Allow risky return draws to be agent-specific. #1250
Conversation
| # Agent specific simulation | ||
| self.pcct.sim_common_Rriksy = False | ||
| self.pcct.initialize_sim() | ||
| self.pcct.simulate() | ||
| # Assety that all columns of Risky are not the same | ||
| self.assertFalse( | ||
| np.all( | ||
| self.pcct.history["Risky"] | ||
| == self.pcct.history["Risky"][:, 0][:, np.newaxis] | ||
| ) | ||
| ) |
There was a problem hiding this comment.
@sbenthall this is, technically, a stochastic test (its probability of failure is veeeery tiny). But I am not sure how else to test the fact that draws should not be the same for all agents. Any idea?
| @@ -61,6 +61,11 @@ def __init__(self, verbose=False, quiet=False, **kwds): | |||
| if not hasattr(self, "PortfolioBisect"): | |||
| self.PortfolioBisect = False | |||
|
|
|||
| # Boolean determines whether, when simulating a given time period, | |||
| # all agents will draw the same risky return factor (true by default) | |||
| if not hasattr(self, "sim_common_Rriksy"): | |||
There was a problem hiding this comment.
I like Rrisky because R is the return factor and risky is its characteristic. Rfree is Rfree and not Free. Is this a battle for another hill?
There was a problem hiding this comment.
No, I mean it's misspelled. You spelled R-RIKsy.
There was a problem hiding this comment.
Crap. You are right. Thanks!
|
This is cool! I guess we've treated |R as an aggregate shock, and this can make it an idiosyncratic shock? Would there be any sense in generalizing |R such that it can have an aggregate and an idiosyncratic component? Also with other shocks we are more explicit about Agg/Ind but not with this one. |
|
Yes, that might be a future direction that we could go in. But my only goal here is to make it easier for me to get estimates of unconditional mean life-cycle profiles. |
Codecov ReportPatch coverage:
📣 This organization is not using Codecov’s GitHub App Integration. We recommend you install it so Codecov can continue to function properly for your repositories. Learn more Additional details and impacted files@@ Coverage Diff @@
## master #1250 +/- ##
==========================================
+ Coverage 73.27% 73.31% +0.03%
==========================================
Files 76 76
Lines 12532 12548 +16
==========================================
+ Hits 9183 9199 +16
Misses 3349 3349
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report in Codecov by Sentry. |
|
These are deep issues. There is a use case for either way of doing it. For instance, if you are interested in "history-contingent" beliefs a la Malmendier and Nagel, it is absolutely necessary for the returns to be at the aggregate level so that the "depression babies" all live through the experience of the stock market crashing (falling by 89 percent), perhaps inducing a scarring effect. If you are interested in more generic issues like the interaction of life cycle profiles with macroeconomic dynamics, a good case can be made for agents to draw "risky return" shocks that are different for each person -- because you deliberately want to suppress Malmendier-Nagel kinds of phenomena as an annoyance instead of an issue. Another case for allowing idiosyncratic risky returns is when we want to model entrepreneurs or firms. The firm or entrepreneur probably needs to experience firm-specific or entrepreneur-specific shocks, as well as aggregate shocks. In any case, my goal is to have a framework that can accommodate all of these options. This is part of the reason I'm keen on building the models backwards. (That might be obscure, but it does have content). |
Our current implementation of models with risky returns impose the restriction that, when simulating, a single economy-wide return is drawn every period.
This PR adds the option for the returns that are drawn in simulation to be agent-specific.
This is useful, for instance, if you are doing a monte-carlo simulation to find the average consumption at age 45 implied by your model. With the current restriction, there is some inefficiency in the fact that agents' lives are correlated through return draws. Drawing a different shock for every agent breaks this correlation, and (I think?) increases the efficiency of the estimator (for a given
AgentCountandT_sim).