Interpreting panel cointegration test using Kao test

[carlton durrant]

Dear forum users

I have two models for which I am testing cointegration using kao test as follows. Can I say model two is not cointegrated while model two is because of the MDF and ADF ?

xtcointtest kao ROA CATETA LOGSIZETA MENIETA MESCNIE LIQLND AQLNTA GDPG LEVDTEQ HHILN
Kao test for cointegration
Ho: No cointegration		Number of	panels	37
Ha: All panels are cointegrated		Number of	periods	8
Cointegrating vector: Same
Panel means: Included		Kernel:		Bartlett
Time trend: Not included		Lags:		1.24 (Newey-West)
AR parameter: Same		Augmented	lags:	1
		Statistic		p-value
Modified Dickey-Fuller t		-6.4144		0
Dickey-Fuller t		-9.5496		0
Augmented Dickey-Fuller t		-3.6762		0.0001
Unadjusted modified Dickey-Fuller	t	-7.7288		0
Unadjusted Dickey-Fuller t		-9.9544		0

model 2
xtcointtest kao ROA LOGSIZEIN LEVTLTA MENIETA NIMLN LIQLND AQLNTA GDPG HHILN
Kao test for cointegration
--------------------------
Ho: No cointegration		Number of	panels	37
Ha: All panels are cointegrated		Number of	periods	8
Cointegrating vector: Same
Panel means: Included		Kernel:		Bartlett
Time trend: Not included		Lags:		1.78 (Newey-West)
AR parameter: Same		Augmented	lags:	1
		Statistic		p-value
Modified Dickey-Fuller t		-0.6945		0.2437
Dickey-Fuller t		-6.0225		0
Augmented Dickey-Fuller t		0.1705		0.4323
Unadjusted modified Dickey-Fuller	t	-8.1348		0
Unadjusted Dickey-Fuller t		-10.3341		0

[Clara Joan Joachim]

Hello, I have questions about Kao cointegration test too. Therefore, I posted here following the same topic.
First, I tested this with all my variable:

Kao test for cointegration
--------------------------
H0: No cointegration                        Number of panels       =      8
Ha: All panels are cointegrated             Number of periods      =     30

Cointegrating vector: Same
Panel means:          Included              Kernel:           Bartlett
Time trend:           Not included          Lags:             2.62 (Newey–West)
AR parameter:         Same                  Augmented lags:   1 (BIC)
------------------------------------------------------------------------------
                                            Statistic         p-value
------------------------------------------------------------------------------
 Modified Dickey–Fuller t                          .               .
 Dickey–Fuller t                                   .               .
 Augmented Dickey–Fuller t                         .               .
 Unadjusted modified Dickey–Fuller t        -22.7420          0.0000
 Unadjusted Dickey–Fuller t                 -12.6351          0.0000
------------------------------------------------------------------------------

I dont understand why it does not shows result for MDF, DF and ADF. Then I tried replacing my variable one by one and get this:

Kao test for cointegration
--------------------------
H0: No cointegration                        Number of panels       =      8
Ha: All panels are cointegrated             Number of periods      =     30

Cointegrating vector: Same
Panel means:          Included              Kernel:           Bartlett
Time trend:           Not included          Lags:             2.25 (Newey–West)
AR parameter:         Same                  Augmented lags:   1 (AIC)
------------------------------------------------------------------------------
                                            Statistic         p-value
------------------------------------------------------------------------------
 Modified Dickey–Fuller t                   -18.4876          0.0000
 Dickey–Fuller t                            -12.3187          0.0000
 Augmented Dickey–Fuller t                   -9.2722          0.0000
 Unadjusted modified Dickey–Fuller t        -22.7654          0.0000
 Unadjusted Dickey–Fuller t                 -12.6060          0.0000
------------------------------------------------------------------------------

Results shows that I could reject my null hypothesis. Question is, why it doesn't work for all variables? I did unit-root test before and confirmed they all cointegrated in level and first difference. Anyone who can help me explain this situation and what should I do?
Thank you!

[Duc-Thai-Nam Nguyen]

Hi Clara,

Taking a step back: Why do we test for cointegration? It is because when first differencing variables to solve for the unit roots, we drop valuable long-run information in levels. Cointegration is meant to help us model non-stationary variables that share a common stochastic trend. Thus, all variables in your cointegration test must be I(1). And thus, a unit root test is needed before you test for cointegration.
In this case, I believe some of the variables are I(0) indeed. My suggestion is: run the unit root tests if you have not, note I(0) and I(1) variables, then test for cointegration on I(1) variables only.

Hope that helps.