Thus, percentage of total gross income from the
major enterprise is utilized as a portfolio effects
variable.
For age and portfolio effects to have the hypoth-
esized relationship on capital structure, their effect
on the responsiveness of Kd function to leverage
must be contrasted to that on Ke function. This
proposition has largely been ignored in the literature.
However, a case can logically be built that these
variables do not have as large an effect on lenders as
on farm owners. The effect of age on Ke arises from
personal preferences not from productive charac-
teristics of the farm. Therefore, age would not be
expected to be perceived by lenders as increasing risk.
For portfolio effects, farm production does not
become more risky. However, lenders are concerned
with security values as much as, or more than with
income, and the risk associated with security values
on specialized farms relative to diversified farms
would be expected to increase as much as the relative
risk of income. Thus, effects of age and specialization
on Kd would not be expected to be as large as on
Ke ∙
In summary, variables in the empirical analysis
include cost of equity, age and percentage of gross
income from major enterprise. The first variable
measures differences between Ke and Kd functions
and the second and third differences in farmers’ risk
preferences.
DATA AND METHODOLOGY
A sample of 121 Georgia farmers was used in this
study. These farm operators were selected through a
stratified random sample and thus represented a
cross-sectional sample of the State’s farmers. Informa-
tion on sales, operating expenses, net taxable farm
income, income tax payments, value of assets, debt,
interest rate paid and family and operator labor use
was obtained from interviews and farm tax records
for 1972. Secondary data on wage rates and land
values supplemented the primary data; the wage data
was the Georgia average rate for hired farm labor for
1972 while the rate of land value appreciation varied
by crop reporting districts. Since optimal capital
structure is a long-run concept, time series data may
be preferable to using data for a single year. Data
from 1972 would, however, be more appropriate than
some other years since that year was representative of
the average recent farming situation.
Statistical analysis of the empirical model in-
volved derivation of a classification function for debt
and no-debt groups through discriminant analysis,
with the three variables discussed previously serving
as discriminating variables. Since farmers with only a
small amount of debt were likely to include cases
which were temporary debtors rather than having a
long-run goal of including debt in their financial
structure, farmers with D/А < .05 were classified in
the no-debt group prior to the statistical analysis.
Besides statistical results of the discriminant
analysis, a breakeven value for each independent
variable was calculated using sample means. This
value was calculated by equating the discriminant
function to zero:
2 _
a+ Σ biXi + bkXk = 0 (4)
ι=l
where
a = constant in the discriminant func-
tion
bi and bk = coefficients in discriminant function
Xi = sample mean for the two variables
and
Xk = variable (i≠k) for which a break-even
value is being calculated.
Solving equation (4) for each variable in turn iden-
tifies a break-even value of the independent variable
in reference to classification into debt and no-debt
groups. The break-even value for each independent
variable is the value for which the average farm
operator would have equal probability of being
classified in the debt and no-debt groups. For higher
values than the break-even value for each respective
value, the farmer would be more likely to be
classified in one of the groups, and for lower values
he would be more likely to be classified in the other
group.
EMPIRICAL RESULTS
Results of the empirical analysis are presented in
Tables 1 and 2. In Table 1, the discriminant equation
for classification from debt into the no-debt group is
presented along with F values for the coefficients. In
Table 2, means of the classification variables in the
sample and in both debt and no-debt subsamples are
presented along with break-even values calculated
with the discriminant function of Table 1 and
equation (4).
Results of the discriminant analysis were very
satisfactory. All variables were significant, with Ke
and age being significant at the one percent level.
Furthermore, 96 out of the 121 sample cases were
classified in the correct group. More importantly, all
coefficient values had the expected influence in the
discriminant analysis. Since the coefficient for Ke is
positive, an increase in Ke would increase the
166