The results in the Cocco, Gomes, Maenhout (2005) Remark are qualitatively similar to those in the original paper, but they do not perfectly line up (quantitatively).
When Matt Zahn and I initially created the REMARK, we had trouble getting Gomes's public Fortran code to run. After changing a few things we got it to run but got different results from those in the paper. Recently, Ruggero Jappelli has translated the fortran code to Matlab, and confirmed that it generates results that very closely match those from the original paper (so the previous inconsistencies were due to the changes we made).
So the puzzle now is why we are getting different results. We are using a different solution method (endogenous gridpoints), but in principle any solution method should converge to the "true" solution. Therefore, I think what is going on is that we probably misinterpreted some normalization of the income process, or that the discretization of some shock is different in a meaningful way.
In any case, the enterprise of resolving these differences is one that I am sure will yield a lot of knowledge (about the properties of the different solution methods, the effects of discretizations, or catching a potential bug?). It would also be good to have results that align with those of the paper.
I am therefore creating this issue for any future portfolio choice enthusiast with time on their hands.
The results in the Cocco, Gomes, Maenhout (2005) Remark are qualitatively similar to those in the original paper, but they do not perfectly line up (quantitatively).
When Matt Zahn and I initially created the REMARK, we had trouble getting Gomes's public Fortran code to run. After changing a few things we got it to run but got different results from those in the paper. Recently, Ruggero Jappelli has translated the fortran code to Matlab, and confirmed that it generates results that very closely match those from the original paper (so the previous inconsistencies were due to the changes we made).
So the puzzle now is why we are getting different results. We are using a different solution method (endogenous gridpoints), but in principle any solution method should converge to the "true" solution. Therefore, I think what is going on is that we probably misinterpreted some normalization of the income process, or that the discretization of some shock is different in a meaningful way.
In any case, the enterprise of resolving these differences is one that I am sure will yield a lot of knowledge (about the properties of the different solution methods, the effects of discretizations, or catching a potential bug?). It would also be good to have results that align with those of the paper.
I am therefore creating this issue for any future portfolio choice enthusiast with time on their hands.