I would like to know what to consider when deciding about lag (p) and horizon (h) to estimate impulse responses with local projections. Jorda (2005) and Plagborg-Moller & Wolf (2019) talk about the details of how to implement local projections, but I could not find enough information about the choice of lags and horizons.
I have quarterly data from 1990 till 2019. I consider choosing lag of 4 because the data is quarterly. Would that be right? However, I am not sure about which horizon to choose to get better results. Any idea about that?