However, if the order of integration of any of the variables is greater
than one, for example an I(2) variable, then the critical bounds provided
by Pesaran et al. (2001) and Narayan (2005) are not valid. They are
computed on the basis that the variables are I(0) or I(1). For this purpose,
it is necessary to test for unit root to ensure that all the variables
satisfy the underlying assumptions of the ARDL bounds testing approach
of cointegration methodology before proceeding to the estimation
stage. In order to overcome the low power problems associated
with conventional unit root tests, especially in small samples,we therefore
prefer the weighted symmetric ADF test (ADF-WS) of Park and
Fuller (1995). It requires much shorter sample sizes than conventional
unit root tests to attain the same statistical power. Leybourne et al.
(2005) have recently noted that ADF-WS has good size and power
properties compared to other tests.
Basically, the ARDL bounds testing approach of cointegration involves
two steps for estimating long-run relationship. The first step
is to investigate the existence of long run relationship among all
variables in the equation. The ARDL model for Eq. (1) may follow
as:
Δcot ¼ α1 þ
Xa1
g¼1
α2gΔcot−g þ
Xb1
h¼0
α3hΔect−h þ
Xc1
i¼0
α4iΔyt−i
þ
Xd1
j¼0
α5jΔy2
t−j þ
Xe1
m¼0
α6mΔopt−m þ
Xf 1
n¼0
α7nΔf dt−n þ δ1cot−1
þδ2ect−1 þ δ3yt−1 þ δ4y2
t−1 þ δ5opt−1 þ δ6f dt−1 þ ε1t
ð2Þ
where ε1t and Δ are the white noise term and the first difference
operator, respectively. An appropriate lag selection based on a criterion
such as Akaike information criterion (AIC) and Schwarz
Bayesian Criterion (SBC). The bounds testing procedure is based
on the joint F-statistic or Wald statistic that is tested the null of
no cointegration, H0:δr=0, against the alternative of H1:δr≠0,
r=1, 2, …, 6. Two sets of critical values that are reported in
Pesaran et al. (2001) provide critical value bounds for all classifications
of the regressors into purely I(1), purely I(0) or mutually
cointegrated. If the calculated F-statistics lies above the upper
level of the band, the null is rejected, indicating cointegration. If
the calculated F-statistics is below the upper critical value, we cannot
reject the null hypothesis of no cointegration. Finally, if it lies
between the bounds, a conclusive inference cannot be made without
knowing the order of integration of the underlying regressors.
Recently, Narayan (2005) argues that existing critical values
which are based on large sample sizes cannot be used for small
sample sizes. Thus, Narayan (2005) regenerated the set of critical
values for the limited data ranging from 30–80 observations by
using the Pesaran et al.'s (2001) GAUSS code. With the limited
النتائج (
العربية) 1:
[نسخ]نسخ!
However, if the order of integration of any of the variables is greaterthan one, for example an I(2) variable, then the critical bounds providedby Pesaran et al. (2001) and Narayan (2005) are not valid. They arecomputed on the basis that the variables are I(0) or I(1). For this purpose,it is necessary to test for unit root to ensure that all the variablessatisfy the underlying assumptions of the ARDL bounds testing approachof cointegration methodology before proceeding to the estimationstage. In order to overcome the low power problems associatedwith conventional unit root tests, especially in small samples,we thereforeprefer the weighted symmetric ADF test (ADF-WS) of Park andFuller (1995). It requires much shorter sample sizes than conventionalunit root tests to attain the same statistical power. Leybourne et al.(2005) have recently noted that ADF-WS has good size and powerproperties compared to other tests.Basically, the ARDL bounds testing approach of cointegration involvestwo steps for estimating long-run relationship. The first stepis to investigate the existence of long run relationship among allvariables in the equation. The ARDL model for Eq. (1) may followas:Δcot ¼ α1 þXa1g¼1α2gΔcot−g þXb1h¼0α3hΔect−h þXc1i¼0α4iΔyt−iþXd1j¼0α5jΔy2t−j þXe1m¼0α6mΔopt−m þXf 1n¼0α7nΔf dt−n þ δ1cot−1þδ2ect−1 þ δ3yt−1 þ δ4y2t−1 þ δ5opt−1 þ δ6f dt−1 þ ε1tð2Þε1t و Δ فيها مصطلح الضوضاء البيضاء والفرق الأولىعامل، على التوالي. تحديد الفارق مناسب استناداً إلى معيارمثل أكايكي المعلومات معيار (AIC) وشفارتزمعيار النظرية الافتراضية (اتفاقية بازل). حدود ويستند إجراء الاختبارعلى إحصاء F المشتركة أو والد الإحصائيات التي يتم اختبارها فارغة (null) منلا كوينتيجريشن، H0:δr = 0، مقابل بديل H1:δr≠0،r = 1, 2,..., 6. مجموعتين من القيم الهامة التي يتم الإبلاغ عنها فيبيسارن et al. (2001) توفير حدود القيمة الحرجة لجميع التصنيفاتمن ريجريسورس إلى I(1) بحتة، I(0) بحتة أو بعضها بعضاcointegrated. واو-الإحصاءات المحسوبة يكذب أعلاه الجزء العلويمستوى من الفرقة، ورفض null، تشير إلى كوينتيجريشن. إذا كانواو-الإحصاءات المحسوبة أقل من القيمة الحرجة العليا، وإننا لا نستطيعرفض الفرضية خالية من لا كوينتيجريشن. وأخيراً، إذا أنها تقعلا يمكن جعل بين الحدود، استنتاج قاطع دونمعرفة ترتيب تكامل ريجريسورس الأساسية.في الآونة الأخيرة، يقول نارايان (2005) أن القائمة القيم الحرجةلا يمكن استخدام القائمة على العينات ذات الأحجام الكبيرة للصغيرةحجم العينة. وهكذا، مجدد نارايان (2005) المجموعة من الحرجالقيم للبيانات المحدودة التي تتراوح ما بين 30 – 80 الملاحظات التي أبداهااستخدام بيسارن et al. (2001) قانون جاوس. مع محدودية
يجري ترجمتها، يرجى الانتظار ..
