How exactly RandomWalk priors work under the hood? Is it true that on each new coefficient we are using all data up to that time point? Or only the latest point?
I’ve put together a notebook showing an equivalent way of construction the
GaussianRandomWalk distribution that may help you understand.