88 cents directly, and invest the rest. The aggregate effect of the investments yields a 1.8 cent increase in consumption for each peso of transfers received. These estimates indicate an estimated rate of return on investment of between 15.52% and 17.55%. Through investments in productive activities, beneficiary households increased their consumption by 34% after five and a half years in the program. Because investments are made in productive assets and income-generating activities, the gains in living standards will likely be maintained even after program benefits end. Our study contributes to the relatively small literature that examines the potential impact of cash transfer programs on long-term living standards. Ravallion and Chen (2005) show that the lion's share of temporary cash transfers given by an anti-poverty program in China were saved. Sadoulet, de Janvry and Davis (2001) argue that payments from the Mexican agricultural support program PROCAMPO are invested in farm production, and estimate an income multiplier in the range of 1.5 to 2.6. While not related directly to cash transfer programs, studies by Yang (2005) and Woodruff and Zenteno (2001), show that remittances are used as capital to invest in micro- enterprises in the Philippines and Mexico, respectively. The paper is organized as follows. Sections 1 and 2 describe the OPORTUNIDADES human development program and review the program's experimental design and data. In section 3, 6 Investment will also depend on expectations about the transfer's temporal horizon and life-cycle considerations (Modigliani and Brumberg 1954; Friedman 1957 and Hall 1978). 2 we present evidence of the causal effect of OPORTUNIDADES on increased investments in agricultural and micro-enterprise activities. Having established that cash transfers lead to more investment, section 4 develops a basic model of investment to illustrate the relationship between consumption, transfers and investment that we then estimate empirically. Section 5 provides a series of robustness tests to support our main findings and section 6 concludes. 1. The Rural OPORTUNIDADES Program The Mexican Government established OPORTUNIDADES (originally called PROGRESA) in 1997. The program was designed to alleviate current poverty and break its intergenerational transmission by inducing parents to invest in the human capital of children. In this sense, OPORTUNIDADES was conceived as a temporary program (i.e. over the course of three to four decades), which would become obsolete once the current generation of beneficiary children reaches adulthood. Cash transfers from OPORTUNIDADES are given to the female head of household7, and are conditional on children attending school, family members obtaining preventive medical care through clinic visits and attending "pláticas" or education talks on health related topics. OPORTUNIDADES is the largest conditional cash transfer program of its kind. It distributed approximately 3 billion US dollars to some 5 million beneficiary households in 20048. The cash transfers come in two forms. The first is a bimonthly fixed food stipend conditional on family members obtaining preventive medical care and is intended for families to spend on more and better nutrition. The second type of transfer comes in the form of educational scholarships and is given conditional on children attending school a minimum of 85% of the time and on not repeating a grade more than twice. Specifically, OPORTUNIDADES provides bimonthly cash scholarships for each child less than 18 years old enrolled in school between the third grade of primary school and the third grade (last) of junior high. High school scholarships are 7Scholarships for beneficiaries in upper-secondary school can be received by the youth themselves. 8www.oportunidades.gob.mx 3 granted to all beneficiaries younger than 21 years old who are enrolled in school. The educational stipend varies by grade and gender. It rises substantially after graduation from primary school and is higher for girls than boys during junior high and high school. Beneficiary children also receive money for school supplies once or twice a year. There is an upper limit in the total transfer received per household. Table 1 details transfer amounts in October 1997 prices. When OPORTUNIDADES was first rolled out in rural areas starting in 1997, program eligibility was determined in two stages (Skoufias et al. 2001). First, the program identified underserved or marginalized communities and then identified low-income households within those communities. Selection criteria for marginalized communities were based on the proportion of households living in very poor conditions, identified by using data from the 1995 census (Conteo de Población y Vivienda). For the selection of eligible households within marginalized communities, OPORTUNIDADES conducted a socio-economic survey, the Encuesta de Características Socioeconómicas de los Hogares (ENCASEH). This census of households was used to classify households as eligible for treatment ("poor") or ineligible ("non-poor") using a proxy means test (PMT). Using the 1997 ENCASEH, the original classification scheme designated approximately 52% of households as eligible ("poor")9. All eligible households living in treatment localities were offered OPORTUNIDADES and a majority (90%) enrolled in the program. Once enrolled, households received benefits for a three- year period conditional on meeting the program requirements. New households were not able to enroll until the next certification period which prevented migration into treatment communities for OPORTUNIDADES benefits. Households in rural areas were "recertified" (re-assessed with a proxy means test) after three years on the program to determine future eligibility. If a household was recertified as eligible, it would continue receiving benefits. If not recertified, the household was guaranteed three more years of support followed by three years of transitional support. Thus, 9Later the Government decided that a subset of the "non-poor" households had been unduly excluded. They expanded the eligibility criteria to include a set of slightly wealthier households in a process called "densification" (Hoddinott and Skoufias 2004). 4 households could expect a minimum of nine years of benefits upon enrolling in the program (OPORTUNIDADES 2003). In order to transfer the cash, OPORTUNIDADES had to verify that households actually completed the required health care visits by having medical providers from the public clinics who administered the services certify that the household had completed the requirements. A similar procedure was followed for the cash transfer associated with school attendance. About 1 percent of households were denied the cash transfer for non-compliance. 2. Experimental Design and Data Our analysis takes advantage of the controlled-randomized evaluation design implemented by the Mexican Government to conduct a rigorous impact evaluation of OPORTUNIDADES. Due to budgetary and logistical constraints, the Government was unable to enroll all eligible families simultaneously. Rather, it needed to phase in enrollment over a period of time. For ease of implementation, the Government decided that it would enroll whole communities at a time and that it would enroll them as fast as possible so that no eligible household would be kept out of the program. As a result of this process, the government randomly chose 320 treatment and 186 control communities in seven states for a total of 506 experimental communities. Eligible households in treatment communities began receiving benefits in April of 1998, and eligible households in control communities were not incorporated until November of 1999 nor were they informed that OPORTUNIDADES would provide benefits to them until two months before incorporation. The data used in this paper comes from household evaluation surveys and administrative records of the amount of money transferred to households. Detailed information on a host of topics was collected in a series of rural evaluation surveys, the Encuesta de Evaluación de los Hogares Rurales (ENCEL). The ENCELs survey all eligible and ineligible households in treatment and control communities. The sub-sample used for the main analysis in this paper is restricted to eligible households classified as poor, although we also use ineligible (non-poor) households for 5 robustness checks. There are three rounds of data during which eligible households in control communities did not receive transfers (October 1998, May 1999 and November 1999), and a straightforward comparison of treatments and controls is possible. Once control communities were incorporated into the program at the end of 1999, an additional three rounds of data were collected during which eligible households in both types of communities were receiving benefits (May 2000, November 2000 and November 2003). In addition, baseline data from the 1997 pre-intervention ENCASEH census is used for a total of seven rounds of data between 1997 and 2003. Our sample consists of 7,658 poor households that were eligible for the program in the 320 treatment communities and 4,644 poor households that were eligible for the program in the 186 control communities (Table 2). We refer to this sample as the "intent to treat" or ITT sample. Baseline data for this sample are described in detail in Table 3A. The sample is well balanced across control and treatment groups. We report the test of difference in means across control and treatment groups, finding no statistically significant differences for 33 out of 35 characteristics measured at baseline. This suggests that the randomization was effective in generating truly exogenous variation in the treatment. We also conduct the analysis with a second sample designed to estimate the effect of treatment on the treated (TOT). In these analyses, we use the approximately 90 percent of eligible households that actually took up the program, 6,819 treatment households and 4,159 control households that took up the program when offered in late 1999. A potential source of bias arises in the TOT sample if observed and unobserved time varying characteristics are driving the household's decision to participate in the program. We assume the group of controls that took up the program when offered in late 1999 would have taken up the program if offered earlier in 1998. The fact that take up rates amongst treatment and controls are almost identical (90%) supports this hypothesis (Table 2 and Graph 1). Furthermore, the TOT sample baseline observables are well balanced, with only 1 statistically significant difference out of 35 characteristics tested (Table 3B). 6 Given this evidence, we conclude that working with the actual beneficiaries does not introduce selection bias coming from heterogeneous take up responses. 3. Investment In this section, we estimate the effect of OPORTUNIDADES on investments in agriculture and micro-enterprise activities. We show that beneficiary households increase agricultural assets and participation in micro-enterprise activities, suggesting that some of the cash transfer was used for investment in productive assets. Section 3.1 presents the empirical specification based on the randomized design of the OPORTUNIDADES evaluation sample, section 3.2 presents comparisons of means, and sections 3.3 and 3.4 discuss the main investment results. 3.1. Specification and Identification Our analysis exploits the exogenous variation introduced by random assignment of treatment to estimate the effect of OPORTUNIDADES on investment in agricultural production and micro-enterprise activity, the two most common domestic income generating activities in this population. We estimate two types of models: one exploiting the community randomization to obtain average treatment effects, and a second which exploits the variation in transfer amounts to estimate the effect of the size of the transfer on investments. To estimate the first specification we use the exogenous variation generated by the randomized phasing in of the program. Beneficiary households in treatment communities received transfers for just over a year and a half longer than beneficiary households in control areas. Given the timing of the ENCEL surveys, we first observe treatment households in October 1998, approximately 6 months following the first transfer payment. Treatment households would have received three bimonthly transfer payments by the time of the first follow up survey. Since control communities did not begin receiving payments until November-December of 1999, there are three rounds of data (October 1998, May 1999 and November 1999) where a simple comparison of eligible households in treatment and control communities is possible. 7 For these three rounds, we estimate the following reduced form: Aijt = o +1Tij + 2tWAVEt + k Xij, + u j + ijt , 97 (1) t k where Aijt denotes productive assets of household i in community j in period t; Tij is a binary indicator equal to 1 if eligible household i lives in a treatment community in the ITT specification, or if the household has received benefits in the TOT specification; WAVEt takes on the value 1 in period t, and the Xij,97 are household and community characteristics measured at baseline. The error term has two components: an idiosyncratic disturbance, ijt, and a community random effect, uj, that accounts for the correlation within communities over time. Note that the dependent variable in the analysis is the level of assets and not investment during the period. If the randomization successfully balanced the treatment and control groups, the level of pre-treatment assets would have been the same in treatment and control households (we explicitly test this assumption in section 3.2). Under this assumption, the coefficient 1is the average treatment effect on productive assets. The second specification takes advantage of the variation in total cumulative transfers introduced both by the randomized phasing in of the program and the variation in the transfer amount based on the demographic structure of the household. The randomization meant that treatment households had been accumulating transfers for about 18 months longer than control households (Graph 1). In addition, households with more children in school and enrolled in higher grades (more female children in higher grades in particular) had higher transfer amounts, and therefore accumulated transfers faster than similar households with fewer children in school or with more male children. Graph 2 illustrates the variation in household demographic structure, plotting shares of households by number of children enrolled in any grade between the third year of primary and the third year of junior high10. 