20
the two groups. Unfortunately, we do not have sufficient information on credit and loan application
history from the surveys to do this.
Consequently, we take two alternative approaches. We first calculate the preprogram value of
each household’s liquid assets (S). We assume that liquid assets include the value of livestock assets,
fixed productive assets and consumable durable goods, plus loans and deposits. We then divide the
sample households into quartiles based on the value of their total liquid asset: Qj, j = [1,2,3,4] where j
= 1 is the group of households with the lowest asset value. We then test whether the program effects
differ among the quartiles using the DID framework. Heterogeneity in treatment effects can be studied
by including interactions between Qj and the treatment dummy variable. Thus, we estimate the
following equation:
. (2)
If a household’s liquidity constraint is indeed being relaxed by participation in the Grain for Green
program, there will be a positive impact by the program on participation in the off-farm labor market (or
on earnings from agriculture). In the empirical model, we anticipate that households that had a lower
level of liquidity before Grain for Green (those households belonging to the lower two quartiles) will
see a greater relaxing of their liquidity constraint when they receive their compensation than households
that had owned a set of liquid assets with a higher value (or those from the top two quartiles).
As a second alternative approach, we utilize a rule developed by Zeldes (1989) to split the
households into liquidity-constrained and -unconstrained groups. Specifically, Zeldes classifies
households into the liquidity-constrained group if their estimated non-housing wealth was less than two