sbenthall
commented
4 years ago Thinking over the previous discussion, I have a couple points to add.
First, from the perspective of the model definition:
- I think it would make sense for death to be a model variable any other, in the sense that it could be an exogenous shock or endogenous state. Why not?
- That means it would be specified as part of a particular frame.
- Where death has additional semantics beyond the normal ones is its particular relationship with the reward function.
- We have not yet discussed how we attach reward functions to frames.
There are many open questions, I think, about how we implement the "magic solver" over problem defined as a sequence of frames.
On the simulation side, the death variable is important for modeling population aggregates. It may effect descriptive statistics of the population, or it may effect partial equilibria as an exogenous shock on each individual. I believe these are the easy cases. As they are relatively easy, I think it would be possible to support a wider range of birth/death models besides the fixed population model, which is motivated by a computer memory constraint but not a scientific modeling constraint.
Topic came up in #836 , related to #798
As we transition to a slick "frame" based design (abolishing the state/poststate distinction, etc.), we have the opportunity to rethink the way agents are born and die within simulations.
We will also have the opportunity to try to provide elegant and flexible options for modelers who are interested in different mortality models.
@mnwhite has a proposal in #836 which is that in simulation, "births" can happen at the start of a period, and "deaths" can happen at the end as a form of final shock. That's a possibility worth talking about, I think.