ARDL - Different Lags for Dynamic Regressors

[Crushtank]

Hi everyone,

for my master's thesis I use EViews 9.5 with its latest updates to estimate an ARDL model. The goal is to find a longterm relationship between the risk premium for real estate (dependent variable) and the risk free rate and rental growth rate as dynamic regressors. The available data is quarterly. For now, I've been following the instructions of Dave Giles on his blog.

I've found a good fitting model (8, 6, 2). However, when I reproduce it in a VAR, one unit root lies outside the unit circle, which makes it unstable. I've added another lag for the last variable in the VAR, making it an 8, 6, 3 ARDL-model, which would make it stable.

Now, I'm having trouble reproducing that in an actual ARDL to find the longterm relationship. I'm not sure if it's valid to let EViews estimate the model automatically based on the AIC and to add this one lag as a fixed regressor. Would the result be any different if EViews added this lag in the dynamic setting by itself?

When I add the lag as a fixed regressor I get the longterm relationship but with the added lag in it, though insignificant. I've attached the result as a JPEG. Could I just ignore it?

I look forward to your answers.