10Households with teenagers enrolled in high school are excluded because high school stipends were granted beginning in the 2001-02 school year (following the experimental period), and because of the low high school enrollment rates (only 33% of 15 to 21 year olds report being enrolled in high school). 8 The second specification uses all six rounds of data from 1998 through 2003 to estimate the differential effects of cumulative transfers on productive assets. Cumulative transfer quintiles are constructed for each household-wave observation based on the total transfer amount a household has accumulated at that point in time. These new variables are constructed as indicator variables for the quintile a household belongs to in each round. It is important to note that they are constructed on the distribution of transfers across all waves of data and not within each round. We then use the cumulative transfer quintile dummy variables to estimate the following specification: 5 Aijt = o + 1pQijt,p + 2tWAVEt + k Xij, + uj + ijt (2) 97 p=1 t k where Qijt,p equals 1 if household i in community j in period t falls in the p-th quintile of the total cumulative transfer distribution. We specify the effect of cumulative transfers non-linearly so as to allow for jumps in investment due to the potential lumpiness of certain types of investment. One potential concern with the specification in equation (2) is that actual total cumulative transfers are determined in part by household decisions to send their children to school. If households decide their children should work instead of going to school, then part of the income generated from work may be invested. In this case, we would obtain a biased estimate of the impact of the transfer amount on investment. In order to correct for this potential endogeneity we use the potential transfer the household would have received if all eligible children were enrolled in school. To compute potential transfers, we take household composition and children's enrollment status at baseline and apply the program rules, assuming the child progresses one grade per year with no school drop outs and no repetition. Our IV is therefore constructed from pre-intervention household demographic structure interacted with timing of incorporation into the program (which was determined randomly). Since the potential transfer amounts include no household behavioral response but are highly correlated with actual transfers, it is a valid instrument of the actual transfer amounts. We also control for the 9 number of household members and share of children directly, so that the IV is not confounded with family composition effects. Graphs 3a and 3b plot the distributions of the actual and potential current and accumulated transfers, respectively. As shown, the distributions of the potential transfers follow those of the actual transfers very closely. As expected, the potential transfers are an over-estimate of the actual transfers (given non-compliance, administrative delays in payments, etc). The simple correlation amongst them is 0.89. If we control for time effects and baseline covariates, the potential transfer explains 55.7% of the total transfer and 65.9% of the variation of the cumulative transfer. Graph 4 plots mean cumulative transfer by quintile for the distributions of actual and potential cumulative transfers. 3.2. Comparison of Means The dependent variables for agricultural assets come from the six rounds of the ENCEL that contain information on animal ownership and amount of land in use over the 12 months preceding the interview (hectares of land for agriculture, grazing and/or forestry purposes). We define production animals to be those whose meat and/or by-products (milk, cheese, eggs, etc.) are sold and consumed. These include goats and sheep, cattle (cows), poultry (chickens, hens and turkeys), pigs and rabbits. We define draft animals to be those traditionally used for farming (plowing) and/or for transportation purposes. These include donkeys, mules, horses and oxen. We transform all of these animals into "cow equivalents" using the ratio of the price of the animal divided by the price of a cow. Therefore, we create cow-equivalent indices of production and draft animals. Land is measured in hectares and includes all plots used by the household11. The dependent variables for (non-agricultural) micro-enterprises come from five rounds of the ENCEL surveys (October 1998 to December 2000). A set of questions ask the household head whether somebody in that household had engaged in a "self-motivated" non-agricultural activity which generates income during the month before the interview. The list of activities includes 11A detailed description of the sample and variables is available upon request. 10 sewing clothes, making food for sale, carpentry and construction, sale of non-food items such as handcrafts, transportation of people or goods in own vehicle, repair of artifacts or machinery, domestic service (wash, iron or cook for a fee), or other activities done on your own. We define micro-enterprise as participation in any of these activities. No micro-enterprise activities were collected in the 1997 baseline survey, and because of lack of comparability (changes in the categories and time frame considered), we do not include the 2003 round. For the purpose of analysis, we classify households according to the amount of land used and agricultural asset ownership at baseline. We distinguish between: (i) households with no agricultural assets, (ii) landless farms (households reporting no land use but animal ownership), (iii) smaller landed farms (households using at least 3 ha of land for agricultural, grazing or forestry purposes, regardless of animal ownership), and (iv) bigger farms (households using more than 3 ha, regardless of animal ownership). Around one tenth of the sample has no agricultural assets, 31% of households are landless but have some animals, 45% have smaller farms and 13% have bigger farms. We check for balance between treatment and control households within farm size categories in Table 3A in the Data Appendix. With random assignment of communities to treatment and control groups, we would expect that ownership of farm assets at baseline would be equal between the two groups. Then, any subsequent changes could be attributed to the program. In effect, when we compare agricultural assets of treatment and control households in the TOT sample prior to intervention, we find no statistically significant differences in mean values (Table 4 Panel A). During the experimental period (October 1998 through November 1999), a simple comparison of the mean values for each dependent variable between households in treatment and control communities gives us an estimate of the program impact (see Table 4 Panel B). We report the results both for whether the household had any of the asset type and for the amount of the asset. We observe significant increases in the likelihood of draft and production animal ownership for all households and in particular for households with no agricultural assets (increases in production 11 animal ownership only), landless households (increases in draft animal ownership only) and bigger farms (increases in both types of animal ownership). There are significant (at the 10% level) increases in land use for landless households (including households with no agricultural assets) and significant increases in the number of production animals for landless and smaller farms. Production animals generally require lower capital investments than draft animals. Thus, it is plausible that households with no farm assets initiate animal production by buying small animals that yield relatively quick returns (for example eggs and meat from chickens or turkeys). For micro-enterprises, the comparison of means between treatment and control households during the experimental period suggests important program effects on the likelihood of engaging in micro-entrepreneurial activity. Specifically, when activities traditionally done by men (construction/carpentry and machinery repair, which are 92% and 77% male respectively) are excluded from the dependent variable, households in the treatment group have 58 percent more micro-enterprises than those in the control group. Results hold for landless and smaller farms. 3.3. Estimated Treatment Effects We estimate the effect of OPORTUNIDADES on both the probability of owning an asset and the amount of the asset owned for draft animals, production animals and land. For the probability of having an asset type, we first report the results for the whole sample and then for a sample that is restricted to households that had none of the asset at baseline. Analogously, for the amount of the asset type, we first report the estimate on the whole sample and then restrict the sample to those households with positive levels of the asset at baseline. The probability of owning an asset conditional on having none at baseline provides insight on whether the program induced households to start a farm business (variation at the extensive margin), whereas the estimated impact of the amount of an asset conditional on having some of that asset at baseline tells us about how the program influenced households to expand their already existing farm assets (variation at the intensive margin). 12 The results of these analyses are presented in Table 5. In Panel I we report the estimated impact of the program on asset ownership (and land use). Each cell reports the estimated impact of the program on an asset type (columns) using a particular model specification (rows). For each of the asset types listed in the columns, we first report the results using the whole sample and then for the sample restricted to those that did not own any of the assets at baseline. The rows report the empirical specification used. Models A and B in the first two rows report the estimated ITT and TOT impacts, respectively, without controls. Models C and D add baseline household and community controls. The list of controls includes household head and spouse's ages and education, head's ethnicity, household demographic composition and size, baseline assets (home ownership, dirt floor and electricity), and community characteristics (male agricultural wage in the community, distance to large urban center and presence of community work associations). The descriptive statistics for the controls are presented in Tables 3A and 3B. In Panel I of Table 5 (Model D - TOT with controls), we observe that the probability of draft animal ownership is 2.2 percentage points higher in treatment households than in control households, conditional on no draft animal ownership in 1997. Similarly, treatment households are 6.2 percentage points more likely to own production animals, conditional on no production animal ownership at baseline. Given the average levels of asset ownership for control households in the estimation sample, these coefficients imply that treatment households are 20.2% more likely to own draft animals and 13.4% more likely to own production animals than control households. Results are robust to the exclusion of covariates (Model B) and/or inclusion of all intent to treat households (Models A and C). Finally, the 5.2 percentage point (15.7%) increase in the conditional probability of land use suggests that landless households are acquiring land. In Model E, we interact the treatment indicator with the level of agricultural assets prior to the intervention. We hypothesize that households with existing assets are more likely to invest more as they have already paid the fixed costs of startup. The estimates from Model E are in line with this hypothesis. Only households owning animals at baseline significantly increase draft animal 13 ownership under treatment. Landless households and small farms increase draft animal ownership by 3.5 and 2.9 percentage points respectively (conditional model). On the other hand, treatment households with no agricultural assets at baseline are 9.4 percentage points more likely to increase production animal ownership than control households. Households with no baseline agricultural assets and landless households increase land use by 4.9 to 6.5 percentage points respectively (10% significance). Panel II in Table 5 presents the estimated impacts of the program on the amount owned of each asset. Again, there is no difference between the ITT and TOT models with or without controls. Model D shows that conditional on having draft animals at baseline, there is an increase in the number of cow equivalent draft animals of 0.062 cows (or equivalently, 0.15 horses or 0.41 mules) for all households (10% significance). Households with small farms increase the number of cow equivalent draft animals by 0.078 cows (or equivalently, 0.19 horses or 0.51 mules), conditional on having draft animals at baseline. Production animals increase by 0.125 cow equivalents (significant at the 10% level), equal to approximately 0.70 goats, 0.79 pigs or 6.76 chickens and/or turkeys. Effects are larger in magnitude for households with big farms at baseline. For big farms in the conditional sample, OPORTUNIDADES increases the number of draft animals by 0.112 cow equivalents (0.27 horses or 0.73 mules), and increases the number of production animals by 0.291 cow equivalents (1.63 goats, 1.84 pigs or 15.75 chickens and/or turkeys). Finally, big farms increase land usage by 0.162 hectares on the unconditional sub-sample12. The effect of OPORTUNIDADES on micro-enterprise participation is estimated with a set of probit regressions. Results are presented in Table 6. Models A and B present the average treatment effect when no controls are included, for the ITT and TOT sub-samples, respectively. Models C and D incorporate household and community controls. For this period, treated households 12For the October 1998, May 1999 and November 2003 rounds of the ENCEL, we have information on land ownership which allows us to classify land into "owned land", if any households member is reported to be the owner; and "non-owned land", if the plot is reported to be rented, borrowed or in tenancy. Results (available upon request) show an increase of approximately 0.034 ha (340m2) in the use of non-owned land for beneficiary households. 