at 20 percent confidence level. This is because there is no much import of CPO currently witnessed in Indonesia from the recent past.
Table 30. Regression Results for Crude Palm Oil Stock
Variable Parameter Estimate t-value
p-value Elasticity
INTERCEP 79.383574
0.639 0.5329 INPOQ
0.890147 21.395 0.0001
3.072841 INPOC
-0.807115 -3.966 0.0014
0.731651 LINPOS
0.703265 9.695 0.0001
1.23549 INPOM
5.448971 1.243 0.2342
0.024206 INPOX
-0.846218 -17.067 0.0001
2.143405 R
2
= 0.9810; F-value = 197.361; ProbF = 0.0001; DW = 1.685 Note: indicates significance at 5 percent level.
From all the variables that determine CPO stock, export and consumption variables were found to be negatively related to the CPO stock while the others
were positively related to it as can be observed in Table 30. It was found that CPO stock had elastic relationships with CPO production, previous CPO stock and
export while the rest had inelastic relationships with it. These relationships were reported in the short run basis. The above exogenous variables explained about
98.1 percent variables that determine CPO stock in Indonesia, meaning it had only about 2 percent of residual error included in its estimates.
7.2. Model Validation
After estimating all the equations, the model was solved simultaneously in a simulation program using SAS Statistical Analysis System v6.12. Historical
simulation of the model’s equations was used to validate the estimated model using the components of the Mean Squared Error MSE and the Theil inequality
coefficients. Table 31 presents the decomposition of the MSE and Theil U coefficient. The decomposition of MSE provides two sets of statistics. The first
decomposition suggested by Theil gives bias UM, variance US, and covariance UC statistics. The second decomposition, as suggested by Maddala, consists of
bias UM, regression UR, and disturbance UD components. An adequate model produces projections in which UM approaches zero,
i.e. the model is without consistent bias; US approaches zero, implying variability of the predicted series closely resembles the variability of the actual series; and
the random deviation UC is a large number. In the second decomposition, the bias and regression components capture the systematic divergence of the
prediction from actual values. Therefore, for a model that fits the data well, the proportion of UM and UR should approach zero. The UD component, which
captures the random divergence of the prediction from the actual values, should approach one. The Theil U coefficient should approach zero when the predicted
series is close to the actual series.
Table 31. Model Validation Statistics
Variable RME Bias
UM Reg
UR Dist
UD Var
US Covar
UC U
INPOA 20.4975
0.000 0.515 0.485 0.133
0.867 0.0909 INPOY
1.8588 0.000 0.000 1.000
0.032 0.968 0.0092
INPOQ 20.7416
0.000 0.395 0.605 0.094
0.906 0.0882 INPOS
16.4653 0.000 0.243 0.757
0.001 0.999 0.0868
INPOC 1.6342
0.000 0.000 1.000 0.011
0.989 0.0081 INPOX
28.5781 0.000 0.302 0.697
0.032 0.967 0.1103
INPOP 3.5065
0.000 0.003 0.997 0.038
0.962 0.0174 The MSE and its decomposition reported in Table 31 show that the
majority of the UM values are close to zero. This suggests that those simulated values are close to their actual values. Consequently, disturbance terms are low,
which implies that errors of these simulated variables are not captured by the randomness contained in the actual data series. Contrary to UM and UD, some of
the UR values are close to zero. In the second decomposition, US component performs well; however, UC values in some instances are fairly low. Compared to
the decomposition of MSE statistic, almost all the Theil’s U-Statistic are close to zero for the endogenous variables for the model. This suggests that overall the
simulation model has reasonably good forecasting ability.
7.3. Evaluation of the Impacts of Crude Palm Oil Export Tax
