Cointegration with non-significant long-run coefficients
Posted: Thu Dec 29, 2011 12:51 pm
Hi,
I would like to pose an interesting (to my view :)) question related to the topic of cointegration.
Recently, by implementing the ARDL approach to cointegration in order to establish long-run relationship between two variables (one dependent and one independent) around a linear time trend (e.g. y=a+bt+cx), I faced the following contradicting result:
While, there was a clear evidence of cointegration (the estimated F-Statistic was well above the related set of critical values) and at the same time the diagnostic testing results of the underlying ADRL specification where excellent (serial correlation, heteros?edasticity, normality and functional form), the resulted long-run coefficient (c in the above example) turned out to be clearly insignificant!!
Continuing with the error correction specification the error correction term appeared to be highly significant!
Therefore, given the existence of the long-run relationship and the significance of the error-correction term, what reason may result to the insignificance of the long-run coefficient? Can be justified theoretically something like that?
Thanks in advance
Variance
I would like to pose an interesting (to my view :)) question related to the topic of cointegration.
Recently, by implementing the ARDL approach to cointegration in order to establish long-run relationship between two variables (one dependent and one independent) around a linear time trend (e.g. y=a+bt+cx), I faced the following contradicting result:
While, there was a clear evidence of cointegration (the estimated F-Statistic was well above the related set of critical values) and at the same time the diagnostic testing results of the underlying ADRL specification where excellent (serial correlation, heteros?edasticity, normality and functional form), the resulted long-run coefficient (c in the above example) turned out to be clearly insignificant!!
Continuing with the error correction specification the error correction term appeared to be highly significant!
Therefore, given the existence of the long-run relationship and the significance of the error-correction term, what reason may result to the insignificance of the long-run coefficient? Can be justified theoretically something like that?
Thanks in advance
Variance