14 have a 2.5 to 2.8 percentage point higher participation in micro-enterprise activities than control households. Given a mean participation rate in micro-enterprise of 5.9% by control households over the sample period, treatment households are approximately 42.3% to 47.7% more likely to engage in micro-enterprise activity. The last column in Table 6 shows the same set of regressions for predominantly female micro-enterprise activity participation. We find a positive treatment effect of 2.8 percentage points without controls, and 3.0 percentage points with controls, or a 58.9% increase in the likelihood of having a micro-enterprise13. 3.4. Estimated Transfer Effects In Table 7, we present the estimates of the intensity of treatment as measured by the quintile of the accumulated transfer distribution (equation (2)). The results are reported in six panels. The first three panels report the estimated effects of the amount of the transfer on the probability of owning the asset type and the second group of three panels reports the results for the impact on the amount of the asset owned. In all of the panels, Model A shows estimates using the quintiles of the cumulative potential transfer, which can be compared to estimates using the cumulative actual transfer in Model B. Note the similarity in magnitude and significance between estimates in models A and B. Model C repeats Model B using the conditional sample; either not owning the asset at baseline for the probability of ownership models or owning some of the asset at baseline for the amount models. Model D further interacts treatment with baseline farm size. A noteworthy pattern emerging from Table 7 is that households receiving higher accumulated transfers per wave have a higher likelihood of investing in agricultural assets. Point estimates on each quintile, while positive and increasing with the quintile, are generally significant only at the top quintiles, suggesting a possible threshold in the amount of accumulated transfers needed to start investing. For production animals, significant increases are achieved at lower 13We have further analyzed female micro-enterprises by activity type (results available upon request). Interestingly, we find large and significant differences in handcraft enterprises amongst treatment and control households, but no differences in domestic services. Handcraft manufacture requires a larger initial capital expenditure than domestic services (which might only imply transportation costs if the jobs are in distant locations) and thus are more likely to be restricted by liquidity constraints. 15 quintiles. This is not surprising since smaller amounts of money are required to purchase a production animal (a turkey, a goat, etc) than a draft animal (a horse or an ox). The effect per quintile on the probability of land use follows the same general pattern. Effects are larger on the conditional estimates, again suggesting that initially landless households are acquiring more land (Model C). The number of hectares used is positive and increasing with the quintile of cumulative transfers except for the top quintile. One possible explanation for this result is that households might be afraid to lose their eligibility status if they accumulate too many easily verifiable assets. Results are similar when conditioning on baseline farm size (Models D in Table 7). 4. Consumption Section 3 presented evidence that OPORTUNIDADES beneficiaries increased farm production and micro-enterprise activity. In this section, we turn our attention to the effect of these investments on living standards. Specifically, we are interested in the effect of the program on income and consumption. One of the limiting factors in this analysis is that we do not have income data, but rather only consumption. Therefore we need to infer the key structural parameters for a consumption model. The parameters of interest are the marginal propensity to consume and to invest (or save), and the return on these investments. We begin with a discussion of the specification and identification in the context of a simple inter-temporal consumption and investment model14. We then describe the estimation and results. We test a number of important identifying assumptions in the next section. 4.1 Model and Empirical Specification In this section we derive the empirical specification for the consumption equation from the set of budget and productive constraints the household faces in the context of an inter-temporal utility maximization model. Consider a setting where household income is derived from a family business and government transfers, and the household is credit constrained. In each period, the household 14Note that it is beyond the scope of this paper to estimate a full household structural savings and investment behavioral model. Instead, we sketch a simple statistical model which allows us to interpret the reduced form estimates obtained in the empirical analysis. 16 decides how much of its income is consumed and how much is invested in the family farm or micro-enterprise so that its budget constraint is: Ct = Yt + PTt - It (3) where Ct is consumption in period t, Yt is income from the household's productive activities, PTt are public transfers, and It denotes investment. For simplicity, we will assume that the family business production is a function of business assets, At, and family labor, L: Yt = (L )At + t (4) where t is a zero mean random productivity shock such as weather or illness. We assume that only family labor is used for production, which is consistent with the data. Moreover, we are not allowing family labor time allocations to be a choice in this model. Let be the depreciation rate. Then, the household can increase its stock of productive assets through investment, and accumulates assets as follows, At = At + It -1 -1 (5) We assume that households are credit and liquidity constrained. This implies that the maximum amount households can invest in a single period is their income in that period and that investment cannot be negative; i.e. It 0 (6) The household chooses consumption and investment to maximize the discounted present value of consumption. The resulting Euler equation at t is: U'(Ct ) = (L)U'(Ct ) + t (It ) (7) +1 where t (It ) is the Lagrange multiplier associated with constraint (6). The optimal levels of consumption and therefore investment equate the marginal utility of current consumption to the marginal utility of future consumption plus the marginal cost of the liquidity constraint on 17 investment. The solution in (7) provides us with the optimal marginal propensity to consume out of income, which we denote t. Consider the first two periods that households receive transfers from OPORTUNIDADES. In this case, the optimal consumption level in the second period after transfers begin is: Ct +1 =(L)[ +(L)(1-)]At +PTt +[(L)(1-)]PTt +t +[(L)(1-)]t -1 +1 +1 (8) where is the marginal propensity to consume and (1 ­ ) is the marginal propensity to invest15. Consumption is a function of initial assets and family labor, current transfers, lagged transfers and productivity shocks. The coefficient on current transfers is just the marginal propensity to consume. The coefficient on lagged transfer consists of three structural parameters. Working backwards, the first is the marginal propensity to invest (1-), which when multiplied times transfers yields the amount of the transfer invested. The second parameter is the marginal productivity of investment, (L), which when multiplied times investment yields income earned from the investment. The third parameter is the marginal propensity to consume, , which when multiplied times income yields the amount of income earned from the investment that is consumed. For each round of data from 1998 through 2003 where there is consumption data available, we estimate (8) as the following reduced form: t Cijt = o +1PTijt +2 PT ijt-s+ 3tWAVEt + k Xij, +vj +ijt (9) 97 s=1 t k where Cijt is monthly per capita adult equivalent consumption16 of household i living in community t j at time t; PTijt is the current monthly OPORTUNIDADES transfer per capita; and PT are 6- ijt-s s=1 months lagged total per capita cumulative transfers, i.e. the total transfer amount the household had accumulated until six months prior to the current round of the ENCEL survey (t). Since 15For simplicity, we assumed that the marginal propensity to consume is constant over time. However, if OPORTUNIDADES significantly relaxed the liquidity constraint, the marginal propensity to consume might fall over time. We tested and rejected this assumption against the alternative of a constant marginal propensity to consume. 16We define adult equivalent household size as the number of children age 12 or younger times 0.5 plus the number family members older than 12. 18 consumption is also a function of initial assets and household labor endowments, we include baseline household demographics, the community wage rate and head and spouse characteristics, i.e. the Xij,97 . As before, the error term is composed of an idiosyncratic disturbance, ijt , and a 17 community random effect, vj. The specification in (9) allows us to identify the marginal propensity to consume and the increase in consumption from investment. The marginal propensity to consume, , is just the coefficient on current transfers, 1. The rate of return in terms of increased income, , is2 1 1-1 , where 2 is the coefficient on lagged transfers, and the rate of return in terms of ( ) consumption is 2 1-1 .( ) 4.2 Estimated Treatment Effects We now turn to the analysis of the effects of current and lagged cumulative transfers on consumption. Our aim here is to determine the proportion of the cash transfer that is consumed directly out of current transfers, with the remainder being saved or invested. We then capture the long run effects of the program on consumption through the total cumulative transfer amount lagged by 6 months, and argue that any increased long run consumption is achieved through productive investments. We estimate (9) for the 4 rounds of the ENCEL for which we have detailed consumption information, including home produced consumption. As has been previously discussed, because reception of the transfer is conditional on school enrollment, the transfer amount received is endogenous to household behavior. Thus, we instrument current and lagged actual transfers with their "potential" counterparts18. The standard errors are clustered at the community level to control for the correlation of the productivity shocks within communities over time. 17 The full list of controls is kept the same as for the investment runs with the exception that baseline household size is replaced with baseline household size adult equivalent for the consumption runs. 18 Separate identification of lagged cumulative transfers from current transfers is possible because treatment households were randomly assigned benefits a year and a half before control households. 19 The consumption measurement used for the analysis is constructed as total household expenditures on food and non-food items, plus home produced consumption19. Expenditures and home produced consumption are separately identifiable in the data. Households are asked about the quantity and expenditures of goods purchased over the week before the interview, as well as how much of their own production of that good was consumed. We use community level prices to impute a value for household production. Previous research on the OPORTUNIDADES program estimates that monetary transfers from the program represent approximately 20% of pre-program consumption (Hoddinott and Skoufias 2004; Skoufias 2005). The summary statistics presented in Table 8 show a similar result, with the mean transfer amount in treatment households equal to 21% of consumption in control households. Comparing the consumption levels of households in treatment and control communities, we observe that treatment households consume approximately 22 pesos more per capita per month, representing 61% of the average per capita monthly transfer of 36 pesos. This simple comparison of consumption levels in treatment and control households suggests that households do not consume the entire transfer amount each month, and that some of the transfer is saved or invested. Table 9 presents least squares (LS) and two stage least squares (2SLS) estimates of equation (9) for the entire sub-sample of households. Instrumental variable estimates (including controls) on monthly transfers per capita indicate that, on average, 0.884 pesos of each peso transferred are consumed. The coefficient on cumulative transfers is 2 =0.018. These results ^ indicate that households consume over four fifths of the transfers directly. The household invests the remaining 0.116 pesos of each peso transferred, and obtains a return of 0.018 on each peso transferred in terms per capita consumption terms. The return on each peso invested ( ^ ) is 15.52% in terms of consumption and 17.55% in terms of income. 19Alternative consumption measures were imputed using quantity consumed times median community prices, and yield comparable results. The consumption measurement used here sums reported expenditures and the imputed value of home production for food and non-food items in a detailed consumption module. 20 The implied long term effects of the transfers on living standards through the investment pathways are dramatic. In 2003, after five and one-half years on the program, households in the treatment group received an average cumulative transfer of $3,444 pesos per capita. Through the increased income from invested transfers, the average beneficiary consumed 62 pesos more per month20. Thus, after five and a half years, transfers increased consumption by 34% through the investment pathway21. Since the 34% increase in consumption is through investment in income generating activities, it should continue even if the household dropped out of the program. These results also suggested that households were liquidity constrained. The estimated rate of return on investment, 17.55%, is substantially higher than the average annual real interest rate in the economy for the entire period (1997 to 2003), which was approximately 6.1%22. We take this result as evidence that these households face imperfect credit markets, since obtaining investments in productive activities would yield net gains. In fact, access to credit for the poor rural households in our sample is very limited. In the two rounds of the evaluation survey (May and November 1999) where questions on access to credit are asked, only 2.4% of households report using credit. 