Asymmetric predictors in dynamic panel models?

[Brenden Beck]

Hi all,

In a longitudinal analysis, I suspect that the impact of an increase in a predictor variable will be different than the impact of a decrease. So, I want to decompose the predictor into two "asymmetric" variables following Allison (2019) where one variable captures increases and is otherwise zero and the other variable captures decreases and is otherwise zero. This is straightforward enough, but I also need to include a lagged dependent variable in my model because my outcome is very path dependent.

Does anyone have an idea on how I can include both the two decomposed predictor variables and the lagged dependent variable? Any advice is much appreciated.

Thanks so much,
Brenden

[Sebastian Kripfganz]

The decomposed predictor variables are just interaction terms with a dummy variable indicating whether there is an increase or decrease, aren't they? They are then just treated as any other regressor, and you can apply the usual estimators for dynamic panel data models. I have a few packages for such estimators on my website:
http://www.kripfganz.de/stata/

[Brenden Beck]

Hi Sebastian,
Thanks so much, this is super helpful. Right now, I have the predictor variables structured as two separate variables, but I'll try this interaction term approach you recommended.
Thanks again,
Brenden