4
existence of a long-run relationship between variables. How long the long run will be will depend on
the speed of adjustment of the considered factors. This can be estimated by using a short-run
dynamic analysis or the error correction model (ECM).
EMPIRICAL RESULTS
UNIT ROOT TEST
As is well known when dealing with time series data, it is necessary to test for the
presence of unit root in the relevant variables in order to avoid the problem of spurious regression.
The Augmented Dickey-Fuller (
ADF
) unit root test is employed for this task. A number of
maximal augmented lag lengths included in unit root test are limited to four as to provide sufficient
degree of freedom, and the optimal lag length is chosen according to the minimum Akaike
Information Criterion (
AIC
) statistics. Drift and trend are included in the estimating regression of
each variable in level and are excluded in the first difference. We assume that there is no structural
break in any employed time series. The results of the
ADF
unit root test are shown in Table 1.
Table 1
Results of the ADF-unit root test
Variable Level First-difference
LJFDI
-2.64(2)
-3.46*** (2)
LUSFDI
-1.94(3)
-3.22** *(2)
LEUFDI
-3.04(4)
-3.31**(2)
Explanatory Variable
LGDP
-1.83 (3)
3.91*** (1)
LTWR
-2.93 (2)
-3.89*** (2)
LTUC
-2.71 (3)
-4.96*** (3)
LRER
-2.04 (2)
-2.26* (1)
LJPY
-2.68 (3)
-2.80*** (1)
LUKP
-2.69 (3)
-3.95*** (2)
Note: ***, * and * indicate 0.01, 0.05 and 0.10 significance levels, respectively.
The
ADF
results suggest that all relevant variables are non-stationary in level. Further
unit root tests are carried out in the first differences. Since the null of non-stationary is rejected in
this step, we can conclude that the relevant series are
I(1)
. This implies that the traditional OLS
estimation cannot be applied to these data because it could lead to spurious result.
LONG-RUN ESTIMATED RESULTS
The finding that all involving data are stationary in the first differences suggests that a
cointegration test should be applied to test whether the long-run relationship between the variables
exists. The Augmented Engle-Granger (AEG) cointegration procedure is conducted for this purpose,
and the estimated results are shown in Table 2.
