Estimating an ARDL model


I am doing a project where I have to regress the interest rate on the GDP using an ARDL model. To determine the best number of lags according to the Akaike Information Criteria I have used the following function:

ARDL <- auto_ardl(estimation_GDP ~ estimation_ir, data= merged_data, max_order =c(12,12))
print(ARDL$top_orders) # The model with the lowest AIC is ARDL(1,0)

The problem that I have is that the best model according to this function is an ARDL(1,0), and the p-value of the coefficients is extremely high. I have also tried to estimate the same model with EViews, and I got that my best model was an ARDL(1,6), so I am not sure if I have done something wrong.

I would highly appreciate your help. Thanks!

I believe your code is regressing GDP on the interest rate, rather than the other way around.

If you read the help file of the auto_ardl function you will see that the default searching method (you can change this via the search_type parameter) is a stepwise regression that doesn't take into account all the possible combinations. So sometimes it finds a local optimum instead of a global one.

Instead, you can either lower down the possible model combinations by restricting via the parameters fixed_order or starting_order) or simply set grid = TRUE that searches all the possible model combinations that it ensures finding the globally best model.

The drawback of this is that it takes some time, depending on the maximum number of orders and the number of independent variables. In your case where you only have 1 independent variable, it will be quite fast.

This topic was automatically closed 54 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.