stochastic vs. deterministic model

[Fontana]

Hi,

basic question, I am having the old problem to decide whether my model is stochastic or deterministic.

My model is new keynesian and has several shock processes which are AR(1), including an error term which is i.i.d and has E(x)=0.

Now, I want to shock the model for instance with permanent shocks or shocks that start at a specific point and last a pre determined period length. Does this make my model deterministic or am I in a stochastic model with deterministic shocks? Do I need to transform the model, to make it deterministic?

Thanks,
Fontana

[AssiaEzzeroug]

Hi,

you need to declare all your shocks as exogenous in the deterministic case.

Best

[Fontana]

But should I use stoch_simul or simul in the end to compute the model? Or should I seperately compute the solutions for a det and a stoch case, respectively?

[AssiaEzzeroug]

Hi,

if your model is deterministic (see my previous answer), you need to use the command "simul". If you didnt change anything about the NK model you were talking about at the beginning of this post then you can use the instruction stoch_simul since your model is stochastic.

Best

[Fontana]

That is exactly my problem, the model is still the same as described above, but I want to shock it with deterministic shocks (ie shocks lasting for a predet. period or permanent shocks). For this purpose, do I need to change my stoch model to a deterministic one?

[pakocica]

Dear Fontana,

I think you don't need to change the model into deterministic one. If I understand it well, your model is stochastic, but you want to observe Impulse Response Functions to specific shocks. That is exactly what the function stoch_simul will do for you.

Are you sure you have read the UsersGuide by Tommaso Mancini carefully enough? You can also read some book to learn more how does Dynare work.

Best!

Pavel

[Fontana]

Ok, I see, that is what I was thinking/guessing.

In fact I am sure I did not read the guide by Thomaso Mancini carefully enough, because there is still plenty of new things I discover everytime I read it. I wasn't able to figure this problem out though, I have also read Collard's 'Dynare in Practice', which was/is quite helpful. I am definitely looking for practical literature on dynare (the material is absolutely new to me), do you have any recommendations?

[otb]

Fontana,

You can keep your model as a stochastic model, but add deterministic shocks by varexo_det designation in the variables section (see pg 7 in the Dynare manual for further details.)

[Fontana]

Thank you otb. I have actually tried varexo_det, it did not return the irfs though.

I included varexo_det after varexo in the mod-file (as described in the manual), so:

....
varexo a m e;
varexo_det i;
....

and declared the shock as in the deterministic case and just used 'stoch_sim' at the end. The Corr Auto corr blocks were all filled with NaNs.

[otb]

I actually haven't used varexo_det myself, so maybe I should let an administrator handle this.

But my understanding is that these deterministic changes specified in varexo_det need to be temporary, so you can't do permanent deterministic changes. This would affect the steady-state which is not allowed since the stochastic model is approximated around a constant steady-state. This could be the source of the problem.

If you're analyzing a one-time unexpected but permanent change, then you should build the model as a deterministic model, and make the model calculate the transition path from an initial to a new steady-state.

otb

[Fontana]

Yes, this was my understanding of it too.

I am going to make the model deterministic now, since my shocks are permanent.

Thank you for your help.

Fontana

[pakocica]

Otb, I think whether you can use permanent shock in the stochastic model or not depends on the size of the shock. If the shock does not change the local dynamics of the model, e.i. does not change the approximation at the point in a relevant way, then the permanent shock should not be a problem. (Moreover, if the first order approximation was infected by the shock too much, you could use second order approximation or a higher order approximation in Dynare++) It can be simply designed so that we put

rw_e = rw_e(-1) + e,

where e is an exogenous variable with single period shock and the desired permanent shock rw_e, in this case defined as an endogenous variable.

Fontana, actually I did not read the guide by Thomaso Mancini carefully enough myself too. I am new to the field too and I started reading the book "Structural Macroeconomics" by D. N. Dejong and Ch. Dave. The book is written a very nice way, it describes all the basic concepts and methods used by Dynare and the book seems to be practical - besides the theory, gives you clear instructions how to design your model.

If someone has good experience with other books, please, give me a reference.

Nice weekend to everyone!

Pavel

[Fontana]

Dear Pavel,

thanks for your reply. I did not know that the size of the shock mattered in any way, but now that you are saying it makes sense.

That book has been recommended to me from others as well, and I think I will give it a try.

Thanks and best regards,
Fontana

[yangv007]

thanks for everybody , i changed my model to determinstic. this problem puzzled for a long time ... great thanks.