Agriculture Reference
In-Depth Information
variables including the constant;
is a
crop-specific error term. The control variables—which include measures of plot size, family
relationships, and farmer and land characteristics—are similar but not identical across the
data sets (see appendix A). We use a logit model to generate maximum likelihood estimates
of the model given by equations (6.9) and (6.10) for fifteen crop-specific contract samples
(j
=
j
is a column vector of unknown coefficients; and
e
ij
, nine on the Great Plains and six in Louisiana.
Table 6.5 presents the logit coefficient estimates from forty-eight separate estimated
equations for the Great Plains (36 equations) and Louisiana (12 equations) data for the two
exogenous variability measures at both the county and regional level. Each entry in table
6.3 is an estimated CV or STD coefficient—that is, an estimate of
15
)
and its associated t-
statistic, derived from a separate estimated equation.
28
For example, the entry in the upper
left cell
j
is the estimated coefficient for REGIONAL CV from the equation using a
sample of dryland corn contracts (
(
−
12.59
)
n
j
= 539) in Nebraska and South Dakota. The remaining
entries in the first column use the same contract sample but replace REGIONAL CV with
the other three measures of
2
j
. The remaining columns represent the same exercise using
contract samples for other crops. For the Nebraska-South Dakota data, we estimate these
equations with nine crop samples using four measures of
σ
2
j
(REGIONAL CV, REGIONAL
STD, COUNTY CV, and COUNTY STD). This results in thirty-six coefficient estimates
presented in the top half of table 6.5. For Louisiana the number of estimated equations and
coefficient estimates for
σ
is twelve because we have data on only six crops and because
the state of Louisiana collects data only for parishes, eliminating the use of REGIONAL
CV and REGIONAL STD.
29
Prediction 6.5 from the risk-sharing model predicts a
positive
coefficient
for
j
2
j
in a region, the more likely the
land contract will be a cropshare—that is, the model implies
σ
—the more variable is the yield for crop
j
crops.
Overall, the estimates fail to support the risk-sharing model. The estimates consistently
show that increases in exogenous crop yield variability do not increase the probability of
cropsharing. In forty-eight estimated equations there is not a single significant and positive
coefficient estimate of
j
>
0 for all
j
. In fact, more than one-half of the coefficient estimates are
negative. Moreover, eleven of these negative estimates are statistically significant, showing
that increases in exogenous risk actually reduces the probability of share contracting. We
also estimate equations (6.9) and (6.10) without control variables (using only CV or STD),
and with a smaller set of control variables than used in table 6.5. Neither of these alternative
specifications change the findings reported in table 6.5 although the specification reported
in table 6.3 consistently gave more precise estimates.
j
The Choice of the Farmer's Cropshare.
In this section we restrict our analysis to the
set of cropshare contracts in order to test prediction 6.3, which states that higher variability
crops result in a lower share to the farmer. For this exercise we use the farmer's contracted
