calculated. For the error terms, the expected error is zero, with the bounds being three standard
deviations of the dependent variable, total crop machinery costs (tcmc), above and below zero.
Based on Chebychev’s inequality (Golan, Judge and Miller) three standard deviations around the
expected value will capture 89% of the observations.
The prior and bounds for the scale coefficients were estimated by regressing the
respective functional form of harvested acres from models 1-4 on ratio R defined below
(equation 7). The upper and lower bound for the estimated scale coefficients was determined to
be three standard errors above and below the estimated coefficients.
Once the empirical models are estimated, they will be tested out-of-sample using a
jackknife procedure, deleting approximately 10% of the observations, re-estimating the model,
and then predicting the deleted observations. This process is repeated until all observations have
been deleted and the model re-estimated to provide an out-of-sample prediction for all
observations (Maddala). The model that predicts the best in an out-of-sample framework is
preferred because when producers not included in this dataset apply this research to estimate per
acre machinery costs for their own operations, they will be predicting out-of-sample. All four
models will be estimated both in-sample and out-of-sample to determine the best model. The
correlation coefficient (CC) (between predicted and actual values) and root mean square error
(RMSE) will be used to measure the relative predictive accuracy of the models.
Data
This research combines the number of field operations with farm financial data from a
sample of Kansas Farm Management Association (KFMA) members. The field operations were
obtained from a survey of the individual farms for field operations performed in the year 2001.
There were 182 farms in this research, with an average, minimum, and maximum harvested acres