4.3 Heterogeneous Treatment Effects Table 10 presents 2SLS estimates of the effects of current and lagged transfers interacted with farm size in 1997, the age of the head of household, head's education, and distance (in Km) to the closest urban center. The purpose of such interactions is to determine whether household heterogeneity affects preferences (the marginal propensity to consume) and/or productivity (the rate of return). Non-agricultural households (households with no agricultural assets at baseline) have the highest marginal propensity to consume, while the other three types of households have a similar 20"Permanent" rise in consumption estimated as average cumulative transfers per capita for beneficiary households in the treatment group times the coefficient on returns from investment, ^2 , estimated as 0.018. 21The percentage increase is measured relative to first round of data where consumption data are available. In this survey eligible households in control communities have a real per-capita expenditure of $183 pesos. 22The annual real interest rate (computed on the 28-day Inter-Bank Equilibrium Interest Rate) was 5.5% in 1997 and 2.8% in 2003. The average annual real interest rate for the whole period is 6.1% (source: Bank of Mexico and National Institute of Geography and Statistics, INEGI). We take this rate as our benchmark since investments are likely to have occurred over the whole period. 21 marginal propensity to consume, not significantly different from 0.867. There is no significant difference in the effect of cumulative transfers on consumption either, with a constant return of 0.018 pesos per peso accumulated. Column 3 in Table 10 indicates that the older the head of the household, the lower the estimated marginal propensity to consume, resulting in a larger remainder left for saving or investment (result not statistically significant). However, this reduction in the marginal propensity to consume becomes smaller with age (positive coefficient on the age squared term). As one expects, there is a larger effect on consumption coming from lagged transfers for households with older heads, and once again, this increase becomes smaller over time. Thus, the transfer amount interacted with the age of the head of the household fits a concave function, which is consistent with a life-cycle model. Column 4 of Table 10 shows that having a positive amount of education reduces the per capita marginal propensity to consume out of current transfers and increases per capita consumption coming from lagged (invested) transfers (the omitted category is no education). Thus, years of education seem to enhance productive activities slightly. Such results are coherent with the effect on age if we assume that basic education has expanded over time in rural communities so that younger people are likely to be more educated than older people. The interaction between distance to an urban center and lagged transfers (Column 5 in Table 10) is negative and significant. Households in more isolated communities may have overall lower return on investments, likely due to the fact that input and output markets are either difficult to access or completely unavailable. The final column of table 10 shows that all the effects are robust to jointly including all interactions in the same specification. 5. Robustness Tests Despite the identification strategy adopted here, our estimates of the marginal propensities to consume are subject to a number of potential biases or alternative interpretations. Even in a perfect randomized setting, general equilibrium effects that arise as a result of the program may 22 drive the changes in household consumption and in saving/investment behavior. In what follows, we discuss a series of potential biases, the reasons why we should worry about them, and how they are resolved in this context. 5.1. Macro Income and Price Effects OPORTUNIDADES infused small rural communities with large amounts of transfers to over half the residents. The transfers could have had multiplier effects on income and cause price inflation in treatment communities. These community macro effects could have caused prices to rise in treatment areas faster than control areas. Therefore, our estimated effects on consumption could be increases in prices rather than in living standards. In addition, it is possible that the observed increases in micro-enterprise and farm activities could be driven by a community macro income effect derived from the presence of OPORTUNIDADES in the community, rather than by increased liquidity from the transfers at the household level. We address this concern in two ways. First, we compare agricultural wages for males, females and children between treatment and control communities. Second, we compare the consumption levels and investments in micro-enterprise and farm activities of ineligible households in treatment and control communities. Table 11 reports the comparisons of means for community wages between treatment and control communities, and the comparisons of mean consumption and investment levels for ineligible households after the implementation of the program. We find no significant differences between control and treatment areas for any of the variables. Moreover, Hoddinot and Skoufias (2004) find no differences for food prices between treatment and control communities, giving more support to the inexistence of local price and macro income effects. 5.2. Labor Supply Effects OPORTUNIDADES transfers could be taken in leisure by reducing labor supply. If so, income in treatment households would be reduced over time, as would consumption. Everything else held constant, treatment households would experience lower increases in consumption with respect to control households, the larger the transfer received. In this case, our estimate of the 23 marginal propensity to consume would be downward biased and so the upper bound for the "real" rate of return will be lower than the one we have considered so far. Parker and Skoufias (2000) study the program effects on time allocation, concluding that OPORTUNIDADES had no effect on adult labor supply. Furthermore, they find no evidence that the program increased leisure time amongst men and women in beneficiary households. However, OPORTUNIDADES should affect the labor supply of school-age children as it is designed to release them from work so that they can attend school. In fact, Parker and Skoufias (2000; 2001) do find evidence of changes in children's labor force participation in salaried and non- salaried activities, especially for boys; and increases in school attendance. Schultz (2004) also finds positive effects on primary and secondary school enrollment for boys and girls. As explained earlier, we control for this potential bias by instrumenting actual transfers with the potential transfers. 5.3. Fertility Effects Fertility decisions may also be affected by program participation. The program might generate perverse fertility incentives in order to maximize future transfers received. It might also alter the costs and/or benefits of having a child, as well as reduce the cost of fertility control (or increase its knowledge and availability). The effect on consumption is thus ambiguous since it will depend on household size. Moreover, larger households that receive larger transfer amounts might increase consumption through the income effect. Skoufias (2005) summarizes the findings of a series of evaluations on OPORTUNIDADES until the end of 1999. He reports no record of statistical evidence on fertility rates. In any case, any endogenous fertility responses will be accounted for by the use of the potential transfer (computed on baseline household demographics) as an instrument for the actual transfer amount received. 5.4. Public and Private Transfers OPORTUNIDADES transfers might also crowd out private transfers. The reduction in these transfers and therefore consumption implies that we would underestimate marginal propensity 24 to consume. Indeed there is some evidence that supports this potential bias. Using only the October 1998 cross-section collected six months after the intervention, Albarran and Attanasio (2002) find that both the likelihood of receiving a private transfer and the amount received -conditional on receiving private transfers, are significantly and negatively affected by the program. To assess the extent of the potential bias, we have replicated Albarran and Attanasio's results for the 3 periods for which data on private transfers are available (October 1998, November 1999 and November 2000). First, the amount of private transfers is very small as only 7% of households receive a private transfer in any given period. The estimation results are presented in Table 12. Panel AI reports the effect of OPORTUNDADES on receiving any private transfer and Panel AII reports the estimated effects on the amount of the private transfer received. We estimate four specifications of the OPORTUNIDADES treatment: in the first row we report the results with a OPORTUNIDADES treatment dummy, in the second row we replace the treatment dummy with the potential OPORTUNIDADES transfer amount, in the third row we add potential transfers squared, and in the fourth row we instrument actual OPORTUNIDADES transfers and actual transfers squared with potential transfers and potential transfers squared. We find that the transferred squared models have significant effects, with the marginal effect size falling with the transfer amount. The same argument could be made for other public transfers. To test this hypothesis, we estimated the effects of OPORTUNIDADES transfers on other public transfers and report the results in Panels BI and BII in Table 12. We find results similar to the private transfer effects. To test the robustness of our main results to the effect of OPORTUNIDADES on private and other public transfers, we estimate a variant of the consumption equation (9) by adding private and other public transfers to OPORTUNIDADES transfers. If the coefficients on current transfers and lagged cumulative transfers do not change by adding private and other transfers, then we will conclude that there is no bias. The results are presented in Panels A and B of Table 13, for private and other public transfers respectively. The first column in Panel A, estimates the original 25 specification with only OPORTUNIDADES transfers in equation (9) using the data only from the waves that collected both private transfers and other public transfers using potential transfers as the instrument. The second column reports the results when private transfers are added to OPORTUNIDADES transfers. There is no difference between the estimated coefficients in the two models. Column 3 reports the results when we additionally add other public transfers to the OPORTUNIDADES and private transfers. Again there are no differences. In Panel B we re- estimate the specifications in Panel A using only other public transfers. The first column reports the base results using only OPORTUNIDADES transfers on data from the waves that collected other public transfers. The second column adds in other public transfers. There is no difference between columns 1 and 2 in Panel B. Finally, column 3 controls for whether the family also receives a public food basket. Again, the estimated coefficients do not change. These results imply that our estimates are not biased due to the exclusion of public and private transfers from the analysis. 5.5. Health Effects We also test the possibility that beneficiary productivity and consumption increased through improvements in health brought on by participation in OPORTUNIDADES. Panel CI of Table 12 estimates the effect of treatment on health, measured by activities of daily living (ADL)23. We show that the program does have small positive effect on health. Given this potential pathway, we again estimate our consumption model controlling for health status (ADL) of the head of household on a limited sub-sample where health data are available. The results are presented in Panel C of Table 13, and again we find that the estimated coefficients on current and cumulative lagged transfers do not change controlling for ADLs. This suggests that our main results are not biased by not controlling for health. 5.6. Length of Benefits Expectations 23Activity of Daily Living measurements are based on an individuals' self-rating of ability to engage in normal daily activities, including ability to carry out vigorous and moderate activities, ability to carry groceries, ability to lift a piece of paper off the ground, ability to walk more than 2km and ability to use the bathroom and bathe unassisted. 26 Friedman's (1957) Permanent Income Hypothesis (PIH) contended that permanent income and not current income was the relevant determinant of consumption, and that permanent consumption was proportional to permanent income (the proportionality hypothesis, which implies that the permanent income elasticity is unity)24. Such a model would suggest that the marginal propensity to consume out of permanent income should be close to 1, and the marginal propensity to consume from transitory income should be close to 0. The implication of the PIH for our results is that if beneficiary households view the transfers as transitory, then they would save most of them; if they view them as permanent, then they would consume most of them. The proportionality hypothesis and its major implications regarding growth and equity in developing countries have been tested many times, with ambiguous results. For example, Bhalla (1979) finds an estimated marginal propensity to consume out of permanent income of 0.61 and an elasticity of 0.79 (and significantly different from 1) using data from rural India. Wolpin (1982) estimates permanent income elasticities in the range of 0.91 to 1.02 for India as well and concludes that "whether or not it is unity, the permanent income elasticity of consumption may be closer to unity than suggested by previous research". Musgrove (1979) estimates 0.881 for urban Colombia, 0.896 for urban Ecuador, and, 0.776 for urban Peru; all significantly different from 1. Finally, Paxson (1992) estimates marginal propensities to consume from 0.56 to 0.84 out of permanent income, being significantly different from 1; and from 0.17 to 0.27 out of transitory income for rice farmers in Thailand. Our estimate of the marginal propensity to consume out of current transfers is 0.88. In view of the estimated magnitudes of marginal propensities to consume in the development literature, we are inclined to think that the rural Mexican households studied here perceive the OPORTUNIDADES transfer as a source of permanent income. Indeed, households know that once they start receiving the benefit they will continue to receive it for at least 9 years provided they 24The Modigliani-Brumberg life cycle hypothesis (LCH) reaches analogous conclusions. 