Sensitivity of convergence


I have a large DSGE model with that uses occasionally binding constraints to model sudden stops. I am trying to use Occbin to solve the model and simulate it for 10000 periods, but it seems that if I use parameter values that make the constraint bind harshly (so that the associated Lagrangian multiplier takes high values when the constraint binds), Occbin does not seem to provide me with solutions. Either Occbin gets stuck, or it tells me that there are “loops” in the solution so it cannot converge.

The large non-linearities as a result of harshly binding constraints are essential to the model, so I would like there to be large changes when the constraint binds. One of the features of the model is that a large temporary shock in period 1 can cause the constraint to bind in period 5 because of sluggish movements in capital and net worth of banks.

Could this feature be a reason why there are “loops” in the solution and it does not converge? Is there an example model/code that uses Occbin to deal with occasionally binding collateral constraints in the context of sudden stops?

Thank you.

I was able to resolve some of this problem by reducing the “simul_check_ahead_periods” from 10000 to 20. Now I do not get “stuck” solving the model and do not get loops.

My understanding was that having a higher “simul_check_ahead_periods” would lead to more convergence than when I have lower “simul_check_ahead_periods” since you look for convergence to the reference regime farther in the future.

Is there a reason why having lower “simul_check_ahead_periods” would lead to more convergence?

That can happen due to numerical reasons. See point 2 at