Executive summary

Annex 3.3: Ex Ante Method to Evaluate the Program: Red de Oportunidades

This appendix is largely based on Bourguignon, Ferreira and Leite (2003).
We propose the use of the methodology presented in Bourguignon, Ferreira and Leite (2003), henceforth BFL model, to simulate the effects of transfers using the actual program design, and transfers of different magnitudes on poverty. Specifically, this model has been evaluated using Mexican, Ecuadorian and Brazilian data producing credibly estimated when predictions were compared with those estimated in the standard Ex Post evaluation

The Model

This model consists of simulating the effects of the Red de Oportunidades program on the basis of a model of the household behavior using national representative dataset.
Denote j = 0 the occupational category “not attending school”; j =1 “attending school and working”, and j = 2 denoting “attending school only”. In this case, the utility function for each category j of the child i is specified by:

where stands for characteristics of both the child and the household; is the household income without the child’s earnings; is the child income earned in alternative j; and is the random term that stands for idiosyncratic preferences.

Then, child earnings, , is defined in alternative j = 0 as the observed market earnings of the child, . In alternative j = 1, children can work and study, spending less time in the labor market than children from category j = 0. In this case, for j = 1, children can only receive a proportion M of the total time of category j = 0. The observed market earnings of a child in j = 1 is, then, on average equal to . Similar specification can be made to children in j = 2 because if he is not working in the labor market it doesn’t prevent him from contributing to domestic production. Then, considering D a proportion of time devoted to domestic production, can be set as .
Replacing these values on equation 1 we have the following:

Where and .

Assuming that we now all parameters of equation 2, the child select its occupation by maximizing its utility function, i.e., . If a CCT is implemented, the amount T is added to household income when child i is studying.
Hence, the equation 2 can be re-written by:

with and .

The Equation 2 must be estimated by using a Multinomial Logistic Model. However, is unobservable for those out of the labor market . Besides that, the value of M or D are also unobservable. To correct that, the BFL model uses a simple approach, which has an advantage of transparence and robustness (OLS regression) to estimate the potential wage, , for all children in order to avoid any type of complexity that would be generated by any correction of potential selection bias.⁶⁸

Finally the simulation of the likely effects of a given CCT program is represented by equation 4, here below, by taking into account the means-test that identify the potential beneficiaries. So taking everything in account, conditionality and means-test, the simulation of the new occupational choice of a child is given by:




This framework can simulate a wide variety of CCT programs conditional to schooling enrollment. Both the means-test and the transfer T could be made dependent of individual or household characteristics (different transfer per age and / or per gender of the child). It is important to have in mind that this model ignores multi-children interaction for the enrollment rate simulation, assuming that households were single-child for the behavioral point of view.

Besides that, the household income is treated as exogenous but this hypothesis can be very unrealistic. It is possible that the means-test may affect adult labor supply if they consider that is more interest to qualify for a CCT program instead of work. However, when the means-test is based on score-based proxy for permanent income this problem is not a real weakness of the model.

As final step, the Ex Ante estimator generated by the BFL model compares the simulated school enrollment and child labor participation of the children with their status-quo. Besides, by applying household ceiling to the transfers, the model allows the estimation of poverty and inequality index based on the “new” income or consumption obtained by the addition of the transfer amount for each beneficiary household in the sample, respecting the household ceiling transfer when necessary.