27 comply with the program requirements. Thus, their expectations regarding the length of the benefits are likely to be rather long term. 5.7. Precautionary Savings We have argued that the evidence on farming and micro-enterprise investment responds to an alleviation of risk aversion and liquidity and credit constraints by the program. However, in an uncertain environment, one might expect households to accumulate wealth to smooth consumption over unexpected shocks. This motive is stronger the higher the uncertainty faced by consumers (Carroll and Kimball 1996). Moreover, if their banking services are scarce, savings are likely to take the form of assets such as land and livestock. For instance, Ravallion and Chen (2005) show that much of the impact of an anti-poverty program in Southwest China did not occur through consumption, but rather through savings, since participants treated income from the program as temporary. If the increase in farm asset ownership, farming and micro-enterprise we observe here is due primarily to precautionary savings rather than investment behavior, the expected increases in long run living standards may not be sustained if households stop saving when the transfer is removed. To rule this possibility out, we test for the existence of a precautionary motive and how the existence of OPORTUNIDADES affected precautionary savings. We estimate agricultural assets (animal ownership and land usage) in 2003 as a function of adverse shocks in the community and the accumulated actual transfer amount received by the household. We construct the history of shocks from natural causes (principally droughts and plagues) from early 1997 to late 2000 and categorize a community as a "high risk area" if it has suffered more than eight shocks during this period25. If there is a precautionary motive, then households in riskier areas (with a larger shock history) should have larger investments in agricultural assets in 2003. A dummy for high risk area is interacted with the accumulated transfer amount. If beneficiary households are using part of the 25The number of shocks per community in this period varies from 0 to 13, with a median of 5 shocks and an average of 5.41 shocks. The proportion of treatment and control communities in high risk areas is the same as the proportion of treatment and control communities in the sample. However, control communities tend to report a larger number of shocks than treatment communities. 28 OPORTUNIDADES transfer for precautionary savings the estimated coefficient on this term should be positive. The coefficient on accumulated actual transfers, which we instrument with its potential counterpart, is simply the "per peso" treatment effect. Results are shown in Panel A of Table 14 for a linear specification in transfers and in Panel B for a quadratic specification. Production animals and land are the channels of wealth accumulation used by these households to ensure their future income. We find that households living in riskier areas are more likely to use land and have production animals, suggesting a precautionary savings motive. However, none of the OPORTUNIDADES money is used with precautionary motives. In fact, the interaction term is negative and significant for land and production animals suggesting that beneficiary households in riskier areas lower their precautionary savings. We find a similar effect for draft animals once we introduce a quadratic term for the transfer amount. Households in treatment areas may be better insured against future adversities because households view the transfer as a permanent source of income. Thus, this analysis rejects the precautionary savings hypothesis as an alternative explanation for the investments results. 6. Conclusion The analysis conducted in this study provides evidence that cash transfer payments from the OPORTUNIDADES program increase consumption not only through direct expenditures out of current transfers, but also through the income generated from investing part of the transfers in farms and micro-enterprises. Beneficiary households experience large increases in participation in micro- enterprise activities and increased investments in farm assets and agricultural activities. Furthermore, households that receive the largest transfers are most likely to invest. Conditional cash transfer programs such as OPORTUNIDADES are designed to alleviate short-term poverty while investing in the human capital of the next generation. The positive benefits of this type of intervention are well established, including increased caloric intake, better health and nutrition, and higher school enrollment for children. This increased human capital is expected to play an important role in breaking the cycle of poverty for younger generations. 29 However, this paper shows that there are also more immediate implications of cash transfers for alleviating long-term poverty for the current generation of beneficiary households. Cash transfers have the potential to increase income permanently by facilitating investments in productive activities. Furthermore, we argue that the primary mechanisms for achieving this result are through reduced liquidity constraints and decreased risk aversion. Although we do not argue that cash transfer programs are necessarily the most desirable policy for promoting micro-enterprise or farm investments for poor households in rural areas, it is clear that increased entrepreneurial activity brought on by cash transfers have increased the potential for self-sufficiency. The results presented here suggest that a permanent rise in consumption through increased economic activity may ultimately reduce long-run welfare dependence and permit beneficiary households to attain a higher standard of living that can be sustained even in the absence of the transfer program. Further understanding of the mechanisms through which cash transfers boost productive investments (softening of liquidity and/or credit constraints, reduction of risk aversion, insurance role) is crucially important in the determination and design of future policies to be undertaken, such as micro-lending, business incentives and advising or the provision of insurance schemes, for the alleviation of poverty. 30 References Aghion, Philippe and Patrick Bolton. 1997. "A Theory of Trickle-Development". The Review of Economic Studies, 64(2):151-172. Albarran, Pedro and Orazio Attanasio. 2002. "Do Public Transfers Crowd out Private Transfers? Evidence from a Randomized Experiment in Mexico". WIDER discussion paper 2002/6. Banerjee, Abhijit. 2004. "Contracting Constraints, Credit Markets, and Economic Development" in Advances in Economics and Econometrics: Theory and Applications, VIII World Congress of the Econometric Society, vol. 3: 1-46, Mathias Dewatripont, Lars P. Hansen and Stephen J. Turnovsky, eds. Cambridge University Press. Banerjee, Abhijit and Andrew F. Newman. 1993. "Occupational Choice and the Process of Development". Journal of Political Economy, 101(2):274-298. Banco de México. 2006. www.banxico.org.mx Bhalla, Surjit S. 1979. "Measurement Errors and the Permanent Income Hypothesis: Evidence from Rural India". American Economic Review, 69(3):295-307. Carroll, Christopher D. and Miles S. Kimball. 1996. "On the Concavity of the Consumption Function". Econometrica, 64(4):981-992. Friedman, Milton. 1957. A Theory of the Consumption Function, Princeton N.J. Princeton University Press. Hall, Robert E. 1978. "Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: Theory and Evidence". Journal of Political Economy, 86(6):971-987. Hoddinott, John and Emmanuel Skoufias. 2004. "The Impact of OPORTUNIDADES on Consumption". Economic Development and Cultural Change, 53(1):37-61. Instituto Nacional de Estadística, Geografía e Informática (INEGI). 2006. www.inegi.gob.mx Lindh, Thomas and Henry Ohlsson. 1998. "Self-employment and wealth inequality". Review of Income and Wealth, 44(1):25-42. Lloyd-Ellis, Huw and Dan Bernhardt. 2000. "Enterprise, Inequality and Economic Development". The Review of Economic Studies, 67(1):147-168. McKenzie, David and Christopher Woodruff. 2006. "Do Entry Costs Provide an Empirical Basis for Poverty Traps? Evidence from Mexican Microenterprises". Economic Development and Cultural Change, forthcoming. Modigliani, Franco and Richard H. Brumberg. 1954. "Utility Analysis and the Consumption Function: An Interpretation of Cross-Section Data" in Post Keynesian Economics, Kenneth K. Kurihara ed., New Brunswick:388-436. Moffit, Robert. 1992. "Incentive Effects of the U.S. Welfare System: A Review". Journal of Economic Literature, 30(1):1-61. Musgrove, Philip. 1979. "Permanent Household Income and Consumption in Urban South America". American Economic Review, 69(3):355-368. OPORTUNIDADES. 2003. "Reglas de Operación 2003". www.progresa.gob.mx OPORTUNIDADES. 2006: www.oportunidades.gob.mx Parker, Susan W. and Emmanuel Skoufias. 2000. "Final Report: The Impact of PROGRESA on Work, Leisure, and Time Allocation". Final Report, International Food Policy Research Institute, Washington D.C. Parker, Susan W. and Emmanuel Skoufias. 2001. "Conditional Cash Transfers and their Impact on Child Work and School Enrollment: Evidence from the PROGRESA Program in Mexico". Economía, 2:45-96. Paxson, Christina. 1992. "Using Weather Variability to Estimate the Response of Savings to Transitory Income in Thailand". American Economic Review, 82(1):15-33. Ravallion, Martin and Shaohua Chen. 2005. "Hidden Impact? Household Saving in Response to a Poor-Area Development Project". Journal of Public Economics, 89(11-12):2183-2204. 31 Sadoulet, Elisabeth, Alain de Janvry and Benjamin Davis. 2001. "Cash Transfer Programs with Income Multipliers: PROCAMPO in Mexico". World Development, 29(6):1043-1056. Schultz, Paul. 2004. "School Subsidies for the Poor: Evaluating the Mexican PROGRESA Poverty Program". Journal of Development Economics, 74(1):199-250. Skoufias, Emmanuel. 2005. "PROGRESA and Its Impacts on the Welfare of Rural Households in Mexico", International Food Policy Research Institute Research Report 139, Washington D.C. Skoufias, Emmanuel, Benjamin Davis and Sergio de la Vega. 2001. "Targeting the Poor in Mexico: An Evaluation of the Selection of Households into PROGRESA". World Development, 29(10):1768-1784. Todd, Petra E. 2004. "Technical Note on Using Matching Estimators to Evaluate the OPORTUNIDADES Program for Six Year Follow-up Evaluation of OPORTUNIDADES in Rural Area". Mimeo, University of Pennsylvania. Wolpin, Kenneth I. 1982. "A New Test of the Permanent Income Hypothesis: The Impact of Weather on the Income and Consumption of Farm Households in India". International Economic Review, 23(3):583-594. Woodruff, Christopher and Rene Zenteno. 2001. "Remittances and Microenterprises in Mexico". University of California at San Diego, Graduate School of International Relations and Pacific Studies Working Paper. Yang, Dean. 2005. "International Migration, Human Capital, and Entrepreneurship: Evidence from Philippine Migrants' Exchange Rate Shocks". World Bank Policy Research Working Paper No. 3578, The World Bank, Washington D.C. 32 APPENDIX 1 ­TABLES Table 1: OPORTUNIDADES Monthly Transfer Amounts at Baseline (October 1997) Transfer Component Level Grade Boys Girls Education Stipend Primary School 3rd year 60 60 4th year 70 70 5th year 90 90 6th year 120 120 Junior High School 1st year 175 185 2nd year 185 205 3rd year 195 225 High School1 1st year 470 540 2nd year 505 575 3rd year 535 610 School Supplies Stipend Primary, 1rst payment 80 80 Primary, 2nd payment 40 40 Junior High School 150 150 High School1 240 240 Nutritional Stipend (per family) 90 Transfer Cap I (per family)2 550 Transfer Cap II (per family)3 700 Source: OPORTUNIDADES (www.oportunidades.gob.mx). Transfer amounts adjusted for inflation every semester according to the Consumer Price Index published by the Bank of Mexico. 1High school stipends only granted beginning in the second semester of 2001 (July 2001). 2Transfer Cap I is the maximum transfer amount awarded for basic education (primary school and junior high) and nutrition. 3Transfer Cap II is the maximum transfer amount given for high school and nutrition. Table 2: Sample Sizes and Take-Up Rates1 Treatment Control All N % N % N % Sample of Eligible (Poor) Households Number Non Take-Up Households 839 10.96 485 10.44 1324 10.76 Number Take-Up Households (Actually Treated -TOT) 6,819 89.04 4,159 89.56 10,978 89.24 Total Number of Households (Intent to Treat -ITT) 7,658 4,644 12,302 Number of Communities 320 63.24 186 36.76 506 1Take-up control households have received at least one bimonthly payment by the time all eligible households should have been incorporated (November 2000). Take-up treatment households must have received their first payment before any eligible control household is phased into the program (November 1999). 2We drop 116 households that receive a total transfer amount higher than the maximum they could have potentially received according to their household demographics and treatment status. Households with heads and/or head's spouses younger than 13 or older than 90 also dropped. Drop outs are balanced between the treatment and control samples. 