search

davidrpugh / assortative-matching-large-firms

Code for solving Eeckhout and Kircher model of assortative matching between heterogenous firms and workers.
MIT License
3 stars 8 forks source link

Add quanecon dep #23

Closed davidrpugh closed 9 years ago

davidrpugh commented 9 years ago

@crisla

I have integrated quantecon.ivp module into our shooting solver code and used some of the module's functionality to compute and plot the residual function for our shooting solver. There are some examples demonstrating use in the Shooting examples notebook. Take a look and see what you think. The shooting solver seems to be quite accurate for the examples that I have tried. You should see how accurate the solver is for your examples.

davidrpugh commented 9 years ago

@crisla

I have noticed that the residuals sometimes blow up to enormous levels in the neighborhood of he boundary conditions (particularly for log-normal workers/firms when approaching x = 0). I suspect that there is a singularity of some sort as x goes to zero which is the root cause of this behavior. It might help to normalize the residuals by the level of either mu or theta. I will try this and see...

crisla commented 9 years ago

The solver doesn't seem to finish for logarithmic distributions and capital. The residuals are ok, but not every node is solved for. The lsoda integrator does say that firm size is negative - maybe for certain distributions and parameters not every firm can exist? Are there some values of x such that there is no possible combination that gives positive profits with positive firm size?

I have to try examples without capital to see if lognormally distributed inputs cause problems.

davidrpugh commented 9 years ago

@crisla

I haven't enabled the logarithmic step sizes on this branch yet. Sounds like the solver is failing at some point during the integration. Currently the solver fails silently! Need to do something about this.