Numerical or model error?

Hi! The following model has two AR(1) processes:

log(theta) = aTheta + bTheta log(theta(-1))+epsTheta log(Chi) = aChi + bTChi log(Chi(-1))+epsChi
Normally, if I shock theta, Chi should show no reaction. But in my case Chi behaves very strange: after three periods it jumps from 0 to -2.7756e-17. At period 28 it jumps back to 0. Is it a dynare/numerical error or can it lie on the model? Thanks in advance!

P.S.: both values from parameterfile:

chiS = 1.181357861634841e+003 chiN = 1.273992578596610
small.mod (2.47 KB)

Usually, anything smaller than 1e-10 is numerical error and indistinguishable from 0.

Thanks a lot!