33 Table 3A: Test of Equality of Means between Potential Treatments and Potential Controls Prior to Program Implementation Sub-Sample of Original Poor at Baseline (October 1997) -Intent to Treat (ITT) Treatment Group Control Group N Mean SD N Mean SD t-stat Intent to Treat Sub-Sample (ITT) -Explanatory Variables Head's Characteristics Age Household Head 7655 41.95 14.133 4643 42.33 14.521 -0.943 Female Head =1 7658 8.16 0.274 4644 8.40 0.277 -0.377 Indigenous Head =1 7646 41.90 0.493 4636 43.70 0.496 -0.324 Head's Education (Years) 5249 4.06 2.306 3143 3.95 2.250 1.148 Never Attended School (Head of Household) =1 7634 31.44 0.464 4637 32.65 0.469 -0.541 Primary School Not Completed (Head of Household) =1 7634 46.28 0.499 4637 46.75 0.499 -0.275 Primary School Completed (Head of Household) =1 7634 16.94 0.375 4637 15.59 0.363 0.988 More than Primary School (Head of Household) =1 7634 5.34 0.225 4637 5.00 0.218 0.516 Spouse's Characteristics Age Spouse of Head 6713 36.27 12.225 4074 36.17 12.223 0.296 Spouse's Education (Years) 4291 4.16 2.124 2551 4.21 2.243 -0.603 Never Attended School (Head's Spouse) =1 6700 36.18 0.481 4067 37.35 0.484 -0.411 Primary School Not Completed (Head's Spouse) =1 6700 41.64 0.493 4067 40.69 0.491 0.477 Primary School Completed (Head's Spouse) =1 6700 18.37 0.387 4067 16.97 0.375 0.986 More than Primary School (Head's Spouse) =1 6700 3.81 0.191 4067 4.99 0.218 -1.953+ Main Entrepreneur's Characteristics Age Main Entrepreneur in the Household 823 0.53 0.499 338 0.54 0.499 -0.178 Education Years Main Entrepreneur in the Household 818 0.35 0.477 335 0.36 0.480 -0.208 Female Main Entrepreneur in the Household =1 818 2.78 2.771 335 2.61 2.582 0.676 Main Entrepreneur is the (likely) Beneficiary Mother =1 822 40.29 13.919 338 40.25 14.299 0.040 Household Characteristics Presence of Children 0 to 7 =1 7645 75.36 0.431 4639 77.04 0.421 -1.396 Presence of Children 8 to 17 =1 7645 69.01 0.462 4639 69.54 0.460 -0.472 Presence of Adult Men 18 to 54 =1 7645 83.69 0.369 4639 83.75 0.369 -0.057 Presence of Adult Female 18 to 54 =1 7645 90.26 0.297 4639 90.80 0.289 -0.699 Presence of Adults Older than 55 =1 7644 27.00 0.444 4639 27.20 0.445 -0.177 Household Size 7658 5.91 2.433 4644 5.94 2.413 -0.457 Home Ownership =1 7655 93.39 0.248 4643 92.10 0.270 1.488 Dirt Floor =1 7641 73.17 0.443 4630 75.72 0.429 -1.015 Electricity =1 7653 57.65 0.494 4642 60.58 0.489 -0.684 No Agricultural Assets =1 7634 10.40 0.305 4629 11.67 0.321 -1.047 Landless Farms =1 7634 10.40 0.305 4629 11.67 0.321 -1.047 Small Landed Farms =1 7634 30.59 0.461 4629 32.45 0.468 -0.828 Big Farms =1 7634 46.86 0.499 4629 40.94 0.492 2.229* 7634 12.16 0.327 4629 14.95 0.357 -1.876 Community Characteristics Village Associations (Community Work) 7658 87.84 0.327 4644 87.75 0.328 0.021 Minimum Distance to Large Urban Centre (Km) 7658 107.73 41.361 4644 105.11 43.721 0.522 Monthly Community Agricultural Male Wage 4426 573.86 172.005 2781 585.23 172.115 -0.463 Notes: +significant at 10%, *significant at 5%, **significant at 1%.T-stat of differences in means computed clustering SE at the community level. Mean of dichotomous variables expressed in percentages. 34 Table 3B: Test of Equality of Means between Actual Treatments and Actual Controls Prior to Program Implementation Sub-Sample of Original Poor at Baseline (October 1997) -Treatment on the Treated (TOT) Treatment Group Control Group N Mean SD N Mean SD t-stat Treatment on the Treated Sub-Sample (TOT) -Explanatory Variables Head's Characteristics Age Household Head 6818 42.01 13.898 4158 42.49 14.381 -1.140 Female Head =1 6819 7.86 0.269 4159 7.89 0.270 -0.042 Indigenous Head =1 6809 41.72 0.493 4151 43.97 0.496 -0.403 Head's Education (Years) 4745 4.03 2.268 2841 3.91 2.206 1.275 Never Attended School (Head of Household) =1 6797 30.38 0.460 4154 32.04 0.467 -0.796 Primary School Not Completed (Head of Household) =1 6797 47.21 0.499 4154 47.64 0.500 -0.257 Primary School Completed (Head of Household) =1 6797 17.18 0.377 4154 15.53 0.362 1.172 More than Primary School (Head of Household) =1 6797 5.22 0.223 4154 4.79 0.214 0.668 Spouse's Characteristics Age Spouse of Head 6037 36.43 12.114 3693 36.47 12.120 -0.111 Spouse's Education (Years) 3917 4.12 2.090 2322 4.15 2.160 -0.287 Never Attended School (Head's Spouse) =1 6027 35.26 0.478 3687 37.08 0.483 -0.669 Primary School Not Completed (Head's Spouse) =1 6027 42.49 0.494 3687 41.58 0.493 0.464 Primary School Completed (Head's Spouse) =1 6027 18.57 0.389 3687 16.79 0.374 1.235 More than Primary School (Head's Spouse) =1 6027 3.68 0.188 3687 4.56 0.209 -1.451 Main Entrepreneur's Characteristics Age Main Entrepreneur in the Household 760 40.38 13.964 318 40.42 14.405 -0.035 Education Years Main Entrepreneur in the Household 757 2.77 2.764 315 2.64 2.555 0.519 Female Main Entrepreneur in the Household =1 761 53.35 0.499 318 53.77 0.499 -0.061 Main Entrepreneur is the (likely) Beneficiary Mother =1 757 35.54 0.479 315 35.87 0.480 -0.069 Household Characteristics Presence of Children 0 to 7 =1 6810 75.89 0.428 4157 77.24 0.419 -1.101 Presence of Children 8 to 17 =1 6810 70.47 0.456 4157 70.99 0.454 -0.435 Presence of Adult Men 18 to 54 =1 6810 84.33 0.364 4157 84.44 0.363 -0.103 Presence of Adult Female 18 to 54 =1 6810 91.17 0.284 4157 91.46 0.280 -0.378 Presence of Adults Older than 55 =1 6810 26.77 0.443 4157 27.54 0.447 -0.662 Household Size 6819 6.00 2.416 4159 6.04 2.405 -0.583 Home Ownership =1 6816 93.87 0.240 4158 92.83 0.258 1.199 Dirt Floor =1 6803 72.59 0.446 4145 75.49 0.430 -1.116 Electricity =1 6815 58.94 0.492 4158 61.21 0.487 -0.529 No Agricultural Assets =1 6798 9.55 0.294 4145 10.78 0.310 -1.012 Landless Farms =1 6798 30.77 0.462 4145 32.06 0.467 -0.570 Small Landed Farms =1 6798 47.16 0.499 4145 42.17 0.494 1.84+ Big Farms =1 6798 12.52 0.331 4145 14.98 0.357 -1.685 Community Characteristics Village Associations (Community Work) 6819 89.18 0.311 4159 87.21 0.334 0.432 Minimum Distance to Large Urban Centre (Km) 6819 107.53 41.857 4159 105.06 43.780 0.486 Monthly Community Agricultural Male Wage 4006 573.79 170.117 2498 579.80 163.320 -0.249 Notes: +significant at 10%, *significant at 5%, **significant at 1%.T-stat of differences in means computed clustering SE at the community level. Mean of dichotomous variables expressed in percentages. 35 Table 4: Exogeneity Test and Average Treatment Effect -Test of Equality of Means between Actual Treatments and Actual Controls Sub-Sample of Original Poor -Treatment on the Treated (TOT) Treatment Group Control Group N Mean SD N Mean SD t-stat A. Exogeneity Test -Dependent Variables at Baseline (1997) AI. Entire Sub-Sample Draft Animals Ownership 6819 35.15 0.477 4159 32.53 0.469 1.109 Production Animals Ownership 6817 81.90 0.385 4158 82.56 0.379 -0.376 Land Use 6819 59.77 0.490 4159 57.30 0.495 0.856 Number Draft Animals 2391 0.98 1.999 1350 0.89 1.422 0.874 Number Production Animals 5576 1.25 2.281 3425 1.23 2.356 0.236 Number Hectares Used 4057 2.68 2.759 2370 2.94 2.851 -1.564 AIII. Landless Households in 97 =1 Draft Animals Ownership 2092 26.00 0.439 1329 22.72 0.419 1.285 Production Animals Ownership 2092 96.99 0.171 1329 97.07 0.169 -0.097 Number Draft Animals 543 0.75 2.714 301 0.56 0.913 1.437 Number Production Animals 2026 0.76 1.494 1287 0.70 1.448 0.796 AIV. Smaller Farms in 97 =1 Draft Animals Ownership 3206 41.08 0.492 1748 39.59 0.489 0.452 Production Animals Ownership 3206 85.90 0.348 1748 88.39 0.320 -1.288 Number Draft Animals 1313 0.97 1.604 691 0.89 1.484 0.689 Number Production Animals 2752 1.20 1.976 1541 1.18 2.086 0.168 Number Hectares Used 3206 1.60 0.749 1748 1.66 0.749 -1.472 AV. Bigger Farms in 97 =1 Draft Animals Ownership 851 61.46 0.487 621 56.52 0.496 1.191 Production Animals Ownership 851 91.89 0.273 621 94.36 0.231 -1.602 Number Draft Animals 522 1.22 1.998 351 1.16 1.562 0.330 Number Production Animals 780 2.68 3.760 585 2.41 3.579 0.851 Number Hectares Used 851 6.74 3.654 621 6.53 3.454 0.664 B. Average Treatment Effect -Dependent Variables Oct 98 to Nov 99 BI. Entire Sub-Sample Draft Animals Ownership =1 19406 29.19 0.455 11805 24.88 0.432 2.013* Production Animals Ownership =1 19406 74.81 0.434 11805 70.99 0.454 2.115* Land Use =1 19409 61.62 0.486 11805 59.06 0.492 0.998 Number of Draft Animals 5641 0.66 1.070 2928 0.62 0.994 0.658 Number of Production Animals 14490 1.02 1.951 8357 0.90 1.829 1.437 Number of Hectares Used 11886 2.10 2.182 6928 2.10 2.171 -0.009 Household has a Micro-Enterprise =1 19409 8.48 0.279 11805 5.85 0.235 1.855+ Household has a Female Micro-Enterprise =1 19409 8.11 0.273 11805 5.13 0.221 2.217* BII. No Agricultural Assets Households in 97 =1 Draft Animals Ownership =1 1814 7.55 0.264 1212 9.16 0.289 -0.937 Production Animals Ownership =1 1814 50.88 0.500 1212 42.16 0.494 2.862** Land Use =1 1816 27.75 0.448 1212 22.03 0.415 1.719+ Number of Draft Animals 137 0.41 0.580 110 0.34 0.266 1.008 Number of Production Animals 921 0.41 0.901 510 0.37 0.803 0.453 Number of Hectares Used 503 1.25 1.403 264 1.36 1.369 -0.854 Household has a Micro-Enterprise =1 1816 7.21 0.259 1212 4.37 0.205 1.203 Household has a Female Micro-Enterprise =1 1816 6.77 0.251 1212 3.63 0.187 1.395 BIII. Landless Households in 97 =1 Draft Animals Ownership =1 5908 24.68 0.431 3768 19.03 0.393 2.438** Production Animals Ownership =1 5908 75.00 0.433 3768 72.03 0.449 1.496 Land Use =1 5908 41.22 0.492 3768 36.73 0.482 1.683+ Number of Draft Animals 1451 0.49 0.888 714 0.44 0.606 1.166 Number of Production Animals 4426 0.77 1.454 2706 0.64 1.316 1.767+ Number of Hectares Used 2425 1.61 1.860 1379 1.55 1.580 0.653 Household has a Micro-Enterprise =1 5908 8.26 0.275 3768 5.52 0.228 1.761+ Household has a Female Micro-Enterprise =1 5908 7.97 0.271 3768 4.99 0.218 1.949* BIV. Smaller Farms in 97 =1 Draft Animals Ownership =1 9171 31.21 0.463 5004 27.26 0.445 1.369 Production Animals Ownership =1 9171 76.91 0.421 5004 73.90 0.439 1.343 Land Use =1 9172 75.44 0.430 5004 76.08 0.427 -0.321 Number of Draft Animals 2847 0.68 1.126 1359 0.58 0.792 1.669+ Number of Production Animals 7046 1.03 1.847 3688 0.87 1.660 1.666+ Number of Hectares Used 6887 1.90 1.903 3798 1.81 1.595 1.121 Household has a Micro-Enterprise =1 9172 8.93 0.285 5004 6.10 0.239 1.579 Household has a Female Micro-Enterprise =1 9172 8.50 0.279 5004 5.14 0.221 1.996* BV. Bigger Farms in 97 =1 Draft Animals Ownership =1 2451 48.51 0.500 1784 40.81 0.492 1.994* Production Animals Ownership =1 2451 83.92 0.367 1784 79.88 0.401 1.804+ Land Use =1 2451 83.80 0.369 1784 83.02 0.376 0.379 Number of Draft Animals 1187 0.83 1.103 728 0.87 1.351 -0.307 Number of Production Animals 2043 1.77 2.953 1423 1.62 2.770 0.690 Number of Hectares Used 2027 3.56 2.832 1460 3.50 3.178 0.213 Household has a Micro-Enterprise =1 2451 8.08 0.273 1784 7.01 0.255 0.891 Household has a Female Micro-Enterprise =1 2451 7.79 0.268 1784 6.56 0.248 1.041 Notes: +significant at 10%, *significant at 5%, **significant at 1%. T-stat of differences in means computed clustering SE at the community level. Mean of dichotomous variables expressed in percentages. Smaller farms are landed households using, at least, 3 Ha. of land; bigger farms use more than 3 Ha. Number of draft or production animals are expressed in equivalent cow units. Continuous variables conditional on being positive. 36 Table 5: Effect of OPORTUNIDADES on Agricultural Investments -Treatment and Control Variation Sub-sample of Original Poor Households in October 1998, May 1999 and November 1999 -Intent to Treat (ITT) and Actual Treatments (TOT) Draft Animal Production Production Animal PANEL I: DICHOTOMOUS OUTCOMES Draft Animal Ownership Animal Ownership Land Use Ownership p(y|y97 =0) Ownership p(y|y97 =0) Land Use p(y|y97 =0) (PROBIT) (PROBIT) (PROBIT) (PROBIT) (PROBIT) (PROBIT) Model A: ITT -no controls Treatment Status 0.043* 0.025+ 0.038* 0.065* 0.029 0.053* (0.021) (0.013) (0.018) (0.027) (0.025) (0.026) Model B: TOT -no controls Treatment Status 0.043* 0.024+ 0.039* 0.062* 0.026 0.050+ (0.021) (0.013) (0.018) (0.029) (0.026) (0.027) Model C: ITT -controls Treatment Status 0.046** 0.023* 0.037* 0.061* 0.028 0.053* (0.018) (0.011) (0.016) (0.027) (0.025) (0.025) Model D: TOT -controls Treatment Status 0.044* 0.022* 0.037* 0.062* 0.028 0.052* (0.017) (0.011) (0.017) (0.028) (0.025) (0.025) Model E: TOT by Farm Size at Baseline (1997) -controls Treatment Status * No Agricultural Assets -0.009 -0.008 0.075** 0.094** 0.063+ 0.065+ (0.034) (0.019) (0.022) (0.033) (0.038) (0.039) Treatment Status * Landless 0.065* 0.035* 0.029+ 0.096 0.045+ 0.049+ (0.026) (0.017) (0.018) (0.075) (0.025) (0.025) Treatment Status * Smaller Farm 0.038+ 0.029+ 0.028 -0.003 -0.002 - (0.022) (0.016) (0.020) (0.041) (0.025) - Treatment Status * Bigger Farm 0.046 0.000 0.035 0.007 0.026 - (0.028) (0.022) (0.023) (0.083) (0.032) - Observations ITT Sub-Sample (Model A and C) 33313 21899 33313 5964 33316 13641 Mean Dependent Variable ITT Sub-Sample (All) 0.275 0.124 0.729 0.494 0.605 0.360 Mean Dependent Variable ITT Sub-Sample (Controls) 0.248 0.108 0.706 0.453 0.588 0.329 Observations TOT Sub-Sample (Models B, D, E) 31211 20477 31211 5457 31214 12708 Mean Dependent Variable TOT Sub-Sample (All) 0.276 0.123 0.734 0.500 0.606 0.361 Mean Dependent Variable TOT Sub-Sample (Controls) 0.248 0.109 0.710 0.461 0.591 0.332 Number of Number of Number of Number of Equivalent Equivalent Equivalent Number of Number of PANEL II: CONTINUOUS OUTCOMES Equivalent Draft Animals Production Production Animals Hectares Hectares Used Draft Animals p(y|y97 >0) Animals p(y|y97 >0) Used p(y|y97 >0) (LS) (LS) (LS) (LS) (LS) (LS) Model A: ITT -no controls Treatment Status 0.030 0.056 0.086 0.103 0.031 0.008 (0.025) (0.035) (0.075) (0.080) (0.089) (0.115) Model B: TOT -no controls Treatment Status 0.029 0.058 0.090 0.103 0.036 0.023 (0.025) (0.037) (0.072) (0.076) (0.091) (0.117) Model C: ITT -controls Treatment Status 0.030 0.060+ 0.084 0.099 0.026 0.014 (0.022) (0.033) (0.067) (0.071) (0.084) (0.112) Model D: TOT -controls Treatment Status 0.031 0.062+ 0.092 0.100 0.038 0.031 (0.023) (0.034) (0.065) (0.068) (0.086) (0.113) Model E: TOT by Farm Size at Baseline (1997) -controls Treatment Status * No Agricultural Assets 0.005 - 0.022 - 0.047 - (0.030) - (0.082) - (0.090) - Treatment Status * Landless 0.019 0.017 0.012 0.007 0.060 - (0.024) (0.047) (0.066) (0.070) (0.070) - Treatment Status * Smaller Farm 0.071** 0.078* 0.119+ 0.125+ 0.036 0.055 (0.023) (0.038) (0.064) (0.069) (0.068) (0.097) Treatment Status * Bigger Farm 0.088** 0.112* 0.234** 0.291** 0.162* 0.182 (0.028) (0.046) (0.076) (0.083) (0.082) (0.111) Observations ITT Sub-Sample (Model A and C) 33313 11385 33313 27301 33193 19573 Mean Dependent Variable ITT Sub-Sample (All) 0.176 0.403 0.722 0.825 1.265 1.763 Mean Dependent Variable ITT Sub-Sample (Controls) 0.152 0.367 0.651 0.741 1.239 1.790 Observations TOT Sub-Sample (Models B, D, E) 31211 10707 31211 25708 31097 18406 Mean Dependent Variable TOT Sub-Sample (All) 0.152 0.367 0.639 0.725 1.240 1.780 Mean Dependent Variable TOT Sub-Sample (Controls) 0.176 0.403 0.715 0.815 1.273 1.767 Notes: +significant at 10%, *significant at 5%, **significant at 1%. SE in parantheses. For the probit models, marginal effects and robust standard errors reported, clustered at the community level. LS regressions include community level random effects. All regressions include wave dummies. Controls in models C to E include: head's and spouse's age, age squared and education level dummies, head's ethnicity (language), household size, dummies controlling for household demographics, baseline assets (home ownership, dirt floor and electricity) and community characteristics (community organizations, distance to urban center and wages). Model E controls for farm size in October 1997 (baseline). Smaller farms use up to 3 Ha. of land; bigger farms use more than 3 Ha. 37 Table 6: Effect of OPORTUNIDADES on Micro-Entepreneurial Investments -Treatment and Control Variation. Sub-sample of Original Poor Households in October 1998, May 1999 and November 1999 -Intent to Treat (ITT) and Actual Treatments (TOT) Female Micro-Enterprise Micro-Enterprise (PROBIT) (PROBIT) Model A: ITT -no controls Treatment Status 0.026+ 0.030* (0.014) (0.013) Model B: TOT -no controls Treatment Status 0.025+ 0.028* (0.014) (0.013) Model C: ITT -controls Treatment Status 0.029* 0.031** (0.012) (0.012) Model D: TOT -controls Treatment Status 0.028* 0.030** (0.012) (0.011) Observations ITT Sub-Sample (Model A and C) 33316 33316 Mean Dependent Variable ITT Sub-Sample (All) 0.076 0.071 Mean Dependent Variable ITT Sub-Sample (Controls) 0.058 0.051 Observations TOT Sub-Sample (Models B and D) 31214 31214 Mean Dependent Variable TOT Sub-Sample (All) 0.075 0.070 Mean Dependent Variable TOT Sub-Sample (Controls) 0.059 0.051 Notes: +significant at 10%, *significant at 5%, **significant at 1%. SE in parentheses. Marginal effects and robust standard errors reported, clustered at the community level. All regressions include wave dummies. Controls in models C and D include head's and spouse's age, age squared and education level dummies, head's ethnicity (language), household size, dummies controlling for household demographics, baseline assets (home ownership, dirt floor and electricity) and community characteristics (community organizations, distance to urban center and wages). 