Zero acceptance rate

I’m running a bayesian estimation with mode 6 in Dynare 4.1.3 with Matlab R2009a. Metropolis gets started and runs with good acceptance rates, but sometimes after a while it reaches a point where acceptance rates go to zero. After they go to zero, they do not converge back to normal value of acceptance rates. It does not seem to be a problem with the mod file because it ran well in other attempts. Does anybody have a clue on what might be happening? Could it be a problem with the memory of the machine? Thanks a lot

try plotting the draws in this case. It may be that the MCMC runs to the boundary of the stability region or to the prior bounds. All new draws are then discarded because they violate the Blanchard-Kahn conditions or the prior.