make tracked macro variables and history available from the AgentPopulation

[sbenthall]

• income levels/shocks
• consumption levels
• wealth quantities
• share ratios

question is how to aggregate over the many HARK agents....

Plot these in the notebook output to get a sense of the macro dynamics

[sbenthall]

pLvl - income level

cNrm - consumption (normalized) , control

Share - share of wealth in risky asset, control

aNrm, mNrm -- state variables

[nicksawhney]

The variables are now all being tracked using the AgentPopulation.agent_df function. How should these be aggregated across the agents? If by subpopulation type, how can the subpopulation type of an agent be determined from the agent itself?

[sbenthall]

The method that currently does this for other 'macro' variables is class_stats():

https://github.com/sbenthall/HARK_ABM_INTRO_public/blob/master/HARK/hark_portfolio_agents.py#L169

You can see what it does: it uses the agent's parameters as elements in a dataframe record, then groups the dataframe records by the relevant parameters (CRRA and DiscFac)

I think taking the mean and standard deviation makes sense for aggregation.

[sbenthall]

This is actually more of a feature for the new Population class as it's about exposing macro agent statistics.

[sbenthall]

depends on #81 , #85, #113

[sbenthall]

Hi @alanlujan91 -- tossing this issue your way since it looks like it's really about functionality for the AgentPopulation.

The HARK AgentType can track the variables of its agents and provide them in a history object.

Does the AgentPopulation class support analogous functionality? How could it? If it did, we could pass that through to the simulation output.