38 Table 7: Effect of OPORTUNIDADES on Agricultural Investments by Quintiles of Cumulative Transfer (Potential and Actual) Sub-sample of Original Poor Households from October 1998 through November 2003 -Treatment on the Treated (TOT) Number Mean Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 of Obs. Dep. Var. I. DRAFT ANIMALS OWNERSHIP (PROBIT) Model A TOT -Controls, Potential Transfer 0.023 0.035+ 0.041+ 0.057* 0.066+ 51327 0.286 Treatment Status * Quintile (0.015) (0.020) (0.022) (0.028) (0.034) Model B: TOT -Controls, Actual Transfer Treatment Status * Quintile 0.014 0.044* 0.045* 0.055* 0.076* 51327 0.286 (0.015) (0.020) (0.023) (0.027) (0.034) Model C: TOT -Controls, Actual Transfer p(y|y 97 =0) Treatment Status * Quintile 0.008 0.019 0.026 0.036+ 0.058* 33649 0.136 (0.010) (0.014) (0.017) (0.020) (0.027) Model D: TOT by Farm Size in 97 -Controls, Actual Transfer p(y|y 97 =0) Treatment Status * Quintile * No Agricultural Assets -0.008 -0.011 -0.005 0.019 0.053 33649 0.136 (0.016) (0.021) (0.028) (0.029) (0.035) Treatment Status * Quintile * Landless 0.020 0.031+ 0.051* 0.055* 0.062+ (0.014) (0.019) (0.024) (0.025) (0.033) Treatment Status * Quintile * Smaller Farm 0.010 0.028 0.024 0.038+ 0.064* (0.014) (0.018) (0.020) (0.023) (0.031) Treatment Status * Quintile * Bigger Farm -0.002 -0.008 0.000 0.005 0.050 (0.018) (0.024) (0.026) (0.027) (0.035) II. PRODUCTION ANIMALS OWNERSHIP (PROBIT) Model A: TOT -Controls, Potential Transfer Treatment Status * Quintile -0.015 0.015 0.030+ 0.057** 0.065** 51333 0.741 (0.013) (0.016) (0.017) (0.021) (0.023) Model B: TOT -Controls, Actual Transfer Treatment Status * Quintile -0.016 0.019 0.027 0.054** 0.079** 51333 0.741 (0.013) (0.016) (0.019) (0.021) (0.023) Model C: TOT -Controls, Actual Transfer p(y|y 97 =0) Treatment Status * Quintile 0.018 0.057* 0.062 0.111* 0.166** 8924 0.524 (0.024) (0.029) (0.038) (0.048) (0.057) Model D: TOT by Farm Size in 97 -Controls, Actual Transfer p(y|y 97 =0) Treatment Status * Quintile * No Agricultural Assets 0.037 0.098** 0.070+ 0.163** 0.197** 8924 0.524 (0.026) (0.034) (0.041) (0.045) (0.051) Treatment Status * Quintile * Landless -0.051 0.004 0.048 0.127 0.085 (0.077) (0.079) (0.104) (0.088) (0.121) Treatment Status * Quintile * Smaller Farm -0.011 0.002 0.022 0.021 0.131* (0.036) (0.041) (0.052) (0.062) (0.063) Treatment Status * Quintile * Bigger Farm -0.010 -0.059 0.119 0.071 0.139 (0.078) (0.082) (0.094) (0.097) (0.118) III. LAND USE (PROBIT) Model A: TOT -Controls, Potential Transfer Treatment Status * Quintile 0.006 0.023 0.024 0.033 0.040 51390 0.619 (0.020) (0.025) (0.027) (0.031) (0.035) Model B: TOT -Controls, Actual Transfer Treatment Status * Quintile 0.009 0.015 0.027 0.039 0.062+ 51390 0.619 (0.020) (0.025) (0.028) (0.030) (0.033) Model C: TOT -Controls, Actual Transfer p(y|y 97 =0) Treatment Status * Quintile 0.033 0.045 0.064+ 0.064+ 0.098* 20868 0.387 (0.021) (0.028) (0.033) (0.036) (0.044) Model D: TOT by Farm Size in 97 -Controls, Actual Transfer p(y|y 97 =0) Treatment Status * Quintile * No Agricultural Assets 0.049 0.058 0.093* 0.043 0.116* 20868 0.387 (0.032) (0.040) (0.046) (0.046) (0.054) Treatment Status * Quintile * Landless 0.031 0.041 0.056+ 0.065+ 0.090* (0.022) (0.028) (0.034) (0.036) (0.044) Treatment Status * Quintile * Smaller Farm - - - - - - - - - - Treatment Status * Quintile * Bigger Farm - - - - - - - - - - 39 Table 7 -continued Number Mean Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 of Obs. Dep. Var. IV. NUMBER OF EQUIVALENT DRAFT ANIMALS (LS) Model A: TOT -Controls, Potential Transfer Treatment Status * Quintile 0.008 0.002 0.003 0.003 0.034 51327 0.217 (0.015) (0.017) (0.019) (0.022) (0.028) Model B: TOT -Controls, Actual Transfer Treatment Status * Quintile 0.007 0.007 0.025 0.013 0.058* 51327 0.217 (0.015) (0.017) (0.020) (0.022) (0.027) Model C: TOT -Controls, Actual Transfer p(y|y 97 >0) Treatment Status * Quintile 0.019 0.019 0.040 0.048 0.137* 17638 0.467 (0.032) (0.035) (0.039) (0.043) (0.055) Model D: TOT by Farm Size in 97 -Controls, Actual Transfer p(y|y 97 >0) Treatment Status * Quintile * No Agricultural Assets - - - - - 17638 0.467 - - - - - Treatment Status * Quintile * Landless 0.002 0.002 0.000 -0.039 0.110 (0.059) (0.063) (0.068) (0.068) (0.075) Treatment Status * Quintile * Smaller Farm 0.012 0.013 0.042 0.061 0.110+ (0.041) (0.044) (0.048) (0.050) (0.060) Treatment Status * Quintile * Bigger Farm 0.040 0.037 0.085 0.093 0.175* (0.058) (0.062) (0.066) (0.065) (0.071) V. NUMBER OF EQUIVALENT PRODUCTION ANIMALS (LS) Model A: TOT -Controls, Potential Transfer Treatment Status * Quintile 0.017 0.063+ 0.072* 0.147** 0.164** 51333 0.740 (0.028) (0.033) (0.036) (0.043) (0.053) Model B: TOT -Controls, Actual Transfer Treatment Status * Quintile 0.029 0.032 0.104** 0.181** 0.193** 51333 0.740 (0.028) (0.033) (0.038) (0.042) (0.052) Model C: TOT -Controls, Actual Transfer p(y|y 97 >0) Treatment Status * Quintile 0.038 0.039 0.113** 0.187** 0.187** 42335 0.837 (0.032) (0.038) (0.043) (0.048) (0.060) Model D: TOT by Farm Size in 97 -Controls, Actual Transfer p(y|y 97 >0) Treatment Status * Quintile * No Agricultural Assets - - - - - 42335 0.837 - - - - - Treatment Status * Quintile * Landless 0.009 0.005 0.049 0.073 0.109 (0.047) (0.053) (0.059) (0.061) (0.070) Treatment Status * Quintile * Smaller Farm 0.049 0.025 0.127* 0.196** 0.185** (0.043) (0.048) (0.052) (0.056) (0.065) Treatment Status * Quintile * Bigger Farm 0.010 0.172* 0.237** 0.410** 0.344** (0.070) (0.076) (0.080) (0.078) (0.084) VI. NUMBER OF HECTARES (LS) Model A: TOT -Controls, Potential Transfer Treatment Status * Quintile 0.121* 0.155* 0.136* 0.205** 0.048 51103 1.481 (0.053) (0.062) (0.067) (0.079) (0.099) Model B: TOT -Controls, Actual Transfer Treatment Status * Quintile 0.135** 0.125* 0.140* 0.247** 0.114 51103 1.481 (0.052) (0.062) (0.070) (0.078) (0.097) Model C: TOT -Controls, Actual Transfer p(y|y 97 >0) Treatment Status * Quintile 0.154* 0.127 0.137 0.260** 0.093 30314 1.965 (0.070) (0.081) (0.091) (0.100) (0.125) Model D: TOT by Farm Size in 97 -Controls, Actual Transfer p(y|y 97 >0) Treatment Status * Quintile * No Agricultural Assets - - - - - 30314 1.965 - - - - - Treatment Status * Quintile * Landless - - - - - - - - - - Treatment Status * Quintile * Smaller Farm 0.124 0.123 0.140 0.224* -0.015 (0.076) (0.086) (0.095) (0.104) (0.126) Treatment Status * Quintile * Bigger Farm 0.223+ 0.094 0.113 0.262+ 0.323* (0.126) (0.137) (0.145) (0.144) (0.158) Notes: +significant at 10%, *significant at 5%, **significant at 1%. SE in parentheses. For the probit models, marginal effects and robust standard errors reported, clustered at the community level. LS regressions include community level random effects. All regressions include wave dummies and controls for head's and spouse's age, age squared and education level dummies, head's ethnicity (language), household size, dummies controlling for household demographics, baseline assets (home ownership, dirt floor and electricity) and community characteristics (community organizations, distance to urban center and wages).Data are for the five waves for which we have data on the dependent variables: October 1998, May 1999, November 1999, May 2000 and November 2003. Smaller farms use up to 3 Ha. of land; bigger farms use more than 3 Ha. 40 Table 8: Consumption and Transfer Amounts -Test of Equality of Means between Actual Treatments and Actual Controls. Sub-Sample of Original Poor from October 1998 to May 19991 -Treatment on the Treated (TOT) Treatment Group Control Group N Mean SD N Mean SD t-stat I. Entire Sub-Sample Monthly Consumption per Capita (Home Production Included) 12667 194.09 96.597 7710 171.91 93.004 4.900 Actual Monthly Transfer per Capita 12667 36.35 21.528 - - - - 6-month Lagged Actual Cumulated Transfer per Capita 12667 141.33 123.927 - - - - Potential Monthly Transfer per Capita 12667 48.61 26.256 - - - - 6-month Lagged Potential Cumulated Transfer per Capita 12667 216.64 187.569 - - - - II. No Agricultural Assets Households in 97 =1 Monthly Consumption per Capita (Home Production Included) 1187 215.02 106.803 785 190.92 102.052 3.203 III. Landless Households in 97 =1 Monthly Consumption per Capita (Home Production Included) 3864 199.84 95.667 2463 182.24 93.550 3.268 IV. Smaller Farms in 97 =1 Monthly Consumption per Capita (Home Production Included) 5986 187.69 94.076 3284 160.99 90.104 5.213 V. Bigger Farms in 97 =1 Monthly Consumption per Capita (Home Production Included) 1592 189.29 97.530 1154 167.67 88.754 3.046 Notes: T-stat of differences in means computed clustering SE at the community level. All consumption values are expressed in per capita adult equivalent units. Smaller farms are landed households using less than 3 Ha. of land; bigger farms use more than 3 Ha. 1Data on home production is only available for two rounds of data, namely October 1998 and May 1999. Table 9: Effect of Current Transfers (Marginal Propensity to Consume) and Lagged Cumulative Transfers on Consumption Sub-sample of Original Poor Households from October 1998 through November 20031-Actual Treatments (TOT) Consumption Consumption Consumption Consumption Linear I Linear II Linear III Linear IV (LS) (LS) (2SLS) (2SLS) Actual Transfer (monthly pc ae) -Preferences 0.649** 0.510** 0.733** 0.884** (0.020) (0.022) (0.034) (0.035) Actual Cumulative Transfer (6 months lag pc ae) -Productivity 0.011** 0.012** 0.018** (0.001) (0.001) (0.001) Full Set of Controls N N N Y Number of Observations 39529 39529 39529 39529 Mean of Dependent Variable (All) 186.598 186.598 186.598 186.598 Mean of Dependent Variable (Controls) 176.070 176.070 176.070 176.070 Notes: +significant at 10%, *significant at 5%, **significant at 1%. SE in parentheses. All regressions include random effects at the community level and wave dummies. The full list of controls additionally includes: head's and spouse's age, education level dummies, head's ethnicity (language), household size adult equivalent and household demographic variables, baseline assets (home ownership, dirt floor and electricity), farm size at baseline and community characteristics (community organizations, distance to urban center and wages). The excluded instruments in the 2SLS models are the monthly pc ae potential transfer and the 6-month lag cumulative potential pc ae transfer. The first stages show significant effects of the excluded instruments on the endogeneous explanatory variables and are available upon request. Consumption is expenditures expressed in per capita adult equivalent units and includes home production. Outliers trimmed at the top and bottom 1% of both the consumption and transfer distributions. pc means "per capita"; ae means "adult equivalent". 1Because of the inclusion of home production, we only have data on consumption for the following rounds: October 1998, May 1999, Nov 2000 and Nov 2003. 