Strange Bug

[hewei2004]

Hi,

It seems that the order in which you write down equations matters.

For example, in the attached RBC1.mod

If you move the following four lines:

x = rho*x(-1) + ex;
da = mu + x(-1) + ea;
dc = ca - ca(-1) + da;
di = ia - ia(-1) + da;

Move them to the beginning of the model, versus the end, or take one of the 4 equations and move it around. Switching order seems to change the value of fixed point interest rate a lot, as is evident in the following matlab commands:
load('RBC1.mat'); dyn_ss(dyn_i_rk)

This is very strange. Why should the order in which the equations are entered in the model section matter. I use dynare++ 4.2.4 linux, order=2. The same phenomenon is also in dynare++ 4.2.2 windows

[SébastienVillemot]

I think that your model is a pathological case.

When I run it through Dynare 4.2.4 (not Dynare++), the "check" command returns the following error message:

Code: Select all

error: One of the eigenvalues is close to 0/0 (the absolute value of numerator and denominator is smaller than 1e-6)


[hewei2004]

May I know what I can do to fix this problem?
I do not want to change the parameter values. Thank you.

[jpfeifer]

It suggests that there is something fundamentally unsound in your model in its current configuration. There may be a typo somewhere, the parametrization may be infeasible but numerically borderline stable, or your model might be wrong (which I don't hope). Just from the Dynare output it is hard to judge.

[hewei2004]

Hi,
The model is correct. The problem is with the following parameter:
beta = 0.9975
It is borderline stable.

If I change beta=0.95, now it is very stable, not borderline stable.

Is there a way to reparameterize things to scale up the eigenvalue so it is not less than 1E-6, Is it just a problem of scaling it seems ?