I have used the ARDL model to model data over ten years. The results of the bounds test indicate that there is a cointegration relationship at the 5% and 10% levels. However, the ARDL and ECM results are not satisfactory. What is your suggestion for me? Should I change the model?
Taranom Rostamian
When working with the ARDL model and encountering unsatisfactory results, it is important to first ensure that the model is correctly specified, including selecting appropriate lag lengths and variables. Conducting diagnostic tests for issues such as autocorrelation and heteroskedasticity as well as checking the stability of the coefficients can help verify the robustness of the model. Additionally, transforming the data or considering alternative cointegration approaches, such as the Johansen cointegration test or VECM, might improve the model’s performance.
If these steps do not lead to better outcomes, exploring models that capture asymmetries in the cointegration relationship, such as Nonlinear ARDL (NARDL), could be beneficial. It is also advisable to consult the recent literature on cointegration analysis for advanced techniques or alternative approaches that might be more suitable for your dataset. Ultimately, the choice of model should align with the theoretical background of the data, its nature, and the specific research questions you are addressing.
Hi
take a look at this paper ...
خطای متداول در کاربرد مدل های سری زمانی: کاربرد نادرست مدل ARDL (مدل خودرگرسیونی و توزیع با وقفه)
please check cointegration test and if all your variables are cointegrated simply use VECM.
- Badges
- Science topic