41 Table 10: Consumption -Heterogeneous Treatment Effects Sub-sample of Original Poor Households from October 1998 through November 20031-Actual Treatments (TOT) Consumption Consumption Consumption Consumption Consumption by Head's Consumption by Head and Linear IV by Farm Size by Head's Age Education by Distance Household (2SLS) (2SLS) (2SLS) (2SLS) (2SLS) Characteristics Actual Transfer (monthly pc ae) -Preferences Linear (Alone) 0.886** 0.868** 1.094** 1.073** 0.872** 1.277** (0.035) (0.038) (0.261) (0.055) (0.034) (0.288) Interacted with Non Agricultural Household at Baseline (1997) =1 0.181* 0.166+ (0.090) (0.091) Interacted with Bigger Farm at Baseline (1997) =1 0.020 0.0001 (0.078) (0.078) Interacted with Head's Age -0.017 -0.018+ (0.011) (0.011) Interacted with Head's Age Squared 0.0002* 0.0002* (0.0001) (0.0001) Interacted with Head has not completed primary =1 -0.251** -0.170* (0.064) (0.069) Interacted with Head has completed primary or more =1 -0.304** -0.194* (0.077) (0.090) Actual Cumulative Transfer (6 months lag pc ae) -Productivity Linear (Alone) 0.018** 0.018** 0.009* 0.016** 0.028** 0.015** (0.001) (0.001) (0.004) (0.001) (0.002) (0.005) Interacted with Non Agricultural Household at Baseline (1997) =1 -0.002 -0.003 (0.002) (0.002) Interacted with Bigger Farm at Baseline (1997) =1 0.002 0.002 (0.001) (0.001) Interacted with Head's Age 0.0004** 0.0004** (0.0001) (0.0001) Interacted with Head's Age Squared -5.35*10-6** -5.30*10-6** (0.0001) (0.0001) Interacted with Head has Not Completed Primary =1 0.003** 0.003* (0.001) (0.001) Interacted with Head has Completed Primary or More =1 0.001 0.001 (0.001) (0.002) Interacted with Min Distance to Urban Center (/100) -0.009** -0.009** (0.001) (0.001) Full Set of Controls Y Y Y Y Y Y Number of Observations 39529 39529 39529 39529 39529 39529 Mean of Dependent Variable (All) 186.598 186.598 186.598 186.598 186.598 186.598 Mean of Dependent Variable (Controls) 176.070 176.070 176.070 176.070 176.070 176.070 Notes: +significant at 10%, *significant at 5%, **significant at 1%. SE in parentheses. All regressions include random effects at the community level and the following controls: wave dummies, head's and spouse's age, education level dummies, head's ethnicity (language), household size adult equivalent and household demographic variables, baseline assets (home ownership, dirt floor and electricity), farm size at baseline and community characteristics (community organizations, distance to urban center and wages). Main effects of the controls added for those specifications with interactions. The excluded instruments are the monthly pc ae potential transfer, the 6-month lag cumulative potential pc ae transfer and its respectives interactions, depending on the specification. The first stages show significant effects of the excluded instruments on the endogeneous explanatory variables and are available upon request. Consumption is expenditures expressed in per capita adult equivalent units and includes home production. Outliers trimmed at the top and bottom 1% of both the consumption and transfer distributions. pc means "per capita"; ae means "adult equivalent". Landless and smaller farms are the omitted farm size categories; no education is the omitted education category. Because of the inclusion of home production, we only have data on consumption for the following rounds: October 1998, May 1999, Nov 2000 and Nov 2003. 1 42 Table 11: Are there Price Effects? Testing for Macro Income and Price Effects Sub-sample of Ineligibles (Non-Poor) Households in October 1998, May 1999 and November 1999 -Intent to Treat (ITT) Treatment Group Control Group N Mean SD N Mean SD t-stat A. Raw Means -no controls AI. Agricultural Investments Draft Animal Ownership 9053 37.23 0.483 5805 35.47 0.478 0.654 Number of Equivalent Draft Animals 3341 0.90 1.523 2047 0.87 1.252 0.320 Production Animal Ownership 9054 75.99 0.427 5805 75.23 0.432 0.447 Number of Production Animals 6858 1.98 3.334 4345 1.98 3.389 0.005 Land Ownership 9054 65.66 0.475 5806 64.74 0.478 0.423 Hectars of Land Used 5819 2.98 3.000 3703 3.19 3.005 -0.975 AII. Micro-Enterprise Investments Micro-Enterprise 9054 9.21 0.289 5806 8.01 0.271 1.068 Female Micro-Enterprise 9054 8.92 0.285 5806 7.34 0.261 1.477 AIII. Consumption1 Monthly Consumption per Capita (Home Production Included) 6155 261.74 197.939 3874 255.97 198.147 0.729 AIV. Community Wages Monthly Male Agricultural Wage 860 605.13 202.668 488 613.94 195.336 -0.520 Monthly Female Agricultural Wage 364 534.53 191.883 197 516.00 188.584 0.884 Monthly Children Agricultural Wage 318 437.42 183.389 160 421.03 162.437 0.811 Notes: T-stat of differences in means computed clustering SE at the community level. All consumption values are expressed in per capita adult equivalent units. Continuous outcomes conditional on being positive 1 Data on home production is only available for two rounds of data, namely October 1998 and May 1999. 43 Table 12: Effect of OPORTUNIDADES on Private Transfers (Panels AI, AII), Other Public Transfers (Panels BI, BII) and Activitities of Daily Living (Panel CI) Sub-sample of Original Poor Households -Actual Treatments (TOT) Potential Potential Cumulative Actual Actual Potential Transfer Transfer Actual Transfer Cumulative Treatment Transfer Squared (6 month Transfer Squared Transfer Number Mean Status (monthly) (monthly) lag) (monthly) (monthly) (6 month lag) of Obs. Dep. Var. AI. Private Transfers1 =1 Model AI.1 -PROBIT (Oct 98, Nov 99) -0.002 20847 0.070 (0.005) Model AI.2 -PROBIT (Oct 98, Nov 99, Nov 00) -0.019 31071 0.067 (0.014) Model AI.3 -PROBIT (Oct 98, Nov 99, Nov 00) -0.087* 0.148* 31071 0.067 (0.034) (0.059) Model AI.4 -2SLS (Oct 98, Nov 99, Nov 00) -0.223** 0.531** 31083 0.067 (0.067) (0.193) AII. Private Transfers Amount (monthly) Model AII.1 -LS (Oct 98, Nov 99) -54.798+ (29.780) Model AII.2 -LS (Oct 98, Nov 99, Nov 00) -0.089 1512 337.618 (0.096) Model AII.3 -LS (Oct 98, Nov 99, Nov 00) -0.587* 0.001* 1512 337.618 (0.234) (0.0004) Model AII.4 -2SLS (Oct 98, Nov 99, Nov 00) -1.235* 0.003* 1512 337.618 (0.498) (0.001) BI. PublicTransfers2 =1 Model BI.1 -PROBIT (Oct 98 - Nov 99) -0.115** 30659 0.369 (0.022) Model BI.2 -PROBIT (Oct 98 - Nov 00) -0.243** 50208 0.346 (0.047) Model BI.3 -PROBIT (Oct 98 - Nov 00) -0.537** 0.611** 50208 0.346 (0.126) (0.190) Model BI.4 -2SLS (Oct 98 - Nov 00) -0.611** 0.820** 50208 0.346 (0.084) (0.202) BII. Public Transfers Amount (monthly) Model BII.1 -LS (Oct 98 - Nov 99) -9.534** 8732 90.544 (3.472) Model BII.2 -LS (Oct 98 - Nov 00) -0.002 13881 91.991 (0.006) Model BII.3 -LS (Oct 98 - Nov 00) -0.036* 6.55*10-5* 13881 91.991 (0.017) (2.99*10-5) Model BII.4 -2SLS (Oct 98 - Nov 00) -0.073* 1.69*10-4* 13881 91.991 (0.032) (7.74*10-5) CI. Activity of Dalily Living3 (ADL) Model CI.1 -LS (May 99 - Nov 99) 0.003+ 40491 0.978 (0.002) Model CI.2 -LS (May 99, Nov 99, Nov 00) 0.073** 59907 0.979 (0.019) Model CI.3 -LS (May 99, Nov 99, Nov 00) 0.065** 0.001 59907 0.979 (0.021) (0.001) Model CI.4 -2SLS (May 99, Nov 99, Nov 00) 0.107** 59907 0.979 (0.038) Model CI.4 -2SLS (May 99, Nov 99, Nov 00) 0.112* -0.0001 59907 0.979 (0.048) (0.003) Notes: +significant at 10%, *significant at 5%, **significant at 1%. For the probit models, marginal effects and robust standard errors clustered at the community level reported. LS and 2SLS regressions include community level random effects. All regressions include the following controls: wave dummies, head's and spouse's age, age squared and education level dummies, head's ethnicity (language), household size adult equivalent, dummies controlling for household demographics, baseline assets (home ownership, dirt floor and electricity), farm size at baseline and community characteristics (community organizations, distance to urban center and wages). Outliers trimmed at the top and bottom 0.5% of the OPORTUNDADES (public) and the private transfers distributions. For other public transfers, outliers trimmed at the top 5% of the distribution. Conditional on being positive. 1Private transfers include both transfers from neighbours, relatives and friends living in the community and transfers received from migrants (conditional on a household member having migrated during the 5 years before the interview). Estimation limited to rounds: Oct 1998, Nov 1999 and Nov 2000 due to non-existent or non-comparable data. 2Public transfers =1 if someone in the household receives help from Procrampo, Niños con Solidaridad and/or DIF food package. Public transfer amount includes the total household monetary value of the transfers received from Procampo and Niños con Solidaridad. Estimation limited to rounds Oct 1998 to Nov 00. 3Estimation limited to the sub-sample of prime age adults (18 to 49 years old) in May 1999, Nov 1999 and Nov 2000 due to non-existent data in other rounds. OPORTUNIDADES actual and potential transfer amounts are expressed in per capita adult equivalent units for this set of regressions. In Panels AI, BI and CI the OPORTUNIDADES potential and actual transfers (and its squared term) have been divided by 1000. 44 Table 13: Potential Biases in the Consumption Equation: Private Transfers, Public Transfers and Increased Productivity. Sub-sample of Original Poor Households in October 1998 through November 2003 -Actual Treatments (TOT) Consumption Consumption Consumption (2SLS) (2SLS) (2SLS) A. Private Transfers 1 -controls Actual Transfer - monthly pc ae 1.055** (0.049) Total Actual Transfers (OPORTUNIDADES + Private) -monthly pc ae 1.035** (0.048) Total Actual Transfers (OPORTUNIDADES + Other Public + Private) -monthly pc ae 1.013** (0.050) Actual Cumulative Transfer -6 months lag pc ae 0.016** 0.015** 0.010** (0.002) (0.002) (0.002) Number of Observations 20282 20282 20267 Mean Consumption 181.476 181.476 181.985 B. Public Transfers 2 -controls Actual Transfer - monthly pc ae 0.913** (0.035) Total Actual Transfers (OPORTUNIDADES + Other Public) -monthly pc ae 0.961** 0.966** (0.038) (0.038) Actual Cumulative Transfer -6 months lag pc ae 0.020** 0.018** 0.018** (0.002) (0.002) (0.002) Receive Food Package DIF =1 7.454** (2.302) Number of Observations 30126 30126 30126 Mean Consumption 181.916 181.916 181.916 C. Increased Productivity: Activity Daily Living (ADL)3 -controls Actual Transfer - monthly pc ae 0.685** 0.685** - (0.046) (0.046) - Actual Cumulative Transfer -6 months lag pc ae 0.015** 0.015** - (0.003) (0.003) - Activity of Daily Living (ADL) Household's Head -1.500 - (7.706) - Number of Observations 13179 13179 - Mean Consumption 175.236 175.236 - Notes: +significant at 10%, *significant at 5%, **significant at 1%. SE in parentheses. All regressions include random effects at the community level and the following controls: wave dummies, head's and spouse's age, age squared and education level dummies, head's ethnicity (language), household size adult equivalent and household demographic variables, baseline assets (home ownership, dirt floor and electricity), farm size at baseline and community characteristics (community organizations, distance to urban center and wages). Consumption is expenditures expressed in per capita adult equivalent units and includes home production. Outliers trimmed at the top and bottom1% of the consumption and the total transfer distributions. pc means "per capita"; ae means "adult equivalent". 1Because of the inclusion of home production and private transfers, estimation is limited to the following rounds: October 1998 and November 2000. Private transfers include both transfers from neighbours, relatives and friends living in the community and transfers received from migrants (conditional on a household member having migrated during the 5 years before the interview). 2Because of the inclusion of home production, estimation is limited to the rounds of October 1998, May 1999 and November 2000. Public transfers include the total household monetary value of the transfers received from Procampo and Niños con Solidaridad. DIF =1 if someone in the househod has received a subsidized food package during the past month. 3Because of the inclusion of home production and ADLs, estimation is limited to the following rounds: May 1999 and November 2000. 45 Table 14: Investment or Precautionary Savings? Sub-sample of Original Poor Households in November 2003 -Actual Treatments (TOT) Number of Number of Equivalent Equivalent Number of Draft Draft Production Production Hectares Animal Animals Animal Animals Used Ownership p(y|y >0) Ownership p(y|y >0) Land Use p(y|y >0) (2SLS) (2SLS) (2SLS) (2SLS) (2SLS) (2SLS) Model A: TOT -controls High Risk Area 0.020 0.086 0.117** 0.097 0.147** -1.545 (0.049) (0.300) (0.036) (0.221) (0.048) (0.970) Total Household Actual Cumulative Transfer (/1000) 0.004* -0.004 0.004* 0.002 0.002 -0.008 (0.002) (0.013) (0.002) (0.008) (0.002) (0.027) High Risk Area * Total HH Actual Cumulative Transfer (/1000) 0.000 -0.020 -0.006** 0.001 -0.008** 0.081 (0.003) (0.016) (0.002) (0.014) (0.003) (0.056) Model B: TOT -controls High Risk Area 0.009 -0.806+ 0.094 -0.288 0.037 -0.509 (0.101) (0.421) (0.078) (0.398) (0.098) (1.870) Total Household Actual Cumulative Transfer (/1000) -0.001 -0.073 0.003 -0.057* 0.006 -0.197+ (0.006) (0.046) (0.005) (0.028) (0.006) (0.113) Total Household Actual Cumulative Transfer Squared (/1000) 0.000 0.002 0.000 0.002* 0.012 0.006+ (0.000) (0.001) (0.000) (0.001) (0.015) (0.003) High Risk Area * Total HH Actual Cumulative Transfer (/1000) 0.002 0.132+ -0.002 0.068 -0.000 -0.099 (0.018) (0.071) (0.012) (0.072) (0.000) (0.324) High Risk Area * Total HH Actual Cumulative Transfer Squared (/1000) -0.000 -0.005* -0.000 -0.002 -0.001 0.006 (0.001) (0.002) (0.000) (0.002) (0.000) (0.012) Observations TOT Sub-Sample 9953 2968 9957 7366 10009 6342 Mean Dependent Variable TOT Sub-Sample 0.303 1.147 0.746 1.066 0.681 3.694 Proportion Households in High Risk Areas 0.093 0.104 0.093 0.097 0.093 0.094 Notes: +significant at 10%, *significant at 5%, **significant at 1%. Community level clustered SE in parantheses. All regressions include the following controls: wave dummies, head's and spouse's age, age squared and education level dummies, head's ethnicity (language), household size, dummies controlling for household demographics, baseline assets (home ownership, dirt floor and electricity) and community characteristics (community organizations, distance to urban center and wages). High risk areas are communities that have suffered more than 8 weather shocks between early 1997 and late 2000. The number of shocks per community in these years varies from 0 to 13, with a median of 5 shocks and an average of 5.41 shocks. 46 APPENDIX 2 -GRAPHS Graph 1: OPORTUNIDADES Program Take up Rates Over Time 100 90 80 70 pu e 60 Takfo 50 ge tanecreP 40 30 20 10 0 199 199 199 199 19 8.04 8.06 8.08 8.10 98.11999.01999.01999.06 2 2 4 1999.08 199 199 200 200 200 200 200 20 9.10 9.12 0.02 0.04 0.06 0.08 0.10 00.12001.02001.04 2 2 2001.06 200 200 200 200 200 200 200 20 1.08 1.10 1.12 2.02 2.04 2.06 2.08 02.12002.12003.02 3.04 0 2 200 Bimester Original Poor -Treatment Original Poor -Control Graph 2: Share of Households by Number of Children Enrolled from 3rd Grade of Primary to 3rd Grade of Junior High School at Baseline (October 1997). 0.45 0.4 0.35 0.3 sd olhe 0.25 usoH Actual Treatments Actual Controls of 0.2 erahS 0.15 0.1 0.05 0 None 1 P 1 JH 2 P 1 P & 1 2 JH 3 P 2 P & 1 P & 2 3 JH More 4 JH 1JH JH Type of Household Note: P =Primary School; JH =Junior High School 47 Graph 3a: Actual and Potential Transfer Distributions Sub-sample of Actual Treatment and Actual Controls 00 eva 20 W erp 00 15 ntuo mArefs 00 10 anrT 0 andie 50 M 0 Oct97 Oct98 May99Nov99May00Nov00May01Nov01May02Nov02May03Nov03 Wave Actual Potential Graph 3b: Cumulative Actual and Potential Transfer Distributions 0 Sub-sample of Actual Treatment and Actual Controls 00 180 00 resfn 150 00 raT 12 veital 00 90 muuC 00 60 00 30 0 Oct97 Oct98 May99Nov99May00Nov00May01Nov01May02Nov02May03Nov03 Wave Actual Potential 48 Graph 4: Mean Actual and Potential Cumulative Transfer Amount by Quintile 20000 18000 16000 14000 nt 12000 Amoure Actual 10000 ansfrT Potential ane 8000 M 6000 4000 2000 0 Q1 Q2 Q3 Q4 Q5 Quintile 49