model will produce unbiased estimates of β, but those estimates need a high variability on the sample.
4.3.2 Fixed Effect Model
Fixed effect model is model which considers eliminated variables can change intercept of cross section and time series. Dummy variables can be added
to the model to make intercept changes possible. Afterward model is estimated by using Ordinary Least Square OLS
Y
it
= α
i
D
i
+ β X
it
+ ε
it
Where: Y
it
= endogenous variable X
it
= exogenous variable α
= intercept D
= dummy variable β
= slope i
= individual i t
= period year t ε
= error
4.3.3 Random Effect Model
Additional of dummy in fixed effect can reduce quantity of degree of freedom. This condition will also reduce efficiency of estimated parameter.
Random effect model can be used to solve this problem. In this model, different parameter between individual and time is included to error. Random effect model
can be explained by this equation:
Y
it
= X
it
β
j
+ ε
it
ε
it
= u
it
+v
it
+ w
it
Where: u
it
~ N0,δu
2
= component of cross section error v
it
~ N0,δv
2
= component of time series error w
it
~ N0,δv
2
= component of combination error It can also be assumed that individual error and combination error is not
correlated each other. Using of random effect model can reduce using of degree of freedom. It has implication that estimated parameter will be more efficient.
Nachrowi and Usman 2006 suggested that it is better to use fixed effects model if we have T time bigger than amount of individual. On contrary, if we
have amount of individual is bigger than amount of time, so it would be better if we use random effects model. Egger 2000 explained that since individual effects
are include in the regressions a decision should be made whether they are treated as random or fixed. A random effects model can be more appropriate when
estimating the flows of trade between a randomly drawn sample of trading partners from a large population. A fixed effects model would be a better model
when estimating flows of trade between an ex ante predetermined selection of countries.
This study deals with the flows of trade between Indonesia and Countries in European Union which is main importer of Indonesia cocoa. Those are
Germany, France, Netherlands, United Kingdom, Belgium, Italy, Spain, Austria, Hungary, Poland and Czech Republic. Therefore the fixed effect will be a more
appropriate model than random specification. The eleven importer countries are selected for the period 1998 - 2011.
4.4 Model Formulation
There are two codes of cocoa which will be analyzed in this paper. Those are HS 1801 and HS 1804. Determination of these codes based on the two highest
cocoa export of Indonesia.
4.4.1 Model Formulation of HS 1801 Cocoa Beans, Whole or Broken, Raw
or Roasted
Analysis used in this research is Gravity Model approach which consists of Dependent variables and some Independent variables. Independent variables
used are GDP of exporter and importer countries, population of exporter and importer countries, physical distance, exchange rate and export tax.
We will divide Analysis of code HS 1801 Cocoa beans, whole or broken, raw or roasted into two analyses. Firstly, export tax is treated as dummy variable
and secondly, export tax is analyzed as percentage value. It is intended to know the effect of the export tax to European Union as whole, before and after export
tax policy and also the effect of export tax in percentage value to trade flows export value. The model formulation could be written as follows:
ln Y
ijt
= β + β
1
lnG
it
+ β
2
lnG
jt
+ β
3
lnS
it
+ β
4
lnS
jt
+ β
5
ln E
ijt
+ β
6
lnL
ij
+ β
7
T
t
+ ε where:
β =
Intercept β
1
, β
2
, β
5
= Parameter of each variable which will be tested statistically and
econometrically t
= 1,…,T between 1998 – 2011