Some clarifications

I had some clarifications regarding specifying some variables,
1- Is it possible to specify a markov transition matrix for a variable ?
2- Is it possible to have an exogenous variable normally distributed with mean different from 0, which is not shocked ?

  1. No. The shock processes in Dynare are continuous variables
  2. This is equivalent to adding a constant to a mean 0 normally distributed variable. You can enter it this way.

Thanks a lot.

Thanks for you reply,
I’m trying to replicate this paper ( with some modifications. Their appendix has the dynare code.

I have coded my model and I get the required steady states for the log-linear model.
I then convert the logged variables N, into its percentage level which is the rate of employment. I then subtract it from one to get the unemployment rates.

However, I am unable to estimate my parameters. I first got a remark that I should use the use_calibration option with the estimated_params_init block. Once I did this the
next error was that “Cannot use parameter values from calibration as they violate the prior bounds”. I have tried it with “s” which is the job separation rate and “bet” the discount factor.

Both give the same errors,
I am wondering what the issue could be. My data file of unemployment rates has been attached too.

Thanks in advance.
shubh.mod (1.2 KB)

I am unable upload my data file it seems, however the data is monthly unemployment rates from germany for the past 6 years.

Please upload everything in one zip-file.

zip file (7.42 KB)

  1. The shocks-block belongs before the estimation command, otherwise the variance will be set to 0 and use_calibration will fail due to stochastic singularity of your calibration.

  2. The data in the Excel file has no headers. I added them

  3. You did not divide your unemployment rate in the data by 100 to make it equivalent to the variable in your model. I also did that.

A running version is attached. (9.12 KB)

Thanks a lot. Also when we have the log-linear model the impulse responses are in terms of the log deviations from steady state which can be interpreted as percentage deviations. However, I was wondering if we could get the impulse responses in level terms. The log-variables converted into level terms also have the irf with 0 steady state, i was wondering how we could get the irf’s for the level variables in terms of the levels itself.


Just add the steady state to them. You must do that manually. See the manual on where the steady states and the IRFs are stored.

Thanks a lot, another query that I had is,
In my data if I have monthly data from 2007 on wards till 2014, then in my model if I shock my exogenous variable in the 12th period will it correspond to a shock in 2008 and be incorporated in the estimation process while using my data ?
Kind regards,

Also, When I have one stochastic shock process and one deterministic shock (say in period 12), Is the deterministic shock anticipated or unanticipated ?
Is there any way to have an unanticipated shock in a later period such that agents dont respond to it during the first 11 periods ?

Maybe I am misinterpreting what you are asking. Please try to explain what you are trying to model.

First post: No, the specified deterministic shocks using varexo_det do not enter estimation at all. The only information used during estimation is the data and all shocks are inferred from the data.

Second post: From the manual entry for varexo_det

[quote]It is possible to mix deterministic and stochastic shocks to build models where agents know
from the start of the simulation about future exogenous changes.[/quote]

That means, there is perfect forsight here: the agents know at time 0 that there will be a one-time shock at time 12.

The “unanticipated shock in a later period such that agents dont respond to it during the first 11 periods” is just your standard stochastic shock.

Thanks a lot Sir,
I mostly had doubts regarding what you explained and things are much clearer to me now, so thanks a lot.

Just to clarify
1 - What did you exactly mean by that all shocks are inferred from the data ? To what extent does the specification of the stochastic shocks matter as in its AR process and standard deviation in estimation.

2 - There is no way to have a surprise deterministic shock, because a fall in the deterministic shocks value in some period after the economy begins will always be taken into account in previous periods because of prefect foresight.


  1. With the shocks-block you specify the stochastic processes of the state-space system. The autocorrelation and variance of the processes is either fixed or estimated. But the actual shock realizations are estimated as the residuals from the state-space system.

  2. Sorry, but I do not know that a “surprise determinstic shock” is. This is an oxymoron.

Thank you, For the second point I meant If it was possible to have an unanticipated shock in the future ?

Estimation allows unanticpated shocks at any point in time in the future. If you are asking whether you can pre-specify the value of an unanticipated shock happening at a future point in time that is taken as given in estimation, the answer is no.

What about the scope of pre-specified values of unanticipated shocks in the deterministic and stochastic models and just analyzing their transition to a new steady state ?
before any estimation takes place.

Also regarding your paper - … edirects=1

My model is calibrated with weekly estimates of parameters and the steady states correspond to the weekly real world values of endogenous variables.
However, my data is the monthly unemployment rate.
I was wondering if this would lead to spurious estimation results. But your paper mentions that for a stock variable the length of the model period or frequency does not matter. And since unemployment rate is an average or stock variable this should not affect the estimation. Am i correct ?

Or should I use the kalman filter with missing data for some weeks as mentioned in your paper ?
My observations are monthly unemployment rates.