I have seen a couple of articles using for instance the log of GDP and it's square in an ARDL model which has both long run and short run while the variables appeared in both part of the model. Is there anything wrong with this or it's better to use only the level of the variable in ARDL assuming short run and long run differs.