Putting bounds on stochastic simulations


I am performing stochastic simulations, and I am wondering whether it is good practice to put bounds on or truncate the parameter distributions. I ran some simulations, and some of the ka values I got, in particular, don’t seem feasible. I have a tvka of 10, and the max individual ka being simulated is >4000.

Are there any general rules on how to decide on the bounds for parameters?

There are different opinions on this matter so my comment here doesn’t represent a consensus view. In my opinion, you have a problem with your model if it gives unreasonable individual parameter values. I’d try to understand why and see if the model can be improved to better fit the data. If you introduce truncation of the individual parameters then I’d argue that such truncation should also be applied during estimation. Otherwise, you don’t simulate from the model that you estimated and there is a risk that your conclusions will be wrong.