When upgrading to Eviews13 from 12, we noticed that our existing VECM models which uses exogenous variables (e.g. date intercept dummies) were producing different forecasts given the same data set. I presume this is most likely due to new functionality introduced with Eviews13 that allows the user to specify between short-run vs long-run co-efficients for the exogenous variables. However, even after re-estimating the VECMs by including the exogenous variables outside the co-integrating equation (like in Eviews12) we were unable to replicate the Eviews12 forecast output in Eviews13.
I have attached a work file named “Eviews12_workfile” to demonstrate this. Opening the work file in Eviews12, The VAR object and forecasts are produced by running the following commands:
Code: Select all
eviews12_var.ec(c,1) 1 2 debit_value credit_value rt_total @ @expand(@month,@droplast)
smpl 2022m08 @last
eviews12_var.forecast(g, e) f
Now if I reopen the same work file in Eviews13, and perform the same steps as above, I will arrive at different forecasts. Keen to understand what is driving this discrepancy and how I can specify the Eviews13 options to reconcile the two results.
I also noticed in Eviews13’s estimation output window for the VECMs, it only shows the co-efficient of the first date dummy as opposed to Eviews12 - which shows all of them. I.e. Specifying
Code: Select all
@ @expand(@month,@droplast)
In the short-run window only shows the slope dummy for ‘@Month=1’ in Eviews13.