Keep in mind that there’s huge fixed costs to simulations. I wrote code in May that is still running. As in, it hasn’t converged yet. This was suppose to be my job market paper.

This summer was spent trying to find ways to make my code run faster. There’s not much economics in that.

There’s not much economics in the algebra involved in getting analytical solutions to your models, either. But at least the end result is an equation that you can give an economic interpretation to.

Also, my sense is that these sorts of papers take a long time to get published. This is probably because the models are not very transparent. People giggle when I say that my paper will have “simulated comparative statics”.

Also, don’t use R to simulate stuff in. Its really slow and there’s not much you can do about it. Otherwise, R rocks.

Is this a macro model? Have you put 200 shocks and 500 free parameters in it? ;p

I’m confused. Is it taking so long because you are using R or because of the model itself?

Also, as an aside, why did you choose R as opposed to MATLAB or Gauss?

EE, I don’t know. That’s the problem. I have heard from reputable people that R is slow for the kind of stuff I’m doing (solving non-linear systems of equations and iterating over loops).

I choose R because I like to go against the grain and R is cool.

Any reason you can’t run the sim on a GPU? Seems like the kind of problem that would benefit from lots of fast parallel floating point units.

I don’t think equation solvers are parallelizable…

Try springing for 8 gb of ram!… (Does ‘R use Ram like stata?) You can probably get 8 gigs these days for sub-$2k…

si… its not hitting a memory limit tho… its the plugging and chugging that’s slowing it down.

Hell, even one of those macro models that Gabriel is referring to will take less time to converge, usually minutes if done in Matlab. What the heck ARE you simulating?

Nothing special… I’m finding an equilibrium but I have to solve a system of non-linear equations for each node in the state-space for every iteration in a value function iteration and then all of that is in a loop trying to find a stationary distribution.

My advisor joked it would be faster to use the economy to “simulate” my model.