Simple question, I hope. I’m currently trying to fit Poisson process models, where the underlying process is (e.g., currently) a local level model with mean reversion (and I may see if an inertia term does anything, when I get this to work without complaining). I model this as a state space model. My current script works, but not perfectly, and I interpret Stan to be telling me that I should reparameterise the innovation process somehow (because the currently constant volatility of this is strikingly linearly correlated with energy__). I suspect the problem has something to do with the fact that I model the log of the underlying process (for the obvious reasons), but t.b.h. I don’t know, and the only pair-plot that looks really badly wrong is what I just mentioned.
So far, my google fishing trip has dragged up nothing, so…
Alternative 1 would be to stare at the model for several days, in the hope that something occurs to me, possibly in vain.
Alternative 2 is to ask here if anybody has gotten such a thing up and r running with no-complaints from Stan, and, if such a person does exist, if they might point me in the direction of a model, or provide wise advice.
Many thanks in advance for any interesting suggestions,
Sean Matthews