Page 1 of 1
0/0 eigenvalues problem
Posted:
Tue Oct 04, 2016 8:02 pm
by lrljack
Hallo All:
I am running a multi-asset allocation model. I encounter the problem of "0/0 eigenvalues". I attach my mod file, the steady state file, and the parameter mat file here. Can you help me to overcome my problems? Actually, I do not want "mu" to be essentially 0. Since r7 is 0, I think lambda7 may be 0.
Thanks!
Re: 0/0 eigenvalues problem
Posted:
Tue Oct 04, 2016 8:30 pm
by jpfeifer
I would focus on the steady state. There are differences in many orders of magnitude for the variables. 10^6 vs 10^(-17) might yield numerical issues when inverting matrices. Is it intended that mu and lambda are essentially 0 in steady state? While you use a steadystate file to compute the steady state, that one essentially just runs a numerical solver. There might be multiple solutions and the one that you currently find yields numerical problems. You should have some intuition whether the steady state you currently have makes sense economically.
Re: 0/0 eigenvalues problem
Posted:
Wed Oct 05, 2016 1:22 pm
by lrljack
Thank you professor. After checking the variance-covariance matrix of the returns, i.e. "sig2", I find that the values are too small. Therefore, I multiply it by 10^4 and the problem disappears. The economic interpretation may be: the variance-covariance matrix is based on values of 100*returns (e.g. instead of using 0.02, the agent use 2). What are your suggestions?
Re: 0/0 eigenvalues problem
Posted:
Fri Oct 07, 2016 3:58 pm
by jpfeifer
If your model is linear or approximated with order=1, then you can just scale everything.