# Deterministic simulations in a model with stock variables

I have a simple RE model with two stock variables, net foreign assets and capital stock, and I look at several deterministic shocks, such as a permanent reduction in an inflation target (i.e. permanent disinflation).

When I solve the model, I need to guess the values of model variables at the end of simulation that enter the “endval” bloc. Some, like inflation, are easy, some, like the two stock variables and the real exchange rate are less obvious.

I went for iteration - first, guessing a set of values for “endval”, then running the deterministic simulation and replacing the original “endval” by values from a period T-40 for sufficiently high T, say, 600. One iteration was enough to obtain paths of model variables converging to a consistent set of “endval” values.

My question is: Is this a correct/conceivable procedure to use in the case of model with stock variables where the end-point steady state values are not immediately obvious?

Dear Jan,

I’m not sure that I understand correctly the problem, so my answer will be quite general

1. The ENDVAL block must contain the asymptotic steady state (and in many cases INITVAL must contain the initial steady state)
2. Either you have an analytical solution for the steady state and you can express ENDVAL values as a function of the parametersof the model, or must rely on a numerical solution provided by STEADY. Note that you can use STEADY twice: once after INITVAL, once after ENDVAL
3. STEADY needs numerical guess values that you provide in INITVAL and, respectively, ENDVAL.
4. Sometimes, it is difficult to come up with good guess values and STEADY doesn’t converge
5. It is an analytical property of stationary nonlinear RE models, that given arbitrary initial and terminal conditions and a large number of iterations between initial and terminal values, that the trajectory will first converge towards the steady state then diverge to finally reach the arbitrary conditions. So, the procedure that you describe is indeed a way to get guess values for the terminal steady state. But, I would recommand that you compute STEADY after ENDVAL to refine these guess values.
6. Of course, STEADY will not work if you have unit roots in your model and I wonder if that maybe the cause of the difficulties that you encounter.

Best

Michel