θ
1
x
1t−1
+
θ
2
x
2t−1
+
θ
3
x
3t−1
estimate the lagged
(and/or differenced) parameters that, when combined, create an unrestricted error correction term
(
Philips 2018
). This combination of estimating the parameters in levels and lags allows for cointegrated
relationships and mixed orders of integration between the parameters.
∆y
t
=
β
0
+
∑
β
i
∆y
t−i
+
∑
β
j
∆x
1t−j
+
∑
β
k
∆x
2t−k
+
θ
0
y
t−1
+
θ
1
x
1t−1
+
θ
2
x
2t−1
+
θ
3
,
(1)
This combination also makes the ARDL model robust in spite of different data structures or
orders of integration (i.e., some variables that are I(0) and others that are I(1)), possibly cointegrated
relationships, separate lag structures for each variable, and small sample sizes (usually less than
100) (
Pesaran and Shin 1997
;
Pesaran and Smith 1998
;
Pesaran et al. 2001
). Such flexibility in the
model makes the ARDL approach suitable for exploratory studies like this one. Another advantage of
the ARDL approach is that it is able to differentiate between short-term relationships (dynamics
and fluctuations over short time periods) and long-term relationships (permanent relationships
over long periods of time) between armed conflict and antiquities looting. The bounds testing
methodology developed by
Pesaran and Shin
(
1997
) and
Pesaran et al.
(
2001
) was designed to work
with mixed orders of integration to determine whether a long-term relationship and cointegration is
present between two variables. The combination of ARDL models and a bounds testing approach
to cointegration addresses potential issues that can arise from data that have different orders of
integration (
Philips 2018
). The ability to differentiate between short and long-term relationships and
the flexibility of this approach to different data structures makes it an appropriate multiple time series
method for analyzing the relationship between antiquities looting and armed conflict.
Initial tests of the data revealed that antiquities looting and regime changes were stationary
(i.e., both variables were I(0)), but that armed conflict was not (i.e., armed conflict was I(1)). It was
also not clear whether or not there were cointegrating relationships among the variables. As such,
to analyze the relationship between antiquities looting and armed conflict, I had to use a method that
could (1) determine whether any cointegrating relationships existed; (2) accommodate mixed orders
of integration between the variables of interest; and (3) allow for cointegration in addition to mixed
orders of integration, if necessary. At both the month and quarter levels of analysis, the data have
small sample sizes (less than 100), which affected the ability of traditional tests to detect cointegration.
Given the complexity of the data, the current study used an ARDL model with the bounds testing
approach to analyze the three hypotheses at both the month and quarter levels. It is important to note
that running the ARDL model in statistical software (e.g., EViews) requires that either armed conflict
or antiquities looting be specified as the dependent variable. As such, the ARDL model was termed
with each as the dependent variable. The procedure used for each model is as follows:
1.
Check the stationarity of the variables and determine their order of integration (m).
2.
Use the ARDL model to analyze the relationship with armed conflict as the dependent variable.
a. Formulate the unrestricted error correction version of the ARDL model in Equation (1) with
armed conflict as the dependent variable.
b. Determine the lag structure of the unrestricted error correction model. This lag structure
selects a lag for each of the endogenous variables in the model.
Arts 2018, 7, 22
11 of 26
c. Make sure the model is well-specified (no serial correlation or autocorrelation and the model
is stable).
3.
Perform a bounds test for cointegrating relationships.
4.
Repeat Step 3 and all component steps with antiquities looting (or cultural property crime) as the
dependent variable.
Dostları ilə paylaş: |