As is known in the literature, the cointegration is investigated for a set of integrated series (at least one unit root) which are in comovement (Granger). However, if all of series are stationary (they have no trend and therefore type I (0)) the cointegration does not exist by definition in this case. The estimation of an VECM for example is reduced to the estimation of a VAR so in my opinion we don't have to go for the cointegration analysis but the causalité analysis can be conducted in the two cases.
As is known in the literature, the cointegration is investigated for a set of integrated series (at least one unit root) which are in comovement (Granger). However, if all of series are stationary (they have no trend and therefore type I (0)) the cointegration does not exist by definition in this case. The estimation of an VECM for example is reduced to the estimation of a VAR so in my opinion we don't have to go for the cointegration analysis but the causalité analysis can be conducted in the two cases.
A VAR will not identify simultaneous contemporaneous effects. You should examine the model for seemingly unrelated time series effects as well. You need to examine it for structural breaks and seasonality before you are finished with your modeling.
Run simple regression analysis , and then the Durbin - Watson auto correlation test is not applicable and no need to worry about auto correlation in regression results , as the variables are stationary.. If you want to test causality, use VAR at level form and apply Granger causality test . Here Vector Error Correction Model for Granger causality is not done as cointegration can not be done .
If all variables are stationary at level, this means there's no long run relationship, a short run relationship may exist and no need for cointegration estimation. But as said above you may need to investigate for causality between them.
if all variables are stationary at level, then that means no co-integration between these times series. you could do further analyses using ARIMA and VAR models.
If the variables are stationary at levels, in this situation regression results are valid in this situation. When variables are unit root at level in this situation regression results are spurious. When the results are valid then no need for co-integration. The idea of co-integration is for integrated series.
When you have a mixture of I(1) and I(0) variables in your VAR, it's advisable to use the bounds testing procedure of Perasan, Shin & Smith (PSS) (2001).
Yes it is the right answer that if variables are all stationary, either OLS , or VAR can be directly applied .Granger causality test can be directly done at level form.data with out any transformation. Cointegration and Error Correction models have not to be applied and not relevant here with Stationary level form data .
If all your variables are stationary at either I(0) OR I(1), kindly run the ARDL model.
Also, if few of your variables are stationary at I(0), while few are stationary at I(1), you can also run ARDL in so far as none of the variables are stationary at I(2). Thanks.
when we have to find out the impact of 5 IDV on stock market volatility. and our DV and 2 IDV are stationary at level and remaining 3 are stationary at 1st difference. than can we apply OLS regression in such situation or ARDL would be used, kindly guide...